paper_id
stringlengths 9
16
| version
stringclasses 26
values | yymm
stringclasses 311
values | created
timestamp[s] | title
stringlengths 6
335
| secondary_subfield
sequencelengths 1
8
| abstract
stringlengths 25
3.93k
| primary_subfield
stringclasses 124
values | field
stringclasses 20
values | fulltext
stringlengths 0
2.84M
|
|---|---|---|---|---|---|---|---|---|---|
1702.07918 | 6 | 1702 | 2018-02-12T18:33:51 | Quantization of noncompact coverings | [
"math.OA"
] | The concept of quantization consists in replacing commutative quantities by noncommutative ones. In mathematical language an algebra of continuous functions on a locally compact topological space is replaced with a noncommutative $C^*$-algebra. Some classical topological notions have noncommutative generalizations. This article is concerned with a generalization of coverings. | math.OA | math | Quantization of noncompact coverings
September 7, 2018
Petr R. Ivankov*
e-mail: * [email protected]
.
A
O
h
t
a
m
[
6
v
8
1
9
7
0
.
2
0
7
1
:
v
i
X
r
a
The concept of quantization consists in replacing commutative quantities by noncom-
mutative ones. In mathematical language an algebra of continuous functions on a locally
compact topological space is replaced with a noncommutative C∗-algebra. Some classical
topological notions have noncommutative generalizations. This article is concerned with
a generalization of coverings.
Contents
1 Motivation. Preliminaries
1.1 Prototype. Inverse limits of coverings in topology . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
1.1.1 Topological construction . . .
1.1.2 Algebraic construction in brief . . .
. . . .
1.2 Locally compact spaces . . . .
1.3 Hilbert modules .
. . . .
. . . .
1.4 C∗-algebras and von Neumann algebras . .
. . .
. . .
. . .
. . .
. . .
. . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
. . .
. . .
. . .
. . .
2 Noncommutative finite-fold coverings
2.1 Basic construction . . .
2.2
. . . .
Induced representation . . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
3 Noncommutative infinite coverings
3.1 Basic construction . . .
3.2
. . . .
Induced representation . . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
1
2
5
. .
5
. .
. . 12
. . 12
. . 13
. . 13
14
. . 14
. . 16
17
. . 17
. . 22
4 Quantization of topological coverings
4.1 Finite-fold coverings .
4.2
Infinite coverings . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
5 Continuous trace C∗-algebras and their coverings
5.1 Basic construction . . .
5.2 Finite-fold coverings .
5.3
Infinite coverings . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
. . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
. . .
6 Noncommutative tori and their coverings
. . . .
. . . .
. . . .
. . . .
6.1 Fourier transformation . . . .
6.2 Noncommutative torus Tn
. . . .
Θ .
. . . .
6.3 Finite-fold coverings .
. . . .
6.4 Moyal plane and a representation of the noncommutative torus . . .
. . . .
6.5
. . . .
. . . .
Infinite coverings . . .
. . . .
6.5.1 Equivariant representation .
6.5.2
Inverse noncommutative limit
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . . .
. . . .
. . . .
. . .
. . .
. . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
. . .
. . . .
. . . .
. . . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
7 Isospectral deformations and their coverings
7.1 Finite-fold coverings .
7.2
Infinite coverings . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . .
. . .
25
. . 25
. . 28
35
. . 36
. . 37
. . 41
45
. . 45
. . 47
. . 49
. . 50
. . 57
. . 61
. . 62
73
. . 75
. . 78
1 Motivation. Preliminaries
Gelfand-Naımark theorem [2] states the correspondence between locally compact Haus-
dorff topological spaces and commutative C∗-algebras.
[2] (Gelfand-Naımark). Let A be a commutative C∗-algebra and let X be the
Theorem 1.1.
spectrum of A. There is the natural ∗-isomorphism γ : A → C0(X ).
So any (noncommutative) C∗-algebra may be regarded as a generalized (noncommutative)
locally compact Hausdorff topological space. Following theorem yields a pure algebraic
description of finite-fold coverings of compact spaces.
[12] Suppose X and Y are compact Hausdorff connected spaces and p : Y → X is
Theorem 1.2.
a continuous surjection. If C(Y ) is a projective finitely generated Hilbert module over C(X ) with
respect to the action
( f ξ)(y) = f (y)ξ(p(y)), f ∈ C(Y ), ξ ∈ C(X ),
then p is a finite-fold covering.
This article contains pure algebraic generalizations of following topological objects:
• Coverings of noncompact spaces,
• Infinite coverings.
2
This article assumes elementary knowledge of following subjects:
1. Set theory [9],
2. Category theory [15],
3. Algebraic topology [15],
4. C∗-algebras, C∗-Hilbert modules [3, 13].
The words "set", "family" and "collection" are synonyms.
3
Following table contains special symbols.
Symbol
A
A+
AG
Aut(A)
A′′
B(H)
C(X )
C0(X )
Cc(X )
Cb(X )
C (resp. R)
G(cid:16)eX X(cid:17)
K = K (H)
H
K(A)
Meaning
Spectrum of a C∗- algebra A with the hull-kernel topology
(or Jacobson topology)
Cone of positive elements of C∗- algebra, i.e. A+ = {a ∈ A a ≥ 0}
Algebra of G - invariants, i.e. AG = {a ∈ A ga = a, ∀g ∈ G}
Group of * - automorphisms of C∗- algebra A
Enveloping von Neumann algebra of A
Algebra of bounded operators on a Hilbert space H
Field of complex (resp. real) numbers
C∗- algebra of continuous complex valued
C∗- algebra of continuous complex valued functions on a locally
functions on a compact space X
compact topological space X equal to 0 at infinity
Algebra of continuous complex valued functions on a
topological space X with compact support
C∗- algebra of bounded continuous complex valued
functions on a locally compact topological space X
Group of covering transformations of covering eX → X [15]
C∗- algebra of compact operators on the separable Hilbert space H
Hilbert space
Pedersen ideal of C∗-algebra A
lim−→
lim←−
M(A)
Mn(A)
N
N0
U(A) ⊂ A
Z
Zn
k ∈ Zn
X\A
X
[x]
fA′
Direct limit
Inverse limit
A multiplier algebra of C∗-algebra A
The n × n matrix algebra over C∗-algebra A
A set of positive integer numbers
A set of nonnegative integer numbers
Group of unitary operators of algebra A
Ring of integers
Ring of integers modulo n
An element in Zn represented by k ∈ Z
Difference of sets X\A = {x ∈ X x /∈ A}
Cardinal number of a finite set X
The range projection of element x of a von Neumann algebra.
Restriction of a map f : A → B to A′ ⊂ A, i.e. fA′ : A′ → B
4
1.1 Prototype. Inverse limits of coverings in topology
1.1.1 Topological construction
This subsection is concerned with a topological construction of the inverse limit in the
category of coverings.
Definition 1.3.
called a covering projection if each point x ∈ X has an open neighborhood evenly covered
[15] Let eπ : eX → X be a continuous map. An open subset U ⊂ X is said
to be evenly covered by eπ if eπ−1(U ) is the disconnected union of open subsets of eX each
of which is mapped homeomorphically onto U by eπ. A continuous map eπ : eX → X is
by eπ. eX is called the covering space and X the base space of the covering.
[15] A fibration p : eX → X with unique path lifting is said to be regular
Definition 1.4.
if, given any closed path ω in X , either every lifting of ω is closed or none is closed.
Definition 1.5.
components of open sets are open.
[15] A topological space X is said to be locally path-connected if the path
Denote by π1 the functor of fundamental group [15].
Theorem 1.6.
Definition 1.7.
Corollary 1.9.
Proposition 1.8.
is said to be the group of covering transformations of p or the covering group. Denote by
[15] Let p : eX → X be a fibration with unique path lifting and assume that a
nonempty eX is a locally path-connected space. Then p is regular if and only if for someex0 ∈ eX ,
π1 (p) π1(cid:16)eX ,ex0(cid:17) is a normal subgroup of π1 (X , p (ex0)).
[15] Let p : eX → X be a covering projection. A self-equivalence is a
homeomorphism f : eX → eX such that p ◦ f = p. This group of such homeomorphisms
G(cid:16)eX X(cid:17) this group.
[15] If p : eX → X is a regular covering projection and eX is connected
and locally path connected, then X is homeomorphic to space of orbits of G(cid:16)eX X(cid:17), i.e. X ≈
eX /G(cid:16)eX X(cid:17). So p is a principal bundle.
[15] Let p : eX → X be a fibration with a unique path lifting. If eX is connected
and locally path-connected and ex0 ∈ eX then p is regular if and only if G(cid:16)eX X(cid:17) transitively
ψ : G(cid:16)eX X(cid:17) ≈ π1 (X , p (ex0)) /π1 (p) π1(cid:16)eX ,ex0(cid:17) .
Remark 1.10. Above results are copied from [15]. Below the covering projection word is
replaced with covering.
acts on each fiber of p, in which case
Definition 1.11.
containing X as a subspace and the closure X of X is Y, i.e X = Y.
[10] A compactification of a space X is a compact Hausdorff space Y
5
The algebraic construction requires following definition
Definition 1.12. A covering π : eX → X is said to be a covering with compactification if there
are compactifications X ֒→ Y and eX ֒→ eY such that:
• There is a covering π : eY → Y,
• The covering π is the restriction of π, i.e. π = πeX
Example 1.13. Let g : S1 → S1 be an n-fold covering of a circle. Let X = eX = S1 × [0, 1).
The map
.
π : eX → X ,
π = g × Id[0,1)
is an n-fold covering. If Y = eY = S1 × [0, 1] then a compactification [0, 1) ֒→ [0, 1] induces
compactifications X ֒→ Y, eX ֒→ eY. The map
π : eY → Y,
π = g × Id[0,1]
ification.
= π. So if n > 1 then π is a nontrivial covering with compact-
is a covering such that πeX
Example 1.14. Let X = C\{0} be a complex plane with punctured 0, which is parametrized
by the complex variable z. Let X ֒→ Y be any compactification. If both {z′n ∈ X }n∈N,
{z′′n ∈ X }n∈N are Cauchy sequences such that limn→∞ z′n = limn→∞ z′n = 0 then form
limn→∞ z′n − z′′n = 0 it turns out
(1.1)
z′′n ∈ Y.
If eX = X then for any n ∈ N there is a finite-fold covering
z′n = lim
n→∞
x0 = lim
n→∞
π : eX → X ,
z 7→ zn.
If both X ֒→ Y, eX ֒→ eY are compactifications, and π : eY → Y is a covering such that
= π then from (1.1) it turns out π−1 (x0) = {ex0} where ex0 is the unique point such
πeX
that following conditions hold:
It turns out(cid:12)(cid:12)(cid:12)π−1 (x0)(cid:12)(cid:12)(cid:12) = 1. However π is an n-fold covering and if n > 1 then(cid:12)(cid:12)(cid:12)π−1 (x0)(cid:12)(cid:12)(cid:12) =
n > 1. It contradicts with(cid:12)(cid:12)(cid:12)π−1 (x0)(cid:12)(cid:12)(cid:12) = 1, and from the contradiction it turns out that for
any n > 1 the map π is not a covering with compactification.
ex0 = lim
n→∞ezn ∈ eY,
n→∞ezn = 0.
lim
6
Definition 1.15. The sequence of regular finite-fold coverings
X = X0 ←− ... ←− Xn ←− ...
is said to be a (topological) finite covering sequence if following conditions hold:
• The space Xn is a second-countable [10] locally compact connected Hausdorff space
for any n ∈ N0,
• If k < l < m are any nonnegative integer numbers then there is the natural exact
sequence
{e} → G (Xm Xl) → G (Xm Xk) → G (Xl Xk) → {e}.
For any finite covering sequence we will use a following notation
S = {X = X0 ←− ... ←− Xn ←− ...} = {X0 ←− ... ←− Xn ←− ...} , S ∈ FinTop.
Example 1.16. Let S = {X = X0 ←− ... ←− Xn ←− ...} be a sequence of locally compact
connected Hausdorff spaces and finite-fold regular coverings such that Xn is locally path-
connected for any n ∈ N. It follows from Lemma 1.6 that if p > q and f pq : Xp → Xq
Corollary 1.9 it turns out
then π1(cid:0) f pq(cid:1) π1(cid:0)Xp, x0(cid:1) is a normal subgroup of π1(cid:0)Xq, f pq (x0)(cid:1). Otherwise from the
G(cid:0)Xp Xq(cid:1) ≈ π1(cid:0)Xq, f pq (x0)(cid:1) /π1(cid:0) f pq(cid:1) π1(cid:0)Xp, x0(cid:1) .
If k < l < m then a following sequence
{e} → π1 (Xl, fml (x0)) /π1 ( fml) π1 (Xm, x0) →
→ π1 (Xk, fmk (x0)) /π1 ( fmk) π1 (Xm, x0) →
→ π1 (Xk, fmk (x0)) /π1 ( flk) π1 (Xl, fml (x0)) → {e}
is exact. Above sequence is equivalent to the following sequence
{e} → G (Xm Xl) → G (Xm Xk) → G (Xl Xk) → {e}
which is also exact. Thus S ∈ FinTop.
is said to be a covering transformation if a following condition holds
inverse limit in the category of topological spaces and continuous maps (cf. [15]).
Definition 1.17. Let {X = X0 ←− ... ←− Xn ←− ...} ∈ FinTop, and let bX = lim←− Xn be the
bπ0 : bX → X0 is the natural continuous map then a homeomorphism g of the space bX
The group bG of such homeomorphisms is said to be the group of covering transformations of
S. Denote by G(cid:16)bX X(cid:17) def= bG.
bπ0 = bπ0 ◦ g.
If
7
verse limit in the category of topological spaces and continuous maps. There is the natural group
Lemma 1.18. Let {X = X0 ←− ... ←− Xn ←− ...} ∈ FinTop, and let bX = lim←−Xn be the in-
isomorphism G(cid:16)bX X(cid:17) ∼= lim←− G (Xn X ). For any n ∈ N there is the natural surjective
homomorphism hn : G(cid:16)bX X(cid:17) → G (Xn X ) andTn∈N ker hn is a trivial group.
Proof. For any n ∈ N there is the natural continuous map bπn : bX → Xn. Let x0 ∈ X0 and
bx0 ∈ bX be such that bπ0 (bx0) = x0. Letbx′ ∈ bX be such that bπ0 (bx′) = x0. If x′n = bπn (bx′) and
xn = bπn (bx0) then πn (xn) = πn (x′n), where πn : Xn → X is the natural covering. Since
πn is regular for any n ∈ N there is the unique gn ∈ G (Xn X ) such that x′n = gnxn. In
result there is a sequence {gn ∈ G (Xn X )}n∈N which satisfies to the following condition
gm ◦ πn
m = πn
m ◦ gn
where n > m and πn
such that
πn
m .
m : Xn → Xm is the natural covering. The sequence {gn} naturally
which for any n > m satisfies to the following condition
From the above equation and properties of inverse limits it follows that there is the unique
defines an elementbg ∈ lim←− G (Xn X ). Let us define an homeomorphism ϕbg : bX → bX by
a following construction. If bx′′ ∈ bX is any point then there is a sequence {x′′n ∈ Xn}n∈N
x′′n = bπn(cid:0)bx′′(cid:1) .
n ∈ Xnon∈N
On the other hand there is the sequencenx′′bg
x′′bg
n = gnx′′n
n (cid:17) = x′′g
m(cid:16)x′′bg
bx′′bg ∈ bX such that
bπn(cid:16)bx′′bg(cid:17) = x′′bg
The required homeomorphism ϕbg is given by
ϕbg(cid:0)bx′′(cid:1) =bx′′bg.
From bπ ◦ ϕbg = bπ it follows that ϕbg corresponds to an element in G(cid:16)bX X(cid:17) which
mapped onto gn for any n ∈ N. Otherwise ϕbg naturally corresponds to the element bg ∈
lim←− G (Xn X ), so one has the natural group isomorphism G(cid:16)bX X(cid:17) ∼= lim←− G (Xn X ).
From the above construction it turns out that any homeomorphism bg ∈ G(cid:16)bX X(cid:17)
uniquely depends on bx′ = bgbx0 ∈ bπ−1
(x0) ≈−→ G(cid:16)bX X(cid:17). Since the covering πn : Xn → X is regular there is the 1-1 map
bπ−1
bπ−1
n (x0) ≈−→ G (Xn X ). The natural surjective map
n (x0)
It follows that there is the 1-1 map ϕ :
n ; ∀n ∈ N.
0
(x0) → π−1
(x0).
0
8
0
ϕn : π−1
induces the surjective homomorphism G(cid:16)bX X(cid:17) → G (Xn X ). If bg ∈ Tn∈N ker hn is
not trivial thenbgbx0 6=bx0 and there is n ∈ N such that bπn (bx0) 6= bπn (bgbx0) = hn (bg)bπn (bx0),
so hn (bg) ∈ G (Xn X ) is not trivial andbg /∈ ker hn. From this contradiction it follows that
Tn∈N ker hn is a trivial group.
Definition 1.19. Let S = {X0 ←− ... ←− Xn ←− ...} be a finite covering sequence. The pair
(cid:16)Y,(cid:8)πYn(cid:9)n∈N(cid:17) of a (discrete) set Y with and surjective maps πYn : Y → Xn is said to be a
coherent system if for any n ∈ N0 a following diagram
Y
πYn
πn
πYn−1
Xn−1
Xn
is commutative.
Definition 1.20. Let S = {X0 ←− ... ←− Xn ←− ...} be a topological finite covering sequence.
A coherent system (cid:0)Y,(cid:8)πYn(cid:9)(cid:1) is said to be a connected covering of S if Y is a connected
topological space and πYn is a regular covering for any n ∈ N. We will use following
notation(cid:0)Y,(cid:8)πYn(cid:9)(cid:1) ↓ S or simply Y ↓ S.
Definition 1.21. Let(cid:0)Y,(cid:8)πYn(cid:9)(cid:1) be a coherent system of S and y ∈ Y. A subset V ⊂ Y is
said to be special if πY0 (V ) is evenly covered by X1 → X0 and for any n ∈ N0 following
conditions hold:
• πYn (V ) ⊂ Xn is an open connected set,
• The restriction πYn V : V → πYn (V ) is a bijective map.
topology of Y.
Remark 1.22. For any n ∈ N0 the space Xn is second-countable, so from the Theorem 1.32
for any point x ∈ Xn there is an open connected neighborhood U ⊂ Xn.
Remark 1.23. If(cid:0)Y,(cid:8)πYn(cid:9)(cid:1) is a covering of S then the set of special sets is a base of the
Lemma 1.24. Let bX = lim←−Xn be the inverse limit of the sequence X0 ←− ... ←− Xn ←− ... in the
category of topological spaces and continuous maps. Any special set of bX is a Borel subset of bX .
n (Un) ⊂ bX is open. If bU is a special set then
Proof. If Un ⊂ Xn is an open set then bπ−1
n ◦bπn(cid:16)bU(cid:17), i.e. bU is a countable intersection of open sets. So bU is a Borel
bU = Tn∈Nbπ−1
n o(cid:17) ↓ S to(cid:16)Y′′,nπY′′
(cid:16)Y′,nπY′
n o(cid:17) ↓ S is a covering f : Y′ → Y′′ such that
Definition 1.25. Let us consider the situation of the Definition 1.20. A morphism from
subset.
πY′′
n ◦ f = πY′
n
for any n ∈ N.
9
1.26. There is a category with objects and morphisms described by Definitions 1.20, 1.25.
Denote by ↓ S this category.
Lemma 1.27. There is the final object of the category ↓ S described in 1.26.
category of topological spaces and continuous maps. Denote by X a topological space
such that
Proof. Let bX = lim←− Xn be the inverse limit of the sequence X0 ←− ... ←− Xn ←− ... in the
• X coincides with bX as a set,
• A set of special sets of bX is a base of the topology of X .
If xn ∈ Xn is a point then there is x ∈ X = bX such that xn = bπn (x) and there is a special
subset bU such that x ∈ bU . From the construction of special subsets it follows that:
• Un = bπn(cid:16)bU(cid:17) is an open neighborhood of xn;
G
n (Un) =
•
g∈ker(G(bX X )→G(Xn X ))
gbU ;
bπ−1
follows that
onto Un.
• For any g ∈ ker(cid:16)G(cid:16)bX X(cid:17) → G (Xn X )(cid:17) the set gbU mapped homeomorphically
If eX ⊂ X is a nontrivial connected
So the natural map πXn : X → Xn is a covering.
component then the map eX → Xn is a covering, hence eX is an object of ↓ S. Let G ⊂ bG
be a maximal subgroup such that GeX = eX . The subgroup G ⊂ bG is normal. If g ∈ bG\G
then geX T eX = ∅, however g is a homeomorphism, i.e. g : eX ≈−→ geX . If x ∈ X then there
is ex ∈ eX such that π0 (x) = π0 (ex), hence there is g ∈ bG such that x = gex and x ∈ geX . It
X = Gg∈J
geX
If (cid:0)Y,(cid:8)πYn(cid:9)(cid:1) is a connected covering
where J ⊂ bG is a set of representatives of bG/G.
of S then there is the natural continuous map Y → bX , because bX is the inverse limit.
Since the continuous map X → bX is bijective there is the natural map π : Y → X . Let
Let GY ⊂ G (Y X ) be such that π(cid:0)GY y(cid:1) = {x}. If bU is a special neighborhood of x
bπ0(cid:16)bU(cid:17) ⊂ X0. It follows that
i.e. bU is evenly covered by π. It turns out the map π : Y → X is continuous. From (1.2) it
turns out that there is g ∈ bG, such that x ∈ geX . The space Y is connected so it is mapped
x ∈ X be such that x ∈ π (Y ), i.e. ∃y ∈ Y which satisfies to a condition x = π (y).
then there is a connected neighborhood V of y which is mapped homeomorphically onto
π−1(cid:16)bU(cid:17) = Gg∈GY
gV,
(1.2)
(1.3)
10
object of the category ↓ S.
into geX , hence there is a continuous map eπ = g−1 ◦ π : Y → eX . The set eπ (Y ) ⊂ eX
contains a nontrivial open subset. Denote by Xeπ ⊂ eX (resp. Xeπ ⊂ eX ) maximal open
subset of eπ (Y ) (resp. minimal closed superset of eπ (Y )). The space eX is connected (i.e.
open and closed), hence from eπ (Y ) 6= eX it turns out Xeπ\ Xeπ 6= ∅. Letex ∈ Xeπ\ Xeπ, and
let eU be a special neighborhood ofex. There is y ∈ Y such that y ∈ eπ−1(cid:16)eU(cid:17) and there is a
special connected neighborhood eV ⊂ Y of y such that eπ(cid:16)eV(cid:17) is mapped homemorphically
onto πX0 (cid:16)eU(cid:17) ⊂ X0. Otherwise eU ⊂ eX is mapped homeomorphically onto πX0 (cid:16)eU(cid:17), it
follows that eV is mapped onto eU . The open set eU is such that:
• eU is a neighborhood of ex,
• eU ⊂ eπ (Y ).
From the above conditions it follows thatex lies in open subset eU ⊂ eπ (Y ), hence ex ∈ Xeπ.
This fact contradicts withex /∈ Xeπ\ Xeπ and from the contradiction it turns out eπ (Y ) = eX ,
i.e. eπ is surjective. From (1.3) it follows that eπ : Y → eX is a covering. Thus eX is the final
Definition 1.28. The final object(cid:16)eX ,nπ eXno(cid:17) of the category ↓ S is said to be the (topo-
logical) inverse limit of ↓ S. The notation(cid:16)eX ,nπ eXno(cid:17) = lim←− ↓ S or simply eX = lim←− ↓ S
Lemma 1.29. Suppose S = {X = X0 ←− ... ←− Xn ←− ...} ∈ FinTop, and bX = lim←− Xn. If X a
topological space which coincides with bX as a set and the topology on X is generated by special sets
then there is the natural isomorphism G(cid:0)X X(cid:1) ≈−→ G(cid:16)bX X(cid:17) induced by the map X → bX .
Proof. Since X coincides with bX as a set, and the topology of X is finer than the topology
of bX there is the natural injective map G(cid:0)X X(cid:1) ֒→ G(cid:16)bX X(cid:17). Ifbg ∈ G(cid:16)bX X(cid:17) and bU
where bπn : bX → Xn is the natural map, and hn : G(cid:16)bX X(cid:17) → G (Xn X ) is given by
the Lemma 1.18. Clearly hn (bg) is a homeomorphism of Xn, so from (1.4) it follows that
bπn(cid:16)bgbU(cid:17) is an open subset of Xn. HencebgbU is special. Sobg maps special sets onto special
sets. Since topology of X is generated by special sets the map bg is a homeomorphism of
X , i.e. bg ∈ G(cid:0)X X(cid:1).
will be used. The space X from the proof of the Lemma 1.27 is said to be the disconnected
inverse limit of S.
is a special set, then for any n ∈ N following condition holds
bπn(cid:16)bgbU(cid:17) = hn (bg) ◦bπn(cid:16)bU(cid:17)
(1.4)
11
1.1.2 Algebraic construction in brief
addition of special subsets. Addition of new sets to a topology is equivalent to addition
The inverse limit of coverings eX is obtained from inverse limit of topological spaces bX
by a change of a topology. The topology of eX is finer then topology of bX , it means that
C0(cid:16)bX(cid:17) is a subalgebra of Cb(cid:16)eX(cid:17). The topology of eX is obtained from topology of bX by
of new elements to C0(cid:16)bX(cid:17). To obtain Cb(cid:16)eX(cid:17) we will add to C0(cid:16)bX(cid:17) special elements
(cf. Definition 3.5). If eU ⊂ eX is a special set andea ∈ Cc(cid:16)eX(cid:17) is positive element such that
bg∈bGbgea, then following condition holds
eaeX\eU
= {0}, and a ∈ Cc (X0) is given by a = ∑
a (eπn (ex)) = ∑
From above equation it follows that
eπn (ex) /∈ eπn(cid:16)eU(cid:17) .
0
bg∈bGbgea (eπn (ex)) =( ea (ex) ex ∈ eU
bg∈bGbgea
∑
bg∈bGbgea2.
2
= ∑
The equation (1.5) is purely algebraic and related to special subsets. From the Theorem
4.13 it follows that the algebraic condition (1.5) is sufficient for construction of C0(cid:16)eX(cid:17).
Thus noncommutative inverse limits of coverings can be constructed by purely algebraic
methods.
(1.5)
1.2 Locally compact spaces
There are two equivalent definitions of C0 (X ) and both of them are used in this article.
Definition 1.30. An algebra C0 (X ) is the C∗-norm closure of the algebra Cc (X ) of com-
pactly supported continuous functions.
Definition 1.31. A C∗-algebra C0 (X ) is given by the following equation
C0 (X ) = {ϕ ∈ Cb (X ) ∀ε > 0 ∃K ⊂ X (K is compact) & ∀x ∈ X \K ϕ (x) < ε} ,
i.e.
(cid:13)(cid:13)(cid:13)ϕX \K(cid:13)(cid:13)(cid:13) < ε.
[4] For a locally compact Hausdorff space X , the following are equivalent:
Theorem 1.32.
(a) The Abelian C∗-algebra C0 (X ) is separable;
(b) X is σ-compact and metrizable;
(c) X is second-countable.
12
Corollary 1.33. If X a is locally compact second-countable Hausdorff space then for any x ∈ X
and any open neighborhood U ⊂ X there is a bounded positive continuous function a : X → R
such that a (x) 6= 0 and a (X \U ) = {0}.
[10] If φ : X → C is continuous then the support of φ is defined to
Definition 1.34.
be the closure of the set φ−1 (C\{0}) Thus if x lies outside the support, there is some
neighborhood of x on which φ vanishes. Denote by supp φ the support of φ.
1.3 Hilbert modules
We refer to [3] for definition of Hilbert C∗-modules, or simply Hilbert modules. Let A be
a C∗- algebra, and let XA be an A-Hilbert module. Let h·, ·iXA be the A-valued product on
XA. For any ξ, ζ ∈ XA let us define an A-endomorphism θξ,ζ given by θξ,ζ(η) = ξhζ, ηiXA
where η ∈ XA. The operator θξ,ζ shall be denoted by ξihζ. The norm completion of a
generated by operators θξ,ζ algebra is said to be an algebra of compact operators K(XA).
We suppose that there is a left action of K(XA) on XA which is A-linear, i.e. action of
K(XA) commutes with action of A.
1.4 C∗-algebras and von Neumann algebras
In this section I follow to [13].
[13] Let A be a C∗-algebra. The strict topology on the multiplier algebra
Definition 1.35.
M(A) is the topology generated by seminorms xa = kaxk + kxak, (a ∈ A).
[13] Let H be a Hilbert space. The strong topology on B (H) is the
Definition 1.36.
locally convex vector space topology associated with the family of seminorms of the form
x 7→ kxξk, x ∈ B(H), ξ ∈ H.
[13] Let H be a Hilbert space. The weak topology on B (H) is the locally
Definition 1.37.
convex vector space topology associated with the family of seminorms of the form x 7→
(xξ, η), x ∈ B(H), ξ, η ∈ H.
Theorem 1.38.
following conditions are equivalent:
[13] Let M be a C∗-subalgebra of B(H), containing the identity operator. The
• M = M′′ where M′′ is the bicommutant of M;
• M is weakly closed;
• M is strongly closed.
Definition 1.39. Any C∗-algebra M is said to be a von Neumann algebra or a W∗- algebra if
M satisfies to the conditions of the Theorem 1.38.
Definition 1.40.
[13] Let A be a C∗-algebra, and let S be the state space of A. For any
s ∈ S there is an associated representation πs : A → B (Hs). The representationLs∈S πs :
A → Ls∈S B (Hs) is said to be the universal representation. The universal representation
can be regarded as A → B (Ls∈S Hs).
13
[13] Let A be a C∗-algebra, and let A → B (H) be the universal represen-
Definition 1.41.
tation A → B (H). The strong closure of π (A) is said to be the enveloping von Neumann
algebra or the enveloping W∗-algebra of A. The enveloping von Neumann algebra will be
denoted by A′′.
Proposition 1.42. [13] The enveloping von Neumann algebra A′′ of a C∗-algebra A is isomorphic,
as a Banach space, to the second dual of A, i.e. A′′ ≈ A∗∗.
[13] Let Λ be an increasing net in the partial ordering. Let {xλ}λ∈Λ be an
Lemma 1.43.
increasing net of self-adjoint operators in B (H), i.e. λ ≤ µ implies xλ ≤ xµ. If kxλk ≤ γ for
some γ ∈ R and all λ then {xλ} is strongly convergent to a self-adjoint element x ∈ B (H) with
kxλk ≤ γ.
For each x ∈ B(H) we define the range projection of x (denoted by [x]) as projection on
[13] For each element x in a von Neumann algebra M there is a unique partial
the closure of xH. If M is a von Neumann algebra and x ∈ M then [x] ∈ M.
Proposition 1.44.
isometry u ∈ M and positive x ∈ M+ with uu∗ = [x] and x = xu.
Definition 1.45. The formula x = xu in the Proposition 1.44 is said to be the polar
decomposition.
1.46. Any separable C∗-algebra A has a state τ which induces a faithful GNS representa-
tion [11]. There is a C-valued product on A given by
(a, b) = τ (a∗b) .
This product induces a product on A/Iτ where Iτ = {a ∈ A τ(a∗a) = 0}. So A/Iτ is
a pre-Hilbert space. Let denote by L2 (A, τ) the Hilbert completion of A/Iτ. The Hilbert
space L2 (A, τ) is a space of a GNS representation of A.
2 Noncommutative finite-fold coverings
2.1 Basic construction
Definition 2.1. If A is a C∗- algebra then an action of a group G is said to be involutive if
ga∗ = (ga)∗ for any a ∈ A and g ∈ G. The action is said to be non-degenerated if for any
nontrivial g ∈ G there is a ∈ A such that ga 6= a.
Definition 2.2. Let A ֒→ eA be an injective *-homomorphism of unital C∗-algebras. Sup-
pose that there is a non-degenerated involutive action G × eA → eA of a finite group G, such
that A = eAG def= na ∈ eA a = ga; ∀g ∈ Go. There is an A-valued product on eA given by
and eA is an A-Hilbert module. We say that a triple(cid:16)A, eA, G(cid:17) is an unital noncommutative
finite-fold covering if eA is a finitely generated projective A-Hilbert module.
ha, bieA = ∑
(2.1)
g (a∗b)
g∈G
14
Remark 2.3. Above definition is motivated by the Theorem 1.2.
conditions hold:
Definition 2.4. Let A, eA be C∗-algebras and let A ֒→ eA be an inclusion such that following
(a) There are unital C∗-algebras B, eB and inclusions A ⊂ B, eA ⊂ eB such that A (resp. B)
is an essential ideal of eA (resp. eB) and A = BT eA,
(b) There is an unital noncommutative finite-fold covering(cid:16)B,eB, G(cid:17),
(c) GeA = eA.
The triple(cid:16)A, eA, G(cid:17) is said to be a noncommutative finite-fold covering with compactification.
The group G is said to be the covering transformation group (of(cid:16)A, eA, G(cid:17) ) and we use the
following notation
(2.2)
G(cid:16)eA A(cid:17) def= G.
Remark 2.5. The Definition 2.4 is motivated by the Lemma 4.1.
Remark 2.6. Any unital noncommutative finite-fold covering is a noncommutative finite-
fold covering with compactification.
conditions hold:
Definition 2.7. Let A, eA be C∗-algebras, A ֒→ eA an injective *-homomorphism and G ×
eA → eA an involutive non-degenerated action of a finite group G such that following
(a) A ∼= eAG def= na ∈ eA Ga = ao,
(b) There is a familyneIλ ⊂ eAoλ∈Λ
MoreoverSλ∈ΛeIλ (resp. Sλ∈Λ(cid:16)ATeIλ(cid:17) ) is a dense subset of eA (resp. A), and for
cation(cid:16)eIλT A,eIλ, G(cid:17).
any λ ∈ Λ there is a natural noncommutative finite-fold covering with compactifi-
of closed ideals of eA such that
We say that the triple(cid:16)A, eA, G(cid:17) is a noncommutative finite-fold covering.
Remark 2.8. The Definition 2.7 is motivated by the Theorem 4.3.
GeIλ =eIλ.
(2.3)
Remark 2.9. Any noncommutative finite-fold covering with compactification is a noncom-
mutative finite-fold covering.
Definition 2.10. The injective *-homomorphism A ֒→ eA from the Definition 2.7 is said to
be a noncommutative finite-fold covering.
15
Definition 2.11. Let(cid:16)A, eA, G(cid:17) be a noncommutative finite-fold covering. The algebra eA
is a Hilbert A-module with an A-valued product given by
ha, bieA = ∑
g∈G
g(a∗b); a, b ∈ eA.
2.2 Induced representation
We say that this structure of Hilbert A-module is induced by the covering(cid:16)A, eA, G(cid:17). Hence-
forth we shall consider eA as a right A-module, so we will write eAA.
2.12. Let(cid:16)A, eA, G(cid:17) be a noncommutative finite-fold covering, and let ρ : A → B (H) be a
representation. If X = eA ⊗A H is the algebraic tensor product then there is a sesquilinear
C-valued product (·, ·)X on X given by
(2.4)
(2.5)
(a ⊗ ξ, b ⊗ η)X =(cid:0)ξ, ha, bieA η(cid:1)H
(a, b ⊗ ξ) 7→ ab ⊗ ξ.
where (·, ·)H means the Hilbert space product on H, and h·, ·ieA is given by (2.4). So X is
a pre-Hilbert space. There is a natural map p : eA ×(cid:16)eA ⊗A H(cid:17) → eA ⊗A H given by
Definition 2.13. Use notation of the Definition 2.11, and 2.12. If eH is the Hilbert comple-
tion of X = eA ⊗A H then the map p : eA ×(cid:16)eA ⊗A H(cid:17) → eA ⊗A H induces the representa-
tioneρ : eA → B(cid:16)eH(cid:17). We say thateρ is induced by the pair(cid:16)ρ,(cid:16)A, eA, G(cid:17)(cid:17).
Remark 2.14. Below anyea ⊗ ξ ∈ eA ⊗A H will be regarded as element in eH.
Lemma 2.15. If A → B (H) is faithful theneρ : eA → B(cid:16)eH(cid:17) is faithful.
Proof. Ifea ∈ eA is a nonzero element then
g (ea∗eaeaea∗) ∈ A
a = heaea∗,eaea∗ieA = ∑
g∈G
is a nonzero positive element. There is ξ ∈ H such that (ξ, aξ)H
> 0. However
(ξ, aξ)H =(cid:16)eaeξ,eaeξ(cid:17)eH
whereeξ =ea∗ ⊗ ξ ∈ eA ⊗A H ⊂ eH. Henceeaeξ 6= 0.
16
g (ea ⊗ ξ) = (gea) ⊗ ξ; ea ∈ eA, g ∈ G, ξ ∈ H.
A ⊂ eA; A ⊗A H ⊂ eA ⊗A H.
(gea)ξ = g(cid:16)ea(cid:16)g−1ξ(cid:17)(cid:17) ; ∀ξ ∈ eH.
2.16. Let (cid:16)A, eA, G(cid:17) be a noncommutative finite-fold covering, let ρ : A → B (H) be a
faithful representation, and let eρ : eA → B(cid:16)eH(cid:17) is induced by the pair (cid:16)ρ,(cid:16)A, eA, G(cid:17)(cid:17).
There is the natural action of G on eH induced by the map
There is the natural orthogonal inclusion H ⊂ eH induced by inclusions
Action of g on eA can be defined by representation as gea = geag−1, i.e.
Definition 2.17. If M(cid:16)eA(cid:17) is the multiplier algebra of eA then there is the natural action of
G on M(cid:16)eA(cid:17) such that for anyea ∈ M(cid:16)eA(cid:17),eb ∈ eA and g ∈ G a following condition holds
We say that action of G on M(cid:16)eA(cid:17) is induced by the action of G on eA.
Lemma 2.18. If an action of G on M(cid:16)eA(cid:17) is induced by the action of G on eA then
Proof. If a ∈ M(cid:16)eA(cid:17)G
eAG.
and b ∈ eAG then ab ∈ eA is such that g (ab) = (ga) (gb) = ab ∈
(gea)eb = g(cid:16)ea(cid:16)g−1eb(cid:17)(cid:17)
M(cid:16)eA(cid:17)G
⊂ M(cid:16)eAG(cid:17) .
(2.6)
3 Noncommutative infinite coverings
3.1 Basic construction
This section contains a noncommutative generalization of infinite coverings.
Definition 3.1. Let
S =(cid:26)A = A0
π1−→ A1
π2−→ ...
πn−→ An
πn+1
−−→ ...(cid:27)
be a sequence of C∗-algebras and noncommutative finite-fold coverings such that:
(a) Any composition πn1 ◦ ... ◦ πn0+1 ◦ πn0 : An0 → An1 corresponds to the noncommu-
tative covering (An0, An1, G (An1 An0 ));
17
(b) If k < l < m then G (Am Ak) Al = Al (Action of G (Am Ak) on Al means that
G (Am Ak) acts on Am, so G (Am Ak) acts on Al since Al a subalgebra of Am);
(c) If k < l < m are nonegative integers then there is the natural exact sequence of
covering transformation groups
{e} → G (Am Al) ι−→ G (Am Ak) π−→ G (Al Ak) → {e}
where the existence of the homomorphism G (Am Ak) π−→ G (Al Ak) follows from
(b).
The sequence S is said to be an (algebraical) finite covering sequence. For any finite covering
sequence we will use the notation S ∈ FinAlg.
(3.1)
(ga) ξ = g(cid:16)a(cid:16)g−1ξ(cid:17)(cid:17) .
Definition 3.2. Let bA = lim−→ An be the C∗-inductive limit [11], and suppose that bG =
lim←− G (An A) is the projective limit of groups [15]. There is the natural action of bG on bA.
A non-degenerate faithful representation bA → B (H) is said to be equivariant if there is an
action of bG on H such that for any ξ ∈ H and g ∈ bG the following condition holds
Example 3.3. Let S be the state space of bA, and let bA → B (Ls∈S Hs) be the universal
representation. There is the natural action of bG on S given by
(gs) (a) = s (ga) ; s ∈ S, a ∈ bA, g ∈ bG.
The action of bG on S induces the action of bG on Ls∈S Hs. It follows that the universal
Example 3.4. Let s be a faithful state which corresponds to the representation bA → B (Hs)
andngn ∈ bGon∈N
corresponds to an equvariant representation bA → B(cid:16)Lg∈bG Hgs(cid:17).
Definition 3.5. Let π : bA → B (H) be an equivariant representation. A positive element
a ∈ B (H)+ is said to be special (with respect to π) if following conditions hold:
(a) For any n ∈ N0 the following series
= bG is a bijection. The state
representation is equivariant.
gns
2n
∑
n∈N
an =
∑
ga
is strongly convergent and the sum lies in An, i.e. an ∈ An;
g∈ker(bG→G(An A))
18
(b) If fε : R → R is given by
fε (x) =(cid:26) 0
x − ε
x ≤ ε
x > ε
(3.2)
then for any n ∈ N0 and for any z ∈ A following series
bn =
∑
g (zaz∗) ,
cn =
dn =
∑
g∈ker(bG→G(An A))
g∈ker(bG→G(An A))
g∈ker(bG→G(An A))
∑
g (zaz∗)2 ,
g fε (zaz∗)
are strongly convergent and the sums lie in An, i.e. bn, cn, dn ∈ An;
(c) For any ε > 0 there is N ∈ N (which depends on a and z) such that for any n ≥ N a
following condition holds
An element a′ ∈ B (H) is said to be weakly special if
n − cn(cid:13)(cid:13)(cid:13) < ε.
(cid:13)(cid:13)(cid:13)b2
(3.3)
a′ = xay; where x, y ∈ bA, and a ∈ B (H) is special.
Lemma 3.6. If a ∈ B (H)+ is a special element and Gn = ker(cid:16)bG → G (An A)(cid:17) then from
ga,
an = ∑
g∈Gn
it follows that a = limn→∞ an in the sense of the strong convergence. Moreover one has a =
infn∈N an.
Proof. From the Lemma 1.43 it follows that the decreasing lower-bounded sequence {an} is
strongly convergent and limn→∞ an = infn∈N an. From an > a it follows that infn∈N an ≥ a.
If infn∈N an > a then there is ξ ∈ H such that
n∈N
an(cid:19) ξ(cid:19) > (ξ, aξ) ,
(cid:18)ξ,(cid:18) inf
n∈Nξ, ∑
ga ξ = inf
g∈Gn
n∈N
(ξ, anξ) = inf
however one has
(cid:18)ξ,(cid:18) inf
n∈N
an(cid:19) ξ(cid:19) = inf
n∈N
It follows that a = infn∈N an.
(ξ, gaξ) = (ξ, aξ) .
∑
g∈Gn
19
Corollary 3.7. Any weakly special element lies in the enveloping von Neumann algebra bA′′ of
bA = lim−→ An. If Aπ ⊂ B (H) is the C∗-norm completion of an algebra generated by weakly special
elements then Aπ ⊂ bA′′.
Lemma 3.8. If a ∈ B (H) is special, (resp. a′ ∈ B (H) weakly special) then for any g ∈ bG the
element ga is special, (resp. ga′ is weakly special).
Proof. If a ∈ B (H) is special then ga satisfies to (a)-(c) of the Definition 3.5, i.e. ga is
special. If a′ is weakly special then form
it turns out that
a′ = xay; where x, y ∈ bA, and a ∈ B (H) is special,
ga′ = (gx) (ga) (gy) ,
π1
−→ A1
π2
−→ ...
πn
−→ An
πn+1
be the C∗-norm completion of algebra generated by weakly special elements. We say that
Aπ is the disconnected inverse noncommutative limit of ↓ S (with respect to π). The triple
i.e. ga′ is weakly special.
Corollary 3.9. If Aπ ⊂ B (H) is the C∗-norm completion of algebra generated by weakly special
elements, then there is a natural action of bG on Aπ.
Definition 3.10. Let S =(cid:26)A = A0
−−→ ...(cid:27) be an algebraical finite
covering sequence. Let π : bA → B (H) be an equivariant representation. Let Aπ ⊂ B (H)
(cid:16)A, Aπ, G(cid:0)Aπ A(cid:1) def= bG(cid:17) is said to be the disconnected infinite noncommutative covering of
and we will write(cid:0)A, A, G(cid:0)A A(cid:1)(cid:1).
Definition 3.11. Any maximal irreducible subalgebra eAπ ⊂ Aπ is said to be a connected
component of S (with respect to π). The maximal subgroup Gπ ⊂ G(cid:0)Aπ A(cid:1) among sub-
groups G ⊂ G(cid:0)Aπ A(cid:1) such that GeAπ = eAπ is said to be the eAπ-invariant group of S. If
Remark 3.12. From the Definition 3.11 it follows that Gπ ⊂ G(cid:0)Aπ A(cid:1) is a normal sub-
S (with respect to π). If π is the universal representation then "with respect to π" is dropped
π is the universal representation then "with respect to π" is dropped.
group.
Definition 3.13. Let
S =(cid:26)A = A0
π1
−→ A1
π2
−→ ...
πn
−→ An
πn+1
−−→ ...(cid:27) ∈ FinAlg,
and let(cid:0)A, Aπ, G(cid:0)Aπ A(cid:1)(cid:1) be a disconnected infinite noncommutative covering of S with
respect to an equivariant representation π : lim−→ An → B (H). Let eAπ ⊂ Aπ be a connected
component of S with respect to π, and let Gπ ⊂ G(cid:0)Aπ A(cid:1) be the eAπ - invariant group
of S. Let hn : G(cid:0)Aπ A(cid:1) → G (An A) be the natural surjective homomorphism. The
representation π : lim−→ An → B (H) is said to be good if it satisfies to following conditions:
20
(a) The natural *-homomorphism lim−→ An → M(cid:16)eAπ(cid:17) is injective,
(b) If J ⊂ G(cid:0)Aπ A(cid:1) is a set of representatives of G(cid:0)Aπ A(cid:1) /Gπ, then the algebraic
direct sum
Mg∈J
geAπ
is a dense subalgebra of Aπ,
(c) For any n ∈ N the restriction hnGπ is an epimorphism, i. e. hn (Gπ) = G (An A).
If π is the universal representation we say that S is good.
Definition 3.14. Let S = {A = A0 → A1 → ... → An → ...} ∈ FinAlg be an algebraical
finite covering sequence. Let π : bA → B (H) be a good representation. A connected
component eAπ ⊂ Aπ is said to be the inverse noncommutative limit of ↓ S (with respect to
π). The eAπ-invariant group Gπ is said to be the covering transformation group of S (with
respect to π). The triple(cid:16)A, eAπ, Gπ(cid:17) is said to be the infinite noncommutative covering of S
(with respect to π). We will use the following notation
def= eAπ,
lim←−π ↓ S
G(cid:16)eAπ A(cid:17) def= Gπ.
conditions hold:
lim←− ↓ S
If π is the universal representation then "with respect to π" is dropped and we will write
def= eA and G(cid:16)eA A(cid:17) def= G.
Definition 3.15. Let S = {A = A0 → A1 → ... → An → ...} ∈ FinAlg be an algebraical
(cid:16)A, eA, G(cid:17),
finite covering sequence. Let π : bA → B (H) be a good representation. Let(cid:16)A, eAπ, Gπ(cid:17)
be the infinite noncommutative covering of S ( with respect to π). Let K(cid:16)eAπ(cid:17) be the
Pedersen ideal of eAπ. We say that S allows inner product (with respect to π) if following
(a) Anyea ∈ K(cid:16)eAπ(cid:17) is weakly special,
(b) For any n ∈ N, andea,eb ∈ K(cid:16)eAπ(cid:17) the series
g∈ker(bG→G(An A))
g(cid:16)ea∗eb(cid:17)
an =
∑
is strongly convergent and an ∈ An.
21
Remark 3.16. If S allows inner product (with respect to π) then K(cid:16)eAπ(cid:17) is a pre-Hilbert
A module such that the inner product is given by
g(cid:16)ea∗eb(cid:17) ∈ A
Dea,ebE = ∑
g∈bG
keak =qkhea,eaik
a norm
where the above series is strongly convergent. The completion of K(cid:16)eAπ(cid:17) with respect to
is an A-Hilbert module. Denote by XA this completion. The ideal K(cid:16)eAπ(cid:17) is a left eAπ-
module, so XA is also eAπ-module. Sometimes we will write eAπ
Definition 3.17. Let S = {A = A0 → A1 → ... → An → ...} ∈ FinAlg and S allows inner
product (with respect to π) then then we say that given by the Remark 3.16 A-Hilbert
XA corresponds to the pair (S, π). If π is the universal representation then we
module eAπ
say that eAXA corresponds to S.
Let π : bA → B(cid:0)Hπ(cid:1) be a good representation. Let(cid:16)A, eAπ, Gπ(cid:17) be an infinite noncom-
mutative covering with respect to π of S. Denote by W π ⊂ B(cid:0)Hπ(cid:1) the bA-bimodule of
3.2 Induced representation
weakly special elements, and denote by
XA instead XA.
(3.4)
If π is the universal representation then we write eW instead eWπ.
Lemma 3.18. Ifea,eb ∈ eWπ are weakly special elements then a series
eWπ = W π\ eAπ.
∑
g∈Gπ
g(cid:16)ea∗eb(cid:17)
is strongly convergent.
Proof. From the definition of weakly special element one has
whereec is a (positive) special element and x, y ∈ bA. A series
ea∗ = xecy
gec
∑
g∈Gπ
22
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
g∈Gπ
Hence one has
i.e. the series
∑
g∈Gπ
∑
g∈Gπ
∑
< ε,
ε
.
∑
<
is strongly convergent. For any ξ ∈ Hπ and ε > 0 there is a finite subset G′ ⊂ Gπ such
that for any finite G′′ which satisfies to G′ ⊂ G′′ ⊂ Gπ following condition holds
kxk(cid:13)(cid:13)(cid:13)∑g∈Gπ gec(cid:13)(cid:13)(cid:13) kyk
g∈G′′\G′(cid:16)geb(cid:17) ξ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈G′′\G′(cid:16)g(cid:16)ea∗eb(cid:17)(cid:17) ξ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g(cid:16)ea∗eb(cid:17)
is strongly convergent and ∑g∈Gπ g(cid:16)ea∗eb(cid:17) ∈ bA′′.
Definition 3.19. Elementea ∈ eAπ is said to be square-summable if the series
g (ea∗ea)
is strongly convergent to a bounded operator. Denote by L2(cid:16)eAπ(cid:17) (or L2(cid:16)eA(cid:17) the C-space
Remark 3.20. Ifeb ∈ bA, andea ∈ L2(cid:16)eA(cid:17) then
g (ea∗ea)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g(cid:16)ebea(cid:17)∗(cid:16)ebea(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ≤(cid:13)(cid:13)(cid:13)eb(cid:13)(cid:13)(cid:13)
g (ea∗ea)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
, (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
i.e. there is the left and right action of bA on L2(cid:16)eA(cid:17).
Remark 3.21. If a, b ∈ L2(cid:16)eAπ(cid:17) then sum ∑g∈Gπ g(cid:16)ea∗eb(cid:17) ∈ bA′′ is bounded and Gπ-
invariant, hence ∑g∈Gπ g(cid:16)ea∗eb(cid:17) ∈ A′′.
Remark 3.22. From the Lemma 3.18 it turns out eWπ ⊂ L2(cid:16)eAπ(cid:17)
3.23. Let A → B (H) be a representation. Denote by eH a Hilbert completion of a pre-
g(cid:16)eaeb(cid:17)∗(cid:16)eaeb(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ≤(cid:13)(cid:13)(cid:13)eb(cid:13)(cid:13)(cid:13)
bAL2(cid:16)eAπ(cid:17) ⊂ L2(cid:16)eAπ(cid:17) , L2(cid:16)eAπ(cid:17) bA ⊂ L2(cid:16)eAπ(cid:17) ,
of square-summable operators.
2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
g∈Gπ
2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
g∈Gπ
∑
g∈Gπ
it turns out
(3.5)
(3.6)
Hilbert space
with a scalar product(cid:16)ea ⊗ ξ,eb ⊗ η(cid:17)eH
= ξ, ∑
g∈Gπ
23
L2(cid:16)eAπ(cid:17) ⊗A H,
g(cid:16)ea∗eb(cid:17)! η!
H
.
(3.7)
following representations
There is the left action of bA on L2(cid:16)eAπ(cid:17) ⊗A H given by
eb (ea ⊗ ξ) =ebea ⊗ ξ
where ea ∈ L2(cid:16)eAπ(cid:17), eb ∈ bA, ξ ∈ H. The left action of bA on L2(cid:16)eAπ(cid:17) ⊗A H induces
bρ : bA → B(cid:16)eH(cid:17) ,
eρ : eAπ → B(cid:16)eH(cid:17) .
Definition 3.24. The constructed in 3.23 representationeρ : eAπ → B(cid:16)eH(cid:17) is said to be in-
duced by (ρ, S, π). We also say thateρ is induced by(cid:16)ρ,(cid:16)A, eAπ, G(cid:16)eAπ A(cid:17)(cid:17) , π(cid:17). If π is an
universal representation we say thateρ is induced by (ρ, S) and/or(cid:16)ρ,(cid:16)A, eA, G(cid:16)eA A(cid:17)(cid:17)(cid:17).
Remark 3.26. There is an action of Gπ on eH induced by the natural action of Gπ on the
eAπ-bimodule L2(cid:16)eAπ(cid:17). If the representation eAπ → B(cid:16)eH(cid:17) is faithful then an action of Gπ
on eAπ is given by
Remark 3.25. If ρ is faithful, then ρ is faithful.
(gea) ξ = g(cid:16)ea(cid:16)g−1eξ(cid:17)(cid:17) ; ∀g ∈ G, ∀ea ∈ eAπ, ∀eξ ∈ eH.
24
(cf. Definitions 3.17 and 3.15)
3.27. If S allows inner product with respect to π then for any representation A → B (H)
XA ⊗A H is a pre-Hilbert space with the product given by
an algebraic tensor product eAπ
(a ⊗ ξ, b ⊗ η) = (ξ, ha, bi η)
Lemma 3.28. Suppose S allows inner product with respect to π and anyea ∈ K(cid:16)eAπ(cid:17) is weakly
special. If eH (resp. eH′) is a Hilbert norm completion of Wπ ⊗A H (resp. eAπ
is the natural isomorphism eH ∼= eH′.
Proof. From K(cid:16)eAπ(cid:17) ⊂ Wπ and taking into account that K(cid:16)eAπ(cid:17) is dense in eAπ
out eH′ ⊂ eH. Ifea ∈ Wπ is a positive element and fε is given by (3.2) then
(a) fε (ea) ∈ K(cid:16)eAπ(cid:17),
(b) limε→0 fε (ea) =ea.
From (a) it follows that fε (ea) ⊗ ξ ∈eAπ
ea ⊗ ξ ∈ eH′. From this fact it follows the natural inclusion eH ⊂ eH′. Mutually inverse
inclusions eH ⊂ eH′ and eH′ ⊂ eH yield the isomorphism eH ∼= eH′.
XA ⊗A H for any ξ ∈ H. From (b) it turns out
XA ⊗A H) then there
XA it turns
3.29. Let Hn be a Hilbert completion of An ⊗A H which is constructed in the section 2.2.
Clearly
L2(cid:16)eAπ(cid:17) ⊗An Hn = L2(cid:16)eAπ(cid:17) ⊗An (An ⊗A H) = L2(cid:16)eAπ(cid:17) ⊗A H.
(3.8)
4 Quantization of topological coverings
4.1 Finite-fold coverings
The following lemma supplies the quantization of coverings with compactification.
then following conditions are equivalent:
Lemma 4.1. If X , eX are locally compact spaces, and π : eX → X is a surjective continuous map,
(i) The map π : eX → X is a finite-fold covering with compactification,
(ii) There is a natural noncommutative finite-fold covering with compactification
(cid:16)C0 (X ) , C0(cid:16)eX(cid:17) , G(cid:17) .
that these algebras satisfy to the condition (a) of the Definition 2.4. From the Theorem 1.2
Proof. (i)=>(ii) Denote by X ⊂ Y, eX ⊂ eY compactifications such that π : eY → Y is a finite-
fold (topological) covering. Let G = G(cid:16)eY Y(cid:17) be a group of covering transformations. If
B = C (Y ) and eB = C(cid:16)eY(cid:17) then A = C0 (X ) (resp. eA = C0(cid:16)eX(cid:17)) is an essential ideal of
B (resp. eB). Taking into account A = C0 (X ) = C0(cid:16)eX(cid:17)T C (Y ) = BT eA one concludes
it turns out that the triple (cid:16)C (Y ) , C(cid:16)eY(cid:17) , G(cid:17) = (cid:16)B,eB, G(cid:17) is an unital noncommutative
finite-fold covering. So the condition (b) of the Definition 2.4 holds. From GeX = eX it
turns out GeA = GC0(cid:16)eX(cid:17) = C0(cid:16)eX(cid:17) = eA, i.e. the condition (c) of the Definition 2.4 holds.
(ii)=>(i) If A = C0 (X ), eA = C0(cid:16)eX(cid:17) and inclusions A ⊂ B, eA ⊂ eB are such that A (resp.
B) is an essential ideal of eA (resp. eB) then there are compactifications X ֒→ Y and eX ֒→ eY
such that B = C (Y ), eB = C(cid:16)eY(cid:17). From the condition (b) of the Definition 2.7 it turns
out that the triple(cid:16)B,eB, G(cid:17) =(cid:16)C (Y ) , C(cid:16)eY(cid:17) , G(cid:17) is an unital noncommutative finite-fold
covering. From the Theorem 1.2 it follows that the *-homomorphism C (Y ) ֒→ C(cid:16)eY(cid:17)
induces a finite-fold (topological) covering π : eY → Y. From condition (c) of of the
Definition 2.4 it turns out GC0(cid:16)eX(cid:17) = C0(cid:16)eX(cid:17) or, equivalently
GeX = eX .
From A = BT eA or, equivalently C0 (X ) = C0(cid:16)eX(cid:17)T C (Y ) and (4.1) it turns out that π is
the restriction of finite-fold covering π, i.e. π = πeX
. So π is a finite-fold covering.
(4.1)
25
Lemma 4.2. Let π : eX → X be a surjective map of topological spaces such that there is a family
of open subsets {Uλ ⊂ X }λ∈Λ such that
(a) X =Sλ∈Λ Uλ,
(b) For any λ ∈ Λ the natural map π−1 (Uλ) → Uλ is a covering.
Then the map π : eX → X is a covering.
Proof. For any point x0 ∈ X there is λ ∈ Λ such that x0 ∈ Uλ. The map π−1 (Uλ) → Uλ is
a covering, it follows that there is an open neighborhood V of x0 such that V ⊂ Uλ and V
is evenly covered by π.
the Definition 2.7.
map, then following conditions are equivalent:
Theorem 4.3. If X , eX are locally compact spaces, and π : eX → X is a surjective continuous
(i) The map π : eX → X is a finite-fold regular covering,
(ii) There is the natural noncommutative finite-fold covering(cid:16)C0 (X ) , C0(cid:16)eX(cid:17) , G(cid:17).
Proof. (i)=>(ii) We need check that(cid:16)C0 (X ) , C0(cid:16)eX(cid:17) , G(cid:17) satisfies to condition (a), (b) of
(a) Covering π is regular, so from the Proposition 1.8 it turns out X = eX /G where
G = G(cid:16)eX X(cid:17) is a covering group. From X = eX /G it follows that C0 (X ) = C0(cid:16)eX(cid:17)G
.
(b) The space X is locally compact, so for any x ∈ X there is a compact neighborhood U .
The maximal open subset U ⊂ U is an open neighborhood of x. So there is family of open
subsets {Uλ ⊂ X }λ∈Λ such that
• X =Sλ∈Λ Uλ,
• For any λ ∈ Λ the closure U λ of Uλ in X is compact.
Since π is a finite-fold covering the set π−1(cid:0)U λ(cid:1) is compact for any λ ∈ Λ. IfeIλ ⊂ C0(cid:16)eX(cid:17)
is a closed ideal given by
eIλ
def= C0(cid:16)π−1 (Uλ)(cid:17) ∼=nea ∈ C0(cid:16)eX(cid:17) ea(cid:16)eX \π−1 (Uλ)(cid:17) = {0}o
theneIλ ⊂ C(cid:0)U λ(cid:1) is an essential ideal of the unital algebra C(cid:0)π−1(cid:0)U λ(cid:1)(cid:1). From Gπ−1 (Uλ) =
π−1 (Uλ) it follows that GeIλ =eIλ. If Iλ = C0 (X )TeIλ then one has
Iλ = C0 (Uλ) ∼= {a ∈ C0 (X ) a (X \Uλ) = {0}}
hence Iλ is an essential ideal of an unital algebra C(cid:0)U λ(cid:1). The restriction map
ππ−1(U λ) : π−1(cid:0)U λ(cid:1) → U λ
26
is a finite-fold covering of compact spaces, so from the Theorem 1.2 it follows that
is an unital noncommutative finite-fold covering. It turns out that
(cid:16)C(cid:0)U λ(cid:1) , C(cid:16)π−1(cid:0)U λ(cid:1)(cid:17) , G(cid:17)
(cid:16)Iλ,eIλ, G(cid:17) =(cid:16)C0 (Uλ) , C0(cid:16)π−1 (Uλ) , G(cid:17)(cid:17)
be a family of closed ideals from the condition (b) of the
Uλ = {x ∈ X ∃a ∈ Iλ; a (x) 6= 0}
is a noncommutative finite-fold covering with compactification. From X =Sλ∈Λ Uλ (resp.
eX =Sλ∈Λ π−1 (Uλ)) it turns out thatSλ∈Λ Iλ (resp. Sλ∈ΛeIλ) is a dense subset of C0 (X )
(resp. C0(cid:16)eX(cid:17)).
(ii)=>(i) LetneIλ ⊂ C0(cid:16)eX(cid:17)oλ∈Λ
Definition 2.7, and let Iλ =eIλT C0 (X ). If eUλ ⊂ eX is a given by
def= nex ∈ eX ∃ea ∈eIλ; ea (ex) 6= 0o
eUλ
then from GeIλ =eIλ it turns out GeUλ = eUλ. If Uλ ⊂ X is given by
then Uλ = π(cid:16)eUλ(cid:17), and eUλ = π−1 (Uλ), hence there is the natural *-isomorphism
Any covering is an open map, so if U λ is the closure of Uλ in X then π−1(cid:0)U λ(cid:1) is the
closure of eUλ in eX . Following conditions hold:
• U λ (resp. π−1(cid:0)U λ(cid:1) ) is a compactification of Uλ, (resp. π−1 (Uλ)),
• Iλ = C0 (Uλ), (resp. eIλ = C0(cid:0)π−1 (Uλ)(cid:1)) is an essential ideal of C(cid:0)U λ(cid:1) (resp.
C(cid:0)π−1(cid:0)U λ(cid:1)(cid:1)),
• The triple(cid:0)C(cid:0)U λ(cid:1) , C(cid:0)π−1(cid:0)U λ(cid:1)(cid:1) , G(cid:1) is an unital noncommutative finite-fold cover-
It follows that the triple (cid:16)Iλ,eIλ, G(cid:17) = (cid:0)C0 (Uλ) , C0(cid:0)π−1 (Uλ)(cid:1) , G(cid:1) is a noncommutative
finite-fold covering with compactification, hence from the Lemma 4.1 it follows that the
natural map π−1 (Uλ) → Uλ is a covering. From (b) of the Definition 2.7 it follows that
Sλ∈Λ Iλ is dense subset of C0 (X ) it turns out X = S Uλ, hence from the Lemma 4.2 it
follows that π : eX → X is a finite-fold covering.
eIλ ∼= C0(cid:16)π−1 (Uλ)(cid:17) .
ing.
27
4.2 Infinite coverings
This section supplies a purely algebraic analog of the topological construction given by
the Subsection 1.1. Suppose that
SX = {X0 ←− ... ←− Xn ←− ...}
is a topological finite covering sequence. From the Theorem 4.3 it turns out that SC0(X ) =
{C0 (X0) → ... → C0 (Xn) → ...} is an algebraical finite covering sequence. The following
theorem and the corollary gives the construction of \C0 (X ) = lim−→ C0 (Xn).
Theorem 4.4.
A is homeomorphic to the projective limit of the state spaces Ωγ of Aγ.
[16] If a C∗-algebra A is a C∗-inductive limit of Aγ (γ ∈ Γ), the state space Ω of
[16] If a commutative C∗-algebra A is a C∗-inductive limit of the commutative
Corollary 4.5.
C∗-algebras Aγ (γ ∈ Γ), the spectrum X of A is the projective limit of spectrums Xγ of Aγ
(γ ∈ Γ).
4.6. From the Corollary 4.5 it turns out \C0 (X ) = C0(cid:16)bX(cid:17) where bX = lim←−Xn. If X is the
disconnected inverse limit of SX then there is the natural bicontinuous map f : X → bX .
The map induces the injective *-homomorphism C0(cid:16)bX(cid:17) ֒→ Cb(cid:0)X(cid:1). It follows that there is
the natural inclusion of enveloping von Neumann algebras C0(cid:16)bX(cid:17)′′ ֒→ C0(cid:0)X(cid:1)′′. Denote
by Gn = G (Xn X ) groups of covering transformations and bG = lim←− Gn. Denote by
π : X → X , πn : X → Xn, πn : Xn → X , πm
projections.
n : Xm → Xn (m > n) the natural covering
Lemma 4.7. Following conditions hold:
(i) If U ⊂ X is a compact set then there is N ∈ N such that for any n ≥ N the restriction
πnU : U ≈−→ πn(cid:0)U(cid:1) is a homeomorphism,
(ii) If a ∈ Cc(cid:0)X(cid:1)+ is a positive element then there there is N ∈ N such that for any n ≥ N
an (πn (x)) =(cid:26) a (x)
x ∈ supp a & πn (x) ∈ supp an
πn (x) /∈ supp an
following condition holds
0
(4.2)
where
an =
∑
ga.
Proof. (i) The set U is compact, hence supp a is a finite disconnected union of connected
compact sets, i.e.
g∈ker(bG→Gn)
U =
MGj=1
V j
28
It is known [15] that any covering is an open map, and any open map maps any closed
cn ≥ cm. Clearly cn ≤ M. The sequence {cn}n∈N is non-decreasing and cn ≤ M.
set onto a closed set, so for any n ∈ N the set πn(cid:0)U(cid:1) is compact. For any n ∈ N denote
by cn ∈ N the number of connected components of πn(cid:0)U(cid:1). If n > m then any connected
component of πn(cid:0)U(cid:1) is mapped into a connected component of πm(cid:0)U(cid:1), it turns out
follows that there is N ∈ N such that cN = M. For any n > N the set πn(cid:0)U(cid:1) is mapped
homeomorphically onto πN(cid:0)U(cid:1), hence from the sequence of homeomorphisms it follows
. . . ∼= πn(cid:0)U(cid:1) ∼= . . . ∼= U
it follows that πnU : U ≈−→ πn(cid:0)U(cid:1) is a homeomorphism.
(ii) The set supp a = U is compact, it follows that from (i) and a > 0 that supp a is mapped
homeomorphically onto supp aN. It turns out that if
It
an =
∑
ga
g∈ker(bG→Gn)
and n ≥ N then an is given by (4.2).
Lemma 4.8. If X is a locally compact Hausdorff space then any positive element a ∈ Cc(cid:0)X(cid:1)+ is
special.
Proof. From the Lemma 4.7 it follows that there is N ∈ N such that the equation (4.2)
holds. It turns out that for any z ∈ C0 (X ), n ≥ N and fε given by (3.2) the series
bn =
∑
g (zaz∗) ,
∑
g (zaz∗)2 ,
are given by
∑
cn =
dn =
g∈ker(bG→Gn)
g∈ker(bG→Gn)
g∈ker(bG→Gn)
bn (πn (x)) =(cid:26) z (πn (x)) a (x) z∗ (πn (x))
cn (πn (x)) =(cid:26) (z (πn (x)) a (x) z∗ (πn) (x))2
dn (πn (x)) =(cid:26) fε (z (πn (x)) a (x) z∗ (πn (x)))
0
0
0
fε (zaz∗)
x ∈ supp a & πn (x) ∈ supp an
πn (x) /∈ supp an
x ∈ supp a & πn (x) ∈ supp an
πn (x) /∈ supp an
x ∈ supp a & πn (x) ∈ supp an
πn (x) /∈ supp an
,
,
.
(4.3)
From (4.3) it turns out b2
n = cn, i.e. a satisfies to the condition (c) of the Definition 3.5.
29
Otherwise (4.2), (4.3) from that an, bn, cn, dn ∈ C0 (Xn) for any n ≥ N. If n < N then
an =
∑
gaN,
bn =
cn =
dn =
g∈G(XN Xn)
∑
g∈G(XN Xn)
∑
g∈G(XN Xn)
∑
g∈G(XN Xn)
gbN,
gcN,
gdN.
Above sums are finite, it turns out an, bn, cn, dn ∈ C0 (Xn) for any n ∈ N0, i.e. a satisfies
to conditions (a), (b) of the Definition 3.5.
Corollary 4.9. If A is a disconnected inverse noncommutative limit of
SC0(X ) = {C0 (X0) → ... → C0 (Xn) → ...}
then C0(cid:0)X(cid:1) ⊂ A.
Proof. From the Lemma 4.8 it follows that Cc(cid:0)X(cid:1) ⊂ A, and taking into account the Defi-
nition 1.30 one has C0(cid:0)X(cid:1) ⊂ A.
Lemma 4.10. Suppose that X is a locally compact Hausdorff space. Let a ∈ C0(cid:0)X (cid:1)′′
that following conditions hold:
+ be such
(a) If fε is given by (3.2) then following series
an =
∑
ga,
∑
g∈ker(bG→Gn)
g∈ker(bG→Gn)
∑
ga2,
g fε (a) ,
bn =
cn =
g∈ker(bG→Gn)
are strongly convergent and an, bn, cn ∈ C0 (Xn),
(b) For any ε > 0 there is N ∈ N such that
Then a ∈ C0(cid:0)X (cid:1)+.
n − bn(cid:13)(cid:13)(cid:13) < ε; ∀n ≥ N.
(cid:13)(cid:13)(cid:13)a2
30
Proof. The dual space C0(cid:0)X(cid:1)∗ of C0(cid:0)X(cid:1) is a space of Radon measures on X . If f : X → R
is given by
f (x) = lim
n→∞
an (πn (x)) = inf
n∈N
an (πn (x))
then from the Proposition 1.42 and the Lemma 3.6 it follows that f represents a, i.e.
following conditions hold:
• The function f defines a following functional
C0(cid:0)X(cid:1)∗ → C,
µ 7→ZX
f dµ
where µ is a Radon measure on X ,
• The functional corresponds to a ∈ C0(cid:0)X(cid:1)∗∗ = C0(cid:0)X(cid:1)′′.
If m > n then
Let M ∈ N be such that for any n ≥ M following condition holds
Let n > M, pn = πn
M : Xm → XM, and letex1,ex2 ∈ Xn be such that
(4.4)
(4.5)
(4.6)
an =
∑
gam,
bn =
g∈G(Xm Xn)
∑
g∈G(Xm Xn)
gbm.
n − bn(cid:13)(cid:13)(cid:13) < 2ε2.
(cid:13)(cid:13)(cid:13)a2
ex1 6=ex2,
pn (ex1) = pn (ex2) = x,
an (ex1) ≥ ε; an (ex2) ≥ ε.
an (ex) ,
bn (ex) .
aM (x) = ∑
n (x)
bM (x) = ∑
ex∈p−1
ex∈p−1
n (x)
n ≥ bn it turns out
an(cid:0)ex′(cid:1) an(cid:0)ex′′(cid:1) ≥
a2
∑
n (ex) +
n (x)×p−1
(ex′,ex′′)∈p−1
ex′6=ex′′
bn (ex) + an (ex1) an (ex2) + an (ex2) an (ex1) =
= bM (x) + 2an (ex1) an (ex2) .
n (x)
31
n (x)
ex∈p−1
≥ ∑
ex∈p−1
n (x)
From (4.4) it turns out
From the above equation and a2
a2
M (x) = ∑
Taking into account an (ex1) ≥ ε, an (ex2) ≥ ε one has
M − bM(cid:13)(cid:13)(cid:13) ≥ 2ε2.
(cid:13)(cid:13)(cid:13)a2
So (4.6) contradicts with (4.5), it follows that
M (x) − bM (x) ≥ 2ε2,
a2
If fε is given by (3.2) and
pn (ex1) = pn (ex2) = x & an (ex1) ≥ ε & an (ex2) ≥ ε ⇒ex1 =ex2.
cn =
∑
g fε (a)
(4.7)
then
g∈ker(bG→Gn)
supp cn =(cid:26)x ∈ Xn
am (ex) ≥ ε(cid:27) =
=nx ∈ Xn ∃ x ∈ X ; πn (x) = x & f (x) ≥ εo .
ex∈πm
inf
m>n
max
n (x)
Indeed fε (a) as a functional on C0(cid:0)X(cid:1)∗ is represented by the following function
f ε : X → R,
x 7→ fε(cid:16) f (x)(cid:17) .
From ex ∈ supp cn it turns out an (ex) ≥ ε and taking into account (4.7) one concludes that
the restriction πn
a bijection supp cn ≈−→ supp cM. The map πn
M (supp cn) = supp cM, so there is
M is a covering and it is known [15] that any
covering is an open map. Any bijective open map is a homeomorphism, hence one has a
sequence of homeomorphisms
Msupp cn is an injective map. Clearly πn
supp cM ≈←− . . . ≈←− supp cn ≈←− . . .
(4.8)
If U ⊂ X is given by
U =
∞\n=M
π−1
n (supp cn)
then from (4.8) it turns out that π M homeomorphically maps U onto supp cM. Moreover
following condition holds
fε (a) (x) =(cid:26) cM (π M (x))
0
x ∈ U
x /∈ U
.
From the Definition 1.31 it turns out D = {x ∈ XM aM (x) ≥ ε} is compact, therefore the
closed subset supp cM ⊂ D is compact, hence U = supp fε (a) ≈ supp cM is also compact. It
From the above equation it follows that fε (a) is a continuous function, i.e. fε (a) ∈ Cb(cid:0)X(cid:1)
turns out fε (a) ∈ Cc(cid:0)X(cid:1). From k fε (a) − ak ≤ ε it follows that a = limε→0 fε (a) and from
the Definition 1.30 it turns out a ∈ C0(cid:0)X(cid:1).
32
(ii) Follows from (i) and the Definitions 3.5, 3.10.
+ of SC0(X ) lies in C0(cid:0)X (cid:1), i.e. a ∈ C0(cid:0)X (cid:1),
Proof. (i) Let {eλ ∈ C0 (X )}λ∈Λ be an approximate unit of C0 (X ). From the Definition 3.5
it follows that bλ = eλaeλ satisfies to conditions of the Lemma 4.10. Otherwise from the
(i) Any special element a ∈ C0(cid:0)X (cid:1)′′
(ii) C0(cid:0)X(cid:1) ⊂ A.
Lemma 4.10 it turns out bλ ∈ C0(cid:0)X(cid:1). From the C∗-norm limit limλ∈Λ bλ = a it follows
that a ∈ C0(cid:0)X(cid:1).
4.12. Let eX ⊂ X be a connected component of X and suppose that
is the maximal among subgroups G′ such that G′ eX = eX . If J ⊂ bG is a set of representatives
of bG/G then from the (1.2) it follows that
X = Gg∈J
geX
and C0(cid:0)X(cid:1) is a C∗-norm completion of the direct sum
C0(cid:16)geX(cid:17) .
G ⊂ G(cid:16)lim←− C0 (Xn) C0 (X )(cid:17)
Corollary 4.11. If SC0(X ) = {C0 (X0) → ... → C0 (Xn) → ...} and A is a disconnected inverse
noncommutative limit of ↓ SC0(X ) then following conditions hold:
Mg∈J
(4.9)
Theorem 4.13. If SX = {X = X0 ←− ... ←− Xn ←− ...} ∈ FinTop and
SC0(X ) = {C0(X ) = C0(X0) → ... → C0(Xn) → ...} ∈ FinAlg
is an algebraical finite covering sequence then following conditions hold:
(i) SC0(X ) is good,
(ii) There are isomorphisms:
• lim←− ↓ SC0(X ) ≈ C0(cid:16)lim←− ↓ SX(cid:17);
• G(cid:16)lim←− ↓ SC0(X ) C0 (X )(cid:17) ≈ G(cid:16)lim←− ↓ SX X(cid:17).
Proof. The proof of this theorem uses a following notation:
• The topological inverse limit eX = lim←− ↓ SX ;
• The limit in the category of spaces and continuous maps bX = lim←−Xn;
33
• The disconnected covering space X of SX ;
• The disconnected covering algebra A of SC0(X );
• A connected component eA ⊂ A;
• The disconnected GX = lim←− G (Xn X ) and the connected GX = G(cid:16)eX X(cid:17) =
G(cid:16)lim←− ↓ SX X(cid:17) covering groups of SX ;
• The disconnected group GC0(X ) = lim←− G (C0 (Xn) C0 (X )) and eA-invariant group
From the Corollary 4.9 it follows that A ⊂ C0(cid:0)X(cid:1). From the Corollary 4.11 it turns out
J ⊂ GX is a set of representatives of GX /G(cid:16)eX X(cid:17)
C0(cid:0)X(cid:1) ⊂ A, hence A = C0(cid:0)X(cid:1). If
then X = Fg∈J geX is the disconnected union of connected homeomorphic spaces, i.e.
geX ≈−→ eX .
(i) We need check (a) - (c) of the Definition 3.13. A = C0(cid:0)X(cid:1) is the C∗-norm comple-
is isomorphic to C0(cid:16)eX(cid:17). The map eX → Xn is a covering for any n ∈ N, it turns out
C0 (Xn) ֒→ Cb(cid:16)eX(cid:17) = M(cid:16)C0(cid:16)eX(cid:17)(cid:17) is injective *-homomorphism. It follows that the nat-
ural *-homomorphism C0(cid:16)bX(cid:17) = limn→∞ C0 (Xn) ֒→ Cb(cid:16)eX(cid:17) = M(cid:16)C0(cid:16)eX(cid:17)(cid:17) is injective,
the condition (a) holds. The algebraic direct sum Lg∈J gC0(cid:16)eX(cid:17) is is a dense sub-
algebra of A, i.e. condition (b) holds. The homomorphism G(cid:16)eX X(cid:17) → G (Xn X ) is
surjective for any n ∈ N. From the following isomorphisms
tion of the algebraical the direct sum (4.9). Any maximal irreducible subalgebra of A
GA.
i.e.
GA ≈ G(cid:16)eX X(cid:17) ,
G (C0 (Xn) C0 (X )) ≈ G (Xn X ) ,
it turns out that GA → G (C0 (Xn) C0 (X )) is surjective, i.e. condition (c) holds.
(ii) From the proof of (i) it turns out
lim←− ↓ SX = eX ; eA = C0(cid:16)eX(cid:17) ,
lim←− ↓ SC0(X ) = eA = C0(cid:16)eX(cid:17) = C0(cid:16)lim←− ↓ SX(cid:17) ,
G(cid:16)lim←− ↓ SX X(cid:17) = GX = GA = G(cid:16)lim←− ↓ SC0(X ) C0 (X )(cid:17) .
34
5 Continuous trace C∗-algebras and their coverings
Let A be a C∗-algebra. For each positive x ∈ A+ and irreducible representation π :
A → B (H) the (canonical) trace of π(x) depends only on the equivalence class of π, so
A → [0, ∞] by x(t) = Tr(π(x)) where A is the space of
that we may define a function x :
equivalence classes of irreducible representations. From Proposition 4.4.9 [13] it follows
that x is lower semicontinuous function in the Jacobson topology.
[13] We say that element x ∈ A has continuous trace if x ∈ Cb( A). We say
Definition 5.1.
that C∗-algebra has continuous trace if a set of elements with continuous trace is dense in
A.
Definition 5.2.
is commutative.
[13] A positive element in C∗ - algebra A is Abelian if subalgebra xAx ⊂ A
[13] We say that a C∗-algebra A is of type I if each non-zero quotient of A
Definition 5.3.
contains a non-zero Abelian element. If A is even generated (as C∗-algebra) by its Abelian
elements we say that it is of type I0.
[13] A positive element x in C∗-algebra A is Abelian if dim π(x) ≤ 1 for every
Proposition 5.4.
irreducible representation π : A → B(H).
Theorem 5.5. [13] For each C∗-algebra A there is a dense hereditary ideal K(A), which is minimal
among dense ideals.
Definition 5.6. The ideal K(A) from the theorem 5.5 is said to be the Pedersen ideal of A.
Henceforth Pedersen ideal shall be denoted by K(A).
Proposition 5.7.
[13] Let A be a C∗ - algebra with continuous trace. Then
(i) A is of type I0;
A is a locally compact Hausdorff space;
(ii)
(iii) For each t ∈ A there is an Abelian element x ∈ A such that x ∈ K( A) and x(t) = 1.
The last condition is sufficient for A to have continuous trace.
Remark 5.8. From [6], Proposition 10, II.9 it follows that a continuous trace C∗-algebra A
is always a CCR-algebra, i.e. for every irreducible representation ρ : A → B(H) following
condition hold
(5.1)
ρ (A) ≈ K
[14] (Dixmier -- Douady). Any stable separable algebra A of continuous trace
Theorem 5.9.
over a second-countable locally compact Hausdorff space X is isomorphic to Γ0 (X ) , the sections
vanishing at infinity of a locally trivial bundle of algebras over X , with fibres K and structure
group Aut(K) = PU = U/T. Classes of such bundles are in natural bijection with the Cech
cohomology group H3(X , Z). The 3-cohomology class δ(A) attached to (the stabilisation of) a
continuous-trace algebra A is called its Dixmier -- Douady class.
35
Remark 5.10. Any commutative C∗-algebra has continuous trace. So described in the
Section 4 case is a special case of described in the Section 5 construction.
5.11. For any x ∈ A denote by ρx : A → B (H) a representation which corresponds to x.
For any a ∈ A denote by supp a ⊂ A the closure of the set
def= (cid:8)x ∈ A ρx (a) 6= 0(cid:9) .
supp a
5.1 Basic construction
Let A be a continuous trace C∗-algebra such that the spectrum A = X of is a second-
countable locally compact Hausdorff space. For any open subset U ⊂ X denote by
A (U ) = {a ∈ A ρx (a) = 0; ∀x ∈ X \U}
where ρx is an irreducible representation which corresponds to x ∈ X . If V ⊂ U then
there is a natural inclusion A (V ) ֒→ A (U ). Let π : eX → X be a topological covering.
Let eU ⊂ eX be a connected open subset homeomorphically mapped onto U = π(cid:16)eU(cid:17),
and suppose that the closure of eU is compact. Denote by eA(cid:16)eU(cid:17) the algebra such that
eA(cid:16)eU(cid:17) ≈ A (U ). If eV ⊂ eU and V = π(cid:16)eV(cid:17) then the inclusion A (V ) ֒→ A (U ) naturally
induces an inclusion ieV
eU
: eA(cid:16)eV(cid:17) ֒→ eA(cid:16)eU(cid:17). Let us consider eU as indexes, and let
A′ =M
(a) − ieU1T eU2
eU eA(cid:16)eU(cid:17) /I,
where L means the algebraic direct sum of C∗-algebras and I is the two sided ideal
(a), for any a ∈ A(cid:16)eU1T eU2(cid:17). There is the
generated by elements ieU1T eU2
natural C∗-norm of the direct sum on LeU eA(cid:16)eU(cid:17) and let us define the norm on A′ =
LeU eA(cid:16)eU(cid:17) /I given by
a′∈I(cid:13)(cid:13)a + a′(cid:13)(cid:13) ; ∀ (a + I) ∈M
Definition 5.12. If A(cid:16)eX(cid:17) is completion of A′ with respect to the given by (5.2) then we
say that A(cid:16)eX(cid:17) is an induced by π : eX → X covering of A.
The action of G(cid:16)eX X(cid:17) on eX induces an action of G(cid:16)eX X(cid:17) on A′, so there is a natural
action of G(cid:16)eX X(cid:17) on A(cid:16)eX(cid:17).
Definition 5.13. We say that the action of G(cid:16)eX X(cid:17) on eX induces the action of G(cid:16)eX X(cid:17)
on A(cid:16)eX(cid:17).
eU eA(cid:16)eU(cid:17) /I
ka + Ik = inf
eU1
eU2
(5.2)
36
From the Proposition 5.7 it follows that A(cid:16)eX(cid:17) is a continuous trace C∗-algebra, and the
spectrum of A(cid:16)eX(cid:17) coincides with eX . If G = G(cid:16)eX X(cid:17) is a finite group then
and the above equation induces an injective *-homomorphism A ֒→ A(cid:16)eX(cid:17).
Lemma 5.14. Let A be a continuous trace C∗-algebra, and let X = A be a spectrum of A. Suppose
that X is a second-countable locally compact Hausdorff space and B is a C∗-algebra such that
A = A(cid:16)eX(cid:17)G
(5.3)
• A ⊂ B ⊂ A′′,
• For any b ∈ B+ and x0 ∈ X such that ρx0 (b) 6= 0 there is an open neighborhood W ⊂ X
of x0 and an Abelian z ∈ A such that
supp z ⊂ W,
tr (zbz) ∈ C0 (X ) ,
tr (zbz) (x0) 6= 0.
Then B = A.
Proof. The spectrum B of B coincides with the spectrum of A as a set. Let V ⊂ X be
B. There is a closed ideal I ⊂ B which
a closed subset with respect to topology of
corresponds to V. Denote by I+ the positive part of I. For any x0 ∈ X \V there is b ∈ I+
such that ρx0 (b) 6= 0. There is an Abelian element z ∈ A such that tr (ρx0 (zbz)) 6= 0. If
W ⊂ X is an open neighborhood of x0 then from the Corollary 1.33 it follows that there is
a bounded positive continuous function a : X → R such that a (x0) 6= 0 and a (X \W) =
{0}. If z = az then z is an Abelian document, tr (zbz) (x0) = (tr (zbz) (x0)) a2 (x0) 6= 0
and supp z ⊂ W. From tr (zbz) ∈ C0 (X ) it turns out that there is an open (with respect
A) neighborhood U of x0 such that tr (ρx (zbz)) 6= 0 for any x ∈ U , i.e.
to topology of
VT U = ∅. It follows that V is a closed subset with respect to the topology of
A. Hence
there is a homeomorhism A ≈ B. Below we apply the method of proof of the Theorem
6.1.11 [13]. Let us consider the set M of elements in B+ with continuous trace, M is
hereditary and the closure of M is the positive part of an ideal J of B. However for any
x ∈ X = B there is an Abelian a ∈ K (A) such that tr (ρx (a)) 6= 0. It turns out J is not
contained in any primitive ideal of B, hence J = B. It turns out B has continuous trace.
From this fact it turns out ρx (A) = ρx (B) ≈ K for any x ∈ X . Taking into account
ρx (A) = ρx (B), homeomorphism A ≈ B and the Theorem 5.9 one has B = A.
5.2 Finite-fold coverings
If π : eX → X is a finite-fold covering, such that X and eX are compact Hausdorff spaces,
of connected open subsets of X evenly covered
then there is a finite family {Uι ⊂ X }ι∈I0
37
by π such that X =Sι∈I0 Uι. There is a partition of unity subordinated to {Uι}, i.e.
1C(X ) = ∑
ι∈I0
aι
by
where aι ∈ C (X )+ is such that supp aι ⊂ Uι. Denote by eι = √aι ∈ C (X )+. For any ι ∈ I0
we select eUι ⊂ eX such that eUι is homemorphically mapped onto Uι. Ifeeι ∈ C(cid:16)eX(cid:17) is given
and G = G(cid:16)eX X(cid:17) then
eeι (ex) =( eι (π (ex)) ex ∈ eUι
ex /∈ eUι
1C(eX ) = ∑
0
gee2
eeι (geeι) = 0; for any nontrivial g ∈ G.
(g,ι)∈G×I0
ι ,
(5.4)
by (2.1).
ι∈Ieeι iheeι
1C(eX ) = ∑
[1] If B is a C∗-subalgebra of A containing an approximate unit for A, then
If I = G × I0 andee(g,ι) = geeι the from the above equation it turns out
whereeeι iheeι means a compact operator induced by the C∗-Hilbert module structure given
Proposition 5.15.
M (B) ⊂ M (A) (regarding B′′ as a subalgebra of A′′).
Lemma 5.16. Let A be a continuous trace algebra, and let A = X be the spectrum of A. Suppose
that X is a locally compact second-countable Hausdorff space, and let π : eX → X be a finite-fold
covering. There is the natural *-isomorphism M (A) ∼= M(cid:16)A(cid:16)eX(cid:17)(cid:17)G
Proof. For any x ∈ X there is an open neighborhood U such that A (U ) ∼= C0 (U ) ⊗ K.
Since X is second-countable there is an enumerable family {Uk}k∈N such that A (Uk) ∼=
C0 (Uk) ⊗ K for any k ∈ N. There is a family(cid:8)ak ∈ C0 (X )+(cid:9)k∈N such that
of multiplier algebras.
• supp ak ⊂ Uk,
•
1Cb(X ) =
∞
∑
k=0
ak
(5.5)
where sum of the series means the strict convergence (cf. Definition 1.35).
There is an enumerable family {ek ∈ K}k∈N of rank-one positive mutually orthogonal
operators such that
ek
(5.6)
∞
∑
k=0
1M(K) =
38
where above sum assumes strict topology (cf. Definition 1.35). The family of products
is enumerable and let us introduce an enumeration ofnujkoj,k∈N
, i.e.
=(cid:8)up(cid:9)p∈N. From (5.5) and (5.6) it follows that
nujk = aj ⊗ ekoj,k∈N
nujkoj,k∈N
1M(A) =
ujk =
up.
(5.7)
∞
∑
j=1
k=1
∞
∑
p=0
If h ∈ A is given by
h =
∞
∑
p=0
1
2p up
and τ : A → C is a state such that τ (h) = 0 then from up > 0 for any p ∈ N it follows
that τ(cid:0)up(cid:1) = 0 for any p ∈ N. However from (5.7) it turns out
τ(cid:0)up(cid:1) ,
1 = τ(cid:16)1M(A)(cid:17) = τ ∞
up! =
∞
∑
p=0
∑
p=0
for any state τ, i.e. h is strictly positive element of A. Similarly one can prove that h is
and above equation contradicts with τ(cid:0)up(cid:1) = 0 for any p ∈ N. It follows that τ (h) 6= 0
strictly positive element of A(cid:16)eX(cid:17) because
1M(A(eX )) =
∞
∑
p=0
up.
From the Proposition 5.15 it follows that there is the natural injective *-homomorphism
f : M (A) ֒→ M(cid:16)A(cid:16)eX(cid:17)(cid:17). Clearly g f (a) = f (ga) = f (a) for any a ∈ A and g ∈ G, it
follows that f (M (A)) ⊂ M(cid:16)eA(cid:17)G
Lemma 2.18 one has M(cid:16)eA(cid:17)G
M(cid:16)eA(cid:17)G
, or equivalently M (A) ⊂ M(cid:16)eA(cid:17)G
⊂ M (A) and M (A) ⊂ M(cid:16)eA(cid:17)G
⊂ M (A). Taking into account mutually inverse inclusions
. Otherwise from the
we conclude that
M (A) ∼= M(cid:16)A(cid:16)eX(cid:17)(cid:17)G
.
Lemma 5.17. Let A be a continuous trace algebra, and let A = X be the spectrum of A. Suppose
that X is a locally compact second-countable Hausdorff space, and let π : eX → X be a finite-fold
covering with compactification. Then the triple (cid:16)A, A(cid:16)eX(cid:17) , G = G(cid:16)eX X(cid:17)(cid:17) is a finite-fold
noncommutative covering with compactification.
39
Proof. We need check conditions (a) - (c) of the Definition 2.4.
(a) There is the action of G on A(cid:16)eX(cid:17) induced by the action of G on eX . From (5.3) it turns
out that A = A(cid:16)eX(cid:17)G
and there is an injective *-homomorphism A ֒→ A(cid:16)eX(cid:17). Denote
by M (A) and M(cid:16)A(cid:16)eX(cid:17)(cid:17) multiplier algebras of A and A(cid:16)eX(cid:17). Denote by X ֒→ Y,
eX ֒→ eY compactifications such that eπ : eY → Y is a (topological) finite covering and
. From Cb (X ) ⊂ M (A), Cb(cid:16)eX(cid:17) ⊂ M(cid:16)A(cid:16)eX(cid:17)(cid:17) and C (Y ) ⊂ Cb (X ), C(cid:16)eY(cid:17) ⊂
π = eπeX
Cb(cid:16)eX(cid:17) it follows that C (Y ) ⊂ M (A), C(cid:16)eY(cid:17) ⊂ M(cid:16)A(cid:16)eX(cid:17)(cid:17). If B = C (Y ) M (A) and
eB = C(cid:16)eY(cid:17) M(cid:16)A(cid:16)eX(cid:17)(cid:17) then A (resp. A(cid:16)eX(cid:17)) is an essential ideal of B (resp. eB). Clearly
A = BT A(cid:16)eX(cid:17).
(b) Since GeY = eY the action G × M(cid:16)A(cid:16)eX(cid:17)(cid:17) → M(cid:16)A(cid:16)eX(cid:17)(cid:17) induces an action G ×eB →
eB. From the Lemma 5.16 on has the natural *-isomorphism M(cid:16)A(cid:16)eX(cid:17)(cid:17)G ∼= M (A). It
follows that B = C (Y ) M (A) ∼= C (Y ) M(cid:16)A(cid:16)eX(cid:17)(cid:17)G
= eBG. From (5.4) it turns out that
there is a finite family {eι ∈ C (Y )}ι∈I such that
1C(eY) = 1eB = ∑
ι∈Ieeι iheeι.
eb = ∑
ι∈Ieeιbι ,
bι =Deb,eeιEeB ∈ B,
It turns out that anyeb ∈ eB is given by
fold noncommutative covering.
i.e. eB is a finitely generated (by {eι}ι∈I ) right B-module. From the Kasparov Stabilization
Theorem [3] it turns out that eB is a projective B module. So(cid:16)B,eB, G(cid:17) is an unital finite-
(c) Follows from GeX = eX .
Theorem 5.18. Let A be a continuous trace algebra, and let A = X be the spectrum of A. Suppose
that X is a locally compact second-countable Hausdorff space, and let π : eX → X be a finite-
fold covering. Then the triple (cid:16)A, A(cid:16)eX(cid:17) , G = G(cid:16)eX X(cid:17)(cid:17) is a finite-fold noncommutative
(a) Follows from X = eX /G.
(b) Let us consider a family {Uλ ⊂ X }λ∈Λ of open sets such that
• X =Sλ∈Λ Uλ,
• The closure U λ of Uλ in X is compact ∀λ ∈ Λ.
Proof. We need check (a), (b) of the Definition 2.7.
covering.
40
Clearly π−1(cid:0)U λ(cid:1) → U λ is a covering, so π−1 (Uλ) → Uλ is a covering with compactifica-
def= A(cid:0)π−1 (Uλ)(cid:1) ⊂ A(cid:16)eX(cid:17) and Iλ =eIλT A then from Gπ−1 (Uλ) = π−1 (Uλ) it
tion. IfeIλ
follows that
GeIλ =eIλ,
Iλ = A (Uλ) ,
i.e. eIλ satisfies to (2.3). From the Lemma 5.17 it follows that there is a finite-fold non-
commutative covering with compactification (cid:16)Iλ,eIλ, G(cid:17) = (cid:0)A (Uλ) , A(cid:0)π−1 (Uλ)(cid:1) , G(cid:1).
From the Definition 5.12 and X = Sλ∈Λ Uλ it follows that Sλ∈Λ Iλ = Sλ∈Λ A (Uλ) (resp.
Sλ∈ΛeIλ =Sλ∈Λ A(cid:0)π−1 (Uλ)(cid:1)) is dense in A (resp. A(cid:16)eX(cid:17).
5.3 Infinite coverings
Let A be a continuous trace C∗-algebra such that the spectrum A = X of is a second-
countable locally compact Hausdorff space. Suppose that
is a topological finite covering sequence. From the Theorem 5.18 it turns out that
SX = {X = X0 ←− ... ←− Xn ←− ...}
SA(X ) = {A = A (X0) → ... → A (Xn) → ...}
algebras.
Lemma 5.19. If SX = {X = X0 ←− ... ←− Xn ←− ...} ∈ FinTop and X is disconnected inverse
is an algebraical finite covering sequence. If bA = lim−→ A (Xn) then from the Theorem 4.4 it
follows that there is the spectrum of bA is homeomorphic to bX = lim←−Xn.
limit of SX , then there is the natural inclusion of bA′′ → A(cid:0)X(cid:1)′′ of von Neumann enveloping
Proof. Surjective maps X → Xn give injective *-homomorphisms A (Xn) ֒→ M(cid:0)A(cid:0)X(cid:1)(cid:1),
which induce the injective *-homomorphism bA ֒→ M(cid:0)A(cid:0)X(cid:1)(cid:1). It turns out the injective
*-homomorphism of von Neumann enveloping algebras bA′′ → A(cid:0)X(cid:1)′′.
5.20. Denote by Gn = G (Xn X ) groups of covering transformations and bG = lim←− Gn.
Denote by πn : X → Xn, πn : Xn → X πm
projections.
positive element in a ∈ A(cid:0)U(cid:1)+ ⊂ A(cid:0)X(cid:1)+ is special.
n : Xm → Xn (m > n) the natural covering
Lemma 5.21. If U ⊂ X is an open subset mapped homeomorphically onto U ⊂ X then any
41
Proof. If Un = πn(cid:0)U(cid:1) then there is a *-isomorphism ϕn : A(cid:0)U(cid:1) ≈−→ A (Un). For any
n ∈ N0 and z ∈ A and fε given by (3.2)
an =
∑
ga = ϕn (a) ,
g∈ker(bG→Gn)
g (zaz∗) = zϕn (a) z∗,
bn =
∑
g∈ker(bG→Gn)
∑
g (zaz∗)2 = (zϕn (a) z∗)2 ,
g fε (zaz∗) = fε (zϕn (a) z∗) .
cn =
dn =
∑
g∈ker(bG→Gn)
g∈ker(bG→Gn)
From the above equations it follows that an, bn, cn, dn ∈ A (Xn) and b2
satisfies to the Definition 3.5.
n = cn, i.e. a
A.
Corollary 5.22. If A is the disconnected inverse noncommutative limit of SA(X ), then A(cid:0)X(cid:1) ⊂
Proof. From the Lemma 5.21 it turns out A(cid:0)U(cid:1) ⊂ A. However A(cid:0)X(cid:1) is the C∗-norm
completion of its subalgebras A(cid:0)U(cid:1) ⊂ A(cid:0)X(cid:1).
Lemma 5.23. If a ∈ A(cid:0)X(cid:1)′′ is a special element and z ∈ K (A) is an Abelian element then
b = tr (zaz) ∈ C0(cid:0)X(cid:1).
Proof. Any Abelian element is positive, hence z = z∗. If fε is given by (3.2) and b′ = zaz
then from (b) of the Definition 3.5 it turns out
b′n =
∑
∑
g∈ker(bG→Gn)
g∈ker(bG→Gn)
∑
c′n =
d′n =
gb′ ∈ A (Xn) ,
gb′2
∈ A (Xn) ,
g fε(cid:16)b′(cid:17) ∈ A (Xn) .
g∈ker(bG→Gn)
From z ∈ K (A) it turns out that supp tr (z) is compact. The map πn : Xn → X is a
n (supp tr (z)) is compact. If supp b′n, supp c′n, supp d′n ⊂
finite-fold covering, it turns out π−1
n (supp tr (z)) it turns out that all sets supp b′n, supp c′n, supp d′n are compact. Taking into
π−1
account that all b′n, c′n, d′n are Abelian one has b′n, c′n, d′n ∈ K (A (Xn)) where K (A (Xn))
42
means the Pedersen ideal of A (Xn). It follows that
bn = tr(cid:0)b′n(cid:1) =
tr(cid:16)gb′2(cid:17) =
g∈ker(bG→Gn)
tr(cid:16)g fε(cid:16)b′(cid:17)(cid:17) =
∑
∑
g∈ker(bG→Gn)
n = tr(cid:16)b′2
g∈ker(bG→Gn)
∑
tr(cid:16)gb′)(cid:17) ∈ Cc (Xn) ,
n(cid:17) = tr(cid:0)b′n(cid:1)2 ∈ Cc (Xn) ,
tr(cid:16)gb′(cid:17)2
fε(cid:16)tr(cid:16)gb′(cid:17)(cid:17) ∈ Cc (Xn) .
∈ Cc (Xn) ,
∑
b2
g∈ker(bG→Gn)
cn = tr(cid:0)c′n(cid:1) =
dn = tr(cid:0)d′n(cid:1) =
∑
g∈ker(bG→Gn)
However from
tr (zaz) ∈ C0(cid:0)X(cid:1) .
ρx (z a z) (a) =(cid:26) ρx (zaz)
0
x ∈ W
x /∈ W
43
(cid:13)(cid:13)(cid:13)b′2
n − c′n(cid:13)(cid:13)(cid:13) < ε.
From the above equations it follows that bn, cn, dn satisfy to the condition (a) of the
Lemma 4.10. From the condition (c) the Definition 3.5 it follows that for any for any ε > 0
there is N ∈ N such that for any n ≥ N following condition holds
(5.8)
Both b′n and c′n are Abelian and the range projection of b′n equals to the range projection
of c′n, i.e. [b′n] = [c′n], it turns out
n − c′n(cid:13)(cid:13)(cid:13) .
condition (b) of the Lemma 4.10. From the Lemma 4.10 it turns out that b = tr (zaz) ∈
Lemma 5.24. If A is the disconnected inverse noncommutative limit of ↓ SA(X ), then A =
n − cn(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13)tr(cid:0)b′(cid:1)2 − tr(cid:0)c′n(cid:1)(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13)b′2
(cid:13)(cid:13)(cid:13)b2
n − cn(cid:13)(cid:13) < ε for any n ≥ N. It means that bn and cn satisfy to
From (5.8) it follows that(cid:13)(cid:13)b2
C0(cid:0)X(cid:1).
A(cid:0)X(cid:1).
Proof. From the Corollary 5.22 it follows that A(cid:0)X(cid:1) ⊂ A. From the Corollary 3.7 it follows
Let π : X → X and let a ∈ A(cid:0)X(cid:1)′′
an open neighborhood of x such that π homeomorphically maps W onto W = π(cid:0)W(cid:1). If
z ∈ K(cid:0)A(cid:0)X(cid:1)(cid:1) is an Abelian element such that supp z ⊂ W and ρx (z a z) 6= 0 then the
g∈bG gz ∈ A is Abelian and supp z ∈ W. If a is special, then from the Lemma
+. Let x ∈ X be such that ρx (a) 6= 0 and let W be
A(cid:0)X(cid:1) ⊂ A ⊂ A(cid:0)X(cid:1)′′ .
element z = ∑
5.23 it turns out that
that
(5.9)
it turns out
tr (z a z) ∈ C0(cid:0)X(cid:1) ,
tr (z a z) (x) 6= 0.
(5.10)
The set of special elements is dense in A+ it turns out that any a ∈ A+ satisfies to (5.10).
Taking into account this fact and (5.9) it turns out
• For any a ∈ A+ and x ∈ X such that ρx (a) 6= 0 there is an open neighborhood
• A(cid:0)X(cid:1) ⊂ A ⊂ A(cid:0)X(cid:1)′′,
W ⊂ X of x and an Abelian z ∈ A(cid:0)X(cid:1) such that
tr (z a z) ∈ C0(cid:0)X(cid:1) ,
supp z ⊂ W ,
tr (z a z) (x) 6= 0.
From the Lemma 5.14 it follows that A = A(cid:0)X(cid:1).
5.25. Let eX ⊂ X be a connected component and let G ⊂ G(cid:16)lim←− C0 (Xn) C0 (X )(cid:17) be
maximal subgroup among subgroups G′ ⊂ G(cid:16)lim←− C0 (Xn) C0 (X )(cid:17) such that G′ eX = eX .
If J ⊂ bG is a set of representatives of bG/G then from the (1.2) it follows that
and the algebraic direct sum
X = Gg∈J
geX .
A(cid:16)geX(cid:17) ⊂ A(cid:0)X(cid:1) .
Mg∈J
is a dense subalgebra of A(cid:0)X(cid:1).
Theorem 5.26. Let A be C∗-algebra of continuous trace, and let X be the spectrum of A. Let
(5.11)
be a topological finite covering sequence, and let
SX = {X = X0 ←− ... ←− Xn ←− ...} ∈ FinTop
SA(X ) = {A = A (X0) → ... → A (Xn) → ...} ∈ FinAlg
be an algebraical finite covering sequence. Following conditions hold:
(i) SA(X ) is good,
(ii) There are isomorphisms:
• lim←− ↓ SA(X ) ≈ A(cid:16)lim←− ↓ SX(cid:17),
• G(cid:16)lim←− ↓ SA(X ) A(cid:17) ≈ G(cid:16)lim←− ↓ SX X(cid:17).
Proof. Similar to the proof ot the Theorem 4.13.
44
6 Noncommutative tori and their coverings
6.1 Fourier transformation
There is a norm on Zn given by
The space of complex-valued Schwartz functions on Zn is given by
k(k1, ..., kn)k =qk2
1 + ... + k2
n.
(6.1)
S (Zn) =na = {ak}k∈Zn ∈ CZn
supk∈Zn (1 + kkk)s ak < ∞, ∀s ∈ No .
given by
Let Tn be an ordinary n-torus. We will often use real coordinates for Tn, that is, view Tn as
Rn/Zn. Let C∞ (Tn) be an algebra of infinitely differentiable complex-valued functions
on Tn. There is the bijective Fourier transformations FT : C∞ (Tn) ≈−→ S (Zn);
7→ bf
(6.2)
where dx is induced by the Lebesgue measure on Rn and · is the scalar product on the
Euclidean space Rn. The Fourier transformation carries multiplication to convolution, i.e.
f
e−2πix·p f (x) dx
bf (p) = FT( f )(p) =ZTn
r+s=pbf (r)bg (s) .
cf g (p) = ∑
T : S (Zn) ≈−→ C∞ (Tn); bf 7→ f is given by
T bf (x) = ∑
( f , g) =ZTn
f gdx = ∑
p∈Znbf (p) e2πix·p.
p∈Znbf (−p)bg (p) .
The inverse Fourier transformation F−1
f (x) = F−1
There is the C-valued scalar product on C∞ (Tn) given by
Denote by S (Rn) be the space of complex Schwartz (smooth, rapidly decreasing) func-
tions on Rn.
S (Rn) =n f ∈ C∞(Rn) : k fkα,β)
+o ,
< ∞ ∀α = (α1, ...αn) , β = (β1, ...βn) ∈ Zn
x∈Rn(cid:12)(cid:12)(cid:12)xαDβ f (x)(cid:12)(cid:12)(cid:12)
k fkα,β = sup
where
(6.3)
xα = xα1
1 · ... · xαn
n ,
∂
∂
∂xβn
∂xβ1
n
1
...
.
Dβ =
The topology on S (Rn) is given by seminorms k · kα,β.
45
Definition 6.1. Denote by S′ (Rn) the vector space dual to S (Rn), i.e. the space of contin-
uous functionals on S (Rn). Denote by h·, ·i : S′ (Rn) × S (Rn) → C the natural pairing.
We say that {an ∈ S′ (Rn)}n∈N is weakly-* convergent to a ∈ S′ (Rn) if for any b ∈ S (Rn)
following condition holds
We say that
lim
n→∞han, bi = ha, bi .
a = lim
n→∞
an
in the sense of weak-* convergence.
Let F and F−1 be the ordinary and inverse Fourier transformations given by
f (t)e2πit·udt
f (t)e−2πit·udt, (cid:16)F−1 f(cid:17) (u) =ZR2N
(F f ) (u) =ZR2N
which satisfy following conditions
F ◦ F−1S (Rn) = F−1 ◦ FS (Rn) = IdS (Rn).
There is the C-valued scalar product on S (Rn) given by
( f , g)L2(Rn) =ZRn
f gdx =ZRn F fF gdx.
which if F -invariant, i.e.
( f , g)L2(Rn) = (F f , F g)L2(Rn) .
There is the action of Zn on Rn such that
and Tn ≈ Rn/Zn. For any x ∈ Rn and C ∈ R the series
gx = x + g; x ∈ Rn, g ∈ Zn
∑
k∈Zn
C
1 + x + kn+1
(6.4)
(6.5)
(6.6)
is convergent, and taking into account (6.3) one concludes that for f ∈ S (Rn) and x ∈ Rn
the series
Dβ f (x + g) (x) = ∑
∑
g∈Zn
g∈Zn(cid:16)gDβ f(cid:17) (x)
is absolutely convergent. It follows that the series
g f
eh = ∑
g∈Zn
46
is point-wise convergent andeh is a smooth Zn - invariant function. The periodic smooth
functioneh corresponds to an element of h ∈ C∞ (Tn). This construction provides a map
(6.7)
S (Rn) → C∞ (Tn) ,
g f .
f 7→ h = ∑
g∈Zn
represented by the Fourier series
If U = (0, 1)n ⊂ Rn is a fundamental domain of the action of Zn on Rn thenehU can be
cp =ZUeh (x) e−2πipx dx = ∑
g∈ZnZU
where bf = F f is the Fourier transformation of f . So ifbh = FTh is the Fourier transfor-
mation of h then for any p ∈ Zn a following condition holds
ehU (x) = ∑
f (x) e−2πipx dx = bf (p)
f (x + g) e−2πipx dx =ZRn
cpe2πipx,
p∈Zn
6.2 Noncommutative torus Tn
Θ
bh (p) = bf (p) .
(6.8)
(6.9)
(6.10)
(6.11)
(6.12)
Denote by · : Rn × Rn → R the scalar product on the Euclidean vector space Rn. Let
Θ be a real skew-symmetric n × n matrix, we will define a new noncommutative product
⋆Θ on S (Zn) given by
and an involution
(cid:16)bf ⋆Θbg(cid:17) (p) = ∑
In result there is an involutive algebra C∞(cid:0)Tn
state on C∞(cid:0)Tn
From C∞(cid:0)Tn
r+s=pbf (r)bg (s) e−πir · Θs.
bf ∗ (p) = bf (−p)).
τ ( f ) = bf (0) .
Θ(cid:1) given by
Θ(cid:1) ≈ S (Zn) it follows that there is a C-linear isomorphism
ϕ∞ : C∞ (Tn
Θ) ≈−→ C∞ (Tn) .
Θ(cid:1) = (S (Zn) , +, ⋆Θ ,∗ ). There is a tracial
such that following condition holds
τ ( f ) =
1
(2π)n ZTn
ϕ∞ ( f ) dx.
47
Θ(cid:1) → B(cid:0)L2(cid:0)C∞(cid:0)Tn
Θ(cid:1) of C∞(cid:0)Tn
Θ(cid:1) is a C∗-algebra and there is a faithful representation
Similarly to 1.46 there is the Hilbert space L2(cid:0)C∞(cid:0)Tn
C∞(cid:0)Tn
C(cid:0)Tn
Θ) → B(cid:16)L2 (C∞ (Tn
Θ(cid:1) , τ(cid:1) instead of L2(cid:0)C∞(cid:0)Tn
We will write L2(cid:0)C(cid:0)Tn
map C∞(cid:0)Tn
Θ(cid:1) → L2(cid:0)C(cid:0)Tn
Θ(cid:1) , τ(cid:1) and since C∞(cid:0)Tn
S (Zn) → L2(cid:0)C(cid:0)Tn
Uk (p) = δkp : ∀p ∈ Zn
Θ(cid:1) , τ(cid:1) and the natural representation
Θ(cid:1) , τ(cid:1)(cid:1) which induces the C∗-norm. The C∗-norm completion
Θ) , τ)(cid:17) .
Θ(cid:1) , τ(cid:1). There is the natural C-linear
Θ(cid:1) ≈ S (Zn) there is a linear map ΨΘ :
Θ(cid:1) , τ(cid:1). If k ∈ Zn and Uk ∈ S (Zn) = C∞(cid:0)Tn
Θ(cid:1) is such that
C (Tn
(6.13)
(6.14)
then
UkUp = e−πik · ΘpUk+p; UkUp = e−2πik · ΘpUpUk.
If ξk = ΨΘ (Uk) then from (6.9), (6.10) it turns out
τ (U∗k
⋆Θ Ul) = (ξk, ξl) = δkl,
by
i.e. the subset {ξk}k∈Zn ⊂ L2(cid:0)C(cid:0)Tn
the Hilbert space L2(cid:0)C(cid:0)Tn
Θ(cid:1) , τ(cid:1). Hence
Θ(cid:1) , τ(cid:1) is naturally isomorphic to the Hilbert space ℓ2 (Zn) given
Θ(cid:1) , τ(cid:1) is an orthogonal basis of L2(cid:0)C(cid:0)Tn
k∈Zn ξk2 < ∞)
ℓ2 (Zn) =(ξ = {ξk ∈ C}k∈Zn ∈ CZn
∑
and the C-valued scalar product on ℓ2 (Zn) is given by
(6.15)
(6.16)
(6.17)
(ξ, η)ℓ2(Zn) = ∑
k∈Zn
ξkηk.
An alternative description of C(cid:0)Tn
Θ(cid:1) is such that if
Θ =
0
θ21
...
θn1
θ12
0
...
θn2
. . .
. . .
. . .
. . .
θ1n
θ2n
...
0
then C(cid:0)Tn
Θ(cid:1) is the universal C∗-algebra generated by unitary elements u1, ..., un ∈ U(cid:0)C(cid:0)Tn
Θ(cid:1)(cid:1)
such that following condition holds
Elements uj are given by
ujuk = e−2πiθjkukuj.
(6.18)
j =0, . . . ,
uj = Uj,
, . . . , 0 .
1{z}
jth−place
48
Definition 6.2. Unitary elements u1, . . . , un ∈ U(cid:0)C(cid:0)Tn
are said to be generators of C(cid:0)Tn
If a ∈ C(cid:0)Tn
Θ(cid:1) is presented by a series
Θ(cid:1). The set {Ul}l∈Zn is said to be the basis of C(cid:0)Tn
Θ(cid:1).
θ(cid:1)(cid:1) which satisfy the relation (6.18)
a = ∑
l∈Zn
clUl; cl ∈ C
and the series ∑l∈Zn cl is convergent then from the triangle inequality it follows that
kak ≤ ∑
l∈Zn cl .
(6.19)
Definition 6.3. If Θ is non-degenerated, that is to say, σ(s, t) def= s · Θt to be symplectic. This
implies even dimension, n = 2N. Then one selects
Θ = θ J
1N
Θ(cid:1) be a
6.3 Finite-fold coverings
def= θ(cid:18) 0
−1N
Θ (cid:1) and C(cid:0)T2N
θ (cid:1) def=
0(cid:19)
where θ > 0 is defined by θ2N def= det Θ. Denote by C∞(cid:0)T2N
θ (cid:1) def= C∞(cid:0)T2N
C(cid:0)T2N
Θ (cid:1).
In this section we write ab instead a ⋆Θ b. Let Θ be given by (6.17), and let C(cid:0)Tn
noncommutative torus. If (k1, . . . , kn) ∈ Nn and
eθ12
eθ21
eθn1 eθn2
e−2πiθrs = e−2πieθrskrks
Θ(cid:1) → C(cid:16)Tn
Θ(cid:1) (resp. v1, ..., vn ∈ C(cid:16)Tn
eΘ(cid:17) given by
eΘ(cid:17)) are unitary generators of C(cid:0)Tn
Θ(cid:1)
eΘ(cid:17) given by
eΘ(cid:17)). There is an involutive action of G = Zk1 × ... × Zkn on C(cid:16)Tn
then there is a *-homomorphism C(cid:0)Tn
where u1, ..., un ∈ C(cid:0)Tn
(resp. C(cid:16)Tn
eΘ =
is a skew-symmetric matrix such that
. . . eθ1n
. . . eθ2n
. . .
. . .
...
0
uj 7→ v
kj
j ; j = 1, ..., n
0
...
0
...
(6.20)
(6.21)
2πi pj
kj vj,
(p1, ..., pn) vj = e
49
. Otherwise there is a following
and a following condition holds C(cid:0)Tn
C(cid:0)Tn
Θ(cid:1) - module isomorphism
Θ(cid:1) = C(cid:16)Tn
eΘ(cid:17)G
vp1
1 · ... · vpn
M
C(cid:16)Tn
eΘ(cid:17) =
eΘ(cid:17) is a finitely generated projective Hilbert C(cid:0)Tn
(p1,...pn)∈Zk1×...×Zkn
n C (Tn
Θ) ≈ C (Tn
Θ(cid:1)-module.
Θ)k1·...·kn
following theorem.
i.e. C(cid:16)Tn
Theorem 6.4. The triple (cid:16)C(cid:0)Tn
finite-fold covering.
Θ(cid:1) , C(cid:16)Tn
eΘ(cid:17) , Zk1 × ... × Zkn(cid:17) is an unital noncommutative
It turns out the
6.4 Moyal plane and a representation of the noncommutative torus
Definition 6.5. Denote the Moyal plane product ⋆θ on S(cid:0)R2N(cid:1) given by
Θy(cid:19) g (u + v) e2πiy·vdydv
( f ⋆θ h) (u) =Zy∈R2N
f(cid:18)u −
1
2
where Θ is given by (6.20).
[8] Denote by S′ (Rn) the vector space dual to S (Rn), i.e. the space of
Definition 6.6.
continuous functionals on S (Rn). The Moyal product can be defined, by duality, on larger
sets than S(cid:0)R2N(cid:1). For T ∈ S′(cid:0)R2N(cid:1), write the evaluation on g ∈ S(cid:0)R2N(cid:1) as hT, gi ∈ C;
then, for f ∈ S we may define T ⋆θ f and f ⋆θ T as elements of S′(cid:0)R2N(cid:1) by
hT ⋆θ f , gi def= hT, f ⋆θ gi
h f ⋆θ T, gi def= hT, g ⋆θ fi
(6.22)
Remark 6.7. It is proven in [7] that the domain of the Moyal plane product can be extended
using the continuity of the star product on S(cid:0)R2N(cid:1). Also, the involution is extended to
by hT∗, gi def= hT, g∗i.
up to L2(cid:0)R2N(cid:1).
Lemma 6.8.
where k·k2 is the L2-norm given by
[7] If f , g ∈ L2(cid:0)R2N(cid:1), then f ⋆θ g ∈ L2(cid:0)R2N(cid:1) and k fkop
< (2πθ)− N
2 k fk2.
and the operator norm kTkop
def= sup{ kT ⋆ gk2/kgk2 : 0 6= g ∈ L2(cid:0)R2N)(cid:1) }
k fk2
def= (cid:12)(cid:12)(cid:12)(cid:12)ZR2N f2 dx(cid:12)(cid:12)(cid:12)(cid:12)
50
1
2
.
(6.23)
Definition 6.9. Denote by S(cid:0)R2N
θ (cid:1) ) the operator algebra which is C-
linearly isomorphic to S(cid:0)R2N(cid:1) (resp. L2(cid:0)R2N(cid:1) ) and product coincides with ⋆θ. Both
S(cid:0)R2N
θ (cid:1) act on the Hilbert space L2(cid:0)R2N(cid:1). Denote by
θ (cid:1) and L2(cid:0)R2N
θ (cid:1) (resp. L2(cid:0)R2N
Ψθ : S(cid:16)R2N(cid:17) ≈−→ S(cid:16)R2N
θ (cid:17)
(6.24)
the natural C-linear isomorphism.
6.10. There is the tracial property [7] of the Moyal product
ZR2N
( f ⋆θ g) (x) dx =ZR2N
f (x) g (x) dx.
(6.25)
The Fourier transformation of the star product satisfies to the following condition.
F ( f ⋆θ g) (x) =ZR2N F f (x − y) F g (y) eπiy·Θx dy.
(6.26)
norm
(6.27)
kTkop
def= sup{ kT ⋆θ gk2/kgk2 : 0 6= g ∈ L2(R2N) }.
Definition 6.11. [7] Let S′(cid:0)R2N(cid:1) be a vector space dual to S(cid:0)R2N(cid:1). Denote by Cb(cid:0)R2N
θ (cid:1) def=
{ T ∈ S′(cid:0)R2N(cid:1) : T ⋆θ g ∈ L2(cid:0)R2N(cid:1) for all g ∈ L2(R2N) }, provided with the operator
Denote by C0(cid:0)R2N
θ (cid:1) the operator norm completion of S(cid:0)R2N
θ (cid:1) .
Remark 6.12. Obviously S(cid:0)R2N
θ (cid:1). But S(cid:0)R2N
θ (cid:1) is not dense in Cb(cid:0)R2N
θ (cid:1), i.e.
C0(cid:0)R2N
θ (cid:1) ( Cb(cid:0)R2N
θ (cid:1) (cf. [7]).
Remark 6.13. L2(cid:0)R2N
θ (cid:1) is the k · k2 norm completion of S(cid:0)R2N
θ (cid:1) hence from the Lemma
θ (cid:17) .
θ (cid:17) ⊂ C0(cid:16)R2N
Remark 6.14. Notation of the Definition 6.11 differs from [7]. Here symbols Aθ, Aθ, A0
are replaced with Cb(cid:0)R2N
θ (cid:1) respectively.
Remark 6.15. The C-linear space C0(cid:0)R2N
θ (cid:1) ֒→ Cb(cid:0)R2N
L2(cid:16)R2N
θ (cid:1) , C0(cid:0)R2N
θ (cid:1) does not isomorphic to C0(cid:0)R2N(cid:1) (cf. [7]).
[7] By plane waves we understand all functions of the form
θ (cid:1) , S(cid:0)R2N
6.8 it follows that
(6.28)
θ
6.16.
for k ∈ R2N. One obtains for the Moyal product of plane waves:
x 7→ exp(ik · x)
exp (ik·) ⋆Θ exp (ik·) = exp (ik·) ⋆θ exp (ik·) = exp (i (k + l) ·) e−πik·Θl
(6.29)
Remark 6.17.
[7] The algebra Cb(cid:0)R2N
Remark 6.18. If {ck ∈ C}k∈N0 is such that ∑∞
turns out k∑∞
k=0 exp (ik·)kop
< ∑∞
θ (cid:1) contains all plane waves.
k=0 ck < ∞, i.e. ∑∞
k=0 ck < ∞ then from kexp (ik·)kop = 1 it
k=0 ck exp (ik·) ∈ Cb(cid:0)R2N
θ (cid:1).
51
this fact and from the Remark 6.18 it follows that there is an injective *-homomorphism
6.19. The equation (6.29) is similar to the equation (6.15) which defines C(cid:0)T2N
θ (cid:1). From
C∞(cid:0)T2N
θ (cid:1) is dense in C(cid:0)T2N
θ (cid:1) so
there is an injective *-homomorphism C(cid:0)T2N
θ (cid:1). The faithful representation
Cb(cid:0)R2N
θ (cid:1) → B(cid:0)L2(cid:0)R2N(cid:1)(cid:1)
θ (cid:1) ֒→ Cb(cid:0)R2N
θ (cid:1) ֒→ Cb(cid:0)R2N
θ (cid:1) → B(cid:0)L2(cid:0)R2N(cid:1)(cid:1) gives a representation π : C(cid:0)T2N
θ (cid:17) → B(cid:16)L2(cid:16)R2N(cid:17)(cid:17) ,
θ (cid:1) ; Uk 7→ exp (2πik·). An algebra C∞(cid:0)T2N
π : C(cid:16)T2N
(6.30)
where Uk ∈ C(cid:0)Tn
Θ(cid:1) is given by the Definition 6.2.
6.20. Let us consider the unitary dilation operators Ea given by
Uk 7→ exp (2πik·)
Ea f (x) def= aN/2 f (a1/2x),
It is proven in [7] that
f ⋆θ g = (θ/2)−N/2E2/θ(Eθ/2 f ⋆2 Eθ/2g).
(6.31)
We can simplify our construction by setting θ = 2. Thanks to the scaling relation (6.31)
any qualitative result can is true if it is true in case of θ = 2. We use the following notation
Lemma 6.21. Let a, b ∈ S(cid:0)R2N(cid:1). For any ∆ ∈ R2N let a∆ ∈ S(cid:0)R2N(cid:1) be such that a∆ (x) =
a (x + ∆). For any m ∈ N there is a constant Ca,b
m such that
def= f ⋆2 g
f×g
where k·k2 is given by (6.23).
Proof. From the definition of Schwartz functions it follows that for any f ∈ S(cid:0)R2N(cid:1) and
any m ∈ N there is C f
m > 0 such that
ka∆ × bk2
<
Ca,b
m
(1 + k∆k)m
f (u) <
C f
m
(1 + kuk)m .
(6.32)
(6.33)
From (6.26) it follows that
F (a∆ × b) (x) =ZR2N F a∆ (x − y) F b (y) eπiy·Θx dy =ZR2N
c (y − ∆ − x) d (y) eπiy·Θx dy
where c (x) = F a (−x), d (x) = F b (x) If ξ = F (a∆ × b) then ξ ∈ L2(cid:0)R2N(cid:1). Let ξ =
ξ1 + ξ2 where ξ1, ξ2 ∈ L2(cid:0)R2N(cid:1) are given by
0
ξ1 (x) =(F (a∆b) (x) kxk ≤ k∆k2
kxk > k∆k2
ξ2 (x) =(0
kxk ≤ k∆k2
F (a∆b) (x) kxk > k∆k2
,
52
(6.34)
.
From (6.33) it turns out
πil
ξ1 (x) ≤ Z (cid:12)(cid:12)(cid:12)(cid:12)c (t − ∆ − x) d (t) e
mj ·Θt(cid:12)(cid:12)(cid:12)(cid:12) dt ≤ZR2N
=ZR2N
(1 + kt − ∆ − xk)M (1 + ktk)M
(1 + ks − ∆ − xk)M (1 + ksk)M ×ZR2N
(1 + kt − ∆ − xk)M
x∈R2N, kxk≤ k∆k2 , s∈R2N
Cc
MCd
2M
Cc
M
Cc
M
sup
≤
2M
Cd
(1 + ktk)2M dt =
(1 + ktk)M dt ≤
(1 + ktk)M dt.
1
Cd
2M
(6.35)
If x, y ∈ R2N then from the triangle inequality it follows that kx + yk > kyk − kxk, hence
(1 + kxk)M (1 + kx + yk)M ≥ (1 + kxk)M (1 + max (0, kyk − kxk))M .
If kxk ≤ kyk2
then kyk − kxk ≥ kyk2 and
2 (cid:19)M
(1 + kxk)M (1 + kx + yk)M >(cid:18)kyk
.
(6.36)
then condition (6.36) also holds, hence(6.36) is always true. Clearly
then condition (6.36) also holds, hence(6.36) is always true. It turns out from
Clearly if kxk > kyk2
if kxk > kyk2
k−x − ∆k > k∆k2 and (6.36) that
convergent, it turns out
where Γ is the Euler gamma function. If M > 2N then the integral C′ =RR2N
2 ≤ 4MC′Cc
MCd
ξ12
2M
k∆kM
!2Zx∈R2N, kxk≤ k∆k2
4MC′Cc
MCd
2M
k∆kM
Γ (N + 1)(cid:18)k∆k
(1+ktk)M dt is
2 (cid:19)2N
1dx =
πN
1
.
If M = 2N + m then from the above equation it turns out that there is C1 > 0 such that
ξ12
2 ≤
C1
k∆km .
53
(6.37)
inf
x∈R2N, kxk≤ k∆k2 , s∈R2N (1 + ks − ∆ − xk)M (1 + ksk)M >(cid:13)(cid:13)(cid:13)(cid:13)
M
,
∆
4(cid:13)(cid:13)(cid:13)(cid:13)
hence from (6.35) it turns out
ξ1 (x) ≤
There is the well known integral
4MCc
MCd
2M
k∆kM ×ZR2N
1
(1 + ktk)M dt
2 (cid:19)2N
πN
Γ (N + 1)(cid:18)k∆k
Zx∈R2N, kxk≤ k∆k2
1dx =
If (·, ·)L2(R2N) is the given by (6.5) scalar product then from (6.6) it turns out
ξ2 (x) ≤(cid:12)(cid:12)(cid:12)(cid:12)Z c (t − ∆ − x) d (t) eπix·Θtdt(cid:12)(cid:12)(cid:12)(cid:12) =
=(cid:12)(cid:12)(cid:12)(cid:12)(cid:16)c (• − ∆ − x) , d (•) eπix·Θ•(cid:17)L2(R2N)(cid:12)(cid:12)(cid:12)(cid:12) =
=(cid:12)(cid:12)(cid:12)(cid:12)(cid:16)F (c (• − ∆ − x)) , F(cid:16)d (•) eπix·Θ•(cid:17)(cid:17)L2(R2N)(cid:12)(cid:12)(cid:12)(cid:12) =
=(cid:12)(cid:12)(cid:12)(cid:12)ZR2N F (c) (• − ∆ − x) (u) F(cid:16)d (•) eπix·Θ•(cid:17) (u) du(cid:12)(cid:12)(cid:12)(cid:12) ≤
≤ZR2N(cid:12)(cid:12)(cid:12)e−i(−∆−x)·uF (c) (u) F (d) (u + πΘx)(cid:12)(cid:12)(cid:12) du ≤
(1 + ku − πΘxk)2M du ≤
(1 + ku − πΘxk)M (1 + kuk)M ×
(1 + kuk)M du.
Since we consider the asymptotic dependence k∆k → ∞ only large values of k∆k are
interesting, so we can suppose that k∆k > 2. If k∆k > 2 then from kxk > k∆k2
it follows
that kπΘxk > 1, and from (6.36) it follows that
(1 + ks − πΘxk)M
x∈R2N, kxk> k∆k2 , s∈R2N
(1 + kuk)3M
≤ZR2N
(1 + ksk)M
×ZR2N
CF (d)
2M
CF (d)
2M
CF (c)
3M
CF (c)
3M
sup
≤
1
1
(1 + kuk)M (1 + ku − πΘxk)M > kπΘxkM ,
,
M
inf
πΘ∆
hence
ξ2 (x) ≤
4 (cid:13)(cid:13)(cid:13)(cid:13)
x∈R2N, kxk> k∆k2 , s∈R2N
(1 + ksk)M (1 + ks − πΘxk)M >(cid:13)(cid:13)(cid:13)(cid:13)
kπΘxkM ZR2N
4 (cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13) π∆
If m ≥ 1 and M = 2N + m then the integral C′RR2N
4 (cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13) π∆
(1 + kuk)M du.
(1+kuk)M du is convergent and
2M C′
kπΘxkM
Taking into account (6.20) and θ = 2 one has
CF (c)
3M CF (d)
M
CF (c)
3M CF (d)
M
ξ2 (x) ≤
2M
1
1
.
kΘzk = k2zk ; ∀z ∈ R2N.
54
(6.38)
It follows that
2 ≤Zx∈R2Nkxk> k∆k2
ξ12
CF (c)
3M CF (d)
M
2M C′
k2πxkM
2
dx.
Since above integral is convergent one has there is a constant C2 such that
4 (cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13) 2π∆
(cid:13)(cid:13)(cid:13) π∆
4 (cid:13)(cid:13)(cid:13)
C2
2M
ξ22
2 ≤
(6.39)
Since ξ1 ⊥ ξ2 one has ξk2 =qξ12
2 + ξ22
that for any m ∈ N there is Cm > 0 such that
2 and taking into account (6.37), (6.39) it follows
From (6.6) it turns out
kξk2 = kF (a∆ × b)k2 ≤
Ca,b
m
(1 + k∆k)m .
ka∆ × bk2 = kF (a∆ × b)k2 ≤
Ca,b
m
(1 + k∆k)m .
Proposition 6.22.
[7] The algebra S(cid:0)R2N, ⋆θ(cid:1) has the (nonunique) factorization property: for
all h ∈ S(cid:0)R2N(cid:1) there exist f , g ∈ S(cid:0)R2N(cid:1) that h = f ⋆θ g.
(i) Let(cid:8)an ∈ Cb(cid:0)R2N
θ (cid:1)(cid:9)n∈N be a sequence such that
Lemma 6.23. Following conditions hold:
• {an} is weakly-* convergent (cf. Definition 6.1),
• If a = limn→∞ an in the sense of weak-* convergence then a ∈ Cb(cid:0)R2N
θ (cid:1).
Then the sequence {an} is convergent in sense of weak topology {an} (cf. Definition 1.37)
and a is limit of {an} with respect to the weak topology. Moreover if {an} is increasing
or decreasing sequence of self-adjoint elements then {an} is convergent in sense of strong
topology (cf. Definition 1.36) and a is limit of {an} with respect to the strong topology.
(ii) If {an} is strongly and/or weakly convergent (cf. Definitions 1.36, 1.37) and a = limn→∞ an
is strong and/ or weak limit then {an} is weakly-* convergent and a is the limit of {an} in
the sense of weakly-* convergence.
Proof. (i) If h·, ·i : S′(cid:0)R2N(cid:1) × S(cid:0)R2N(cid:1) → C is the natural pairing then one has
lim
n→∞han, bi = ha, bi ; ∀b ∈ S(cid:16)R2N(cid:17) .
(6.40)
55
Let ξ, η ∈ L2(cid:0)R2N(cid:1) and let (cid:8)xj ∈ S(cid:0)R2N
that there are following limits
θ (cid:1)(cid:9)j∈N, (cid:8)yj ∈ S(cid:0)R2N
θ (cid:1)(cid:9)∈N be sequences such
xj = ξ,
lim
j→∞
lim
j→∞
yj = η
(6.41)
in the topology of the Hilbert space L2(cid:0)R2N(cid:1). If (·, ·) : L2(cid:0)R2N(cid:1) × L2(cid:0)R2N(cid:1) → C is the
Hilbert pairing then from (6.22), (6.41) it follows that
hence, taking into account (6.40) one has
lim
n→∞
(anξ, η) = lim
n→∞
(anξ, η) = lim
j→∞(cid:10)anxj, yj(cid:11) = lim
j→∞(cid:10)an, xj ⋆θ yj(cid:11) ,
j→∞(cid:10)an, xj ⋆θ yj(cid:11) = lim
lim
lim
j→∞(cid:0)anxj, yj(cid:1) = lim
n→∞
j→∞(cid:10)a, xj ⋆θ yj(cid:11) = (aξ, η) ,
(6.42)
i.e. {an} is weakly convergent to a. If {an} is an increasing sequence then an < a for any
n ∈ N and from the Lemma 1.43 it turns out that {an} is strongly convergent. Clearly
the strong limit coincides with the weak one. Similarly one can prove that is {an} is an
decreasing then {an} is strongly convergent.
(ii) If b ∈ S(cid:0)R2N(cid:1) then from the Proposition 6.22 it follows that b = x ⋆θ y where x, y ∈
S(cid:0)R2N(cid:1). The sequence {an} is strongly and/or weakly convergent it turns out that
(x, an ⋆θ y) = han, x ⋆θ yi = han, bi
is convergent. Hence {an} is weakly-* convergent.
There are elements fmn ∈ S(cid:0)R2(cid:1) which have very useful properties. To present fmn
explicitly, we use polar coordinates q + ip = ρeiα, where p, q ∈ R2 and ρ = (p, q) ∈ R2
Note that kρk2 = kpk2 + kqk2.
fmn = 2 (−1)nr n!
m!
(cid:16)kρk2(cid:17) e−kρk2
eiα(m−n) kρkm−n Lm−n
fnn (ρ, α) = 2 (−1)n Ln(cid:16)kρk2(cid:17) e−kρk2/2
n
,
n
where Lm−n
C∗-norm completion of linear span of fmn (cf. [8]).
θ(cid:1) is the
, Ln are Laguerre functions. From this properties it follows that C0(cid:0)R2
[8] Let m, n, k, l ∈ N. Then fmn ⋆θ fkl = δnk fml and f ∗mn = fnm. Thus fnn is
Lemma 6.24.
an orthogonal projection and fmn is nilpotent for m 6= n. Moreover, h fmn, fkli = 2N δmk δnl. The
family { fmn : m, n ∈ N0 } ⊂ S(cid:0)R2(cid:1) ⊂ L2(R2) is an orthogonal basis.
Proposition 6.25. [7,8] Let N = 1. Then S(cid:0)R2N
θ(cid:1) has a Fréchet algebra isomorphism
θ (cid:1) = S(cid:0)R2
cmn2(cid:19)1/2
θ2k(cid:16)m + 1
2(cid:17)k
2(cid:17)k(cid:16)n + 1
with the matrix algebra of rapidly decreasing double sequences c = (cmn) such that, for each k ∈ N,
rk(c) def= (cid:18) ∞
∑
m,n=0
(6.43)
56
is finite, topologized by all the seminorms (rk); via the decomposition f = ∑∞
in the { fmn} basis. The twisted product f ⋆θ g is the matrix product ab, where
m,n=0 cmn fmn of S(R2)
(ab)mn
def=
∞
∑
k=0
amkbkn.
(6.44)
For N > 1, C∞(cid:0)R2N
θ (cid:1) is isomorphic to the (projective) tensor product of N matrix algebras of this
kind, i.e.
S(cid:16)R2N
θ (cid:17) ∼= S(cid:16)R2
θ(cid:17) ⊗ · · · ⊗ S(cid:16)R2
θ(cid:17)
}
N−times
{z
with the projective topology induced by seminorms rk given by (6.43).
Remark 6.26. If A is C∗-norm completion of the matrix algebra with the norm (6.43) then
A ≈ K, i.e.
Form (6.45) and (6.46) it follows that
θ(cid:17) ≈ K.
C0(cid:16)R2
θ(cid:17)
θ(cid:17) ⊗ · · · ⊗ C0(cid:16)R2
}
N−times
{z
C0(cid:16)R2N
θ (cid:17) ∼= C0(cid:16)R2
where ⊗ means minimal or maximal tensor product (K is nuclear hence both products
coincide).
≈ K ⊗ · · · ⊗ K
≈ K
(6.47)
N−times
{z
}
(6.45)
(6.46)
6.5 Infinite coverings
Let us consider a sequence
SC(Tn
Θ) =(cid:26)C (Tn
Θ) = C(cid:16)Tn
Θ0(cid:17) π1
−→ ...
π j
−→ C(cid:16)Tn
−−→ ...(cid:27) .
Θj(cid:17) π j+1
(6.48)
of finite coverings of noncommutative tori. The sequence (6.48) satisfies to the Definition
3.1, i.e. SC(Tn
6.27. Let Θ = Jθ where θ ∈ R\Q and
Θ) ∈ FinAlg.
J =(cid:18) 0
−1N
1N
0(cid:19) .
numbers such that pk > 1 for any k, and let mj = Πj
there is a sequence of *-homomorphisms
Θ (cid:1). Let {pk ∈ N}k∈N be an infinite sequence of natural
k=1 pk. From the 6.3 it follows that
Denote by C(cid:0)T2N
Sθ =(cid:26)C(cid:16)T2N
θ (cid:1) def= C(cid:0)T2N
θ (cid:17) → C(cid:16)T2N
such that
1(cid:17) → C(cid:16)T2N
θ/m2
2(cid:17) → ... → C(cid:18)T2N
θ/m2
j(cid:19) → ...(cid:27)
θ/m2
(6.49)
57
θ/m2
j−1(cid:19)(cid:19) and gener-
j−1(cid:19) →
θ/m2
j(cid:19)(cid:19) such that the *-homomorphism C(cid:18)T2N
θ/m2
θ/m2
j(cid:19) is given by
(a) For any j ∈ N there are generators uj−1,1, ..., uj−1,2N ∈ U(cid:18)C(cid:18)T2N
ators uj,1, ..., uj,2N ∈ U(cid:18)C(cid:18)T2N
C(cid:18)T2N
There are generators u1, ..., u2N ∈ U(cid:0)C(cid:0)T2N
C(cid:18)T2N
uj 7→ up1
(b) For any j ∈ N the triple (cid:18)C(cid:18)T2N
1(cid:19) is given by
pj
j,k; ∀k = 1, ..., 2N.
, C(cid:18)T2N
1,j; ∀j = 1, ..., 2N,
uj−1,k 7→ u
θ/m2
θ/m2
θ/m2
j−1
finite-fold covering,
θ (cid:1)(cid:1) such that *-homomorphism C(cid:0)T2N
θ (cid:1) →
j(cid:19) , Zpj(cid:19)(cid:19) is a noncommutative
(c) There is the sequence of groups and epimorphisms
which is equivalent to the sequence
Z2N
m1 ← Z2N
m2 ← ...
G(cid:16)C(cid:16)T2N
θ/m2
1(cid:17) C(cid:16)T2N
θ/m2
2(cid:17) C(cid:16)T2N
j(cid:19) C(cid:16)T2N
θ (cid:17)(cid:17) ← ...
θ (cid:17)(cid:19) ← ... .
θ/m2
j(cid:19), bG
The sequence (6.49), is a specialization of (6.48), hence Sθ ∈ FinAlg. Denote by \C(cid:0)T2N
θ (cid:1) def=
θ (cid:1)(cid:19) . The group bG is Abelian because it
lim−→ C(cid:18)T2N
def= lim←− G(cid:18)C(cid:18)T2N
is the inverse limit of Ablelian groups. Denote by 0bG (resp. "+") the neutral element of bG
(resp. the product operation of bG).
6.28. For anyea ∈ S(cid:0)R2N
θ (cid:1) from (6.7) it turns out that the series
aj =
θ/m2
∑
θ/m2
θ (cid:17)(cid:17) ← G(cid:16)C(cid:16)T2N
← G(cid:18)C(cid:18)T2N
j(cid:19) C(cid:0)T2N
mj(cid:17)ea
g∈ker(cid:16)Z2N→Z2N
is point-wise convergent and aj satisfies to following conditions:
• aj ∈ S′(cid:0)R2N(cid:1),
• aj is a smooth mj - periodic function.
58
ck =
1
m2N
j
On the other hand
1
mj! .
j Fea k
m2N
k
mj · x! dx =
ZR2Nea (x) exp 2πi
ea = lim
mj(cid:17). Letea ∈ S(cid:0)R2N
j→∞
aj
aj = ∑
g∈Gj
gea
(6.50)
(6.51)
(6.52)
It follows that the above series is weakly-* convergent (cf. Definition 6.1) and from the
Lemma 6.23 it turns out that the series is weakly convergent. From (6.8) it follows that
ck exp 2πi
k
mj ·!
aj = ∑
k∈Z2N
where {ck ∈ C}k∈Z2N are rapidly decreasing coefficients given by
in sense of weakly-* convergence, and from the Lemma 6.23 it follows that (6.51) is a limit
in sense of the weak topology.
Lemma 6.29. Let Gj = ker(cid:16)Z2N → Z2N
θ (cid:1) and let
where the sum the series means weakly-* convergence. Following conditions hold:
(i) aj ∈ C∞(cid:0)R2N(cid:1),
(ii) The series (6.52) is convergent with respect to the strong topology (cf. Definition 1.36),
(iii) There is a following strong limit
Proof. (i) From 6.50 it turns out that
aj.
ea = lim
j→∞
(6.53)
aj = ∑
k∈Z2N
ckUk
where {ck} is a rapidly decreasing sequence, hence aj ∈ C∞(cid:0)R2N(cid:1).
(ii) From the Lemma 6.23 it turns out that the series
is strongly convergent, and the series (6.52) is weakly convergent. If k = max(cid:16)1,pkck(cid:17)
then for any η ∈ L2(cid:0)R2N(cid:1) and any subset G ⊂ Z2N following condition holds
c = ∑
g∈Z2N
g (ea∗ea)
gea! η(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
59
∑
g∈G
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
≤ k kηk2
that
where k·k2 is given by (6.23). If ξ ∈ L2(cid:0)R2N(cid:1) then for any ε > 0 there iseb ∈ S(cid:0)R2N(cid:1) such
(6.54)
From the Lemma 6.21 it follows that for any m ∈ N there is a constant Cm > 0 such that
(6.55)
(cid:13)(cid:13)(cid:13)ξ −eb(cid:13)(cid:13)(cid:13)2
(1 + kgk)m ; ∀g ∈ Z2N
ε
2k
Cm
<
<
If m > 2N then there is M ∈ N such that if G0 =
k(gea) bk2
where kgk is given by (6.1).
{−M, . . . , M}2N ⊂ Z2N such that
∑
g∈Z2N\G0
Cm
(1 + kgk)m
<
ε
2
.
(6.56)
It follows that
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
∑
<
< ε.
<
∑
<
ε
2
.
<
ε
2
(1 + kgk)m
g∈Gj\(GjT G0)
Otherwise from (6.54)-(6.56) one has
(cid:13)(cid:13)(cid:0)aj −ea(cid:1) ξ(cid:13)(cid:13)2
Above equation means that the series (6.52) is strongly convergent.
(iii) If j ∈ N is such that mj > M then
g∈Gj\(GjT G0)eaeb(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
g∈GjT G0eaeb(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
=(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
g∈Gjea − ∑
g∈Gj\G0ea(cid:16)ξ −eb(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
g∈GjT G0ea ξ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
<(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
+(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈Gj\(GjT G0)eaeb(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
∑
g∈Gjea − ∑
+ k(cid:13)(cid:13)(cid:13)ξ −eb(cid:13)(cid:13)(cid:13)2
gea −ea ξ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
gea ξ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
=(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
=(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
∑
<(cid:13)(cid:13)(cid:13)(cid:0)aj −ea(cid:1)eb(cid:13)(cid:13)(cid:13)2
+ k(cid:13)(cid:13)(cid:13)ξ −eb(cid:13)(cid:13)(cid:13)2
(cid:13)(cid:13)(cid:0)aj −ea(cid:1) ξ(cid:13)(cid:13)2
=(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
geaeb(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
gea −eaeb(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
∑
Corollary 6.30. Anyea ∈ S(cid:0)R2N
θ (cid:1) lies in \C(cid:0)T2N
θ (cid:1)′′.
Proof. There is a strong limit (6.52), i.e.ea = limj→∞ aj. For any j ∈ N one has aj ∈ \C(cid:0)T2N
θ (cid:1)
it turns outea = limj→∞ aj ∈ \C(cid:0)T2N
θ (cid:1)′′.
where 0 is the neutral element of Z2N. However from mj > M it turns out G0T(cid:0)Gj\{0}(cid:1) =
Above equation means that there is the strong limit (6.53).
∅, so from (6.54)-(6.56) one has
∑
<(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈Gj\{0}
g∈Gj\{0}
+
< ε.
ε
2
+
ε
2
g∈Gj
Cm
<
g∈Gj
60
6.5.1 Equivariant representation
θ/mj
k
Denote bynU
the basis of C(cid:16)T2N
θ/mj(cid:17)(cid:17)ok∈Z2N
∈ U(cid:16)C(cid:16)T2N
θ/mj(cid:17) → B(cid:0)L2(cid:0)R2N(cid:1)(cid:1) given by
there is the representation πj : C(cid:16)T2N
(cid:17) = exp 2πi
πj(cid:16)U
C(cid:16)T2N
There is a following commutative diagram.
θ/mj(cid:17)
C(cid:16)T2N
θ/mj+1(cid:17)
mj ·! .
θ/mj
k
k
πj+1
θ/mj(cid:17). Similarly to (6.30)
πj
θ (cid:1) → B(cid:0)L2(cid:0)R2N(cid:1)(cid:1). There is the
B(cid:0)L2(cid:0)R2N(cid:1)(cid:1)
This diagram defines a faithful representation bπ : \C(cid:0)T2N
action of Z2N × R2N → R2N given by
The action naturally induces the action of Z2N on both L2(cid:0)R2N(cid:1) and B(cid:0)L2(cid:0)R2N(cid:1)(cid:1). Oth-
erwise the action of Z2N on B(cid:0)L2(cid:0)R2N(cid:1)(cid:1) induces the action of Z2N on \C(cid:0)T2N
θ (cid:1). There is
G(cid:16) \C(cid:0)T2N
θ (cid:1)(cid:17)
θ (cid:1) C(cid:0)T2N
the following commutative diagram
(k, x) 7→ k + x.
Z2N
Gj = G(cid:16)C(cid:16)T2N
θ/mj(cid:17) C(cid:0)T2N
θ (cid:1)(cid:17) ≈ Z2N
mj
From the above diagram it follows that there is the natural homomorphism Z2N ֒→ bG,
and Z2N is a normal subgroup. Let J ⊂ bG be a set of representatives of bG/Z2N, and
suppose that 0bG ∈ J. Any g ∈ bG can be uniquely represented as g = gJ + gZ where g ∈ J,
gZ ∈ Z2N. For any g1, g2 ∈ bG denote by ΦJ (g1, g2) ∈ J, ΦZ (g1, g2) ∈ Z2N, such that
Let us define an action of bG onLg∈J L2(cid:0)R2N(cid:1) given by
, ..., 0, ... =0, ..., ΦZ (g1, g2) ξ
}
{z
, ..., 0, ... .
g10, ...,
g1 + g2 = ΦJ (g1, g2) + ΦZ (g1, g2) .
Φ J(g1+g2)th−place
{z}
gth
2 −place
ξ
61
Let X ⊂Lg∈J L2(cid:0)R2N(cid:1) be given by
η ∈Mg∈J
X =
L2(cid:16)R2N(cid:17) η =0, ...,
, ..., 0, ...
{z}
bG−place
Taking into account that X ≈ L2(cid:0)R2N(cid:1), we will write L2(cid:0)R2N(cid:1) ⊂Lg∈J L2(cid:0)R2N(cid:1) instead
of X ⊂ Lg∈J L2(cid:0)R2N(cid:1). This inclusion and the action of bG on Lg∈J L2(cid:0)R2N(cid:1) enable us
θ (cid:1) → B(cid:16)Lg∈J gL2(cid:0)R2N(cid:1)(cid:17)
writeLg∈J gL2(cid:0)R2N(cid:1) instead ofLg∈J L2(cid:0)R2N(cid:1). If bπ⊕ : \C(cid:0)T2N
is given by
0th
ξ
.
bπ⊕ (a) (gξ) = g(cid:16)bπ(cid:16)g−1a(cid:17) ξ(cid:17) ; ∀a ∈ \C(cid:0)T2N
θ (cid:1), ∀g ∈ J, ∀ξ ∈ L2(cid:16)R2N(cid:17)
6.5.2 Inverse noncommutative limit
then bπ⊕ is an equivariant representation.
If ea ∈ S(cid:0)R2N
θ (cid:1) then from the Corollary 6.30 it turns out ea ∈ \C(cid:0)T2N
θ (cid:1)′′. Since bπ⊕ is
a faithful representation of \C(cid:0)T2N
θ (cid:1), one has an injective homomorphism S(cid:0)R2N
θ (cid:1) ֒→
bπ⊕(cid:16) \C(cid:0)T2N
θ (cid:1)(cid:17)′′ of involutive algebras.
For anyea ∈ S(cid:0)R2N
θ (cid:1) following condition holds
g′
gbπ⊕ (ea) = ∑
g∈ker(bG→Gj)
g′′bπ (ea) = ∑
g′′∈ker(Z2N→Gj)
g′∈J
g∈J
gP.
∑
∑
where
P =
∑
g∈ker(Z2N→Gj)
gbπ (ea) .
If J ⊂ bG is a set of representatives of bG/Z2N and g′, g′′ ∈ J are such that g′
6= g′′ then
operators g′P, g′′P act on mutually orthogonal Hilbert subspaces g′L2(cid:0)R2N(cid:1), g′′L2(cid:0)R2N(cid:1)
of the direct sumLg∈J gL2(cid:0)R2N(cid:1), and taking into account kPk = kgPk one has
g∈ker(Z2N→Gj)bπ (ea)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈ker(bG→Gj)bπ⊕ (ea)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Lemma 6.31. Let a ∈ S(cid:0)R2N
= kPk =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
gP(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
θ (cid:1), and let a∆ ∈ S(cid:0)R2N
θ (cid:1) be given by
a∆ (x) = a(x + ∆); ∀x ∈ R2N
=(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
g∈J
(6.58)
(6.57)
∑
∑
.
62
where ∆ ∈ R2N. For any m ∈ N there is a dependent on a real constant Cm > 0 such that for any
j ∈ N following condition holds
≤
Cm
k∆km .
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
g∈ker(bG→Gj)bπ⊕ (a∆a)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
=(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Proof. From (6.57) it follows that
g∈ker(Z2N→Gj)bπ (a∆a)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
From (6.33) it follows that for any f ∈ S(cid:0)R2N(cid:1) and any m ∈ N there is C f
g∈ker(bG→Gj)bπ⊕ (a∆a)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
∑
.
m such that
f (u) <
C f
m
(1 + kuk)m .
Let M = 2N + 1 + m. From (6.19) and (6.50) it follows that
Otherwise from (6.26) it follows that
From the above equations it turns out
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
g∈ker(cid:16)Z2N→Z2N
mj(cid:17)
.
∑
1
j
1
j
≤
m2N
gbπ (a∆a)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
l∈Z2N(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)F (a∆a) l
mj!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
F (a∆a) (x) =ZR2N F a∆ (x − y) F a (y) eπiy·Θx dy.
l∈Z2N(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)F (a∆a) l
mj!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
dt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≤
+ ∆ − t! F a (t) e
mj ·Θt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj! c (t) e
mj ·Θt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj! c (t) e
mj ·Θt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj! c (t) e
l∈Z2N(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Z F a l
l∈Z2NZ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
b t − ∆ −
Z (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
b t − ∆ −
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2
Z (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
b t − ∆ −
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
mj ·Θt
m2N
m2N
m2N
dt =
≤
dt+
∑
l
∑
l
1
j
1
j
∑
l
l
πil
πil
∑
∑
mj
=
l
πil
dt
=
πil
63
=
1
m2N
j
+
1
m2N
j
where b (u) = F a (−u), c (u) = F a (u). From (6.33) it turns out
=
1
m2N
j
≤
1
m2N
j
∑
l
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
N∆
mj
m2N
≤
l
l
j
l
πil
1
2M
Cc
∑
∑
Cb
M
m2N
dt ≤
mj ·Θt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj! c (t) e
b t − ∆ −
(1 + ktk)2M dt =
Z (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2
ZR2N
(cid:16)1 +(cid:13)(cid:13)(cid:13)t − ∆ − l
mj(cid:13)(cid:13)(cid:13)(cid:17)M
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2
ZR2N
(1 + ktk)M dt ≤
mj(cid:13)(cid:13)(cid:13)(cid:17)M
(cid:16)1 +(cid:13)(cid:13)(cid:13)t − ∆ − l
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2
(cid:16)1 +(cid:13)(cid:13)(cid:13)s − ∆ − l
mj(cid:13)(cid:13)(cid:13)(cid:17)M
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2 , s∈R2N
×ZR2N
(1 + ksk)M ×
(1 + ktk)M dt.
(1 + ktk)M
MCc
Cb
Cb
M
sup
Cc
2M
2M
1
where N∆
mj
of points with integer coordinates inside 2N-dimensional cube
mj
can be estimated as a number
l
j
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
=(cid:12)(cid:12)(cid:12)nl ∈ Z2N (cid:13)(cid:13)(cid:13) l
mj(cid:13)(cid:13)(cid:13) ≤ k∆k2 o(cid:12)(cid:12)(cid:12). The number N∆
<(cid:13)(cid:13)mj∆(cid:13)(cid:13)2N .
Z b t − ∆ −
N∆
mj
∑
1
1
j
m2N
From M > 2N it turns out the integralRR2N
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2 , s∈R2N
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
≤ C′1
where
sup
l
k∆k2N
l
1
MCc
C′1 = Cb
mj(cid:13)(cid:13)(cid:13)(cid:17)M
(cid:16)1 +(cid:13)(cid:13)(cid:13)s − ∆ − l
2MZR2N
(1 + ktk)M dt.
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2 , s∈R2N 1 +(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
!M
mj(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
s − ∆ −
l
l
(1+ktk)M dt is convergent, hence
l
mj! c (t) e
πil
mj ·Θt
dt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≤
(1 + ksk)M
It turns out from the (6.36) that
inf
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
(1 + ksk)M >(cid:13)(cid:13)(cid:13)(cid:13)
M
.
∆
4(cid:13)(cid:13)(cid:13)(cid:13)
64
From M = 2N + 1 + m it turns out
1
m2N
j
∑
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
l
mj(cid:13)(cid:13)(cid:13)(cid:13)≤ k∆k2
where C1 = C′1/4m. Clearly
Z b t − ∆ −
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
l
mj! c (t) e
1
m2N
j
=
1
m2N
j
∑
l
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
∑
l
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Z b t − ∆ −
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
b • − ∆ −
C1
k∆km
l
l
πil
πil
πil
=
mj ·Θt
mj ·Θt
mj! c (t) e
mj! , c (•) e
dt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≤
dt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj ·Θ•!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj! , c (•) e
mj!! , F(cid:18)c (•) e
l
l
where (·, ·) means the given by (6.5) scalar product. From the F -invariance of (·, ·) it
follows that
=
1
m2N
j
=
1
m2N
j
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
∑
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
l
≤
1
m2N
j
≤
1
m2N
j
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
sup
l
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2 , s∈R2N
1
m2N
j
∑
l
l
l
πil
πil
πil
=
=
∑
∑
b • − ∆ −
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
F b • + ∆ −
mj! (u) F(cid:18)c (•) e
e−i(cid:18)∆− l
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
ZR2N F (b) • − ∆ −
ZR2N(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
(cid:16)1 +(cid:13)(cid:13)(cid:13)u − Θ πl
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
mj(cid:13)(cid:13)(cid:13)(cid:17)M
mj(cid:13)(cid:13)(cid:13)(cid:17)M
(cid:16)1 +(cid:13)(cid:13)(cid:13)u − Θ πl
(cid:16)1 +(cid:13)(cid:13)(cid:13)s − Θ πl
ZR2N
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
mj ·Θ•!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj ·Θ•(cid:19)!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
mj ·Θ•(cid:19) (u) du(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≤
mj!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
F (b) (u) F (c) u + Θ
mj(cid:13)(cid:13)(cid:13)(cid:17)2M du ≤
(1 + kuk)M ×
(1 + kuk)M du.
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
(1 + kuk)3M
(1 + ksk)M
ZR2N
mj(cid:19)·u
CF (b)
3M
CF (b)
3M
CF (c)
2M
CF (c)
2M
du ≤
m2N
×
∑
∑
πl
1
j
1
1
l
l
l
≤
65
Since we consider the asymptotic dependence k∆k → ∞ only large values of k∆k are
it follows
M
,
,
πl
πl
∑
M
∆
inf
l
1
m2N
j
πl
Θ
s − Θ
u − Θ
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
hence, taking into account (6.38), one has
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2 , s∈R2N
interesting, so we can suppose that k∆k > 2. If k∆k > 2 then from(cid:13)(cid:13)(cid:13) l
mj(cid:13)(cid:13)(cid:13) > k∆k2
that(cid:13)(cid:13)(cid:13)Θ πl
>(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
mj(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
mj(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
!M
mj(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
!M
4(cid:13)(cid:13)(cid:13)(cid:13)
>(cid:13)(cid:13)(cid:13)(cid:13)Θ
mj ·Θ•!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≤
mj! , c (•) e
mj(cid:13)(cid:13)(cid:13) > 1, and from (6.36) it follows that
(1 + kuk)M 1 +(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(1 + ksk)M 1 +(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
b • + ∆ −
ZR2N
(1 + kuk)M =
(cid:13)(cid:13)(cid:13) 2πl
mj(cid:13)(cid:13)(cid:13)
mj(cid:13)(cid:13)(cid:13)(cid:13)>1
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
ZR2N
(1 + kuk)M du! ,
mj(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13) 2πl
mj(cid:13)(cid:13)(cid:13)(cid:13)>1
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
2m . Since M ≥ 2N + 1 the integralRR2N
M =ZR2N−{x∈R2N kxk>1}
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
l
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
C′2
k∆kM
C′2
k∆kM
fmj (x) dx
1
m2N
j
m2N
j
1
m2N
j
∑
l
∑
l
≤
1
M
1
M
πil
1
1
1
l
3m CF (c)
where C′2 = CF (b)
The infinite sum in the above equation can be represented as an integral of step function,
in particular following condition holds
(1+kuk)M du is convergent.
1
where fmj is a multidimensional step function such that
∑
l∈Z2N,(cid:13)(cid:13)(cid:13)(cid:13)
l
mj(cid:13)(cid:13)(cid:13)(cid:13)>1
1
mj (cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13) πΘl
fmj 2πl
mj ! =
From
fmj (x) <
M
1
mj(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13) 2πl
mj (cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13) 2πhx
2
M
66
it follows that
ZR2N−{x∈R2N kxk>1}
fmj (x) dx <ZR2N−{x∈R2N kxk>1}
2
k2πxkM dx.
From m > 2N + 1 it turns out the integral
ZR2N−{x∈R2N kxk>1}ZR2N−{x∈R2N kxk>1}
2
k2πxkm dx
is convergent, hence
< C′′2 =ZR2N−{x∈R2N kxk>1}
l
mj! c (w + ∆) e
i l
mj ·Jw
2
k2πxkM dx.
dt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≤
C2
k∆km
1
(1+kuk)M du. In result for any m > 0 there is
b t + ∆ −
mj! c (t) e
dt <
Cm
πil
l
k∆km .
mj ·Θt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
1
m2N
j
∑
l∈Z2N,(cid:13)(cid:13)(cid:13)(cid:13)
l
mj(cid:13)(cid:13)(cid:13)(cid:13)>1
1
m
mj(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)J l
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
ZR2N
From above equations it follows that
l
j
1
∑
m2N
Cm ∈ R such that
b w −
l∈Z2N, (cid:13)(cid:13)(cid:13)(cid:13)
mj(cid:13)(cid:13)(cid:13)(cid:13)> k∆k2
where M = 2N + 1 + m and C2 = C′2C′′2RR2N
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈ker(bG→Gj)bπ⊕ (a∆a)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
l∈Z2NZ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Lemma 6.32. If a in S(cid:0)R2N
(i) For any j ∈ N0 the following series
m2N
∑
∑
<
1
j
θ (cid:1) is positive then following conditions hold:
(ii) For any ε > 0 there is N ∈ N such that for any j ≥ N the following condition holds
j(cid:19), i.e. aj, bj ∈ C∞(cid:18)T2N
θ/m2
j(cid:19);
θ/m2
aj =
∑
ga,
ga2
∑
bj =
g∈ker(bG→Gj))
g∈ker(bG→Gj)
are strongly convergent and the sums lie in C∞(cid:18)T2N
(cid:13)(cid:13)(cid:13)a2
j − bj(cid:13)(cid:13)(cid:13) < ε.
67
Proof. (i) Follows from the Lemmas 6.23 and/or 6.29.
(ii) Denote by Jj = ker(cid:0)Z2N → Gj(cid:1) = mjZ2N. If
ga,
ga2
aj = ∑
g∈Jj
bj = ∑
g∈Jj
then
a2
j − bj = ∑
g∈Jj
ga ∑
g′∈Jj\{g}
g′′a .
From (6.58) it follows that ga = ag where ag (x) = a (x + g) for any x ∈ R2N and g ∈ Z2N.
Hence the equation (6.59) is equivalent to
(6.59)
a2
j − bj = ∑
g∈Jj
ag ∑
g′∈Jj\{g}
ag′ = ∑
g∈Z2N
amjg ∑
g′∈Z2N\{g}
g′a ∑
g∈Z2N\{0}
amjg′ =
amjg .
= ∑
g′∈Jj
Let m > 1 and M = 2N + 1 + m. From the Lemma 6.31 it follows that there is C ∈ R such
that
From the triangle inequality it follows that
<
C
∑
g∈Jj
k∆kM .
g (aa∆)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
j − bj(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ga ∑
(cid:13)(cid:13)(cid:13)a2
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g(cid:16)a amjg′(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
∑
g∈Jj
∑
g∈Jj
≤ ∑
g′∈Z2N\{0}
g∈Z2N\{0}
≤
amjg′(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)mj g′(cid:13)(cid:13)M .
C
≤ ∑
g′∈Z2N\{0}
From M > 2N it turns out that the series
is convergent and
C′ = ∑
g′∈Z2N\{0}
C
kg′kM
∑
g∈Z2N\{0}
C′
mM
j
.
C
(cid:13)(cid:13)mjg(cid:13)(cid:13)M =
68
If ε > 0 is a small number and N ∈ N is such mN > Mq C′
follows that for any j ≥ N the following condition holds
ε
then from above equations it
(cid:13)(cid:13)(cid:13)a2
j − bj(cid:13)(cid:13)(cid:13) < ε.
θ(cid:17)
θ(cid:17) ⊗ · · · ⊗ S(cid:16)R2
}
N−times
Lemma 6.33. Let us consider a dense inclusion
⊂ S(cid:16)R2N
θ (cid:17)
of algebraic tensor product which follows from (6.45). If a ∈ S(cid:0)R2N
θ (cid:1) is a positive such that
S(cid:16)R2
{z
•
a =
M1
∑
j=0
k=0
c1
jk fjk ⊗ · · · ⊗
cN
jk fjk
MN
∑
j=0
k=0
θ (cid:1) then from the Lemmas 6.23 and/or
θ (cid:1) then z then from
j(cid:19) it follows that z can be regarded as
θ/m2
where cl
jk ∈ C and fjk are given by the Lemma 6.24,
• For any l = 1, . . . , N the sum ∑Ml
j=0
k=0
cl
jk fjk is a rank-one operator.
then a is special.
θ/m2
θ/m2
bj =
Proof. Clearly a is a rank-one operator. If a ∈ S(cid:0)R2N
6.29 it turns out that a satisfies to (a) of the Definition 3.5. If z ∈ C(cid:0)T2N
θ (cid:1) ֒→ C(cid:18)T2N
the injective *-homomorphism C(cid:0)T2N
j(cid:19). Denote by
j(cid:19), i.e. z ∈ C(cid:18)T2N
element of C(cid:18)T2N
g (zaz∗) = z ∑
ga z∗,
g∈ker(bG→Gj)
g∈ker(bG→Gj)
g (zaz∗)2 = z ∑
g (az∗za) z∗,
g∈ker(bG→Gj)
where fε is given by (3.2). From a in S(cid:0)R2N
θ (cid:1) it turns out aj = ∑
j(cid:19).
C(cid:18)T2N
j(cid:19), hence bj = zajz∗ ∈ C(cid:18)T2N
g∈ker(bG→Gj)
g∈ker(bG→Gj)
g fε (zaz∗)
dj =
cj =
∑
θ/m2
∑
∑
g∈ker(bG→Gj) ga ∈
If ξ ∈ H is eigenvector of a such that
θ/m2
69
aξ = kak ξ then η = zξ is an is eigenvector of η = zaz∗ such that zaz∗η = kzaz∗k η. It
follows that (zaz∗)2 = kzaz∗ where k ∈ R+ is given by
.
.
<
δ2 <
θ/m2
k′ =
θ/m2
2
∑
∑
ε
4
,
ε
4
,
ε
4
.
∑
g∈bG
g∈bG
kzaz∗k
∑
g∈bG
g∈bG
Definition 3.5. Let ε > 0, and let δ > 0 be such that
Hence cj = kbj and cj ∈ C∞(cid:18)T2N
k = kzaz∗k2
kzaz∗k
j(cid:19). Similarly fε (zaz∗) = k′ (zaz∗) where
max (0, kzaz∗k − ε)
j(cid:19), it follows that a satisfies to the condition (b) of the
+ 2δ2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Hence dj = k′bj and dj ∈ C∞(cid:18)T2N
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
a(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
a(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
δ4(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
zaz∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g (az∗za)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ga2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(kzk + δ)2(cid:16)δ2 + 2δ kzk(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈bG
The algebra C∞(cid:0)T2N
θ (cid:1) is a dense subalgebra of C(cid:0)T2N
θ (cid:1), so there is y ∈ C∞(cid:0)T2N
θ (cid:1) such
a(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ga y∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ga (z − y)∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
< δ2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
bj − y ∑
(z − y) ∑
g∈ker(bG→Gj)
g∈bG
g∈bGg a(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)∑
and taking into account δ4(cid:13)(cid:13)(cid:13)∑
g∈bGg zaz∗(cid:13)(cid:13)(cid:13) < ε
g∈bGg a(cid:13)(cid:13)(cid:13)
+ 2δ2(cid:13)(cid:13)(cid:13)∑
2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ga y∗
j −y ∑
g∈ker(bG→Gj)
≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g (az∗za) y∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(z − y) ∑
g∈bG
cj − y ∑
g∈ker(bG→Gj)
that kz − yk < δ. From
4 one has
g∈bG
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
<
ε
4
.
∑
<
From
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
2
b2
∑
(6.60)
<
g (az∗za) (z − y)∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
< δ2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g (az∗za)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈bG
∑
70
<
ε
4
.
(6.61)
and taking into account(cid:13)(cid:13)(cid:13)∑
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
From kyk < kzk + δ it turns out
4 one has
g∈bG g (az∗za)(cid:13)(cid:13)(cid:13) δ2 < ε
g (az∗za) y∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
cj − y ∑
g∈ker(bG→Gj)
g (az∗za) y∗ − y ∑
g∈ker(bG→Gj)
∑
≤
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈bG
4 one has
y ∑
g∈ker(bG→Gj)
and taking into account (kzk + δ)2(cid:0)δ2 + 2δ kzk(cid:1)(cid:13)(cid:13)(cid:13)∑
g (az∗za) y∗ − y ∑
g∈ker(bG→Gj)
g (ay∗ya) y∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
≤ (kzk + δ)2(cid:16)δ2 + 2δ kzk(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ga2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g∈bG ga2(cid:13)(cid:13)(cid:13) < ε
g (ay∗ya) y∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
y ∑
g∈ker(bG→Gj)
From y ∈ C∞(cid:0)T2N
θ (cid:1) and a ∈ S(cid:0)R2N
θ (cid:1) it follows that yay∗ ∈ S(cid:0)R2N
(g (yay∗))2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g (yay∗)
∑
g∈ker(bG→Gj)
g∈ker(bG→Gj)
(cid:13)(cid:13)(cid:13)b2
j − cj(cid:13)(cid:13)(cid:13) < ε,
From (6.60)-(6.63) it follows than for any j ≥ N following condition holds
i.e. a satisfies to the condition (c) of the Definition 3.5.
−
∑
ε
4
.
ε
4
.
2
Lemma 6.32 it turns out the existence of N ∈ N such that for any j ≥ N following
condition holds
<
(6.62)
θ (cid:1), hence from the
<
(6.63)
Mg∈J
Corollary 6.34. If Abπ⊕ is the disconnected inverse noncommutative limit of Sθ with respect to
bπ⊕ then
Proof. From the Lemma 6.33 it turns out that Abπ⊕ contains all elements
θ (cid:17)
⊂ S(cid:16)R2N
fj1k1 ⊗ · · · ⊗ fjNkN ∈ S(cid:16)R2
gC0(cid:16)R2N
θ (cid:17) ⊂ Abπ⊕
θ(cid:17)
θ(cid:17) ⊗ · · · ⊗ S(cid:16)R2
}
N−times
{z
(6.64)
71
with the norm k·k2 given by (6.23). One can construct the Hilbert direct sum
where fjl kl (l = 1, . . . , N) are given by the Lemma 6.24. However the linear span of given
θ (cid:1) ⊂ Abπ⊕. From the Corollary 3.9 it
turns out
by (6.64) elements is dense in C0(cid:0)R2N
θ (cid:1), hence C0(cid:0)R2N
θ (cid:17) ⊂ Abπ⊕.
gC0(cid:16)R2N
Mg∈J
6.35. From the Lemma 6.8 it turns out that L2(cid:0)R2N
gL2(cid:16)R2N
X =
x ∈ ∏
g∈J
If a ∈ X is a special element and b = ∑
X =Mg∈J
θ (cid:1) ⊂ B(cid:0)L2(cid:0)R2N(cid:1)(cid:1) is a Hilbert space
θ (cid:17) ⊂ ∏
B(cid:16)gL2(cid:16)R2N(cid:17)(cid:17) (cid:13)(cid:13)(cid:0)..., xgk, ...(cid:1)(cid:13)(cid:13)2 =s ∑
θ (cid:1) then
B(cid:16)gL2(cid:16)R2N(cid:17)(cid:17) ,
< ∞
g∈J(cid:13)(cid:13)xg(cid:13)(cid:13)2
g∈J
2
.
g∈bG ga2 ∈ C(cid:0)T2N
a2dx = kak2
2
τ (b) =ZR2N
g
2
θ/m2
b = ∑
< ∞ for a special element a. In result we have the following lemma.
j(cid:19) lies in X =Lg∈J L2(cid:0)R2N
where τ is given by (6.10), or (6.11). On the other hand τ (b) < ∞ for any b ∈ C(cid:0)T2N
θ (cid:1) it
follows that kak2
Lemma 6.36. The special element a ∈ lim−→ C(cid:18)T2N
θ (cid:1). Moreover if
g∈bG g(cid:0)a2(cid:1) ∈ C(cid:0)T2N
θ (cid:1) then
kak2
where τ is the tracial state on C(cid:0)T2N
θ (cid:1) given by (6.10), (6.12) and k·k2 is given by (6.23).
Remark 6.37. From L2(cid:0)R2N
θ (cid:1) ⊂ C0(cid:0)R2N
θ (cid:1) it follows that any special element in B(cid:0)L2(cid:0)R2N
θ (cid:1)(cid:1)
lies in C0(cid:0)R2N
θ (cid:1).
6.38. Let Abπ⊕ be the disconnected inverse noncommutative limit of Sθ with respect to bπ⊕
of Sθ. From the Corollary 6.34 it follows that
2 = τ (b) < ∞
From the Remark 6.37 it follows that
In result we have
θ (cid:17) ⊂ Abπ⊕\ B(cid:16)L2(cid:16)R2N(cid:17)(cid:17) .
C0(cid:16)R2N
θ (cid:17) .
Abπ⊕\ B(cid:16)L2(cid:16)R2N(cid:17)(cid:17) ⊂ C0(cid:16)R2N
θ (cid:17) .
Abπ⊕\ B(cid:16)L2(cid:16)R2N(cid:17)(cid:17) = C0(cid:16)R2N
72
(6.65)
Similarly for any g ∈ J on has
θ (cid:17) .
Abπ⊕\ B(cid:16)gL2(cid:16)R2N(cid:17)(cid:17) = gC0(cid:16)R2N
The algebra C0(cid:0)R2N
subalgebra.
θ (cid:1) is irreducible. Clearly C0(cid:0)R2N
θ (cid:1) ⊂ Abπ⊕ is a maximal irreducible
Theorem 6.39. Following conditions hold:
(ii)
(i) The representation bπ⊕ is good,
G lim←−bπ⊕
(iii) The triple(cid:0)C(cid:0)T2N
θ (cid:1) , C0(cid:0)R2N
θ (cid:1) , Z2N(cid:1) is an infinite noncommutative covering of Sθ with
Proof. (i) There is the natural inclusion Abπ⊕ ֒→ ∏g∈J B(cid:0)gL2(cid:0)R2N(cid:1)(cid:1) where ∏ means the
θ (cid:17) ;
↓ Sθ = C0(cid:16)R2N
θ (cid:17)! = Z2N,
lim←−bπ⊕
↓ Sθ C(cid:16)T2N
Cartesian product of algebras. This inclusion induces the decomposition
respect to bπ⊕.
Abπ⊕ ֒→ ∏
g∈J(cid:16)Abπ⊕\ B(cid:16)gL2(cid:16)R2N(cid:17)(cid:17)(cid:17) .
g∈J
Abπ⊕ ֒→ ∏
From (6.65) it turns out Abπ⊕T B(cid:0)gL2(cid:0)R2N(cid:1)(cid:1) = gC0(cid:0)R2N
θ (cid:1), hence there is the inclusion
θ (cid:17) .
gC0(cid:16)R2N
From the above equation it follows that C0(cid:0)R2N
θ (cid:1) ⊂ Abπ⊕ is a maximal irreducible subal-
gebra. From the Lemma 6.36 it turns out that algebraic direct sum Lg∈J gC0(cid:0)R2N
θ (cid:1) is a
dense subalgebra of Abπ⊕, i.e. the condition (b) of the Definition 3.13 holds. Clearly the
map \C(cid:0)T2N
θ (cid:1)(cid:1) is injective, i.e.
holds. If G ⊂ bG is the maximal group such that GC0(cid:0)R2N
θ (cid:1) then G = Z2N.
The homomorphism Z2N → Z2N
Definition 3.13 holds.
(ii) and (iii) Follows from the proof of (i).
θ (cid:1) → M(cid:0)C0(cid:0)R2N
θ (cid:1) = C0(cid:0)R2N
is surjective, it turns out that the condition (c) of the
the condition (a) of the Definition 3.13
mj
7 Isospectral deformations and their coverings
A very general construction of isospectral deformations of noncommutative geometries
is described in [5]. The construction implies in particular that any compact spin-manifold
73
M whose isometry group has rank ≥ 2 admits a natural one-parameter isospectral de-
formation to noncommutative geometries Mθ. We let (C∞ (M) , H = L2 (M, S) , /D) be
the canonical spectral triple associated with a compact spin-manifold M. We recall that
A = C∞(M) is the algebra of smooth functions on M, S is the spinor bundle and /D is the
Dirac operator. Let us assume that the group Isom(M) of isometries of M has rank r ≥ 2.
Then, we have an inclusion
with T2 = R2/2πZ2 the usual torus, and we let U(s), s ∈ T2, be the corresponding
unitary operators in H = L2(M, S) so that by construction
T2 ⊂ Isom(M) ,
U(s) /D = /D U(s).
Also,
U(s) a U(s)−1 = αs(a) , ∀ a ∈ A ,
(7.1)
where αs ∈ Aut(A) is the action by isometries on the algebra of functions on M.
We let p = (p1, p2) be the generator of the two-parameters group U(s) so that
The operators p1 and p2 commute with D. Both p1 and p2 have integral spectrum,
U(s) = exp(i(s1 p1 + s2 p2)) .
Spec(pj) ⊂ Z , j = 1, 2 .
One defines a bigrading of the algebra of bounded operators in H with the operator T
declared to be of bidegree (n1, n2) when,
where αs(T) = U(s) T U(s)−1 as in (7.1).
αs(T) = exp(i(s1n1 + s2n2)) T , ∀ s ∈ T2 ,
Any operator T of class C∞ relative to αs (i. e. such that the map s → αs(T) is of class
C∞ for the norm topology) can be uniquely written as a doubly infinite norm convergent
sum of homogeneous elements,
withbTn1,n2 of bidegree (n1, n2) and where the sequence of norms bTn1,n2 is of rapid decay
in (n1, n2). Let λ = exp(2πiθ). For any operator T in H of class C∞ we define its left twist
l(T) by
(7.2)
l(T) = ∑
and its right twist r(T) by
T = ∑
n1,n2 bTn1,n2 ,
n1,n2 bTn1,n2 λn2 p1 ,
n1,n2 bTn1,n2 λn1 p2 ,
r(T) = ∑
74
Since λ = 1 and p1, p2 are self-adjoint, both series converge in norm. Denote by
C∞ (M)n1,n2 ⊂ C∞ (M) the C-linear subspace of elements of bidegree (n1, n2).
One has,
Lemma 7.1.
[5]
a) Let x be a homogeneous operator of bidegree (n1, n2) and y be a homogeneous operator of
bidegree (n′1, n′2). Then,
l(x) r(y) − r(y) l(x) = (x y − y x) λn′1n2 λn2 p1+n′1 p2
In particular, [l(x), r(y)] = 0 if [x, y] = 0.
b) Let x and y be homogeneous operators as before and define
x ∗ y = λn′1n2 xy ;
(7.3)
(7.4)
then l(x)l(y) = l(x ∗ y).
The product ∗ defined in (7.4) extends by linearity to an associative product on the linear
space of smooth operators and could be called a ∗-product. One could also define a
deformed 'right product'. If x is homogeneous of bidegree (n1, n2) and y is homogeneous
of bidegree (n′1, n′2) the product is defined by
x ∗r y = λn1n′2 xy .
Then, along the lines of the previous lemma one shows that r(x)r(y) = r(x ∗r y).
We can now define a new spectral triple where both H and the operator D are unchanged
while the algebra C∞ (M) is modified to l(C∞ (M)) . By Lemma 7.1 b) one checks that
l (C∞ (M)) is still an algebra. Since D is of bidegree (0, 0) one has,
[D, l(a)] = l([D, a])
which is enough to check that [D, x] is bounded for any x ∈ l(A). There is a spectral triple
(l (C∞ (M)) , H, D).
l (C∞ (M)), and denote by ρ : C (M) → L2 (M, S) (resp. πθ : C (Mθ) → B(cid:0)L2 (M, S)(cid:1) )
Denote by C (Mθ) the operator norm completion (equivalently C∗-norm completion) of
natural representations.
7.1 Finite-fold coverings
covering. There are two closed paths ω1, ω2 : [0, 1] → M given by
ω1 (t) = ϕ (t, 0) x0, ω2 (t) = ϕ (0, t) x0.
Let M be a spin - manifold with the smooth action of T2. Let π : eM → M be a finite-fold
covering. Let ex0 ∈ eM and x0 = π (ex0). Denote by ϕ : R2 → R2/Z2 = T2 the natural
There are lifts of these paths, i.e. maps eω1,eω2 : [0, 1] → eM such that
eω1 (0) = eω2 (0) =ex0,
π (eω1 (t)) = ω1 (t) ,
π (eω2 (t)) = ω2 (t) .
75
Since π is a finite-fold covering there are N1, N2 ∈ N such that if
γ1 (t) = ϕ (N1t, 0) x0, γ2 (t) = ϕ (0, N2t) x0.
and eγ1 (resp. eγ2) is the lift of γ1 (resp. γ2) then both eγ1, eγ2 are closed. Let us select
minimal positive values of N1, N2. If prn : S1 → S1 is an n listed covering and prN1,N2
covering given by
the
ϕθ : C∞ (Mθ) → C∞(cid:16)eMeθ(cid:17) ,
ϕeθ (C∞ (Mθ))n1,n2 ⊂ C∞(cid:16)eMeθ(cid:17)n1 N1, n2 N2
.
76
prN1×prN2
−−−−−−→→ S1 × S1 = T2
eT2 = S1 × S1
π
M
eM
prN1N2 × π
then there is the action eT2 × eM → eM such that
eT2 × eM
T2. × M
where eT2 ≈ T2. Let ep = (ep1,ep2) be the generator of the associated with eT2 two-
parameters group eU (s) so that
eU (s) = exp (i (s1ep1 + s2ep2)) .
The covering eM → M induces an involutive injective homomorphism
ϕ : C∞ (M) ֒→ C∞(cid:16)eM(cid:17) .
Since eM → M is a covering C∞(cid:16)eM(cid:17) is a finitely generated projective C∞ (M)-module, i.e.
there is the following direct sum of C∞(cid:16)eM(cid:17)-modules
C∞(cid:16)eM(cid:17)M P = C∞ (M)n
ϕ (C∞ (M))n1,n2 ⊂ C∞(cid:16)eM(cid:17)n1 N1, n2N2
such that
Let θ,eθ ∈ R be such that
If λ = e2πiθ,eλ = e2πieθ then λ =eλN1 N2. There are isospectral deformations C∞ (Mθ) , C∞(cid:16)eMeθ(cid:17)
and C-linear isomorphisms l : C∞ (M) → C∞ (Mθ),el : C∞(cid:16)eM(cid:17) → C∞(cid:16)eMeθ(cid:17). These iso-
morphisms and the inclusion ϕ induce the inclusion
, where n ∈ Z.
θ + n
N1N2
eθ =
(7.5)
.
From (7.5) it follows that
el(cid:16)C∞(cid:16)eM(cid:17)(cid:17)M l (P) = l (C∞ (M))n ,
or equivalently C∞(cid:16)eMeθ(cid:17)M l (P) = C∞ (Mθ)n ,
then
p ∈ Mn (C∞ (Mθ)) such that
C∞(cid:16)eMeθ(cid:17) = pC∞ (Mθ)n
C(cid:16)eMeθ(cid:17) = pC (Mθ)n ,
i.e. C∞(cid:16)eMeθ(cid:17) is a finitely generated projective C∞ (Mθ) module. There is a projection
If C(cid:16)eMeθ(cid:17) (resp. C (Mθ) ) is the operator norm completion of C∞(cid:16)eMeθ(cid:17) (resp. C∞ (Mθ) )
i.e. C(cid:16)eMeθ(cid:17) is a finitely generated projective C (Mθ) module. Denote by G = G(cid:16)eM M(cid:17)
the group of covering transformations. Sinceel is a C-linear isomorphism the action of
G on C∞(cid:16)eM(cid:17) induces a C-linear action G × C∞(cid:16)eMeθ(cid:17) → C∞(cid:16)eMeθ(cid:17). According to the
definition of the action of eT2 on eM it follows that the action of G commutes with the
action of eT2. It turns out
then g(cid:16)eaeb(cid:17) =
(gea)(cid:16)geb(cid:17) ∈ C∞(cid:16)eM(cid:17)n1+n′1,n2+n′2
for any n1, n2 ∈ Z and g ∈ G.
. One has
gC∞(cid:16)eM(cid:17)n1,n2
= C∞(cid:16)eM(cid:17)n1,n2
If ea ∈ C∞(cid:16)eM(cid:17)n1,n2
, eb ∈ C∞(cid:16)eM(cid:17)n′1,n′2
el (ea)el(cid:16)eb(cid:17) =eλn′1n2el(cid:16)eaeb(cid:17) ,
eλn2ep1 l(cid:16)eb(cid:17) =eλn′1n2 l(cid:16)eb(cid:17)eλn2ep1,
el (gea)el(cid:16)geb(cid:17) = geaeλn2ep1 gebeλn′2ep1 =eλn′1n2 g(cid:16)eaeb(cid:17)eλ(n2+n′2)ep1.
g(cid:16)el (ea)el(cid:16)eb(cid:17)(cid:17) = g(cid:16)eλn′1n2el(cid:16)eaeb(cid:17)(cid:17) =eλn′1n2 g(cid:16)eaeb(cid:17)eλ(n2+n′2)ep1.
From above equations it turns out
On the other hand
el (gea)el(cid:16)geb(cid:17) = g(cid:16)el (ea)el(cid:16)eb(cid:17)(cid:17) ,
i.e.
g corresponds to automorphism of C∞(cid:16)eMeθ(cid:17).
of automorphisms of C∞(cid:16)eMeθ(cid:17). Clearly form ea ∈ C∞(cid:16)eMeθ(cid:17)n1,n2
77
It turns out that G is the group
it follows that ea∗ ∈
C∞(cid:16)eMeθ(cid:17)−n1,−n2
. One has
g(cid:16)(cid:16)el (ea)(cid:17)∗(cid:17) = g(cid:16)eλ−n2fp1ea∗(cid:17) = g(cid:16)eλn1n2ea∗eλ−n2fp1(cid:17) =eλn1n2 g(cid:16)el (ea∗)(cid:17) .
On the other hand
theorem.
It follows that g corresponds to the involutive automor-
(cid:16)gel (ea)(cid:17)∗ =(cid:16)(gea)eλn2fp1(cid:17)∗ =eλ−n2fp1 (ga∗) =eλn1n2(cid:16)ga∗eλ−n2fp1(cid:17) =eλn1n2 g(cid:16)el (ea∗)(cid:17) ,
i.e. g(cid:16)(cid:16)el (ea)(cid:17)∗(cid:17) = (cid:16)gel (ea)(cid:17)∗.
phism of C∞(cid:16)eMeθ(cid:17). Since C∞(cid:16)eMeθ(cid:17) is dense in C(cid:16)eMeθ(cid:17) there is the unique involutive
action G × C(cid:16)eMeθ(cid:17) → C(cid:16)eMeθ(cid:17). From the above construction it turns out the following
Theorem 7.2. The triple(cid:16)C (Mθ) , C(cid:16)eMeθ(cid:17) , G(cid:16)eM M(cid:17)(cid:17) is an unital noncommutative finite-
Let SM =(cid:8)M = M0 ← M1 ← ... ← Mn ← ...(cid:9) ∈ FinTop be an infinite sequence of spin
- manifolds and regular finite-fold covering. Suppose that there is the action T2 × M → M
given by (7.1). From the Theorem 7.2 it follows that there is the algebraical finite covering
sequence
7.2 Infinite coverings
fold covering.
SC(Mθ) =(cid:8)C (Mθ) → ... → C(cid:0)Mn
θn(cid:1) → ...(cid:9) .
So one can calculate a finite noncommutative limit of the above sequence. This article does
not contain detailed properties of this noncommutative limit, because it is not known yet
by the author of this article.
Acknowledgment
Author would like to acknowledge members of the Moscow State University Seminar
"Algebras in analysis" leaded by professor A. Ya. Helemskii and others for a discussion
of this work.
References
[1] Charles A. Akemann, Gert K. Pedersen, Jun Tomiyama. Multipliers of C∗-algebras.
Journal of Functional Analysis Volume 13, Issue 3, July 1973, Pages 277-301, 1973.
[2] W. Arveson. An Invitation to C∗-Algebras, Springer-Verlag. ISBN 0-387-90176-0, 1981.
[3] B. Blackadar. K-theory for Operator Algebras, Second edition. Cambridge University
Press 1998.
78
[4] Chun-Yen Chou. Notes on the Separability of C∗-Algebras. TAIWANESE JOURNAL OF
MATHEMATICS Vol. 16, No. 2, pp. 555-559, April 2012 This paper is available online
at http://journal.taiwanmathsoc.org.tw, 2012.
[5] Alain Connes, Giovanni Landi. Noncommutative Manifolds the Instanton Algebra and
Isospectral Deformations, arXiv:math/0011194, 2001.
[6] J. Dixmier. Traces sur les C∗-algebras. Ann. Inst. Fourier, 13, 1(1963), 219-262, 1963.
[7] V. Gayral, J. M. Gracia-Bondía, B. Iochum, T. Schücker, J. C. Varilly. Moyal Planes are
Spectral Triples. arXiv:hep-th/0307241, 2003.
[8] José M. Gracia-Bondia, Joseph C. Varilly. Algebras of Distributions suitable for phase-
space quantum mechanics. I. Escuela de Matemáatica, Universidad de Costa Rica,San
José, Costa Rica J. Math. Phys 29 (1988), 869-879, 1988.
[9] Paul R. Halmos Naive Set Theory. D. Van Nostrand Company, Inc., Prineston, N.J.,
1960.
[10] James R. Munkres. Topology. Prentice Hall, Incorporated, 2000.
[11] G.J. Murphy. C∗-Algebras and Operator Theory. Academic Press 1990.
[12] Alexander Pavlov, Evgenij Troitsky. Quantization of branched coverings. arXiv:1002.3491,
2010.
[13] Pedersen G.K. C∗-algebras and their automorphism groups. London ; New York : Aca-
demic Press, 1979.
[14] Jonathan Rosenberg. Continuous-trace algebras from the bundle theoretic point of view.
Journal of the Australian Mathematical Society, Volume 47, Issue 3 December 1989 ,
pp. 368-381, 1989.
[15] E.H. Spanier. Algebraic Topology. McGraw-Hill. New York 1966.
[16] Zirô Takeda. Inductive limit and infinite direct product of operator algebras. Tohoku Math.
J. (2) Volume 7, Number 1-2 (1955), 67-86. 1955.
79
|
1305.2321 | 1 | 1305 | 2013-05-09T13:20:45 | Type-Decomposition of a Synaptic Algebra | [
"math.OA"
] | A synaptic algebra is a generalization of the self-adjoint part of a von Neumann algebra. In this article we extend to synaptic algebras the type-I/II/III decomposition of von Neumann algebras, AW*-algebras, and JW-algebras. | math.OA | math |
Type-Decomposition of a Synaptic Algebra
David J. Foulis∗ and Sylvia Pulmannov´a†
Abstract
A synaptic algebra is a generalization of the self-adjoint part of a von
Neumann algebra. In this article we extend to synaptic algebras the
type-I/II/III decomposition of von Neumann algebras, AW∗-algebras,
and JW-algebras.
1
Introduction
Soon after laying rigorous Hilbert-space based foundations for quantum me-
chanics in his celebrated book [22], John von Neumann wrote in an unpub-
lished letter to Garrett Birkhoff,
"I would like to make a confession which may seem immoral: I
do not believe absolutely in Hilbert space any more."
As is authoritatively documented in [25], by the time this letter was written
(1935), von Neumann had begun to focus on what is now called a type
II1 factor as the appropriate mathematical basis for quantum mechanics.
Later von Neumann's advocacy of type II1 factors was supplemented by the
discovery that type III factors occur naturally in relativistic quantum field
theory [15].
∗Emeritus Professor, Department of Mathematics and Statistics, University of Mas-
sachusetts, Amherst, MA; Postal Address: 1 Sutton Court, Amherst, MA 01002, USA;
[email protected].
†Mathematical Institute, Slovak Academy of Sciences, Stef´anikova 49, SK-814 73
Bratislava, Slovakia; [email protected]. The second author was supported by Re-
search and Development Support Agency under the contract No. APVV-0178-11 and
grant VEGA 2/0059/12.
1
In this article we are going to study the type I/II/III decomposition theory
for a so-called synaptic algebra, which is a proper generalization of the self-
adjoint part of a von Neumann algebra. We believe that our work casts
considerable light on just what makes a type I/II/III decomposition work,
not only in von Neumann algebras, but in many related algebraic structures
as well. We note that a synaptic algebra can host the probability measures
that were a main concern of von Neumann [25, §2] and it can serve as a value
algebra for quantum-mechanical observables.
A synaptic algebra (from the Greek sunaptein, meaning to join together)
[5, 8, 13, 24] unites the notions of an order-unit normed space [1, p. 69], a
special Jordan algebra [21], a convex effect algebra [14], and an orthomodular
lattice [2, 19]. The generalized Hermitian algebras, introduced and studied
in [7, 11], are special cases of synaptic algebras, and numerous additional
examples can be found in the papers cited above.
The JW-algebras of D. Topping [27] are important special cases of synap-
tic algebras and they will motivate much of our work in this article. One
of the significant ways in which synaptic algebras are more general than
JW-algebras is that, whereas the orthomodular lattice (OML) of projections
in a JW-algebra is complete [27, Theorem 4], the OML of projections in a
synaptic algebra need not be complete.
Our purpose in this article is twofold: First Project: To show that a
synaptic algebra with a complete OML of projections has sufficiently many
properties in common with a JW-algebra to enable Topping's proof of his
version of a type-I/II/III decomposition theorem [27, Theorem 13]. Sec-
ond Project: To show how the type-decomposition theory developed in [10]
applies to a synaptic algebra with a projection lattice satisfying the much
weaker central orthocompleteness condition. For both projects, our main
tool will be the notion of a type determining (TD) subset of the projection
lattice (Section 6 below).
2 Some basic properties of a synaptic algebra
To carry out our two type-decomposition projects, we shall need only a por-
tion of the theory of synaptic algebras as developed in [5, 8, 13, 24], and as
a convenience for the reader, we devote this and the next two sections to a
sketch of some of the parts of this theory that we shall require and to some
of the corresponding notation and nomenclature. We use the notation := for
2
"equals by definition" and "iff" abbreviates "if and only if."
2.1 Standing Assumption. In this article we assume that A is a synaptic
algebra with enveloping algebra R [5, Definition 1.1].
2((a + b)2 − a2 − b2) = 1
To help fix ideas, the reader might want to keep in mind the case in which
R is a von Neumann algebra and A is the self-adjoint part of R. Some of the
important properties of A and R are as follows:
• R is a real or complex linear associative algebra with unit element 1 and
A is a real vector subspace of R. To avoid trivialities, we shall assume that
0 6= 1.
• A is a partially ordered real vector space under ≤ and 0 ≤ 1 ∈ A.
• Let a, b ∈ A. Then the product ab as calculated in R may or may not
belong to A. We write aCb iff a and b commute (i.e. ab = ba) and we define
C(a) := {b ∈ A : aCb}. If B ⊆ A, then C(B) := Tb∈B C(b).
• If a ∈ A, then 0 ≤ a2 ∈ A. Thus, A is a special Jordan algebra under
the Jordan product a ◦ b := 1
2 (ab + ba) ∈ A for
all a, b ∈ A. Therefore, if a, b ∈ A, then ab + ba = 2(a ◦ b) ∈ A and
aCb ⇒ ab = ba = a ◦ b = b ◦ a ∈ A. Also, if ab = 0, then aCb and ba = 0.
Moreover, aba = 2a ◦ (a ◦ b) − a2 ◦ b ∈ A and if 0 ≤ b, then 0 ≤ aba.
• With the operations and partial order inherited from A, the set C(A),
called the center of A, is a synaptic algebra with unit element 1. As such, it
is a commutative associative partially-ordered normed real linear algebra and
it is its own enveloping algebra. We call A a commutative synaptic algebra
iff A = C(A).
• An element p ∈ A is called a projection iff p2 = p, and the set of all
projections in A is denoted by P . Under the partial order inherited from A,
P is an orthomodular lattice (OML) [2, 19] with smallest element 0, largest
element 1, and p 7→ p⊥ := 1 − p as the orthocomplementation.
• If p ∈ P , then, with the operations and partial order inherited from A,
pAp := {pap : a ∈ A} is a synaptic algebra with pRp as its enveloping
algebra and with p as its unit element. The OML of projections in pAp is
pAp ∩ P = {q ∈ P : q = pqp} = {q ∈ P : q = qp} = {q ∈ P : q ≤ p}.
An arbitrary cartesian product of synaptic algebras is again a synaptic
algebra with coordinatewise operations and relations and with the cartesian
product of the enveloping algebras of the factors as its enveloping algebra.
3
3 Orthomodular lattices
In this section we review some facts about orthomodular lattices (OMLs)
that we shall need in our study of the orthomodular projection lattice P of
the synaptic algebra A. More details in regard to OMLs can be found in
[2, 19].
3.1 Standing Assumption. In this section, we assume that L is an OML
with smallest element 0, largest element 1, and p 7→ p⊥ as its orthocomple-
mentation.
Let p, q ∈ L. We say that q dominates p, or equivalently, that p is a
subelement of q iff p ≤ q. If p ≤ q and p 6= q, we write p < q. As usual, p ∨ q
and p ∧ q will denote the supremum (least upper bound) and the infimum
(greatest lower bound), respectively, of p and q in L. The two elements
p, q ∈ L are said to be orthogonal, in symbols p ⊥ q, iff p ≤ q⊥.
The p-interval in L, defined and denoted by L[0, p] := {q ∈ L : q ≤ p} is
a sublattice of L and it is an OML in its own right with q 7→ q⊥p := p ∧ q⊥
as its orthocomplementation. If Q ⊆ L[0, p], then Q has a supremum in L
iff it has a supremum in L[0, p], and the two suprema, if they exist, coincide.
Likewise for infima, provided that Q is not empty. Therefore, if the OML L
is complete (i.e., every subset of L has a supremum and an infimum in L)
then the OML L[0, p] is also complete.
The elements p, q ∈ L are called (Mackey) compatible iff there are ele-
ments p1, q1, d ∈ L such that p1 ⊥ q1, p1 ⊥ d, p2 ⊥ d, p = p1 ∨ p2, and
q = q1 ∨ q2. For instance, if p ≤ q, or if p ⊥ q, then p and q are compatible.
The elements p and q are compatible iff p can be written as p = x ∨ y with
x ∈ L[0, q] and y ∈ L[0, q⊥]. The set of all elements in L that are compatible
with every element in L is called the center of L. The center of L is a sub-
lattice of L, closed under orthocomplementation, and as such it is a boolean
algebra (a complemented distributive lattice). Computations in L are facil-
itated by the fact that, if one of the elements p, q, r ∈ L is compatible with
the other two, then the distributive relations p ∨ (q ∧ r) = (p ∨ q) ∧ (p ∨ r)
and p ∧ (q ∨ r) = (p ∧ q) ∨ (p ∧ r) hold [4].
If c belongs to the center of L and p ∈ L, then p ∧ c belongs to the center
of L[0, p]. If, conversely, for every p ∈ L, every element in the center of L[0, p]
has the form p ∧ c for some c in the center of L, then L is said to have the
relative center property [3].
4
3.2 Remark. The OML L can be regarded as a lattice effect algebra [6, 26]
by defining the orthosum p⊕q for p, q ∈ L iff p ⊥ q, in which case p⊕q := p∨q.
Then the partial order on L coincides with the effect-algebra partial order, the
orthocomplementation on L coincides with the effect-algebra orthosupple-
mentation, and the structure of L as an effect algebra determines its structure
as an OML. In this way the theory of effect algebras [6, 9, 10, 12, 14, 18, ?, 26]
can be applied to L.
A family (pi)i∈I ⊆ L in L is said to be pairwise orthogonal iff, for all
i, j ∈ I, i 6= j ⇒ pi ⊥ pj. Regarding L as an effect algebra and using
standard effect-algebra terminology (e.g., [9, p. 286]), we have the following:
A family in L is orthogonal iff it is pairwise orthogonal, such an orthogonal
family is orthosummable iff it has a supremum in L, and if the family is
orthosummable, then its orthosum is its supremum.
If every orthogonal family in an effect algebra is orthosummable, then the
effect algebra is called orthocomplete [18]. By a theorem of S. Holland [17],
L is orthocomplete as an effect algebra iff it is complete as a lattice.
The OML L is said to be modular iff, for all p, q, r ∈ L, p ≤ r ⇒
p ∨ (q ∧ r) = (p ∨ q) ∧ r; it is called locally modular [27, p. 28] iff, for every
nonzero central element c ∈ L, there is a nonzero p ∈ L[0, c] such that L[0, p]
is modular.
If p, q ∈ L, p ∨ q = 1, and p ∧ q = 0, then p and q are called complements
of each other in L. For instance, p and p⊥ are complements in L. Two
elements of L that share a common complement are said to be perspective.
If p and q are perspective in the OML L[0, p ∨ q], then p and q are called
strongly perspective. The transitive closure of the relation of perspectivity is
an equivalence relation on L called projectivity; thus p and q are projective
iff there is a finite sequence e1, e2, ..., en ∈ L such that p = e1, q = en, and ei
is perspective to ei+1 for i = 1, 2, ..., n − 1.
3.3 Lemma. Let r ∈ L and p, q ∈ L[0, r]. Then: (i) If p and q are perspective
in L[0, r], then p and q are perspective in L.
(ii) If p and q are strongly
perspective in L, then they are perspective in L. (iii) If p and q are strongly
perspective in L, then they are strongly perspective in L[0, r].
Proof. (i) Suppose there exists x ∈ L[0, r] such that p ∨ x = q ∨ x = r and
p ∧ x = q ∧ x = 0 and put y := x ∨ r⊥. Then p ∨ y = p ∨ (x ∨ r⊥) = r ∨ r⊥ = 1,
and since p, x ≤ r, p ∧ y = p ∧ (x ∨ r⊥) = (p ∧ x) ∨ (p ∧ r⊥) = 0 ∨ 0 = 0.
Likewise q ∨ y = 1 and q ∧ y = 0.
5
(ii) Part (ii) follows from (i) with r := p ∨ q.
(iii) If p and q are strongly perspective in L, then they are perspective
in L[0, p ∨ q] = (L[0, r])[0, p ∨ q], whence they are strongly perspective in
L[0, r].
3.4 Theorem. The following conditions are mutually equivalent:
(i) L is modular.
(ii) If p, q ∈ L are perspective, then p and q are strongly perspective.
(iii) If p, q ∈ L, p ≤ q, and p is perspective to q, then p = q.
Proof. That (i) ⇔ (ii) follows from [16, Theorem 2] and the equivalence (i)
⇔ (iii) follows from [27, Lemma 20].
3.5 Theorem. Suppose that L is both complete and modular. Then:
(i) L is a continuous geometry [23].
(ii) Perspectivity is transitive on L, i.e., perspectivity coincides with pro-
jectivity.
(iii) If p, q ∈ L, p ≤ q, and p and q are projective in L, then p = q.
(iv) Any orthogonal family of nonzero elements in L such that any two of
the elements in the family are projective is necessarily finite.
Proof. Part (i) is a classic result of I. Kaplansky [20], (ii) is [23, Theorem
5.16], (iii) is [23, Theorem 4.4], and (iv) is [23, Theorem 3.8].
4 The orthomodular lattice of projections
Owing to the fact that P ⊆ A, the OML P of projections in A acquires
several special properties, among which are the following:
For all p, q ∈ P : (i) p ≤ q ⇔ pq = p ⇔ qp = p ⇔ p = qpq ⇔ p = pqp. (ii)
If p ≤ q, then q − p = q ∧ p⊥ = qp⊥ = p⊥q. (iii) If pCq, then p ∧ q = pq = qp
and p ∨ q = p + q − pq. (iv) p ⊥ q iff p + q ≤ 1 iff p + q = p ∨ q iff pq=0. (v) p
and q are compatible iff pCq. (vi) C(A) = C(P ) [8, p. 242]. (vii) The center
of the OML P is P ∩ C(P ) = P ∩ C(A) and it coincides with the boolean
algebra of projections in the center C(A) of A. (viii) A projection c ∈ P is
6
central, i.e., it belongs to the center P ∩ C(A) = P ∩ C(P ) of P , iff P =
P [0, c] + P [0, c⊥] := {x + y : x ∈ P [0, c], y ∈ P [0, c⊥]}. (ix) If d ∈ P ∩ C(A),
then pd = p ∧ d belongs to the center of P [0, p]. (x) If c ∈ P ∩ C(A), then
the center of P [0, c] is {cd : d ∈ P ∩ C(A)} = (P ∩ C(A))[0, c]. (xi) If P
is complete, then it has the relative center property [13, Theorem 8.7]; hence
the center of P [0, p] is {pd : d ∈ P ∩ C(A)}. (xii) The p-interval P [0, p] is
the OML of projections in the synaptic algebra pAp.
If p1, p2, ..., pn is a finite orthogonal sequence in P , then we refer to p1 +
p2 + · · · + pn = p1 ∨ p2 ∨ · · · ∨ pn as an orthogonal sum.
4.1 Definition. The family (pi)i∈I ⊆ P is called centrally orthogonal
iff
there is an orthogonal family (ci)i∈I ⊆ P ∩ C(A) of projections in the center
C(A) = C(P ) of A such that pi ≤ ci for every i ∈ I. We say that P
is centrally orthocomplete iff every centrally orthogonal family in P has a
supremum in P .
Clearly, every centrally orthogonal family is orthogonal, and if P is com-
plete, then it is centrally orthocomplete.
If P is centrally orthocomplete,
then the center P ∩ C(P ) = P ∩ C(A) is a complete boolean algebra; more-
over, for each a ∈ A, there is a smallest central projection c ∈ P ∩ C(A) such
that a = ac [13, Lemma 6.5 and Definition 6.6].
4.2 Definition. Suppose that P is centrally orthocomplete. For each a ∈ A,
the smallest central projection c ∈ P ∩ C(A) such that a = ac is called the
central cover of a and denoted by γa.
Thus, if P is centrally orthocomplete, a ∈ A, and c ∈ P ∩ C(A), then a =
ac ⇔ γa ≤ c. The restriction of the central cover mapping γ : A → P ∩ C(A)
to P is order preserving, it preserves arbitrary existing suprema in P , and if
p ∈ P and c ∈ P ∩ C(A), then γ(p ∧ c) = γp ∧ c [13, Theorem 6.7].
We note that, if P is centrally orthocomplete, then a family (pi)i∈I ⊆ P
is centrally orthogonal iff the family (γpi)i∈I of central covers is orthogonal,
and it follows that (pi)i∈I is centrally orthogonal iff it is pairwise centrally
orthogonal in the sense that, for i, j ∈ I, i 6= j implies that the pair consist-
ing of pi and pj is centrally orthogonal. The following lemma and theorem
address the issue of how these notions relativize to an interval P [0, p].
4.3 Lemma. Let p ∈ P , let (pi)i∈I be a family of projections in P [0, p], and
suppose that (pi)i∈I is centrally orthogonal in P . Then (pi)i∈I is centrally
orthogonal in P [0, p].
7
Proof. By hypothesis, there exists a pairwise orthogonal family (ci)i∈I of
central projections in P such that pi ≤ ci for all i ∈ I. Then (pci)i∈I is a
pairwise orthogonal family of central projections in P [0, p] and pi ≤ pci for
all i ∈ I.
4.4 Theorem. Suppose that P is centrally orthocomplete, let p ∈ P , let
(pi)i∈I be a family of projections in P [0, p], and suppose that at least one of
the following conditions holds: P is complete or p ∈ P ∩C(A). Then: (i) The
family (pi)i∈I is centrally orthogonal in P [0, p] iff it is centrally orthogonal in
P . (ii) P [0, p] is centrally orthocomplete.
Proof. Assume the hypotheses.
If P is complete, then it has the relative
center property, so the center of P [0, p] is {pc : c ∈ P ∩ C(A)}, and the same
conclusion holds if p ∈ P ∩ C(A).
(i) Suppose that (pi)i∈I is centrally orthogonal in P [0, p]. Then there
exists a family (ci)i∈I of central projections in P such that pci ⊥ pcj for
i, j ∈ I with i 6= j and pi ≤ pci for all i ∈ I. Then, for i 6= j, pi ≤ pci ≤ c ⊥
j ci ∈
P ∩ C(A) and pj ≤ pcj ≤ c ⊥
i cj; hence (pi)i∈I
is pairwise centrally orthogonal in P . Since P is centrally orthocomplete,
(pi)i∈I is centrally orthogonal in P . The converse follows from Lemma 4.3,
and (i) is proved. Since P is centrally orthocomplete, (ii) follows from (i).
i cj ∈ P ∩ C(A) with c ⊥
j ci ⊥ c ⊥
Let c1, c2, ..., cn ∈ P ∩ C(A) be a finite sequence of central projections
with ci ⊥ cj for i 6= j and c1 + c2 + · · · + cn = 1. Then P is the (internal)
direct sum of the OMLs P [0, ci], in symbols
P = P [0, c1] ⊕ P [0, c2] ⊕ · · · ⊕ P [0, cn],
in the sense that (1) every projection p ∈ P can be written uniquely as an
orthogonal sum p = p1 + p2 + · · · + pn = p1 ∨ p2 ∨ · · · ∨ pn with pi ∈ P [0, ci]
for i = 1, 2, ..., n and (2) all operations and relations for P can be computed
"coordinatewise" in the obvious sense. This direct sum decomposition of P is
reflected by a corresponding direct sum decomposition A = c1A ⊕ c2A ⊕ · · · ⊕
cnA of the synaptic algebra A into the direct summands ciA = ciAci = Aci,
where again every a ∈ A can be written uniquely as a = a1 + a2 + · · · + an
with ai ∈ ciA for i = 1, 2, ..., n and all synaptic operations and relations can
be computed "coordinatewise." In this case, P is isomorphic as an OML to
the cartesian product P [0, c1] × P [0, c2] × · · · × P [0, cn] and A is isomorphic
as a synaptic algebra to c1A × c2A × · · · × cnA.
8
We note that, if c ∈ P ∩ C(A), then P = P [0, c] ⊕ P [0, c⊥] and A =
cA ⊕ c⊥A. Thus, the direct summands of P (respectively, of A) are of the
form P [0, c] (respectively, cA) for central projections c ∈ P ∩ C(A).
The OML P , is called irreducible, and the synaptic algebra is said to be
a factor, iff P ∩ C(A) = {0, 1}. Thus A is a factor iff it admits no nontrivial
direct-sum decomposition. It can be shown that A is a factor iff the center
C(A) is the set of all real multiples of the unit element 1.
By regarding P as an effect algebra, we obtain the following.
4.5 Theorem ([9, Theorem 6.14]). Suppose that P is centrally orthocom-
plete, let (pi)i∈I be a centrally orthogonal family in P with p := Wi∈I pi, and
let X be the cartesian product X :=×i∈IP [0, pi] organized into an OML
with coordinatewise operations and relations. Define the mapping Φ : X →
P [0, p] by Φ((ei)i∈I) := Wi∈I ei for every (ei)i∈I ∈ X. Then Φ is an OML-
isomorphism of X onto P [0, p] and for q ∈ P [0, p], Φ−1(q) = (q ∧ γpi)i∈I =
(q ∧ pi)i∈I .
5 Symmetries and equivalence of projections
By a symmetry in A we mean an element s ∈ A such that s2 = 1 [13].
Two projections p, q ∈ P are said to be exchanged by a symmetry s ∈ A iff
sps = q, or equivalently, iff sqs = p. We note that p and q are exchanged by
a symmetry s ∈ A iff p⊥ = 1 − p and q⊥ = 1 − q are exchanged by s.
5.1 Theorem. Let p, q ∈ P . Then: (i) If p and q are exchanged by a
symmetry in A, then p and q are strongly perspective in P . (ii) If p and q are
strongly perspective in P , then p and q are perspective in P . (iii) If p and q
are perspective in P , then there are symmetries s, t ∈ A such that tspst = q.
Proof. Part (i) follows from [13, Theorem 5.11], (ii) is a consequence of
Lemma 3.3 (ii), and (iii) follows from [13, Theorem 5.12 (i)].
5.2 Corollary. Let p, q ∈ P with p ⊥ q. Then the following conditions are
mutually equivalent: (i) There are symmetries s, t ∈ A such that tspst = q.
(ii) p and q are exchanged by a symmetry in A. (iii) p and q are strongly
perspective in P . (iv) p and q are perspective in P .
Proof. That (i) ⇒ (ii) follows from [13, Theorem 5.12 (ii)] and (ii) ⇒ (iii)
⇒ (iv) ⇒ (i) follows from Theorem 5.1.
9
5.3 Lemma. Let r ∈ P , let p, q ∈ P [0, r]. Then p and q are exchanged by a
symmetry in A iff p and q are exchanged by a symmetry in rAr.
Proof. Let p, q ∈ P [0, r], let s be a symmetry in A with sps = q, and put
t := (p ∨ q)s(p ∨ q). Clearly, tpt = q, tqt = p and rt = tr = t. Also, since
s(p ∨ q)s = sps ∨ sqs = q ∨ p = p ∨ q, it follows that t2 = p ∨ q, whence
t3 = (p ∨ q)t = t. Put s1 := t + r − t2. Then s1 = rs1r ∈ rAr and s 2
1 = r,
whence s1 is a symmetry in rAr. Moreover, as rp = pr = p and t2p = pt2 = p,
we have s1ps1 = tpt = q. Conversely, by a straightforward calculation, if p
and q are exchanged by a symmetry u in rAr, then s := u + 1 − r is a
symmetry in A that exchanges p and q.
5.4 Theorem (Generalized Comparability). Suppose that P is complete and
let e, f ∈ P . Then there exists a symmetry s ∈ S and a central projection
c ∈ P ∩ C(A) such that secs ≤ f c, sf c⊥s ≤ ec⊥, se⊥c⊥s ≤ f ⊥c⊥, and
sf ⊥cs ≤ e⊥c.
Proof. Since P is complete, [13, Theorem 8.6] applies, so there exists c ∈
P ∩C(A) and a symmetry s ∈ S such that secs ≤ f c and sf c⊥s ≤ ec⊥. Thus,
f c⊥ ≤ sec⊥s, so se⊥c⊥s = s(c⊥ − ec⊥)s = c⊥ − sec⊥s ≤ c⊥ − f c⊥ = f ⊥c⊥.
By a similar computation, sf ⊥cs ≤ e⊥c.
If x = snsn−1 · · · s1 ∈ R is a finite product of symmetries sn, sn−1, ..., s1 ∈
A then we define x∗ := s1s2 · · · sn ∈ R to be the product of the same sym-
metries, but in the reverse order. We note that xx∗ = x∗x = 1 and, for any
a ∈ A, xax∗ ∈ A.
Let p, q ∈ P . Then by definition, p and q are equivalent, in symbols,
p ∼ q, iff there is a finite sequence of projections e1, e2, ..., en ∈ P such that
p = e1, q = en, and for each i = 1, 2, ..., n − 1, the projections ei and ei+1 are
exchanged by a symmetry si ∈ A. Clearly, p ∼ q iff there is a finite product
x ∈ R of symmetries in A such that q = xpx∗.
5.5 Lemma. If p, q ∈ P , then p ∼ q iff p and q are projective in P .
Proof. As a consequence of parts (i) and (ii) of Theorem 5.1, if p ∼ q, then
p and q are projective, and the converse follows from Theorem 5.1 (iii).
5.6 Lemma. Let p, q ∈ P and let x be a finite product of symmetries in A.
Then: (i) For a ∈ A, the mapping a 7→ xax∗ is a linear, order, and Jordan
automorphism of A. (ii) For p ∈ P , p ∼ xpx∗ and the mapping p 7→ xpx∗ is
10
an OML-automorphism of P . (iii) If p ∈ P , then for r ∈ P [0, p], the mapping
r 7→ xrx∗ is an OML-isomorphism of P [0, p] onto P [0, xpx∗]. (iv) If p, q ∈ P
and p ∼ q, then P [0, p] is isomorphic as an OML to P [0, q].
Proof. Parts (i) and (ii) follow from [13, Theorem 5.3 (i) and (ii)]. Parts (iii)
and (iv) follow from (ii).
The projections p and q are said to be related iff there are nonzero sub-
projections 0 6= p1 ≤ p and 0 6= q1 ≤ q such that p1 ∼ q1; otherwise they
are unrelated. If there exists a projection q1 ≤ q such that p ∼ q1, we say
that p is subequivalent to q, in symbols, p (cid:22) q. A projection h ∈ P is called
invariant iff it is unrelated to its orthocomplement h⊥.
5.7 Lemma. Suppose that P is centrally orthocomplete and let p, q, h ∈ P .
Then: (i) If p ∼ q, then γp = γq. (ii) If p (cid:22) q, then γp ≤ γq. (iii) h is
invariant iff it is central. (iv) If P is complete, then γp = W{q ∈ P : q (cid:22) p}.
(v) If γp ⊥ γq, then p and q are unrelated. (vi) If P is complete, then p and
q are unrelated iff γp ⊥ γq.
Proof. Assume that P is centrally orthocomplete, so the central cover map-
ping γ exists. To prove (i), it will be sufficient to prove that, if p and q are
exchanged by a symmetry s ∈ A, then γp = γq. So assume that sps = q.
Since γp ∈ C(A), it follows that qγp = spsγp = sp(γp)s = sps = q, whence
γq ≤ γp. Likewise, γp ≤ γq, and (i) is proved. Part (ii) is an immediate
consequence of (i) and the fact that, for e, f ∈ P , e ≤ f ⇒ γe ≤ γf . Part
(iii) follows from [13, Theorem 7.5], (iv) is a consequence of [13, Theorem
7.7], (v) follows from [13, Corollary 7.6], and (vi) follows from [13, Corollary
7.8].
We denote the set of natural numbers by N := {1, 2, 3, ...}.
5.8 Lemma (Cf.
[27, Lemma 21]). If e1, e2, e3, ... is an infinite orthogonal
sequence of projections and if ei and ei+1 are exchanged by a symmetry si ∈ A
for all i ∈ N, then for all i, j ∈ N, the projections ei and ej are exchanged by
a symmetry in A.
Proof. It will be sufficient to prove by induction on n ∈ N that, if i ∈ N and
i ≤ n, then ei and en are exchanged by a symmetry in A. For n = 1, this is
obvious. Assume that it is true for n and suppose that i ∈ N with i ≤ n + 1.
We have to prove that ei and en+1 are exchanged by a symmetry in A.
11
Obviously, we can assume that i ≤ n, whence by the induction hypothesis,
there is a symmetry s ∈ A such that seis = en. But snensn = en+1, so
snseissn = en+1. Since ei ⊥ en+1, we infer from Corollary 5.2 that ei and
en+1 are exchanged by a symmetry in A.
5.9 Lemma. Suppose s and t are symmetries in A, f ∈ P , and tsf st <
f . Define f1 := f and fn := (ts)n−1f (st)n−1 for 2 ≤ n ∈ N. Then:
(i) f = f1 > f2 > f3 > · · · . (ii) The sequence e1, e2, e3, ... ∈ P [0, f ] de-
fined by en := fn − fn+1 > 0 for n ∈ N is orthogonal and for all i, j ∈ N, ei
and ej are exchanged by a symmetry in A.
Proof. For n ∈ N, we have fn+1 = tsfnst.
(i) We prove by induction on n ∈ N that fn+1 < fn. For n = 1, we have
f2 = tsf st < f = f1. Assume by the induction hypothesis that n > 1 and
fn < fn−1. Then fn+1 = tsfnst < tsfn−1st = fn.
(ii) Suppose i, j ∈ N with i < j. Then fj ≤ fi+1 and fj−1 ≤ fi, whence
ei + ej ≤ ei + ej + fi+1 − fj = fi − fj+1 ≤ fi ≤ 1, and it follows that
ei ⊥ ej. Also, for all n ∈ N, tsenst = ts(fn − fn+1)st = tsfnst − tsfn+1st =
fn+1 − fn+2 = en+1, and since en ⊥ en+1, Corollary 5.2 implies that en and
en+1 are exchanged by a symmetry. Therefore, by Lemma 5.8, ei and ej are
exchanged by a symmetry for all i, j ∈ N.
6 Type-determining sets, orthodensity, and
faithful projections
Material in this section is adapted from [10, §3, §4].
6.1 Standing Assumption. Henceforth in this article, we assume that the
OML P is centrally orthocomplete. Therefore the center P ∩ C(A) is a com-
plete boolean algebra and the central cover mapping γ : A → P ∩ C(A) exists.
6.2 Definition. Let Q ⊆ P . Then:
(1) The set of all suprema of centrally orthogonal families of projections in
Q is denoted by [Q]. We understand that [∅] = {0}.
(2) Qγ := {q ∧ c : q ∈ Q, c ∈ P ∩ C(A)}.
(3) Q↓ := Sq∈Q P [0, q]. If Q↓ ⊆ Q 6= ∅, then Q is called an order ideal.
12
(4) Q is type determining (TD) iff [Q] ⊆ Q and Qγ ⊆ Q.
(5) Q is strongly type determining (STD) iff [Q] ⊆ Q and Q↓ ⊆ Q.
(6) Q is projective iff for all q ∈ Q, if q is projective to p ∈ P , then p ∈ Q.
(7) Q is orthodense in P iff every projection in P is the supremum of an
orthogonal family of projections in Q.
(8) Q is an OML-ideal iff Q is an order ideal and p, q ∈ Q ⇒ p ∨ q ∈ Q.
An OML-ideal is a p-ideal iff it is projective [19, p. 75].
We note that Q ⊆ P is TD (respectively, STD) iff Q = [Q] = Qγ (re-
spectively, iff Q = [Q] = Q↓). Clearly, STD ⇒ TD, and the intersection of
TD subsets (respectively, STD subsets, projective subsets) of P is again TD
(respectively, STD, projective). Since [∅] = {0}, 0 belongs to every TD set.
If p ∈ P , then the p-interval P [0, p] is both an OML ideal and an STD
subset of P , but it is projective iff p ∈ C(A). Also, the center P ∩ C(P ) is a
projective TD subset of P , but it is STD iff P is boolean. By [10, Theorem
4.1] [Qγ] is the smallest TD subset of P that contains Q, and [Q↓] is the
smallest STD subset of P that contains Q.
Since two projections are projective iff they are equivalent (Lemma 5.5),
it follows that Q ⊆ P is projective iff, for all p ∈ P , p ∼ q ∈ Q ⇒ p ∈ Q.
Clearly, Q is projective iff, for every symmetry s ∈ A, we have sQs ⊆ Q.
6.3 Lemma. Let p ∈ P and suppose that one of the following conditions
holds: P is complete or p ∈ P ∩ C(A). Then, if Q ⊆ P is TD (respec-
tively, STD, projective) it follows that Q ∩ P [0, p] is TD (respectively, STD,
projective) both in P and in the projection lattice P [0, p] of pAp.
Proof. Assume the hypotheses. Then, as in the proof of Theorem 4.4, the
center of P [0, p] is {pc : c ∈ P ∩ C(A)}; moreover, by Theorem 4.4, P [0, p] is
centrally orthocomplete and a family in Q ∩ P [0, p] is centrally orthogonal in
P [0, p] iff it is centrally orthogonal in P . Consequently, Q ∩ P [0, p] is closed
under the formation of suprema of centrally orthogonal families. If d belongs
to the center of P [0, p], then d = pc for some c ∈ P ∩ C(A), whence, for any
q ∈ Q∩P [0, p], we have qd = qpc = qc ∈ Q∩P [0, p], and therefore Q∩P [0, p]
is TD in P [0, p]. If Q is STD, it is clear that Q ∩ P [0, p] is STD in P [0, p].
Suppose Q is projective, let q ∈ Q ∩ P [0, p], and suppose q and a projection
r ∈ P [0, p] are exchanged by a symmetry in pAp. Then, by Lemma 5.3, q
and r are exchanged by a symmetry in A, whence r ∈ Q ∩ P [0, p].
13
6.4 Definition. A nonempty class L of OMLs is called an OML type class
iff the following conditions are satisfied: (1) If L ∈ L and c belongs to the
center of L, then L[0, c] ∈ L. (2) L is closed under the formation of arbitrary
(3) If L1 and L2 are isomorphic OMLs and L1 ∈ L,
cartesian products.
then L2 ∈ L. If, in addition to (2) and (3), L satisfies (1′) if L ∈ L, then
p ∈ L ⇒ L[0, p] ∈ L, then L is called a strong OML type class.
Some examples of strong OML type classes are the following: The class of
all boolean algebras, all modular OMLs, all complete OMLs, all σ-complete
OMLs, and all atomic OMLs. Obviously, the intersection of (strong) OML
type classes is again a (strong) OML type class. For instance, the class of
all complete modular OMLs is a strong OML type class. The class of all
locally modular OMLs provides an example of an OML type class that is not
strong; however the class of all complete locally modular OMLs is a strong
OML type class.
6.5 Theorem. If Q is a OML type class (respectively, a strong OML type
class), then Q := {q ∈ P : P [0, q] ∈ Q} is a projective TD set (respectively,
a projective STD set).
Proof. Assume that Q is a OML type class and Q := {q ∈ P : P [0, q] ∈
Q}. Suppose that (qi)i∈I is a centrally orthogonal family in Q. Since P is
centrally orthocomplete (Assumption 6.1), q := Wi∈I qi exists in P . For every
i ∈ I, P [0, qi] ∈ Q, whence X :=×i∈IP [0, qi] ∈ Q. By Theorem 4.5, X is
isomorphic as an OML to P [0, q], so P [0, q] ∈ Q, and therefore q ∈ Q. Thus,
[Q] ⊆ Q.
Let q ∈ Q and c ∈ P ∩ C(A). Then P [0, q] ∈ Q and, q ∧ c = qc belongs
to the center of P [0, q], whence P [0, qc] = (P [0, q])[0, qc] ∈ Q, and so qc ∈ Q.
This proves that Qγ ⊆ Q, so Q is a TD-set. To prove that Q is projective, let
s ∈ A be a symmetry. Then P [0, sqs] is isomorphic as an OML to P [0, q] ∈ Q,
whereupon P [0, sqs] ∈ Q, and we have sqs ∈ Q.
To complete the proof, suppose that Q is a strong OML type class, let
q ∈ Q and suppose p ∈ P [0, q]. Then P [0, p] ∈ Q, and it follows that
(P [0, p])[0, q] = P [0, q] ∈ Q, whence p ∈ Q.
If Q ⊆ P , we understand that γ(Q) := {γq : q ∈ Q}. The following
theorem is an adaptation to our present context of [10, Theorem 4.5 and
Corollary 4.6].
14
6.6 Theorem. Let Q ⊆ P be a TD set. Then: (i) Q ∩ γ(Q) = Q ∩ C(A) ⊆
γ(Q) ⊆ P ∩ C(A). (ii) There is a unique central projection cQ ∈ P ∩ C(A)
such that γ(Q) = (P ∩ C(A))[0, cQ]. (iii)There is a unique central projection
cQ∩C(A) ∈ P ∩C(A) such that Q∩γ(Q) = Q∩C(A) = (P ∩C(A))[0, cQ∩C(A)].
(iv) Both γ(Q) and Q ∩ γ(Q) = Q ∩ C(A) are TD subsets of P .
6.7 Definition (Cf. [10, Definition 4.7]). Let Q be a TD subset of P . Then
the central projection cQ in Theorem 6.6 (ii) is called the type-cover of Q,
and the central projection cQ∩C(A) in Theorem 6.6 (iii) is called the restricted
type-cover of Q.
The type cover cQ and the restricted type cover cQ∩C(A) will play significant
roles in Sections 8 and 9 below.
6.8 Lemma ([10, Lemma 4.8]). Let Q ⊆ P be a TD set. Then: (i) cQ is the
largest projection in γ(Q) and every central subprojection of cQ belongs to
γ(Q). (ii) cQ∩C(A) is the largest central projection in Q. (iii) cQ∩C(A) ≤ cQ.
(iv) The smallest central projection c ∈ P ∩ C(A) such that Q ⊆ P [0, c]
(v) The smallest central projection d ∈ P ∩ C(A) such that
is c = cQ.
Q ∩ C(A) ⊆ P [0, d] is d = cQ∩C(A). (vi) If c ∈ P ∩ C(A), then c ⊥ cQ iff
Q ∩ P [0, c] = {0}. (vii) If c ∈ P ∩ C(A), then c ⊥ cQ∩C(A) iff Q ∩ (P ∩
C(A))[0, c] = {0}.
6.9 Definition. A projection f ∈ P is faithful iff γf = 1.
Clearly, f ∈ P is faithful iff the only central projection c ∈ P ∩ C(A) such
that f ≤ c is c = 1. The next lemma clarifies how faithfulness relativizes to
a direct summand of P .
6.10 Lemma ([10, Lemma 3.5]). Let c ∈ P ∩ C(A) and let f ∈ P [0, c].
Then the following conditions are mutually equivalent: (i) f is faithful in the
projection lattice P [0, c] of cA. (ii) γf = c. (iii) γ(P [0, f ]) is the center of
P [0, c]. (iv) f has a nonzero component in every nonzero direct summand of
P [0, c], i.e., if 0 6= d ∈ (P ∩ C(A))[0, c], then f d 6= 0.
6.11 Lemma. Let Q ⊆ P and let c := W γ(Q). Then Q ⊆ P [0, c] and the
following conditions are mutually equivalent: (i) If p ∈ P , d ∈ γ(Q), and
pd 6= 0, then Q ∩ P [0, p] 6= {0}. (ii) If 0 6= p ∈ P [0, c], then Q ∩ P [0, p] 6= {0}.
(iii) Q is orthodense in P [0, c].
15
Proof. If q ∈ Q, then q ≤ γq ≤ c, so Q ⊆ P [0, c]. The rest of the lemma
follows from [12, Lemma 5.3] by taking η = γ.
6.12 Theorem (Cf. [27, Propositions 13 and 16]). Suppose that P is com-
plete, Q↓ ⊆ Q ⊆ P , Q is projective, and c = W{γq : q ∈ Q}. Then (i)
Q ⊆ P [0, c]. (ii) Q is orthodense in P [0, c]. (iii) c = W Q. (iv) If Q is TD,
then c = cQ ∈ γ(Q).
Proof. Assume the hypotheses. Part (i) follows from Lemma 6.11. To prove
(ii), it will be sufficient to show that part (i) of Lemma 6.11 holds. Thus,
suppose that p ∈ P , q ∈ Q, and p ∧ γq 6= 0. Since P is complete, [13,
Corollary 7.8] applies, whence p and q are related, i.e., there are nonzero
subprojections 0 6= p1 ≤ p and 0 6= q1 ≤ q such that p1 ∼ q1. As q1 ∈ Q↓ ⊆ Q
and Q is projective, it follows that p1 ∈ Q, and (ii) is proved. By (ii), c is
the supremum of an orthogonal family in Q, by (i), c is an upper bound for
Q, whence (iii) holds. If Q is TD, then by Lemma 6.8 (i), cQ is the largest
projection in γ(Q), whence cQ = W γ(Q) = c.
6.13 Corollary. Suppose that P is complete, Q ⊆ P , Q is projective, and
Q is STD. Then cQ = W Q, Q ⊆ P [0, cQ], and Q is orthodense in P [0, cQ].
6.14 Theorem. Let Q ⊆ P be TD and let 0 6= p ∈ P . Then the following
two conditions are equivalent: (i) There exists 0 6= q ∈ Q ∩ P [0, p] such that
γq = γp. (ii) For all d ∈ P ∩ C(A), if pd 6= 0, then Q ∩ P [0, pd] 6= {0}.
Proof. (i) ⇒ (ii). Assume (i) and let d ∈ P ∩C(A) with pd 6= 0. By (i), there
exists 0 6= q ∈ Q ∩ P [0, p] with γq = γp. Put q0 := qd = q ∧ d ≤ p ∧ d = pd.
Then q0 ∈ Qγ ⊆ Q, and since d ∈ P ∩ C(A), we have 0 6= pd ≤ γ(pd) =
(γp)d = (γq)d = γ(qd) = γq0, whence q0 6= 0.
(ii) ⇒ (i). Assume (ii), let (qi)i∈I be a maximal centrally orthogonal
family in Q ∩ P [0, p], and put q := Wi∈I qi ∈ [Q] ⊆ Q. Then q ≤ p, so
γq ≤ γp. Taking d = 1 in (ii), we find that Q ∩ P [0, p] 6= {0}, whence
q 6= 0.
If γq = γp, then we are done, so, we assume that γq < γp and
this time we put d = γp − γq = γp(γq)⊥ in (ii). Then, as p ≤ γp, we
have pd = p(γq)⊥, whence, if pd = 0, then p ≤ γq, so γp ≤ γq ≤ γp,
contradicting γq < γp. Therefore, pd 6= 0, and it follows from (ii) that there
exists 0 6= q0 ∈ Q ∩ P [0, pd]. But then, q0 ≤ p and q0 ≤ d ≤ (γq)⊥ ≤ (γqi)⊥
for all i ∈ I, contradicting the maximality of (qi)i∈I.
16
6.15 Corollary. Let c ∈ P ∩ C(A) and let Q be a TD subset of P . Then the
following two conditions are equivalent: (i) c ∈ γ(Q). (ii) Q has a nonzero
intersection with every nonzero direct summand of P [0, c].
Proof. Assume the hypotheses. If c = 0, then (i) and (ii) are both true, so
we assume that c 6= 0 and put p := c in Theorem 6.14. Then conditions
(i) and (ii) in Theorem 6.14 are equivalent to conditions (i) and (ii) in the
corollary.
6.16 Lemma (Cf. [27, Proposition 15]). Suppose that P is complete, Q ⊆ P ,
Q is projective, Q is STD, c ∈ γ(Q), and 0 6= p ∈ P [0, c]. Then there exists
0 6= q ∈ Q ∩ P [0, p] with γq = γp.
Proof. Assume the hypotheses and suppose that d ∈ P ∩ C(A) with pd 6= 0.
By Theorem 6.14, it will be sufficient to prove that Q∩P [0, pd] 6= {0}. But, by
Theorem 6.6 (ii), c ≤ cQ, whence, if 0 6= pd ∈ P [0, c], then 0 6= pd ∈ P [0, cQ],
and it follows from Corollary 6.13 that pd is the supremum of an orthogonal
family in Q. Therefore, since pd 6= 0, it follows that Q ∩ P [0, pd] 6= {0}.
7 Abelian, modular, locally modular, and
complete projections
In this section we study some important examples of TD and STD subsets
of P . Many of the results in this section are generalizations to a synaptic
algebra of results due to D. Topping for JW-algebras [27]. Often the proofs
of these results are more or less the same as Topping's proofs, but we include
these proofs here in the interest of a more coherent account. The assumption
that P is centrally orthocomplete is still in force.
7.1 Definition. Let p ∈ P .
(1) p is abelian (also called boolean [10, p. 1551] iff P [0, p] is a boolean
(We shall regard P [0, 0] = {0} as a "degenerate" boolean
algebra.
algebra, hence 0 is an abelian projection in A.) We denote the set of
all abelian projections in P by B.
(2) p is modular iff P [0, p] is a modular OML. We denote the set of all
modular projections in P by M.
17
(3) p is locally modular iff P [0, p] is a locally modular OML. We denote the
set of all locally modular projections in P by M0.
(4) p is complete iff P [0, p] is a complete OML. We denote the set of all
complete projections in P by T .
7.2 Theorem. (i) B ⊆ M ⊆ M0. (ii) The sets B, M , and T are projective
(iv) If c ∈ P ∩ C(A), then
STD sets.
c ∈ M0 ⇔ c ∈ γ(M).
(iii) M0 is a projective TD set.
Proof. Part (i) is obvious. Since the class B of all boolean OMLs, the class
M of all modular OMLs, and the class T of all complete OMLs are strong
OML type classes and the class M0 of all locally modular OMLs is an OML
type class, (ii) and (iii) follow from Theorem 6.5, and part (iv) follows from
Corollary 6.15.
7.3 Lemma. Let p, q ∈ P . Then: (i) p ∈ B iff pAp is a commutative
synaptic algebra. : (ii) If p ∈ B, then p ∧ q is an abelian projection in the
(iii) If p ∈ P [0, q], then p ∈ B iff p is an abelian
synaptic algebra qAq.
projection in the synaptic algebra qAq.
Proof. (i) By [13, Theorem 4.5], pAp is a commutative synaptic algebra iff
its lattice of projections P [0, p] is a boolean algebra.
(ii) Suppose p ∈ B, i.e., P [0, p] is boolean. Then q ∧ p ∈ P [0, q], which
is the lattice of projections in qAq, and (P [0, q])[0, q ∧ p] = P [0, q ∧ p]. But,
since P [0, p] is boolean, so is the sublattice P [0, q ∧ p] ⊆ P [0, p], whence q ∧ p
is abelian in qAq.
(iii) If p is abelian in qAq, then (P [0, q])[0, p] = P [0, p] is boolean, whence
p ∈ B. Conversely, suppose that p ∈ P [0, q], i.e., p ≤ q. Then, if p ∈ B, it
follows from (ii) that p = p ∧ q is abelian in qAq.
7.4 Theorem (Cf. [27, Theorem 11]). Let p ∈ P and consider the following
two conditions: (i) Every orthogonal family of nonzero projections in P [0, p],
any two of which are exchanged by a symmetry in A, is necessarily finite.
(ii) p ∈ M . Then (i) ⇒ (ii), and if p ∈ T , then (ii) ⇒ (i).
Proof. To prove that (i) ⇒ (ii), it will be sufficient to show that if (ii) fails,
then (i) fails. So assume that P [0, p] is not modular. Then by Theorem 3.4,
there exist projections e, f ∈ P [0, p] such that e < f and e is perspective
to f in P [0, p]. Thus by Lemma 3.3 (i), e is perspective to f in P , whence
18
by Theorem 5.1 (iii), there are symmetries s, t ∈ A such that stets = f .
Therefore, tsf st = e < f , and by Lemma 5.9 (ii), (i) fails.
Conversely, assume that P [0, p] is complete, that (ii) holds, and that
(ei)i∈I is an orthogonal family of nonzero projections in P [0, p] any two of
which are exchanged by a symmetry in A. Therefore, by Corollary 5.2, any
two projections in (ei)i∈I are strongly perspective in P , whence by Lemma
3.3, they are strongly perspective, hence perspective, in P [0, p]. Since the
OML P [0, p] is modular and complete, it follows from Theorem 3.5 (iv) that
(ei)i∈I is finite.
[27, Lemma 23]). Suppose that p ∈ T , but p /∈ M . Then
7.5 Lemma (Cf.
there is a projection e ∈ P [0, p] with the following properties: (i) e is the
supremum of an infinite sequence of nonzero projections in P [0, p] any two
of which are exchanged by a symmetry in A. (ii) There is a symmetry s ∈ A
with ses ∈ P [0, p] and ses ⊥ e.
Proof. Assume the hypotheses. Then by Theorem 7.4, there is an infinite
sequence e1, e2, e3, ... of nonzero projections in P [0, p], any two of which are
exchanged by a symmetry in A, whence also by a symmetry in pAp (Lemma
5.3). Putting e := W∞
n=1 e2n, we have (i). To prove (ii), we work in the
synaptic algebra pAp and its complete OML P [0, p] of projections. Let f :=
W∞
n=1 e2n−1. Then e ⊥ f , whence by [13, Theorem 5.15] (a weak form of
additivity for exchangeability by symmetries), there is a symmetry t ∈ pAp
such that tet = f . By Lemma 5.3 again, there is a symmetry s ∈ A with
ses = f , and (ii) is proved.
7.6 Theorem (Cf. [27, Theorem 12]). (i) If p, q ∈ M and p ∨ q ∈ T , then
p ∨ q ∈ M . (ii) If P is complete, then M is both a projective STD set and a
p-ideal in P .
Proof. (i) Assuming the hypothesis of (i), we have to prove that P [0, p ∨ q]
is modular; hence we may drop down to the synaptic algebra (p ∨ q)A(p ∨ q)
with complete projection lattice P [0, p ∨ q]. Thus, changing notation, we can
(and do) assume that P is complete, that p, q ∈ M with p ∨ q = 1, and we
have to prove that P is modular. By [13, Theorem 5.9 (ii)] (the symmetry
parallelogram law ) p⊥ = 1 − p = (p ∨ q) − p is exchanged by a symmetry in
A with the modular projection q − (p ∧ q) ≤ q, so p⊥ is modular.
Now, aiming for a contradiction, we assume that P is not modular. There-
fore by Lemma 7.5 (with p=1), there is a projection e ∈ P such that e is
19
the supremum of an infinite sequence of nonzero projections in P any two of
which are exchanged by a symmetry in A, and there is a symmetry t ∈ A
with tet ⊥ e. Applying Theorem 5.4 to the pair e, p, we find that there is a
symmetry s ∈ A and a central projection c ∈ P ∩ C(A) such that secs ≤ pc
and se⊥c⊥s ≤ p⊥c⊥. From the latter inequality and the fact that p⊥ ∈ M,
we infer that e⊥c⊥ ∈ M. But tec⊥t = tetc⊥ ≤ e⊥c⊥, whence ec⊥ ∈ M.
Moreover, the pair of modular projections ec and ec⊥ is centrally orthogonal,
hence e = ec + ec⊥ ∈ M, contradicting Theorem 7.4.
(ii) Part (ii) follows immediately from (i).
7.7 Theorem (Cf.
p, q ∈ P . Then:
[27, Corollary 21]). Assume that P is complete and let
(i) If p ∈ M and p ∼ q, then q ∈ M , there is a projection r ∈ M such that
p, q ∈ P [0, r], and p is perspective to q in P [0, r].
(ii) If p ∈ M p ∼ q, then q ∈ M and p and q are perspective in P .
(iii) On the set M , perspectivity is transitive.
(iv) If q ≤ p ∈ M and q ∼ p, then q = p (i.e., p is finite [27, p.23]).
(v) If p, q ∈ M , then p ∼ q iff p and q are exchanged by a symmetry in A.
(vi) If p, q ∈ M , p (cid:22) q, and q (cid:22) p, then p ∼ q.
Proof. (i) Assume p ∈ M and p ∼ q. Since M is projective, q ∈ M. Also
there exist projections p = e1, e2, ..., en = q such that ei is exchanged by a
symmetry in A with ei+1 for i = 1, 2..., n − 1. Since p ∈ M, it follows from
Lemma 5.6 (iii) that e1, e2, ...en ∈ M, and by Theorem 7.6, r := e1 ∨ e2 ∨ · · ·∨
en ∈ M. By Lemma 5.3, for i = 1, 2, ..., n − 1, ei is exchanged with ei+1 by
a symmetry in rAr. Since P is complete, so is P [0, r]; hence, we may apply
Theorem 3.5 (ii) to rAr and its complete modular projection lattice P [0, r]
and infer that p = e1 is perspective to q = en in P [0, r].
(ii) By (i) and Lemma 3.3, p and q are perspective in P .
(iii) Suppose that p, q, r ∈ M with p perspective to q and q perspective
to r in P . Then by Theorem 5.1 (iv), p ∼ q and q ∼ r, so p ∼ r, and by (ii),
p is perspective to r in P .
(iv) Assume that q ≤ p ∈ M and q ∼ p. By (i) there exists r ∈ M such
that q ≤ p ∈ P [0, r] and p is perspective to q in P [0, r]; hence p = q by
Theorem 3.4 applied to the modular OML P [0, r].
20
(v) Suppose that p ∈ M p ∼ q. Then q ∈ M and applying Theorem 5.4
we infer that there is a symmetry s ∈ A and a central projection c ∈ P ∩C(A)
such that spcs ≤ qc and sqc⊥s ≤ pc⊥. Since p ∼ q, there is a finite product
of symmetries x such that xpx∗ = q. Thus, spcs ≤ qc = xpcx∗, whence
e := x∗spcsx ≤ pc with sxex∗s = pc. Therefore, e ≤ pc with e ∼ pc, and
since pc ∈ M, e = pc by (iv), and it follows that spcs = xex∗ = xpcx∗ = qc.
Likewise, f := xsqc⊥sx∗ ≤ xpc⊥x∗ = qc⊥ with sx∗f xs = qc⊥, and we deduce
that f = qc⊥, whence sqc⊥s = x∗f x = x∗qc⊥x = pc⊥, so spc⊥s = qc⊥.
Consequently, sps = spcs + spc⊥s = qc + qc⊥ = q. Conversely, if p and q are
exchanged by a symmetry, then p ∼ q.
(vi) By hypothesis, there are finite products of symmetries u and x such
that q1 := upu∗ ≤ q and p1 := xqx∗ ≤ p. Thus, xq1x∗ ≤ xqx∗ = p1 ≤ p
with xq1x∗ = xupu∗x∗ ∼ p. By (iv), xq1x∗ = p, and therefore q1 = x∗px.
Consequently, q = x∗p1x ≤ x∗px = q1, so q1 = q, whence p ∼ q.
Examination of the results in [27] required for Topping's proof of his
version of the type-I/II/III decomposition theorem for a JW-algebra [27,
Theorem 13] now shows that all of these results either have been obtained
above (often assuming that P is complete) or follow easily from the results
above. Therefore, we claim that our first project has been accomplished. We
now focus on our second project.
8 The fundamental direct-decomposition
theorem
The assumption that P is centrally orthocomplete is still in force.
8.1 Standing Assumption. In this section and the next, we assume that
Q is a TD subset of P .
We note that our subsequent results, apart from Theorem 9.5, do not
require completeness of the OML P , nor do they require that Q is STD. The
terminology in the following definition is borrowed from [27, pp. 28 -- 29].
8.2 Definition. Let c ∈ P ∩ C(A). Then:
(1) c is type-Q iff c ∈ Q.
(2) c is locally type-Q iff c ∈ γ(Q).
21
(3) c is purely non-Q iff no nonzero subprojection of c belongs to Q.
(4) c is properly non-Q iff no nonzero central subprojection of c belongs
to Q.
If c ∈ P ∩ C(A) and if c is type-Q (respectively, locally type-Q, purely
non-Q, etc.), one also says that the direct summand P [0, c] of P and the
direct summand cA of A are type-Q (respectively, locally type-Q, purely
non-Q, etc.).
We note that, by Theorem 7.2 (iv), for central projections c ∈ P ∩ C(A),
the notion of local modularity introduced in Definition 7.1 (3) is consistent
with Definition 8.2 (2), i.e., c ∈ M0 iff c is locally type-M.
8.3 Theorem ([10, Theorem 5.2]). Let c ∈ P ∩ C(A). Then:
(i) c is type-Q iff c ∈ Q ∩ γ(Q) = Q ∩ C(A) iff every central subprojection
of c belongs to Q ∩ C(A) iff c ≤ cQ∩C(A).
(ii) If Q is STD, then c is type-Q iff P [0, c] ⊆ Q.
(iii) c is locally type-Q iff every central subprojection of c belongs to γ(Q)
iff c ≤ cQ.
(iv) c is purely non-Q iff Q ∩ P [0, c] = {0} iff c ≤ (cQ)⊥.
(v) c is properly non-Q iff the only central projection in Q ∩ P [0, c] is 0 iff
c ≤ (cQ∩C(A))⊥.
8.4 Corollary. Let c, d ∈ P ∩ C(A). Then: (i) If c is type-Q, then c is
locally type-Q. (ii) If c is purely non-Q, then c is properly non-Q. (iii) If c is
both type-Q and properly non-Q, then c = 0. (iv) If c is both locally type-Q
and purely non-Q, then c = 0. (v) If c is type-Q (respectively, locally type-Q,
purely non-Q, properly non-Q), then so is c ∧ d. (vi) If both c and d are
type-Q (respectively, locally type-Q, purely non-Q, properly non-Q), then so
is c ∨ d.
8.5 Lemma ([10, Lemma 5.5]). (i) There exists a unique central projection
c, namely c = cQ, such that A = cA ⊕ c⊥A, cA is locally type-Q, and c⊥A
is purely non-Q; moreover, Q ⊆ P [0, cQ]. (ii) There exists a unique central
projection d, namely d = cQ∩C(A), such that A = dA ⊕ d⊥A, dA is type-Q,
and d⊥A is properly non-Q; moreover, Q ∩ C(A) ⊆ P [0, cQ∩C(A)].
22
The following theorem results from combining the direct decompositions
in parts (i) and (ii) of Lemma 8.5. We regard this theorem as the fundamental
direct-decomposition theorem for the synaptic algebra A.
8.6 Theorem ([10, Theorem 5.6]). Corresponding to the TD set Q, there
exist unique pairwise orthogonal central projections c1, c2 and c3, namely
c1 = cQ∩C(A), c2 = cQ ∧(cQ∩C(A))⊥, and c3 = (cQ)⊥, such that c1 + c2 + c3 = 1;
A = c1A ⊕ c2A ⊕ c3A;
c1A is type-Q; c2A is locally type-Q, but properly non-Q; and c3 is purely
non-Q. Moreover, Q ∩ C(A) = (P ∩ [C(A))[0, c1], Q ⊆ P [0, c1 + c2], and
(P ∩ C(A))[0, c2 + c3] ∩ Q = {0}.
9 The type-I/II/III decomposition theorem
The assumption that P is centrally orthocomplete is still in force.
9.1 Standing Assumption. In this section, we continue to assume that
Q ⊆ P is TD, and we also assume that K ⊆ P is TD and that Q ⊆ K.
Since Q ⊆ K, we have cQ ≤ cK and cQ∩C(A) ≤ cK∩γ(K).
9.2 Definition. Let c ∈ P ∩ C(A). Then, with respecct to the pair of TD
sets Q ⊆ K:
(1) c is type I iff it is locally type-Q.
(2) c is type II iff it is locally type K, but purely non-Q.
(3) c is type III iff it is purely non-K.
(4) c is type IK (respectively, type IIK) iff it is type I (respectively, type II)
and also type-K.
(5) c is type I eK (respectively, type II eK) iff it is type I (respectively, type II
and also properly non-K).
If c ∈ P ∩ C(A) and if c is type I (respectively, type II, type III, etc.), one
also says that the direct summand P [0, c] of P and the direct summand cA
of A are type I (respectively, type II, type III, etc.).
23
9.3 Lemma. Let c ∈ P ∩ C(A). Then the following conditions are mutually
equivalent: (i) c is type I. (ii) There is a projection q ∈ Q such that γq = c.
(iii) There is a projection q ∈ Q ∩ P [0, c] that is faithful in P [0, c]. (iv) Every
nonzero direct summand of P [0, c] contains a nonzero projection in Q. (v)
c ≤ cQ.
Proof. (i) ⇔ (ii) is the definition of c being locally type-Q, (ii) ⇔ (iii) follows
from Lemma 6.10, and (i) ⇔ (iv) ⇔ (v) follows from Theorem 8.3 (iii).
The following is the type-I/II/III decomposition theorem for synaptic al-
gebras. It is obtained by combining the fundamental direct-decomposition
theorems for Q and for K.
9.4 Theorem ([10, Theorem 6.4]). Corresponding to the pair of TD sets Q
and K with Q ⊆ K, there are unique pairwise orthogonal central projections
c I, c II and c III, namely c I = cQ, c II = cK ∧ (cQ)⊥, and c III = (cK)⊥, such
that c I + c II + c III = 1;
A = c IA ⊕ c IIA ⊕ c IIIA;
and c IA, cIIA, and c IIIA are of types I, II, and III, respectively. Moreover,
there are further decompositions
c IA = c IKA ⊕ c I eKA and c IIA = c IIKA ⊕ c II eKA,
where c IK, c I eK
and II eK , respectively; these decompositions are also unique; and
are central projections of types IK , I eK , IIK ,
, cIIK, and c II eK
c IK = cQ ∧ cK∩γ(K), c I eK = cQ ∧ (cK∩γ(K))⊥,
cIIK = cK∩γ(K) ∧ (cQ)⊥, c II eK = cK ∧ (cK∩γ(K))⊥ ∧ (cQ)⊥.
Furthermore, the type IK direct summand decomposes as
c IKA = c11A ⊕ c21A,
where c11 and c21 are central projections, c11 is type-Q (hence also of type-K),
and c21 is type-K, locally type-Q, but properly non-Q. The latter decomposi-
tion is also unique, and
c11 = cQ∩C(A),
c21 = cK∩γ(K) ∧ cQ ∧ (cQ∩C(A))⊥.
24
9.5 Theorem. With the notation of Theorem 9.4:
(i) Q ⊆ P [0, c I] and K ⊆ P [0, c I + c II].
(ii) If P is complete, Q is projective, and Q is STD, then c I = W Q and Q
is orthodense in P [0, c I].
(iii) If P is complete, K is projective, and K is STD, then c I + c II = W K
and K is orthodense in P [0, c I + c II].
Proof. (i) As c I = cQ, and c I + c II = cQ + (cK − cQ) = cK, (i) follows from
Lemma 6.8 (iv). In view of part (i), parts (ii) and (iii) follow from Corollary
6.13.
In Theorem 9.4, the unique five-fold direct-sum decomposition
A = c IKA ⊕ c I eKA ⊕ c IIKA ⊕ c II eKA ⊕ c IIIA
of A into direct summands of types IK, I eK, IIK, II eK and III is a generalization
of the classic type-I/II/III decomposition for a von Neumann algebra (see
Remark 9.6 below); moreover,the additional decomposition c IKA = c11A ⊕
c21A into direct summands of type-Q and of type-K, locally type-Q, but
properly non-Q yields a six-fold direct decomposition of A,
A = c11A ⊕ c21A ⊕ c I eKA ⊕ c IIKA ⊕ c II eKA ⊕ c IIIA.
Of course, if A is a factor, then it is of precisely one of these six types.
9.6 Remark. If R is a von Neumann algebra and A is the synaptic algebra
of all self-adjoint elements of A, then one obtains the classic type-I/II/III
decomposition of A (and also of R) by taking Q = B, the TD set of abelian
projections in A, and taking K to be the set of all finite projections in A.
9.7 Remark. If A is a JW-algebra, regarded as a synaptic algebra, then one
obtains Topping's version of a type-I/II/III decomposition [27, Theorem 13]
by taking Q = B and K = M.
25
References
[1] Alfsen, E.M., Compact Convex Sets and Boundary Integrals, Springer-
Verlag, New York, 1971, ISBN 0-387-05090-6.
[2] Beran, L., Orthomodular Lattices, An Algebraic Approach, Mathematics
and its Applications, Vol. 18, D. Reidel Publishing Company, Dordrecht,
1985.
[3] Chevalier, G., Around the relative center property in orthomodular lat-
tices, Proc. Amer. Math. Soc. 112 (1991), 935 -- 948.
[4] Foulis, D.J., A note on orthomodular lattices, Portugal. Math. 21 (1962)
65 -- 72.
[5] Foulis, D.J., Synaptic algebras, Math. Slovaca 60, no. 5 (2010) 631 -- 654.
[6] Foulis, D.J. and Bennett, M.K., Effect algebras and unsharp quantum
logics, Found. Physics 24, no. 10 (1994) 1331 -- 1352.
[7] Foulis, D.J. and Pulmannov´a, S., Generalized Hermitian Algebras, Int.
J. Theor. Phys. 48, no. 5 (2009) 1320 -- 1333.
[8] Foulis, D.J. and Pulmannov´a, S., Projections in synaptic algebras, Order
27, no. 2 (2010) 235 -- 257.
[9] Foulis, D.J. and Pulmannov´a, S., Centrally orthocomplete effect alge-
bras, Algebra Univers. 64 (2010) 283 -- 307, DOI 10.007/s00012-010-0100-
5.
[10] Foulis, D.J. and Pulmannov´a, S., Type-decomposition of an effect al-
gebra, Found. Physics 40 (2010) 1543 -- 1565, DOI 10.1007/s10701-009-
9344-3.
[11] Foulis, D.J. and Pulmannov´a, S., Regular elements in generalized Her-
mitian Algebras, Math. Slovaca 61, no. 2 (2011) 155 -- 172.
[12] Foulis, D.J. and Pulmannov´a, S., Hull mappings and dimension effect
algebras, Math. Slovaca 61, no. 3 (2011) 1 -- 38.
[13] Foulis, D.J., and Pulmannov´a, S., Symmetries in synaptic algebras,
arXiv:1304.4378.
26
[14] Gudder, S., Pulmannov´a, S., Bugajski, S., and Beltrametti, E., Convex
and linear effect algebras, Rep. Math. Phys. 44, no. 3 (1999) 359 -- 379.
[15] Haag, R., Local Quantum Physics, Fields, Particles, Algebras, Springer,
Berlin, 1992.
[16] Holland, S.S, Jr., Distributivity and perspectivity in orthomodular lat-
tices, Trans. Amer. Math. Soc. 112 (1964) 330-343.
[17] Holland, S.S., Jr., An m-orthocomplete orthomodular lattice is m-
complete, Proc. Amer. Math. Soc. 24 (1970) 716 -- 718.
[18] Jenca, G. and Pulmannov´a, S., Orthocomplete effect algebras, Proc.
Amer. Math. Soc. 131 (2003) 2663 -- 2671.
[19] Kalmbach, G., Orthomodular Lattices, Academic Press, London, New
York, 1983.
[20] Kaplansky, I., Any orthocomplemented complete modular lattice is a
continuous geometry, Ann. of Math. 61 (1955) 524 -- 541.
[21] McCrimmon, K., A taste of Jordan algebras, Universitext, Springer-
Verlag, New York, 2004, ISBN: 0-387-95447-3.
[22] Neumann, J. von, Mathematische Grundlagen der Quantenmechanik
Springer, Heidelberg, 1932.
[23] Neumann, J. von, Continuous geometry, Princeton Univ. Press, Prince-
ton 1960.
[24] Pulmannov´a, S., A note on ideals in synaptic algebras, Math. Slovaca
62, no. 6 (2012) 1091-1104.
[25] R´edei, Mikl´os, Why John von Neumann did not like the Hilbert space
formalism of quantum mechanics (and what he liked instead), Stud. Hist.
Phil. Mod. Physics 27, no. 4 (1996) 493 -- 510.
[26] Riecanov´a, Z., Subdirect decompositions of lattice effect algebras, In-
ternat. J. Theoret. Phys. 42, no. 7 (2003) 1425-1433.
[27] Topping, D.M., Jordan Algebras of Self-Adjoint Operators, A.M.S. Mem-
oir No 53 AMS, Providence, Rhode Island, 1965.
27
|
1109.4073 | 1 | 1109 | 2011-09-19T16:06:26 | The structure of crossed products by endomorphisms | [
"math.OA"
] | We describe simplicity of the Stacey crossed product A\times_\beta \N in terms of conditions of the endomorphism \beta. Then, we use a characterization of the graph C*-algebras C*(E) as the Stacey crossed product C*(E)^\gamma\times_{\beta_E}\N to study its ideal properties, in terms of the (non-classical) C*-dynamical system (C*(E)^\gamma, \beta_E). Finally, we give sufficient conditions for the Stacey crossed product A\times_\beta \N being a purely infinite simple C*-algebra. | math.OA | math |
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
EDUARD ORTEGA AND ENRIQUE PARDO
Abstract. We describe simplicity of the Stacey crossed product A ×β N in terms of con-
ditions of the endomorphism β. Then, we use a characterization of the graph C ∗-algebras
C ∗(E) as the Stacey crossed product C ∗(E)γ ×βE
N to study its ideal properties, in terms of
the (non-classical) C ∗-dynamical system (C ∗(E)γ , βE). Finally, we give sufficient conditions
for the Stacey crossed product A ×β N being a purely infinite simple C ∗-algebra.
In [12], Cuntz defined the fundamental Cuntz algebras On. He also represented these
algebras as crossed products of a UHF-algebra by an endomorphism, and he used this rep-
resentation to prove the simplicity of his algebras. In a subsequent paper [13] he saw this
construction as a full corner of an ordinary crossed product. However Cuntz did not explain
what kind of crossed product by an endomorphism was. Later, Paschke [26] gave an elegant
generalization of Cuntz's result, and described the crossed product of a unital C ∗-algebra by
an endomorphism β : A → A, written A ×β N, as the C ∗-algebra generated by A and an
isometry V , such that V aV ∗ = β(a). Endomorphisms of C ∗-algebras appeared elsewhere (cf.
[7], [14] and the references given there), and this led Stacey to give a modern description of
their crossed products in terms of covariant representations and universal properties [33]. He
also verified that the candidate proposed in [12] had the required property. See [3] and [11]
for further study and generalization of the Stacey's crossed product.
Cuntz's representation of the On as crossed products by an endomorphism aimed to prove
the simplicity of these C ∗-algebras. Paschke gave conditions on the C ∗-algebra A and in
the isometry to obtain a simple crossed product [26, Proposition 2.1], later improved in [11,
Corollary 2.6]. But it is in [32, Theorem 4.1] where the most powerful result about the
simplicity of the Stacey crossed product is given. Namely, If A is a unital C ∗-algebra and β is
an injective ∗-endomorphism, then A ×β N is simple and β(1) is a full projection in A if and
only if βn is outer for every n > 0 and there are no non-trivial ideals I of A with β(I) ⊆ I.
Schweizer used the representation of the Stacey crossed product as Cuntz Pimsner algebra
given by Muhly and Solel [21].
The theory of graph C ∗-algebras C ∗(E) has been developed by a number of researchers (see
[8], [9] and [28], among others) in an attempt to produce a far-reaching and yet accessible
Date: November 5, 2018.
2000 Mathematics Subject Classification. Primary 16D70, 46L35; Secondary 06A12, 06F05, 46L80.
This research was supported by the NordForsk Research Network "Operator Algebras and Dynamics"
(grant #11580). The first author was partly supported by MEC-DGESIC (Spain) through Project MTM2008-
06201-C02-01/MTM. The second author was partially supported by the DGI and European Regional Develop-
ment Fund, jointly, through Project MTM2008-06201-C02-02 and by PAI III grants FQM-298 and P07-FQM-
02467 of the Junta de Andaluc´ıa. Both authors were partially supported by the DGI and European Regional
Development Fund, jointly, through Project MTM2011-28992-C02-02, by the Consolider Ingenio "Mathemat-
ica" project CSD2006-32 by the MEC and by 2009 SGR 1389 grant of the Comissionat per Universitats i
Recerca de la Generalitat de Catalunya.
1
2
EDUARD ORTEGA AND ENRIQUE PARDO
Indeed, graph algebras do
generalization of the Cuntz-Krieger algebras of finite matrices.
provide a large and interesting class of examples of C ∗-algebras, both simple and non-simple
ones. For example, Cuntz's algebras are C ∗-algebras of a graph.
In [4] an Huef and Raeburn study the crossed products of an Exel system, and they prove
that the relative Cuntz-Pimsner algebra of an Exel system is isomorphic to a Stacey crossed
product of its core. This result leads them to a realization of the graph algebra C ∗(E) as a
Stacey crossed product C ∗(E)γ ×βE
N by an endomorphism of the core, extending the work
of Kwa´sniewski on finite graphs [20].
In the case of Leavitt path algebras (see e.g. [2]), this result appears in more simple form
in [6, Section 2], where the authors give a representation of Leavitt path algebras of a finite
graph without sinks and sources as a fractional skew monoid rings (the algebraic analog of
the crossed product by an endomorphism).
The aim of this paper is to study the simplicity of the non-unital crossed product. Our
fundamental technique is seeing the Stacey crossed product A×β N as a full corner of a crossed
product by an automorphism P (A∞ ×β∞
Z)P (see [13, 33]), where P is a full projection
of the multipliers that is invariant under the canonical gauge-action. Therefore, we can
define the associated Connes' Spectrum of the endomorphism in a similar way we do it for
an automorphism (see [22, 23, 16]) and construct a parallel Connes' spectrum theory for
endomorphisms. Hence, following the results of Olesen and Pedersen [23, 25] we characterize
simplicity for the Stacey crossed product A ×β N.
As an example, we use the characterization of graph C ∗-algebras C ∗(E) as Stacey crossed
product C ∗(E)γ ×βE
N [4], where in this case C ∗(E)γ the core, that is a (non-unital) AF-
algebra, and βE is a corner isomorphism. However, although the characterization of the
simplicity of C ∗(E) is well understood in terms of properties of the graph [8], our intention is to
describe this characterization in terms of the non-classical C ∗-dynamical system (C ∗(E)γ, βE).
We also give conditions on the C ∗-dynamical system to satisfy the Cuntz-Krieger uniqueness
theorem: for any faithful covariant representation (π, V ) of (C ∗(E)γ, βE) we have C ∗(π, V ) ∼=
C ∗(E). Finally, by using ideas from [29, 16], we give sufficient conditions on A and the
endomorphism β in order to guarantee that A ×β N is simple and purely infinite. The main
difference between these previous results and ours is that we do not ask the C ∗-algebra A to
be simple.
The contents of this paper can be summarized as follows: In Section 1 we give the basic
definitions of a Stacey crossed product. We use the characterization of the Stacey crossed
product as a Cuntz Pimsner algebra [21] to describe the gauge invariant ideals, using a result
from Katsura [18]. Then we define the Connes' spectrum of an endomorphism [22], a technical
device that, with the help of results from Olsen and Pedersen ([23, 25]), allows us to give
necessary and sufficient conditions to state the simplicity of a Stacey crossed product. In
Section 2 we apply our results to graph C ∗-algebras. We recall the definition of the graph
endomorphism βE : C ∗(E)γ → C ∗(E)γ of the core of the graph C ∗-algebra [4, Theorem 9.3],
used to prove that C ∗(E) ∼= C ∗(E)γ ×βE
N. Then, we characterize condition (L) of the graph
E in terms of the endomorphism βE: every cycle has an entry. Condition (L) in E implies
that C ∗(E) satisfies the Cuntz-Krieger uniqueness theorem (see e.g. [28, Section 2]). Thus,
we use our previous results to give the (well-known) necessary and sufficient condition of the
graph E for the graph C ∗-algebra C ∗(E) being simple. Finally, in Section 3, we give sufficient
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
3
conditions on the C ∗-algebra A and the endomorphism β : A −→ A for A ×β N being a unital
simple and purely infinite C ∗-algebra.
1. Simple Stacey crossed product
The pair (A, β), where A is a C ∗-algebra and β : A → A an injective endomorphism, is
called a C ∗-dynamical system.
Definition 1.1. We say that (π, V ) is a Stacey covariant representation of (A, β) if π :
A → B(H) is a non-degenerated representation and V is an isometry of B(H) such that
π(β(a)) = V π(a)V ∗ for every a ∈ A. We say that (π, V ) is faithful if π is faithful, and we
denote by C ∗(π, V ) the C ∗-algebra generated by {π(A)V n(V m)∗}n,m≥0.
Stacey showed in [33] that there exists a C ∗-algebra that is generated by a universal Stacey
covariant representation (ι∞, V∞). We call A ×β N := C ∗(ι∞, V∞) the Stacey crossed product
of A by the endomorphism β.
Remark 1.2. Observe that, if β is an automorphism, then A ×β N is the usual crossed
product A ×β Z.
Given z ∈ T, we define an automorphism in A×β N by the rule γz(a) = a and γz(V∞) = zV∞
for every a ∈ A. It defines the gauge action γ : T → Aut(A ×β N). An ideal I of A ×β N
is said to be gauge invariant if γz(I) = I for every z ∈ T. We define a canonical faithful
We say that the endomorphism β : A −→ A is extendible if, given any strong convergent
sequence {xn}n≥0 ⊂ A, then the sequence {β(xn)}n≥0 converges in the strong topology (i.e.,
conditional expectation E : A ×β N −→ A as E(x) :=RT γz(x)dz for every x ∈ A ×β N.
β extends to bβ : M(A) −→ M(A)). Observe that, if β is injective, then bβ(a) ∈ A implies that
a ∈ A. Indeed, let {an} be a sequence that converges in the strong topology and such that
{β(an)} converges in norm topology. Since β is isometric (β is injective) then {an} converges
in the norm topology too.
We define the inductive system {Ai, γi}i≥0 given by Ai := A and γi = β for every i ≥ 0.
Let A∞ := lim−→{Ai, γi}. For any i ≥ 0, ϕi : Ai −→ A∞ denotes the (injective) canonical map.
The diagram
A
β
A
β
β
/ A
β
/ A
β
β
/ A
β
/ A
β
β
/ · · ·
/ · · ·
gives rise to an automorphism β∞ : A∞ −→ A∞.
M(A∞).
Observe that, if β is an extendible endomorphism, then ϕ0 extends to cϕ0 : M(A) −→
Proposition 1.3 (cf.
an extendible and injective endomorphism, then A ×β N ∼= P (A∞ ×β∞
[32, Proposition 3.3]). If A is a C ∗-algebra and β : A −→ A is
Z)P , where P =
Z). Moreover, P is a full projection, so that A ×β N is strongly
cϕ0(1M (A)) ∈ M(A∞ ×β∞
Morita equivalent to A∞ ×β∞
Z.
/
/
/
/
/
/
4
EDUARD ORTEGA AND ENRIQUE PARDO
Therefore, there there exist a bijection between the ideals of A∞ ×β∞
Z and A ×β N given
by
I 7−→ P IP
and
J 7−→ (A∞ ×β∞
Z)J(A∞ ×β∞
Z) .
Moreover, if U∞ is the unitary that implements the automorphism β∞, then V∞ = P U∞P is
the isometry implementing β. Since γz(P ) = P for every z ∈ T, the canonical gauge action
γ : T −→ Aut (A∞ ×β∞
Z) restricts to the gauge action of A ×β N.
Lemma 1.4. If A is a C ∗-algebra and β : A −→ A is an extendible and injective endomor-
phism, then there exists an order preserving bijection between gauge invariant ideals of A×β N
and A∞ ×β∞
Z.
Now, we will describe the gauge invariant ideals in terms of the C ∗-dynamical system
(A, β). Given an endomorphism β : A −→ A, it is easy to check that β(A) is a hereditary
sub-C ∗-algebra of A if and only if β(A)Aβ(A) = β(A).
Definition 1.5. Let A be a C ∗-algebra and let β : A → A an endomorphism such that
β(A) is a hereditary sub-C ∗-algebra of A. We say that an ideal I of A is β-invariant if
β(A)Iβ(A) = β(I). We say that A is β-simple if there are no non-trivial β-invariant ideals.
Lemma 1.6. Let A be a C ∗-algebra, and let β : A → A an endomorphism such that β(A) is
a hereditary sub-C ∗-algebra of A. If I is a β-invariant ideal of A, then it is also βn-invariant
for every n > 0.
Proof. Let I be an ideal such that β(A)Iβ(A) = β(I). We will prove the result by induction
on n. The case n = 1 being clear, suppose that βn−1(A)Iβn−1(A) = βn−1(I). Observe that,
since β(A) is a hereditary sub-C ∗-algebra of A, we have that β(A) = β(A)Aβ(A). Thus,
βn(A)Iβn(A) = βn−1(β(A)Aβ(A))Iβn−1(β(A)Aβ(A))
⊆ βn(A)βn−1(A)Iβn−1(A)βn(A)
= βn(A)βn−1(I)βn(A) = βn−1(β(A)Iβ(A)) = βn(I) .
Therefore, βn(A)Iβn(A) = βn(I) as desired.
(cid:3)
Remark 1.7. Notice that the converse of the above Lemma is not true in general. Let
A = C0(Z) and let β : C0(Z) → C0(Z) be the automorphism that sends χ{i} (the characteristic
function at i) to χ{i+1} for every i ∈ Z. It is clear that C0(Z) is β-simple, but I = C0(2 · Z)
is a β2-invariant ideal.
Observe also that, if I is a β-invariant ideal, then β(I) is a hereditary sub-C ∗-algebra of
A, but the above example also shows that the converse it is not true.
Remark 1.8. Let β be an injective and extendible endomorphism such that β(A) is hered-
itary.
C ∗-algebra of A∞ such that β∞A = β. Indeed, it is enough to check that, given any n ∈ N
If we set the projection P = cϕ0(1M (A)) = (1, P1, P2, . . .) ∈ M(A∞), where Pn =
bβn(1M (A)), then we have that A ∼= ϕ0(A) = P A∞P . Hence, we can see A as a hereditary sub-
and a ∈ A, then P ϕn(a)P = cϕn(PnaPn) ∈ ϕ0(A). But since PnaPn ∈ βn(A)Aβn(A) = βn(A)
(by Lemma 1.6), we have that P ϕn(a)P ∈ ϕn(βn(A)) = ϕ0(A).
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
5
In [27] Pimsner introduced a class of C ∗-algebras (later improved by Katsura [17]) generated
by C ∗-correspondences (X, ϕX) over A, called Cuntz-Pimsner algebras and denoted by OX .
In particular this class includes crossed products and graph C ∗-algebras. Katsura [18] studies
gauge-invariant ideals of Cuntz-Pimsner algebras;
in particular, when X is a Hilbert A-
[1]), he obtain a bijection between gauge invariant ideals of the Cuntz-
bimodule (see e.g.
Pimsner algebra OX and invariant ideals I of A with respect to the correspondence X (i.e.,
ϕX(I)X = XI) [18, Theorem 10.6].
Let β : A −→ A is an injective endomorphism such that β(A) is a hereditary sub-C ∗-
algebra.
If we set X := βA = β(A)A with left-action ϕX given by the endomorphism β,
and right inner product given by < x, y >A= x∗y for every x, y ∈ A, then we have a C ∗-
correspondence. We have that ϕX(A) ⊆ K(X) (the compact operators of X), and since β(A)
is a hereditary sub-C ∗-algebra, it follows that β(A)Aβ(A) = β(A), whence ϕX(A) = K(X).
Therefore, since β is injective and ϕX(A) = K(X), we can define a left inner product as
A < x, y >:= ϕ−1
X (θx,y) for every x, y ∈ A. Hence, X has a natural structure of Hilbert
A-bimodule.
Lemma 1.9 (cf. [21]). If A is a C ∗-algebra, β : A −→ A is an injective endomorphism such
that β(A) is a hereditary sub-C ∗-algebra of A and X = βA is the Hilbert A-bimodule defined
above, then OX
∼= A ×β N.
Thus, we can apply Katsura's description of the gauge invariant ideals, and we see that an
ideal I of A is invariant with respect to the correspondence X if and only if β(A)I = β(I)A.
Lemma 1.10. If A is a C ∗-algebra and β : A −→ A is an injective endomorphism such that
β(A) is a hereditary sub-C ∗-algebra of A, then I is a β-invariant ideal of A if and only if
β(A)I = β(I)A.
Proof. First, suppose that β(A)I = β(I)A, and observe that β(I) ⊆ I. Thus we have that
β(I)Aβ(I) = β(A)Iβ(A) ⊆ β(A)Aβ(A) = β(A) .
Then multiplying at both sides by β(I) we have β(A)Iβ(A) = β(I)Aβ(I) ⊆ β(I), and
therefore β(A)Iβ(A) = β(I).
In the other side, suppose that β(A)Iβ(A) = β(I). From β(I) ⊆ I it follows β(I)A ⊆
β(A)I. Now, let {en} ⊂ I+ be an approximate unit of I, and let a ∈ A and y ∈ I. We claim
that β(en)β(a)y = β(ena)y converges to β(a)y, whence β(a)y ∈ β(I)A. Indeed, let z ∈ I
such that β(a)yy∗β(a∗) = β(z). Given ε > 0 there exists n ∈ N such that kenz − zk < ε/2.
Then we have
kβ(ena)y − β(a)yk2 = k(β(ena)y − β(a)y)(β(ena)y − β(a)y)∗k
≤ kβ(ena)yy∗β(a∗en) − β(a)yy∗β(a∗en)k + kβ(ena)yy∗β(a∗) − β(a)yy∗β(a∗)k
≤ kβ(enzen) − β(zen)k + kβ(enz) − β(z)k
= kenzen − zenk + kenz − zk < ε/2 + ε/2 = ε .
Thus β(ena)y converges to β(a)y, as desired.
(cid:3)
Proposition 1.11. If A is a C ∗-algebra and β : A −→ A is an injective endomorphism
such that β(A) is a hereditary sub-C ∗-algebra of A, then there is a bijection between gauge
6
EDUARD ORTEGA AND ENRIQUE PARDO
invariant ideals of A ×β N and β-invariant ideals of A. Thus, A is β-simple if and only if
A∞ is β∞-simple.
Proof. First statement holds from [18, Theorem 10.6] and Lemmas 1.9 & 1.10. Last statement
follows from [23, Lemma 6.1].
(cid:3)
Remark 1.12. The bijection stated in Proposition 1.11 sends I 7→ (A ×β N) · I · (A ×β N)
and K 7→ K ∩ A.
Finally we give necessary and sufficient conditions for the simplicity of a Stacey crossed
product. The main technical device we use is the Connes' spectrum of an endomorphism.
This is just a reformulation of the Connes' spectrum for automorphisms (see [22, 16]). We
will see that for nice endomorphisms (extendible and hereditary image) the Connes' spectrum
of β and that of the associated automorphism β∞ coincide. Therefore, we will be able to use
results by Olesen and Pedersen to determine the conditions for the simplicity of the Stacey
crossed products.
Definition 1.13. Let A be a C ∗-algebra and let β : A → A be an endomorphism. Then we
say that:
(1) β is inner if there exists an isometry W ∈ M(A) such that β = Ad W .
(2) β is outer if it is not inner.
(3) β is weakly properly outer if for every β-invariant ideal I and every n > 0 the restriction
endomorphism βn
I is outer.
Recall [15, Definition 2.1] that an automorphism α of a C ∗-algebra I is said to be properly
outer if for every nonzero α-invariant two-sided ideal I of A and for every unitary multiplier
u of I, kαI − AduIk. When β is an automorphism, the notion of weakly properly outerness
is weaker than the properly outer notion by Elliott [15], later studied by Olesen and Pedersen
[25, Theorem 10.4].
Definition 1.14. Let A be a C ∗-algebra, let β : A → A be an extendible injective endomor-
phism and let γ : T −→ Aut (A ×β N) be the gauge action. We define the Connes' spectrum
of β as
T(β) := {t ∈ T : γt(I) ∩ I 6= 0 for every 0 6= I ✁ A ×β N} .
Remark 1.15. Observe that T(β) is a closed subgroup of T. Hence can only be {1}, T or a
finite subgroup.
This definition of the Connes' spectrum coincide with the one given by Olesen and Olesen
& Pedersen [22, 23] when β is an automorphism. Moreover, using that the bijection between
ideals of A ×β N and these of A∞ ×β∞
Z given by
I 7−→ P IP
and
J 7−→ (A∞ ×β∞
Z)J(A∞ ×β∞
Z) ,
and the fact that the canonical gauge action γ : T −→ Aut (A∞ ×β∞
Z) restricts to the
gauge action of A ×β N (since γz(P ) = P for every z ∈ T), the following lemma easily follows.
Lemma 1.16. If A is a C ∗-algebra and β : A → A is an extendible injective endomorphism
with β(A) being a hereditary sub-C ∗-algebra of A, then T(β) = T(β∞).
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
7
Let β be an extendible endomorphism such that β = Ad V , where V is an isometry of
M(A). Then we can construct a unitary of M(A∞) U :=Pi≥0 bϕi(V ) such that β∞ = Ad U.
Now, let us see a result following from [22].
Theorem 1.17. Let A be a C ∗-algebra and let β : A → A be an extendible injective endo-
morphism with β(A) being a hereditary sub-C ∗-algebra of A. Let us consider the following
statements:
(1) T(βn) = T for every n > 0.
(2) Given a ∈ A e (the unitization of A) and any B hereditary sub-C ∗-algebra of A, for
every n > 0 we have that
inf {kxaβn(x)k : 0 ≤ x ∈ B with kxk = 1} = 0 .
(3) βn is outer for every n > 0.
Then, (1) ⇒ (2) ⇒ (3). Moreover, if A is β-simple, then (3) ⇒ (1) (and thus all they are
equivalent).
Proof. (1) ⇒ (2) This is [25, Theorem 10.4 and Lemma 7.1]. If T(βn) = T then T(βn
∞) = T
for every n > 0, so βn
∞ is properly outer for every n > 0. Since any hereditary sub-C ∗-algebra
B of A is also a hereditary sub-C ∗-algebra of A∞, (see Remark 1.8), we can apply [25, Proof
of Lemma 7.1] to B. Thus, since βn
∞A = βn, we have the result.
(2) ⇒ (3) Suppose that βn = Ad W for an isometry W ∈ M(A). Fix ε > 0, and take
b ∈ A+ with kbk = 1. Set c := fε(b), where fε(t) : [0, 1] −→ R+ is the continuous function
that is fε(0) = 0, constant 1 for t ≥ ε and linear otherwise. Then, we have that xc = cx = x
for every x ∈ (b − ε)+A(b − ε)+. Hence, given any 0 ≤ x ∈ (b − ε)+A(b − ε)+ with kxk = 1,
we have that
kx(cW ∗)βn(x)k2 = kx(cW ∗)W xW ∗k2 = kxcxW ∗k2
= kx2W ∗k2 = kx2W ∗W x2k = kx4k = kxk4 = 1 ,
which contradicts the hypothesis, since cW ∗ ∈ A.
Now, suppose that A is β-simple. We are going to prove that (3) ⇒ (1). By [25, Theorem
10.4] we have that T(β∞) = T if and only if T(βn
∞) = T for every n ∈ N. Let us suppose that
T(β) = T(β∞) 6= T. Hence, T(β∞) is a finite subgroup, and thus the complement T(β∞)⊥ 6=
{0}. Therefore, by [25, Theorem 4.5], for every k ∈ T(β∞)⊥ we have that βk
∞ = Ad U,
where U ∈ M(A∞). But then V = P UP ∈ M(A) is an isometry such that βk = Ad V , a
contradiction.
(cid:3)
Then it follows the characterization of simplicity.
Corollary 1.18. Let A be a C ∗-algebra and let β : A → A be an extendible injective endo-
morphism with β(A) being a hereditary sub-C ∗-algebra of A. Then A ×β N is simple if and
only if A is β-simple and βn is outer for every n > 0.
Proof. A ×β N is simple if and only if A∞ ×β∞
Z is simple if and only if A∞ is β∞-simple
and T(β∞) = T [23, Theorem 6.5] if and only if A is β-simple and T(β) = T. Therefore, by
Theorem 1.17 we have that A is β-simple and T(β) = T if and only if A is β-simple and βn
is outer for every n > 0.
(cid:3)
8
EDUARD ORTEGA AND ENRIQUE PARDO
Example 1.19. The following example is [25, Theorem 9.1]. Let A be a C ∗-algebra with a
faithful bounded trace, let A −→ B(H) be a faithful non-degenerate representation of A, and
let V be a non-unitary isometry of B(H) with V V ∗ ∈ M(A) such that V AV ∗ + V ∗AV ⊆ A.
Then suppose that there are no non-trivial ideals I of A such that V IV ∗ + V ∗IV ⊆ I. Then
we claim that the C ∗-algebra B := C ∗({AV n(V ∗)m}n,m≥0) ⊆ B(H) is simple. Indeed, let us
define the endomorphism β : A −→ A by β(a) = V aV ∗ for every a ∈ A that is extendible
(since V V ∗ ∈ M(A)). Clearly satisfies that β(A) is a hereditary sub-C ∗-algebra of A, and
it does not have any non-trivial β-invariant ideal. Now, since τ is a faithful bounded trace
of A, we can extended it to a faithful bounded trace ¯τ of M(A). Hence, M(A) has no non-
unitaries isometries. Therefore, by Theorem 1.17 A ×β N is simple, whence the natural map
A ×β N −→ B is an isomorphism.
2. Graph C ∗-algebras
In this section, we apply the above results to determine the simplicity of certain graph
C ∗-algebras. Though their simplicity is well understood in terms of properties of the graph,
we are going to deduce it from the properties of their associated C ∗-dynamical systems.
We use the conventions of [28]. Let E = (E0, E1, r, s) be a countable directed graph;
r, s : E1 → E0 denote the range and the source maps of an edge. We say that E is column-
finite if s−1(v) < ∞ for every v ∈ E0. A vertex v ∈ E0 is a sink (source) if s−1(v) = 0
(r−1(v) = 0). A vertex v ∈ E0 is called singular if is either a source or an infinite receiver.
We denote by E0
sing the set of all singular vertices. A path α of length n is a concatenation of
n edges en · · · e1 with r(ei) = s(ei+1) for i = 1, ..., n − 1. Given a path α we denote by α its
length. Let En be the set of all paths of length n, and E∗ = ∪n≥0En the set of all the paths
of finite length in E. Finally, given α, η ∈ E∗, we say that α ∈ η if there exist ρ, γ ∈ E∗ such
that η = ραγ.
Recall that the graph C ∗-algebra C ∗(E) is the universal C ∗-algebra generated by orthogonal
projections {Pv}v∈E 0 and partial isometries {Se}e∈E 1, satisfying the following conditions:
(CK1)
(CK2)
S∗
e Sf = δe,f · Ps(e)
SeS∗
e
Pv = Xr(e)=v
for every e, f ∈ E1
for every v ∈ E0
sing.
See [28] for a survey on graph C ∗-algebras. One can naturally define a group homomorphism
γ : T → Aut C ∗(E), given by γz(Pv) = Pv and γz(Se) = zSe for every z ∈ T, v ∈ E0 and
e ∈ E1; it is the so-called gauge action on C ∗(E). An ideal I of C ∗(E) is said to be a gauge
invariant ideal if γz(I) = I for every z ∈ T (see [8] and [9]). The core sub-C ∗-algebra of
C ∗(E) is defined as
C ∗(E)γ := {x ∈ C ∗(E) : γz(x) = x for every z ∈ T} .
We can give another description of the core. For every n ∈ N and v ∈ E0, define
Fn(v) := {SηS∗
ρ : η, ρ ∈ En with s(η) = s(ρ) = v} ∼= Mkn,v (C)
for some kn,v ∈ N, and let Fn = ⊕v∈E 0Fn(v). Now, if we index the vertices {vi}i≥0, then we
i,j≥0 Fi(vj) for every n, n ≥ 0. These are finite dimensional sub-C ∗-algebras
define Cn.m :=Pn,m
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
9
of C ∗(E)γ with Cn,m ⊆ Cn,m+1 and Cn,m ⊆ Cn+1,m for every n, m ≥ 0. Hence,
C ∗(E)γ = [n,m≥0
Cn,m
is an AF-algebra.
We recall the following result from [4], that allows to present certain graph C ∗-algebras
as C ∗-dynamical systems (we would like to thank the authors for showing us the result even
before the releasing of the manuscript).
Theorem 2.1 ([4, Theorem 9.3]). Let E be a column finite graph without sinks. If we define
the endomorphism βE : C ∗(E)γ → C ∗(E)γ as βE(z) = T zT ∗ for every z ∈ C ∗(E)γ, where
T = Xe∈E 1
s−1(s(e))1/2Se
is an isometry of M(C ∗(E)), then we have that C ∗(E) ∼= C ∗(E)γ ×βE
N.
Notice that the endomorphism βE : C ∗(E)γ → C ∗(E)γ is injective, extendible and βE(C ∗(E)γ)
is a hereditary sub-C ∗-algebra of C ∗(E)γ.
Definition 2.2. A subset H ⊆ E0 is said to be hereditary if, whenever η ∈ E∗ with r(η) ∈ H,
then s(η) ∈ H. We say that H is saturated if, whenever r−1(v) < ∞ and {s(r−1(v))} :=
{z ∈ E0 : z = s(e) for some e ∈ r−1(v)} ⊆ H, then v ∈ H.
By [8, Theorem 4.1], there exists a bijection between hereditary and saturated subsets
of E0 and gauge invariant ideals of C ∗(E), H 7−→ KH, where KH := span {SηS∗
ν : η, ν ∈
E∗ with s(η) = s(ν) ∈ H}. The inverse map is K 7−→ HK, where HK := {v ∈ E0 : Pv ∈ K}.
Now, given a hereditary and saturated subset of E0, we define
By Remark 1.12, IH is a βE-invariant ideal of C ∗(E)γ, and it is easy to see that
IH := KH ∩ C ∗(E)γ.
IH := Xv∈H,n≥0
Fn(v).
On the other side, if K is a gauge invariant ideal of C ∗(E), since I := K ∩ C ∗(E)γ is a
βE-invariant ideal, then we have that the set
HI := {v ∈ E0 : Pv ∈ I}
is a subset of HK. Moreover, since Pv ∈ C ∗(E)γ for every v ∈ E0, it is clear that HK ⊆ HI,
whence HK = HI. Thus, HI is an hereditary and saturated subset of E0.
In particular, if I is a βE-invariant ideal of C ∗(E)γ, then K := (C ∗(E)) · I · (C ∗(E)) is a
gauge invariant ideal of C ∗(E) and
IHI = IHK = KHK ∩ C ∗(E)γ = K ∩ C ∗(E)γ = I.
Conversely, if H is a hereditary and saturated subset of E0, since IH := KH ∩ C ∗(E)γ, we
conclude that
HIH = HKH = H.
10
EDUARD ORTEGA AND ENRIQUE PARDO
Summarizing, there exists a bijection between the hereditary and saturated subsets of E0
and the βE-invariant ideals of C ∗(E)γ defined by the maps
H 7−→ IH = Xv∈H,n≥0
Fn(v)
and
I 7−→ HI = {v ∈ E0 : Pv ∈ I}.
One could be tempted to think that there is a bijection between hereditary sets of E0 and
the ideal of C ∗(E)γ such that βE(I) ⊆ I, but this is not the case (see Examples 2.5).
Theorem 2.3 (cf. [8, Theorem 4.1]). If E is a column finite graph without sinks, then there
is a bijection between the closed gauge invariant ideals of C ∗(E), the hereditary and saturated
subsets of E0 and the βE-invariant ideals of C ∗(E)γ.
Corollary 2.4. Let E be a column finite graph without sinks, then E0 has no non-trivial
hereditary and saturated subsets if and only if C ∗(E)γ does not have a proper βE-invariant
ideals.
Example 2.5. In the following examples we would like to illustrate some consequences of
Corollaries 1.18 and 2.4 and determine the simplicity of some graph C ∗-algebras. We would
like to remark again that this is well-known by [8, Proposition 5.1]. However, one can slightly
modify some of the examples to get new simple C ∗-algebra that probably do not arise as
graph C ∗-algebras.
(1) Consider the graph E
•v0
/ •v1
/ •v2
/ · · ·
Then E0 has no non-trivial hereditary and saturated subsets. We have that C ∗(E)γ ∼=
C0(N ∪ {0}) and the endomorphism βE : C ∗(E)γ → C ∗(E)γ sends χ{i} (the charac-
teristic function at i) to χ{i+1} for every i ≥ 0. Then, since E0 does not have non-
trivial saturated and hereditary subsets, C0(N ∪ {0}) is βE-simple. Moreover, since
M(C0(N∪{0})) is a commutative C ∗-algebra, it does not have non-unitary isometries.
Hence, C ∗(E) is simple.
(2) Consider the graph E
e
•v0
/ •v1
/ •v2
/ · · ·
Then C ∗(E)γ = K (the compact operators of a countable infinite dimensional Hilbert
space H), that is simple. Therefore C ∗(E)γ is (βE−) simple, and thus E0 has no
non-trivial hereditary and saturated subsets. Moreover, it is not difficult to see that
βE = Ad W where W is the shift operator of H, whence βE is inner and C ∗(E) is not
a simple C ∗-algebra.
(3) This is the graph C ∗-algebra picture of the algebra On. Let E be the graph
(n)
•v
/
/
/
/
/
/
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
11
with n loops. We have that C ∗(E)γ is isomorphic to the n-infinity UHF-algebra
Un :=N∞
i=1 Mn and βE(x) = P ⊗ x for every x ∈ Un, where P =
1/n · · ·
...
1/n · · ·
1/n
...
1/n
.
Therefore C ∗(E)γ is βE-simple, since Un is simple. Moreover, since C ∗(E)γ is a unital
and finite C ∗-algebra it does not have non-unitary isometries, and therefore C ∗(E) is
a simple C ∗-algebra.
(4) An example of the different behaviour of βE and β2
E can be found when the graph E
is
•v
) •w
In this case Pv and Pw generate two orthogonal ideals of C ∗(E)γ, Iv and Iw respectively,
n=1 M2, and such that C ∗(E)γ = Iv ⊕ Iw. We
both isomorphic to the CAR-algebra N∞
have that βE(x, y) = (y, P ⊗ x) for every (x, y) ∈ Iv ⊕ Iw, where P =(cid:18) 1/2 1/2
1/2 1/2 (cid:19).
Therefore C ∗(E)γ is βE-simple, but β2
E(Iw) ⊆ Iw. Moreover, since
C ∗(E)γ is a unital and finite C ∗-algebra, it does not have non-unitary isometries, and
therefore C ∗(E) is a simple C ∗-algebra.
E(Iv) ⊆ Iv and β2
(5) Consider the graph E
e
/ •w
•v
f
em, Semf S∗
Observe that C ∗(E)γ = span {Pv, SemS∗
emf : m ≥ 0}. Then, C ∗(E)γ is a
commutative C ∗-algebra isomorphic to C(X) where X = {1/n : 1 ≤ n} ∪ {0}, and the
endomorphism acts βE(χ[0,1/n]) = χ[0,1/(n+1)] for every n ≥ 1. Since {v} is a saturated
and hereditary subset, there exists a proper βE-invariant ideal, that corresponds to
the ideal C0(X \ {0}). Given n ∈ N let In be the ideal of C ∗(E)γ generated by χ[0,1/n].
Observe that βE(In) ⊆ In (in particular βE(In) = In+1). Therefore C ∗(E)γ posses a
infinite countably family of different ideals I such that βE(I) ⊆ I.
Definition 2.6. A C ∗-dynamical system (A, β) is said to satisfy the Cuntz-Krieger uniqueness
theorem if for every faithful Stacey covariant representation (π, V ) of (A, β) we have that
C ∗(π, V ) ∼= A ×β N.
Recall that the graph E satisfies condition (L) if every cycle has an entry. A graph E
satisfies condition (L) if and only if, given any ∗-homomorphism η : C ∗(E) → B such that
η(Pv) 6= 0 for every v ∈ E0, we have that η is injective (see e.g. [28, Section 2]). Thus,
Theorem 2.7 (cf. [28, Theorem 2.4] & [24, Theorem 2.5]). Let E be a column finite graph
without sinks. Then the following statements are equivalent:
(1) The graph E satisfies condition (L).
(2) (C ∗(E)γ, βE) satisfies the Cuntz-Krieger uniqueness theorem.
(3) T(βE) = T.
Now we will see that for the dynamical system (C ∗(E)γ, βE) associated to a graph C ∗-
algebras, the results of Olesen and Pedersen [24, Theorem 2.5 & Theorem 4.6] reduces to a
simpler way.
,
,
)
l
l
/
12
EDUARD ORTEGA AND ENRIQUE PARDO
Proposition 2.8. Let E be a column finite graph without sinks and let (C ∗(E)γ, βE) be its
associated C ∗-dynamical system. If βE is weakly properly outer then E satisfies condition (L).
Proof. Suppose that βE is weakly properly outer and that E does not satisfy condition (L),
i.e., there exists a cycle α without an entry. We can suppose that α = en · · · e1 with ei ∈ E1
and vi = r(ei) for i = 1, . . . , n, such that s(e1) = r(en) = vn and r(ei) 6= vn for every i 6= n.
Let Hα = {vi}i=1,...,n, and let I be the ideal of C ∗(E)γ generated by {Pv}v∈Hα. Observe that,
since α does not have any exit, by (CK2) we have that
I =Xk≥0
Fk(vn) .
Given w ∈ E0, let {ηi,w}νw
i=1 ⊆ En be the set of paths such that s(ηi,w) = w (a finite number
since E is column finite),. Given any z ∈ E0
vn, where
E0
vn := {z ∈ E0 : exists η ∈ E∗ with s(η) = vn and r(η) = z} ,
consider all the paths {γj,z}j∈∆z ⊆ E∗ such that s(γj,z) = vn and r(γj,z) = z. Observe that
1 ≤ {γj,z}j∈∆z ≤ ∞. Given any path γj,z, we define κi,z := s−1(s(fn)) · · · s−1(s(f1)) < ∞
for fn · · · f1 = ηi,z with fi ∈ E1. Then, define the formal sums (we still not determine where
their converge to)
Vw :=
Vvk =
νwXj∈∆w,i=1
νvkXi=1
κ−1/2
i,vk
κ−1/2
i,w Sηi,wγj,wS∗
γj,wα
if w ∈ E0
vn \ Hα ,
Sηi,vk ek···e1S∗
ek···e1α
for 1 ≤ k ≤ n − 1
and
Vvn =
νvnXi=1
κ−1/2
i,vn Sηi,vn S∗
α .
vn
γj,z : γi,w = γj,z for z, w ∈ E0
I = span {Sγi,w S∗
E0
are elements of I of norm less or equal to 1. Observe that
We claim that Pw∈E 0
vn and k, l ∈ N such that γk,w = γl,z then (Pu∈E 0
νwXi=1
γl,z = VwSγk,w S∗
( Xu∈E 0
Vu)Sγk,w S∗
γl,z =
vn
Vw converges with the strong topology in M(I). Indeed, recall that
vn}, so it is enough to see that for every v, w ∈
Vu)
γl,z and Sγk,w S∗
Vu)Sγk,wS∗
γl,z (Pu∈E 0
vn
κ−1/2
i,w Sηi,wγk,w S∗
γl,zα ∈ I .
vn
Now we have that
k
νwXi=1
κ−1/2
i,w Sηi,wγk,w S∗
γl,zαk2 = k(
κ−1/2
i,w Sηi,wγk,wS∗
γl,zα)∗
νwXi=1
κ−1/2
i,w Sηi,wγk,w S∗
γl,zαk
νwXi=1
νwXi=1
= k
κ−1
i,wSγl,zαS∗
γl,zαk ≤ kSγl,z S∗
γl,z k = 1
as desired. Analogously, we can see that Sγk,wS∗
of norm less or equal than 1. Define V :=Pw∈E 0
vn
γl,z (Pu∈E 0
Vu) converges to an element of I
Vw. Then we have that V ∗V = 1M (I). One
vn
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
13
can easily check that V zV ∗ = βE
the hypothesis.
n(z) for every z ∈ I, so βE
n
I is inner, a contradiction with
(cid:3)
Remark 2.9. In the proof of Proposition 2.8 we prove that, if E does not satisfy condition
(L), i.e., there exists a loop α without entries, then given any vertex v ∈ E0 of the loop α we
have that
I :=Xn≥0
Fk(v) ,
is a βE-invariant ideal of C ∗(E)γ such that βE
n
I is inner for some n > 0. Then HI is a
hereditary and saturated subset of E0 containing Hα := {v ∈ E0 : v ∈ α} such that I = IHI .
But since IHI ⊆ IH for every hereditary and saturated subset H of E0 containing Hα we have
that HI is the minimal hereditary and saturated subset of E0 containing Hα. Thus I is a
minimal βE-invariant ideal such that βE
n
I is inner.
Observe also that in general I is a non-simple sub-C ∗-algebra of C ∗(E)γ. For example, if
n > 1, then the ideal generated by Pvi for every 1 ≤ i ≤ n is proper (one can check easily
that it cannot contain Pvj for i 6= j).
Proposition 2.10. Let E be a column finite graph without sinks and let (C ∗(E)γ, βE) be its
associated C ∗-dynamical system. If E satisfies condition (L) then βE is weakly properly outer.
E I is inner for some n > 0. Hence, I =Pv∈HI ,k≥0 Fk(v), where HI is a
Proof. Suppose that E satisfies condition (L) and there exist a non-zero βE-invariant ideal I
of C ∗(E)γ such that βn
hereditary and saturated subset of E0. So, there is a gauge-invariant ideal of C ∗(E), say KHI ,
generated by {Pv}v∈HI . Recall that the core K γ
is precisely I. Now, there exists an isometry
HI
n(z) = W zW ∗ for every z ∈ I. Since I contains an approximate unit
W ∈ M(I) such that βE
for KHI (see for example [5, Lemma 3.4]) we can see M(I) as a sub-C ∗-algebra of M(KHI ).
Define U := W ∗T n a unitary in M(KHI ) (we are also using that T I ⊆ IT ). Then, for
every z ∈ I we have z = UzU ∗ Observe that, given any y ∈ I, we have that that yU = Uy.
We claim that HI ⊆ E0 cannot have sources. Indeed, if v ∈ HI is a source, then Pv ∈ I,
and hence
Pv = UPvU ∗ = W ∗T nPvT n∗W .
with r(ηl) = r(ρl) = v and ηl = ρl + n. But this contradicts the fact that v is a source.
i=1 be the set of all the paths in En with r(ηi,v) = v. So,
ρl ∈ PvKHI Pv ⊆ PvC ∗(E)Pv, with λl ∈ C and ηl, ρl ∈ E∗
We have that W ∗T nPv =Pl λlSηlS∗
Now, given any v ∈ E0, let {ηi,v}νv
given v ∈ HI and i ≤ νv, we define
0 6= Xi,v := S∗
ηi,vU ∈ Ps(ηi,v)C ∗(E)γPv .
If µ, γ ∈ Em with s(µ) = s(γ) ∈ HI then we have that SµS∗
1 ≤ k ≤ νw and 1 ≤ l ≤ νz, we have that
γ ∈ I. So, for every w, z ∈ HI,
γ)X ∗
l,z = S∗
(1) Xk,w(SµS∗
(2) Xk,wX ∗
(3) X ∗
ηk,w U(SµS∗
l,z = δw,z · δk,l · Ps(ηk,w),
γXl,z = U ∗Sηk,wSµS∗
k,wSµS∗
γS∗
γ)U ∗S∗
ηl,z = S∗
ηk,w(SµS∗
γ)Sηl,z ,
ηl,z U = δr(µ),s(ηk,w) · δr(γ),s(ηj,z ) · Sηk,wSµS∗
γS∗
ηl,z ,
while given v ∈ HI and 1 ≤ i ≤ νv, we have that
Xi,vX ∗
i,vXi,v = S∗
ηi,v UU ∗Sηi,v S∗
ηi,v U = S∗
ηi,v Sηi,vS∗
ηi,v U = S∗
ηi,v U = Xi,v ,
14
EDUARD ORTEGA AND ENRIQUE PARDO
so Xi,v is a partial isometry in Ps(ηi,v )C ∗(E)γSηi,v S∗
ηi,v .
Now, choose any v ∈ HI and 1 ≤ i ≤ νv, and consider the isometry Xi,vSηi,v ∈ Ps(ηi,v )C ∗(E)Ps(ηi,v).
Given any ε > 0, there exist m ∈ N, λj ∈ C \ {0} and αj, βj ∈ E∗ with αj = βj + n and
r(αj) = r(βj) = s(ηi,v) such that
kXi,vSηi,v −
mXj=1
λjSαj S∗
βj k < ε .
Suppose that β1 ≥ βi for every i ≤ m. Then we have that
kps(ηi,v) − y∗yk = k(Xi,vSηi,v)∗Xi,vSηi,v − y∗yk
≤ k(Xi,vSηi,v )∗Xi,vSηi,v − y∗Xi,vSηi,v k + ky∗Xi,vSηi,v − y∗yk
≤ k(Xi,vSηi,v )∗ − y∗kkXi,vSηi,v k + kXi,vSηi,v − ykky∗k
≤ ε · 1 + ε(1 + ε)
Thus, if ε < 1/4, we have that y∗y is invertible in Ps(ηi,v )C ∗(E)γPs(ηi,v ). Hence, ySβj 6= 0 for
every 1 ≤ j ≤ m. Thus,
β1 − ySβ1S∗
kSβ1S∗
β1y∗k = k(Xi,vSηi,v )Sβ1S∗
≤ k(Xi,vSηi,v )Sβ1S∗
β1(Xi,vSηi,v )∗ − ySβ1S∗
β1(Xi,vSηi,v)∗ − (Xi,vSηi,v )Sβ1S∗
β1y∗k
β1y∗k+
+ k(Xi,vSηi,v )Sβ1S∗
β1y∗ − ySβ1S∗
β1y∗kε
β1kε + kSβ1S∗
β1y∗k
≤ k(Xi,vSηi,v )Sβ1S∗
≤ ε + ε(1 + ε)
β1y∗ is invertible in Sβ1S∗
and therefore ySβ1S∗
jSγj ,
where γj ∈ En with s(γj) = r(γj) = r(β1) = s(ηi,v) for every 1 ≤ j ≤ m′. Hence, the γjs
are cycles. Let γ = γ1. Since by assumption E satisfies condition (L), we have that γ has
an entry. Therefore, there exists η ∈ E∗ such that γ /∈ η. So, we have that Sβ1ηS∗
β1η ∈
β1C ∗(E)γSβ1S∗
ηSγ = 0, that contradicts the fact that Sβ1S∗
Sβ1S∗
β1 is
invertible in Sβ1S∗
(cid:3)
β1 and hence S∗
β1C ∗(E)γSβ1S∗
β1.
β1. So, 0 6= S∗
β1ηSβ1 = S∗
β1C ∗(E)γSβ1S∗
j=1 λ′
β1ySβ1 =Pm′
Summarizing, we have the following result.
Theorem 2.11. Let E be a column finite graph without sinks. Then the following statements
are equivalent:
(1) The graph E satisfies condition (L).
(2) (C ∗(E)γ, βE) satisfies the Cuntz-Krieger uniqueness theorem.
(3) T(βE) = T.
(4) There is no βE-invariant ideal I of C ∗(E)γ and n ∈ N such that βn
E I = Ad V , where
V ∈ M(I) is an isometry.
(5) βE is weakly properly outer.
Finally, using the characterization of simplicity of C ∗(E) in terms of properties of the graph
E [8, Proposition 5.1], the representation of an Huef and Raeburn of the graph C ∗-algebra
C ∗(E) as the Stacey crossed product C ∗(E)γ ×βE
N, joint with Corollary 1.18 and Theorem
2.11, we conclude the desired result.
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
15
Theorem 2.12 (cf. [8, Proposition 5.1]). Let E be a column finite graph without sinks. Then
the following statements are equivalent:
(1) C ∗(E)γ ×βE
(2) E does not have non-trivial hereditary and saturated subsets and satisfies condition
N is simple.
(L).
(3) C ∗(E)γ does not have any proper βE-invariant ideal and βn
E is outer for every n ≥ 1.
3. Pure infiniteness
In Theorem 1.17 we have given necessary and sufficient conditions on the endomorphism β
for the simplicity of the C ∗-algebra A ×β N. If A is a unital C ∗-algebra and β(1) 6= 1 we have
then that A ×β N contains a proper isometry, and if in addition A ×β N is simple, we have
that it is a properly infinite C ∗-algebra. We will see that for a broad class of unital real rank
zero C ∗-algebras A we have that A ×β N turns out to be purely infinite. Our result generalize
and unify similar results given in [29] and [16].
Lemma 3.1. Let A be a unital C ∗-algebra, let β : A −→ A be an injective endomorphism,
and suppose that does not exist any proper ideal I of A such that β(I) ⊆ I. Then, given any
non-zero a ∈ A+ there exists n ∈ N such that a + β(a) + · · · + βn(a) is a full positive element
in A.
Proof. Consider the ideal I := span {xβn(a)y : n ≥ 0, x, y ∈ A} 6= 0.
β(I) ⊆ I and then, by hypothesis, we have that I = A. Therefore we can write
It clearly satisfies
1 =
kXi=1
xiβnk(a)yk
where xi, yi ∈ A and ni ∈ N for every i ∈ {1, . . . , k}. Then, taking n = maxi{ni}, we have or
desired result.
(cid:3)
Let T (A) be the set of tracial states of A, which is a compact space with the ∗-weak
topology. We say that A has strict comparison if: (i) T (A) 6= ∅; (ii) Whenever p ∈ AqA such
that τ (p) < τ (q) for every τ ∈ T (A), we have that p . q. For example, every unital exact and
stably finite C ∗-algebra of real rank zero that is Z-stable has strict comparison [30, Corollary
4.10].
Recall that a (non-necessarily simple) C ∗-algebra A is said to be purely infinite if and only
if all positive elements are properly infinite [19] ; in particular, every projection of A (if it has
any) must be properly infinite. Also recall that a unital simple C ∗-algebra is purely infinite
if and only if has real rank zero and every projection is infinite [35]. The following lemma is
a slight modification of [29, Lemma 3.2].
Lemma 3.2 (cf. [29, Lemma 3.2]). Let A be a unital C ∗-algebra that either has strict compar-
ison or is purely infinite. Let β : A −→ A be an injective endomorphism such that β(1) 6= 1
and β(A) is a hereditary sub-C ∗-algebra and let A ×β N = C ∗(A, V ). If does not exist any
proper ideal I of A such that β(I) ⊆ I, then for every full projection p ∈ A there exist a
partial isometry u ∈ A and m ∈ N such that (V ∗)mu∗puV m = (V ∗)mV m = 1.
16
EDUARD ORTEGA AND ENRIQUE PARDO
Proof. We claim that there exists m ∈ N such that V m(V m)∗ . p. Observe that if A is
purely infinite then p is a properly infinite full projection. So, we have that V V ∗ ∈ ApA = A.
Hence, V V ∗ . p, so that m = 1 holds.
Now suppose that A has strict comparison. Then T (A) is non-empty and compact. So,
given any k ∈ N we set
α = inf {τ (p) : τ ∈ T (A)}
and
γk = sup {τ (V k(V ∗)k) : τ ∈ T (A)} .
Observe that, since p is full, we have that α > 0. Now, we claim that there exists n ∈ N such
that γn < 1. Indeed, it is enough to prove that there exists n ∈ N such that 1 − V n(V ∗)n is
a full projection. Let us construct the ideal
I := span {x(V l(V ∗)l − V l+1(V ∗)l+1)y : l ≥ 0, x, y ∈ A} 6= 0 .
It is clear that β(I) ⊆ I. Therefore, by Lemma 3.1, there exists n ∈ N such that
(1−V V ∗)+· · ·+βn−1(1−V V ∗) = (1−V V ∗)+· · ·+(V n−1(V ∗)n−1 −V n(V ∗)n) = 1−V n(V ∗)n ,
is a full projection. Therefore γn < 1. By the same argument as in the proof of [29, Lemma
3.2], we have that τ (V nl(V ∗)nl) ≤ γl
n for every l ∈ N. Then, there exists l ∈ N such that
τ (V nl(V ∗)nl) ≤ γl
n < α ≤ τ (p). Since A has strict comparison, we have that V nl(V ∗)nl . p.
So, there exists a partial isometry u ∈ A such that u∗u = V nl(V ∗)nl and uu∗ ≤ p. Therefore
(V ∗)nlu∗puV nl = (V ∗)nl(V nl(V ∗)nl)V nl = 1, so we are done.
(cid:3)
Lemma 3.3. Let A be a C ∗-algebra of real rank zero, and let β : A → A be an extendible
injective endomorphism with β(A) being hereditary such that T(β) = T. Then, given any
a ∈ A∼ and any B hereditary sub-C ∗-algebra of A we have that
inf {kpaβ(p)k : p is a non-zero projection of B} = 0 .
Proof. Let a ∈ A+ and let B be a hereditary sub-C ∗-algebra of A. Given ε > 0, by Theorem
1.17 there exists x ∈ B+ with kxk = 1 such that kxaβ(x)k < ε/2. Given δ > 0, let
fδ : [0, 1] −→ [0, 1] be such that f (t) = 1 for every t ∈ [1 − δ/2, 1] and such that fδ(t) − t < δ
for every 0 ≤ t ≤ 1. Take δ > 0 such that kfδ(x)aβ(fδ(x))k < ε. Let C = {y ∈ B : fδ(x)y =
yfδ(x) = y} 6= 0. Notice that C is a hereditary sub-C ∗-algebra of B. Since C has real rank
zero, there exists a non-zero projection p ∈ C, and by construction pfδ(x) = fδ(x)p = p.
Therefore
kpaβ(p)k = kpfδ(x)aβ(fδ(x)p)k ≤ kfδ(x)aβ(fδ(x))k < ε .
(cid:3)
Corollary 3.4. Let A be a C ∗-algebra of real rank zero, and let β : A −→ A be an extendible
injective endomorphism with β(A) being hereditary such that T(βn) = T for every n > 0.
Then, given any ε > 0, a1, . . . , ak ∈ A∼ and n1, . . . , nk ∈ N and a projection p ∈ A, there
exists a projection q ∈ pAp such that
kqaiβni(q)k < ε
for every i ∈ {1, . . . , k} .
A C ∗-algebra A is said to be weakly divisible if given any projection p ∈ A, there exists
a unital ∗-homomorphism M2 ⊕ M3 −→ pAp [31, Lemma 5.2]. Conditions for a non-type I
real rank zero C ∗-algebra being weakly divisible are given in [31, Theorem 5.8]. In particular,
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
17
every simple non-type I C ∗-algebra of real rank zero is weakly divisible. Observe that, if A
is weakly divisible or purely infinite, then the following statement holds:
(†)
Given any n ∈ N and p ∈ A there exists p1, . . . , pn ∈ A non-zero pairwise orthogonal
subprojections of p with p ∈ ApiA for all i
Proposition 3.5. Let A be a unital C ∗-algebra of real rank zero satisfying (†), let β : A −→ A
be an injective endomorphism such that β(A) is a hereditary sub-C ∗-algebra of A, and let
A ×β N = C ∗(A, V ). If does not exist any proper ideal I of A such that β(I) ⊆ I, then given
any non-zero projection p ∈ A there exist a full projection q ∈ A and c ∈ A ×β N such that
q = cpc∗.
Proof. By Lemma 3.1 there exists n ∈ N such that p+β(p)+· · ·+βn(p) is a full positive element
of A. Since A satisfies (†) there exist non-zero orthogonal projections p0, . . . , pn ∈ A such
that p0 + · · ·+ pn ≤ p with p ∈ ApiA for all i ∈ {0, . . . , n}. Observe that p + β(p) + · · ·+ βn(p)
lies in the ideal generated by q′ := p0 + β(p1) + · · ·+ βn(pn), so q′ is also a full positive element
of A. Denote p′
i := βi(pi) for every i ∈ {0, . . . , n}. Now we are going to use induction on n to
construct a projection q ∈ A such that p′
n ∈ AqA. The case n = 0 is clear. Now,
suppose that there exists a projection qk−1 such that p′
0 + · · · + p′
k−1 ∈ Aqk−1A.
0 + · · · + p′
Using the Riesz decomposition of V (A) [34] we have p′
k ∼ ak ⊕ bk such that ak . qk−1 and
kvk ≤ p′
bk . 1 − qk−1. Let vk be the partial isometry such that v∗
k ≤ 1 − qk−1. If we
define the projection qk := qk−1 + vkv∗
k ∈ AqkA. Therefore
the projection q := qn is full. If we define c := p0 + v1V 1p1 + · · · + vnV npn, then we have that
k, then we have that p′
1 + · · · + p′
k and vkv∗
cpc∗ = cc∗ = p0 + v1β(p1)v∗
1 + · · · + vnβn(pn)v∗
n = q ,
as desired.
(cid:3)
Theorem 3.6. Let A be a unital C ∗-algebra of real rank zero satisfying (†) that has strict
comparison, let β : A −→ A be an injective endomorphism such that β(1) 6= 1 and β(A) is a
hereditary sub-C ∗-algebra of A. If A ×β N is simple and β(1) is a full projection of A, then
A ×β N is purely infinite simple C ∗-algebra.
Proof. It is enough to prove that given a positive element x ∈ A×β N there exist a, b ∈ A×β N
such that axb = 1. Let E : A ×β N −→ A be the canonical faithful conditional expectation.
So, 0 6= E(x) = c ∈ A+. Then, for kck > ε > 0 we have that the hereditary sub-C ∗-algebra
(c − ε)+A(c − ε)+ ⊆ c1/2Ac1/2 has real rank zero. Hence, there exists a non-zero projection
p = c1/2yc1/2 ∈ c1/2Ac1/2. Then, q = y1/2cy1/2 is a projection, and E(y1/2xy1/2) = y1/2cy1/2 =
q. Thus, we can assume that E(x) = q is a non-zero projection. Given 1/2 > ε > 0, there
exists x′ = (V ∗)md−m + · · ·+ q + · · ·+ dmV m, with dj ∈ A+ for every j, such that kx − x′k < ε.
By Corollary 1.18, Theorem 1.17 and Corollary 3.4, there exists a non-zero projection p ∈ qAq
such that
kpdiβi(p)k < ε/2m
and
kβi(p)d−ipk < ε/2m
for every i ∈ {1, . . . , m}. Therefore
kpxp − pk ≤ kpxp − px′pk + kpx′p − pk ≤ ε + ε < 1 .
Then, pxp is invertible in p(A×β N)p, whence there exists y ∈ p(A×β N)p such that ypxp = p.
Since we are assuming that A ×β N is simple and β(1) is a full projection, [32, Theorem 4.1]
18
EDUARD ORTEGA AND ENRIQUE PARDO
implies that there are no non-trivial ideals I of A such that β(I) ⊆ I. Thus, by Proposition
3.5, there exist c ∈ A ×β N and a full projection q ∈ A such that cpc∗ = q.
By Lemma 3.2, there exist m ∈ N and a partial isometry u ∈ A such that (V ∗)mu∗quV m = 1
and therefore
(V ∗)mu∗(cypxpc∗)uV m = (V ∗)mu∗cpc∗uV m = (V ∗)mu∗quV m = 1 .
Thus, if we set a := (V ∗)mu∗cyp and b := pc∗uV m we have axb = 1, as desired.
(cid:3)
When A is a purely infinite C ∗-algebra, we generalize the result of [16].
Corollary 3.7. Let A be a unital purely infinite C ∗-algebra of real rank zero, let β : A −→ A
be an injective endomorphism such that β(1) 6= 1 is a full projection and β(A) is a hereditary
sub-C ∗-algebra of A. Then A ×β N is a simple purely infinite C ∗-algebra if and only if A ×β N
is simple.
Proof. The proof works in the same way as that of Theorem 3.6, but reminding that Lemma
3.2 and condition (†) are also satisfied for purely infinite C ∗-algebras.
(cid:3)
Finally, we can use Corollary 3.7 to characterize when a crossed product by an automor-
phism A ×α Z is simple and purely infinite.
[16, Theorem 3.1]). Let A be a unital purely infinite C ∗-algebra of real
Corollary 3.8 (cf.
rank zero, and let α : A −→ A be an automorphism. Then A ×α Z is a simple purely infinite
C ∗-algebra if and only if A ×α Z is simple.
Proof. The proof is a verbatim of the proof of [16, Theorem 3.1]. We only have to prove that
there exist projections p, e ∈ A and partial isometries t, s ∈ A such that
s∗s = α(p) ,
ss∗ = e < p ,
t∗t = 1 − α(p)
and
tt∗ = 1 − e .
Indeed, since 1 is a properly infinite projection, there exist mutually orthogonal projections
p1, p2, p3 ∈ A, all them Murray-von Neumann equivalent to 1. Observe that α(pi) are mutually
orthogonal full properly infinite projection of A. Then, we have that α(p1) ∼ e < α(p2) for
some projection e ∈ A. Since α(p3) ⊥ α(p1) and e ⊥ α(p1), by [10, Proposition 2.5] we have
that α(p1) and e are homotopic equivalent, and hence 1 − α(p1) and 1 − e are homotopic
equivalent, thus Murray-von Neumann too. Thus, setting p := α(p1) we have proved the
claim.
By the proof of [16, Theorem 3.1], the dynamical system (A, α) is exterior equivalent to
(A, ρ), where ρ is the automorphism defined by ρ(x) = (s + t)α(x)(s + t)∗ for every x ∈ A.
So, it is enough to prove that A ×ρ Z is simple and purely infinite. Notice that T(α) = T(ρ),
and that A is ρ-simple since it is α-simple. Hence, A ×ρ Z is a simple C ∗-algebra. Then
p(A ×ρ Z)p ∼= pAp ×ρ N is a full simple hereditary sub-C ∗-algebra. Now, we have that pAp
is a purely infinite C ∗-algebra of real rank zero, and by construction ρ(p) = sα(p)s∗ is a
full projection of pAp. Thus, by Theorem 3.7 we have that pAp ×ρ N is a purely infinite
C ∗-algebra, whence so is A ×α Z.
(cid:3)
algebras On [12]. Let Um be the m-infinity UHF algebraN∞
Example 3.9. This is a generalization of Example 2.5(3) and Cuntz's construction of the
n=1 Mm, and let B = Um ⊕· · ·⊕Um
be the direct sum of n copies of Um, that is a nuclear unital weakly divisible C ∗-algebra of real
rank zero that absorbs Z and hence has strict comparison. Let us consider the endomorphism
THE STRUCTURE OF CROSSED PRODUCTS BY ENDOMORPHISMS
19
β : B −→ B given by β(x1, . . . , xn) = (P1 ⊗x2, P2 ⊗x3 · · · , Pn ⊗x1) for every (x1, . . . , xn) ∈ B,
where P1, · · · , Pn ∈ Mm are rank 1 projections. Hence, β is injective. Observe that β(1) 6= 1
is a full projection of B. It is clear that B is β-simple and βk is outer for any k > 0, since
B is a unital finite C ∗-algebra. Hence, B ×β N is simple by Theorem 1.17, and thus applying
Theorem ?? it is also a purely infinite C ∗-algebra, in particular it is a Kirchberg algebra.
Now, we use the modification of the Pimsner-Voiculescu six-term exact sequence given in
[29],
K0(B)
1−β ∗
/ K0(B)
/ K0(B ×β N)
K1(B ×β N)
K1(B)
1−β ∗
K1(B)
Notice that the induced map β∗ : Z[1/m]n −→ Z[1/m]n is given by
β∗(x1, . . . , xn) = (x2/m, . . . , xn/m, x1/m) ,
for every (x1, . . . , xn) ∈ Z[1/m]n. Then, we can easily compute K0(B ×β N) = Z/(mn − 1)Z
and K1(B ×β N) = 0. Thus, using the Kirchberg-Phillips classification theorems, we conclude
that B ×β N is stably isomorphic to the Cuntz algebra Omn.
Acknowledgments
Part of this work was done during visits of the first author to the Department of Mathe-
matics and Statistics of the University of Otago (New Zealand), and of the second author to
the Institutt for Matematiske Fag, Norges Teknisk-Naturvitenskapelige Universitet (Norway).
Both authors thank the host centers for their kind hospitality.
References
[1] B. Abadie, S. Eilers, R. Exel, Morita equivalence for crossed products by Hilbert C ∗-bimodules.
Trans. Amer. Math. Soc. 350 (1998), no. 8, 3043-3054.
[2] G. Abrams and G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005),
319 -- 334.
[3] S. Adji, M. Laca, M. Nielsen and I. Raeburn, Crossed products of endomorphisms and the Toeplitz
algebras of ordered groups. Proc. Amer. Math. Soc. vol. 112, no. 4 (1994), 1133 -- 1141.
[4] A. an Huef and I. Raeburn, Exel and Stacey crossed products and Cuntz Pimsner algebras. Private
notes.
[5] A. an Huef and I. Raeburn, The ideal structure of Cuntz-Krieger algebras. Ergodic Theory Dynam.
Systems 17 (1997), no. 3, 611 -- 624.
[6] P. Ara, M.A. Gonzalez-Barroso, K.R. Goodearl and E. Pardo, Fractional skew monoid rings,
J. Algebra 278 (2004), no. 1, 104 -- 126.
[7] W. B. Arveson, Continuous analogues of Fock space, Mem. Amer. Math. Soc, vol. 409, Amer. Math.
Soc, Providence, RI, 1989.
[8] T. Bates, D. Pask, I. Raeburn and W. Szyma´nski, The C ∗-algebras of row-finite graphs. New York
J. Math. 6 (2000), 307 -- 324
[9] T. Bates, J.H. Hong, I. Raeburn and W. Szyma´nski. The ideal structure of the C ∗-algebras of
infinite graphs. Illinois J. Math. 46 (2002), no. 4, 1159 -- 1176.
[10] E. Blanchard, R. Rohde and M. Rørdam. Properly infinite C(X)-algebras and K1-injectivity. J.
Noncommutative Geometry 2 (2008), no. 3, 263 -- 282.
[11] S. Boyd, N. Keswani and I. Raeburn, Faithful representations of crossed products of endomorphisms.
Proc. Amer. Math. Soc. vol. 118, no. 2 (1993), 427 -- 436.
/
/
O
O
o
o
o
o
20
EDUARD ORTEGA AND ENRIQUE PARDO
[12] J. Cuntz, Simple C -algebras generated by isometries, Comm. Math. Phys. 57 (1977), 173 -- 185.
[13] J. Cuntz, The internal structure of simple C ∗-algebras, Proc. Sympos. Pure Math., vol. 38, Amer.
Math. Soc, Providence, RI, 1982, pp. 85 -- 115.
[14] S. Doplicher and J. E. Roberts, Endomorphisms of C ∗ -algebras, cross products and duality for
compact groups, Ann. of Math. (2) 130 (1989), 75 -- 119.
[15] G. A. Elliott, Some simple C ∗-algebras constructed as crossed products with discrete outer automor-
phism groups, Publ. Res. Inst. Math. Sci. 16 (1980), no. 1, 299 -- 311.
[16] J. Jeong, K. Kodaka and H. Osaka, Purely Infinite Simple C ∗-Crossed Products II, Canad. Math.
Bull. 39 vol.2 (1996), 203 -- 210.
[17] T. Katsura, On C ∗-algebras associated with C ∗-correspondences. J. Funct. Anal. 217 (2004), no. 2,
366 -- 401.
[18] T. Katsura, Ideal structure of C ∗-algebras associated with C ∗-correspondences. Pacific J. Math. 230
(2007), no. 1, 107 -- 145.
[19] R. Kirchberg and M. Rørdam, Non-simple purely infinite C ∗-algebras. Amer. J. Math. 122 (2000),
no. 3, 637 -- 666.
[20] Kwa´sniewski, Cuntz-Krieger uniqueness theorem for crossed products by Hilbert bimodules. Preprint.
[21] P. S. Muhly and B. Solel, On the simplicity of some Cuntz Pimsner algebras, 83 (1998), 53 -- 73.
[22] D. Olesen, Inner ∗-Automorphisms of Simple C ∗-Algebras, Comm. Math. Phys. 44 (1975), 175 -- 190.
[23] D. Olesen and G.K. Pedersen, Applications of the Connes Spectrum to C ∗-dynamical systems, I, J.
Funct. Anal. 30 (1978), 179 -- 197.
[24] D. Olesen and G.K. Pedersen, Applications of the Connes Spectrum to C ∗-dynamical systems, II,
J. Funct. Anal. 36 (1980), 18 -- 32.
[25] D. Olesen and G.K. Pedersen, Applications of the Connes Spectrum to C ∗-dynamical systems, III,
J. Funct. Anal. 45 (1982), no. 3, 357 -- 390.
[26] W. L. Paschke, The crossed product by an endomorphism, Proc. Amer. Math. Soc. 80 (1980), 113 -- 118.
[27] M.V. Pimsner, A class of C ∗-algebras generalizing both Cuntz -- Krieger algebras and crossed products
by Z, in: D. Voiculescu (Ed.), Free Probability Theory, Fields Inst. Communication, Vol. 12, American
Mathematical Society, Providence, RI, 1997, pp. 189 -- 212.
[28] I. Raeburn, Graph algebras. CBMS Regional Conference Series in Mathematics, 103. American Math-
ematical Society, Providence, 2005.
[29] M. Rørdam, Classification of certain infinite simple C ∗-algebras. J. Funct. Anal. 131, (1995), 415 -- 458.
[30] M. Rørdam, The stable and the real rank of Z-absorbing C ∗-algebras. Internat. J. Math. 15 (2004),
no. 10, 1065 -- 1084.
[31] M. Rørdam and F. Perera, AF-embeddings into C ∗-algebras of real rank zero. J. Funct. Anal. 217
(2004), no. 1, 142 -- 170.
[32] J. Schweizer, Dilations of C ∗-correspondences and the simplicity of Cuntz-Pimsner algebras. J. Funct.
Anal. 180 (2001), no. 2, 404 -- 425.
[33] P.J. Stacey , Crossed products of C ∗-algebras by ∗-endomorphisms. J. Austral. Math. Soc. (Seres A)
54 (1993), no. 2, 204 -- 212.
[34] S. Zhang , A Riesz decomposition property and ideal structure of multiplier algebras. J. Operator
Theory 24 (1990), no. 2, 209 -- 225.
[35] S. Zhang , A property of purely infinite simple C ∗-algebras. Proc. Amer. Math. Soc. 109 (1990), no.
3, 717 -- 720.
Department of Mathematical Sciences, NTNU, NO-7491 Trondheim, Norway
E-mail address: [email protected]
Departamento de Matem´aticas, Facultad de Ciencias, Universidad de C´adiz, Campus de
Puerto Real, 11510 Puerto Real (C´adiz), Spain.
E-mail address: [email protected]
URL: https://sites.google.com/a/gm.uca.es/enrique-pardo-s-home-page/
|
1806.10983 | 2 | 1806 | 2018-08-29T14:37:38 | Uniqueness questions for C*-norms on group rings | [
"math.OA",
"math.GR"
] | We provide a large class of discrete amenable groups for which the complex group ring has several C*-completions, thus providing partial evidence towards a positive answer to a question raised by Rostislav Grigorchuk, Magdalena Musat and Mikael R{\o}rdam. | math.OA | math |
UNIQUENESS QUESTIONS FOR C*-NORMS ON GROUP RINGS
VADIM ALEKSEEV AND DAVID KYED
Abstract. We provide a large class of discrete amenable groups for which the complex
group ring has several C*-completions, thus providing partial evidence towards a positive
answer to a question raised by Rostislav Grigorchuk, Magdalena Musat and Mikael Rørdam.
1. Introduction
The interplay between group theory and operator algebras dates back to the seminal papers
by Murray and von Neumann [MvN36] and by choosing different completions of a discrete
countable group Γ one obtains interesting analytic objects; for instance the Banach algebra
ℓ1(Γ), the full and reduced C*-algebras C*(Γ) and C*
r (Γ), and the group von Neumann algebra
LΓ. In general there are many norms on, say, ℓ1(Γ) such that the completion with respect to
this norm gives a C*-algebra, and the question of when the C*-completion is unique (in which
case Γ is said to be C*-unique) has been studied by various authors [LN04, Boi84, Bar83]. A
C*-unique discrete group is evidently amenable and it is, to the best of the authors' knowledge,
an open question whether the converse is true, although it is known to be false in the more
general context of locally compact groups [LN04]. More recently, the paper [GMR18] put
emphasis on the question of when the complex group algebra CΓ has a unique C*-completion.
As is easily seen [GMR18, Proposition 6.7], if Γ is locally finite (i.e. if every finitely generated
subgroup is finite) then CΓ has a unique C*-completion, and [GMR18, Question 6.8] asks if
the converse is true. The present paper provides partial evidence towards a positive answer
to this, in that we prove that for the following classes of non-locally finite groups have several
C*-completions.
Theorem A (see Proposition 2.4 and Corollary 3.7). The class of countable groups Γ for
which CΓ does not have a unique C*-norm includes the following:
(i) Infinite groups of polynomial growth.
(ii) Torsion free, elementary amenable groups with a non-trivial, finite conjugacy class.
(iii) Groups with a central element of infinite order.
The key to the proof of of (i) and (ii) is the so-called strong Atiyah conjecture (see Sec-
tion 3.1) which predicts a concrete restriction on the von Neumann dimension of kernels of
2010 Mathematics Subject Classification. 16S34, 46L05, 46L10.
Key words and phrases. Group rings, C*-norms, the Atiyah conjecture.
1
2
VADIM ALEKSEEV AND DAVID KYED
elements in the complex group algebra under the left regular representation -- notably these
are predicted to be either zero or one if the group in question is torsion free.
Acknowledgements. The authors would like to thank Mikael Rørdam and Thomas Schick
for helpful discussions revolving around the topics of the present paper as well as the anony-
mous referee for the valuable remarks improving Corollary 2.3 and Proposition 3.3. DK
gratefully acknowledge the financial support from the Villum Foundation (grant no. 7423)
and from the Independent Research Fund Denmark (grant no. 7014-00145B).
2. Basic results on C*-uniqueness
In what follows, all discrete groups are implicitly assumed to be at most countable. We
will use several operator algebras associated to a discrete group Γ: the maximal C*-algebra
C*(Γ), the reduced C*-algebra C*
r (Γ) and the von Neumann algebra LΓ. For more information
on these, we refer to [BO08, §2.5]. We recall that LΓ = (λ(CΓ))′′ ⊂ B(ℓ2Γ) is generated by
the left regular representation λ : CΓ → B(ℓ2Γ) and carries a canonical, faithful, normal trace
given by τ (x) = hxδe , δei. In what follows, tr will denote the normalized trace on Mn(C)
while Tr will denote the non-normalized trace.
We begin by formally introducing the notion of C*-uniqueness. In order to avoid a notational
conflict with the already existing notions studied in [LN04, Boi84], we emphasize that we are
investigating the uniqueness of C*-norms on the complex group algebra in contrast to the
ℓ1-algebra.
Definition 2.1. Let Γ be a discrete group. CΓ is said to be:
(i) C*-unique if it carries a unique C*-norm;
(ii) C*
r -unique if no C*-norm on CΓ is properly majorised by the reduced C*-norm.
Γ is said to be algebraically C*- (respectively C*
r -)unique if CΓ is C*- (respectively C*
r -)unique.
Amenable groups are characterized by the property that the maximal and reduced C*-
algebras coincide, and thus a nonamenable group is never algebraically C*-unique; on the
other hand, for amenable groups the above notions coincide. Note also that the class of C*-
simple groups, which has recently received a lot of attention [BKKO17, LB17], falls within the
class of algebraically C*
r -unique groups. As already mentioned in the introduction, algebraic
C*-uniqueness appeared in the recent paper [GMR18] in which the authors observed that
locally finite groups have this property and asked if this characterizes the class of locally finite
groups. Below we prove a few basic permanence results regarding algebraic C*-uniqueness,
but before doing so we give an alternative characterization, which is straightforward algebraic
adaptation of the similar result for ℓ1-algebras [Bar83, Proposition 2.4].
Lemma 2.2. Let Γ be a discrete group. Then CΓ is C*-unique (respectively C*
and only if every nontrivial closed, two-sided ideal in C*(Γ) (respectively C*
non-trivially.
r -unique) if
r (Γ)) intersects CΓ
UNIQUENESS QUESTIONS FOR C*-NORMS ON GROUP RINGS
3
Proof. We give the proof for the statement about algebraic C*-uniqueness; the other case is
obtained by replacing C*(Γ) by C*
r (Γ) throughout the proof. Assume that there is a non-
trivial ideal J P C*(Γ) intersecting CΓ trivially and denote by q : C*(Γ) → C*(Γ)/J the
quotient map. Composing q with the inclusion CΓ ֒→ C*(Γ) yields a faithful representation
of CΓ and it defines a C*-norm on it that is properly majorised by the maximal norm by
non-triviality of J. Conversely, if there is a C*-norm on CΓ which is properly majorised by
the norm coming from C*(Γ), then C*(Γ) surjects onto the corresponding quotient, and the
kernel of this surjection is a non-trivial ideal intersecting CΓ trivially.
(cid:3)
Corollary 2.3. Let Γ and Λ be discrete groups. If C(Γ × Λ) is C*-unique (respectively C*
r -
unique), then so are CΓ and CΛ.
r (Γ) be a non-trivial ideal intersecting CΓ trivially. Then J ⊗max C*
Proof. Let J P C*
r (Λ) P
C*(Γ) ⊗max C*(Λ) = C*(Γ × Λ) is a non-trivial ideal intersecting C(Γ × Λ) = CΓ ⊗alg CΛ
trivially. The same proof with C* replaced by C*
r and ⊗max replaced by ⊗min works for the
reduced case.
(cid:3)
Proposition 2.4. If Γ is a discrete group with a central element of infinite order then CΓ is
not C*
r -unique.
Proof. Denote by Z the subgroup in Γ generated by a central element of infinite order. Then
r (Z) ∼= C(S1) and LZ ∼= L∞(S1) via the Fourier transform and we denote by p ∈ LZ the
C*
projection corresponding to the characteristic function of the upper half circle {eiθ θ ∈ [0, π]}.
Define π := λΓp; i.e. the left regular representation of Γ restricted to the invariant subspace
pℓ2(Γ). Choosing a non-zero function f ∈ C(S1) supported in the lower half circle we obtain
a non-zero element x ∈ C*
r (Γ) induced
by π is not the one induced by λΓ. We now only need to see that π is faithful on CΓ. To
this end, consider the trace-preserving conditional expectation E : LΓ → LZ [BO08, Lemma
1.5.11] and assume that a ∈ CΓ is in the kernel of π. Then a∗a is also in the kernel of π and
since E is an LZ-bimodule map [BO08, Proposition 1.5.7] we get
r (Γ) with xp = 0 and hence the norm on C*
r (Z) ⊂ C*
0 = E(λΓ(a∗a)p) = E(λΓ(a∗a))p.
However, E(CΓ) ⊂ CZ ≃ Pol(z, ¯z) ⊂ C(S1) and therefore E(λΓ(a∗a)) = 0 and since E is
trace-preserving and the trace on LΓ is faithful we conclude that a∗a, and hence a, is zero. (cid:3)
Corollary 2.5. An abelian group is algebraically C*-unique if and only if it is locally finite
(i.e. pure torsion).
Remark 2.6. The result in Corollary 2.5 was also observed, independently and with different
proofs, by Rostislav Grigorchuk, Magdalena Musat and Mikael Rørdam (unpublished).
Remark 2.7. The class of locally finite groups has many stability properties -- for instance
it is closed under subgroups, quotients and extensions and, moreover, being virtually locally
4
VADIM ALEKSEEV AND DAVID KYED
finite is the same as being locally finite. However, verifying these properties for the class of
C*-unique groups seems to be a much bigger challenge.
3. The strong Atiyah conjecture and C*
r -uniqueness
3.1. The strong Atiyah conjecture. The key to our main result is the so-called strong
Atiyah conjecture which is briefly described in the following. A good general reference is
[Lüc02, Chapter 10] where all of the results below can be found, and to which we also refer
for the original references. Let Γ be a discrete group and denote by
Z the additive
subgroup in Q generated by the set
FIN(Γ)
1
Λ 6 Γ a finite subgroup(cid:27) .
(cid:26) 1
Λ (cid:12)(cid:12)(cid:12)
Given a matrix A ∈ Mn(CΓ) we denote by LA ∈ LΓ ⊗ Mn(C) ⊂ B(ℓ2(Γ)n) the bounded
operator given by left multiplication with A (via the left regular representation of Γ). The
strong Atiyah conjecture then predicts that
dimLΓ ker(LA) := (τ ⊗ Tr)(Pker LA) ∈
1
FIN(Γ)
Z.
Here dimLΓ(−) denotes the von Neumann dimension of the (right) Hilbert LΓ-module ker(LA)
defined as the non-normalized trace of the kernel projection Pker LA; see [Lüc02] for details on
this. It should be noted that the strong Atiyah conjecture is false in general [Lüc02, Theorem
10.23], but is known to hold for all groups which have a bound on the order of finite subgroups
and belong to Linnell's class C [Lüc02, Theorem 10.19], the latter being the smallest class
of groups which contain all free groups, is closed under directed unions and extensions by
elementary amenable groups (i.e., if Λ P Γ, Λ ∈ C and Γ/Λ is elementary amenable, then
Γ ∈ C ). The above discussion motivates the following notion.
Definition 3.1. Let Γ be a countable group. The torsion multiplier of Γ is defined as
θ(Γ) =
1
lcm{H H 6 Γ finite}
∈ [0, 1].
In this definition, and in what follows, we use the convention that the least common multiple
∞ = 0. Note that if Γ has an
(lcm) of an infinite set of natural numbers is infinity and that 1
upper bound on the set of finite subgroups, then
1
FIN(G)
Z = {nθ(Γ) n ∈ Z} ,
1
FIN(G) Z has 0 as an accumulation point otherwise. In view of this, the strong Atiyah
and
conjecture for a group Γ with θ(Γ) > 0 implies that the possible kernel dimensions are properly
quantized in the sense that they can only take values in the discrete set {nθ(Γ) n ∈ N} ⊂
R. Theorem A (i) and (ii) will follow directly from our main technical result, Theorem 3.6
below. The key idea in the proof is to play the aforementioned "quantization" of the kernel
UNIQUENESS QUESTIONS FOR C*-NORMS ON GROUP RINGS
5
dimensions against an abundance of central projections in LΓ with small traces which provide
representations of C*
r (Γ) with non-trivial kernels. To quantify this, we need the following
definition.
Definition 3.2. The central granularity of Γ is defined as
σ(Γ) = inf{τ (p) p ∈ Proj(Z(LΓ)), p 6= 0} ∈ [0, 1].
We note that σ(Γ) < 1 if and only if Z(LΓ) is nontrivial which is equivalent Γ not being
icc1. The next proposition computes the central granularity of Γ in group-theoretic terms.
Recall that the FC-centre Γfc is the normal subgroup of Γ consisting of all elements with finite
conjugacy classes.
Proposition 3.3. Let Γfc P Γ be the FC-centre of Γ. Then
σ(Γ) =
1
Γfc
,
where the right-hand side is interpreted as 0 if Γfc = ∞.
Proof. Γfc is an increasing union of a sequence of finitely generated normal subgroups Λn P Γ;
to see this, note that Γfc is clearly an increasing union of a sequence of finitely generated
subgroups Λ′
n, and defining Λn to be generated by the Γ-conjugacy classes of a finite system
of generators for Λ′
n yields the desired sequence of finitely generated subgroups which are
normal in Γ. We now have two cases to consider:
(i) all Λn are finite (equivalently, Γfc is a torsion group),
(ii) Λn is infinite for some n.
Λn Pg∈Λn g, we get a projection pn ∈ LΓfc with τ (pn) = 1
In case (i), setting pn := 1
Λn ;
moreover, pn is central in LΓ since Λn is normal in Γ. This proves that σ(Γ) = 0 if Γfc is an
infinite torsion group (in this case Λn → ∞). If Γfc is finite, then the sequence stabilizes,
1
and therefore we get a central projection p in LΓ with trace
Γfc . The centre of LΓ consists of
elements whose associated Fourier series in ℓ2(Γ) = L2(LΓ, τ ) are supported only on Γfc and
are constant along conjugacy classes, and is therefore contained in the centre of LΓfc; hence
> σ(Γ) > σ(Γfc). But we also have LΓfc = CΓfc which by representation theory
we get
1
Γfc
of finite groups is isomorphic to a direct sum of matrix algebras Lπ Mdπ (C) with the trace
1
given by Lπ
Γfc = σ(Γfc); this proves
d2
Γfc tr; thus, the minimal central projection has trace
the claim.
π
In case (ii) we fix an n ∈ N such that Λn =: Λ is infinite and note that since Λ is generated
by a finite number of elements with finite conjugacy classes, its centralizer CΓ(Λ) is of finite
index in Γ. We now claim that LΛ has a diffuse von Neumann subalgebra and thus projections
of arbitrarily small trace. This can be seen as follows: if LΛ has a direct summand of type
II1, it is clear because such von Neumann algebras are diffuse. Otherwise LΛ is of type I,
1recall that a group is icc if all non-trivial conjugacy classes are infinite
6
VADIM ALEKSEEV AND DAVID KYED
but then Λ is virtually abelian [Lüc97, Lemma 3.3], and hence, being infinite by assumption
and finitely generated by construction, contains a copy of Z which generates a diffuse von
Neumann algebra LZ ∼= L∞(S1).
In view of the above, for an arbitrary ε > 0 there is a
projection p ∈ LΛ ⊂ LΓfc of trace τ (p) <
ε
[Γ:CΓ(Λ)] . Now let
q := _g∈Γ
gp,
where gp := gpg−1. Then q is a central projection in LΓ. Moreover, p is invariant under the
centralizer CΓ(Λ) and upon choosing coset representatives g1, . . . , g[Γ:CΓ(Λ)] for Γ/CΓ(Λ) we
obtain that
[Γ:CΓ(Λ)]
q =
gip
_i=1
and hence τ (q) 6 [Γ : CΓ(Λ)] · τ (p) < ε. Thus σ(Γ) = 0.
(cid:3)
Lemma 3.4. Let Γ be a discrete non-icc group. For every ε > 0 there exists a nonzero
projection p ∈ Z(LΓ) with τ (p) < σ(Γ) + ε and a nonzero, central element x ∈ C*
r (Γ) with
xp⊥ = 0.
Proof. Since Γ is non-icc, Z(LΓ) 6= C1 so σ(Γ) < 1. Let ε > 0 be given and assume, without
r (Γ))′′ = Z(CΓ)′′, as can bee seen
loss of generality, that σ(Γ) + ε < 1. One has Z(LΓ) = Z(C*
for instance by using Kaplansky's density theorem together with the center valued trace, and
noting that Z(CΓ) consists of the elements whose coefficients are constant along conjugacy
classes. By Gelfand duality, Z(C*
r (Γ)) is isomorphic to the C*-algebra C(Z) of continuous
functions on its Gelfand spectrum Z, which is a compact Hausdorff space; it is metrizable
because C*
r (Γ) is separable. The canonical trace τ thus gives a regular Borel probability
r (Γ))′′ ∼= L∞(Z, µ)
measure µ on Z [Rud66, Theorem 2.14] and an isomorphism Z(LΓ) = Z(C*
compatible with the natural inclusions. Projections in Z(LΓ) correspond via this isomorphism
to measurable subsets of Z (up to null sets), and we therefore obtain a measurable subset
A ⊂ Z such that 0 < µ(A) < σ(Γ) + ε/2. By regularity of µ, there exists U ⊇ A open such
that
0 < µ(A) 6 µ(U ) < σ(Γ) + ε < 1.
Now, there is a non-zero element x ∈ C(Z) vanishing on the compact set K := Z \ U (for
instance, the distance function to K); letting p be the projection corresponding to U finishes
the proof.
(cid:3)
The following lemma gives a concrete description of the decomposition of the left regular
representation of a discrete group Γ over the cosets of a finite index normal subgroup Λ.
Lemma 3.5. Let Λ P Γ be a normal subgroup of finite index. For every choice of coset
representatives g1, . . . , g[Γ:Λ] ∈ Γ there exists a trace-preserving inclusion of von Neumann
algebras π : (LΓ, τ ) ֒→ (M[Γ:Λ](LΛ), τ ⊗ tr) which restricts to corresponding inclusions at the
UNIQUENESS QUESTIONS FOR C*-NORMS ON GROUP RINGS
7
level of reduced C*-algebras and complex group rings, and which for x ∈ LΛ is given by
π(x) = diag(g1x, g2x, . . . , g[Γ:Λ]x),
(3.1)
where gx = gxg−1 is the conjugation action of g ∈ Γ on LΛ.
Proof. Choose coset representatives g1, g2, . . . , g[Γ:Λ] of Γ/Λ and consider the isomorphisms of
Hilbert spaces
ℓ2(Γ) ∼=
[Γ:Λ]
Mi=1
ℓ2(g−1
i Λ) ∼=
[Γ:Λ]
Mi=1
ℓ2(Λ).
These induce a ∗-isomorphism π : B(ℓ2Γ)
It is routine to check that π
restricts to a trace-preserving inclusion of CΓ into M[Γ:Λ](CΛ) which automatically implies
the corresponding results for the reduced C*-algebras and von Neumann algebras. Finally, for
h ∈ Λ we have
∼=−→ Mn(B(ℓ2(Λ))).
[Γ:Λ]
π(h) = diag(λ(h), . . . , λ(h)) ∈
B(ℓ2(g−1
i Λ)),
Mi=1
and thus formula (3.1) follows in view of the identity
hg−1
i h′ = g−1
i
(gih)h′,
h, h′ ∈ Λ.
(cid:3)
Theorem 3.6. Let Λ be a discrete group satisfying the strong Atiyah conjecture and let Λ P Γ
be a finite index inclusion of Λ into a group Γ as a normal subgroup. If [Γ : Λ]2 · σ(Λ) < θ(Λ)
then CΓ is not C*
r -unique.
Proof. The assumption [Γ : Λ]2 · σ(Λ) < θ(Λ) forces Λ to be non-icc and applying Lemma 3.4
we get a projection p ∈ Z(LΛ) with τ (p) < θ(Λ)
r (Λ)
with xp⊥ = 0. We are going to construct a representation of C*
r (Γ) which is injective on CΓ
but with x in the kernel. To this end, consider a set of coset representatives g1, . . . , g[Γ:Λ] for
Γ/Λ and the ∗-homomorphism π : LΓ → M[Γ:Λ](LΛ) provided by Lemma 3.5. From this we
gip ∈ Z(LΛ), and cutting π with the complement of
[Γ:Λ]2 and a non-zero central element x ∈ C*
obtain a central projection q := W[Γ:Λ]
i=1
q := diag(q, . . . , q) ∈ Z(M[Γ:Λ](LΛ)), we get a representation
πq : C*
r (Γ) → B(cid:16)ℓ2(Λ)[Γ:Λ] q⊥(cid:17) ,
a 7→ π(a)q⊥.
i=1
As q⊥ = V[Γ:Λ]
gi(p⊥) and xp⊥ = 0, it follows that x ∈ ker πq in view of (3.1). Let a ∈
CΓ ∩ ker πq. This means that π(a)q⊥ = 0, and thus the kernel projection r of π(a) satisfies
r > q⊥. Therefore
(τ ⊗ Tr)(r) > (τ ⊗ Tr)(q⊥) > [Γ : Λ](1 − [Γ : Λ]τ (p)) > [Γ : Λ] − θ(Λ).
On the other hand, the assumption [Γ : Λ]2 · σ(Λ) < θ(Λ) forces an upper bound on the order
of finite subgroups in Λ, i.e. θ(Λ) > 0, and since Λ is furthermore assumed to satisfy the strong
8
VADIM ALEKSEEV AND DAVID KYED
Atiyah conjecture we obtain (using the notation of Section 3.1) that
dimLΛ(ker(LA)) = (τ ⊗ Tr)(Pker LA) ∈ {nθ(Λ) n ∈ Z}
for any matrix A ∈ M[Γ:Λ](CΛ). Thus (τ ⊗ Tr)(r) 6 [Γ : Λ] − θ(Λ) unless π(a) = 0. This
proves that πq is injective on CΓ and hence completes the proof.
(cid:3)
As a corollary, we deduce that some important families of groups are not C*
r -unique. In
particular this includes the groups mentioned in Theorem A (i) and (ii), and together with
Proposition 2.4 this completes the proof of Theorem A.
Corollary 3.7. All groups in following classes are not C*
r -unique:
(i) Torsion free, non-icc groups satisfying the strong Atiyah conjecture; in particular all
elementary amenable, non-icc, torsion free groups.
(ii) Virtually polycyclic groups with infinite FC-centre; in particular, all infinite groups of
polynomial growth.
Proof. To see (i), note that the existence of a non-trivial finite conjugacy class implies the
existence of a non-trivial central element in CΓ (namely the sum of the elements in the finite
conjugacy class) and hence a non-trivial projection in Z(LΓ); thus σ(Γ) < 1. Moreover, since
Γ is torsion free, θ(Γ) = 1 and since Γ is assumed to satisfy the strong Atiyah conjecture it
follows C*
r -unique by Theorem 3.6. The last statement in (i) follows directly from this since
the elementary amenable groups are contained in Linnell's class C (see Section 3.1) for which
the strong Atiyah conjecture is known to hold in the presence of a bound on the order of finite
subgroups [Lüc02, Theorem 10.19].
To see (ii), let Λ P Γ be a normal finite index polycyclic subgroup of Γ. As Γ has infinite
FC-centre, so does Λ and the FC-centre of Λ is moreover finitely generated by polycyclicity. A
classical result by Hirsch [Hir46, Theorem 3.21] implies that the orders of finite subgroups of Λ
are bounded; thus θ(Λ) > 0. On the other hand, σ(Λ) = 0 by Proposition 3.3 (ii). Moreover,
polycyclic groups, being elementary amenable, satisfy the strong Atiyah conjecture. Thus, CΓ
follows non-C*
r unique by Theorem 3.6.
Finally, the claim about infinite groups of polynomial growth follows by first observing that
by Gromov's theorem [Gro81] these are exactly finitely generated virtually nilpotent groups.
As finitely generated nilpotent groups are polycyclic, the claim follows once we argue that
virtually nilpotent groups automatically have infinite FC-centre. To see this, recall that a
finitely generated virtually nilpotent group Γ contains a finite index torsion free nilpotent
normal subgroup Λ (by polycyclicity and [Hir46, Theorem 3.21]). Now it follows that the
centre of Λ is infinite, and therefore so is the FC-centre Λfc; but as Λ P Γ is a finite index
inclusion, Λfc ⊆ Γfc. Thus, Γfc is infinite.
(cid:3)
UNIQUENESS QUESTIONS FOR C*-NORMS ON GROUP RINGS
9
References
[Bar83]
Bruce A. Barnes. The properties ∗-regularity and uniqueness of C ∗-norm in a general ∗-algebra.
Trans. Amer. Math. Soc., 279(2):841 -- 859, 1983.
[BO08]
[BKKO17] Emmanuel Breuillard, Mehrdad Kalantar, Matthew Kennedy, and Narutaka Ozawa. C ∗-simplicity
and the unique trace property for discrete groups. Publ. Math. Inst. Hautes Études Sci., 126:35 -- 71,
2017.
Nathanial P. Brown and Narutaka Ozawa. C ∗-algebras and finite-dimensional approximations, vol-
ume 88 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI,
2008.
Joachim Boidol. Group algebras with a unique C ∗-norm. J. Funct. Anal., 56(2):220 -- 232, 1984.
[Boi84]
[GMR18] Rostislav Grigorchuk, Magdalena Musat, and Mikael Rørdam. Just-infinite C ∗-algebras. Comment.
[Gro81]
[Hir46]
[LB17]
[LN04]
Math. Helv., 93(1):157 -- 201, 2018.
Mikhael Gromov. Groups of polynomial growth and expanding maps. Inst. Hautes Études Sci.
Publ. Math., (53):53 -- 73, 1981.
Kurt A. Hirsch. On infinite soluble groups. III. Proc. London Math. Soc. (2), 49:184 -- 194, 1946.
Adrien Le Boudec. C ∗-simplicity and the amenable radical. Invent. Math., 209(1):159 -- 174, 2017.
Chi-Wai Leung and Chi-Keung Ng. Some permanence properties of C ∗-unique groups. J. Funct.
Anal., 210(2):376 -- 390, 2004.
[Lüc97] Wolfgang Lück. Hilbert modules and modules over finite von Neumann algebras and applications
to L2-invariants. Math. Ann., 309(2):247 -- 285, 1997.
[Lüc02] Wolfgang Lück. L2-invariants: theory and applications to geometry and K-theory, volume 44 of
Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Math-
ematics. Springer-Verlag, Berlin, 2002.
Francis J. Murray and John von Neumann. On rings of operators. Ann. of Math. (2), 37(1):116 -- 229,
1936.
[MvN36]
[Rud66] Walter Rudin. Real and complex analysis. McGraw-Hill Book Co., New York, 1966.
Vadim Alekseev, Technische Universität Dresden, Fakultät Mathematik, Institut für Ge-
ometrie, 01062 Dresden, Germany
E-mail address: [email protected]
David Kyed, Department of Mathematics and Computer Science, University of Southern
Denmark, Campusvej 55, 5230 Odense M, Denmark
E-mail address: [email protected]
|
1211.5101 | 1 | 1211 | 2012-11-21T17:51:03 | Real Operator Algebras and Real Completely Isometric Theory | [
"math.OA",
"math.FA"
] | This paper is a continuation of the program started by Ruan in 2003, of developing real operator space theory. In particular, we develop the theory of real operator algebras. We also show among other things that the injective envelope, C*-envelope and non-commutative Shilov boundary exist for a real operator space. We develop real one-sided M-ideal theory and characterize one-sided M-ideals in real C*-algebras and real operator algebras with contractive approximate identity. | math.OA | math | REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC
THEORY
SONIA SHARMA
Abstract. This paper is a continuation of the program started by Ruan in [11] and [12], of
developing real operator space theory. In particular, we develop the theory of real operator algebras.
We also show among other things that the injective envelope, C ∗-envelope and non-commutative
Shilov boundary exist for a real operator space. We develop real one-sided M -ideal theory and
characterize one-sided M -ideals in real C ∗-algebras and real operator algebras with contractive
approximate identity.
2
1
0
2
v
o
N
1
2
]
.
A
O
h
t
a
m
[
1
v
1
0
1
5
.
1
1
2
1
:
v
i
X
r
a
1. Introduction
In functional analysis, the underlying objects of study are vector spaces over a field, where the
field is usually either the field of real numbers, R, or the field of complex numbers, C. The field of
complex numbers has been preferred over real numbers, since the field of real numbers is a little
more restrictive. For instance, every polynomial over the field of reals has a roots in C, but need
not have any root in R, or a n× n matrix need not have real eigenvalues. Thus, usually most of the
theory is developed with the assumption that the underlying field is C. The theory of real spaces,
however, occurs naturally in all areas of mathematics and physics. They come up naturally in the
theory of C ∗-algebras, for instance the self-adjoint part of every C ∗-algebra is a real space, also in
graded C ∗-algebras and in the theory of real TROs in graded C ∗-algebras. See [6, 9, 13, 16] for
the theory of real C ∗-algebras and real W ∗-algebras. They also come up in JB∗-triples [8] and KK
theory [1, 4]. Thus it becomes important to study the analogues theory for the case when the field
is the real scalars and know which results hold true and which results fail.
The theory of real operator spaces is the study of subspaces of bounded operators on real Hilbert
spaces. In the general theory of (complex) operator spaces, the underlying Hilbert space is assumed
to be a complex Hilbert space.
In two recent papers [11, 12], Ruan studies the basic theory of real operator spaces. He shows
that with appropriate modifications, many complex results hold for real operator spaces. It is shown
among other things, that Ruan's characterization, Stinespring's theorem, Arveson's extension the-
orem, and injectivity of B(H) for real Hilbert space H, hold true for real operator spaces. In [12],
Ruan defines the notion of complexification of a real operator space and studies the relationship
between the properties of real operator spaces and their complexification. We want to continue this
program, and develop more theory of real operator spaces and real operator algebras since there
appears to be a gap in the literature here. This is unfortunate because a researcher who is familiar
with the complex operator space theory and is facing a problem which involves real operator spaces,
must then face the daunting task of reconstructing a large amount of the real theory from scratch.
Date: June 19, 2018.
1
2
SONIA SHARMA
In section 2, we further develop the real operator space theory, investigate operator space struc-
tures such as minimal and maximal operator spaces structures on real operator spaces. We also see
that unlike in the Banach space theory (see [9, Proposition 1.1.6]), if X is a complex operator space
then (X ∗)r is not completely isometrically isomorphic to (Xr)∗, where Xr denotes X considered
as a real operator space. Also, unlike in the complex operator space theory, ℓ2(R) does not have a
unique operator space structure. In section 3, we briefly consider real operator algebras and their
complexification. We show that the BRS characterization theorem and Meyer's Theorem hold for
real operator algebras. In section 4, we study the relation between the real injective envelope and
the injective envelope of its complexification. We also show among other things that the injective
envelope, C ∗-envelope and non-commutative Shilov boundary exist for a real operator space. In
Section 5, we begin to develop the theory of real one-sided M -ideals and show that several results
from one-sided M -ideal theory [2], are true in the real case. We also show that a subspace J
in a real operator space X is a right M -ideal if and only if Jc is a (complex) right M -ideal in
Xc, which allows us to characterize one-sided M -ideals in real C ∗-algebras and in real operator
algebras with one-sided contractive approximate identity. We also infer that a real operator space
X is M -embedded if and only if Xc is M -embedded. This facilitates in generalizing results in
one-sided M -embedded theory from [15] to real operator spaces. For instance, we show that real
one-sided M -embedded TRO are of the form A ∼= ⊕◦
i,jK(Hi, Hj) completely isometrically, for some
real Hilbert spaces Hi, Hj.
Much of the work presented in this paper was done in author's thesis in 2009 (see [14]).
2. Real Operator Spaces
A (concrete) real operator space is a closed subspace of B(H), for some real Hilbert space H.
An abstract real operator space is a pair (X,k.kn), where X is a real vector space such that
there is a complete isometry u : X −→ B(H), for some real Hilbert space H. As in the case of
complex operator spaces, Ruan's norm characterization hold for real operator spaces, and we say
that (X,k.kn) is an abstract real operator space if and only if it satisfies
(i) kx ⊕ ykn+m = max{kxkn ,kykm},
(ii) kαxβkn ≤ kαkkxkn kβk,
for all x ∈ Mn(X), y ∈ Mm(X) and α, β ∈ Mn(R).
Let X ⊂ B(H), then Xc ⊂ B(H)c and B(H)c ∼= B(Hc) completely isometrically, where Hc is
a complex Hilbert space (see e.g. discussion on page 1051 from [12]). Thus there is a canonical
matrix norm structure on Xc inherited from B(Hc), and Xc is a complex operator space with this
canonical norm structure. The space B(Hc) can be identified with a real subspace of M2(B(H))
via
(2.1)
Thus the matrix norm on the complexification is given by
y
B(Hc) = B(H) + iB(H) =(cid:26)(cid:20) x −y
k[xkl + iykl]k =(cid:13)(cid:13)(cid:13)(cid:13)
Xc =(cid:26)(cid:20) x −y
x (cid:21) : x, y ∈ B(H)(cid:27) ∈ M2(B(H)).
xkl (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) xkl −ykl
x (cid:21) : x, y ∈ X(cid:27) ∈ M2(X).
ykl
,
and we have the following complete isometric identification
y
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
3
This canonical complex operator space matrix norm structure on its complexification Xc =
X + iX which extends the original norm on X, i.e., kx + i0kn = kxkn and satisfies the reasonable
(in the sense of [12]) condition
kx + iyk = kx − iyk ,
for all x + iy ∈ Mn(Xc) = Mn(X) + iMn(X) and n ∈ N. By [12, Theorem 2.1], the operator
space structure on Xc is independent of the choice of H. Moreover, by [12, Theorem 3.1], any
other reasonable in the above sense operator space structure on Xc is completely isometric to the
canonical operator space structure on Xc.
Let T be a (real linear) bounded operator between real operator spaces X and Y , then define the
complexification of T , Tc : Xc −→ Yc as Tc(x + iy) = T (x) + iT (y), a complex linear bounded oper-
ator. It is shown in [12, Theorem 3.1] that if T is a complete contraction (respectively, a complete
isometry) then Tc is a complete contraction (respectively, complete isometry) with kTckcb = kTkcb.
This is not true in the case of a Banach space, that is, the complexification of a contraction on a
real Banach space is bounded, but is not necessarily a contraction, and kTk 6= kTck, in general. If
π : Xc −→ Yc is linear, then as in [12], define a linear map π : Xc −→ Yc as π(x + iy) = π(x − iy).
Let Re(π) = π+π
. Then Re(π) and Im(π) are (real) linear maps which map
2
X into Y such that Re(π) = Re(π), Im(π) = Im(π), and π = Re(π) + iIm(π).
2.1. Minimal Real Operator Space Structure. A real C ∗-algebra is a norm closed ∗-subalgebra
of B(H), where H is a real Hilbert space. By [9, Proposition 5.13], every real C ∗-algebra A is a
fixed point algebra of (B,−), i.e., A = {b ∈ B : b = b}, where B is a (complex) C ∗-algebra, and
"-" is a conjugate linear ∗-algebraic isomorphism of B with period 2. Moreover, B = A + iA is the
complexification of A.
and let Im(π) = π−π
2i
Let A be a commutative real C ∗-algebra. Then define the spectral space of A as,
Ω = {ρA : ρ is nonzero multiplicative linear functional on Ac}.
In other words, Ω is the set of all non-zero complex valued multiplicative real linear functionals on
A. Then using the "−" on Ac, define "−" on Ω as,
ρ(a) = ρ(a).
Then every commutative real C ∗-algebra A is of the form
A ∼= C0(Ω,−) = {f ∈ C0(Ω) : f (t) = f (t) ∀ t ∈ Ω},
where Ω is the spectral space of A, and "−" is a conjugation on Ω defined above (see e.g. [9, 5.1.4]).
Also Ac = C0(Ω).
If Ω is any compact Hausdorff space then there is a canonical real C ∗-algebra,
C(Ω, R) = {f : Ω −→ R : f is continuous}.
For instance, if A is a commutative real C ∗-algebra and a ∈ A is self adjoint, then the real C ∗-
algebra generated by a in A, C ∗(a), is of the form C(Ω, R), where Ω is the spectral space of A. But
not every commutative real C ∗-algebra A is of the form C(Ω, R). To see this, let Ω = S2 ⊂ R3, the
3-dimensional sphere. Let A = {f : Ω −→ C : f (−t) = f (t) ∀ t ∈ Ω}. Then Ac = C(Ω), so A is a
real C ∗-algebra. But A is not ∗-isomorphic to C(Ω, R) since Asa = {f : Ω −→ R : f (t) = f (−t)} ≇
C(Ω, R).
We can define an operator space structure on C(Ω,−) by the canonical structure it inherits as
a subspace of C(Ω). Then C(Ω) is the operator space complexification of C(Ω,−), in the sense
4
SONIA SHARMA
defined above (see e.g., [9, Proposition 5.1.3]). Let E be a real Banach space. Then E can be
embedded isometrically into a real commutative C ∗-algebra A of the form C(Ω, R). For instance,
take Ω = Ball(E∗) where E∗ = {f : E −→ R, f continuous}. Since commutative real C ∗-algebras
are real operator spaces, there is an operator space matrix structure on E via the identification
Mn(E) ⊆ Mn(A). This operator space structure is called the minimal operator space structure
since it is the smallest operator space structure on E. To see this, let E be the operator space
sitting inside C(Ω, R) and F denote the Banach space E, with a different operator space structure.
Let u : F −→ E be the identity map. So u is an isometry, ku(x)kE = kxkE = kxkF , and for any
[xij] ∈ Mn(E) and Ω = Ball(E∗),
kun[xij]kMn(E) = k[u(xij)]kMn(E)
= sup{k[u(xij)(t)]kMn(R) : t ∈ Ω}
= sup{Xi,j
u(xij)(t)wj vi : ~v, ~w ∈ l2
= sup{ku(Xi,j
= sup{kXi,j
≤ k[xij]kMn(F ).
xijwjvi)kE : ~v, ~w ∈ l2
xijwjvikF : ~v, ~w ∈ l2
n(R)}
n(R), t ∈ Ω}
n(R)}
This implies that ku : F −→ Ekcb ≤ 1.
Let E be a Banach space and let x, y ∈ E. Define
kx + iyk = sup{kαx + βyk : α2 + β2 ≤ 1, α, β ∈ R}.
Then kx + iyk = kx − iyk and kx + i0k = kxk, and thus with this new norm Ec is a complexification
of the Banach space E. This norm is called the w2-norm in [5]. Also note that for any z + iw ∈ C,
z + iw = sup{αz + βw : α2 + β2 ≤ 1, α, β ∈ R}.
So,
kx + iyk = sup{αf (x) + βf (y) : α2 + β2 ≤ 1, α, β ∈ R and f ∈ Ball(E∗)}
= sup{f (x) + if (y)
: f ∈ Ball(E∗)}
= sup(cid:26)(cid:13)(cid:13)(cid:13)(cid:13)
f (y)
f (x) (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) f (x) −f (y)
: f ∈ Ball(E∗)(cid:27) .
Proposition 2.1. Let E be a real Banach space and Ec be the complexification of E with the norm
defined above. Then (Min(E))c = Min(Ec), completely isometrically.
Proof. Let π : E −→ C(Ω, R) be the canonical isometry. Then Min(π) : Min(E) −→ C(Ω, R) is a
complete isometry, and so, Min(π)c : Min(E)c −→ C(Ω) is a complete isometry. Further,
kπc(x + iy)k = sup{π(x)(f ) + iπ(y)(f ) : f ∈ Ω = Ball(E∗)}
= sup{f (x) + if (y) : f ∈ Ball(E∗)}
= sup{kαx + βyk : α2 + β2 ≤ 1, α, β ∈ R}
= kx + iykEc .
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
5
So πc : Ec −→ C(Ω) is an isometry, and hence Min(πc) : Min(Ec) −→ C(Ω) is a complete isometry.
So we have the following diagram which commutes.
Min(E)c
c.i.
C(Ω)
Id
Id
Min(Ec)
c.i.
/ C(Ω)
Hence (Min(E))c = Min(Ec), completely isometrically.
Lemma 2.2. Let A and B be real C ∗-algebras, and let π : A −→ B be a homomorphism. Then π
is a ∗-homomorphism if and only if it is completely contractive. Further, π is a complete isometry
if and only if it is one-one.
Proof. Let π : A −→ B be a ∗-homomorphism, then πc : Ac −→ Bc is a ∗-homomorphism. Hence
πc is a complete contraction, by [3, Proposition 1.2.4], so π = πcA is a complete contraction. A
similar argument using the complexification proves the converse. The last assertion follows from
[3, Proposition 1.2.4] and that πc is one-one if π is.
(cid:3)
(cid:3)
The following proposition has been noted in [12].
Proposition 2.3. Let X be a real operator space, then (Xc)∗ = (X ∗)c, completely isometrically.
Proposition 2.4. If X is a real operator space then X ⊂ X ∗∗ completely isometrically via the
canonical map iX .
Proof. Let X be a real operator space and let π : Xc ֒→ (Xc)∗∗ be the canonical embedding. By
Proposition 2.3, (Xc)∗∗ = (X ∗∗)c, completely isometrically via, say, θ. Then θ ◦ π is a complete
isometry such that (θ ◦ π)(z) = Re(π(z)) + iIm(π(z)), for all z ∈ Xc. So the restriction of θ ◦ π to
X is a complete isometry on X such that, for all f ∈ X ∗ and x ∈ X, (θ ◦ π)(x) = (θ ◦ π)(x), and
((θ ◦ π)(x))(f ) = (Re(π(x)))(f ) = f (x) = iX (x). Thus iX is a complete isometry.
(cid:3)
The maximal operator space structure is the largest operator space structure that can be put on
a real operator space, and its matrix norms are defined exactly as in the complex case.
k[xij]k = sup{k[u(xij)]k : u ∈ Ball(B(E, Y )), Y a real operator space}.
If we put the maximal operator space structure on E, then it has the universal property that for
any real operator space Y , and u : E −→ Y bounded linear, we have
ku : E −→ Y k = ku : Max(E) −→ Y kcb
i.e., B(E, Y ) = CB(Max(E), Y )
Lemma 2.5. Let K be a compact Hausdorff space then C(K, R)∗∗ is a (real) commutative C ∗-
algebra of the form C(Ω, R).
Proof. Let u : C(K, R) −→ C(K, R)c = C(K) be the inclusion map. Then u∗∗ : C(K, R)∗∗ −→
C(K)∗∗ is a ∗-monomorphism. The second dual of a (real or complex) commutative C ∗-algebra
is a commutative C ∗-algebra. Let C(K)∗∗ ∼= C(Ω), ∗-isomorphically. Then C(Ω, R) sits inside
C(Ω) as a real space, in fact, as the real part such that, C(Ω, R)c = C(Ω). It is enough to show
that u∗∗(f ) = (u∗∗(f )) for all f ∈ C(K, R)∗∗. We use a weak∗-density argument. First, note that
/
/
/
6
SONIA SHARMA
u∗∗C(K,R) = u and u is selfadjoint, i.e., u(g) = u(g) ∀ g ∈ C(K, R). Let f ∈ C(K, R)∗∗, then there
exists a net {fλ} in C(K, R) converging weak∗ to f . Then u∗∗(fλ) weak∗
−→ u∗∗(f ). This implies that
u∗∗(fλ)(ω) converges pointwise to u∗∗(f )(ω) in C for all ω ∈ Ω. Hence u∗∗(fλ)(ω) −→ u∗∗(f )(ω) in
C for all ω ∈ Ω. So u∗∗(fλ) weak∗
−→ u∗∗(f ). But u∗∗(fλ) = u(fλ) = u∗∗(fλ) = u(fλ). So u∗∗(fλ) weak∗
−→
u∗∗(f ). Hence, by uniqueness of limit, u∗∗(f ) = u∗∗(f ). This shows that the map u∗∗ is real, and
hence it maps into C(Ω, R). Let f ∈ C(Ω, R) = (C(K)∗∗)sa. Let {fλ} ∈ C(K) be a net which
converges weak∗ to f . Then {fλ} also converges weak∗ to f , and so does gλ = fλ+fλ
∈ C(K, R).
Thus f ∈ Ran(u)
Proposition 2.6. Let E be a real Banach space, then
⊂ Ran(u∗∗). Hence u∗∗ maps onto C(Ω, R).
weak∗
(cid:3)
2
Min(E∗) = Max(E)∗ and Min(E)∗ = Max(E∗), completely isometrically.
Proof. We have that Mn(Max(E)∗) ∼= CB(Max(E), Mn(R)) ∼= B(E, Mn(R)), isometrically, for each
n. On the other hand,
isometrically, where ⊗ denotes the Banach space injective tensor product. Thus Min(E∗) =
Max(E)∗.
Mn(Min(E∗)) ∼= Mn(R) ⊗E∗ ∼= B(E, Mn(R)),
Let K be a compact Hausdorff space, then by Lemma 2.5,
Min(C(K, R))∗∗ = C(K, R)∗∗ = C(Ω, R).
On the other hand, Min(C(K, R)∗∗) = Min(C(Ω, R)) = C(Ω, R). Hence Min(C(K, R))∗∗ =
Min(C(K, R)∗∗).
Let E be a real Banach space, and suppose that Min(E) ֒→ C(K, R) completely isometrically.
By taking the duals, we get the following commuting diagram
C(K, R)∗∗
Min(C(K, R)∗∗)
Min(E∗∗)
Min(E)∗∗.
Let u denote the map from Min(E)∗∗ to Min(E∗∗). Since all the maps except u, in the above
diagram are complete isometries and since the diagram commutes, it forces u to be a complete
isometry. Hence Min(E)∗∗ = Min(E∗∗). Applying the first identity, we proved above, to E∗, we
get Min(E∗∗) = Max(E∗)∗. Hence, Max(E)∗∗ = Min(E)∗∗. Let X = Max(E∗) and Y = Min(E)∗,
then since X ∗ = Y ∗, this implies X ∗∗ = Y ∗∗ completely isometrically. By the commuting diagram
below
X Id
Y
X ∗∗
/ Y ∗∗
it is clear that X = Y , completely isometrically.
(cid:3)
We write ℓ1
2(R) is isometrically isomorphic to ℓ∞
2(R) for the two-dimensional real Banach space R ⊕1 R, and ℓ∞
2(R) and (ℓ1
2 (R) for R ⊕∞ R.
2 (R) via (x, y) 7→ (x + y, x − y). We also have that
2 (R), isometrically. From [10] we know that there is a unique
2(R))∗ ∼= ℓ∞
Then ℓ1
2 (R))∗ ∼= ℓ1
(ℓ∞
o
o
O
O
o
o
O
O
/
/
_
_
/
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
7
operator space structure on the two-dimensional complex Banach space, ℓ1
this is not true in the case of real operator spaces.
2(C). We see next that
Proposition 2.7. The operator space structure on l1
2(R) is not unique.
Proof. We consider the maximal and the minimal operator space structures on l2
facts stated above and Proposition 2.6, we have that
1(R). Using the
2 (R)∗,
completely isometrically. So the maximal operator space matrix norm on l1
2 (R)∗) ∼= Min(l∞
2(R)) ∼= Max(l∞
2 (R))∗ = l∞
Max(l1
2(R) is given by
k[(aij, bij)]kmax = sup{k[aijdkl + bijekl]k :
On the other hand, Min(l1
2 (R)) = l∞
(x, y) 7→ (x + y, x − y). So the matrix norm on Min(l1
2(R)) ∼= Min(l∞
[dkl], [ekl] ∈ Ball(Mm(R)), m ∈ N}.
2 (R), completely isometrically via the map
2(R)) is
k[(aij, bij)]kmin = max{k[(aij + bij)]k ,k[(aij − bij)]k}.
0
1 0 (cid:21). Then
1
0
= √2.
k(A, B)kM2(Min(l1
Let [dkl] = A and [ekl] = B, then
2)) = max{kA + Bk ,kA − Bk}
It is clear that k[(aij, bij)]kmin ≤ k[(aij, bij)]kmax . Let A =(cid:20) 1
0 −1 (cid:21) and B =(cid:20) 0 1
= max(cid:26)(cid:13)(cid:13)(cid:13)(cid:13)
1 −1 (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
,(cid:13)(cid:13)(cid:13)(cid:13)
−1 −1 (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) 1
(cid:20) 1 −1
(cid:27)
(cid:20) 0 1
+
1 0 (cid:21)
(cid:20) 1
0 −1 (cid:21)
1 1 (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) 1 1
1 (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) 1 −1
−1
On adding and rearranging the rows and columns we see that this norm is the same as
2)) ≥(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) 1 −1
1 (cid:21)
−1
= max(cid:26)(cid:13)(cid:13)(cid:13)(cid:13)
2)) ≥ 2 > √2 = k(A, B)kM2(Min(l1
k(A, B)kM2(Max(ℓ1
1 0 (cid:21)
(cid:20) 0 1
0
−(cid:20) 1
0 −1 (cid:21)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) 1 1
1 1 (cid:21)
0
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
0
,(cid:13)(cid:13)(cid:13)(cid:13)
0
0
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:27)
= 2.
0
0
So k(A, B)kM2(Max(l1
space structures on l1
2(R).
2)). Hence, there are two different operator
(cid:3)
If X is a complex operator space then, it is also a real operator space, and hence we can talk
about the dual of X both as a real operator space X ∗
r , as well as a complex operator space X ∗,
and ask the question, whether these two spaces are the same real operator spaces. Then by [9],
(X ∗)r is isometrically isomorphic to (Xr)∗. We see next that these spaces need not be completely
isometrically isomorphic.
8
SONIA SHARMA
Consider
Proposition 2.8. Let X = C, be a (complex) operator space with the canonical operator space
structure. Then (Xr)∗ and (X ∗)r are isometrically isomorphic but not necessarily completely iso-
metrically isomorphic.
Proof. It follows from [9, Proposition 1.1.6] that (Cr)∗ ∼= (C∗)r, isometrically.
Note that C∗ is completely isometrically isomorphic to C via the map φz −→ z.
Consider the canonical map θ : C∗ −→ C∗
r given by θ(φ) = Re(φ). By the identification C ∼= C∗,
we can view the above map as θ(z)(y) = Re(y ¯z). If there is any complete isometric isomorphism,
say ψ, then since ψ is an onto isometry between 2-dimensional real Hilbert spaces, it is unitarily
equivalent to θ. Any unitary from C to C, is a rotation by an angle α. So, u is multiplication
by eiα, which is a complete isometry with the canonical operator space matrix norm structure on
C. Then θ = u−1ψu is a complete isometry. Thus ψ is a complete isometry if and only if θ is a
complete isometry. Hence it is enough to show that θ is not a complete isometry.
Consider x =(cid:20) 1 i
0 0 (cid:21) . Then kxk = √2. Since θ2(x) ∈ M2(C∗) ∼= CB(C, M2(R)), we have that
kθ2(x)k = sup{kθ2(x)([zkl])k : [zkl = xkl + iykl] ∈ Mn(C)}.
(cid:20) Re[xkl + iykl] Re[ixkl − ykl]
(cid:20) [xkl]
kθ2(x)[xkl + iykl]k = (cid:13)(cid:13)(cid:13)(cid:13)
= (cid:13)(cid:13)(cid:13)(cid:13)
(xklαl − yklβl)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
Then the norm of the square of ~v produced by the action of (cid:20) [xkl]
(xkl + iykl)(αl + iβl)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Xk=1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Xl=1
≤
≤ k[xkl + iykl]k2 .
(cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:21) is given by
Let ~v be a row vector of length 2n, whose first n entries are αi and the last n entries are βi.
0
[−ykl]
[−ykl]
0
0
[−ykl]
So (cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) [xkl]
instance if [zkl] = IM2(R), then kθ2(x)k ≥ 1. Thus kθ2(x)k = 1 ≤ √2 = kxk.
≤ k[xkl + iykl]k, and hence kθ2(x)k ≤ 1. In fact kθ2(x)k is equal to 1, for
Remark. We end this section with a list of several results from the operator space theory which
can be generalized for the real operator spaces using the exact same proof as in the complex setting.
Various constructions using real operator spaces like taking the quotient, infinite direct sums, c0-
direct sums, mapping spaces CB(X, Y ) and matrix spaces MI,J (X) can be defined analogously,
and are real operator spaces. All the results and properties of matrix spaces hold true for the
real operator spaces (see e.g. [3, 1.2.26]). Further, we can define Hilbert row and Hilbert column
operator space structure on a real Hilbert space by replacing C with R, in the usual definition.
Then B(H, K) ∼= CB(H c, K c) and B(H, K) ∼= CB(K r, H r) completely isometrically, for real
Hilbert spaces H, K. Also (H c)∗ ∼= H r and (H r)∗ ∼= H c. We can show that if u : X −→ Z
is completely bounded between real operator spaces X and Z, and Y is any subspace of Ker(u),
then the canonical map u : X/Y −→ Z induced by u is completely bounded. If Y = Ker(u) then
(cid:3)
n
n
Xk=1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Xl=1
(cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
n
n
0
0
0
0
0
.
2
2
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
9
u is a complete quotient if and only if u is a completely isometric isomorphism. The duality of
subspaces and quotients hold in the real case, i.e., X ∗ ∼= Y ∗/X ⊥ and (Y /X)∗ ∼= X ⊥ completely
isometrically, where Y is a subspace of the real operator space X. It is also true that the trace
class operator S1(H) is the predual of B(H) for every real Hilbert space H. If X is a real operator
space then Mm,n(X)∗∗ ∼= Mm,n(X ∗∗) completely isometrically for all m, n ∈ N. If X and Y are
real operator spaces and if u : X −→ Y ∗ is completely bounded, then its (unique) w∗-extension
u : X ∗∗ −→ Y ∗ is completely bounded with kuk = kuk. Hence CB(X, Y ∗) = w∗CB(X ∗∗, Y ∗)
completely isometrically.
3. Real Operator Algebras
Definition 3.1. An (abstract) real operator algebra A is an algebra which is also an operator
space, such that A is completely isometrically isomorphic to a subalgebra of B(H) for some (real)
Hilbert space H, i.e., there exists a (real) completely isometric homomorphism π : A −→ B(H).
For any n, Mn(A) ⊂ Mn(B(H)) = B(H n) is a real operator algebra with product of two elements,
[aij] and [bij] of Mn(A), given by
n
[aij][bij] = [
aikbkj].
Xk=1
Every real operator algebra can be embedded (uniquely up to a complete isometry) into a complex
operator algebra via Ruan's 'reasonable' complexification. Let A be a real operator algebra and
Ac = A + iA be the operator space complexification of A. Then Ac is an algebra with a natural
product
(x + iy)(v + iw) = (xv − yw) + i(xw + yv).
Suppose that π : A −→ B(H) is a complete isometric homomorphism, for some real Hilbert
space H. Then πc : Ac −→ B(H)c is a (complex) complete isometry, and it is easy to see that πc is
also a homomorphism. Thus Ac is a complex operator algebra if A is a real operator algebra. As in
[12], B(Hc) = B(H)+iB(H) has a reasonable norm extension {k.kn}n∈N, these norms are inherited
by Ac via the complete isometric homomorphism πc : Ac −→ B(H)c. Thus the matrix norms on
Ac satisfy kx + i0kn = kxkn and kx + iyk = kx − iyk , for all x + iy ∈ Mn(Xc) = Mn(X) + iMn(X)
and n ∈ N. The conjugation "-" on Ac satisfies xy = ¯x¯y, for all x, y ∈ Ac.
Remarks. 1) The complexification of a real operator algebra is unique, up to complete isometry
by [12, Theorem 3.1].
2) If A is approximately unital then so is Ac. Indeed if et is an approximate unit for A, then
for any x + iy ∈ Ac, ket(x + iy) − (x + iy)k ≤ ketx − xk + kety − yk. Thus et is an approximate
identity for Ac.
Now we show that there is a real version of the BRS theorem which characterizes the real operator
algebras.
Theorem 3.2. (BRS Real Version) Let A be a real operator space which is also an approximately
unital Banach space. Then the following are equivalent:
(i) The multiplication map m : A ⊗h A −→ A is completely contractive.
(ii) For any n, Mn(A) is a Banach algebra. That is,
" n
Xk=1
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
aikbkj#(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Mn(A)
≤ k[aij]kMn(A) k[bij]kMn(A) ,
10
SONIA SHARMA
for any [aij] and [bij] in Mn(A).
(iii) A is a real operator algebra, that is, there exist a real Hilbert space H and a completely
isometric homomorphism π : A −→ B(H).
Proof. The equivalence between (i) and (ii), and that (iii) implies these, follows from the property
that the Haagerup tensor product of real operator spaces linearizes completely bounded bilinear
maps, and the fact that each Mn(A) is an operator algebra.
(iii) ⇒ (ii) Let π : A −→ B(H). Then by [12, Theorem 2.1], πc : Ac −→ B(H)c is a complete
isometric homomorphism. Let [aij], [bij] ∈ Mn(A) ⊂ Mn(Ac). Then by the BRS theorem for
complex operator algebras,
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
" n
Xk=1
aikbkj#(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Mn(A)
(ii) ⇒ (iii) Since A is approximately unital, by the above remark, Ac is also approximately unital.
Let θ : Ac −→ M2(A) be θ(x + iy) = (cid:20) x
−y x (cid:21). Then θ is a complete isometric homomorphism
and each amplification, θn is an isometric homomorphism. Let a = [aij], b = [bij] ∈ Mn(Ac). Then
≤ k[aij]kMn(A) k[bij]kMn(A) .
y
kabk = kθn(ab)k = kθn(a)θn(b)k ≤ kθn(a)kkθn(b)k = kakkbk .
Thus by the BRS theorem for complex operator algebras, there exists a completely isometric
homomorphism π : Ac −→ B(K), for some complex Hilbert space K. Let K = Hc. Define
π1 = π+π
. Then π1, π2 are (complex) linear maps such that π1 = π1, π2 = π2, and
2
π(x + iy) = (π1(x) − π2(y)) + i(π1(y) + π2(x)). Let π be the composition of π with the canonical
identification B(K) ֒→ M2(B(H)) (see e.g. (2.1)), so
and π2 = π−π
2i
π(x + iy) =(cid:20) π1(x) − π2(y) −π1(y) − π2(x)
π1(x) − π2(y) (cid:21) ∈ M2(B(H)).
π1(y) + π2(x)
The restriction of π to A, say π◦, is a complete isometric inclusion from A into M2(B(H)). Also,
for x, v ∈ A
π◦(x)π◦(v) = (cid:20) π1(x) −π2(x)
π2(x)
π2(v)
π1(x) (cid:21)(cid:20) π1(v) −π2(v)
π1(v) (cid:21)
= (cid:20) π1(x)π1(v) − π2(x)π2(v) −π1(x)π2(v) − π2(x)π1(v)
π1(x)π1(v) − π2(x)π2(v) (cid:21)
= (cid:20) π1(xv) −π2(xv)
π1(xv) (cid:21) = π◦(xv)
π1(x)π2(v) + π2(x)π1(v)
π2(xv)
Thus π◦ is a completely isometric homomorphism from A into M2(B(H)) ∼= B(H 2).
Theorem 3.3. Let A be a complex operator algebra. Then A is a complexification of a real operator
algebra B, i.e., A = Bc completely isometrically if and only if there exists a complex conjugation
"−" on A such that
(i) "−" is a complete isometry, i.e., k[xij]kn = k[xij]kn for all [xij] ∈ Mn(A) and n ∈ N,
(ii) xy = ¯x¯y for all x, y ∈ A.
(cid:3)
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
11
Proof. If A = Bc, for a real operator algebra B, then clearly A satisfies the conditions in (i) and
(ii) above. Suppose that A is a complex operator algebra such that (i) and (ii) hold. Since A is a
complex operator space such that the matrix norms satisfy (i), by [12, Theorem 3.2] there exists a
real operator space B such that A = B + iB completely isometrically. Now the conjugation on A
is x + iy = x − iy, and B = Re(A) = {x ∈ A : x = ¯x}. So if x, y ∈ B, then xy = ¯x¯y = xy. Thus B
is a subalgebra. Since A is a complex operator algebra, it is also a real operator algebra, and B is
a (real) closed subalgebra of A. Thus B is a (real) operator algebra.
(cid:3)
Let A ⊂ B(H) be a real operator algebra, for some real Hilbert space H. Define the unitization
of A as A1 = SpanR{A, IH} ⊂ B(H). Then A ⊂ A1 ⊂ B(H) is a closed subalgebra, and A1 is a
unital real operator algebra.
Lemma 3.4. Let A ⊂ B(H) be a real operator algebra. Then (Ac)1 = (A1)c ⊂ B(H)c, completely
isometrically.
Proof. Clearly both (Ac)1 and (A1)c are subsets of B(H)c. Since A ⊂ A1, Ac ⊂ (A1)c and IH ∈
(A1)c. So (Ac)1 = Span{Ac, IH} ⊂ (A1)c. If x ∈ (A1)c, then
′
′
x = (αa + α
IH) + i(βb + β
= (αa + iβb) + (α
′
′
+ iβ
Thus (Ac)1 = (A1)c.
IH)
)IH ∈ Span{Ac + IH} ⊂ (Ac)1.
(cid:3)
The following result shows that the unitization of real operator algebras is independent of the
choice of the Hilbert space H.
Theorem 3.5. (Real Version of Meyer's Theorem) Let A ⊆ B(H) be a real operator algebra, and
suppose that IH /∈ A. Let π : A −→ B(K) be a completely contractive homomorphism, where K
is a real Hilbert space. We extend π to π◦ : A1 −→ B(K) by π◦(a + λIH) = π(a) + λIK, a ∈ A,
λ ∈ C. Then π◦ is a completely contractive homomorphism.
Proof. Consider πc : Ac −→ B(K)c ∼= B(Kc), which is a completely contractive homomorphism.
Now extend πc to (πc)◦ : (Ac)1 −→ B(Kc) by (πc)◦(a + λIHc ) = πc(a) + λIKc, a ∈ Ac, λ ∈ C. Then
by the Meyer's Theorem for complex operator algebras ([3, Corollary 2.1.15]), (πc)◦ is a completely
contractive homomorphism. Let a + λIH ∈ A1, then (πc)◦(a + λIH ) = πc(a) + λIK = π(a) + λIK =
π◦(a + λIH). Thus (πc)◦A1 = π◦ and hence π◦ is a completely contractive homomorphism.
(cid:3)
4. Real Injective Envelope
In this section we study in more detail the real injective envelope of real operator spaces, which
is mentioned by Ruan in [11].
Definition 4.1. Let X be a real operator space and let Y be a real operator space, such that
there is a complete isometry i : X −→ Y . Then the pair (Y, i) is called an extension of X. An
injective extension (Y, i) is a real injective envelope of X if there is no real injective space Z such
that i(X) ⊂ Z ⊂ Y . We denote a real injective envelope by (I(X), i) or simply by I(X).
By the Arveson-Wittstock-Hahn-Banach theorem for real operator spaces, [11, Theorem 3.1],
B(H) is an injective real operator space for any real Hilbert space H. Thus a real operator space
X ⊂ B(H) is injective if and only if it is the range of a completely contractive idempotent map
from B(H) onto X.
12
SONIA SHARMA
Definition 4.2. If (Y, i) is an extension of X, then Y is a rigid extension if IY is the only
completely contractive map which restricts to an identity map on X. We say that (Y, i) is an
essential extension of X, if whenever u : X −→ Z is a completely contractive map, for some real
operator space Z, such that u ◦ i is a complete isometry, then u is a complete isometry.
Theorem 4.3. If a real operator X is contained in a real injective operator space W , then there
is an injective envelope Y of X such that X ⊂ Y ⊂ W .
To prove this theorem we need to define some more terminology, and we also need the following
two lemmas, which are the real analogies of [3, Lemma 4.2.2] and [3, Lemma 4.2.4], respectively. The
proof of Lemma 4.5 uses the fact that, if X is a real operator space and H is any real Hilbert space,
then a bounded net (ut) in CB(X, B(H)) converges in weak∗-topology to a u ∈ CB(X, B(H)) if
and only if
hut(x)ζ, ηi → hu(x)ζ, ηi for all x ∈ X, ζ, η ∈ H.
Definition 4.4. Let X is a subspace of a real operator space W . An X-projection on W is a
completely contractive (real) idempotent map φ : W → W which restricts to the identity map on
X. An X-seminorm on W is a seminorm of the form p(·) = ku(·)k, for a completely contractive
(real) linear map u : W → W which restricts to the identity map on X. Define a partial order ≤
on the sets of all X-projections, by setting φ ≤ ψ if φ ◦ ψ = ψ ◦ φ = φ. This is also equivalent to
Ran(φ) ⊂ Ran(ψ) and Ker(ψ) ⊂ Ker(φ).
Lemma 4.5. Let X be a subspace of a real injective operator space W .
(i) Any decreasing net of X-seminorms on W has a lower bound. Hence there exists a minimal
X-seminorm on W , by Zorn's lemma. Each X-seminorm majorizes a minimal X-seminorm.
(ii) If p is a minimal X-seminorm on W , and if p(·) = ku(·)k, for a completely contractive linear
map on W which restricts to the identity map on X, then u is a minimal X-projection.
Lemma 4.6. Let (Y, i) be an extension of real operator space X such that Y is injective. Then the
following are equivalent:
(i) Y is an injective envelope of X,
(ii) Y is a rigid extension of X,
(iii) Y is an essential extension of X.
Using the rigidity property of injective envelopes and a standard diagram chase, we can show
that if (Y1, i1) and (Y2, i2) are two injective envelopes of a real operator space X then Y1 and Y2 are
completely isometrically isomorphic via some map u such that u ◦ i1 = i2. Hence the real injective
envelope, if exists, is unique. The argument in [3, Theorem 4.2.6], and Lemma 4.5 and Lemma 4.6,
prove Theorem 4.3. Thus the real injective envelope exists.
Lemma 4.7. Let X be a real operator space. Then X is real injective iff Xc is (complex) injective.
Proof. First suppose that X is real injective, then there exists a completely contractive idempotent
P , from B(H) onto Z, for some real Hilbert space H. The complexification of P , Pc : B(H)c −→ Xc
is clearly a (complex) completely contractive idempotent onto Xc. Since B(H)c ∼= B(Hc), com-
pletely isometrically, Zc is a (complex) injective operator space. Conversely, let Xc be a (complex)
injective space and Q : B(K) −→ Xc be a completely contractive (complex) linear surjective idem-
potent. Let K = Hc where H is a real Hilbert space, so Q : B(Hc) ∼= B(H)c −→ Xc. Consider
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
13
Re(Q) = Q+ ¯Q
2
, where ¯Q(T + iS) = Q(T − iS). For any T + iS ∈ B(H)c,
Q
(T + iS) = Q(Q(T − iS)) = Q2(T − iS) = Q(T − iS) = Q(T + iS).
2
Let x+iy ∈ Xc and suppose that Q(T +iS) = x−iy for some T, S ∈ B(H). Then Q(T −iS) = x+iy.
Thus Q is an idempotent onto Xc. So QQ(T +iS) = Q(T +iS) and QQ(T +iS) = Q(T +iS), for all
T + iS ∈ B(H)c. Thus for T ∈ B(H), (Re(Q))2(T ) = Q2(T )+QQ(T )+QQ(T )+Q
=
If x ∈ X ⊂ Xc, then Q(x) = x and Q(x) = x, so Re(Q)(x) = x. This shows that
Re(Q)(T ).
Re(Q) : B(H) −→ X is a (real) linear completely contractive idempotent onto X. Hence X is real
injective.
= 2Q(T )+2Q(T )
(T )
(cid:3)
4
4
2
The next result is a real analogy of a Choi-Effros theorem (see e.g., [3, Theorem 1.3.13]).
It
is shown in the last paragraph of [11, pg. 492]) that the argument in the complex version of the
theorem can be reproduced to prove part (i) of the following result.
Theorem 4.8 (Choi-Effros). Let A be a unital real C ∗-algebra and let φ : A −→ A be a selfadjoint,
completely positive, unital, idempotent map. Then
(i) R = Ran(φ) is a unital real C ∗-algebra with respect to the original norm, involution, and
vector space structure, but new product r1 ◦φ r2 = φ(r1r2),
(ii) φ(ar) = φ(φ(a)r) and φ(ra) = φ(rφ(a)), for r ∈ R and a ∈ A,
(iii) If B is the C ∗-algebra generated by the set R, and if R is given the product ◦φ, then φ B is a
∗-homomorphism from B onto R.
Proof. Let φ : A −→ A be a selfadjoint, completely positive, unital idempotent map. Then φ
is completely contractive, by [11, Proposition 4.1], and hence φc : Ac −→ Ac is a completely
contractive, unital idempotent onto Ran(φ)c. By the Choi-Effros Lemma for complex operator
systems, [3, Theorem 1.3.13], Ran(φ)c is a C ∗-algebra with a new product given by (r1 + ir2) ◦
(s1 + is2) = φc((r1 + ir2)(s1 + is2)), r1, r2, s1, s2 ∈ R. For r, s ∈ R, r ◦ s = φc(rs) = φ(rs) ∈ R.
By [9, Proposition 5.1.3], R is a real C ∗-algebra with this product. Further, φ(ar) = φc(ar) =
φc(φc(a)r) = φ(φ(a)r), and similarly φ(ra) = φ(rφ(a)), for all a ∈ A, r ∈ R.
Let C = C ∗(Rc) be the (complex) C ∗-algebra generated by Rc in Ac, then by [3, Theorem 1.3.13
(iii)], (φc)C is a ∗-homomorphism from C onto Rc. Let B = C ∗(R) be the real C ∗-subalgebra
of A generated by R. It is easy to see that C ∗(Rc) = C ∗(R)c. Clearly, since C ∗(R) ⊂ C ∗(Rc),
C ∗(R)c ⊂ C ∗(Rc). Also,
SC{s1s2 . . . sn : n ∈ N} = SR{r1r2 . . . rn : n ∈ N} + iSR{r
1r
2 . . . r
′
′
′
n : n ∈ N},
′
′
′
′
1r
2 . . . r
i ∈ R, and S means "Span". If a ∈ C ∗(Rc) ⊂ Ac then a = x + iy is the
where si ∈ Rc, and ri, r
limit of at ∈ SC{s1s2 . . . sn : n ∈ N}. Then at = xt + iyt, where xt ∈ SR{r1r2 . . . rn : n ∈ N},
n : n ∈ N}. Also, if we suppose that Ac ⊂ B(H)c, for some real Hilbert space
yt ∈ SR{r
H, then it is easy to see that xt −→ x, yt −→ y. Hence, (φc)B = φB is a ∗-homomorphism from
B onto R.
Remark. Let A and B be real C ∗-algebras, and let φ : A −→ B be a unital completely contractive
map. Then φc is a (complex) completely contractive linear map between complex C ∗-algebras Ac
and Bc. So φc is completely positive and hence selfadjoint. Since φ = φcA, φ is also selfadjoint.
Thus a completely contractive unital map between real C ∗-algebras is selfadjoint. As a result,
(cid:3)
14
SONIA SHARMA
we can replace the completely positive and selfadjoint condition in Theorem 4.8 above, with the
condition that φ is completely contractive.
Theorem 4.9. X be a unital real operator space, then there is an injective envelope I(X) which is
a unital real C ∗-algebra.
Proof. Let X ⊂ B(H) for some real Hilbert space H. Since B(H) is injective, we can find an
injective envelope of X such that X ⊂ I(X) ⊂ B(H). As I(X) is injective, so the identity map
on I(X) extends to φ : B(H) −→ B(H) such that φ is a completely contractive idempotent onto
I(X). By Theorem 4.8 and the remark above, Ran(φ) = I(X) becomes a unital real C ∗-algebra
with the new product.
(cid:3)
Proposition 4.10. Let X be a real (or complex) Banach space, then Min(I(X)) = I(Min(X)),
completely isometrically.
Proof. Let X be a real Banach space. Since I(X) is an injective Banach space, and contractive
maps into Min(X) are completely contractive, it clear that Min(I(X)) is a real injective operator
space. Let i : X −→ I(X) be the canonical isometry, and let j : I(X) −→ C(Ω, R) be an isometric
embedding of I(X), for some compact, Hausdorff space Ω. Then j : Min(I(X)) −→ C(Ω, R) and
j◦i : Min(X) −→ C(Ω, R) are complete isometries. Thus (Min(I(X)), i) is a real injective extension
of Min(X). Further suppose that u : Min(I(X)) −→ Min(I(X)) is a complete contraction which
restricts to the identity map on Min(X). Then by the rigidity of I(X), u is an isometry into
Min(I(X)), and hence a complete isometry. Thus (Min(I(X)), i) is a rigid extension of Min(X),
and hence I(Min(X)) = Min((I(X))), completely isometrically.
(cid:3)
Definition 4.11. Let X be a real unital operator space. Then we define a C ∗-extension of X to be
a pair (B, j) consisting of a unital real C ∗-algebra B, and a complete isometry j : X −→ B, such
that j(X) generates B as a C ∗-algebra. A C ∗-extension (B, i) is a C ∗-envelope of X if it has the
the following universal property: Given any C ∗-extension (A, j) of X, there exists a (necessarily
unique and surjective) real ∗-homomorphism π : A −→ B, such that π ◦ j = i.
Using Theorem 4.8, Theorem 4.9, and the argument in [3, 4.3.3], we can show that the C ∗-
subalgebra of I(X) generated by i(X) is a C ∗-envelope of X, where the pair (I(X), i) is an injective
envelope of X. Thus the C ∗-envelope exists for every unital real operator space X.
A real operator system is a (closed) subspace S of B(H), H a real Hilbert space, such that S
contains IH, and S is selfadjoint, i.e., x∗ ∈ S if and only if x ∈ S. Note that a positive element in
B(H), H a real Hilbert space, need not be selfadjoint. For instance, consider the 2 × 2 matrices
over R, then x = (cid:20) 2 −1
2 (cid:21) is positive, i.e., hxζ, ζi ≥ 0 for all ζ ∈ R2, but x 6= x∗. Thus, we say
that an element x ∈ S(X) ⊂ B(H) is positive, if for all ζ, η ∈ H, hxζ, ηi = hζ, xηi (selfadjoint), and
hxζ, ζi ≥ 0. If x ∈ B(H, K), H and K real Hilbert spaces, then
(cid:20) 1
x∗ 1 (cid:21) ≥ 0 ⇐⇒ kxk ≤ 1.
(4.1)
In [11], Ruan considers real operator systems and shows that a unital selfadjoint map between two
real operator systems is completely contractive if and only if it is completely positive. It is also
shown that the Stinespring theorem, the Arveson's Extension Theorem, and the Kadison-Schwarz
inequality hold true, with an added hypothesis that the maps be selfadjoint. We can show using
1
x
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
15
the Stinespring theorem that Proposition 1.3.11 and Proposition 1.3.12 from [3], are also true in
the real setting.
If X ⊂ B(H) is a real operator space, then we can define the real Paulsen system as
y∗ µ (cid:21) : x, y ∈ X and λ, µ ∈ R(cid:27) ⊂ M2(B(H)).
X ⋆ RIH (cid:21) =(cid:26)(cid:20) λ x
S(X) =(cid:20) RIH X
The next lemma is the real version of Paulsen lemma, and it can be proved using the argument in
[3, Lemma 1.3.15], Equation (4.1), and that the map φ, defined below, is selfadjoint. This lemma
shows that as a real operator system (i.e., up to complete order isomorphism) S(X) only depends
on the operator space structure of X, and not on its representation on H.
Lemma 4.12. For i = 1, 2, let Hi and Ki be real Hilbert spaces, and Xi ⊂ B(Ki, Hi). Suppose
that u : X1 → X2 is a real linear map. Let Si be the real Paulsen systems associated with Xi inside
B(Hi ⊕ Ki). If u is contractive (resp. completely contractive, completely isometric), then
φ :(cid:20) λ x
y∗ µ (cid:21) →(cid:20)
λ
u(y)∗
u(x)
µ (cid:21)
is positive (resp. completely positive and completely contractive, a complete order injection) as a
map from S1 to S2.
Let X ⊂ B(H) be a real operator space and let S(X) ⊂ M2(B(H)) be the associated real
Paulsen system. Then I(S(X)) ⊂ M2(B(H)) is a unital C ∗-algebra, by Theorem 4.9, and there is a
completely positive idempotent map φ from M2(B(H)) onto I(S(X)). Let p and q be the canonical
projections IH ⊕ 0 and 0 ⊕ IH, then φ(p) = p and φ(q) = q. So,
I(S(X)) =(cid:20) pI(S(X))p pI(S(X))q
qI(S(X))p qI(S(X))q (cid:21) .
Using Lemma 4.6 and Lemma 4.12, and the argument in [3, Theorem 4.4.3], we can show that the
1-2-corner, pI(S(X))q, of I(S(X)) is an injective envelope of X. As a corollary, we get the following
which is the real analogue of the Hamana-Ruan characterization of injective operator spaces.
Theorem 4.13. A real operator space X is injective if and only if X ∼= pA(1 − p) completely
isometrically, for a projection p in an injective real C ∗-algebra A.
A real TRO is a closed linear subspace Z of B(K, H), for some real Hilbert spaces K and
H, satisfying ZZ ⋆Z ⊂ Z. For x, y, z ∈ Z, xy∗z is called the triple or ternary product on Z,
sometimes written as [x, y, z]. A subtriple of a TRO Z is a closed subspace Y of Z satisfying
Y Y ⋆Y ⊂ Y . A triple morphism between TROs is a linear map which respects the triple product:
thus T ([x, y, z]) = [T x, T y, T z]. In the construction of the real injective envelope, discussed above,
let Z = pI(S(X))q, then ZZ ⋆Z ⊂ Z with the product of the C ∗-algebra I(S(X)). In terms of the
product in B(H), [x, y, z] = P (xy∗z) for x, y, z ∈ Z. So if X is a TRO, then the triple product on
X coincides with the triple product on X coming from I(X). Thus pI(S(X))q = I(X) is a TRO.
If two TROs X and Y are completely isometrically isomorphic, via say u, then by Lemma 4.12, we
can extend u to a complete order isomorphism between the Paulsen systems. Further, this map
extends to a completely isometric unital surjection u between the the injective envelopes I(S(X))
and I(S(Y )), which are (real) unital C ∗-algebras. By Lemma 2.2, u is a ∗-isomorphism, and hence
a ternary isomorphism between when restricted to X. Thus u is a triple isomorphism. Thus a real
operator space can have at most one triple product (up to complete isometry).
16
SONIA SHARMA
Define T (X) to be the smallest subtriple of I(X) containing X. Then it is easy to see that
T (X) = Span{x1x∗
2x3x∗
4 . . . x2n+1 : x1, x2, . . . x2n+1 ∈ X}.
Let B = T (X)⋆T (X), T (X) regarded as a subtriple of I(X) in I(S(X)). Then B is a C ∗-subalgebra
of 2-2-corner of I(S(X)), and hence of I(S(X)). Define hy, zi = y∗z for y, z ∈ T (X), a B-valued
inner product. This inner product is called the Shilov inner product on X.
Let X be a real operator space. If P is a projection, i.e., P = P 2 and P ∗ = P (equivalently
5. One-Sided Real M -Ideals
kPk ≤ 1), then define linear mappings
νc
P : X −→ C2(X) : x 7→(cid:20)
P (x)
x − P (x) (cid:21) ,
µc
P : C2(X) −→ X :(cid:20) x
y (cid:21) 7→ P (x) + (Id − P )(y).
Then µc
P ◦ νc
P = I.
Definition 5.1. A complete left M -projection on X is a linear idempotent on X such that the map
νc
P : X −→ C2(X) : x 7→(cid:20)
Proposition 5.2. If X is a real operator space and P : X −→ X is a projection, then P is a
complete left M -projection if and only if µc
x − P (x) (cid:21) is a complete isometry.
P are both completely contractive.
P and νc
P (x)
Proof. If νc
P is completely isometric, then
kP (x) + y − P (y)k =(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20)
P (x)
y − P (x) (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
P (x)
P (y)
x − P (x)
y − P (y)
≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
,
=(cid:13)(cid:13)(cid:13)(cid:13)
y (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) x
and thus µc
the fact that µc
P ◦ νc
P = I.
P is contractive. These calculations work as well for matrices. The converse follows from
(cid:3)
Proposition 5.3. The complete left M -projections in a real operator space X are just the mappings
P (x) = ex for a completely isometric embedding X ֒→ B(H) and an orthogonal projection e ∈
B(H).
Proof. If P : X −→ X is a complete left M -projection, then fix an embedding X ⊂ B(H) for some
real Hilbert space H. By the definition, the mapping
σ : X ֒→ B(H ⊕ H) : x 7→(cid:20)
is completely isometric. We have that
P (x)
(I − P )(x) 0 (cid:21)
0
σ(P (x)) =(cid:20) P (x) 0
0 (cid:21) =(cid:20) 1 0
0 0 (cid:21) σ(x),
0
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
17
and thus e =(cid:20) 1 0
converse follows from the following:
0 0 (cid:21) ∈ B(H ⊕ H) is the desired left projection relative to the embedding σ. The
x − P (x) (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
=(cid:13)(cid:13)(cid:13)(cid:13)(cid:2) x∗e x∗ − x∗e (cid:3)(cid:20)
= kx∗ex + x∗(1 − e)xk = kx∗xk = kxk2 .
x − ex (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
= (cid:13)(cid:13)(cid:13)(cid:13)
ex
x − ex (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
P (x)
(cid:20)
ex
2
2
(cid:20)
(cid:13)(cid:13)(cid:13)(cid:13)
(cid:3)
Let X be a real operator space. We say a map u : X −→ X is a left multiplier of X if there
exists a linear complete isometry σ : X −→ B(H) for some real Hilbert space H, and an operator
S ∈ B(H) such that
σ(u(x)) = Sσ(x),
for all x ∈ X. We denote the set of all left multipliers of X by Mℓ(X). Define the multiplier norm
of u, to be the infimum of kSk over all such possible H, S, σ. We define a left adjointable map of X
to be a linear map u : X −→ X such that there exists a linear complete isometry σ : X −→ B(H)
for some real Hilbert space H, and an operator A ∈ B(H) such that
σ(u(x)) = Aσ(x) for all x ∈ X, and A∗σ(X) ⊂ σ(X).
The collection of all left adjointable maps of X is denoted by Aℓ(X). Every left adjointable map
of X is a left multiplier of X, that is, Aℓ(X) ⊂ Mℓ(X).
Theorem 5.4. Let X be a real operator space and let u : X −→ X be a linear map. Then the
following are equivalent:
(i) u is a left multiplier of X with norm ≤ 1.
(ii) The map τu : C2(X) −→ C2(X) :(cid:20) x
(iii) There exists a unique 'a' in the 1-1-corner of I(S(X)) such that kak ≤ 1 and u(x) = ax for
(cid:21), is completely contractive.
y (cid:21) 7→(cid:20) u(x)
y
all x ∈ X.
′
By a direct application of the argument in [3, Theorem 4.5.2], we get that (i) ⇒ (ii) and (iii)
⇒ (i). For the implication (ii) ⇒ (iii), using the machinery we developed for real operator spaces
in the last section, we can replicate the elegant proof due to Paulsen mentioned in [3, Theorem
4.5.2]. Note that the map Φ
in [3, Theorem 4.5.2], is selfadjoint, therefore by the real version of
the Arveson's extension theorem from [11], Φ
extends to a completely positive and selfadjoint map
Φ, on the C ∗-algebra M . By the real version of the Stinespring's Theorem [11, Theorem 4.3], the
argument in [3, Proposition 1.3.11] can be reproduced, and hence [3, Proposition 1.3.11] holds for
real C ∗-algebras. Since Φ fixes the C ∗-subalgebra
0
I11
0
′
C
0
0
B =
I(X)⋆
I(X)
I22
of M , so Φ is a ∗-homomorphism on B. By [3, Proposition 1.3.11], Φ : M −→ M is a bimodule
map over B. The rest of the argument follows verbatim.
Theorem 5.5. Let X be a real operator space then Mℓ(X) is a real operator algebra. Further,
Aℓ(X) is a real C ∗-algebra.
18
SONIA SHARMA
Proof. We use the completely isometric embeddings X ⊂ I(X) ⊂ S(X), and the notation from
Section 4. Let
IMl(X) = {a ∈ pI(S(X))p : aX ⊂ X}.
Then IMl(X) is a subalgebra of the real C ∗-algebra pI(S(X))p, and hence is a real operator
algebra. Define θ : IMl(X) −→ Mℓ(X) as θ(a)(x) = ax for any x ∈ X. Then θ is an isometric
isomorphism. Using the canonical identification Mn(Mℓ(X)) ∼= Mℓ(Cn(X)), define a matrix norm
on Mn(Mℓ(X)) for each n. With these matrix norms, and a matricial generalization of the argument
[3, 4.5.4]), θ is a complete isometric isomorphism. Hence all the
after Theorem 5.4 (see e.g.
'multiplier matrix norms' are norms, and Mℓ(X) ∼= IMl(X) is a real operator algebra. Since
Aℓ(X) = Mℓ(X) ∩ Mℓ(X)⋆, we have that
Aℓ(X) ∼= {a ∈ pI(S(X))p : aX ⊂ Xand a∗X ⊂ X}.
Hence Aℓ(X) is a real C ∗-algebra.
Theorem 5.6. If P is a projection on a real operator space X, then the following are equivalent:
(cid:3)
P is completely contractive.
(i) P is a complete left M -projection.
(ii) τ c
(iii) P is an orthogonal projection in the real C ∗-algebra Aℓ(X).
(iv) P ∈ Mℓ(X) with the multiplier norm ≤ 1.
(v) The maps νc
P are completely contractive.
P and µc
The above theorem can be easily seen from Proposition 5.2, Proposition 5.3, and Theorem 5.4.
Definition 5.7. A subspace J of a real operator space X is a right M -ideal if J ⊥⊥ is the range of
a complete left M -projection on X ∗∗.
Proposition 5.8. A projection P : X −→ X is a complete left M -projection if and only if Pc is a
(complex) complete left M -projection on Xc.
Proof. We first note that C2(Xc) ∼= C2(X)c, completely isometrically, via the shuffling map
Also,
x2
(τP )c(cid:18)(cid:20) x
(cid:20) x1 −x2
x1 (cid:21)
(cid:20) y1 −y2
y1 (cid:21)
v (cid:21) + i(cid:20) y
y2
.
7→
(cid:20) x1
y1 (cid:21) −(cid:20) x2
y2 (cid:21)
(cid:20) x2
y2 (cid:21)
(cid:20) x1
y1 (cid:21)
w (cid:21)
(cid:21) + i(cid:20) P (y)
w (cid:21)(cid:19) = (cid:20) P (x)
(cid:21)
v + iw (cid:21)(cid:19) .
= (cid:20) P (x) + iP (y)
= τ(Pc)(cid:18)(cid:20) x + iy
v + iw
v
If P is a complete left M -projection, then by Theorem 5.6, τP and hence, (τP )c is completely con-
tractive. By the above τ(Pc) is completely contractive and so, Pc is a complete left M -projection.
Conversely,
if Pc is a complete left M -projection, then τ(Pc) is completely contractive. Since
τ(Pc)C2(X) = τP , τP is a complete contraction and hence P is a complete left M -projection, by
Theorem 5.6.
(cid:3)
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
19
Corollary 5.9. A subspace J in a real operator space X is a right M -ideal if and only if Jc is a
(complex) right M -ideal in Xc.
yt
Proof. Since(cid:20) xt −yt
xt (cid:21) converge weak∗ in (Xc)∗∗ if and only if both (xt) and (yt) converge weak∗
)c = (J ⊥⊥)c. If J is a real right M -ideal and if
in X ∗∗, if J ⊂ X, then (Jc)⊥⊥ = Jc
P : X ∗∗ −→ J ⊥⊥ is a (real) left M -projection, then by the above corollary Pc : (X ∗∗)c −→ (J ⊥⊥)c
is a (complex) left M -projection. Let Q be the induced map from (Xc)∗∗ onto (Jc)⊥⊥. So the
diagram
= (J
w∗
w∗
(X ∗∗)c
Pc
(J ⊥⊥)c
(Xc)∗∗
/ (Jc)⊥⊥
Q
commutes and thus Q is an idempotent. Also, since the diagram
C2((X ∗∗)c)
τ(Pc )
C2((X ∗∗)c)
c.i.
c.i.
C2((Xc)∗∗)
/ C2((Xc)∗∗)
τQ
commutes, and τ(Pc) is a complete contraction, so τQ is a complete contraction. Hence Jc is a
right M -ideal in Xc. Conversely, if P is a complete left M -projection from (Xc)∗∗ = (X ∗∗)c onto
(Jc)⊥⊥ = (J ⊥⊥)c, then let Q = Re(P ). Then a similar argument as in Lemma 4.7 shows that
Q is an idempotent from X ∗∗ onto J ⊥⊥. Also since τQ is the restriction of τP to C2(X ∗∗), τQ is
completely contractive. Thus J is a real right M -ideal.
(cid:3)
Corollary 5.10. The right M -ideals in a real C ∗-algebra A are precisely the closed right ideals in
A.
Corollary 5.11. The right M -ideals in an approximately unital real operator algebra are precisely
the closed right ideals with a left contractive approximate identity.
Note that by Corollary 5.9, and Proposition 2.3, it is clear that X is right M -ideal in X ∗∗ if and
only if Xc is right M -ideal in (Xc)∗∗.
We say that a real operator space X is right M -embedded if X is a right M -ideal in X ∗∗.
Thus by the above lines it is clear that a real operator space X is right M -embedded if and only if
Xc is right M - embedded.
Lemma 5.12. Let X be a real operator space and Y ⊂ X. Then (X/Y )c ∼= Xc/Yc, completely
isometrically.
Proof. Let φ : X −→ X/Y be the canonical complete quotient map. We claim that φ : Xc −→
(X/Y )c given by x1 + ix2 7→ (x1 + Y ) + i(x2 + Y ) is a complete quotient. Since φ is a com-
plete contraction, so is φc, by [12, Theorem 2.1]. Thus if we denote the open unit ball of an
operator space Z by UZ , then φc(UXc ) ⊂ U(X/Y )c . Let (x1 + Y ) + i(x2 + Y ) ∈ U(X/Y )c, then
/
/
O
O
O
O
/
/
/
O
O
O
O
/
20
SONIA SHARMA
x2 + Y
(cid:3)
x2 + Y
−x2 x1 (cid:21) 7→ (cid:20) x1 + Y
(cid:13)(cid:13)(cid:13)(cid:13)
−x2 + Y x1 + Y (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)
(cid:20) x1 + Y
< 1. Since φ is a complete quotient, φ2 : M2(X) −→ M2(X/Y ) is a
quotient and φ : (cid:20) x1
−x2 + Y x1 + Y (cid:21). Thus kx1 + ix2k < 1. Similar argu-
x2
ment at each matrix level proves the claim. Since Ker(φc) = Yc, (X/Y )c ∼= Xc/Yc completely
isometrically.
Theorem 5.13. Let X be a real right M -embedded operator space and X ⊂ Y , then Y and X/Y
are right M -embedded.
Proof. Since Yc ⊂ Xc, by [15, Theorem 2.6], Yc is right M -embedded and hence Y is right M -
embedded. Again by [15, Theorem 2.6], Xc/Yc is right M -embedded. By the above lemma, (X/Y )c
is right M -embedded, and thus, X/Y is right M -embedded.
(cid:3)
Lemma 5.14. Let Xk be real operator spaces, then (⊕◦
Proof. Let φ : (⊕◦
k(xk)k ,k(yk)k ≤ k(xk + iyk)k ≤ k(xk)k + k(yk)k, φ is well defined and onto. Also,
k(Xk)c completely isometrically.
k(Xk)c be the canonical map (xk) + i(yk) 7→ (xk + iyk). Since
k(xk) + i(yk)k = (cid:13)(cid:13)(cid:13)(cid:13)
(xk) (cid:21)(cid:13)(cid:13)(cid:13)(cid:13)M2(⊕∞
(cid:20) (xk)
−(yk)
−yk xk (cid:21)(cid:19)(cid:13)(cid:13)(cid:13)(cid:13)⊕∞
= (cid:13)(cid:13)(cid:13)(cid:13)
(cid:18)(cid:20) xk
Thus, φ is an isometry. Similar argument at each matrix level shows that φ is a complete isometry.
(cid:3)
kXk)c −→ ⊕◦
= k(xk + iyk)k⊕∞
k (Xk)c .
k M2(Xk)
kXk)c ∼= ⊕◦
(yk)
yk
k Xk)
The following result follows from above the lemma, [15, Proposition 2.12], and the fact that a
real operator space X is right M -embedded if and only if Xc is right M -embedded.
Theorem 5.15. Let A be a real right M -embedded TRO, then A ∼= ⊕◦
isometrically, for some real Hilbert spaces Hi, Hj.
i,jK(Hi, Hj) completely
Acknowledgments. We thank Dr. David Blecher, for proposing this project and continually sup-
porting the work. We are grateful for his insightful comments and many suggestions and corrections.
References
[1] B. Blackadar, K-theory for operator algebras, Mathematical Sciences Research Institute Publications, 5, Cam-
bridge University Press, Cambridge, 1998.
[2] D. P. Blecher, E. G. Effros, and V. Zarikian, One-sided M -ideals and multipliers in operator spaces, I, Pacific J.
Math. 206 (2002), 287-319.
[3] D. P. Blecher and C. Le Merdy, Operator algebras and their modules -- an operator space approach, Oxford Univ.
Press, Oxford (2004).
[4] A. Connes, J. Cuntz, E. Guentner, N. Higson, J. Kaminker and J. E. Roberts, Noncommutative geometry,
Lecture Notes in Mathematics, 1831, Springer-Verlag, Berlin, 2004.
[5] A. Defant and A. Floret, Tensor norms and operator ideals, North-Holland Mathematics Studies, 176. North-
Holland Publishing Co., Amsterdam, 1993.
[6] K. Goodearl, Notes on real and complex C ∗-algebras, Shiva Mathematics Series, 5. Shiva Publishing Ltd.,
Nantwich, 1982.
[7] P. Harmand, D. Werner, and W. Werner, M -ideals in Banach spaces and Banach algebras, Lecture Notes in
Math., 1547, Springer-Verlag, Berlin -- New York, 1993.
REAL OPERATOR ALGEBRAS AND REAL COMPLETELY ISOMETRIC THEORY
21
[8] T. Ho, J. M. Moreno, A. M. Peralta and B. Russo, Derivations on real and complex JB ∗-triples, J. London
Math. Soc. (2) 65 (2002), no. 1, 85-102.
[9] B. R. Li, Real operator algebras, World Scientific Publishing Co., Inc., River Edge, NJ, 2003.
[10] V. I. Paulsen, Representations of function algebras, abstract operator spaces, and Banach space geometry, J.
Funct. Anal. 109 (1992), no. 1, 113 -- 129.
[11] Z. J. Ruan, On real operator spaces, Acta Math. Sin. (Engl. Ser.) 19 (2003), no. 3, 485 -- 496.
[12] Z. J. Ruan, Complexification of real operator spaces, Illinois J. Math. 47 (2003), no. 4, 1047 -- 1062.
[13] H. Schroder, K-theory for real C ∗-algebras and applications, Pitman Research Notes in Mathematics Series, 290,
John Wiley & Sons, Inc., New York, 1993.
[14] S. Sharma, Ph.D. Thesis, University of Houston, 2009.
[15] S. Sharma, Operator spaces which are one-sided M -ideals in their bidual, Studia Math. 196 (2010), no. 2, 121-141.
[16] E. Stormer, Real structure in the hyperfinite factor, Duke Math. J. 47 (1980), no. 1, 145 -- 153.
Department of Mathematics, SUNY Cortland, Cortland, NY 13045
E-mail address, Sonia Sharma: [email protected]
|
1711.01016 | 1 | 1711 | 2017-11-03T03:48:05 | Semiflat Orbifold Projections | [
"math.OA"
] | We compute the semiflat positive cone $K_0^{+SF}(A_\theta^\sigma)$ of the $K_0$-group of the irrational rotation orbifold $A_\theta^\sigma$ under the noncommutative Fourier transform $\sigma$ and show that it is determined by classes of positive trace and the vanishing of two topological invariants. The semiflat orbifold projections are 3-dimensional and come in three basic topological genera: $(2,0,0)$, $(1,1,2)$, $(0,0,2)$. (A projection is called semiflat when it has the form $h + \sigma(h)$ where $h$ is a flip-invariant projection such that $h\sigma(h)=0$.) Among other things, we also show that every number in $(0,1) \cap (2\mathbb Z + 2\mathbb Z\theta)$ is the trace of a semiflat projection in $A_\theta$. The noncommutative Fourier transform is the order 4 automorphism $\sigma: V \to U \to V^{-1}$ (and the flip is $\sigma^2$: $U \to U^{-1},\ V \to V^{-1}$), where $U,V$ are the canonical unitary generators of the rotation algebra $A_\theta$ satisfying $VU = e^{2\pi i\theta} UV$. | math.OA | math | Semiflat Orbifold Projections
S. Walters
In Memory of Mom
0
(Aσ
ABSTRACT. We compute the semiflat positive cone K+SF
θ) of the K0-group of the
irrational rotation orbifold Aσ
θ under the noncommutative Fourier transform σ and
show that it is determined by classes of positive trace and the vanishing of two
topological invariants. The semiflat orbifold projections are 3-dimensional and
come in three basic topological genera: (2, 0, 0), (1, 1, 2), (0, 0, 2).
(A projection is
called semiflat when it has the form h + σ(h) where h is a flip-invariant projection
such that hσ(h) = 0.) Among other things, we also show that every number in
(0, 1) ∩ (2Z + 2Zθ) is the trace of a semiflat projection in Aθ. The noncommutative
Fourier transform is the order 4 automorphism σ : V → U → V −1 (and the flip is σ2:
U → U −1, V → V −1), where U,V are the canonical unitary generators of the rotation
algebra Aθ satisfying VU = e2πiθUV .
7
1
0
2
v
o
N
3
]
.
A
O
h
t
a
m
[
1
v
6
1
0
1
0
.
1
1
7
1
:
v
i
X
r
a
CONTENTS
Introduction
1.
2. Topological Invariants
3. Density of Topological Types
4. Proof of Main Theorem
5. Proofs of Trace Results
6. Determination of Flat and Semiflat Projections
References
2
4
6
10
12
15
17
Date: Oct. 19, 2017.
2000 Mathematics Subject Classification. 46L80, 46L40, 46L35, 46L85, 55N15.
Key words and phrases. C*-algebras, automorphisms, noncommutative torus, rotation alge-
bra, orbifold, K-group, Connes Chern characters.
1
Semiflat Orbifold Projections
2
1. INTRODUCTION
For each irrational number θ in (0, 1) the irrational rotation C*-algebra Aθ is
the universal (and unique) C*-algebra generated by unitaries U,V satisfying the
Heisenberg relation
VU = e2πiθUV.
The noncommutative Fourier transform is the canonical order four automor-
phism σ of Aθ defined by the equations
σ(U ) = V −1, σ(V ) = U.
The flip automorphism Φ = σ2 of the rotation C*-algebra Aθ is defined by Φ(U ) =
U −1, Φ(V ) = V −1, and it was studied extensively in [1] [2] [3] [9] [10].
If one represents the unitaries U,V on the Hilbert space L2(R) in the canonical
way as complex phase multiplication and translation operators, the automor-
phism σ corresponds exactly to the classical Fourier transform on L2, hence the
name.
In this paper we show that the semiflat positive cone K+SF
θ) of the K0-group of
the irrational rotation orbifold Aσ
θ (a sort of noncommutative sphere [12]) is deter-
mined by classes of positive trace and the vanishing of two topological invariants
(Theorem 1.4). We also determine semiflat and flat projections by their canoni-
cal traces and topological invariants up to Fourier-invariant unitary equivalence
(Theorems 1.11 and 1.12); namely, Fourier invariant projections as well as pro-
jections that are orthogonal to their transform (of which there are two kinds).
First, let us define these.
(Aσ
0
Definition 1.1. By a cylic (or σ-cyclic) projection in Aθ we mean a projection g that
is orthogonal to its σ-orbit -- that is, g,σ(g),σ2(g),σ3(g) are mutually orthogonal.
The associated σ-invariant projection
f = σ∗(g) := g + σ(g) + σ2(g) + σ3(g)
(1.1)
will be called flat (or σ-flat). We refer to g as a cyclic projection for f .
Flat projections have been used in [16] to obtain the K-inductive structure of
the Fourier transform σ.
Definition 1.2. A projection h is semicyclic (with respect to σ) if hσ(h) = 0 and
σ2(h) = h (i.e., h is flip invariant). The associated Fourier invariant projection
f = h + σ(h) is called semiflat. Two projections are σ-unitarily equivalent when
they are unitarily equivalent by a unitary that is σ-invariant.
The sum of two orthogonal flat (resp., semiflat) projections is flat (resp., semi-
flat). If g is a cyclic projection, then g + σ2(g) is semicyclic.
Semiflat Orbifold Projections
3
We will see that the trace and the topological genus of semiflat projections
determines them up to σ-unitary equivalence. (The meaning of topological genus
is given in Definition 2.1.)
Definition 1.3. The semiflat positive cone, denoted K+SF
nonzero classes in positive cone K+
θ) given by semiflat projections in Aσ
θ.
θ), consists of the
0 (Aσ
(Aσ
0
(We excluded the zero class for simplicity, although we could have included it.)
Theorem 1.4. (Main Theorem.) Let θ be irrational. The semiflat positive cone
K+SF
θ) consists of the K0-classes x of positive trace τ(x) and
(Aσ
0
ψ10(x) = ψ11(x) = 0.
Further, the semiflat projections come in three basic topological genera: (2, 0, 0),
(1, 1, 2), (0, 0, 2).
Thus, in general, a semiflat projection has genus that is an integral linear
combination of these three basic genus types.
Corollary 1.5. Let θ be irrational and let f , h be two semiflat projections in Aσ
θ.
Then f and h are σ-unitarily equivalent if and only if they have the same trace and
same genus.
It is well known that the K0-group of Aθ is Z2, and that the unique tracial state τ
of Aθ induces a group isomorphism τ∗ : K0(Aθ) → Z + Zθ when θ is irrational. (This
is a classic theorem of Pimsner and Voiculescu [6], and Rieffel [8] from 1980-81.)
Further, it is also known that each number in (0, 1) ∩ (Z + Zθ) is the trace of a
projection in Aθ, namely a Powers-Rieffel projection [8]. For our purposes here,
we prove the following related results for cyclic, flat, and semiflat projections.
Theorem 1.6. Let θ be irrational. Each number in (0, 1
cyclic projection in Aθ.
4 ) ∩ (Z + Zθ) is the trace of a
Theorem 1.7. Let θ be irrational. Each number in (0, 1) ∩ (4Z + 4Zθ) is the trace of
a flat projection in Aθ.
The analogous results hold for semicyclic and semiflat projections.
Theorem 1.8. Let θ be irrational. Each number in (0, 1
semicyclic projection in Aθ.
2 ) ∩ (Z + Zθ) is the trace of a
Theorem 1.9. Let θ be irrational. Each number in (0, 1) ∩ (2Z + 2Zθ) is the trace of
a semiflat projection in Aθ.
The next result shows that the vanishing of all the topological invariants of a
Fourier invariant projection means that it must be flat.
Theorem 1.10. Let e be a Fourier invariant projection in Aθ where θ is irrational.
If the topological invariants of e vanish (i.e., ψ∗∗(e) = 0), then e is flat.
Semiflat Orbifold Projections
4
Proof. By Lemma 3.6 the trace of e has to be in 4Z + 4Zθ. By Theorem 1.7,
τ(e) would also be the trace of some flat projection f . Since the Connes-Chern
character invariant T4 (mentioned below) of e and f are equal, they are unitarily
equivalent by a σ-invariant unitary, and hence e is flat also.
This characterizes what one might call the flat positive cone K+F
nonzero classes in the positive cone K+
Aσ
θ.
We also obtain a result that identifies cyclic and flat projections up to Fourier
θ), the
θ) represented by flat projections in
0 (Aσ
(Aσ
0
invariant unitary equivalence simply by means of the trace.
Theorem 1.11. Let θ be any irrational number, and let g1 and g2 be two cylic
projections in Aθ.
(1) Then g1 and g2 are σ-unitarily equivalent iff g1 and g2 have the same trace.
(2) Two flat projections f1 = σ∗(g1) and f2 = σ∗(g2) are σ-unitarily equivalent iff
they have the same trace iff g1 and g2 σ-unitarily equivalent.
We also have the corresponding result for semiflat projections and their semi-
cyclic components.
Theorem 1.12. Let θ be any irrational number, and let g and h be two semicyclic
projections in AΦ
θ .
(1) Then g and h are σ-unitarily equivalent iff they are Φ-unitarily equivalent iff
T2(g) = T2(h).
(2) Two semiflat projections f = g + σ(g) and f ′ = h + σ(h) are σ-unitarily equiv-
alent iff τ(g) = τ(h) and
φ00(g) = φ00(h),
φ11(g) = φ11(h),
φ01(g) + φ10(g) = φ01(h) + φ10(h).
Lastly, we show that all possible trace values are realized by Fourier invariant
projections.
Theorem 1.13. (See Theorem 5.7.) Let θ be irrational. Each number in (0, 1) ∩ (Z +
Zθ) is the trace of a Fourier invariant projection in Aθ.
2. TOPOLOGICAL INVARIANTS
In this section we recall the topological invariants for the flip and the Fourier
transform associated with the "twisted" unbounded traces on the canonical smooth
dense *-subalgebra A∞
θ.
The flip automorphism Φ has associated unbounded Φ-traces defined on the
basic unitaries U mV n by
φi j(U mV n) = e(−θ
2 mn) δm−i
2
δn− j
2
(2.1)
Semiflat Orbifold Projections
5
for i j = 00, 01, 10, 11, m, n ∈ Z, where δb
1 when a divides b, and 0 otherwise.
linear functionals defined on the canonical smooth dense *-subalgebra A∞
are Φ-invariant and satisfy the Φ-trace condition
a is the divisor delta function defined to be
(See [9] or [10].) These are (unbounded)
θ which
φi j(xy) = φi j(Φ(y)x)
for all x, y in A∞
θ. In addition, they are Hermitian maps: they are real on Hermitian
elements. Clearly, on the fixed point subalgebra A∞,Φ
θ under the flip they
give rise to (unbounded) trace functionals. Together with the canonical trace τ
one has the Connes-Chern character
of A∞
θ
T2 : K0(AΦ
θ ) → R5,
T2(x) = (τ(x);φ00(x),φ01(x),φ10(x),φ11(x))
(2.2)
the injectivity of which was shown in [9] (Proposition 3.2) for irrational θ.1 We
may sometimes refer to T2(x), or simply the φi j(x), as the Φ-topological invariant(s)
of the class. For the identity element one has T2(1) = (1; 1, 0, 0, 0).
For the Fourier transform σ, one has five basic unbounded twisted trace func-
θ of Aθ, defined as follows
tionals defined on the canonical smooth *-subalgebra A∞
on generic unitary elements:
ψ10(U mV n) = e(−θ
ψ11(U mV n) = e(−θ
4 (m + n)2) δm−n
,
4 (m + n)2) δm−n−1
2
2
,
ψ20(U mV n) = e(−θ
ψ21(U mV n) = e(−θ
ψ22(U mV n) = e(−θ
2 mn) δm
δn
2,
2 mn) δm−1
δn−1
2 mn) δm−n−1
.
2
2
2
2
,
(2.3)
(2.4)
(2.5)
(See [11]2.) These maps were calculated in [11] and were used in [12], [13]. The
Fourier Connes-Chern character is the group homomorphism
T4 : K0(Aσ
θ) → C6,
T4(x) = (τ(x); ψ10(x),ψ11(x); ψ20(x),ψ21(x),ψ22(x))
where τ is the canonical trace on Aσ
θ, the (orbifold) fixed point subalgebra with re-
spect to the Fourier transform. For irrational θ the map T4 is injective, so defines
a complete invariant for projections in the fixed point algebra Aσ
θ (up to Fourier
invariant unitary equivalence).3 For the identity one has T4(1) = (1; 1, 0; 1, 0, 0).
Definition 2.1. By the topological genus (or simply genus) of a semiflat projection
f we mean the triple (ψ20( f ),ψ21( f ),ψ22( f )).
1In [9] we worked with the crossed product algebra Aθ ⋊Φ Z2, but since this is strongly Morita
equivalent to the fixed point algebra, the injectivity follows.
2In [11] our Fourier transform was the inverse of the one used in this paper, so the unbounded
traces in [11] are "conjugate" to those above. See also the proof of Lemma 4.1.
3When θ is rational, one needs to include the Chern number arising from Connes' cyclic 2-
cocycle to the T4 invariant to ensure injectivity -- however, for our purposes, this is not necessary.
Semiflat Orbifold Projections
6
The injectivity of the Fourier Connes-Chern map T4 was shown in [12] for
a dense Gδ set of θ's, but later it was shown in [7], and independently in [5],
θ) ∼= Z9 for all θ, which gives the injectivity of T4 for all irrational θ.
that K0(Aσ
This allows us to conclude that since Aσ
θ has the cancellation property for any
irrational θ, two projections e and e′ in Aσ
θ are σ-unitarily equivalent if and only
if T4(e) = T4(e′).
One easily checks the following relations between the Φ and σ unbounded
traces
ψ20 = φ00,
ψ21 = φ11,
ψ22 = φ01 + φ10.
(2.6)
We will need to use the parity automorphism γ of Aθ defined by
γ(U ) = −U,
γ(V ) = −V
which will be useful because it commutes with the Fourier transform and has the
property of switching the signs of the topological maps ψ11,ψ22 (while preserving
the others). It also has the useful property
φ00γ = φ00,
φ11γ = φ11,
φ01γ = −φ01,
φ10γ = −φ10.
(2.7)
The topological numbers of σ-invariant projections are quantized. Indeed, in
view of [11] and [12], the ψ10,ψ11 invariants of such projections take values in the
lattice subgroup Z + Z( 1−i
Z, and ψ22
in Z.
2 ) of C; the ψ20,ψ21 invariants take values in 1
2
Analogous results can probably be established for the Cubic and Hexic trans-
forms studied in [4] and [15]. For example, for the Hexic transform ρ (the canon-
ical order 6 automorphism), there are three kinds of 'flat' projections: g + ρ(g)
(where g is ρ2-invariant), g + ρ(g) + ρ2(g) (where g is ρ3-invariant, ρ3 being the
flip), and g + ρ(g) + · · · + ρ5(g) (where g is ρ-cyclic).
3. DENSITY OF TOPOLOGICAL TYPES
In this section we establish key lemmas needed for the proof of the main the-
orem. For the reader's convenience, we quote the part of Lemma 3.1 from [9] (p.
594) that is relevant to the proofs below.
Lemma 3.1. Let α = rθ + s be irrational in the interval ( 1
2, 1) where r, s are integers.
With U r and V being unitaries satisfying VU r = e2πiαU rV , there exists a Powers-
Rieffel projection
e = V g(U r) + f (U r) + g(U r)V −1
Semiflat Orbifold Projections
7
of trace α that is flip-invariant, where f , g are certain smooth functions. Further, if
r is even, then
φi j( f (U r)) = 0,
φi j(g(U r)V −1) =(0
1
2
δ j−1
2
δi
2
if s is even,
if s is odd.
If r is odd, one has
φi j( f (U r)) = 1
2 (−1)iδ j
2,
φi j(g(U r)V −1) = 1
4(−1)i(s+1)δ j−1
2
.
(N.B., this slightly more simplified version of the lemma was obtained by setting
"p = q = 0" in the notation of Lemma 3.1 of [9].)
Lemma 3.2. Let θ be irrational. There are flip-invariant Powers-Rieffel projections
e, e′, e′′ in Aθ with Φ-invariants
T2(e) = (τ(e); 0, 1, 0, 0),
T2(e′) = (τ(e′); 1
2 , 1
2, 1
2, 1
2),
T2(e′′) = (τ(e′′); 1, 0, 0, 0)
such that the set of traces of each type is dense in (0, 1
2).
Proof. Using Lemma 3.1 with α = rθ + s in ( 1
Powers-Rieffel projection
2, 1) where r is even and s odd, the
has invariant T2(e1) = (rθ + s; 0, 1, 0, 0). Indeed, in this case φi j( f (U r)) = 0 and
e1 = V g(U r) + f (U r) + g(U r)V −1
Thus, φ01(e1) = 1 and the other φi j(e1) = 0.
φi j(e1) = 2φi j(g(U r)V −1) = δ j−1
2
δi
2.
Now let us take any other irrational α′ = r′θ + s′ in ( 1
2, 1) but this time with both
r′ and s′ even. The corresponding Powers-Rieffel projection (Lemma 3.1)
e2 = V g(U r′
) + f (U r′
) + g(U r′
)V −1
2, 1), choosing 1 > α > α′ > 1
has φi j(e2) = 0 and T2(e2) = (α′; 0, 0, 0, 0). Since the sets of such α and α′ are dense
in ( 1
2 ). Upon
picking a flip-invariant unitary w such that we2w∗ ≤ e1, we obtain the flip-invariant
projection
2 we get a dense set of traces {α−α′} in (0, 1
e = e1 − we2w∗
with invariant T2(e) = (τ(e); 0, 1, 0, 0) and traces dense in (0, 1
jections e in the statement of the lemma.
2 ), giving us the pro-
Note that
T2(1 − e2) = (1; 1, 0, 0, 0) − (α′; 0, 0, 0, 0) = (1 − α′; 1, 0, 0, 0)
whose traces are dense in (0, 1
of the lemma.
2), which gives us the projections e′′ in the statement
Semiflat Orbifold Projections
8
Now let's suppose that r and s are odd and let e0 be the associated Powers-
δ j−1
2
2
and
Rieffel projection of trace β = rθ + s ∈ ( 1
2, 1). Then φi j(e0) = 1
2(−1)iδ j
2 + 1
which in view of (2.7) gives
T2(e0) = (β; 1
2, 1
2, − 1
2, 1
2)
T2(γe0) = (β; 1
2, − 1
2, 1
2, 1
2).
Subtracting from γe0 subprojections equivalent to e2's of traces α′ less than β (as
done previously), we obtain flip-invariant projections e′′′ such that
T2(e′′′) = T2(γe0) − T2(e2) = (β − α′; 1
2 , − 1
2, 1
2, 1
2)
where the set of traces {β − α′} is dense in (0, 1
2 ). Adding projections Φ-unitarily
equivalent to e orthogonally to e′′′ one gets flip-invariant projections e′ such that
T2(e′) = (τ(e′); 1
2 , 1
2 , 1
2 , 1
2 )
with traces τ(e′) is dense in (0, 1
2 ).
In view of Theorem 1.9, the projections e, e′, e′′ in Lemma 3.2 can be conjugated
by suitable flip-invariant unitaries u so that, for instance, h = ueu∗ is under a
semicyclic projection g of some suitably larger trace.
(Recall that this means
gσ(g) = 0 where g is flip invariant.) Clearly, the T2 invariants of e and h are the
same (since u is flip-invariant), with the difference that h is now a semicyclic
projection. Therefore, we obtain the following.
Corollary 3.3. Let θ be irrational. There are semicyclic projections h, h′, h′′ in Aθ
with invariants
T2(h) = (τ(h); 0, 1, 0, 0),
T2(h′) = (τ(h′); 1
2, 1
2, 1
2, 1
2),
T2(h′′) = (τ(h′′); 1, 0, 0, 0)
such that the set of traces of each type is dense in (0, 1
2).
Now consider the semiflat projection f = h + σ(h) associated to the first type of
projection h in this corollary. We have
likewise ψ21( f ) = 2φ11(h) = 0, and
ψ20( f ) = 2ψ20(h) = 2φ00(h) = 0
ψ22( f ) = 2ψ22(h) = 2φ01(h) + 2φ10(h) = 2.
Therefore f is semiflat of topological genus (0, 0, 2).4 In addition, the traces of
such f form a dense set in (0, 1).
In the same way, from h′ and h′′ we obtain semiflat projections f ′ and f ′′ with
respective topological genera (1, 1, 2) and (2, 0, 0). We will say that a class of pro-
jections has trace density when the set of its traces is dense in (0, 1). We have
therefore addressed part of the following result.
4The author had some difficulty finding semiflat projections of genus (0, 0, 2).
Semiflat Orbifold Projections
9
Lemma 3.4. Let θ be irrational. Each triple
(1, 1, 2),
(2, 0, 0),
(0, 0, 2),
(0, 2, 0),
(0, 0, −2),
is the topological genus of semiflat projections in Aσ
(−1, −1, −2),
θ with dense traces in (0, 1).
(−2, 0, 0),
(0, −2, 0),
Applying the automorphism γ to semiflats of genus (0, 0, 2) we obtain semiflat
projections of genus (0, 0, −2) with trace density.
Since (0, 2, 0) = 2(1, 1, 2) + (−2, 0, 0) + 2(0, 0, −2) is a positive linear combination
of genera with trace density, we immediately get semiflat projections of genus
(0, 2, 0) with trace density. To complete the proof Lemma 3.4 we must deal with
the remaining genera:
These, however, follow from the lemma.
(−2, 0, 0),
(−1, −1, −2),
(0, −2, 0).
Lemma 3.5. For each semiflat projection of genus (a, b, c), there is a semiflat pro-
jection of genus (−a, −b, −c). If the former has trace density, so does the latter.
Proof. Let p = h +σ(h) be a semiflat projection of genus (a, b, c) (with trace density).
We show how to construct semiflat projections of its negative genus (−a, −b, −c)
(with trace density). By Theorem 1.7, choose a flat projection
k = g + σ(g) + σ2(g) + σ3(g)
where g is a cyclic projection such that 2τ(h) = τ(p) < τ(k) = 4τ(g). Choose a flip-
invariant unitary u such that uhu∗ ≤ g +σ2(g), where g +σ2(g) is flip-invariant and
semicyclic. The difference projection
h := g + σ2(g) − uhu∗
is semicyclic as well (and is flip invariant) and its associated semiflat projection
p := h + σ( h) has genus (−a, −b, −c) (since the ψi j invariants of g and k vanish).
Further, the traces
τ( p) = 2τ( h) = 4τ(g) − 2τ(h) = τ(k) − τ(p)
are dense in (0, 1) (since the set of traces τ(k) and τ(p) are each dense in (0, 1)).
This completes the proof of Lemma 3.4.
Lemma 3.6. Let x be a class in K0(Aσ
If all ψi j(x) = 0, then τ(x) ∈ 4Z + 4Zθ.
θ). If ψ10(x) = ψ11(x) = 0, then τ(x) ∈ 2Z + 2Zθ.
The proof of this lemma is contained in the proof of Theorem 1.4 given in
Section 4 below (see paragraph following equation (4.3) below).
Semiflat Orbifold Projections
10
4. PROOF OF MAIN THEOREM
In this section we prove the main theorem on the determination of the semiflat
positive cone of K0 of the Fourier orbifold Aσ
θ. Before doing so we state a lemma
that is essentially a paraphrase of a result from [11], which was originally stated
for crossed products, for our fixed point subalgebra situation.
Lemma 4.1. The range of the homomorphism T4 : K0(Aσ
nine vectors
θ) → C6 is spanned by the
V1 = (2; 0, 0; 2, 0, 0)
V2 = (2; 1 + i, 0; 0, 0, 0)
V3 = (1; 1, 0; 1, 0, 0)
V4 = (2; 0, 0; 0, 2, 0)
V5 = (2; 0, 1 + i; 0, 0, 0)
V6 = (1; 0, 1; 0, 1, 0)
2, 1
2 − 1
V7 = (θ; 1
2 , 1)
2 i;
2 i; − 1
2 i, − 1
2 − 1
V8 = (θ; − 1
2 , 1
2 + 1
2 i, − 1
V9 = (θ; − 1
2 i;
2 − 1
2 + 1
1
2 − 1
2 i, 1
2, −1)
2 , − 1
2 , 1).
1
Proof. These can be obtained from the range of the associated homomorphism
calculated for the crossed product Aθ ⋊σ Z4 in [11] (see character table on page
645). The only difference that we need to take into account is that the "Fourier
transform" used in [11] was the inverse of the one used in the current paper --
and this has the effect of taking the complex conjugates of the ψ10,ψ11 values
obtained in [11], which correspond, respectively, to the values of the maps "T10"
and "λ1/4T11" used in the character table therein. Further, we need to multiply
all entries in that character table by 4 in view of the normalizations used for
the unbounded traces in [11]. Once these are taken into account, we obtain
the above 9 vectors from those in [11] in view of the canonical isomorphism
K0(Aσ
θ) ∼= K0(Aθ ⋊σ Z4).
We are now ready to prove the Main Theorem 1.4.
Proof. (Proof of Theorem 1.4.) Fix a class x in K0(Aσ
θ) such that τ(x) > 0 and
ψ10(x) = ψ11(x) = 0. If τ(x) > 1, by Theorem 1.9 we can subtract from x the sum
of a finite number of K0-classes of semiflat projections so that the difference has
positive trace less than 1. Further, the fact that τ(x) 6= 1 will follow from the
computation below which show that the vanishing of ψ10(x) and ψ11(x) implies
that the trace of x is a multiple of 2 -- see, for example, equation (4.1) below.
Therefore, with no loss of generality we may assume that 0 < τ(x) < 1.
Write T4(x) as an integral linear combination of the nine vectors in Lemma 4.1:
Semiflat Orbifold Projections
11
T4(x) =
9
∑
j=1
N jVj
for some integers N j. Reading off the ψ10 and ψ11 coordinates of x, we have
ψ10(x) = N2(1 + i) + N3 + N7( 1
2 − 1
2 i) + N8(− 1
2 − 1
2 i) + N9(− 1
2 + 1
2 i) = 0
and
ψ11(x) = N5(1 + i) + N6 + N7( 1
2 − 1
2 i) + N8(− 1
2 − 1
2 i) + N9(− 1
2 + 1
2 i) = 0.
Solving these gives
N6 = N3,
N5 = N2,
N7 = N9 − N3,
N8 = 2N2 + N3.
In terms of the integers N1, N2, N3, N4, N9, the total trace is
τ(x) = ∑
j
N jτ(Vj) = 2N1 + 4N2 + 2N3 + 2N4 + (2N2 + 2N9)θ
(4.1)
which is a multiple of 2 as we had noted at the beginning of the proof. Simplify-
ing, one gets the ψ2k invariants
ψ20(x) = 2N1 − N2 + N9
ψ21(x) = 2N4 − N2 + N9
ψ22(x) = 2N9 − 2N2 − 2N3.
Therefore one gets
T4(x) = (2N1 + 4N2 + 2N3 + 2N4 + (2N2 + 2N9)θ; 0, 0;
(4.2)
2N1 − N2 + N9, 2N4 − N2 + N9, 2N9 − 2N2 − 2N3)
= N1(2; 0, 0; 2, 0, 0) + N2(4 + 2θ; 0, 0; −1, −1, −2) + N3(2; 0, 0; 0, 0, −2)
(4.3)
+ N4(2; 0, 0; 0, 2, 0) + N9(2θ; 0, 0; 1, 1, 2).
We digress momentarily to note that in view of this calculation, the condition
ψ10(x) = ψ11(x) = 0 has yielded the conclusion that τ(x) is in 2Z + 2Zθ (as can
be seen from (4.2)), thus establishing the first assertion of Lemma 3.6.
If, in
addition, the remaining invariants ψ2k(x) vanish, then it is easy to check that τ(x)
is in 4Z + 4Zθ, which establishes the second assertion of Lemma 3.6. Thus, in
particular, if all the topological invariants of a K0-class x vanish, then its trace is
a multiple of 4 - and we know from Theorem 1.7 that any such number is the
trace of a flat projection.
Returning to our current proof, in view of Lemma 3.4 we can pick semiflat
projections f1, f2, f3, f4, f9 with the respective topological genera appearing in the
Semiflat Orbifold Projections
12
last equality in (4.3), and of appropriately small trace, such that
T4 x − ∑
i=1,2,3,4,9
Ni[ fi]! = (α; 0, 0; 0, 0, 0)
where α < 1 is some positive number in Z + Zθ. We now have a class with all its
topological invariants vanishing, so by Lemma 3.6, α = 4α′ for some α′ ∈ Z + Zθ.
By Theorem 1.7, there is flat projection f of trace 4α′ and T4[ f ] = (4α′; 0, 0; 0, 0, 0).
This gives
T4(x) = T4 [ f ] + ∑
i=1,2,3,4,9
Ni[ fi]!
and by the injectivity of T4, the class x is a finite non-negative integral linear
combination of classes of semiflat projections (at least one of them nonzero).
Since τ(x) < 1, Lemma 5.1 allows us to write x = [e] for a single semiflat projec-
tion e (since the projections f , fi could all be unitarily combined into a sum of
orthogonal semiflat projections, which is also semiflat).
5. PROOFS OF TRACE RESULTS
In this section we prove Theorems 1.7 and 1.9 stated in the Introduction. To
do this we begin with two lemmas.
Lemma 5.1. Let e, e1, . . . , en be Fourier invariant projections in Aθ such that
t := τ(e1) + · · · + τ(en) < τ(e).
There are σ-invariant unitaries w1, . . . , wn such that
1 + · · · + wnenw∗
n
w1e1w∗
is a Fourier invariant subprojection of e of trace t.
Proof. Since the order structure on K0(Aσ
τ, the hypothesis implies that there is a projection Q in Mm(Aσ
θ) is determined by the canonical trace
θ) such that
[e1 ⊕ · · · ⊕ en ⊕ Q] = [e] = [e ⊕ Or]
in K0(Aσ
property of Aσ
θ), where r = n + m − 1 and Or is the zero r × r matrix. By the cancellation
θ, there is a unitary W in Mr+1(Aσ
θ) such that
W (e1 ⊕ · · · ⊕ en ⊕ Q)W ∗ = e ⊕ Or.
From this, one obtains a set f1, . . . , fn of pairwise orthogonal subprojections of e
such that [ f j] = [e j] for each j, which by cancellation again gives unitaries w j in
Aσ
θ such that f j = w je jw∗
j . One therefore gets the projection
w1e1w∗
1 + · · · + wnenw∗
n
Semiflat Orbifold Projections
13
contained in e and with trace t.
We cite the following lemma from [14].
Lemma 5.2. (See Theorem 1.6 of [14].) Let θ be irrational, p/q a rational (in
reduced form) approximant of θ such that 0 < qqθ − p < 1. Then for each positive
integer k such that kqθ− p < 1
4 , there exists a cyclic projection in Aθ of trace kqθ− p.
(Of course, one would then have the corresponding flat projection whose trace
is 4kqθ − p.)
Proposition 5.3. Let θ be any irrational number in (0, 1). Then each number in
(0, 1) ∩ (4Z + 4Zθ) is the trace of a σ-flat projection.
Proof. Fix t ∈ (0, 1) ∩ (4Z + 4Zθ). With no loss of generality we can assume t =
4k(nθ−m) where k, n ≥ 1 and m ≥ 0. (If t = 4ℓ(r −sθ) where r, s ≥ 0, then t = 4ℓ[s(1−θ)−
(s − r)] so that we could obtain a σ-flat projection in A1−θ of this trace which maps
to Aθ via a Fourier compatible isomorphism that gives one a σ-flat projection of
trace t.)
Since 0 ≤ m
n < θ, one can choose a pair of consecutive convergents p
q , p′
q′ of θ
q < θ < p′
such that 0 < m
n < p
θ = p′(qθ − p) + p(p′ − q′θ),
each as sums of positive terms. Thus
q′ where p′q − pq′ = 1. We can write
1 = q′(qθ − p) + q(p′ − q′θ)
t = 4k[np′(qθ − p) + np(p′ − q′θ) − mq′(qθ − p) − mq(p′ − q′θ)]
= 4a(qθ − p) + 4b(p′ − q′θ)
where a = k(np′ − mq′) and b = k(np − mq) are positive integers. Now we are in
the situation of Lemma 5.2 which gives us σ-flat projections f1 and f2 in Aθ
with respective traces 4a(qθ − p) and 4b(p′ − q′θ). Using Lemma 5.1 there exists
a Fourier invariant unitary w such that the orthogonal sum f1 + w f2w∗ is a flat
projection with trace t.
Corollary 5.4. Let θ be irrational. Then each number in (0, 1
of a cyclic projection.
4) ∩ (Z + Zθ) is the trace
Theorem 5.5. Let θ be irrational. Then each number in (0, 1
of a semicyclic projection.
2) ∩ (Z + Zθ) is the trace
Proof. First, note that each number 2x in (2Z + 2Zθ) ∩ (0, 1
cyclic projection. Since x < 1
of trace x. The projection g + σ2(g) is then semicyclic of trace 2x.
2) is the trace of a semi-
4, the preceding corollary gives a cyclic projection g
Now fix t ∈ (0, 1
2 ) such that
t < 2x. By the claim just proved, there exists a semicyclic projection h of trace
2) ∩ (Z + Zθ). By density, pick 2x ∈ (2Z + 2Zθ) ∩ (0, 1
Semiflat Orbifold Projections
14
2x. Picking a flip-invariant subprojection e of h of trace t (as t < 2x), one gets a
semicyclic projection e of trace t.
Lemma 5.6. Let θ ∈ (0, 1) be irrational. Then for each pair of integers m, n there
exists a Fourier invariant projection of trace (m2 + n2)θ mod 1.
Proof. By Theorem 1.1 of [13] for any rational p/q such that 0 < qqθ− p < 1, there
is a Fourier invariant projection of trace qqθ − p. Applying this with p = 0, q = 1,
one obtains a Fourier invariant projection of trace θ. Fix m, n and let
The unitaries
θ′ = (m2 + n2)θ mod 1.
are easily checked to satisfy
2 mnθ)V −nU m,
2 mnθ)U nV m
eU = e( 1
eVeU = e(θ′)eUeV ,
eV = e( 1
σ(eU) =eV −1,
σ(eV ) = eU
so that σ induces the Fourier transform on the rotation C*-subalgebra Aθ′ of Aθ
generated by eU and eV . Therefore, by what was just noted, there exists a Fourier
invariant projection in Aθ′ (hence in Aθ) of trace θ′, as required.
Theorem 5.7. Let θ be irrational in (0, 1). Each t ∈ (0, 1) ∩ (Z + Zθ) is the trace of
a Fourier invariant projection.
Proof. Write t = mθ − n < 1, and assume, with no loss of generality, that m > 0. By
Lagrange's Theorem, write m = m2
4 as a sum of four integer squares.
By Lemma 5.6, there are Fourier invariant projections e of trace (m2
2)θ− n1 < 1
and f of trace (m2
4)θ − n2 < 1 for some nonnegative integers n1, n2. The class
[e] + [ f ] in K0(Aσ
3 + m2
3 + m2
2 + m2
1 + m2
1 + m2
θ) has positive trace
(m2
1 + m2
2 + m2
3 + m2
4)θ − n1 − n2 = t + k < 2
(5.1)
If k = 0, then since the sum of the traces
where k = n − n1 − n2 is either 0 or 1.
is less than 1, Lemma 5.1 implies that there are unitaries u, v in
of e and f
Aσ
θ such that ueu∗ + v f v∗ is a Fourier invariant projection of trace t.
If k = 1,
then [e] + [ f ] − [1] = [e] − [1 − f ] has positive trace t (from (5.1)), which means that
[e] > [1 − f ] in K0(Aσ
θ), so that 1 − f is unitarily equivalent to a subprojection of
e by a unitary w in Aσ
θ: w(1 − f )w∗ ≤ e. One therefore gets the Fourier invariant
projection e − w(1 − f )w∗ of trace t.
Semiflat Orbifold Projections
15
6. DETERMINATION OF FLAT AND SEMIFLAT PROJECTIONS
In this section we prove Theorems 1.11 and 1.12.
Theorem 6.1. Let θ be any irrational number, and let g1 and g2 be two cyclic
projections in Aθ.
(1) Then g1 and g2 are unitarily equivalent by a unitary in Aσ
θ if and only if g1
and g2 have equal traces.
(2) Two flat projections f1 = σ∗(g1) and f2 = σ∗(g2) are σ-unitarily equivalent iff
they have the same trace iff g1 and g2 σ-unitarily equivalent.
If g1 and g2 are σ-unitarily equivalent then clearly so
Proof. We prove (2) first.
are f1 and f2. So we start with two σ-unitarily equivalent flat projections f1, f2.
Since the projections g1, g2 have the same trace, they are Murray von Neumann
equivalent in Aθ (by a theorem of Rieffel). Let v ∈ Aθ be a partial isometry such
that vv∗ = g1, v∗v = g2 and g1v = v = vg2. As the projections g1 and g2 are cyclic, one
has
v∗σ j(v) = 0,
vσ j(v∗) = 0
for j = 1, 2, 3. The first of these follows by replacing v by g1v and likewise the
second by replacing v by vg2. This lends us the Fourier invariant element
w = v + σ(v) + σ2(v) + σ3(v)
which acts as a partial isometry between the flat projections
ww∗ = f1,
w∗w = f2.
Further, one has
g1w = vv∗[v + σ(v) + σ2(v) + σ3(v)] = v,
wg2 = [v + σ(v) + σ2(v) + σ3(v)]v∗v = v
where we noted that vv∗v = v. These give f1w = w = w f2.
As the complements 1 − f1 and 1 − f2 are also equivalent projections in Aσ
θ (hav-
ing same T4's), there is a σ-invariant partial isometry x such that xx∗ = 1 − f1 and
x∗x = 1 − f2, as well as (1 − f1)x = x = x(1 − f2). Since one has wx∗ = 0 = w∗x, the
element w + x is a σ-invariant unitary that is easily checked to satisfy
g1(w + x) = (w + x)g2.
This proves (2).
To see (1), assume g1 and g2 have equal trace. Since their corresponding flat
projections f1 = σ∗(g1), f2 = σ∗(g2) also have equal traces, we have T4( f1) = T4( f2) =
(trace; 0, 0; 0, 0, 0) (recalling that the topological invariants ψ⋆ of flat projections all
vanish). By the injectivity of T4, it follows that f1 and f2 are σ-unitarily equivalent,
and the result follows from (2).
Semiflat Orbifold Projections
16
Remark 6.2. In view of (2), it also follows that g1 is σ-unitarily equivalent to
σ j(g2) for any j.
Theorem 6.3. Let θ be any irrational number, and let g and h be two semicyclic
projections in AΦ
θ .
(1) Then g and h are σ-unitarily equivalent iff they are Φ-unitarily equivalent iff
(2) Two semiflat projections f = g + σ(g) and f ′ = h + σ(h) are equivalent in Aσ
θ
T2(g) = T2(h).
iff τ(g) = τ(h) and
φ00(g) = φ00(h),
φ11(g) = φ11(h),
φ01(g) + φ10(g) = φ01(h) + φ10(h),
Proof. For (1) it is enough to assume that g and h are Φ-unitarily equivalent
(since it is already known from [9] that this is equivalent to T2(g) = T2(h)). Let u
be a partial isometry in AΦ
θ such that uu∗ = g, u∗u = h (and also gu = u = uh). The
orthogonalities gσ(g) = hσ(h) = 0 give u∗σ(u) = 0 = uσ(u∗). Letting w = u + σ(u), we
get
ww∗ = g + σ(g) =: f ,
w∗w = h + σ(h) =: f ′
(and f w = w = w f ′) so that w is a σ-invariant partial isometry. Further, gw = u = wh
(as uu∗u = u). Since 1 − f and 1 − f ′ are also equivalent projections in Aσ
θ (since they
have equal T4's), there is a σ-invariant partial isometry w′ such that w′w′∗ = 1 − f
and w′∗w′ = 1 − f ′. One then forms the σ-invariant unitary
W = w + w′
which satisfies gW = W h.
For statement (2), if f = g + σ(g) and f ′ = h + σ(h) are equivalent in Aσ
θ then in
view of (2.6) one has the assertion stated on the φi j values (and the trace). The
converse follows likewise since the T4 invariants of f and f ′ are equal in view of
the hypothesis.
Remark 6.4. We emphasize that the semicyclic projections g, h in (2) of the pre-
ceding theorem are not necessarily equivalent even in the flip fixed point algebra
AΦ
θ . Indeed, given any g one can consider a semicyclic projection h arising from
the equation
T2(h) = T2(g) + (0; 0, n, −n, 0)
for any integer n (where h is obtained in the same manner that lead to Corollary
3.3). Such h gives rise to a semiflat projection f ′ = h+σ(h) equivalent to f = g+σ(g)
(in Aσ
θ), but h is neither equivalent to g nor to σ(g). This contrasts with statement
(2) of Theorem 1.11.
Acknowledgement. Too bad NSERC changed its rules which discouraged the
author's applying for a research grant, or else it would have been (happily) ac-
knowledged here! Nevertheless, NSERC is thanked for some 20 years of grant
support (and the momentum it gave the author).
Semiflat Orbifold Projections
17
REFERENCES
[1] O. Bratteli, G. A. Elliott, D. E. Evans, A. Kishimoto, Non-commutative spheres I, Internat. J. Math. 2
(1990), no. 2, 139 -- 166.
[2] O. Bratteli, G. A. Elliott, D. E. Evans, A. Kishimoto, Non-commutative spheres II: rational rotation, J.
Operator Theory 27 (1992), 53 -- 85.
[3] O. Bratteli, A. Kishimoto, Non-commutative spheres III: Irrational rotation, Comm. Math. Phys. 147
(1992), 605 -- 624.
[4] J. Buck and S. Walters, Connes-Chern characters of hexic and cubic modules, J. Operator Theory 57
(2007), 35 -- 65.
[5] S. Echterhoff, W. L uck, N. C. Phillips, S. Walters, The structure of crossed products of irrational rotation
algebras by finite subgroups of SL2(Z), J. Reine Angew. Math. (Crelle's Journal) 639 (2010), 173-221.
[6] M. Pimsner and D. Voiculescu, Exact sequences for K-groups and Ext-groups of certain crossed product
C*-algebras, J. Operator Theory 4 (1980), 93-118.
[7] A. Polishchuk, Holomorphic bundles on 2-dimensional noncommutative toric orbifolds, in Noncommuta-
tive Geometry and Number Theory, by C. Consani and M. Marcolli (eds.), Vieweg Publ. (2006), 341-359.
[8] M. Rieffel, C*-algebras associated with irrational rotations, Pacific J. Math. 93, No. 2 (1981), 415 -- 429.
[9] S. G. Walters, Projective modules over the non-commutative sphere, J. London Math. Soc. (2) 51 (1995),
no. 3, 589-602.
[10] S. G. Walters, Inductive limit automorphisms of the irrational rotation algebra, Comm. Math. Phys. 171
(1995), 365 -- 381.
[11] S. Walters, Chern characters of Fourier modules, Canad. J. Math. 52 (2000), No. 3, 633 -- 672.
[12] S. Walters, K-theory of non commutative spheres arising from the Fourier automorphism, Canad. J.
Math. 53, No. 3 (2001), 631 -- 672.
[13] S. Walters, The AF structure of non commutative toroidal Z/4Z orbifolds, J. Reine Angew. Math. (Crelle's
Journal) 568 (2004), 139 -- 196. arXiv: math.OA/0207239.
[14] S. Walters, Decomposable Projections Related to the Fourier and flip Automorphisms, Math. Scand. 107
(2010), 174-197.
[15] S. Walters, Toroidal Orbifolds of Z3 and Z6 Symmetries of Noncommutative Tori, Nuclear Physics B 894
(2015), 496-526. DOI: http://dx.doi.org/10.1016/j.nuclphysb.2015.03.008
[16] S. Walters, The K-inductive Structure of the Noncommutative Fourier Transform, preprint (2017), 13
pages.
DEPAR TMENT OF MATHEMATICS AND STATISTICS, UNIVERSITY OF NOR THERN B.C., PRINCE
GEORGE, B.C. V2N 4Z9, CANADA.
E-mail address: [email protected] or
URL: hilbert.unbc.ca or web.unbc.ca/walters
[email protected]
|
1101.1235 | 1 | 1101 | 2011-01-06T15:39:59 | Equivalence of Fell Systems and their Reduced C*-Algebras | [
"math.OA",
"math.FA",
"math.KT"
] | This paper is aimed at investigating links between Fell bundles over Morita equivalent groupoids and their corresponding reduced C*-algebras. Mainly, we review the notion of Fell pairs over a Morita equivalence of groupoids, and give the analogue of the Renault's Equivalence Theorem for the reduced C*-algebras of equivalent Fell systems. Eventually, we will use this theorem to connect the reduced C*-algebra of an S^1-central groupoid extension to that of its associated Dixmier-Douady bundle. | math.OA | math |
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
Abstract. This paper is aimed at investigating links between Fell bundles over Morita equivalent groupoids and their
corresponding reduced C∗-algebras. Mainly, we review the notion of Fell pairs over a Morita equivalence of groupoids, and
give the analogue of the Renault's Equivalence Theorem for the reduced C∗-algebras of equivalent Fell systems. Eventually,
we will use this theorem to connect the reduced C∗-algebra of an S1-central groupoid extension to that of its associated
Dixmier-Douady bundle.
Introduction
A Fell system consists of a pair (G, E), where E is a Fell bundle over the groupoid G. The notion of (Morita)
equivalence of Fell systems was first introduced by S. Yamagami in [22], and by then it was studied by Muhly
in [9] and very recently by Muhly and Williams in [12] where the authors prove that if (Γ, F) and (G, E) are
equivalent, then their full C∗-algebras C∗(Γ; F) and C∗(G; E) are Morita equivalent (see [9, Theorem 11], and [12,
Theorem 6.4]). However, it has not been known so far whether an equivalence of Fell systems gives rise to a
Morita equivalence between the associated reduced C∗-algebras.
The first motivation of our work came from twisted K-theory: to every groupoid G and every cocycle α ∈
α(G) are defined as the C∗-algebraic
r(Γα, Lα) (cf. [21]). Moreover, it is known that when α ∼ β, then not only Γα is
r(Γβ, Lβ) are Morita equivalent
β(G). This has led us to a generalisation of the so-called Renault's
C2(G•, T) is associated a Fell system (Γα, Lα), and the twisted K-groups K∗
K-groups of the reduced C∗-algebra C∗
Morita equivalent to Γβ but also the associated reduced C∗-algebras C∗
(see [21, Proposition 3.3]); so that K∗
equivalence Theorem for reduced groupoid C∗-algebras ( [18, Theorem 13]) to Fell systems.
r(Γα, Lα) and C∗
α(G) (cid:27) K∗
We recall from [21] some concepts related to groupoids such as generalized homomorphisms and Dixmier-Douady
bundles in §1, and we review the basics of Fell systems and their reduced C∗-algebras from [5] and [21] in §2.
In §3, we discuss the notion of equivalence of Fell systems of [9] and [12] from another formalism that better
suits with the construction of the linking Fell systems introduced in §4. The equivalence theorem for the reduced
C∗-algebras of Fell systems is proved in §5, and then, in §6, we apply this theorem to link the C∗-algebra associated
to an S1-central extension of a groupoid G to the reduced cross-product A ⋊r G, where A is some Dixmier-Douady
bundle.
1. Preliminaries
Although we assume that the reader is familiar with the language of groupoids (see for instance [16]), we
recall some of their basics used substantially throughout this paper. All the groupoids we are working with are
supposed to be Hausdorff, locally compact, second countable, and are equipped with Haar systems. They are
also assumed to have finite-dimentional base spaces, in the sense of [4].
1.1. Given two groupoids G
r
s
/ G(0) and Γ
r
s
/ Γ(0) , a generalized morphism Z : Γ −→ G consists of a
/ G(0) (s and r are called generalized source map
(locally compact Hausdorff) space Z, two maps Γ(0)
and generalized range map of Z, respectively), a left action of Γ on Z with respect to r, a right action of G on Z with
Z
r
s
The first author is supported by the German Research Foundation (DFG) and the Universit´e franco-allemande (DFH-UFA) via the International
Research Training Group 1133 "Geometry and Analysis of Symmetries".
1
/
/
/
/
/
/
o
o
/
2
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
respect to s, such that the two actions commute, and Z −→ Γ(0) is a right G-principal bundle. Such a morphism is
a Morita equivalence if in addition, Z −→ G(0) is a left Γ-principal bundle; in this case, we say that Γ and G are Morita
equivalent, and we write Γ ∼Z G. The terminology of generalized morphism is justified by the fact that any strict
groupoid homomorphism (see [16]) f : Γ −→ G induces a generalized one Z f : Γ −→ G, where Z f := Γ(0) × f,G(0),r G,
with generalized source and range s(y, g) := s(g) and r(y, g) := y, while the actions are γ · (s(γ), g) := (r(γ), f (γ)g),
and (y, g) · g′ := (y, gg′).
If Γ ∼Z G, then G ∼Z−1 Γ, where Z−1 is Z as topological space, and if ♭ : Z −→ Z−1 is the identity map,
1.2.
the generalized source and range are s♭(♭(z)) := r(z) and r♭(♭(z)) := s(z). The left G-action on Z−1 is given by
g · ♭(z) := ♭(zg−1) for (z, g−1) ∈ Z ∗ G, and the right Γ-action is ♭(z) · γ := ♭(γ−1z) whenever (γ−1, z) ∈ Γ ∗ Z.
If
Z2 by the
Γ ∼Z1
equivalence relation (z1, z2) ∼ (z1 g1, g−1
G, where Z1 ×G1 Z2 is the quotient of the fibre product space Z1 ×
G, then Γ ∼Z1×G1
G1 ∼Z2
s1,G(0)
1 ,r2
Z2
1 z2).
1.3. A Dixmier-Douady bundle A over G is a locally trivial bundle A −→ G(0) with fibre the C∗-algebra K of
compact operators on the separable infinite-dimensional Hilbert space H = l2(N), together with an action α by
automorphisms of G on A; that is, a continuous family of isomorphisms of C∗-algebras αg : As(g) −→ Ar(g) such
that αgh = αg ◦ αh whenever g and h are composable, and αg−1 = α−1
g . Such a bundle is represented by the triple
(A, G, α). Let HG := L2(G) ⊗ H, where L2(G) is the G-equivariant C0(G(0))-Hilbert module obtained by completing
G(g). We say that two Dixmier-Douady bundles
A and B are Morita equivalent, and write A ∼ B, if A ⊗ K(HG) (cid:27) B ⊗ K(HG). The set of Morita equivalence classes
of Dixmier-Douady bundles forms an abelian group Br(G) called the Brauer group of G. We refer to [6], [20], or [8]
for more details about the structures of Br(G).
Cc(G) with respect to the scalar product hξ, ηi(x) =RGx ξ(g)η(g)dµx
2. Fell systems and their reduced C∗-algebras
If p : E −→ G is a Banach bundle, we set
E[2] :=n(e1, e2) ∈ E × E (p(e1), p(e2)) ∈ G(2)o .
Let m : G(2) −→ G denote the partial multiplication of the groupoid G
r
s
/ G(0) . Then m∗E −→ G(2) is a Banach
bundle.
Definition 2.1. (cf. [5], [21, Appendix A]) . A multiplication on E consists of a continuous map E[2] ∋ (e1, e2) 7−→
(cid:0)(p(e1), p(e2)), e1e2(cid:1) ∈ m∗E satisfying the following properties:
(i) the induced map Eg × Eh −→ Egh is bilinear for all (g, h) ∈ G(2);
(ii) (associativity) (e1e2)e3 = e1(e2e3) whenever the multiplication makes sense; and
(iii) ke1e2k ≤ ke1kke2k, for every (e1, e2) ∈ E[2].
A ∗-involution on E is a continuous 2-periodic map ∗ : E ∋ e 7−→ e∗ ∈ E such that
(iv) p(e∗) = p(e)−1, and
(v) for all g ∈ G, the induced map ∗ : Eg −→ Eg−1 is conjugate linear.
Finally, we say that p : E −→ G is a Fell bundle if in addition the following conditions hold:
2e∗
1, ∀(e1, e2) ∈ E[2];
(vi) (e1e2)∗ = e∗
(vii) ke∗ek = kek2, ∀e ∈ E; in particular, Ex is a C∗-algebra, for x ∈ G(0);
(viii) e∗e ≥ 0, ∀e ∈ E; and
(ix) (fullness) the image of the map Eg × Eh −→ Egh spans a dense subspace of Egh, for all (g, h) ∈ G(2).
If we are given such a Fell bundle, we say that (G, E) is a Fell system.
Example 2.2. If A is a Dixmier-Douady bundle over G with action α, we get a Fell system (G, s∗A), where s∗A is the
C∗-bundle over G obtained by pulling back A through the source map s : G −→ G(0). The multiplication on s∗A is given by
As(g) × As(h) ∋ (a, b) 7−→ αh−1 (a)b ∈ As(gh) = As(h), and the ∗-involution is As(g) ∋ a 7−→ αg(a)∗ ∈ As(g−1)
= Ar(g).
/
/
/
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
3
Given a Fell system (G, E), we turn the space Cc(G; E) of compactly supported continuous sections of E into a
convolution algebra by setting
(ξ ∗ η)(g) :=ZGr(g)
ξ(γ)η(γ−1 g)dµr(g)
G (γ), and ξ∗(g) := ξ(g−1)∗, for ξ, η ∈ Cc(G; E), g ∈ G.
(1)
Let kξk1 := supx∈G(0)RGx kξ(g)kdµx
G(g). We next define the I-norm k · kI by kξkI := max{kξk1, kξ∗k1}. Then, the
completion L1(G; E) of Cc(G; E) with respect to k · kI is a Banach ∗-algebra. Its envelopping C∗-algebra C∗(G; E) is
called the full C∗-algebra of (G, E).
Note that we have a C∗-bundle over the base G(0), defined as the pull-back of E along the identity map G(0) ֒→ G;
we denote it by E(0). We can view E(0) as the restriction EG(0) , once we have identified G(0) with a subset of G.
Moreover, equipped with the pointwise norm, A := C0(G(0); E(0)) is a C∗-algebra. We will usually write Ax for the
C∗-algebra which is the fibre of E(0) over x ∈ G(0).
The following proposition is proved, for instance, in [8], (and in [5] in the case of proper groupoids), so we omit
the proof.
Proposition 2.3. Cc(G; E) is a pre-Hilbert (left) A-module under the operations
( f · ξ)(g) := f (r(g))ξ(g), for f ∈ A, ξ ∈ Cc(G; E), a ∈ G, and
Ahξ, ηi(x) :=ZGx
ξ(g)η(g)∗dµx
G(g), for ξ, η ∈ Cc(G; E), x ∈ G(0).
(2)
(3)
Let L2(G; E) be the Hilbert A-module obtained by completing Cc(G; E) with respect to the norm
kξk2 := kAhξ, ξik1/2, for ξ ∈ Cc(G; E).
Then, left multiplication by an element of Cc(G; E) (i.e. the map πl(ξ) : η 7−→ ξ ∗ η, ξ, η ∈ Cc(G; E)) is a bounded
A-linear operator with respect to the norm k · k2 which is adjointable (see [8]). Hence πl extends to a ∗-monorphism
The extension of πl to L1(G; E) is known as the left regular representation of L1(G; E).
πl : Cc(G; E) −→ L(L2(G; E)) := LA(L2(G; E)).
Definition 2.4. (cf. [21, A.3]). Under the above notations, the closure of the image πl(Cc(G; E)) in L(L2(G; E)) with respect
to the operator norm is called the reduced C∗-algebra of the Fell system (G, E), and is denoted by C∗
r(G; E); i.e.
r(G; E) := πl(Cc(G; E)) ⊂ L(L2(G; E))).
C∗
Remark 2.5. One can think of C∗
norm k · kr given by kξkr := sup{kπl(ξ)ηk2 η ∈ Cc(G; E), kηk2 ≤ 1}.
r(G; E) as the completion of the convolution ∗-algebra Cc(G; E) with respect to the reduced
r(G; s∗A) associated to the Fell bundle
Remark 2.6. If (A, G, α) is a Dixmier-Douady bundle, then the reduced C∗-algebra C∗
s∗A, denoted by A ⋊r G, is called the reduced crossed productof (A, G, α); it plays an important role in twisted K-theory
of groupoids ( [21], [20]).
Alternatively, we will sometimes use another definition of the reduced norm, which is a generalisation of that
of [18]. Suppose we are given a right Fell system (G, E). Then, for all x ∈ G(0), consider the inclusion ix : Gx −→ G.
Then, as in [21, A.3], we define the (right) Hilbert Ax-module L2(Gx; E) as the completion of Cc(Gx; i∗
xE) with respect
ξ(g)∗η(g)dµx(g) (the right action being (ξ · a) : Gx ∋ g 7−→ ξ(g) · a ∈ Eg). The
to the inner product hξ, ηiAx := RGx
following lemma is very easy to prove.
Lemma 2.7. Let (G, E) be as above. Then, for all x ∈ G(0), left multiplication by elements of Cc(G; E) gives a ∗-representation
πG
x : Cc(G; E) −→ LAx (L2(Gx; E)). Moreover, we have
kξkC∗(G;E) := kξkr = sup
x∈G(0)
{kπG
x (ξ)k, ∀ξ ∈ Cc(G; E)
4
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
3. Equivalence of Fell systems
In this section, we are presenting the notion of equivalences of Fell bundles over Morita equivalent groupoids.
Our definitions are slight modifications of those given by P. Muhly and D. Williams in [12, §.6].
Suppose that Z is a (right) principal G-space; that is, there is a principal G-action α : Z ∗ G −→ Z. If π : X −→ Z
is a Banach bundle, and if p : E −→ G is a Fell bundle, we set
X ∗ E := {(u, e) ∈ X × E (π(u), p(e)) ∈ Z ∗ G}.
Definition 3.1. A right Fell G-pair over the principal G-space Z is a pair (X, E) consisting of a Fell bundle E over G, a
Banach bundle π : X −→ Z, and a continuous map X ∗ E ∋ (u, e) 7−→ ue ∈ α∗X, such that
(i) (bilinearity) for all (z, g) ∈ Z ∗ G, the induced map Xz × Eg −→ Xzg is bilinear, and is compatible with the scalar
multiplication; i.e. (λu)e = u(λe) = λ(ue), ∀λ ∈ C, (u, e) ∈ Xz × Eg;
(ii) (associativity) if (z, g) ∈ Z ∗ G and (g, h) ∈ G(2), one has u(e1e2) = (ue1)e2, ∀(u, e1, e2) ∈ Xz × Eg × Eh;
(iii) kuek = kukkek, ∀(u, e) ∈ Xz × Eg;
(iv) (faithfullness) the induced map Xz × Xg −→ Xzg spans a dense subspace of Xzg.
We also say that (G, E) acts on X on the right over Z.
Likewise, one defines a left Fell G-pair (E, X) over a principal left G-space Z.
Remark 3.2. Notice that if (X, E) is a right Fell G-pair over Z, then for every z ∈ Z, Xz is a right Es(z)-module.
Now suppose that Γ ∼Z G. Then there are a continuous Γ-valued inner product Γ < ·, · >: Z ×G(0) Z−1 −→ Γ, and
a continuous G-valued inner product < ·, · >G: Z−1 ×Γ(0) Z −→ G, defined as follows
• for (z, ♭(z′)) ∈ Z ×Γ(0) Z−1, Γ < z, z′ > is the unique element of Γ such that z = Γ < z, z′ > ·z′;
• for (♭(z), z′) ∈ Z−1 ×Γ(0) Z, < z, z′ >G is the unique element of G such that z′ = z· < z, z′ >G.
Observe that these functions are well defined, for Z −→ Γ(0) is a G-principal bundle, and Z −→ G(0) is a Γ-principal
bundle. Furthermore, they satisfy the following equalities (cf. [13, §.6.1]):
Γ < z, z′ >−1 = Γ < z′, z >, ∀(z, ♭(z′)) ∈ Z ×G(0) Z−1,
< z, z′ >−1
G
=< z′, z >G, ∀(♭(z), z′) ∈ Z−1 ×Γ(0) Z, and
z· < z′, z" >G = Γ < z, z′ > ·z", ∀(z, ♭(z′), z") × Z ×G(0) Z−1 ×Γ(0) Z.
(4)
(5)
(6)
Lemma 3.3. Let Γ ∼Z G. Then, any right Fell G-pair (X, E) over Z gives rise to the left Fell G-pair (E, X) over the inverse
Z−1, where X is defined as the conjugate bundle of X. A similar statement holds for a left Γ-pair over Z.
Proof. By definition X is X as space. If ♭ : X −→ X denotes the identity map, we define the projection ¯π : X −→ Z−1
by ¯π(♭(u)) := ♭(π(u)). The fibre X♭(z) is the conjugate Banach space of Xz; the left G-action on X is g · ♭(u) := ♭(u · g−1),
while the left action of E on X is given by Eg × X♭(z) ∋ (e, ♭(u)) 7−→ ♭(u · e∗) ∈ Xg·♭(u).
(cid:3)
Let us fix some notations that will be used in the sequel. Suppose Γ ∼Z G. If (X, E) is a Fell G-pair, we define
the topological spaces X ∗ X := {(u, ♭(u′)) ∈ X × X (π(u), ¯π(♭(u′))) ∈ Z ×G(0) Z−1} and X ∗ X := {(♭(u), u′) ∈ X × X
( ¯π(♭(u)), π(u′)) ∈ Z−1 ×Γ(0) Z}. Observe that the space Z ×G(0) Z−1 is a locally compact groupoid with base Z as
follows: the product is (z, ♭(z′)) · (z′, ♭(z")) := (z, ♭(z")), the source of (z, ♭(z′)) is z′, its range is z, and its inverse is
(z′, ♭(z)). Similarly Z−1 ×Γ(0) Z is a locally compact groupoid with base Z−1.
If Γ ∼Z G, and if (G, E) and (Γ, F) are Fell systems, we denote by E>G and FΓ< the Fell bundles over Z−1 ×Γ(0) Z and
Z ×G(0) Z−1, respectively, obtained by pulling back E −→ G along the continuous map < ·, · >G, and F −→ Γ along
the continuous map Γ < ·, · >, respectively. Note that, for instance, the fiber of E>G over (♭(z), z′) is isomorphic to
E<z,z′ >G .
Definition 3.4. Assume Γ ∼Z G and (X, E) is a Fell G-pair over Z. An E-valued inner product on X is a continuous map
h·, ·iE : X ∗ X −→ E>G , (♭(u), u′) 7−→ hu, u′iE, such that
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
5
(i) for (♭(z), z′) ∈ Z−1 ×Γ(0) Z, the induced map h·, ·iE : X♭(z) × Xz′ −→ E<z,z′ >G is linear in both the first and the second
variable;
(ii) (E-linearity) if (♭(z), z′) ∈ Z−1 ×Γ(0) Z and (z, g) ∈ Z∗G, then hu, u′iE ·e = hu, u′·eiE, ∀(♭(u), u′, e) ∈ X♭(z) ×Xz′ ×Eg;
(iii) hu, u′i∗
(iv) (positivity) for all z ∈ Z and u ∈ Xz, hu, uiE ≥ 0 in E<z,z>G
= Es(z); and the equality hu, uiE = 0 implies u = 0.
= hu′, uiE ∈ E<z,z′ >−1
= E<z′,z>G ;
E
G
In this case, we say that X is a right (G, E)-inner product module over Z.
Likewise, if (F, X) is a left Fell Γ-pair, one defines an F-valued inner product Fh·, ·i : X ∗ X −→ FΓ<, all the actions being
considered on the left.
Remark 3.5. Observe that conditions (ii) and (iii) of the definition imply that hu · e, u′iE = e∗ · hu, u′iE, whenever the
multiplications and the inner product are defined. Moreover, for all z ∈ Z, Xz is a pre-Hilbert As(z)-module.
Definition 3.6. An equivalence between two Fell systems (Γ, F) and (G, E) is a pair (Z, X) such that Γ ∼Z G, X is a left
(F, Γ)-inner product module and a right (G, E)-inner product module over Z, with the following properties
(i) (equivariance) for all (γ, z, g) ∈ Γ ∗ Z ∗ G, the multiplication Fγ × Xz × Eg −→ Xγzg is associative; i.e. f · (u · e) =
( f · u) · e, ∀( f, u, e) ∈ Fγ × Xz × Eg;
(ii) (compatibility) for all (z, ♭(z′), z") ∈ Z ×G(0) Z−1 ×Γ(0) Z and (u, ♭(u′), u") ∈ Xz × X♭(z′) × Xz", Fhu, u′i · u" =
u · hu′, u"iE in Xz·<z′,z">G
= XΓ<z,z′>·z";
(iii) the F-valued inner product is full; i.e., the image of the induced map Xz × X♭(z′) −→ FΓ<z,z′ > spans a dense subspace
of FΓ<z,z′ >;
(iv) the E-valued inner product is full.
In this case, we write (Γ, F) ∼(Z,X) (G, E).
It follows from Definition 3.6 and Lemma 3.3 that if (Γ, F) ∼(Z,X) (G, E), then (G, E) ∼(Z−1,X) (Γ, F).
Remark 3.7.
Furthermore, it is starightforward that for all z ∈ Z, (the completion with respect to the inner products of) Xz is an
imprimitivity Br(z)-As(z)-bimodule.
Example 3.8.
we identify G).
If (G, E) is a Fell system, then (G, E) ∼(ZG,E) (G, E), where ZG is the space of morphisms G(1) (with which
G, the generalized source and range maps being
Indeed, ZG implements a Morita equivalence G ∼ZG
/ G(0) , together with the canonical left and right actions given by partial
r
the source and range maps s and r of G
s
G
= {g−1 g ∈ G}. It is easy to see that the inner products G < ·, · >: ZG ×G(0) Z−1
multiplications. Notice that Z−1
< ·, · >G: Z−1
of a Fell bundle. Now, the conjugate bundle E −→ Z−1
Eg × Eh−1 ∋ (e1, e∗
2) 7−→ e1e∗
2 ∈ Egh−1 , and Eg−1 × Eh ∋ (e∗
of Definition 3.6 are satisfied.
G −→ G and
G ×G(0) ZG −→ G are G < g, h >= gh−1 and < g, h >G= g−1h, respectively. E acts on itself over ZG by definition
G is given fibrewise by Eg−1 = {e∗ e ∈ Eg}. The inner products are
1e2 ∈ Eg−1 h. It is straightforward that all the conditions
1, e2) 7−→ e∗
By virtue of Remark 3.7 and Example 3.8, equivalence of Fell systems is symmetric and reflexive. Also, it is
not hard to show that it is transitive, so that it defines an equivalence relation among the collection of Fell systems
(cf. [8]).
In the sequel, we will need the following result.
/
/
/
6
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
Proposition 3.9.
inductive limit topologies 1 under the following operations:
If (Γ, F) ∼(Z,X) (G, E), Cc(Z; X) is a full pre-inner product Cc(Γ; F)-Cc(G; E)-bimodule with respect to the
(ξ · φ)(z) :=ZΓr(z)
(φ · η)(z) :=ZGs(z)
ξ(γ)φ(γ−1 · z)dµr(z)
Γ (γ),
φ(z · g)η(g−1)dµs(z)
G (g),
Cc(Γ;F)hφ, ψi(γ) :=ZGs(z)
FDφ(z · g), ψ(γ−1 · z · g)E dµs(z)
hφ, ψiCc(G;E)(g) :=ZΓr(z)Dφ(γ−1 · z), ψ(γ−1 · z · g)EE
G (g), where r(z) = r(γ), and
dµr(z)
Γ (γ), where s(z) = r(g),
Proof. See [12] or [8].
We will adopt the following notations.
(7)
(8)
(9)
(10)
(cid:3)
Notations 3.10. 1. For the sake of simplicity, we will sometimes write
2. As in [18], if ξ ∈ Cc(G; E), η ∈ Cc(Γ; F), and φ, ψ ∈ Cc(Z−1; X), we will write ξ : φ and φ : η for the left and right actions of
Cc(G; E) and Cc(Γ; F) on Cc(Z−1; X), respectively, and we will write
for hφ, ψiCc(Γ;F).
h· , ·i for Cc(Γ;F)h·, ·i and h· , ·i
⋆
⋆
for h·, ·iCc(G;E).
hhφ , ψii for Cc(G;E)hφ, ψi and hhφ , ψii
⋆
⋆
Proposition 6.10] that guarantees the existence of a net { fλ}λ∈Λ in Cc(Γ; F) of the form fλ = Pnλ
Remark 3.11. We should note that the proof of Proposition 3.9 is mostly based on the crucial result proved in [12,
i i, with each
φλ
i ∈ Cc(Z; X), which is an approximate identity with respect to the inductive limit topology for both the left action of
Cc(Γ; F) on itself and on Cc(Z; X). By symmetry, a similar statement holds for (G, E). In particular, by Example 3.8, the
same result shows that for any Fell system (G, E), Cc(G; E) admits an approximate identity for the inductive limit topology.
hφλ
i
, φλ
i=1⋆
4. The linking Fell system
In this section, we use some constructions from [13, Chapter 6] and [10, §.2]. If Γ ∼Z G, then form the Linking
/ M(0) by setting: M := Γ ⊔ Z ⊔ Z−1 ⊔ G, and M(0) := Γ(0) ⊔ G(0), the source and range maps sM
groupoid M
and rM being the obvious ones. The partial multiplication of M is given by
M(2) −→ M,
∈ Γ(2) :
∈ Γ ∗ Z :
(γ1, γ2)
(γ, z)
(z, ♭(z′)) ∈ Z ×G(0) Z−1 :
(z, g)
(♭(z), z′) ∈ Z−1 ×Γ(0) Z :
(♭(z), γ)
(g, ♭(z))
(g1, g2)
∈ Z−1 ∗ Γ :
∈ G ∗ Z−1 :
∈ G(2) :
∈ Z ∗ G :
∈ Γ
∈ Z
∈ Γ
∈ Z
∈ G
γ1γ2
γ.z
z.♭(z′) := Γhz, z′i
z.g
♭(z).z′ := hz, z′iG
♭(z).γ := ♭(γ−1.z) ∈ Z−1
∈ Z−1
g.♭(z) := ♭(zg−1)
g1 g2
∈ G
,
1Since we are dealing with Banach ∗-algebras, the only properties we take into account here are the continuity of the actions and the pre-inner
products with respect to the inductive limit topologies, the compatibility between the actions and the pre-inner products, and the fullness of the
latters.
/
/
/
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
7
so that M(2) = Γ(2) ⊔ Γ ∗ Z ⊔ Z ×G(0) Z−1 ⊔ Z ∗ G ⊔ Z−1 ∗ Γ ⊔ Z−1 ×Γ(0) Z ⊔ G ∗ Z−1 ⊔ G(2). Finally, the inversion in M is
defined by
M −→ M,
Γ
Z
Z−1
G
∋ γ
∋ z
∋ ♭(z)
∋ g
∈ Γ
7−→ γ−1
7−→ ♭(z) ∈ Z−1
7−→ z
7−→ g−1
∈ Z
∈ G
.
With these structures, M
( [13, Proposition 6.2.2]).
Now, let µΓ and µG be left Haar systems on Γ and G, respectively. Then, if Γ ∼Z G, there exists a full r-system 2
µZ = {µy
is a locally compact Hausdorff groupoid with open source and range maps
Z}y∈Γ(0) of Radon measures on Z determined by
/ M(0)
Z(φ) :=ZGs(z)
µy
φ(z · g)dµs(z)
G (g),
(11)
for all y ∈ Γ(0) and φ ∈ Cc(Z), where z is some arbitrary element of the fibre Zy = r−1(y). Furthermore, µZ is a
Z (z) =
Z (z) (see [13, §.6.4], [18]). Similarly, considering the inverse Z−1 : G −→ Γ, the Haar system µΓ
= (r♭)−1(x) =
left Haar system on Z for the left action of Γ; that is, for all γ ∈ Γ and φ ∈ Cc(Z), we have RZr(γ)
RZs(γ)
Z−1 }x∈G(0) on Z−1 for left action of G. Note that we have suppµx
φ(γ.z)dµs(γ)
φ(z)dµr(γ)
Z−1
induces a left Haar system µZ−1 = {µx
Z−1
x , and that for φ ∈ Cc(Z−1) and ♭(z) ∈ Z−1
x , we have
Z−1 (φ) :=ZΓs♭(♭(z)) =Γr(z)
µx
φ(♭(γ−1z))dµr(z)
Γ (γ).
(12)
Moreover, µΓ, µG, µZ, and µZ−1 induces a left Haar system µM on M as it is shown in the following proposition.
Proposition 4.1. Assume Γ ∼Z G, and µΓ and µG are left Haar systems on Γ and G, respectively. Then, under the above
constructions, there is a left Haar system µM = {µω
M}ω∈M(0) on the linking groupoid M determined by
µω
M(F) :=
for all ω ∈ M(0) and F ∈ Cc(M).
µω
Γ (FΓ) + µω
Z(FZ),
µω
Z−1 (FZ−1) + µω
G(FG),
if ω ∈ Γ(0), and
if ω ∈ G(0),
Proof. See [13, Proposition 6.4.5], or [18, Lemma 4].
(13)
(cid:3)
Proposition 4.2. Suppose (Γ, F) ∼(Z,X) (G, E). Then we define a Banach bundle L over the linking groupoid M, where L
is the topological space L := F ⊔ X ⊔ X ⊔ E, the projection is given by
pL : L −→ M,
F ∋ f
X ∋ u
X ∋ ♭(v)
E ∋ e
2See for instance [17] for the definition.
∈ Γ
∈ Z
7−→ pF( f )
7−→ π(u)
7−→ ♭(π(v)) ∈ Z−1
7−→ pE(e)
∈ G
.
(14)
/
/
/
8
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
Moreover, pL : L −→ M is a Fell bundle with respect to the multiplication L[2] −→ m∗L and involution (∗) : L −→ L
respectively given by
Fγ1 × Fγ2
Fγ × Xz
Xz1 × X♭(z2)
Xz × Eg
X♭(z) × Fγ
X♭(z1) × Xz2
Eg × X♭(z)
Eg × Eh
∋ ( f1, f2)
∋ ( f, u)
∋ (u, ♭(v))
∋ (u, e)
∋ (♭(u), f )
∋ (♭(u), v)
∋ (e, ♭(u))
∋ (e1, e2)
and
∈ Fγ1γ2 , for (γ1, g2) ∈ Γ(2)
∈ Xγ·z, for (γ, z) ∈ Γ ∗ Z
∈ FΓ<z1,z2>, for (z1, ♭(z2)) ∈ Z ×G(0) Z−1
∈ Xzg, for (z, g) ∈ Z ∗ G
7−→ f1 f2
7−→ f · u
7−→ Fhu, vi
7−→ u · e
7−→ ♭( f ∗ · u) ∈ X♭(γ−1z), for (♭(z), γ) × Z−1 ∗ Γ
7−→ hu, viE
7−→ ♭(u · e∗)
7−→ e1e2
∈ E<z1,z2>G , for (♭(z1), z2) ∈ Z−1 ×Γ(0) Z
∈ X♭(zg−1), for (g, ♭(z)) ∈ G ∗ Z−1
∈ Egh, for (g, h) ∈ G(2)
(∗) : L → L,
Fγ
Xz
X♭(z)
Eg
∋ f
∋ u
∋ ♭(v)
∋ e
7−→ f ∗
∈ Fγ−1 , for γ ∈ Γ
7−→ ♭(u) ∈ X♭(z), for z ∈ Z
7−→ v
7−→ e∗
∈ Xz, for ♭(z) ∈ Z−1
∈ Eg−1 , for g ∈ G
L is called the linking Fell bundle, and (M, L) is the linking Fell system.
.
(15)
(16)
Proof. It is clear that pL : L −→ M is a Banach bundle. Next, observe that all of the conditions of Definition 2.1
are verified by the operations (15) and (16) by merely applying Definition 3.6 to the equivalences (Z, X) and
(Z−1, X).
(cid:3)
At this point, we can do integration on M with values on the the linking Fell bundle L. We then can form
the convolution ∗-algebra Cc(M; L). Note that we have an isomorphism of convolution ∗-algebras Cc(M; L) (cid:27)
Cc(Γ; F) ⊕ Cc(Z; X) ⊕ Cc(Z−1; X) ⊕ Cc(G; E); so that an element ξ ∈ Cc(M; L) can be written as a matrix
where ξ11 := ξΓ ∈ Cc(Γ; F), ξ12 := ξZ ∈ Cc(Z; X), ξ21 := ξZ−1 ∈ Cc(Z−1; X), and ξ22 := ξG ∈ Cc(G; E). With respect to
this decomposition, the involution in Cc(M; L) is given by
ξ11
ξ21
ξ12
ξ22 ,
ξ =
=
ξ∗ =
ξ∗
11
ξ∗
12
ξ∗
21
ξ∗
22
ξ∗
11
♭ ◦ ξ12 ◦ ♭
♭ ◦ ξ21 ◦ ♭
ξ∗
22
,
11 and ξ∗
where ξ∗
Furthermore, routine calculations (cf. [8]) show that the convolution in Cc(M; L) is given by
22 are the images of ξ11 and ξ22 under the standard involutions in Cc(Γ; F) and Cc(G; E), respectively.
ξ11 ∗ η11 +
hξ12 , η∗
⋆
21i
ξ11 · η12 + ξ12 · η22
(η∗
11 · ξ∗
21)∗ + (η∗
21 · ξ∗
22)∗
hξ∗
21 , η12i
⋆
+ ξ22 ∗ η22
.
(17)
η11
η12
η21
ξ22
ξ21
ξ12
ξ11
∗
=
(x) = RΓr(z) hφ(γ−1 · z), ψ(γ−1 · z)iEdµr(z)
η22
Suppose (Γ, F) ∼(Z,X) (G, E). For x ∈ G(0), we also denote by X −→ Zx the pull-back of X −→ Z along the
inclusion Zx ֒→ Z. Then, L2(Zx; X) is the completion of Cc(Zx; X) with respect to the Ax-valued inner product
Γ (γ), where s(z) = x, and the right Ax-action (φ · a)(z) := φ(z)a, for
hφ , ψi
⋆
φ ∈ Cc(Zx; X), a ∈ Ax. Thus, L2(Zx; X) is a Hilbert Ax-module. Similarly, for all y ∈ Γ(0), one can form the Hilbert
By-module L2(Z−1
The following proposition will be crucial in the proof of the equivalence theorem (Theorem 5.5).
y ; X).
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
9
Proposition 4.3.
representation RΓ
get a representation RG
y : C∗
r(G; E) −→ LBy (L2(Z−1
y ; X)).
Suppose (Γ, F) ∼(Z,X) (G, E). For x ∈ G(0), the left action of Cc(Γ; F) on Cc(Zx; X) induces a ∗-
r(Γ; F). Similarly, for all y ∈ Γ(0), we
x : Cc(Γ; F) −→ LAx (L2(Zx; X)) that factors through the C∗-algebra C∗
Proof. Let ξ ∈ Cc(Γ; F); then for φ, ψ ∈ Cc(Zx; X), simple calculations give hξ · φ , ψi
(x). It follows
⋆
that the Ax-linear operator Cc(Zx; X) ∋ φ 7−→ ξ · φ ∈ Cc(Zx; X) is adjointable, and then bounded with respect to the
norm k · kL2(Zx;X), which gives the ∗-representation RΓ
Now, let z0 ∈ Zx, and let y := r(z0). Then, to complete the proof it suffices to check that for all ξ ∈ Cc(Γ; F),
kRΓ
Consider the (left) Hilbert By-module Xz0, and form the interior tensor product L2(Γy; F) ⊗By Xz0 which is a right
Hilbert Ax-module under the operations defined on simple tensors by: (ξ ⊗ u) · a := ξ ⊗ (ua), and hξ ⊗ u, η ⊗ vi :=
hu, hξ, ηiBy · viAx . Then, the map
y : Cc(Γ; F) −→ LBy (L2(Γy; F)) is the representation defined in Lemma 2.7.
x : Cc(Γ; F) −→ LAx (L2(Zx; X)), ξ −→ (RΓ
x(ξ) : φ 7−→ ξ · φ).
y(ξ)k, where πΓ
(x) = hφ , ξ∗ · ψi
⋆
x(ξ)k ≤ kπΓ
uz0 : L2(Γy; F) ⊗By Xz0 −→ L2(Zx; X), Xi
ξi ⊗ ui 7−→Xi
ξi · ui,
(18)
where for ξ ∈ Cc(Γy; F) and u ∈ Xz0 , (ξ · u)(z) := ξ(Γ < z, z0 >) · u ∈ XΓ<z,z0>·z0 , is an isomorphism of Hilbert
Ax-modules. The map (18) is clearly Ax-linear and injective. To see that it is surjective, first notice that the well
defined map Zx ∋ z 7−→ Γ < z, z0 >∈ Γy, is a homeomorphism of Γ-spaces (its inverse being Γy ∋ γ 7−→ γ · z0 ∈ Zx).
Next, for all z ∈ Zx, the linear span of the image of FΓ<z,z0> × Xz0 ∋ ( f, u) 7−→ f · u ∈ XΓ<z,z0>·z0 is dense in XΓ<z,z0>·z0
by definition of a Fell pair; so that, using the Weierstrass theorem,
spannη · u : Zx ∋ z 7−→ η(Γ < z, z0 >) · u ∈ XΓ<z,z0>·z0 η ∈ Cc(Γy; F), u ∈ Xz0o
is dense in Cc(Zx; X) in the inductive limit topology. It follows that any φ ∈ Cc(Zx; X) is the inductive limit of some
Pi ηi · ui = uz0 (Pi ηi ⊗ ui). We then have an isomorphism of C∗-algebras
such that uz0 (T) (Pi ξi · ui) := uz0 (T (Pi ξi ⊗ ui)), for all T ∈ LAx (L2(Γy; F) ⊗By Xz0 ). Furthemore, the following
uz0 : LAx (L2(Γy; F) ⊗By Xz0) −→ LAx (L2(Zx; X))
diagram is commutative
Cc(Γ; F)
πΓ
y
RΓ
x
/ LAx (L2(Zx; X))
uz0
LBy (L2(Γy; F))
/ LAx (L2(Γy; F) ⊗By Xz0 )
where the lower horizontal arrow is the map T 7−→ T ⊗ id (cf. for instance [7, p.50]). Indeed, let ξ ∈ Cc(Γ; F), and
φ ∈ Cc(Zx; X). Without loss of generality, we can suppose that φ = η · u; then,
uz0 (πΓ
y(ξ) ⊗ id)φ = (πΓ
y(ξ) ⊗ id)(η ⊗ u) = (πΓ
y(ξ)η) ⊗ u = (ξ ∗ η) · u = ξ · (η · u) = RΓ
x(ξ)(η ⊗ u) = RΓ
x(ξ)φ,
which completes the proof since uz0 is an isomorphism and kπΓ
y(ξ) ⊗ idk ≤ kπΓ
y(ξ)k (see [7, p.50]).
(cid:3)
5. The equivalence theorem for reduced C∗-algebras of Fell systems
We start this section by the following observations. Let (G, E) be a Fell system and let A := C0(G(0); E(0)) be as
usual. Suppose we are given a bounded continuous section f ∈ Cb(G(0); E(0)). Then, for ξ ∈ Cc(G; E), we define an
element L f ξ =: f ξ ∈ Cc(G; E) by setting:
L f ξ(g) := f (r(g))ξ(g) ∈ Eg, for all g ∈ G.
Also, we define an element ξ f ∈ Cc(G; E) by
G ∋ g 7−→ ξ f (g) := ξ(g) f (s(g)) ∈ Eg.
(19)
(20)
Notice that Cb(G(0); EG(0) ) is a C∗-algebra under pointwise operations and the supremum norm ( [1, Lemma 3.2]).
/
/
O
O
10
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
Lemma 5.1. For all f ∈ Cb(G(0); E(0)), we have L f ∈ L(L2(G; E)), where L f is the element defined by (19). Moreover, the
map L : Cb(G(0); E(0)) ∋ f 7−→ L f ∈ LA(L2(G; E)) is a ∗-homomorphism.
Proof. L f is clearly continuous; also it is bounded since f is a bounded section (it is straightforward that kL f kop ≤
k f k, where k · kop is the operator norm in L(L2(G; E))). If ξ, η ∈ Cc(G; E) and x ∈ G(0), then
AhL f ξ, ηi(x) =ZGx
=ZGx
L f ξ(g−1)∗η(g−1)dµx
G(g)
ξ∗(g) f (r(g−1))∗η(g−1)dµx
G(g)
hence, L f is adjointable with adjoint L∗
f
L f1 L f2 , ∀ f1, f2 ∈ Cb(G(0); E(0)).
:= L f ∗ . Moreover, L f1 f2 (ξ) = L f1 (L f2 (ξ)), ∀ξ ∈ Cc(G; E); thus L f1 f2
=
(cid:3)
= Ahξ, L f ∗ηi(x);
Proposition 5.2. Let (G, E) be as above. Then, Cb(G(0); E(0)) is a C∗-subalgebra of M(C∗
r(G; E)).
Proof. If πl(ξ) ∈ C∗
r(G; E) and f ∈ Cb(G(0); E(0)), we put
L f (πlξ) := πl(L f ξ) = πl( f ξ), and R f (πlξ) := πl(ξ f ).
(21)
We verify that with these formulas, we obtain a double centralizer (L f , R f ) ∈ M(C∗
for ξ, η ∈ Cc(G; E) and g ∈ G, one has
r(G; E)). To see this, observe that
( f (ξ ∗ η))(g) = f (r(g))ZGr(g)
ξ(h)η(h−1 g)dµr(g)
G (h)
=ZGr(g)
=ZGr(g)
= f ξ ∗ η;
f (r(h))ξ(h)η(h−1 g)dµr(g)
G (h)
( f ξ)(h)η(h−1 g)dµr(g)
G (h)
and similarly one shows that (ξ ∗ η) f = ξ ∗ η f . Moreover, we have
(ξ f ∗ η)(g) =ZGr(g)
=ZGr(g)
=ZGr(g)
ξ(h) f (s(h))η(h−1 g)dµr(g)
G (h)
ξ(h) f (r(h−1 g))η(h−1 g)dµr(g)
G (h)
ξ(h)( f η)(h−1 g)dµr(g)
G (h)
so that R f (πl(ξ))πl(η) = πl(ξ)L f (πl(η)), and by continuity, for every f ∈ Cb(G(0); E(0)), the pair (L f , R f ) verifies
R f (a)b = aL f (b) for all a, b ∈ C∗
r(G; E); i.e. (L f , R f ) ∈ M(C∗
r(G; E)).
(cid:3)
= (ξ ∗ f η)(g);
In what follows, we identify the double centralizer (L f , R f ), and hence the element f ∈ Cb(G(0); E(0)), with
r(G; E) under the formulas: L f πl(ξ) := πl(L f ξ) = πl( f ξ), and
L f ∈ L(L2(G; E)), by considering L f as a multiplier of C∗
πl(ξ)L f := R f (πl(ξ)) = πl(ξ f ).
Now, consider the field of C∗-algebras
M(E) := ax∈G(0)
M(Ex)
over G(0). Then, denote by Cstr
b (G(0); M(E)) the unital C∗-algebra (under pointwise operations and the supremum
norm) consisting of all the bounded strictly continuous sections of M(E) over G(0) (see [1, p.7] for details). Note that
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
11
the unit 1 ∈ Cstr
identity map. From Proposition 5.2 we obtain the following corollary.
b (G(0); M(E)) is the section given by 1 : G(0) ∋ x 7−→ (idEx , idEx ) ∈ M(Ex), where idEx : Ex −→ Ex is the
Corollary 5.3. Let (G, E) be as above. Then Cstr
b (G(0); M(E)) is a unital C∗-subalgebra of M(C∗
r(G; E)).
7−→ L f ∈ M(C∗
Proof. The map Cb(G(0); EG(0) ) ∋ f
r(G; E)) is non-degenerate; indeed, by considering the left Fell
G-pair (E, E) determined by the full maps Eg × Eh −→ Egh, we see that for f ∈ Cb(G(0); E(0)) ⊂ C0(G(0); E(0)) and
ξ ∈ Cc(G; E), the element L f ξ ∈ Cc(G; E) is nothing but the canonical action of C0(G(0); E(0)) on Cc(G; E) defined by
the formula ( f · ξ)(g) := f (r(g))ξ(g). It follows that if {ai}i∈I is an approximate identity of Cb(G(0); E(0)), then thanks
to [11, Lemma 6.1], for all ξ ∈ Cc(G; E), ai · ξ −→ ξ in Cc(G; E) with respect to the inductive limit topology. Thus
Lai πl(ξ) = πl(ai·ξ) −→ πl(ξ) in C∗
r(G; E). Now, from [14, §.3.12.10
and §.3.12.12], the map L extends to a unital strictly continuous ∗-homomorphism M(Cb(G(0); E(0))) −→ M(C∗
r(G; E));
this map is again denoted by L. Furthermore, from [1, Lemma 3.1], we have that M(C0(G(0); E(0))) = Cstr
b (G(0); M(E)),
which settles the result.
r(G; E). Whence, L(Cb(G(0); E(0)))C∗
r(G; E) is dense in C∗
(cid:3)
Proposition 5.4. Suppose (Γ, F) ∼(Z,X) (G, E). Let χ
Then we get two elements χ
Now define
G(0) 1 of Cstr
Γ(0) 1 and χ
Γ(0) and χ
b (M(0); M(L)), where 1 ∈ Cstr
G(0) be the characteristic functions of Γ(0) and G(0) respectively.
b (M(0); M(L)), defined by scalar multiplication.
Then pΓ and pG are complementary full projections 3 in M(C∗
r(M; L)).
pΓ := Lχ
Γ(0)
1, and pG := Lχ
G(0)
1 ∈ M(C∗
r(M; L)).
Proof. It is straightforward that χ
images pΓ and pG are complementary projections of M(C∗
Γ(0) 1 and χ
G(0) 1 are complementary projections of Cstr
b (M(0); M(E)). Hence, their
r(M; L)), by virtue of Corollary 5.3.
Now, let ξ, η ∈ Cc(M; L). Then
πM
l (ξ)pΓπM
l (η) = πM
l (ξ ∗ pΓη) = πM
l (ξpΓ ∗ η)
= πM
l
ξ11 ∗ η11
ξ11 · η12
ξ21 : η11
hξ∗
21 , η12i
⋆
.
So, to check that pΓ is full, we just have to show that
span
πM
l
ξ11 ∗ η11
ξ21 : η11
ξ11 · η12
hξ∗
21 , η12i
⋆
ξ11 ∈ Cc(Γ; F), ξ21 ∈ Cc(Z−1; X),
η12 ∈ Cc(Z; X), η11 ∈ Cc(Γ; E)(cid:9)
(22)
is dense in C∗
r(M; L). But this is not hard to verify, by using the previous results. Indeed, the existence of an
approximate identity in Cc(Γ; F) for both the left actions of Cc(Γ; F) on itself and on Cc(Z; X) shows that elements
of the form ξ11 ∗ η11, for ξ11, η11 ∈ Cc(Γ; F) span a dense subspace of Cc(Γ; F) and that elements of the form
ξ11 · η12, for η12 ∈ Cc(Z; X), span a dense subspace of Cc(Z; X). Also, that elements of the form ξ21 : η11, where
ξ21 ∈ Cc(Z−1; X), η11 ∈ Cc(Γ; F), span a dense subspace of Cc(Z−1; X) follows from the existence of an approximate
identity in Cc(Γ; F) for the right action of Cc(Γ; F) on Cc(Z−1; X) (cf. Remark 3.11). Finally, since Cc(Z; X) is a
full pre-inner product Cc(Γ; F)-Cc(G; E)-bimodule (Proposition 3.9), the image of h· , ·i
is a dense subspace of
⋆
Cc(G; E). We then have shown that C∗
r(M; L). In a similar fashion, we get that
C∗
r(M; L) is dense in C∗
r(M; L), which completes the proof.
r(M; L) is dense in C∗
r(M; L)pGC∗
r(M; L)pΓC∗
(cid:3)
3Recall from [2] that a projection p ∈ M(A) is said to be full if pAp is not contained in any proper closed two-sided ideal of A; that is, span{ApA}
is dense in A (see for instance [3] or [15, p.50]). In this case, we say that pAp is a full corner of A. Two projections p, q ∈ M(A) are complementary if
p + q = 1, in which case pAq is a pAp-qAq-imprimitivity bimodule; i.e. pAp and qAq are Morita equivalent. Conversely, two C∗-algebras A and B
are Morita equivalent if and only if there is a C∗-algebra C with complementary full corners isomorphic to A and B, respectively (cf. [3, Theorem
1.1], [15, Theorem 3.19]).
12
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
Theorem 5.5. Let Γ and G be locally compact Hausdorff groupoids. Suppose (Γ, F) ∼(Z,X) (G, E). Then the isomorphisms
of convolution ∗-algebras
and
Cc(Γ; F) ∋ ξ11 7−→
Cc(G; E) ∋ η22 7−→
ξ11
0
0
0
0
0 ∈ pΓCc(M; L)pΓ,
η22 ∈ pGCc(M; L)pG
0
extend to two isomorphisms of C∗-algebras
In particular, C∗
isomorphic to the completion Xr of Cc(Z; X) in the norm
r(Γ; F) and C∗
C∗
r(Γ; F) −→ pΓC∗
r(G; E) are Morita equivalent with imprimitivity bimodule pΓC∗
r(M; L)pΓ, and C∗
r(G; E) −→ pGC∗
r(M; L)pG.
r(M; L)pG which is isometrically
kφkE := khφ , φi
⋆
k1/2
r(G;E), for φ ∈ Cc(Z; X).
C∗
Proof. That the maps defined by (23) and (24) are isomorphisms of convolutions ∗-algebras is obvious.
As previously, let us put B := C0(Γ(0); F(0)) and A := C0(G(0); E(0)). Then
C0(M(0); L(0)) (cid:27) B ⊕ A,
as C∗-algebras. Now, with respect to this decomposition, simple calculations show that
for all ξ =
ξ11
ξ21
pΓCc(M; L)pΓ, then
so that
ξ12
η11
η21
B⊕Ahξ, ηi =(cid:18)Bhξ11, η11i +
ξ22 and η =
B⊕A*
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ξ11
0
ξ11
0
hξ∗
⋆
21 , η∗
21iΓ(0)(cid:19) ⊕(cid:18)hξ12 , η12i
⋆
G(0) + Ahξ22, η22i(cid:19) ,
0
0
η12
ξ11
0
η22 in Cc(M; L).
In particular, suppose that ξ =
0
0 ,
+ = Bhξ11, ξ11i ⊕ 0,
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)L2(M;L)
0
= kξ11kL2(Γ;F);
0
ξ11
0
(26)
0
0 ∈
(27)
thus, (23) extends to an isometric B-linear map uΓ of B-modules
uΓ : L2(Γ; F) −→ pΓL2(M; L)pΓ,
where pΓL2(M; L)pΓ is the completion of pΓCc(M; L)pΓ with respect to the norm of L2(M; L). Similarly, for
ξ22 ∈ Cc(G; E), we get
(23)
(24)
(25)
(28)
and hence
0
0
0
0
B⊕A*
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
0
0
0
0
ξ22 ,
ξ22
+ = 0 ⊕ Ahξ22, ξ22i,
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)L2(M;L)
ξ22
= kξ22kL2(G;E);
0
so that (24) extends to an isometric A-linear map uG of A-modules
uG : L2(G; E) −→ pGL2(M; L)pG.
Furthermore, since uΓ and uG are surjective, then from [7, Theorem 3.5], they are unitaries in LB(L2(Γ; F), pΓL2(M; L)pΓ)
and LA(L2(G; E), pGL2(M; L)pG), respectively; in other words,
L2(Γ; F) ≈ pΓL2(M; L)pΓ
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
13
as Hilbert B-modules, and
L2(G; E) ≈ pGL2(M; L)pG
as Hilbert A-modules, here the sign "≈" stands for unitarily equivalent. Moreover, it is very easy to see that the
following diagrams commute:
Cc(Γ; F)
(cid:27)
/ pΓCc(M; L)pΓ
Cc(G; E)
(cid:27)
/ pGCC(M; L)pG
(29)
πΓ
l
πM
l
πG
l
πM
l
(cid:27) /
L(cid:0)L2(Γ; F)(cid:1)
/ L(cid:0)pΓL2(M; L)pΓ(cid:1)
It then only remains to check that for ξ =
ξ11
0
0
L(cid:0)L2(G; E)(cid:1)
0 and η =
0
0
(cid:27) /
0
/ L(cid:0)pGL2(M; L)pG(cid:1)
η22 , we have kξkC∗
r(M;L) = kη22kC∗
r(Γ;F)
r(G;E) which will lead to the desired isomorphisms of C∗-algebras (25) since pΓ and pG are
and kηkC∗
complementary (cf. Proposition 5.4). However, by symmetry it suffices to check one of the latter equalities. To
this end, we will use the constructions of Lemma 2.7.
Note that we have
r (M;L) = kξ11kC∗
Cc(Mω; L) =
Cc(Γy; F) ⊕ Cc(Z−1
y ; X),
Cc(Zx; X) ⊕ Cc(Gx; E),
if ω = y ∈ Γ(0);
if ω = x ∈ G(0)
In other words, elements of Cc(My; L), for y ∈ Γ(0) , are of the form
y ; X), while elements of Cc(Mx; L), for x ∈ G(0), are of the form
Cc(Gx; E). Then, for all y ∈ Γ(0), and η, ζ ∈ Cc(My; L), one has
Cc(Z−1
η11
η21
0
0
0
0 with η11 ∈ Cc(Γy; F) and η21 ∈
η22 with η12 ∈ Cc(Zx; X) and η22 ∈
η12
hη, ζiBy
=ZΓy
η11(γ)∗ζ11(γ)∗d(µΓ)y(γ) +ZZ−1
y
Fhη21(♭(z)), ζ21(♭(z))id(µZ−1)y(♭(z)),
where (µZ−1)y is the Radon measure on Z−1 with support Z−1
Z−1 −→ Z, ♭(z) 7−→ z; it is then given by
y , which is the image of µy on Z under the "inversion"
(µZ−1 )y(φ) =ZGr♭(♭(z))
φ(φ(g−1 · ♭(z)))dµr♭(♭(z))
G
(g), for φ ∈ Cc(Z−1).
So, by using Notations 3.10, we get hξ, ηiBy
In the same way, we verify that L2(Mx; L) = L2(Zx; X) ⊕ L2(Gx; E). Thus, for all ξ ∈ Cc(M; L), we have
+ hhη21 , ζ21ii
⋆
= hη11, ζ11iBy
(y); hence L2(My; L) = L2(Γ; F) ⊕ L2(Z−1
y ; X).
r (M;L) = max
r(M;L) = max
kξkC∗
sup
y∈Γ(0)
kπM
y (ξ)k, sup
x∈G(0)
kπM
0
ξ11
0
0 ∈ Cc(M; L), and y ∈ Γ(0), then πM
In particular, if ξ =
Now, let x ∈ G(0), and suppose η ∈ Cc(Mx; L) is such that kηkL2 (Mx;L) ≤ 1; i.e. maxnkη12kL2(Zx;X), kη22kL2(G;E)o ≤ 1.
Then, from a simple calculation we obtain
y(ξ11) ⊕ 0, so that
r(Γ;F), sup
x∈G(0)
y (ξ) = πΓ
kξ11kC∗
kξkC∗
kπM
(30)
.
hπM
x (ξ)η, πM
x (ξ)ηiAx
= hξ11 · η12 , ξ11 · η12i
⋆
(x) = hRΓ
x(ξ11)η12 , RΓ
x(ξ11)η12i
⋆
(x);
hence, by applying Proposition 4.3, we get kπM
from (30), we get kξkC∗
r(M;L) = kξ11kC∗
r(Γ;F).
x (ξ)ηkL2(Mx;L)
= kRΓ
x (ξ11)η12kL2(Zx;X) ≤ kξ11kC∗
r(Γ;F). Therefore,
(cid:3)
.
x (ξ)k
x (ξ)k
/
/
14
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
Recall ( [6, p.14]) that if A is a Dixmier-Douady bundle over G, and Γ ∼Z G, we define the pull-back AZ over Γ
as the quotient space of the pull-back s∗A := {(z, a) ∈ Z × A s(z) = p(a)} by G, where G acts on s∗A (on the right)
by (z, a) · g := (z · g, α−1
g (a)).
Corollary 5.6. Assume that (A, G, α) is a Dixmier-Douady bundle, and that Γ ∼Z G. Then A ⋊r G ∼Morita AZ ⋊r Γ.
Proof. Observe that for γ ∈ Γ, the fibre (s∗
s∗A −→ Z. Then, the Fell system (Γ, s∗
ΓAZ) acts on (Z, s∗A) on the left via
ΓAZ)γ = AZ
sΓ(γ) is identified with (ZsΓ(γ) ×G(0) A)/G. Consider the C∗-bundle
ZsΓ(γ) ×G(0) A
G
× As(z) ∋ ([z, a], b) 7−→ ab ∈ As(γz) = As(z),
where (γ, z) ∈ Γ ∗ Z. Also, we have a right Fell G-pair (s∗
G
A, s∗A) over Z determined by the right action
Next, define the inner products in the obvious way: if (z, ♭(z′)) ∈ Z ×G(0) Z−1, we set
As(z) × AsG(g) ∋ (a, b) 7−→ α−1
g (a)b ∈ As(zg) = AsG(g).
As(z) × As(z) ∋ (a, ♭(b)) 7−→ [z, ab∗] ∈ (s∗
ΓAZ)Γ<z,z′> =
ZsΓ(Γ<z,z′ >) ×G(0) A
G
,
and if (♭(z), z′) ∈ Z−1 ×Γ(0) Z, we put
As(z) × As(z′) ∋ (♭(a), b) 7−→ α−1
<z,z′ >G
(a)∗b ∈ (s∗
GA)<z,z′>G
= AsG(<z,z′>G) = As(z′).
(31)
(32)
(33)
(34)
It is not hard to check that the settings ( 31), ( 32), ( 33), and ( 34) give an equivalence of Fell systems (Γ, s∗
(G, s∗
A). We thus complete the proof by applying Theorem 5.5.
ΓAZ) ∼(Z,s∗ A)
(cid:3)
G
Remark 5.7. In particular, it results from the last corollary that twisted K-theory ( [21]) is invariant under Morita
equivalences of locally compact Hausdorff groupoids; i.e. if A ∈ Br(G), and Γ ∼Z G, one has K∗
A(G) (cid:27) K∗
AZ (Γ).
6. The reduced C∗-algebra of an S1-central extension
Let G be groupoid. Recall that ( [21], [6], [20]) an S1-central extension of G is a pair ( S1
π
/eΓ
where S1
/ Γ is a central groupoid extension, and Γ ∼P G; that iseΓ(0) = Γ(0), S1 acts continuously
on eΓ, and π : eΓ −→ Γ is an S1-principal bundle. Such an object is symbolized as (eΓ, P) if there is no risk
of confusion. We say that (eΓ1, P1) and (eΓ2, P2) are Morita equivalent if there exists an S1-equivariant Morita
equivalence Z :eΓ1 −→eΓ2
4 such that the following diagrams commute (in terms of generalized morphisms):
π
/ Γ , P),
/eΓ
Z/S1
Γ1
@
@
@
@
@
P1
@
@
@
Γ2
and Γ1
P2
G
Z/S1
A
A
A
Zδ1
A
A
A
A
Γ2
Zδ2
A
Z2
The set of Morita equivalence classes of S1-central extensions of G is an Abelian group denoted by Ext(G, S1).
Note that the inverse of a class [E] in Ext(G, S1) is the class of the opposite Eop defined as follows: if E = (eΓ, P), then
Eop := (eΓop, P), whereeΓop iseΓ as a topological groupoid but the S1-action is the conjugate one; i.e. t · γop := (t−1 γ)op.
It is known ( [6], [19], [8], [21]) that elements of Ext(G, S1) are in bijection to those of of the 2-cohomology group
H2(G•, S1); more precisely, there is an ismorphism of abelian groups Ext(G, S1) (cid:27) H2(G•, S1); we refer to [6], [19], [8]
4Let πi :eΓi −→ Γi, i = 1, 2 be an S1-principal bundle. A generalised morphism Z :eΓ1 −→eΓ2 is said to be S1-equivariant if there is an action of
S1 on Z such that (λ γ1) · z · γ2 = γ1 · (λz) · γ2 = γ1 · z · (λ γ2), for any (λ, γ1, z, γ2) ∈ S1 ×eΓ1 × Z ×eΓ2 such that these products make sense. Is is shown
( [21]) that such a morphism induces a generalized morphism Z/S1 : Γ1 −→ Γ2.
/
/
/
/
/
/
/
/
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
15
for more details.
Given an S1-central extension E = (eΓ, P) of G, we form the Fell system (L, Γ) as follows:
If [ γ, t] denote the class of ( γ, t) ∈eΓ × C in L, then we get a line bundle over Γ by setting L ∋ [ γ, t] 7−→ π( γ) ∈ Γ.
Next, we define the multiplication and the ∗-involution on L as Lγ1 × Lγ2 ∋ ([ γ1, t1], [ γ2, t2]) 7−→ [ γ1 γ2, t1t2] ∈ Lγ1γ2 ,
for (γ1, g2) ∈ Γ(2), and Lγ ∋ [ γ, t] 7−→ [ γ−1, t−1] ∈ Lγ−1 , respectively.
L :=eΓ ×S1 C := (eΓ × C)/( γ,t)∼(λ· γ,λ−1t),λ∈S1.
Definition 6.1. The reduced C∗-algebra C∗
C∗
r(E) := C∗
r(Γ; L).
r(E) of E is defined as the reduced C∗-algebra of the Fell system (L, Γ); i.e.
Proposition 6.2. (Compare with [21, Proposition 3.3]). Suppose E1 ∼ E2; i.e.
C∗
r(E1) ∼Morita C∗
r(E2).
[E1] = [E2] in Ext(G, S1). Then
πi
Proof. Suppose Ei = ( S1
equivalence, then take X := Z ×S1 C = Z × C/(z,t)∼(λz,λ−1t). Then, X is a line bundle over Z/S1, the projection being
the map [z, t] 7−→ [z], where [z] is the class of z in the quotient space Z/S1. Furthermore, it is easy to verify
that (Z/S1, X) implements an equivalence of Fell systems (Γ1, L1) ∼ (Γ2, L2). Therefore, our assertion follows from
Theorem 5.5.
/ Γi , δi, Pi), and Li :=eΓ1 ×S1 C. If Z :eΓ1 −→eΓ2 is an S1-equivariant Morita
/eΓi
(cid:3)
Let us now recall some constructions that we will need in the next result (see for instance [6, 20, 8] for more
of
= {µy
eΓ
Cc(eΓy; H)S1
being a countably generated continuous field of infinite-dimensional Hilbert spaces over the finite dimensional
( γ). Denote by HeΓ
with respect to h·, ·iy, and let HeΓ :=`y∈Γ(0) HeΓ
details). From an S1-central extension E = (eΓ, P) of G, one constructs a Dixmier-Douady bundle (AE, G, αE) in
}y∈Γ(0) be a Haar system oneΓ. For any y ∈ Γ(0), define the space Cc(eΓy; H)S1
the following way. Let µeΓ
compactly supported continuous H-valued S1-equivariant functions oneΓx as
:=nξ ∈ Cc(eΓy; H) ξ(t · γ) = t−1ξ( γ), ∀t ∈ S1, γ ∈eΓyo .
Next, define a scalar product h·, ·iy on Cc(eΓy; H)S1 by hξ, ζiy :=ReΓy hξ( γ), ζ( γ)idµy
y := L2(eΓy; H)S1 the
eΓ
y. Then HeΓ −→ Γ(0),
Hilbert space obtained by completing Cc(eΓy; H)S1
locally compact space Γ(0), is a locally trivial Hilbert bundle (cf. [4, Th´eor`eme 5]). Moreover,eΓ acts continuously
and by unitaries on HeΓ under the operation: Cc(eΓs(γ); H)S1 ∋ ξ 7−→(cid:16) γ · ξ :eΓr(γ) ∋ h 7−→ ξ( γ−1 h) ∈ H(cid:17) ∈ Cc(eΓr(γ); H)S1 ,
for γ ∈ eΓ. Consider the continuous field of elementary C∗-algebras KeΓ := `y∈Γ(0) K(HeΓ
y). Then KeΓ −→ Γ(0)
is a locally trivial C∗-bundle with fibre K, according to [4, Th´eor`eme 8] (by comparing the field HeΓ with the
trivial Hilbert bundle). Furthermore, there is a continuous action α of Γ by automorphisms on KeΓ given by:
KeΓ
s(γ) ∋ T 7−→ γ−1T γ ∈ KeΓ
r(γ), where γ is any lift of γ oneΓ, which gives us an element (KeΓ, Γ, α) in Br(Γ). Finally,
we define AE over G as the pull-back of KeΓ through the Morita equivalence G ∼P−1 Γ; i.e. AE := (KeΓ)P−1 . This
construction gives a homomorphism of abelian groups Ψ : Ext(G, S1) −→ Br(G). Conversely, from a Dixmier-
Douady bundle over G, it is not hard to build an S1-central extension of G, and then to construct a homomorphism
Br(G) −→ Ext(G, S1) which is inverse to Ψ ( [6], [21]).
Theorem 6.3. Let E ∈ Ext(G, S1), and let (AE, G, αE) in Br(G) be its corresponding Dixmier-Douady bundle over G. Then,
under the above constructions and notations, we have
Proof. Write E = ( S1
we will denote A := KeΓ. From Corollary 5.6, we have
AE ⋊r G ∼Morita C∗
r(Eop).
π
/ Γ , Z), where Z : G −→ Γ is a Morita equivalence. For the sake of simplicity,
/eΓ
AE ⋊r G ∼Morita A ⋊r Γ := C∗
r(Γ; s∗KeΓ).
/
/
/
/
16
EL-KAIOUM M. MOUTUOU AND JEAN-LOUIS TU
Thus, we only have to show that
C∗
r(Γ; s∗A) ∼Morita C∗
r(Γ; L) =: C∗
r(Eop), where L :=eΓop ×S1 C.
However, again in view of the Renault's equivalence Theorem 5.5, it suffices to build an equivalence between the
Fell systems (Γ, s∗A) and (Γ, L).
Consider the Banach bundle X := s∗HeΓ over Γ defined as the pull-back of the HilberteΓ-bundle HeΓ −→ Γ(0) through
the source map of Γ. We claim that X implements the desired equivalence over Γ; that is, that
(Γ, s∗A) ∼(Γ,X) (Γ, L).
(36)
From theeΓ-action on HeΓ defined in the discussion before the theorem, we get a left action of s∗A on X given by
K(L2(eΓs(γ1), H)S1 ) × L2(eΓs(γ2), H)S1 −→ L2(eΓs(γ1γ2), H)S1
7−→ T · ξ := γ−1
(T
ξ)
,
2 T( γ2ξ),
(35)
(37)
(38)
and a right action of L on X
L2(eΓs(γ2), H)S1 ×(cid:16)eΓop
(ξ ,
×S1 C(cid:17) −→ L2(eΓs(γ2γ3), H)S1
7−→ ξ · [ γ, λ] := γ−1
3
γ3
[ γ, λ])
· λξ
where γ3 is any lift of γ3 ineΓ. The maps ( 37) and ( 38) are continuous since the Γ-actions are continuous. Also,
they are full since the actions are, in fact, isomorphisms.
We now construct the s∗A-valued and L-valued inner products X ∗ X −→ s∗AΓ< and X × X −→ L>Γ , respectively.
Note that, as in Example 3.8, Γ−1 = {γ−1 γ ∈ Γ}, if (γ, ♭(γ′)) ∈ Γ ×Γ(0) Γ−1 (in other words, s(γ) = s(γ′)), then
Γ < γ, γ′ >= γγ′−1, and if (♭(γ′), γ") ∈ Γ−1 ×Γ(0) Γ (i.e. r(γ′) = r(γ")), then < γ′, γ" >Γ= γ′−1γ". We then define these
inner products as
Xγ × Xγ′ −→ As(γγ′−1)
(ξ, ♭(η))
= K(L2(eΓr(γ′), H)S1
where for ζ, ζ′ ∈ L2(eΓy, H)S1 , θζ,ζ′ ∈ K(L2(eΓy, H)S1 ) is the rank one operator
∋ ζ" 7−→ (hζ′, ζ"ix)ζ ∈ L2(eΓy, H)S1
L2(eΓy, H)S1
7−→ s∗Ahξ, ηi := θ γ′ξ, γ′η
)
and
, y ∈ Γ(0);
Xγ′ × Xγ" −→ Lγ′−1γ"
(♭(ξ), η)
γ′−1γ"
7−→ hξ, ηiL :=h γ′ −1 γ", h γ′ξ, γ"ηir(γ′)i
×S1 C
=eΓop
(39)
(40)
where, as usual, γ′ and γ" are any lifts of γ′ and γ", respectively. Recall that for y ∈ Γ(0), the scalar product h·, ·i(y)
on HeΓ
= Xy is defined as
y
hξ, ηi(y) =Z
eΓy
hξ(h), η(h)idµy
eΓ
(h) ∈ C.
The algebraic properties of these maps are easy to check. The map ( 39) is full, for spannθζ,ζ′ ζ, ζ′ ∈ L2(eΓr(γ′), H)S1o
is the ideal of finite-rank operators on L2(eΓr(γ′), H)S1 and the map L2(eΓs(γ′), H)S1 −→ L2(eΓr(γ′), H)S1 given by the
eΓ-action is an isomorphism of Hilbert spaces. The map (40) is clearly surjective. Thus, it only remains to verify
that the compatibility condition (cf. Definition 3.6 (ii)) holds; that is, for any triple (γ, γ′−1, γ") ∈ Γ ×Γ(0) Γ−1 ×Γ(0) Γ,
ξ · hξ′, ξ"iL = s∗ Ahξ, ξ′i · ξ", ∀(ξ, ♭(ξ′), ξ") ∈ Xγ × Xγ′ × Xγ".
(41)
One has
EQUIVALENCE OF FELL SYSTEMS AND THEIR REDUCED C∗-ALGEBRAS
ξ · hξ′, ξ"iL = ξ ·h γ′−1 γ", h γ′ξ′, γ"ξ"i(r(γ′))i
= γ"−1 γ′ · (h γ′ξ′, γ"ξ"i(r(γ′)))ξ
= γ"−1 · (h γ′ξ′, γ"ξ"i(r(γ′)))( γ′ξ)
which completes the proof.
= γ"−1 · θ γ′ξ, γ′ξ′ ( γ"ξ")
= s∗ Ahξ, ξ′i · ξ",
References
17
(cid:3)
[1] Akemann, C.A., Pedersen, G.K., Tomiyama, J., Multipliers of C∗-Algebras. Journal of Functional Analysis, 13, 277-301 (1973).
[2] Brown, L., Stable Isomorphism of Hereditary Subalgebras of C∗-Algebras. Pacific Journal of Mathematics, vol. 71, No. 2 (1977), 335-348.
[3] Brown, L., Green, P., Rieffel. M., Stable Isomorphism and Strong Morita Equivalence of C∗-Algebras. Pacific Journal of Mathematics, Vol. 71, No. 2
(1977).
[4] Dixmier, J., Douady, A., Champs continus d'espaces hilbertiens et de C∗-alg`ebres. Bull. Soc. Math. France 91 (1963), 227-284.
[5] Kumjian, A., Fell bundles over groupoids. Proceeding of The American Mathematical Society. Vol. 126, Num. 4, (1998), pp. 1115-1125.
[6] Kumjian, A., Muhly, P. S., Renault, J. N., Williams, D. P., The Brauer group of a locally compact groupoid. American Journal of Mathematics 120
(1998), 901-954.
[7] Lance, E. Hilbert C∗-Modules. A Toolkit for operator algebraists. Cambridge University Press, (1995).
[8] Moutuou, E., PhD. Thesis, in preparation.
[9] Muhly, P.S., Bundles over groupoids, Groupoids in analysis, geometry, and physics (Boulder, CO,1999), Contemp. Math., vol. 282, Amer. Math.
Soc., Providence, RI, 2001, pp. 6782.
[10] Muhly, P., Renault, J., Williams, D., Equivalence and isomorphism for groupoid C∗-algebras. J. Operator Theory, 17 (1987), 3-22.
[11] Muhly, P., Williams, D., Renault's Equivalence Theorem for Groupoid Crossed Products. New York Journal of Mathematics Monographs, vol 3,
SUNY, Albany NY, (2008).
[12] Mulhy, P., Williams, D., Equivalence and Desintegration Theorems For Fell Bundles And Their C∗-Algebras. Dissertationes Mathematicae (2008).
[13] Paravicini, W., KK-Theory for Banach Algebras and Proper Groupoids. PhD Thesis, (2007).
[14] Pedersen, G., C∗-Algebras and Their automorphism Groups. London Mathematical Society Monographs, vol. 14, Academic Press, London,
(1979).
[15] Raeburn, I., Williams, D., Morita Equivalence and Continuous-Trace C∗-Algebras. Mathematical Surveys and Mononographs, vol. 60, American
Mathematical Society, (1998).
[16] Renault, J., A Groupoid approach to C∗-algebras. Lecture Notes in Mathematics, 793, Springer (1980).
[17] Renault, J., Repr´esentations des produits crois´es d'alg`ebres de groupoıdes. J. Operator Theory, 18 (1987), 67-97.
[18] Sims, A., Williams, D., Renault's Equivalence Theorem for Reduced Groupoid C∗-Algebras. Electronic: arXiv:1002.3093 (2010).
[19] Tu, J. L., Groupoid cohomology and extensions. Trans. Amer. Math. Soc. 358 (2006), 4721-4747.
[20] Tu, J. L., Twisted K-theory and Poincar´e duality. Trans. Amer. Math. Soc. 361 (2009), 1269-1278.
[21] Tu, J. L, Xu, P. , Laurent-Gengoux, C.,Twisted K-Theory of differentiable stacks. Ann. Scient. ´Ec. Norm. Sup. 4e s´erie, t. 37 (2004), p.841-910.
[22] Yamagami, S., On the ideal structure of C∗-algebras over locally compact groupoids, unpub-lished manuscript, 1987.
Department of Mathematics, Paderborn University, Warburger Str. 100, D-33098 Paderborn, Germany, and
Universit´e Paul Verlaine - Metz, LMAM - CNRS UMR 7122, B atiment A, Ile du Saulcy, 57000 Metz, France
E-mail address: [email protected]
Universit´e Paul Verlaine - Metz, LMAM - CNRS UMR 7122, B atiment A, Ile du Saulcy, 57000 Metz, France
E-mail address: [email protected]
|
1809.01788 | 1 | 1809 | 2018-09-06T01:47:20 | Noncommutative weighted individual ergodic theorems with continuous time | [
"math.OA",
"math.FA"
] | We show that ergodic flows in noncommutative fully symmetric spaces (associated with a semifinite von Neumann algebra) generated by continuous semigroups of positive Dunford-Schwartz operators and modulated by bounded Besicovitch almost periodic functions converge almost uniformly. The corresponding local ergodic theorem is also discussed. | math.OA | math | NONCOMMUTATIVE WEIGHTED
INDIVIDUAL ERGODIC THEOREMS
WITH CONTINUOUS TIME
VLADIMIR CHILIN AND SEMYON LITVINOV
Abstract. We show that ergodic flows in noncommutative fully symmetric
spaces (associated with a semifinite von Neumann algebra) generated by con-
tinuous semigroups of positive Dunford-Schwartz operators and modulated by
bounded Besicovitch almost periodic functions converge almost uniformly. The
corresponding local ergodic theorem is also discussed.
.
A
O
h
t
a
m
[
1
v
8
8
7
1
0
.
9
0
8
1
:
v
i
X
r
a
1. Introduction
In the classical ergodic theory, Besicovitch-weighted individual ergodic theorems
have been studied quite extensively (see, for example, [33, 2, 3, 25, 11]).
In the noncommutative setting, first individual ergodic theorem with bounded
Besicovitch weights in a von Neumann algebra was obtained in [19]. Later, in [7],
a similar result concerning the so-called bilaterally almost uniform convergence (in
Egorov's sense) was established in the L1-space associated with a semifinite von
Neumann algebra. In [29], utilizing the approach of [20], a multi-parameter version
of [7, Theorem 4.6] was proved for every noncommutative Lp-space with 1 < p < ∞.
Recently, almost uniform convergence (in Egorov's sense) of Besicovitch-weighted
ergodic averages in fully symmetric spaces of measurable operators was established
in [8, Theorem 4.7 and Sec. 6].
Note that all of the above were concerned with bounded Besicovitch sequences
- generated by the well-studied Besicovitch almost periodic functions [1] - leaving
open the problem what happens in the case of actions of continuous semigroups,
when one has to turn to Bisicovitch almost periodic functions. To this end, a
noncommutative local ergodic Besicovitch-weighted theorem was first considered in
[30]; see remarks following Theorem 3.2.
Let {Tt}t≥0 be a (continuous) semigroup of positive Dunford-Schwartz operators
in a noncommutative (L1 + L∞)-space. Our goal is to show that the corresponding
ergodic averages modulated by a bounded Besicovitch (zero-Besicovitch) almost
periodic function β(t), t ≥ 0, in a noncommutative fully symmetric space converge
almost uniformly as t → ∞ (Theorem 4.2) (respectively, as t → 0 (Theorem 4.3)).
Since the results appear to be new for the commutative setting, a relevant discussion
is given in the last section of the article.
Date: September 5, 2018.
2010 Mathematics Subject Classification. 47A35(primary), 46L52(secondary).
Key words and phrases. Dunford-Schwartz operator, continuous semigroup, bounded Besicov-
itch function, almost uniform convergence.
1
2
VLADIMIR CHILIN AND SEMYON LITVINOV
2. Preliminaries
Let M be a semifinite von Neumann algebra equipped with a faithful normal
semifinite trace τ . Denote by P(M) the complete lattice of projections in M. If
1 is the identity of M and e ∈ P(M), we write e⊥ = 1 − e. Also, if M acts in
eα the projection on
a Hilbert space H, then, given {eα}α∈Λ ⊂ P(M), denote Vα∈Λ
the subspace Tα∈Λ
eα ∈ P(M) and
eαH. Note that Vα∈Λ
τ(cid:18)(cid:20) ∞^n=1
en(cid:21)⊥(cid:19) ≤
∞Xn=1
τ (e⊥
n )
for any sequence {en} ⊂ P(M).
Let L0 = L0(M, τ ) be the ∗-algebra of τ -measurable operators affiliated with
M, and let k · k∞ be the uniform norm in M. Equipped with the measure topology
given by the system
N (ǫ, δ) = {x ∈ L0 : kxek∞ ≤ δ for some e ∈ P(M) with τ (e⊥) ≤ ǫ},
ǫ > 0, δ > 0, of (closed) neighborhoods of zero, L0 is a complete metrizable
topological ∗-algebra [31].
Let Lp = Lp(M, τ ), 1 ≤ p ≤ ∞, (L∞ = M) be the noncommutative Lp-space
associated with (M, τ ), and let k · kp be the standard norm in the space Lp, 1 ≤
p < ∞. For detailed accounts on the noncommutative Lp-spaces, p ∈ {0} ∪ [1, ∞),
see [35, 31, 36, 32].
A net {xα} ⊂ L0 is said to converge almost uniformly (a.u.) (bilaterally almost
uniformly (b.a.u.)) to x ∈ L0 if for any given ǫ > 0 there is a projection e ∈ P(M)
such that τ (e⊥) ≤ ǫ and k(x − xα)ek∞ → 0 (respectively, ke(x − xα)ek∞ → 0).
It is well-known that if a sequence in L0 converges in measure, then it has a
subsequence converging a.u. Besides, a sequence in L0 converging in Lp for some
1 ≤ p ≤ ∞ also converges in measure.
A linear operator T : L1 + M → L1 + M is called a Dunford-Schwartz operator
(writing T ∈ DS) if
kT (x)k1 ≤ kxk1 ∀ x ∈ L1 and kT (x)k∞ ≤ kxk∞ ∀ x ∈ L∞.
If a Dunford-Schwartz operator T is positive, that is, T (x) ≥ 0 whenever x ≥ 0, we
will write T ∈ DS+. Note that positive absolute contractions in L1, considered in
[37] and then in [7, 29, 30], can be uniquely extended to positive Dunford-Schwartz
operators - see [5].
Given x ∈ L1 + M and T ∈ DS, denote
An(x) =
1
n
n−1Xk=0
T k(x).
The following fundamental result is due to Yeadon [37, Theorem 1].
Theorem 2.1. Let T ∈ DS+. Then for every x ∈ L1
e ∈ P(M) such that
+ and λ > 0 there exists
τ (e⊥) ≤
kxk1
λ
and
sup
n
keAn(x)ek∞ ≤ λ.
NONCOMMUTATIVE WEIGHTEDINDIVIDUAL ERGODIC THEOREMSWITH CONTINUOUS TIME3
Let a semigroup {Tt}t≥0 ⊂ DS be strongly continuous in L1, that is,
for all x ∈ L1.
kTt(x) − Tt0 (x)k1 → 0 whenever t → t0
Then, given x ∈ L1 and t > 0, there exists
At(x) =
Ts(x)ds ∈ L1
1
t Z t
0
(see the argument preceding (3)).
Here is a continuous extension of Theorem 2.1 (cf. [20, Remark 4.7]):
Theorem 2.2. If {Tt}t≥0 ⊂ DS+ is strongly continuous in L1, then, given x ∈ L1
+
and λ > 0, there exists e ∈ P(M) such that
τ (e⊥) ≤
2kxk1
λ
and
sup
t>0
keAt(x)ek∞ ≤ λ.
Proof. Let N (Q) be the set of all natural (respectively, rational) numbers, and let
n
0 Ts/m(x)ds. We have
m ∈ Q, where n, m ∈ N. Denote y =R 1
nZ n
Ts(x)ds =
0 ≤ A n
(x) =
m
1
m
0
Ts/m(x)ds
m
n Z n
n(cid:18)Z 1
1
0
0
=
By Theorem 2.1, given λ > 0, there is f ∈ P(M) such that
n−1
Ts/m(x)ds + · · · +Z n
n (cid:13)(cid:13)(cid:13)(cid:13)f
kxk1
and
sup
λ
1
n
Ts/m(x)ds(cid:19) =
n−1Xk=0
1/m(y) f(cid:13)(cid:13)(cid:13)(cid:13)∞
n−1Xk=0
T k
1
n
≤ λ,
T k
1/m(y).
τ (f ⊥) ≤
kyk1
λ
implying, that
≤
sup
0<r∈Q
kf Ar(x)f k∞ ≤ λ.
If t > 0 and 0 < rn → t, rn ∈ Q, then we have Arn (x) → At(x) in L1, hence in
(x) → At(x) a.u. for a subsequence {rnk } ⊂ {rn}. Thus,
measure. Therefore Arnk
it is possible to find g ∈ P(M) such that
τ (g⊥) ≤
kxk1
λ
and kgArnk
(x)gk∞ → kgAt(x)gk∞ as k → ∞.
Letting e = f ∧ g, we obtain the desired inequalities.
(cid:3)
3. Convergence in the space L1(M, τ )
Let C be the field of complex numbers, and let C1 = {z ∈ C : z = 1}. A
function p : R+ → C is called a trigonomertic polynomial if p(t) =
n ∈ N, {wj}n
1 ⊂ C, and {λj}n
1 ⊂ C1.
A Lebesgue measurable function β : R+ → C will be called bounded Besicovitch
β(t) < C < ∞, and for every ǫ > 0 there is a
(zero-Besicovitch) function if sup
t≥0
wj λt
j, where
nPj=1
trigonometric polynomial pǫ such that
(1)
lim sup
t→∞
β(s) − pǫ(s)ds < ǫ
1
t Z t
0
4
VLADIMIR CHILIN AND SEMYON LITVINOV
(respectively,
(2)
lim sup
t→0
1
t Z t
0
β(s) − pǫ(s)ds < ǫ ).
Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1, and assume that
β : R+ → C be a Lebesgue measurable function with kβk∞ < ∞. Fix x ∈ L1.
Then for any given y ∈ M the function ϕx,y(t) = τ (Tt(x)y) is continuous on R+.
Therefore, if µ is the Lebesgue measure on R+, then the map Ux : R+ → L1 defined
as Ux(t) = Tt(x) is weakly µ-measurable [38, Ch.V, § 4]. Since, in addition, Ux(R)
is a separable subset in L1, Pettis theorem [38, Ch.V, § 4] entails that the map Ux is
strongly µ-measurable and the real function kUx(t)k1 = kTt(x)k1 is µ-measurable
on R+. Since kTt(x)k1 ≤ kxk1, it follows that kTs(x)k1 is an integrable function on
[0, t] for any t > 0. Consequently, kβ(s)Ts(x)k1 = β(s) · kTs(x)k1 is also integrable
on [0, t] for any t > 0. By [38, Ch.V, § 5, Theorem 1], the function β(s)Ts(x) is
Bochner µ-integrable on [0, t], t > 0. Therefore, for any x ∈ L1 and t > 0 there
exists
(3)
Bt(x) =
β(s)Ts(x)ds ∈ L1.
1
t Z t
0
Lemma 3.1. Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1, and
let β : R+ → C be a Lebesgue measurable function such that sup
β(t) ≤ C < ∞.
t≥0
If x ∈ L1 and ǫ > 0, then there is a projection e ∈ P(M) satisfying inequalities
τ (e⊥) ≤
4kxk1
ǫ
and
sup
t>0
keBt(x)ek∞ ≤ 48Cǫ.
Proof. We have x = (x1 − x2) + i(x3 − x4), where xj ∈ L1
each j = 1, 2, 3, 4. By Theorem 2.2, given j, there exists ej ∈ P(M) such that
+ and kxjk1 ≤ kxk1 for
τ (e⊥
j ) ≤
kxjk1
ǫ
and
sup
t>0(cid:13)(cid:13)(cid:13)ej
1
t Z t
0
Ts(xj )ds ej(cid:13)(cid:13)(cid:13)∞
≤ 2ǫ.
Next, we have 0 ≤ Re β(s) + C ≤ 2C and 0 ≤ Im β(s) + C ≤ 2C, s ≥ 0, implying
that
0 ≤ [Re β(s) + C]Ts(xj) ≤ 2CTs(xj ) and 0 ≤ [Im β(s) + C]Ts(xj) ≤ 2CTs(xj )
0 ≤
[Re β(s) + C]Ts(xj )ds ≤ 2C
0 ≤
[Im β(s) + C]Ts(xj )ds ≤ 2C
for each j. This, together with the decomposition
for all s ≥ 0, hence
and
0
1
1
t Z t
t Z t
t Z t
0
0
1
0
1
t Z t
t Z t
0
1
Ts(xj )ds
Ts(xj )ds.
1
t Z t
0
Bt(xj ) =
[Re β(s) + C]Ts(xj )ds + i
[Im β(s) + C]Ts(xj)ds
− C(1 + i)
implies that
1
t Z t
0
Ts(xj)ds,
kejBt(xj)ejk∞ ≤ 12Cǫ, j = 1, 2, 3, 4.
sup
t>0
NONCOMMUTATIVE WEIGHTEDINDIVIDUAL ERGODIC THEOREMSWITH CONTINUOUS TIME5
Now, letting e =
4Vj=1
ej, we obtain the desired inequalities.
(cid:3)
The following is a noncommutative counterpart of the classical notion of continu-
ity at zero of the maximal operator (see [26, Proposition 1.1 and ensuing remarks]).
Definition 3.1. Let (X, k · k) be a Banach space. A sequence of linear maps
Mn : X → L0 is called bilaterally uniformly equicontinuous in measure (b.u.e.m.)
at zero if, given ǫ > 0 and δ > 0, there exists γ > 0 such that for every kxk < γ
there exists a projection e ∈ P(M) satisfying conditions
τ (e⊥) ≤ ǫ and sup
n
keMn(x)ek∞ ≤ δ.
In order to establish a.u. convergence - which is generally stronger than b.a.u.
convergence - of the averages Bt(x), we will need the following lemma a proof of
which can be found in [27, Lemma 3.2]; see also [28].
Lemma 3.2. Let Mn : L1 → L1 be sequence of linear maps that is b.u.e.m. at
zero. If {xm} ⊂ L1 is such that kxmk1 → 0, then for every ǫ > 0 and δ > 0 there
are e ∈ P(M) and xm0 ∈ {xm} satisfying conditions
τ (e⊥) ≤ ǫ and sup
n
kMn(xm0 )ek∞ ≤ δ.
In what follows we shall assume that M has separable predual. Denote by ν
the normalized Lebesgue measure on C1. Let fM be the von Neumann algebra of
essentially bounded ultraweakly measurable functions ef : (C1, ν) → M equipped
with the trace
eτ (ef ) =ZC1
τ (ef (z))dν(z), ef ≥ 0.
Repeating the argument in [10, Lemma 2] (see also [7, Lemma 4.1]), we obtain
Let fL1 be the Banach space of Bochner ν-integrable functions ef : (C1, ν) →
L1(M, τ ). As the predual of fM [34, Theorem 1.22.13], the space fL1 is isomor-
phic to L1(fM,eτ ).
Lemma 3.3. If fL1 ∋ eft → ef ∈ fL1 a.u. as t → ∞, then eft(z) → ef (z) a.u. as
t → ∞ for ν-almost all z ∈ C1.
the following.
Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1. Pick λ ∈ C1 and
define
T (λ)
t
Then it is easily verified that {T (λ)
(ef )(z) = Tt(ef (λtz)),
T (λ)
ef ∈ L1(fM,eτ ) + fM, and z ∈ C1, we have
)(ef )(z),
t }t≥0 is a semigroup.
that is, {T (λ)
t T (λ)
s
t ≥ 0, ef ∈ L1(fM,eτ ) + fM, z ∈ C1.
t }t≥0 ⊂ DS+(fM,eτ ). Besides, given t, s ≥ 0,
t+s(ef )(z) = Tt+s(ef (λt+sz)) = Tt(Ts(ef (λsλtz))) = Tt(T (λ)
= (T (λ)
s
(ef )(λtz))
Proposition 3.1. The semigropup {T (λ)
t }t≥0 is strongly continuous in L1(fM ,eτ ).
6
VLADIMIR CHILIN AND SEMYON LITVINOV
kT (λ)
We have
Proof. Let us show that if 0 ≤ sn → 0 and ef ∈ L1(fM,eτ ), then
sn (ef ) − ef kL1( fM,eτ ) =ZC1
kTsn (ef (λsn z)) − ef (z)k1 dν(z) → 0 as n → ∞.
kTsn(ef (λsn z)) − ef (z)k1 ≤ kTsn(ef (λsn z)) − Tsn (ef (z))k1 + kTsn (ef (z)) − ef (z)k1
Since {Tt}t≥0 is strongly continuous in L1, it follows that kTsn (ef (z)) − ef (z)k1 → 0
≤ kef (λsn z) − ef (z)k1 + kTsn(ef (z)) − ef (z)k1.
for ν-almost all z ∈ C1. Besides,
Thus, it remains to verify that
where kef (z)k1 ∈ L1(C1, ν), which allows us to conclude that
ν − a.e.,
kTsn (ef (z)) − ef (z)k1 ≤ 2kef (z)k1
ZC1
kTsn(ef (z)) − ef (z)k1 dν(z) → 0 as n → ∞.
ZC1
kef (λsn z)) − ef (z)k1 dν(z) → 0 as n → ∞.
mXi=1
xiχ[λi,λi+1)(z), z ∈ C1, xi ∈ L1, i = 1, . . . , m
eh(z) =
be a simple Bochner measurable function, where {λi}m
C1. Then, given s ∈ R, we have
i=1 is a partition of the circle
Let
(4)
(5)
(6)
mXi=1
xiχ[λsλi,λsλi+1)(z),
eh(λsz) =
where {λsλi}n
as n → ∞ for all i = 1, . . . , m, implying that
i=1 is another partition of C1. Since sn → 0, it follows that λsn λi → λi
ZC1
keh(λsn z) −eh(z)k1 dν(z) → 0 as n → ∞.
Let A be the subalgebra of the σ-algebra of Lebesgue measurable sets of (C1, ν)
generated by the arcs of the circle C1. Since any A ∈ A is a finite union of pairwise
disjoint arcs of C1, a simple Bochner measurable function
mXi=1
xiχAi(z), Ai ∈ A, i = 1, . . . , m
eh(z) =
has the form (4) for some partition {λi}l
i=1 of C1. Therefore, (5) holds for any
Since the σ-algebra generated by the subalgebra A coincides ν-almost everywhere
with the σ-algebra Σ of Lebesgue measurable sets in (C1, ν), given an arbitrary
i=1 xiχAi(z), Ai ∈ Σ, i = 1, . . . , m,
k=1 of simple Bochner measurable functions of the
simple Bochner measurable function eh of the form (6).
simple Bochner measurable function eg(z) = Pm
there exists a sequence {ehk(z)}∞
form (6) such that
ZC1
keg(z) −ehk(z)k1 dν(z) → 0 as k → ∞.
NONCOMMUTATIVE WEIGHTEDINDIVIDUAL ERGODIC THEOREMSWITH CONTINUOUS TIME7
Therefore
ZC1
keg(λsn z) −eg(z)k1 dν(z) ≤ZC1
keg(λsn z) −ehk(λsn z)k1 dν(z)
+ZC1
= 2ZC1
kehk(λsn z) −ehk(z)k1 dν(z) +ZC1
kehk(z) −eg(z)k1 dν(z) +ZC1
implies that (5) holds for any simple Bochner measurable function eg.
As ef ∈ L1(fM,eτ ), there exists a sequence {egk(z)}∞
kehk(z) −eg(z)k1 dν(z)
kehk(λsn z) −ehk(z)k1 dν(z)
able functions for which
k=1 of simple Bochner measur-
ZC1
kef (z) −egk(z)k1 dν(z) → 0
as k → ∞.
Then, repeating the previous argument, we conclude that the convergence in (5)
Finally, let t ≥ 0, tn > 0, tn ↓ t and denote sn = tn − t. Then we have
holds for ef . Therefore, it now follows that
kT (λ)
sn (ef ) − ef kL1( fM,eτ ) → 0
(ef )kL1( fM,eτ ) = kT (λ)
≤ kT (λ)
kT (λ)
tn (ef ) − T (λ)
t
as n → ∞.
t
t+sn(ef ) − T (λ)
(ef )kL1( fM,eτ )
sn (ef ) − ef kL1( fM,eτ ) → 0
as n → ∞. The case tn ↑ t > 0 is similar.
(cid:3)
Lemma 3.4. Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1, and
nPj=1
let p(t) =
then
wjλt
j be a trigonometric polynomial. If x ∈ L1 and
Pt(x) =
1
t Z t
0
p(s)Ts(x)ds,
(i) the averages Pt(x) converge a.u. as t → ∞;
(ii) the averages Pt(x) converge a.u. to p(0)x as t → 0.
Proof. (i) Fix λ ∈ C1 and let
T (λ)
t
(ef )(z) = Tt(ef (λtz)),
In view of Proposition 3.1, {T (λ)
t }t≥0 ⊂ DS+(fM,eτ ) is a strongly continuous semi-
group on L1(fM ,eτ ). Then, by [8, Corollary 5.2], given ef ∈ L1(fM,eτ ), the averages
t ≥ 0, ef ∈ L1(fM,eτ ) + fM, z ∈ C1.
t Z t
T (λ)
s
(7)
1
0
converge a.u. as t → ∞. Therefore, by Lemma 3.3, the averages
(ef )ds
t Z t
0
1
1
t Z t
0
T (λ)
s
(ef )(z)ds =
Ts(ef (λsz))ds
8
VLADIMIR CHILIN AND SEMYON LITVINOV
converge a.u. as t → ∞ for ν-almost all z ∈ C1. In particular, letting ef (z) = zx,
we conclude that the averages
1
t Z t
0
z
λsTs(x)ds
converge a.u. as t → ∞ for some 0 6= z ∈ C1, implying that the averages
1
t Z t
0
λsTs(x)ds
converge a.u. as t → ∞. Therefore, by linearity, the averages Pt(x) converge a.u.
as t → ∞.
(ii) Now, by [9, Theorem 5.1], if ef ∈ L1(fM,eτ ), it follows that the averages (7)
converge a.u. to ef as t → 0. Then, letting ef (z) = zx, we see as above that
λsTs(x)ds → x a.u.
1
t Z t
0
as t → 0, and the result follows by linearity.
(cid:3)
Here is a Besicovitch-weighted noncommutative individual ergodic theorem for
flows in L1 generated by L1-strongly continuous semigroups of positive Dunford-
Schwartz operators:
Theorem 3.1. Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1,
and let β(t) be a bounded Besicovitch function with kβk∞ < C < ∞. Then, given
x ∈ L1, the averages (3) converge a.u. to some bx ∈ L1 as t → ∞.
Proof. Assume first that x ∈ L1 ∩ M. Fix ǫ > 0 and choose a trigonometric
polynomial p = pǫ to satisfy condition (1). Let {Pt(x)}t>0 be the corresponding
averages from Lemma 3.4. Then we have
kBt(x) − Pt(x)k∞ ≤
1
t Z t
0
β(s) − p(s)ds kxk∞ < ǫ kxk∞
for all big enough values of t. Since, by Lemma 3.4, the averages Pt(x) converge
a.u., it follows that the net {Bt(x)}t>0 is a.u. Cauchy as t → ∞.
Now, let x ∈ L1. Without loss of generality, assume that x ∈ L1
the spectral family of x. Given m ∈ N, if we define ym =R m
then {ym} ⊂ L1
+ ∩ M, {xm} ⊂ L1
+ and kxmk1 → 0.
+, and let {eλ} be
0 λdeλ and xm = x−ym,
Fix ǫ > 0 and δ > 0. If {tn} is a sequence of positive rational numbers which
is dense in (0, ∞), then, by Lemma 3.1, the sequence {Btn} is b.u.e.m. at zero on
L1
+, hence on L1 (see [26, Lemma 4.1]). Applying Lemma 3.2, we find a projection
e ∈ P(M) and xm0 ∈ {xm} such that
τ (e⊥) ≤
ǫ
2
and
sup
n
kBtn(xm0 )ek∞ ≤
δ
3
.
If t > 0, then tnk → t for a subsequence {tnk }, and it easily verified that
kBt(xm0 ) − Btnk
(xm0 )k1 → 0.
(xm0 ) → Bt(xm0 ) in measure, which implies that there is a subse-
Therefore Btnk
quence {tnkl
} such that Btnkl
(xm0 ) → Bt(xm0 ) a.u.
NONCOMMUTATIVE WEIGHTEDINDIVIDUAL ERGODIC THEOREMSWITH CONTINUOUS TIME9
Since kBtnkl
(xm0 )ek∞ < δ/3 for each l, it follows from [8, Lemma 5.1] that
(8)
kBt(xm0 )ek∞ ≤
sup
t>0
δ
3
.
Because ym0 ∈ L1 ∩ M, the net Bt(ym0 ) is a.u. Cauchy as t → ∞. Therefore, there
exist g ∈ P(M) and t0 > 0 such that
(9)
for all t, t′ ≥ t0.
τ (g⊥) ≤
ǫ
2
and k(Bt(ym0) − Bt′ (ym0))gk∞ ≤
δ
3
.
If h = e ∧ g, then τ (h⊥) ≤ ǫ and, in view of (8) and (9), we have
k(Bt(x) − Bt′ (x))hk∞ ≤ k(Bt(ym0) − Bt′ (ym0))hk∞
+ kBt(xm0 )hk∞ + kBt′(xm0 )hk∞ ≤ δ
for all t, t′ ≥ t0. Thus, the net {Bt(x)} is a.u. Cauchy as t → ∞. Since L0 is
complete with respect to a.u. convergence (see proof of [7, Theorem 2.3]), there
β(t) < C < ∞ for all t ≥ 0, each map C −1Bt is a contraction in L1, which,
is bx ∈ L0 such that Bt(x) → bx a.u. In particular, Bt(x) → bx in measure. Since
because the unit ball of L1 is closed in measure topology, implies that bx ∈ L1. (cid:3)
The following theorem is a local ergodic theorem in L1 for L1-continuous semi-
groups of positive Dunford-Schwartz operators modulated by bounded zero-Besicovitch
functions.
Theorem 3.2. Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1, and
let β(t) be a zero-bounded Besicovitch function with kβk∞ < C < ∞. Then, given
x ∈ L1, the averages (3) converge a.u. to α(x) x as t → 0 for some α(x) ∈ C.
Proof. Assume first that x ∈ L1 ∩ M. Fix ǫ > 0 and choose a trigonometric
If {Pt(x)}t>0 are the averages from
polynomial p = pǫ to satisfy condition (2).
Lemma 3.4, then
kBt(x) − Pt(x)k∞ ≤
1
t Z t
0
β(s) − p(s)ds kxk∞ < ǫ kxk∞
for all small enough values of t. Since, by Lemma 3.4, the averages Pt(x) converge
a.u. as t → 0, it follows that the net {Bt(x)}t>0 is a.u. Cauchy as t → 0. Then,
repeating the proof of the Theorem 3.1, we obtain that for any x ∈ L1 the averages
{Bt(x)}t>0 converge a.u. to some bx ∈ L1 as t → 0.
Let pn be a trigonometric polynomial to satisfy condition (2) with ǫ = 1/n. If
(x)}t>0 are the corresponding averages from Lemma 3.4, then there is tn > 0
{P (n)
such that
t
(10)
kBt(x) − P (n)
t
(x)k1 ≤
1
t Z t
0
β(s) − p(s)ds kxk1 <
kxk1
n
for all 0 < t < tn.
By Lemma 3.4, P (n)
t
(x) → pn(0) x a.u., hence Bt(x) − P (n)
t
a.u., implying that
Bt(x) − P (n)
t
(x) → bx − pn(0) x
(x) → bx − pn(0) x in measure as t → 0.
10
VLADIMIR CHILIN AND SEMYON LITVINOV
Since the unit ball of L1 is closed in measure topology, (10) entails that
kbx − pn(0) xk1 ≤
kxk1
n
,
n→∞
Now, let 0 ≤ x ∈ L1, x 6= 0, en = {x ≤ n}, n ∈ N, and let xn = x en. It is clear
hence bx = k · k1 − lim
pn(0) x, and we conclude thatbx = α(x) x for some α(x) ∈ C.
that {xn} ⊂ L1 ∩ M and kx − xnk1 → 0 as n → ∞. As shown above, Bt(x) → bx
a.u. as t → 0 for some bx ∈ L1 and Bt(xn) → α(xn) xn as t → 0 for for every n and
some α(xn) ∈ C. Consequently, Bt(x) − Bt(xn) → bx − α(xn) xn in measure.
Besides,
kBt(x) − Bt(xn)k1 ≤
β(s)ds kx − xnk1 ≤ C kx − xnk1.
1
t Z t
0
Since the unit ball of L1 is closed in measure topology, it follows that
(11)
Choose k ∈ N such that xk = x ek 6= 0. If n > k, then we have
kbx − α(xn) xnk1 ≤ C kx − xnk1 → 0 as n → ∞.
so, (11) implies that
kbx ek − α(xn) xkk1 = kbx ek − α(xn) xn ekk1 ≤ kbx − α(xn) xnk1,
Therefore, there exists lim
n→∞
bx ek = k · k1 − lim
as n → ∞, hence bx = α(x) x, in view of (11).
n→∞
α(xn) xk.
If x ∈ L1, we employ the decomposition x = x1 − x2 + i(x3 − x4), 0 ≤ xi ∈ L1,
(cid:3)
i = 1, 2, 3, 4, and apply the above argument to each xi.
α(xn) = α(x), implying that kα(xn) xn − α(x) xk1 → 0
A weaker version of Theorem 3.2 - for b.a.u. convergence and with no identifica-
tion of the limit - was announced in [30]. However, some steps leading to the main
Theorem 3.1 of the paper, such as the fact given above in Proposition 3.1, were left
unjustified.
4. Extension to noncommutative fully symmetric spaces
Now we will extend the results of Section 3 to the noncommutative fully sym-
metric spaces E ⊂ L1 + L∞ with 1 /∈ E, in particular, to the spaces Lp, 1 < p < ∞.
Let x ∈ L0, and let {eλ}λ≥0 be the spectral family of projections for the absolute
value x = (x∗x)1/2 of x. If t > 0, then the t-th generalized singular number of x
(the non-increasing rearrangement of x) is defined as
(see [18]).
µt(x) = inf{λ > 0 : τ (e⊥
λ ) ≤ t}
Let µ be the Lebesgue measure on (0, ∞). It is well known that
Lp = Lp(M, τ ) =nx ∈ L0 :Z ∞
0
µt(x)pdµ(t) < ∞o,
and kxkp = kµt(x)kp, x ∈ Lp, 1 ≤ p < ∞ (see, for example, [18]).
Let (E, k · kE) be a symmetric function space on ((0, ∞), µ) (see, for example,
[22, Ch.II, § 4]). Define
E(M) = E(M, τ ) = {x ∈ L0 : µt(x) ∈ E}
NONCOMMUTATIVE WEIGHTEDINDIVIDUAL ERGODIC THEOREMSWITH CONTINUOUS TIME11
and set
It is shown in [21] that (E(M), k · kE(M)) is a Banach space and that conditions
kxkE(M) = kµt(x)kE, x ∈ E(M).
x ∈ E(M), y ∈ L0, µt(y) ≤ µt(x) for all t > 0
imply that y ∈ E and kykE ≤ kxkE, in which case (E(M), k · kE(M)) is said to be
a noncommutative symmetric space.
A noncommutative symmetric space (E(M), k · kE(M)) is called fully symmetric
if conditions
x ∈ E(M), y ∈ L0, Z s
0
µt(y)dt ≤Z s
0
µt(x)dt ∀ s > 0 (writing y ≺≺ x)
imply that y ∈ E and kykE ≤ kxkE. For example, Lp(M) = Lp, 1 ≤ p ≤ ∞, and
the Banach spaces
(L1 ∩ L∞)(M) = L1 ∩ M with kxkL1∩M = max{kxk1, kxk∞} and
(L1 + L∞)(M) = L1 + M with
kxkL1+M = inf(cid:8)kyk1 + kzk∞ : x = y + z, y ∈ L1, z ∈ M(cid:9) =Z 1
0
are noncommutative fully symmetric spaces (see [14]).
µt(x)dt
Since, given a symmetric function space E = E(0, ∞),
L1(0, ∞) ∩ L∞(0, ∞) ⊂ E ⊂ L1(0, ∞) + L∞(0, ∞),
with continuous embedding [22, Ch.II, § 4, Theorem 4.1], it follows that
L1(M) ∩ M ⊂ E(M) ⊂ L1(M) + M,
with continuous embedding.
Define
Rτ = {x ∈ L1 + M : µt(x) → 0 as t → ∞}.
It is known that Rτ is the closure of L1 ∩ M in L1 + M [14, Proposition 2.7],
in particular, (Rτ , k · kL1+M) is a noncommutative fully symmetric space [8]. In
addition, if τ (1) = ∞, then a symmetric space E(M, τ ) is contained in Rτ if and
only if 1 /∈ E(M, τ ) [8, Proposition 2.2].
Every noncommutative fully symmetric space E = E(M) is an exact interpo-
lation space for the Banach couple (L1(M), M) [13]. Therefore T (E) ⊂ E and
kT kE→E ≤ 1 for T ∈ DS; in particular,
T (Rτ ) ⊂ Rτ
and kT kL1+M→L1+M ≤ 1.
Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1, β : R+ → C
a Lebesgue measurable function with kβk∞ < C < ∞, and let Bt(x), x ∈ L1,
t > 0, be given by (3). We have β(t) = β1(t) − β2(t) + i (β3(t) − β4(t)), where
βj : R+ → R+ is Lebesgue measurable function such that kβjk∞ < C < ∞ for each
j = 1, . . . , 4. Denote
B(j)
t
(x) =
Then for each j,
1
t Z t
0
βj(s)Ts(x)ds, x ∈ L1, t > 0.
kB(j)
t
(x)k1 ≤ Ckxk1 ∀ x ∈ L1 and kB(j)
t
(x)k∞ ≤ Ckxk∞ ∀ x ∈ L1 ∩ L∞.
12
VLADIMIR CHILIN AND SEMYON LITVINOV
Consequently, C −1B(j)
is a positive absolute contraction in L1, which, by [5, Propo-
sition 1.1], admits a unique extension to a positive Dunford-Schwartz operator
D(j)
to the Banach space
L1 + L∞. By linearity, Bt admits a unique extension to L1 + L∞, which we will
also denote by Bt.
, t > 0, is the unique extension of B(j)
. Therefore, CD(j)
t
t
t
t
Since a noncommutative fully symmetric space E = E(M) is an exact interpo-
lation space for the Banach couple (L1, M), it now follows that Bt(E) ⊂ E and
kBtkE→E ≤ C for every t > 0.
Theorem 4.1. Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1, and
let β(t) be a bounded Besicovitch function such that kβ(t)k∞ < C. Then, given
Proof. Without loss of generality assume that x ≥ 0, and let {eλ}λ≥0 be the spectral
λdeλ. Then we
x ∈ Rτ , the averages (3) converge a.u. to some bx ∈ Rτ as t → ∞.
family of x. Given m ∈ N, denote xm =R ∞
1/m λdeλ and ym =R 1/m
Fix ǫ > 0. By Theorem 3.1, Bt(xm) → bxm ∈ L1 a.u. as t → ∞ for each m.
have 0 ≤ ym ≤ m−1 1, xm ∈ L1, and x = xm + ym for all m.
Therefore, there exists a sequence {em} ⊂ P(M) such that
0
Then it follows that, for for some t(m) > 0,
2m and k(Bt(xm) −bxm)emk∞ → 0
τ (e⊥
m) ≤
ǫ
as
t → ∞.
k(Bt(xm) − Bt′(xm))emk∞ <
1
m
∀ t, t′ ≥ t(m).
Since kymk∞ ≤ m−1, we have
k(Bt(x) − Bt′ (x))emk∞ ≤ k((Bt(xm) − Bt′ (xm))emk∞ + k(Bt(ym) − Bt′ (ym))emk∞
<
1
m
+ kBt(ym)emk∞ + kBt′(ym)emk∞ ≤
1 + 2C
m
for each m and all t, t′ ≥ t(m).
If e = Vm∈N
em, then
τ (e⊥) ≤ ǫ
and k(Bt(x) − Bt′ (x))ek∞ <
1 + 2C
m
∀ t, t′ ≥ t(m).
This means that {Bt(x)}t>0 is a Cauchy net with respect to a.u. convergence.
Since L0 is complete with respect to a.u. convergence [7, Remark 2.4], we conclude
the Fatou property [16, §4], its unit ball is closed in measure topology [15, Theorem
that the net {Bt(x)}t>0 converges a.u. to some bx ∈ L0.
As Bt(x) → bx a.u., it is clear that Bt(x) → bx in measure. Since L1 + M satisfies
4.1], and the inequality kBt(x)kL1+M ≤ C kxkL1+M implies that bx ∈ L1 + M. In
addition, Bt(x) → bx in measure implies that µs(Bt(x)) → µs(bx) almost everywhere
on ((0, ∞), µ) for each s > 0 (this can be shown as in the commutative case; see,
for example, [22, Ch.II, § 2, Property 11◦]).
Since C −1Bt ∈ DS+, we have C −1 µs(Bt(x)) = µs(C −1Bt(x)) ≺≺ µt(x). This
means that
C −1Z u
0
µs(Bt(x))ds =Z u
0
µs(C −1Bt(x))ds ≤Z u
0
µs(x)ds ∀ u > 0, t > 0.
NONCOMMUTATIVE WEIGHTEDINDIVIDUAL ERGODIC THEOREMSWITH CONTINUOUS TIME13
Z u
imply that
Now, Fatou property for L1((0, u), µ) and µs(Bt(x)) → µs(bx) in measure on ((0, u), µ)
µs(bx)ds ≤ C Z u
that is, µs(bx) ≺≺ µs(C x). Therefore, since Rτ is a fully symmetric space and
C x ∈ Rτ , it follows that bx ∈ Rτ .
The following is an application of Theorem 4.1 to fully symmetric spaces.
µs(x)ds =Z u
µs(C x)ds ∀ u > 0,
(cid:3)
0
0
0
Theorem 4.2. Let E = E(M, τ ) be a noncommutative fully symmetric space such
that 1 /∈ E, and let {Tt}t≥0 and β(t) be as in Theorem 4.1. Then, given x ∈ E, the
Proof. Since 1 /∈ E, it follows that E ⊂ Rτ . Thus, by Theorem 4.1, the averages
averages (3) converge a.u. to some bx ∈ E as t → ∞.
{Bt(x)}t>0 converge a.u. to some bx ∈ Rτ as t → ∞. Since µs(bx) ≺≺ µs(C x)
bx ∈ E.
(see the proof of Theorem 4.1) and E is a fully symmetric space, we conclude that
(cid:3)
Repeating the proofs of Theorems 4.1 and 4.2, we obtain the following extension
of Theorem 3.2.
Theorem 4.3. Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1, β(t)
be a bounded zero-Besicovitch function, and let E = E(M, τ ) be a noncommutative
fully symmetric space such that 1 /∈ E. Then, given x ∈ E, the averages (3)
converge a.u. to α(x) x as t → 0 for some α(x) ∈ C.
Remark 4.1. Employing the approach in the proof of [8, Theorem 5.7] (see also
[9]), one can see that the assertion of Theorem 4.2 (Theorem 4.3) remains valid
when the semigroup R+ is replaced by Rd
+, d ∈ N, so that the averages
Bt(x) =
β(s)Ts(x)ds, x ∈ E,
1
td Z[0,t]d
converge a.u. to some bx ∈ E as t → ∞ (respectively, to α(x) x as t → 0).
5. Applications to noncommutative Orlicz, Lorentz and
Marcinkiewicz spaces
Below we give applications of Theorems 4.2 and 4.3 to noncommutative Orlicz,
Lorentz and Marcinkiewicz spaces.
1. Let Φ be an Orlicz function, that is, Φ : [0, ∞) → [0, ∞) is left-continuous,
convex, increasing and such that Φ(0) = 0 and Φ(u) > 0 for some u 6= 0 (see, for ex-
0 λdeλ its spectral decomposition,
ample [17, Ch.2, §2.1]). If 0 ≤ x ∈ L0 and x =R ∞
one can define Φ(x) =R ∞
The noncommutative Orlicz space associated with (M, τ ) is the set
0 Φ(λ)deλ ∈ L0.
LΦ = LΦ(M, τ ) =nx ∈ L0 : Φ(a−1 x) ∈ L1 for some a > 0o.
The Luxemburg norm of an operator x ∈ LΦ is defined as
kxkΦ = inf(cid:8)a > 0 : kΦ(a−1 x)k1 ≤ 1(cid:9) .
It is known that (LΦ, k · kΦ) is a noncommutative fully symmetric space (see, for
example, [6, Corollary 2.2]). In addition,
LΦ =(cid:8)x ∈ L0 : µt(x) ∈ LΦ(0, ∞)(cid:9) and kxkΦ = kµt(x)kΦ ∀ x ∈ LΦ.
14
VLADIMIR CHILIN AND SEMYON LITVINOV
[6, Corollary 2.1]). The pair (LΦ, k · kΦ) is called the noncommutative Orlicz space
(see, for example [23]).
If τ (1) < ∞, then LΦ ⊂ L1. If τ (1) = ∞ and Φ(u) > 0 for all u 6= 0, then
Φ(a−1 1) /∈ L1 for each a > 0, hence 1 /∈ LΦ. Therefore, Theorems 4.2 and 4.3
imply the following.
Theorem 5.1. Let Φ be an Orlicz function such that Φ(u) > 0 for all u > 0.
Let {Tt}t≥0 ⊂ DS+ be a strongly continuous semigroup in L1, and let β(t) be a
bounded Besicovitch (zero-Besicovitch) function. Then, given x ∈ LΦ, the averages
same α(x) ∈ C).
(3) converge a.u. to some bx ∈ LΦ as t → ∞ (respectively, to α(x) x as t → 0 for
2. Let µ be the Lebesgue measure on the interval (0, ∞). Let L0(0, ∞) be the
algebra of (equivalence classes of) almost everywhere finite real-valued measurable
functions f on the measure space ((0, ∞), µ) with µ{f > λ} < ∞ for some λ > 0.
The non-increasing rearrangement of a function f ∈ L0(0, ∞) is the function f ∗ on
(0, ∞) defined by
(12)
f ∗(t) = inf{λ > 0 : µ{f > λ} ≤ t}
(see, for example, [22, Ch.II, § 2]). Let ψ be a concave function on [0, ∞) with
ψ(0) = 0 and ψ(t) > 0 for all t > 0, and let
Λψ(0, ∞) =(cid:26)f ∈ L0(0, ∞) : kf kΛψ =Z ∞
f ∗(t)dψ(t) < ∞(cid:27)
0
be the corresponding Lorentz space.
It is well known that (Λψ(0, ∞), k · kΛψ ) is a fully symmetric function space; in
addition, if ψ(∞) = lim
t→∞
ψ(t) = ∞, then 1 /∈ Λϕ(0, ∞) [22, Ch.II, § 5]).
Define the noncommutative Lorentz space (see, for example, [4]) as
Λψ = Λψ(M, τ ) =(cid:8)x ∈ L0 : µt(x) ∈ Λψ(0, ∞)(cid:9)
and set
kxkΛψ = kµt(x)kΛψ , x ∈ Λψ.
It is known [21] that (Λψ, k · kΛψ ) is a noncommutative fully symmetric space. If
τ (1) < ∞, then Λψ ⊂ L1. If τ (1) = ∞ and ψ(∞) = ∞, then it follows that 1 /∈ Λψ.
Therefore, Theorems 4.2 and 4.3 imply the following.
Theorem 5.2. Let ψ be a concave function on [0, ∞) with ψ(0) = 0 and ψ(t) > 0
for all t > 0, and let ψ(∞) = ∞. Let {Tt}t≥0 and β(t) be as in Theorem 5.1. Then,
given x ∈ Λψ, the averages (3) converge a.u.
(respectively, to α(x) x as t → 0 for same α(x) ∈ C).
to some bx ∈ Λψ(M, τ ) as t → ∞
3. Let ψ be as above, and let
Mψ(0, ∞) =(cid:26)f ∈ L0(0, ∞) : kf kMψ = sup
0<s<∞
1
ψ(s)Z s
0
f ∗(t)dt < ∞(cid:27)
be the corresponding Marcinkiewicz space. It is known that (Mψ(0, ∞), k·kMψ ) is a
ψ(t)
fully symmetric function space. Besides, 1 /∈ Mψ(0, ∞) if and only if lim
t = 0
t→∞
[22, Ch.II, § 5]). Also, if ψ(+0) > 0 and ψ(∞) < ∞, then Mψ(0, ∞) = L1(0, ∞) as
sets.
Define the noncommutative Marcinkiewicz space as
Mψ = Mψ(M, τ ) =(cid:8)x ∈ L0 : µt(x) ∈ Mψ(0, ∞)(cid:9)
NONCOMMUTATIVE WEIGHTEDINDIVIDUAL ERGODIC THEOREMSWITH CONTINUOUS TIME15
and set
kxkMψ = kµt(x)kMψ , x ∈ Mψ(M, τ )
(see, for example, [4]). It is known [21] that (Mψ, k · kMψ ) is a noncommutative
fully symmetric space. If τ (1) < ∞ or ψ(+0) > 0 and ψ(∞) < ∞, then Mψ ⊂ L1.
If τ (1) = ∞ and lim
t→∞
ψ(t)
t = 0, then 1 /∈ Mψ.
Thus, the corresponding version of Theorem 5.2 holds for Marcinkiewicz spaces
Mψ(M, τ ) if we replace condition ψ(∞) = ∞ by lim
t→∞
ψ(t)
t = 0.
6. The commutative case
In this section we present applications of Theorems 4.2 and 4.3 to fully symmetric
function spaces.
Let (Ω, ν) = (Ω, A, ν) be a Maharam measure space (see, for example, [24, Ch.1,
§ 1.2, Sections 1.1.7, 1.1.8]), that is,
(i) ν is a countably additive function defined on a σ-algebra A of subsets of a set
Ω with the values in the extended half-line [0, ∞];
(ii) if A ⊂ Ω and A ∩ F ∈ A for all F ∈ A with ν(F ) < ∞, then A ∈ A;
(iii) if A ⊂ F ∈ A and ν(F )=0, then A ∈ A;
(iv) for any A ∈ A there exists F ∈ A such that F ⊂ A and ν(F ) < ∞;
(v) the Boolean algebra bA of classes of ν-almost everywhere equal sets in A is
order complete.
Clearly, every complete σ-finite measure space is Maharam measure space.
It is well known that the ∗-algebra L∞(Ω, ν) of (equivalence classes of) essen-
tially bounded measurable complex-valued functions defined on a Maharam mea-
sure space (Ω, ν) is a commutative von Neumann algebra with a semifinite normal
commutative von Neumann algebra M with a semi-finite normal faithful trace τ
is ∗-isomorphic to the ∗-algebra L∞(Ω, ν) for some Maharam measure space (Ω, ν)
faithful trace ν(f ) = RΩ f dν, 0 ≤ f ∈ L∞(Ω, ν). The converse is also true: any
such that τ (f ) =RΩ f dν, 0 ≤ f ∈ L∞(Ω, ν) (see, for example, [12, Ch.7, § 7.3]).
If (Ω, ν) is a Maharam measure space and M = L∞(Ω, ν) with τ (f ) =RΩ f dν,
0 ≤ f ∈ L∞(Ω, ν), then L0 = L0(M, τ ) = L0(Ω, ν) is the algebra of (equivalence
classes of) almost everywhere finite measurable complex-valued functions on (Ω, ν)
with ν{f > λ} < ∞ for some λ > 0. The non-increasing rearrangement of a
function f ∈ L0(Ω, ν) is the function f ∗ on (0, ∞) defined by (12).
Let (E(0, ∞), k · kE) be a symmetric function space on ((0, ∞), µ). Let
and
E(Ω, ν) = E(M, τ ) =(cid:8)f ∈ L0(Ω, ν) : f ∗(t) ∈ E(0, ∞)(cid:9)
kf kE(Ω) = kf kE(M) = kf ∗(t)kE, f ∈ E(Ω, ν).
Then (E(Ω, ν), k · kE(Ω)) is a symmetric function space on (Ω, ν). Recall that this
space is a fully symmetric function space if conditions
f ∈ E(Ω, ν), g ∈ L0(Ω, ν), Z s
0
g∗(t)dt ≤Z s
0
f ∗(t)dt ∀ s > 0
imply that g ∈ E(Ω, ν) and kgkE(Ω) ≤ kf kE(Ω). Note that if ν(Ω) < ∞, then
E(Ω, ν) ⊂ L1(Ω, ν).
16
VLADIMIR CHILIN AND SEMYON LITVINOV
A net {fα} ⊂ L0(Ω, ν) converges almost uniformly (a.u.) to f ∈ L0(Ω, ν) if for
any ǫ > 0 there is a set A ∈ A such that τ (Ω \ A) ≤ ǫ and k(f − fα) χAk∞ → 0,
where χA is the characteristic function of A.
Now we can state the following corollary of Theorems 4.2 and 4.3.
Theorem 6.1. Let (Ω, ν) be a Maharam measure space. Let E = E(Ω, ν) be a fully
symmetric function space on (Ω, ν) such that 1 /∈ E if ν(Ω) = ∞. If {Tt}t≥0 ⊂ DS+
is a strongly continuous semigroup in L1(Ω, ν) and β(t) a bounded Besicovitch (zero-
Besicovitch) function, then, given f ∈ E, the averages (3) converge a.u. to some
bf ∈ E as t → ∞ (respectively, to α(f ) f as t → 0 for some α(f ) ∈ C).
References
[1] A. S. Besicovitch, Almost Periodic Functions, Cambridge Univ. Press, 1932.
[2] J. R. Baxter, J. H. Olsen, Weighted and subsequential ergodic theorems, Canad. J. Math.,
35 (1983), 145-166.
[3] A. Bellow, V. Lozert, The weighted pointwise ergodic theorem and the individual ergodic
theorem along sequences, Trans. Amer. Math. Soc., 288(1) (1985), 307-345.
[4] V. I. Chilin, A. M. Medzhitov, and F. A. Sukochev, Isometries of non-commutative Lorentz
spaces, Math. Z., 200 (1989), 527-545.
[5] V. Chilin, S. Litvinov, Ergodic theorems in fully symmetric spaces of τ -measurable operators,
Studia Math., 288(2) (2015), 177-195.
[6] V. Chilin, S. Litvinov, Individual ergodic theorems in noncommutative Orlicz spaces, Posi-
tivity, 21(1) (2017), 49-59. DOI 10.1007/s11117-016-0402-8.
[7] V. Chilin, S. Litvinov, and A. Skalski, A few remarks in noncommutative ergodic theory, J.
Oper. Th., 53(2) (2005), 331-350.
[8] V. Chilin, S. Litvinov, Individual ergodic theorems for semifinite von Neumann algebras,
arXiv:1607.03452v4 [math.OA].
[9] V. Chilin, S. Litvinov, Local ergodic theorems in symmetric spaces of measurable operators,
arXiv:1805.02614 [math.FA].
[10] N. Dang-Ngoc, A random ergodic theorem in von Neumann algebras, Pure Appl. Math. Sci.,
86 (1982), 605-608.
[11] C. Demeter and R. L. Jones, Besicovitch weights and the necessity of duality restrictions in
the weighted ergodic theorem, Chapel Hill Ergodic Theory Workshops, Contemp. Math. 356,
Amer. Math. Soc., Providence, RI, 2004, 127-135.
[12] J. Dixmier, Von Neumann Algebras, North-Holland Mathematical Library 27, North-Holland
Publishing Co., Amsterdam-New York, 1981.
[13] P. G. Dodds, T. K. Dodds, and B. Pagter, Fully symmetric operator spaces, Integr. Equat.
Oper. Th., 15 (1992), 942-972.
[14] P. G. Dodds, T. K. Dodds, and B. Pagter, Noncommutative Kothe duality, Trans. Amer.
Math. Soc., 339(2) (1993), 717-750.
[15] P. G. Dodds, T. K. Dodds, F. A. Sukochev, and O. Ye. Tikhonov, A Non-commutative
Yoshida-Hewitt theorem and convex sets of measurable operators closed locally in measure,
Positivity, 9 (2005), 457-484.
[16] P. G. Dodds, B. Pagter, The non-commutative Yosida-Hewitt decomposition revisited, Trans.
Amer. Math. Soc., 364(12) (2012), 6425-6457.
[17] G. A. Edgar, L. Sucheston, Stopping Times and Directed Processes, Cambridge University
Press, 1992.
[18] T. Fack, H. Kosaki, Generalized s-numbers of τ -measurable operators, Pacific. J. Math., 123
(1986), 269-300.
[19] E. Hensz, On some ergodic theorems for von Neumann algebras, in: Probability Theory and
Vector Spaces III, Lublin, August 1983, Lecture Notes in Math., Springer, 1983, 119-123.
[20] M. Junge, Q. Xu, Noncommutative maximal ergodic theorems, J. Amer. Math. Soc., 20(2)
(2007), 385-439.
[21] N. J. Kalton, F. A. Sukochev, Symmetric norms and spaces of operators, J. Reine Angew.
Math., 621 (2008), 81-121.
NONCOMMUTATIVE WEIGHTEDINDIVIDUAL ERGODIC THEOREMSWITH CONTINUOUS TIME17
[22] S. G. Krein, Ju. I. Petunin, and E. M. Semenov, Interpolation of Linear Operators, Transla-
tions of Mathematical Monographs, Amer. Math. Soc., 54, 1982.
[23] W. Kunze, Noncommutative Orlicz Spaces and Generalized Arens Algebras, Math. Nachr.,
147 (1990), 123-138.
[24] A. G. Kusraev, Dominated Operators, Springer, Netherlands, 2000.
[25] M. Lin, J. Olsen, and A. Tempelman, On modulated ergodic theorems for Dunford-Schwartz
operators, Illinois J. Math., 43(3) (1999), 542-567.
[26] S. Litvinov, Uniform equicontinuity of sequences of measurable operators and non-
commutative ergodic theorems, Proc. Amer. Math. Soc., 140(7) (2012), 2401-2409.
[27] S. Litvinov, Almost uniform convergence in noncommutative Dunford-Schwartz ergodic the-
orem, arXiv:1606.04501v2 [math.OA].
[28] S. Litvinov, Almost uniform convergence in noncommutative Dunford-Schwartz ergodic the-
orem, C. R. Acad. Sci. Paris, Ser. I 355 (2017), 977-980.
[29] F. Mukhamedov, M. Mukhamedov, and S. Temir, On multiparameter weighted ergodic the-
orem for noncommutative L
p-spaces, Math. Anal. Appl., 343 (2008), 226-232.
[30] F. Mukhamedov, A. Karimov, On noncommutative weighted local ergodic theorems on Lp-
spaces, Period. Math. Hungar., 55(2) (2007), 223-235.
[31] E. Nelson, Notes on non-commutative integration, J. Funct. Anal., 15 (1974), 103-116.
[32] G. Pisier, Q. Xu, Noncommutative Lp-spaces, Handbook of the geometry of Banach spaces,
2 (2003), 1459-1517.
[33] C. Ryll-Nardzewski, Topics in ergodic theory, Lecture Notes in Math., 472 (1975), Springer-
Verlag, 131-156.
[34] S. Sakai, C ∗-algebras and W ∗-algebras, Springer-Verlag, Berlin Heidelberg New York, 1971.
[35] I. E. Segal, A non-commutative extension of abstract integration, Ann. of Math., 57 (1953),
401-457.
[36] F. J. Yeadon, Non-commutative Lp-spaces, Math. Proc. Camb. Phil. Soc., 77 (1975), 91-102.
[37] F. J. Yeadon, Ergodic theorems for semifinite von Neumann algebras I, J. London Math.
Soc., 16(2) (1977), 326-332.
[38] K. Yosida, Functional Analysis, Springer Verlag, Berlin-Gottingen-Heidelberg, 1965.
National University of Uzbekistan, Tashkent, Uzbekistan
E-mail address: [email protected]; [email protected]
Pennsylvania State University, 76 University Drive, Hazleton, PA 18202, USA
E-mail address: [email protected]
|
1612.06256 | 2 | 1612 | 2018-03-18T12:14:52 | Triviality of Equivariant Maps in Crossed Products and Matrix Algebras | [
"math.OA"
] | We consider a "twisted" noncommutative join procedure for unital $C^*$-algebras which admit actions by a compact abelian group $G$ and its discrete abelian dual $\Gamma$, so that we may investigate an analogue of Baum-Dabrowski-Hajac noncommutative Borsuk-Ulam theory in the twisted setting. Namely, under what conditions is it guaranteed that an equivariant map $\phi$ from a unital $C^*$-algebra $A$ to the twisted join of $A$ and $C^*(\Gamma)$ cannot exist? This pursuit is motivated by the twisted analogues of even spheres, which admit the same $K_0$ groups as even spheres and have an analogous Borsuk-Ulam theorem that is detected by $K_0$, despite the fact that the objects are not themselves deformations of a sphere. We find multiple sufficient conditions for twisted Borsuk-Ulam theorems to hold, one of which is the addition of another equivariance condition on $\phi$ that corresponds to the choice of twist. However, we also find multiple examples of equivariant maps $\phi$ that exist even under fairly restrictive assumptions. Finally, we consider an extension of unital contractibility (in the sense of Dabrowski-Hajac-Neshveyev) "modulo $k$." | math.OA | math |
Triviality of Equivariant Maps in Crossed Products and Matrix Algebras
Benjamin Passer∗
Technion-Israel Institute of Technology
Haifa, Israel
[email protected]
Abstract
We consider a "twisted" noncommutative join procedure for unital C ∗-algebras which admit
actions by a compact abelian group G and its discrete abelian dual Γ, so that we may investigate
an analogue of Baum-D
abrowski-Hajac noncommutative Borsuk-Ulam theory in the twisted
'
setting. Namely, under what conditions is it guaranteed that an equivariant map φ from a
unital C ∗-algebra A to the twisted join of A and C ∗(Γ) cannot exist? This pursuit is motivated
by the twisted analogues of even spheres, which admit the same K0 groups as even spheres
and have an analogous Borsuk-Ulam theorem that is detected by K0, despite the fact that the
objects are not themselves deformations of a sphere. We find multiple sufficient conditions for
twisted Borsuk-Ulam theorems to hold, one of which is the addition of another equivariance
condition on φ that corresponds to the choice of twist. However, we also find multiple examples
of equivariant maps φ that exist even under fairly restrictive assumptions. Finally, we consider
abrowski-Hajac-Neshveyev) "modulo k."
an extension of unital contractibility (in the sense of D
'
Keywords: Borsuk-Ulam, twisted join, twisted suspension, free actions, torsion
MSC2010: 46L85, 46L65
1 Introduction
If X and Y are compact Hausdorff spaces which admit free actions of a nontrivial finite group G,
then continuous, equivariant maps from X to Y are severely restricted by dimension and homotopy
invariants. Such claims are rooted in the Borsuk-Ulam theorem, which concerns spheres Sk and the
free Z/2Z action x 7→ −x. In this setting, an equivariant map f is just an odd function, that is, a
function that satisfies f (−x) = −f (x) for each x in the domain.
Theorem 1.1 (Borsuk-Ulam Theorem). If n is a positive integer, then the following hold.
1. Every continuous function from Sn to Rn admits at least one pair of opposite points x and −x
with the same image.
2. Every continuous, odd function from Sn to Rn has a zero.
3. There is no continuous, odd function from Sn to Sn−1.
∗Partially supported by a Zuckerman Fellowship at the Technion, the Simons Foundation grant 346300, and the
Polish Government MNiSW 2015-2019 matching fund. This work is part of a project sponsored by EU grant H2020-
MSCA-RISE-2015-691246-QUANTUM DYNAMICS.
1
4. Every continuous, odd function from Sn−1 to Sn−1 is homotopically nontrivial.
5. Every continuous, odd function from Sn−1 to Sn−1 has odd degree.
Items 1-4 above are equivalent, and they follow from the stronger item 5. Generalizations and
consequences of the Borsuk-Ulam theorem abound in combinatorics and algebraic topology, in which
more general spaces, groups, and actions replace spheres, Z/2Z, and the antipodal action. More
recently, significant progress has been made in generalizing these results to noncommutative topology,
opening up many new potential problems. The reason for this is twofold: first, the behavior of
noncommutative C∗-algebras can be quite different than the picture painted by compact Hausdorff
topology, and second, the noncommutative setting allows discussion of compact quantum group
actions in addition to actions of compact groups. The aim of such results [22, 20, 1, 18, 4, 17, 16, 3, 6]
is to restrict the existence or homotopy properties of equivariant homomorphisms between unital
C∗-algebras. As such, the noncommutative join constructions in [5] play a major role in this story.
Definition 1.2 ([5]). Let A and B be unital C∗-algebras. Then the noncommutative join of A and
B is
A ⊛ B = {f ∈ C([0, 1], A ⊗min B) : f (0) ∈ C ⊗ B, f (1) ∈ A ⊗ C}.
If (H, ∆) is a compact quantum group and δ : A → A ⊗ H is a free coaction of H on A, then the
equivariant noncommutative join of A and H is
A ⊛δ H = {f ∈ C([0, 1], A ⊗min H) : f (0) ∈ C ⊗ H, f (1) ∈ δ(A)}.
Further, A ⊛δ H admits a free coaction δ∆, which applies id ⊗ ∆ pointwise on A ⊗min H.
In the above definition, freeness is meant in the sense of [8] (see also [2]). When H is equal
to C(G) for a compact group G, the two joins A ⊛ C(G) and A ⊛δ C(G) are isomorphic, and if
A ⊛ C(G) is given the diagonal action of G, the natural choice of isomorphism is equivariant for the
diagonal action and δ∆. That is, the equivariant join is needed precisely when the quantum group H
is not commutative, to avoid applying a diagonal action. Further, the iterated join ⊛
i=1C(Z/2Z) is
isomorphic to C(Sn−1), and the induced diagonal action (at each additional join) is just the antipodal
action of Z/2Z on C(Sn−1). The conjectures of Baum, D
abrowski, and Hajac in [1] therefore directly
'
generalize the Borsuk-Ulam theorem.
n
Conjecture 1.3. ([1, Conjecture 2.3]) Let A be a unital C∗-algebra with a free action δ of a
nontrivial compact quantum group (H, ∆). Also, let A ⊛δ H denote the equivariant noncommutative
join of A and H, with the induced action of (H, ∆) given by δ∆. Then both of the following hold.
1. There does not exist a (δ, δ∆)-equivariant, unital ∗-homomorphism from A to A ⊛δ H.
2. There does not exist a (∆, δ∆)-equivariant, unital ∗-homomorphism from H to A ⊛δ H.
We note that when H = C(Z/2Z), A ⊛δ C(Z/2Z) is equivariantly isomorphic to the unreduced
suspension of A, ΣA, discussed in [4]. More specifically, ΣA is the (non-equivariant) noncommutative
join A ⊛ C(Z/2Z), and the action presented in [4] is the diagonal action of Z/2Z.
Conjecture 1.3 type 1 holds when H is a compact quantum group with a torsion character
other than a counit, as in [6, Corollary 2.7], from a direct application of the case H = C(Z/kZ)
in [17, Corollary 2.4]. Conjecture 2 is false, and counterexamples exist for compact groups acting
on nuclear C∗-algebras from [3, Theorem 2.6]. However, there are certain key examples for which
the type 2 conjecture holds, as in [3, 10, 1, 6]. In [17], we proposed a different type of join (and
similarly, unreduced suspension) for C∗-algebras with free actions of Z/kZ, replacing the tensor
product with a crossed product. We generalize this definition and adopt new terminology/notation
to avoid confusion with Definition 1.2.
2
If Γ is a discrete abelian group and G is its compact abelian Pontryagin dual, then an action α
of G on A is equivalent to a grading of A by Γ, or a coaction δ : A → A ⊗ C∗(Γ) of the compact
quantum group C∗(Γ) = C(G). For γ ∈ Γ, which we may view as a character on G, the spectral
subspace
is written in coaction form as the γ-isotypic subspace
Aγ = {a ∈ A : αg(a) = γ(g)a for all g ∈ G}
Aγ = {a ∈ A : δ(a) = a ⊗ γ},
and the subspaces Aγ produces a grading of A. When γ = 1, Aγ is called the fixed-point subalgebra.
Moreover, freeness of the (co)action is equivalent to a saturation property
∀γ ∈ Γ,
1 ∈ AγA∗γ.
(1.4)
See [2, Theorem 0.4] for the equivalence of freeness and saturation in greater generality, as well as
[19, 11, 9] for some useful discussion of saturation properties in the context of group actions, and
[14] for related conditions when Γ is finite (but not necessarily abelian).
Definition 1.5. Let Γ be a discrete abelian group with G =bΓ. Suppose A is a unital C∗-algebra
and β is an action of Γ on A. Then the β-twisted join of A and C∗(Γ) ∼= C(G) is
J(A, β) = {f ∈ C([0, 1], A ⋊β Γ) : f (0) ∈ C∗(Γ), f (1) ∈ A}.
(1.6)
When Γ = Z/2Z, we call J(A, β) the β-twisted unreduced suspension of A and C∗(Z/2Z) ∼= C(Z/2Z).
If β is the trivial action, then J(A, β) is just A ⊛ C(G).
There is a natural grading of A ⋊β Γ by Γ, which extends to a grading of J(A, β) pointwise, that
places each γ ∈ Γ in the γ-isotypic subspace and each a ∈ A in the fixed-point subalgebra. This
grading is determined from the action bβ of the compact group G =bΓ on A ⋊β Γ, given by
bβg(a) = a, a ∈ A
bβg(γ) = g(γ), γ ∈ Γ.
If α is an action of G on A which commutes with β, then we extend α to A ⋊β Γ so that α fixes all
a0 + a1µ 7→ α(a0) − α(a1)µ.
a grading of A ⋊β Γ by C∗(Γ), generated by the following rule: if a is in the τ -isotypic subspace of
A from α, and γ ∈ Γ, then aγ = γβ(a) is in the τ γ-isotypic subspace of A ⋊β Γ. In particular, if
elements of the group Γ. The composition αbβ = bβα produces an action of G on A ⋊β Γ, and hence
Γ = Z/2Z is generated by µ, then αbβ is generated by the automorphism
Similarly, if Γ = Z/kZ is generated by µ and ω = e2πi/k, then αbβ is generated by the automorphism
β of Γ = bG on A, then let eα denote the pointwise application of the action αbβ on J(A, β).
In the non-twisted setting, when β is trivial, eα corresponds to the diagonal action of G on
A ⊛ C(G). Hence, if G = Z/kZ, α is free, and β is trivial, then there are no (α,eα)-equivariant unital
∗-homomorphisms from A to J(A, β), as seen in [17, Corollary 2.4]. To extend nonexistence results
about equivariant maps A → J(A, β) to cases where β is nontrivial, we know from [17, Example
3.7] that some assumption on β, or the equivariant map in question, is still necessary. Here we
consider two assumptions that insist β is not too far removed from the trivial action, in an attempt
to generalize known examples and theorems.
Definition 1.7. If α is an action of a compact abelian group G on A that commutes with an action
a0 + a1µ + . . . + ak−1µk−1
7→ α(a0) + ωα(a1)µ + . . . + ωk−1α(ak−1)µk−1.
3
Question 1.8. Suppose α and β are commuting actions of Z/kZ on a unital C∗-algebra A, and α
is free. Consider the following conditions on β.
1. The action β is not free.
2. The individual automorphisms of β are connected within Aut(A) to the trivial automorphism.
from A to J(A, β)?
Is either condition sufficient to guarantee that there are no (α,eα)-equivariant, unital ∗-homomorphisms
The conditions were determined through the computation of various examples, as in [17, 16], chief
among them odd-dimensional θ-deformed spheres (defined in [13, 15]) and twisted versions thereof. In
section 2 we consider more restrictive assumptions that guarantee nonexistence of equivariant maps
from A to J(A, β). However, neither condition of Question 1.8 is actually sufficient in general, as
shown in Theorems 3.1 and 3.3. Finally, in the remainder of section 3 we apply the usual embedding
A ⋊β Z/kZ ֒→ Mk(A) to see consequences of the aforementioned results for deforming the diagonal
inclusion A ֒→ Mk(A) to finite-dimensional and one-dimensional representations. Extending the
definition of unital contractibility to unital contractibility "modulo k", we find that the connection
between contractibility properties and examples of equivariant maps A → J(A, β) is not as direct
as in [6, Corollary 2.7].
2 Nonexistence
The noncommutative Borsuk-Ulam theorem in [17, Corollary 2.4] can be proved using an iteration
procedure, which is not as immediate in the twisted setting. Specifically, while morphisms A → B
induce morphisms A ⊛ C(G) → B ⊛ C(G), the same does not generally hold for twisted joins
without additional equivariance requirements. In this section, we consider sufficient conditions that
guarantee there are no equivariant morphisms A → J(A, β). To avoid unnecessary repretition, Γ
will always refer to a discrete abelian group, and G will be its compact abelian Pontryagin dual
groupbΓ.
If β is an action of Γ on A, then we may extend β to A ⋊β Γ so that each group element is in
the fixed-point subalgebra. We will also refer to this action, as well as its pointwise application on
J(A, β), by β.
Lemma 2.1. Suppose βA and βB are actions of Γ on unital C∗-algebras A and B. If φ : A → B is
a (βA, βB)-equivariant, unital ∗-homomorphism, then the rule
a ∈ A 7→ φ(a)
γ ∈ Γ 7→ γ
(2.2)
produces a homomorphism A ⋊βA Γ → B ⋊βB Γ. If Jφ : J(A, βA) → J(B, βB) denotes the pointwise
application of this map, then Jφ is (βA, βB)-equivariant.
If, in addition, αA and αB are actions of G =bΓ on A and B which commute with βA and βB,
respectively, and φ is also (αA, αB)-equivariant, then Jφ is pointwise (fαA,fαB)-equivariant.
Proof. The function ψ : A ⋊βA Γ → B ⋊βB Γ defined by (2.2) is a unital ∗-homomorphism by the
universal property of crossed products. The associated homomorphism
id ⊗ ψ : C([0, 1]) ⊗ (A ⋊βA Γ) 7→ C([0, 1]) ⊗ (B ⋊βB Γ)
respects the boundary conditions of the twisted join, as ψ(C∗(Γ)) = C∗(Γ) and ψ(A) ⊂ B. Therefore,
it induces a homomorphism Jφ : J(A, βA) → J(B, βB), which is pointwise (βA, βB)-equivariant by
design. Finally, the actions fαA and fαB are pointwise applications of αAcβA and αBcβB, so it suffices
to prove that ψ is (αAcβA, αBcβB)-equivariant. For g ∈ G,
4
(αBcβB)g(cid:16)ψ(cid:16)X aγ · γ(cid:17)(cid:17) = (αBcβB)g(cid:16)X φ(aγ) · γ(cid:17)
=X(αB)g(φ(aγ)) · g(γ)
=X φ((αA)g(aγ)) · g(γ)
= ψ(cid:16)X(αA)g(aγ) · g(γ)(cid:17)
= ψ(cid:16)(αAcβA)g(cid:16)X aγ · γ(cid:17)(cid:17) ,
so ψ is (αAcβA, αBcβB)-equivariant, and Jφ is (fαA,fαB)-equivariant.
The additional equivariance demanded in Lemma 2.1 is automatic if β is trivial. It follows that
the result below generalizes [17, Corollary 2.4].
Theorem 2.3. Let Γ = G = Z/kZ, k ≥ 2, and suppose A is a unital C∗-algebra with two commuting
actions α and β of Z/kZ, where α is free. Then there is no unital ∗-homomorphism φ : A → J(A, β)
Proof. We let A0 := A, which admits actions α0 := α and β. The twisted join A1 := J(A, β) admits
which is both (α,eα)-equivariant and (β, β)-equivariant.
the pointwise action of β (still denoted β), and we let α1 =fα0. Iterating this procedure, we define
An as an iterated twisted join of A via the rule
An = J(An−1, β),
αn = ]αn−1.
Suppose φ0 := φ : A0 → A1 is both (α0, α1)-equivariant and (β, β) equivariant. Then repeated
applications of Lemma 2.1 produce (αn−1, αn)-equivariant and (β, β)-equivariant maps φn : An−1 →
An. From composition of these maps in a chain, we find that for any n ∈ Z+, there is an (α0, αn)-
equivariant, (β, β)-equivariant map Φn : A0 → An.
Next, we apply a similar iteration procedure for maps into C∗(Z/kZ). Let B0 := C∗(Z/kZ)
be graded by Z/kZ in the usual way, inducing an action ρ0. Since φ0 : A0 → J(A0, β) is (α,eα)-
equivariant and (β, β)-equivariant, evaluation at the t = 0 endpoint of (1.6) shows that there is a
map ψ0 : A0 → B0 which is (α0, ρ0)-equivariant and (β, triv)-equivariant. Define
Bn = J(Bn−1, triv),
ρn = ]ρn−1,
and note that iteration of Lemma 2.1 once more establishes that for each n, there is a (αn, ρn)-
equivariant, (β, triv)-equivariant map ψn : An → Bn. Since Bn is defined using a trivial twist in
its iterated join procedure, we note that Bn ∼= ⊛
i=1 C(Z/kZ) = C((Z/kZ)∗n+1), and ρn is the
diagonal action.
n+1
N such that there exist a1, b1, . . . , aN , bN ∈ Aγ with P aib∗i
Finally, the composition ψn ◦ Φn : A → C(Z/kZ∗n) is (α, diag)-equivariant. Since the action α
is free, fix a generator γ ∈ Z/kZ and note that by (1.4), 1 ∈ AγA∗γ. In particular, there is a finite
invertible. On the other hand, from
the increasing connectivity of the iterated joins of Z/kZ, the number of elements in the γ-isotypic
subspace of C(Z/kZ∗n) required to generate an invertible grows without bound as n increases (see
[7] and [12, Definition 4.3.1 and Proposition 4.4.3]). It follows that for large n, the existence of the
(α, diag)-equivariant map ψn ◦ Φn : A → C((Z/kZ)∗n) leads to a contradiction.
Theorem 2.3 does not assume any condition on the freeness or non-freeness of β. However,
we note that the pointwise action β on J(A, β) is not free, as at the endpoint t = 0 of (1.6), β
corresponds to the trivial action on C∗(Z/kZ). Therefore, if the original action β on A is free, there
cannot be a (β, β)-equivariant unital ∗-homomorphism from A to J(A, β), regardless of any other
action α. It follows that Theorem 2.3 is useful when β is not free, so it may be safely viewed inside
the framework of Question 1.8, condition 1.
5
If A = C(X) and k ∈ Z+ is prime, then any non-free action of Z/kZ has a nonempty fixed point
set. When this set is an equivariant retract of X, where X is acted upon freely, we may produce a
twisted Borsuk-Ulam theorem as follows.
Proposition 2.4. Let k ∈ Z+ be prime, and let X be a compact Hausdorff space with two com-
muting Z/kZ actions α and β, where α is free and β is not. Let Y = Fix(β) 6= ∅ be equipped
with the restricted action γ = αY , and suppose there is an (α, γ)-equivariant continuous function
f : X → Y . Then there is no unital, (α,eα)-equivariant ∗-homomorphism from C(X) to J(C(X), β).
Proof. Suppose φ : C(X) → J(C(X), β) is (α,eα)-equivariant. First, we have that the dual map
f∗ : C(Y ) → C(X) to f : X → Y is (γ, α)-equivariant. Second, the restriction map C(X) → C(Y )
is certainly (α, γ)-equivariant and (β, triv)-equivariant, so it may be applied pointwise on the twisted
joins by Lemma 2.1. This gives
C(Y )
f ∗
−−−→(γ,α)
C(X)
φ
−−−→(α, eα)
J(C(X), β)
pointwise Y
−−−−−−−−→
( eα,eγ)
J(C(Y ), triv)
∼=−−−−−→(eγ,diag)
C(Y ) ⊛ C(Z/kZ).
The composition is (γ, diag)-equivariant, contradicting [17, Corollary 2.4] (or, rather, the topological
result [21, Corollary 3.1]).
The following proposition is motivated by a topological picture:
if A = C(X) is commutative
and α is a free action of Z/kZ which permutes the components of X, then for non-free actions β on
X, there might not exist equivariant maps A → J(A, β) due to the existence of finite-order unitaries.
The arguments depend upon a standard matrix expansion map
E : A ⋊β Z/kZ → Mk(A),
(2.5)
defined by mapping a ∈ A to the diagonal matrix with entries a, β1(a), . . . , βk−1(a) and mapping a
generator µ ∈ Z/kZ to a {0, 1}-valued matrix S which induces the shift en 7→ en+1 in the standard
basis of Ck. The action β on A ⋊β Z/kZ then corresponds to the entrywise application of β on the
subalgebra E(A ⋊β Z/kZ) ⊆ Mk(A).
Proposition 2.6. Fix k ≥ 2 and a generator µ of \Z/kZ ∼= Z/kZ. Suppose A is a unital C∗-algebra
with a free action α of Z/kZ, such that there is a unitary x in the µ-isotypic subspace with xk = 1.
Further, let β be an action of Z/kZ on A which commutes with α, such that the ideal I generated
by terms βm(a)− a, m ∈ Z/kZ, a ∈ A is proper, and the K0(A/I)-class of the unit 1 is not divisible
by k. Then there is no (α,eα)-equivariant unital ∗-homomorphism from A to J(A, β).
Proof. If φ : A → J(A, β) is (α,eα)-equivariant, then φ(x) determines a path of unitaries ut ∈
subspace of A ⋊β Z/kZ for αbβ, so u0 is of the form cµ for some c ∈ C with ck = 1, and u1 is in the
t = 1, connecting u0 ∈ C∗(Z/kZ) to u1 ∈ A. Now, ut is also in the µ-isotypic
µ-isotypic subspace of A for α. Let vt = c−1ut and define the projections
A ⋊β Z/kZ with uk
Pt =
1 + vt + v2
t + . . . + vk−1
k
t
for t ∈ [0, 1]. Apply the expansion map E of (2.5) and the entrywise quotient A → A/I, where I
identifies any a ∈ A with βm(a) for any m. Then the quotient of E(Pt) produces a path of projections
in Mk(A/I) connecting a matrix T ∈ Mk(C) to a diagonal matrix with entries a + I, a + I, . . . , a + I
Sk for an order k shift S ∈ Uk(C). Since
for some a ∈ A. The matrix T is of the form
S has eigenvalues 1, ω, . . . , ωk−1 for a primitive kth root of unity ω, it follows that T has rank
1, and T is equivalent in K0(A/I) to the unit 1. On the other hand, the diagonal matrix with
entries a + I, . . . , a + I is equivalent in K0(A/I) to the sum of a + I with itself k times. Therefore,
1 ∈ K0(A/I) is divisible by k.
k−1Pn=0
1
k
6
In Proposition 2.6, the assumptions imply that α is free and β is not free. Below we write a
commutative subcase of the result.
Corollary 2.7. Fix k ≥ 2 and let X be a disconnected compact Hausdorff space. Let α and β
be two commuting actions of Z/kZ on X such that Fix(β) is nonempty and there is a clopen set
Y ⊆ X such that Y ∪ α1(Y ) ∪ . . . ∪ αk−1(Y ) = X and the union is disjoint. Then there is no
(α,eα)-equivariant unital ∗-homorphism from C(X) to J(C(X), β).
Proof. Fix a primitive kth root of unity µ, and define a function f on X by assigning f (p) = µm
if and only if y ∈ αm(Y ). Then f ∈ C(X) is a unitary in the µ-isotypic subspace of α, such that
f k = 1. Since Z := Fix(β) is not empty, the ideal in C(X) generated by βm(a) − a, 0 ≤ m ≤ k − 1,
a ∈ C(X) is proper. Specifically, C(X)/I ∼= C(Z). Next, there is a map K0(C(Z)) → K0(C) ∼= Z
induced by evaluation at a point q ∈ Z, which sends the unit 1 ∈ C(Z) to 1, so 1 is indivisible by k
in K0(C(Z)). Finally, we may apply Proposition 2.6.
If k is prime, note that Fix(β) is empty if and only if β is free. Therefore, these results also fall
under the umbrella of Question 1.8, condition 1. Finally, following the K-theoretic computations in
[18, 16], we see that θ-deformed spheres, and certain twisted unreduced suspensions thereof, admit
twisted Borsuk-Ulam theorems. First, we recall the definitions, as in [13, 15, 16].
Definition 2.8. For an antisymmetric matrix θ ∈ Mn(R), let
C(S2n−1
θ
) := C∗(z1, . . . , zn zk normal, zkzj = e2πiθjk zjzk, z1z∗1 + . . . + znz∗n = 1)
denote the θ-sphere of dimension 2n − 1.
(n − 1) × (n − 1) minor of θ, let
If θ is such that zn is central, then for ρ the top left
C(S2n−2
ρ
) := C(S2n−1
θ
)/hzn − z∗ni
denoted the ρ-sphere of dimension 2n − 2. If θ is instead such that zn anticommutes with all of the
other zi, then let
R2n−2
ρ
:= C(S2n−1
θ
)/hzn − z∗ni
denote the "twisted" analogue of the ρ-sphere of dimension 2n− 2. Each of the above objects admits
an antipodal action of Z/2Z, denoted α, which negates each generator in the presentation. For
appropriate choices of ρ and ω, we have that
) ∼= C(S2n−3
) ⊛ C(Z/2Z) ∼= J(C(S2n−3
C(S2n−2
), triv)
ω
ω
ρ
and
R2n−2
ρ
∼= J(C(S2n−3
ω
), α).
In both cases, the antipodal action on the (2n−3)-dimensional ω-sphere induces the antipodal action
on the (twisted) join, in the usual way.
The antipodal action α is free, and both the trivial action and the antipodal action satisfy
condition 2 of Question 1.8.
Example 2.9. (Consequences of [18, Corollary 4.12] and [16, Theorem 1.8]) Let A be an algebra in
Definition 2.8 of dimension k, and let B be an algebra in Definition 2.8 of dimension k + 1. If Z/2Z
acts on A and B via the antipodal action, then there is no equivariant, unital ∗-homomorphism from
A to B.
Let C be another algebra in Definition 2.8 of dimension k, and suppose φ : A → C is equivariant.
If k is odd, then the K1-groups of both algebras are isomorphic to Z, and the induced map φ∗ on
K1 is nontrivial. If k is even, then the K0-groups of both algebras are isomorphic to Z ⊕ Z, with
the first component generated by the unit 1, and the induced map φ∗ on K0 is such that Ran(φ∗)
is not cyclic.
7
3 Existence
Recall the following assumptions of Question 1.8 for commuting Z/kZ-actions α and β on a unital
C∗-algebra A, where α is free.
1. The action β is not free.
2. The individual automorphisms of β are connected within Aut(A) to the trivial automorphism.
When we discuss continuous paths in Aut(A) or Hom(A, B), we will always mean continuous
with respect to the pointwise norm topology. Both conditions assert that β is in some sense similar to
the trivial action, and the conditions are motivated by examples and counterexamples from [17, 16]
and the previous section. In particular, condition 2 may be thought of as the demand that β is
equivariant maps A → J(A, β).
Theorem 3.1. Let A = C(S1) be generated by the coordinate unitary z, and equip A with the
"orientation-preserving." However, we find that neither condition is sufficient to rule out (α,eα)-
antipodal action α : z 7→ −z and the conjugation action β : z 7→ z∗ of Z/2Z. There is an (α,eα)-
equivariant, unital ∗-homomorphism from A to J(A, β). Since α is free and β is not free, condition
1 of Question 1.8 is insufficient.
Proof. Let z = x + iy, so that α negates both x and y, but β fixes x and negates y. Also, let
C∗(Z/2Z) be generated by the self-adjoint unitary µ. It follows that in A ⋊β Z/2Z, yµ = −µy and
xµ = µx. The points at, bt ∈ A ⋊β Z/2Z defined by
at = tx,
bt = ty +p1 − t2µ,
for t ∈ [0, 1] are self-adjoint, commute with each other, and satisfy
a2
t + b2
t = t2x2 +(cid:16)t2y2 + tp1 − t2yδ + tp1 − t2δy + (1 − t2)(cid:17)
= t2(x2 + y2) + (1 − t2)
= 1.
Since a0 = 0 and b0 = µ belong to C∗(Z/2Z), and similarly a1 = x and b1 = y belong to A, it
follows that f (t) = at + ibt is a unitary element in J(A, β). Further, eα(f ) = −f , since eα is defined
as the pointwise application of αbβ, which negates x, y, and µ in A ⋊β Z/2Z. Finally, the unital
∗-homomorphism φ : A → J(A, β) defined by φ(z) = f is (α,eα)-equivariant.
Next, consider the universal C∗-algebra
A ∼= C∗(x, y x = x∗, y = y∗, x2 + y2 = 1).
While A is itself noncommutative, there is an obvious surjection from A onto C(S1). Moreover, A
admits a Z/2Z action generated by
α :
x 7→ −x
y
7→ −y
,
(3.2)
analogous to the antipodal action on the quotient C(S1). Using a rotation argument motivated by
the commutative quotient, we find that the automorphism which generates α is connected within
Aut(A) to the trivial automorphism. Specifically, if s, t ∈ R have s2 + t2 = 1, then
(sx + ty)2 + (−tx + sy)2 = (s2x2 + stxy + styx + t2y2) + (t2x2 − stxy − styx + s2y2)
= (s2 + t2)(x2 + y2)
= 1,
8
so there is an endomorphism Rs,t : A → A defined by R(s,t)(x) = sx + ty and R(s,t)(y) = −tx + sy.
The inverse of R(s,t) is R(s,−t), as
s(sx − ty) + t(tx + sy) = (s2 + t2)x = x,
−t(sx − ty) + s(tx + sy) = (s2 + t2)y = y,
and similarly for the reverse composition. Therefore, R(s,t) is an automorphism for each (s, t) ∈ S1.
Finally, α1 = R(−1,0) is connected via a path R(s,t) ∈ Aut(A) to the trivial automorphism R(1,0).
Theorem 3.3. Let A = C∗(x, y x = x∗, y = y∗, x2 + y2 = 1), equip A with the action α of Z/2Z
from A to J(A, β). Since α is free and the automorphism generating β is connected within Aut(A)
to the trivial automorphism, condition 2 of Question 1.8 is insufficient.
that negates x and y, and let α = β. Then there is an (α,eα)-equivariant, unital ∗-homomorphism
Proof. Let C∗(Z/2Z) be generated by the self-adjoint unitary µ, and define the self-adjoint elements
at, bt ∈ A ⋊β Z/2Z by
√1 − t2
√2
µ.
at = tx +
µ,
bt = ty +
√1 − t2
√2
Since µ anticommutes with both x and y, it follows that
a2
t + b2
1 − t2
2 ! + t2y2 +
t = t2x2 +
= t2(x2 + y2) + (1 − t2)
= 1.
1 − t2
2 !
We also have that a0 = 1√2
µ = b0 ∈ C∗(Z/2Z), and similarly, a1 = x and b1 = y belong to
A. Therefore f (t) = at and g(t) = bt define elements f, g ∈ J(A, β). Since f and g are negated
∗-homomorphism.
by eα, the map φ : A → J(A, β) defined by φ(x) = f , φ(y) = g is an (α,eα)-equivariant, unital
Next, we expand upon some consequences of the two previous theorems.
Remark 3.4. In both Theorem 3.1 and Theorem 3.3, the chosen equivariant morphism A → J(A, β)
is such that evaluation at the t = 1 endpoint of the twisted join produces the usual embedding
A ֒→ A ⋊β Z/2Z. Further, since evaluation at t = 0 produces a map A → C∗(Z/2Z), which has a
commutative codomain, and the largest commutative quotient of A is C(S1), there is a factorization
A → C(S1)
Λ
−→ C∗(Z/2Z).
The morphism Λ is dual to a continuous function λ : Z/2Z → S1, i.e. a selection of two points in
S1. Since S1 is path connected, we may apply a path λt connecting λ = λ0 to λ1, which selects two
identical points. This produces a path connecting Λ to a one-dimensional representation A → C.
We conclude that the inclusion map A ֒→ A ⋊β Z/2Z may be connected to a one-dimensional
representation.
Remark 3.5. Let A be as in Theorem 3.3, so A again has C(S1) as its largest commutative
quotient. The commutator ideal of A is invariant, and the induced action on the quotient C(S1) is
the antipodal action. However, if α = β is the antipodal action on C(S1), then there is no (α,eα)-
equivariant unital ∗-homomorphism C(S1) → J(C(S1), β). In particular, J(C(S1), β) is a twisted
analogue of a 2-sphere that appears in Definition 2.8 and Example 2.9. Therefore, the failure of
Borsuk-Ulam theorems is not preserved in quotient procedures.
9
In the non-twisted setting, if a compact group G acts on A, then existence of an equivariant
morphism A → A ⊛ C(G) is equivalent to the existence of a path in Hom(A, A) between an equiv-
ariant endomorphism and a one-dimensional representation of A (see, e.g., the more general result
[6, Lemma 2.5]). Since the identity endomorphism is certainly equivariant for any action, it is of
perhaps independent interest whether or not idA may be connected to a one-dimensional represen-
tation, regardless of the presence of an action. If such a path does exist, then A is called unitally
contractible, as in [6, Definition 2.6]. In the commutative case A = C(X), this property corresponds
to contractibility of X. Moreover, it is immediate that if equivariant maps A → A ⊛ C(G) cannot
exist, then A cannot be unitally contractible. To consider analogous concepts in the twisted set-
ting, we extend the notion of unital contractibility "modulo k" by consideration of finite-dimensional
representations.
Definition 3.6. Let A be a unital C∗-algebra A. Then A is called unitally contractible modulo
k if the embedding a ∈ A 7→ a ⊗ Ik ∈ A ⊗ Mk(C) is connected within Hom(A, A ⊗ Mk(C)) to
a k-dimensional representation, that is, a map ρ ∈ Hom(A, C ⊗ Mk(C)). Similarly, A is called
strongly unitally contractible modulo k if the embedding a 7→ a ⊗ Ik may be connected within
Hom(A, A ⊗ Mk(C)) to a one-dimensional representation ρ ∈ Hom(A, C ⊗ C).
Example 3.7. Fix k ≥ 2. The matrix algebra Mk(C) admits a free Z/kZ action, and it is con-
tractible modulo k, but it is not strongly unitally contractible modulo k.
Proof. Let V be diagonal with entries 1, ω, . . . , ωk−1, where ω is a primitive kth root of unity.
Conjugation by V is a free Z/kZ action of Mk(C), as the matrix W which acts on the standard basis
of Cn by ei 7→ ei+1 is unitary and in the ω-isotypic subspace of the action. Further, the embeddings
id⊗ 1 and 1⊗ id of Mk(C) into Mk(C)⊗ Mk(C) ∼= Mk2 (C) are conjugate. That is, there is a unitary
U ∈ Mk2 (C) such that for each M ∈ Mk(C), U (M ⊗ 1)U∗ = 1 ⊗ M . Because the unitary group of
Mk2 (C) is path connected, the two embeddings are connected via M 7→ Ut(M ⊗ 1)U∗t , where U0 = I,
U1 = U , and each Ut is unitary. Therefore, Mk(C) is unitally contractible modulo k. However, Mk(C)
has no one-dimensional representations, so it cannot be strongly unitally contractible modulo k.
In analogy with [6, Corollary 2.7], we seek a connection between Borsuk-Ulam theorems and
(strong) unital contractibility of A modulo the same k. Certainly, if A has K-theory invariants
which remain nontrivial under the quotient G/kG, then A is not unitally contractible modulo k.
Example 3.8. The circle C(S1) is not unitally contractible modulo k for any k. This holds even
though Theorem 3.1 produces an equivariant map C(S1) → J(C(S1), β) such that evaluation at
t = 1 gives the usual inclusion A ֒→ A ⋊β Z/2Z. In particular, it is crucial for this example that β
is orientation-reversing.
We may, however, adapt the counterexample in Theorem 3.3 to show the C∗-algebra A used
therein is strongly unitally contractible modulo 2.
Example 3.9. The C∗-algebra A = C∗(x, y x = x∗, y = y∗, x2 + y2 = 1) is strongly unitally
contractible modulo 2.
a1
Proof. It is known from Theorem 3.3 that if α = β is the action generated by x 7→ −x, y 7→ −y,
evt=1(φ(a)) = a for each a ∈ A. Expanding the crossed product A ⋊α Z/2Z via E : a0 + a1µ 7→
then there is an (α,eα)-equivariant map φ : A → J(A, α). An examination of the proof shows that
α(a1) α(a0)(cid:19) then shows that the embedding
(cid:18) a0
ψ1 : a ∈ A 7→(cid:18)a
α(a)(cid:19) ∈ A ⊗ M2(C)
10
may be connected via a path ψt ∈ Hom(A, A⊗ M2(C)) to a homomorphism ψ0 such that Ran(ψ0) ⊆
E(C∗(Z/2Z)).
Both endpoints of the above path must be adjusted to show strong unital contractibility modulo
2. First, the automorphism generating the action α is connected in Aut(A) via a path αt ∈ Aut(A)
to α0 = id. Therefore
a ∈ A 7→(cid:18)a
αt(a)(cid:19) ∈ A ⊗ M2(C)
may be used to connect ψ1 to the diagonal embedding A ֒→ A ⊗ M2(C). Second, ψ0 maps A maps
to a subalgebra of M2(C) isomorphic to C∗(Z/2Z), which is commutative. Using the factorization
technique of Remark 3.4, we see that ψ0 may be connected to a one-dimensional representation.
Gluing all of the constructed paths together shows that A is indeed strongly unitally contractible
modulo 2.
Finally, we have seen that the nontriviality of equivariant maps proved in [17, Corollary 2.4] for
Z/kZ actions is removed when crossed products replace tensor products, or when a matrix algebra
is introduced, even under a fairly stringent assumption on the twist.
Acknowledgments
I am very grateful to Orr Shalit and Baruch Solel for their support during my current postdoc at
Technion-Israel Institute of Technology, and to the organizers of the Simons Semester at IMPAN.
References
[1] Paul F. Baum, Ludwik Dąbrowski, and Piotr M. Hajac. Noncommutative Borsuk-Ulam-type
conjectures. Banach Center Publ., 106:9 -- 18, 2015.
[2] Paul F. Baum, Kenny De Commer, and Piotr M. Hajac. Free actions of compact quantum
groups on unital C∗-algebras. Doc. Math., 22:825 -- 849, 2017.
[3] Alexandru Chirvasitu and Benjamin Passer. Compact group actions on topological and non-
commutative joins. arxiv:1604.02173. To appear in Proceedings of the AMS.
[4] Ludwik D
abrowski. Towards a noncommutative Brouwer fixed-point theorem. J. Geom. Phys.,
'
105:60 -- 65, 2016.
[5] Ludwik D
abrowski, Tom Hadfield, and Piotr M. Hajac. Equivariant join and fusion of non-
'
commutative algebras. SIGMA Symmetry Integrability Geom. Methods Appl., 11:Paper 082, 7,
2015.
[6] Ludwik D
abrowski, Piotr M. Hajac, and Sergey Neshveyev. Noncommutative Borsuk-Ulam-
'
type conjectures revisited. arxiv:1611.04130.
[7] Albrecht Dold. Simple proofs of some Borsuk-Ulam results. In Proceedings of the Northwestern
Homotopy Theory Conference (Evanston, Ill., 1982), volume 19 of Contemp. Math., pages 65 --
69. Amer. Math. Soc., Providence, RI, 1983.
[8] David Alexandre Ellwood. A new characterisation of principal actions. J. Funct. Anal.,
173(1):49 -- 60, 2000.
[9] Elliot C. Gootman, Aldo J. Lazar, and Costel Peligrad. Spectra for compact group actions. J.
Operator Theory, 31(2):381 -- 399, 1994.
11
[10] Piotr M. Hajac and Tomasz Maszczyk. Pullbacks and nontriviality of associated noncommuta-
tive vector bundles. arxiv:1601.00021. To appear in Journal of Noncommutative Geometry.
[11] Akitaka Kishimoto. Simple crossed products of C∗-algebras by locally compact abelian groups.
Yokohama Math. J., 28(1-2):69 -- 85, 1980.
[12] Jiří Matoušek. Using the Borsuk-Ulam Theorem. Universitext. Springer-Verlag, Berlin, 2003.
Lectures on topological methods in combinatorics and geometry, Written in cooperation with
Anders Björner and Günter M. Ziegler.
[13] Kengo Matsumoto. Noncommutative three-dimensional spheres. Japan. J. Math. (N.S.),
17(2):333 -- 356, 1991.
[14] Susan Montgomery. Hopf algebras and their actions on rings, volume 82 of CBMS Regional
Conference Series in Mathematics. Published for the Conference Board of the Mathematical
Sciences, Washington, DC; by the American Mathematical Society, Providence, RI, 1993.
[15] T. Natsume and C. L. Olsen. Toeplitz operators on noncommutative spheres and an index
theorem. Indiana Univ. Math. J., 46(4):1055 -- 1112, 1997.
[16] Benjamin Passer. Anticommutation in the presentations of theta-deformed spheres. J. Math.
Anal. Appl., 445(1):855 -- 870, 2017.
[17] Benjamin Passer. Free actions on C*-algebra suspensions and joins by finite cyclic groups.
Indiana Univ. Math. J., 67(1):187 -- 203, 2018.
[18] Benjamin W. Passer. A noncommutative Borsuk-Ulam theorem for Natsume-Olsen spheres. J.
Operator Theory, 75(2):337 -- 366, 2016.
[19] N. Christopher Phillips. Freeness of actions of finite groups on C∗-algebras.
In Operator
structures and dynamical systems, volume 503 of Contemp. Math., pages 217 -- 257. Amer. Math.
Soc., Providence, RI, 2009.
[20] Ali Taghavi. A Banach algebraic approach to the Borsuk-Ulam theorem. Abstr. Appl. Anal.,
pages Art. ID 729745, 11, 2012.
[21] A. Yu. Volovikov. Coincidence points of mappings of Z n
p -spaces. Izv. Ross. Akad. Nauk Ser.
Mat., 69(5):53 -- 106, 2005.
[22] Makoto Yamashita. Equivariant comparison of quantum homogeneous spaces. Comm. Math.
Phys., 317(3):593 -- 614, 2013.
12
|
1011.3739 | 2 | 1011 | 2010-11-22T13:51:13 | On positive definite preserving linear transformations of rank $r$ on real symmetric matrices | [
"math.OA"
] | We study on what conditions on $B_k,$ \ a linear transformation of rank $r$ \label{form} T(A)=\sum_{k=1}^r\tr(AB_k)U_k where $U_k,\ k=1,2,..., r$ are linear independent and all positive definite; is positive definite preserving. We give some first results for this question. For the case of rank one and two, the necessary and sufficient conditions are given. We also give some sufficient conditions for the case of rank $r.$ | math.OA | math |
www.emis.de/journals/BJMA/
ON POSITIVE DEFINITE PRESERVING LINEAR
TRANSFORMATIONS OF RANK r ON REAL SYMMETRIC
MATRICES
DOAN THE HIEU1 ∗ AND -HUYNH DINH TUAN2
Abstract. We study on what conditions on Bk, a linear transformation of
rank r
r
T (A) =
tr(ABk)Uk
(0.1)
Xk=1
where Uk, k = 1, 2, . . . , r are linear independent and all positive definite; is
positive definite preserving. We give some first results for this question. For
the case of rank one and two, the necessary and sufficient conditions are given.
We also give some sufficient conditions for the case of rank r.
1. Introduction
One of active topics in linear algebra is the linear preserver problems (LPPs)
involving linear transformations on matrix space that have special properties:
leaving some functions, subsets, relations . . . invariant. For more details about
LPPs: the history, the results and open problems we refer the reader to [2], [4],
[5], [8], and references therein.
On real symmetric or complex Hermitian matrices, the LPP of positive defi-
niteness is still open and seems to be complicated. In [10], we solved this problem
on real symmetric matrices with some additional assumptions.
In this paper, we consider this problem on real symmetric matrices based on
the rank of linear transformations. It is showed that a linear transformation T of
rank r preserving positive definiteness can be expressed in the following form
r
T (A) =
tr(ABk)Uk
Xk=1
where Uk, k = 1, 2, . . . , r are linear independent and all positive definite and
Bk, k = 1, 2, . . . , r are all symmetric.
Of course, any linear transformation of form (0.1) may be not positive definite
preserving in general. We address the question on what conditions on Bk, T of
form (0.1) is positive definite preserving and give some first results. For the case
of rank one and two, the necessary and sufficient conditions are given. We also
give some sufficient conditions for the case of rank r.
Date: Received: xxxxxx; Revised: yyyyyy; Accepted: zzzzzz.
∗ Corresponding author.
2010 Mathematics Subject Classification. Primary 15A86; Secondary 15A18, 15A04.
Key words and phrases. Linear preserver problems, Symmetric matrix, Positive definite.
1
2
D. T. HIEU, H. D. TUAN
2. Some basic lemmas
Lemma 2.1. For any A ∈ Sn(R) of rank r, there exists linear independent (pair-
i, ki ∈ {−1, 1}.
wise orthogonal) vectors x1, x2, . . . , xr such that A = Pr
Moreover if A is positive semi-definite, then A =Pr
Lemma 2.2. Let B ∈ Sn(R).
i=1 kixixt
xixt
i.
i=1
(1) If B is non-zero positive semi-definite, then for every positive definite
matrix A ∈ Sn(R), we have tr(AB) > 0.
(2) If B is not positive semi-definite, then there exists positive definite matrix
A ∈ Sn(R), such that tr(AB) < 0.
and {y1, y2, . . . , yr} are systems of linear independent vectors. Then,
i=1 xixt
i=1 yiyt
i, where {x1, x2, . . . , xn}
Proof.
(1) By Lemma 2.1, A =Pn
AB =Pn
iPr
i=1 xixt
i=1 yiyt
i =Pi,j xixt
tr(xixt
iyjyt
tr(AB) =Xi,j
iyjyt
i, B =Pr
j) =Xi,j
j. It is not hard to check that
hxi, yji2 ≥ 0.
Since {x1, x2, . . . , xn} is a basis of Rn, we have tr(AB) > 0.
(2) Suppose QBQt = D = diag(µ1, µ2, . . . , µn), where Q is an orthogonal ma-
trix. Since B is not positive semi-definite, there exists a diagonal matrix
C > 0, such that tr(CD) < 0 and let A = QtCQ. Obviously, A > 0 and
tr(AB) = tr(DC) < 0.
(cid:3)
The following lemma is well-known in the literature of the theory of quadratic
forms.
Lemma 2.3. Let A ∈ Sn(R) be positive definite and B ∈ Sn(R). Then there ex-
ists an invertible matrix W, such that W AW t = In and W BW t = diag(µ1, µ2, . . . , µn).
Remark 2.4. Since W A = (W t)−1, (W t)−1(A−1B)W t = diag(µ1, µ2, . . . , µn).
Thus, µ1, µ2, . . . , µn are eigenvalues of A−1B.
3. Positive definite preserving linear transformations of rank r
Let T : Sn(R) −→ Sn(R) be a linear transformation of rank r and {U1, U2, . . . , Ur}
is a basis of Im T. Suppose T (Eii) = Pr
bk
ji)Uk, bk
every A ∈ Sn(R) we can verify
ij = bk
ji, k = 1, 2, . . . r; i, j = 1, 2, · · · , n. Let Bk = (bk
k=1 bk
iiUk and T (Eij + Eji) = Pr
ij +
ij)n×n, then for
k=1(bk
T (A) =
r
Xk=1
tr(ABk)Uk.
Of course, a transformation of form (0.1) is linear. Moreover, if T is positive
definite preserving, then Uk, k = 1, 2, . . . , r can be chosen to be positive definite.
ON POSITIVE DEFINITE PRESERVING LINEAR TRANSFORMATIONS . . .
3
3.1. The case of rank one and two. By virtue of Lemma 2.2, the case of rank
1 is easy to prove.
Theorem 3.1. A linear transformation T of rank 1 is positive definite preserving
if and only if for every A ∈ Sn(R), T has the form.
T (A) = tr(AB)U,
(3.1)
where U > 0 and B is a non-zero positive semi-definite matrix.
In the rest of this subsection, we give the necessary and sufficient condition for
the case T is of rank 2.
Consider a linear transformation of rank 2 on Sn(R)
T (A) = tr(AB1)U1 + tr(AB2)U2,
(3.2)
where U1, U2 are linear independent and positive definite. By virtue of Lemma
2.3, there exists an invertible matrix W such that W U1W t = In and W U2W t =
diag(µ1, µ2, . . . , µn); µi > 0, i = 1, 2, . . . , n. Let µmin = min{µ1, µ2, . . . , µn} and
µmax = max{µ1, µ2, . . . , µn}, then we have
Theorem 3.2. The linear transformation (3.2) is positive definite preserving if
and only if B1 + µminB2 and B1 + µmaxB2 are non-zero and positive semi-definite.
Proof. Consider the linear transformation T1(A) = W T (A)W t. It is easy to see
that
T1(A) = tr(AB1)In + tr(AB2) diag(µ1, µ2, . . . , µn),
and T is positive definite preserving if and only if T1 is. Thus, we need only to
prove the theorem for T1. First we observe that
tr(AB1)In + tr(AB2) diag(µ1, µ2, . . . , µn) > 0
⇔ tr(AB1) + tr(AB2)µi > 0, ∀i = 1, 2, . . . , n
⇔ tr(AB1) + tr(AB2)µmin > 0, tr(AB1) + tr(AB2)µmax > 0.
(3.3)
But, the fact that (3.3) is equivalent to B1 + µminB2 > 0 and B1 + µmaxB2 > 0,
is followed by Lemma 2.2.
(cid:3)
3.2. The case of rank r. With a proof similar to the above, we have the fol-
lowing
Theorem 3.3. Let T be the linear transformation of form (3.1), where Ui, i =
1, 2, . . . , Ur are linear independent and positive definite. Furthermore, suppose
that there exists an invertible matrix W, such that W UkW t = diag(λ1k, λ2k . . . , λnk), i =
k=1 λikBk, i =
1, 2, . . . r, then T is positive definite preserving if and only if Pr
1, . . . , n are all non-zero and positive semi-definite.
By virtue of Lemma 2.2, we have
Theorem 3.4. Consider a linear transformation of form (0.1), where Uk > 0. If
Bk ≥ 0, i = 1, 2, . . . , r and are all non-zero, then T is positive definite pereserving.
4
D. T. HIEU, H. D. TUAN
For a matrix X ∈ Sn(R), denote by λmin(X), λmax(X) the smallest and largest
eigenvalues of X, respectively while λm-m(X) can be λmin(X) or λmax(X). The
following is a sufficient condition for a linear transformation of form (0.1) to be
positive definite preserving.
Theorem 3.5. Let T be the linear transformation of form (0.1), where Ui, i =
k=1 λm-m(Uk)Bk are
all non-zero and positive semi-definite, then T is positive definite preserving.
1, 2, . . . , Ur are linear independent and positive definite. If Pr
Proof. By the assumption and Lemma 2.2, we have Pr
for any positive definite matrix A. But,
r
k=1 tr(ABk)λm-m(Uk) > 0
λmin[tr(ABk)Uk]
(3.4)
λmin" r
Xk=1
tr(ABk)Uk# ≥
Xk=1
≥ min{
The theorem is proved.
r
Xk=1
tr(ABk)λm-m(Uk)} > 0.
(3.5)
(cid:3)
References
[1] Cao, C. and Tang, X., Determinant Preserving Transformations on Symmetric matrix
Spaces, Electronic Journal of Linear Algebra, Vol.11, 205-211 (2004).
[2] Horn, R. A. and Johnson, C. R., Topics in Matrix Analysis, Cambridge University Press,
Cambridge (1991).
[3] Johnson, C., Matrix Theory and Applications, American Mathematical Society, 1990.
[4] Li, C. K. and Tsing, N. K., Linear preserver problems: a brief introduction and some
special techniques, Directions in matrix theory (Auburn, AL, 1990). Linear Algebra Appl.
162/164 217-235 (1992).
[5] Li, C.-K. and Pierce, S., Linear preserver problems, Amer. Math. Monthly, 108: 591-605
(2001).
[6] Loewy, R., Linear maps which preserve a balanced nonsingular inertia class, Linear Algebra
Appl. 134:165-179 (1990).
[7] Loewy, R., Linear maps which preserve an inertia class, Siam J. Matrix Anal. Appl. 11:107-
112 (1990).
[8] Loewy, R., A survey of linear preserver problems-chapter 3: Inertia preservers, Linear
Multilinear Algebra 33: 2230 (1992).
[9] Pierce, S. and Rodman, L., Linear Preservers of the class of Hermitian matrices with
balanced inertia, Siam J. Matrix Anal. Appl. Vol. 9, No. 4, 461-472 (1988).
[10] Huynh Dinh Tuan-Tran Thi Nha Trang-Doan The Hieu, Positive definite preserving linear
transformations on symmetric matrix spaces, preprint, http://arxiv.org/abs/1008.1347.
1 Hue Geometry Group, College of Education, Hue University; 34 Le Loi, Hue,
Vietnam.
E-mail address: [email protected]
2 Hue Geometry Group, College of Education, Hue University; 34 Le Loi, Hue,
Vietnam.
E-mail address: [email protected]
|
1908.00770 | 2 | 1908 | 2019-09-10T08:19:16 | Classification of tiling $C^*$-algebras | [
"math.OA"
] | We prove that Kellendonk's $C^*$-algebra of an aperiodic and repetitive tiling with finite local complexity is classifiable by the Elliott invariant. Our result follows from showing that tiling $C^*$-algebras are $\mathcal{Z}$-stable, and hence have finite nuclear dimension. To prove $\mathcal{Z}$-stability, we extend Matui's notion of almost finiteness to the setting of \'etale groupoid actions following the footsteps of Kerr. To use some of Kerr's techniques we have developed a version of the Ornstein-Weiss quasitiling theorem for general \'etale groupoids. | math.OA | math |
CLASSIFICATION OF TILING C ∗-ALGEBRAS
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
Abstract. We prove that Kellendonk's C ∗-algebra of an aperiodic and repetitive
tiling with finite local complexity is classifiable by the Elliott invariant. Our result
follows from showing that tiling C ∗-algebras are Z-stable, and hence have finite
nuclear dimension. To prove Z-stability, we extend Matui's notion of almost finite-
ness to the setting of ´etale groupoid actions following the footsteps of Kerr. To
use some of Kerr's techniques we have developed a version of the Ornstein-Weiss
quasitiling theorem for general ´etale groupoids.
1. Introduction
Due to the recent completion of the classification program for simple C ∗-algebras
with finite nuclear dimension [12,18,53], the time is ripe to look for naturally occurring
In this paper, we consider C ∗-algebras
examples to which this program applies.
associated to repetitive aperiodic tilings with finite local complexity (FLC). A tiling
is a covering of Rd by a collection of labelled compact subsets, called tiles, that only
meet on their boundary. Aperiodic tilings give rise to two C ∗-algebras: a crossed
product C(Ω) ⋊ Rd and a quantised algebra that arises from an abstract transversal
to the translation action on the tiling space Ω. The latter algebra is due to Kellendonk
[25], and was developed to give a new picture (up to stable isomorphism) of algebras
studied by Bellissard in order to understand the spectrum of Schodinger operators
with aperiodic potential [2].
Tiling algebras are particularly well-suited to the classification program since their
Elliott invariant is well understood [1,14,17,25,26,43,49]. Our strategy is in line with
the most recent techniques developed for the classification of C ∗-algebras coming
from minimal actions of general amenable groups acting on compact metric spaces,
studied by Kerr and others [6,27]. The main result of this paper is the following.
Theorem 1 (c.f. Theorem 8.4). The class of C ∗-algebras consisting of those which
are associated to any aperiodic and repetitive tiling with FLC is classified by the Elliott
invariant.
The classification of tiling algebras is highly relevant to mathematical physics.
Bellissard showed that gaps in the spectrum of Schrodinger operators with aperiodic
potential are topological in nature in the sense that they are unchanged under per-
turbation of the Hamiltonian [2]. In particular, he proved that these gaps are labelled
by the range of the trace of K-theory elements of a crossed product tiling C ∗-algebra
that is strongly Morita equivalent to Kellendonk's algebra. Standard results from
the classification program for simple C ∗-algebras then imply that the isomorphism
class of the C ∗-algebra Bellissard associated to a Schrodinger operator is completely
determined by the operator itself.
This research was partially supported by EPSRC grants EP/R013691/1, EP/R025061/1,
EP/N509668/1 and EP/M506539/1.
1
2
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
Before discussing the specific techniques used in this paper, we wish to put our
methods and results into the context of the classification program for simple nuclear
C ∗-algebras. The classification program may be regarded as a C ∗-analogue of the
Connes-Haagerup classification of injective factors. It has its origin in Elliott's clas-
sification of AF-algebras by K-theory from the 1970s [10] but was put forward as a
conjecture only in the late 1980s [11]. Whilst initially regarded as very speculative,
an outburst of subsequent intensive research produced a great deal of positive evi-
dence. Around 1995, Kirchberg and Phillips proved the Elliott conjecture for simple
nuclear purely infinite C ∗-algebras (now known as Kirchberg algebras) satisfying the
UCT [38]. A key role in this classification was played by Kirchberg's absorption re-
∼= A.
sult [30,31], which says that A is a Kirchberg algebra if and only if A ⊗ O∞
In 2003, Rørdam [44] proved that there are infinite simple nuclear C ∗-algebras that
fail the Elliott conjecture. Subsequently, Toms discovered examples with no infinite
projections that the Elliott invariant cannot classify [54,55].
Unlike in the von Neumann algebra case, where there are only very few injective
factors with traces - one finite and one semifinite such factor - the classification of sim-
ple nuclear C ∗-algebras with traces poses a major challenge. Inspired by Kirchberg's
absorption result, Jiang and Su constructed a simple unital nuclear C ∗-algebra Z
with unique trace which is KK-equivalent to the complex numbers [24]. They proved
that A ⊗ Z ∼= A for simple AF-algebras and this was subsequently extended to large
classes of other examples. Since A and A ⊗ Z have the same Elliott invariants (under
mild assumptions), Z-stability appears as a minimal classifiability requirement. It
was becoming more and more clear that a new regularity hypothesis was necessary
to continue Elliott's classification program.
The necessary regularity condition came in 2010 when Winter and the third named
author introduced nuclear dimension in [60] (thanks to Winter's handwriting, it was
also known as unclear dimension). This is a non-commutative dimension theory
which builds on previous definitions by Kirchberg and Winter but which is finite for
purely infinite simple C ∗-algebras.
It turned out that simple C ∗-algebras without
traces have finite nuclear dimension if and only if they are purely infinite. Hence,
finiteness of nuclear dimension captures the correct classifiability condition for infi-
nite C ∗-algebras. Moreover, Winter proved that finite nuclear dimension for simple
C ∗-algebras implies Z-stablity, as well as far reaching regularity properties of their
Cuntz semigroup [58]. Shortly afterwards, Toms and Winter conjectured that for
simple nuclear C ∗-algebras, finiteness of nuclear dimension, Z-stablity, and strict
comparison, a regularity condition for the Cuntz semigroup of the algebra, should all
be equivalent.
After many partial results, it is now known by the fundamental work of [4] that
finiteness of nuclear dimension and Z-stablity are indeed equivalent. To establish
classifiability of concrete examples one is therefore free to either to find an upper
bound for the nuclear dimension or to show Z-stablity. Whilst it initially appeared
difficult to prove Z-stablity directly, there are now relatively easy methods to do so
thanks to Hirshberg and Orovitz's application [21] of the breakthroughs of Matui and
Sato [34].
Kellendonk's tiling algebras are defined using a principal ´etale groupoid with com-
pact and totally disconnected unit space. The notion of almost finiteness for second
countable ´etale groupoids with compact and totally disconnected unit spaces was
introduced by Matui in [33]. The definition was designed as a weakening of the AF
CLASSIFICATION OF TILING C ∗-ALGEBRAS
3
(approximately finite) property of groupoids, motivated by the fact that AF ´etale
groupoids with totally disconnected unit spaces were already known to be completely
classified up to isomorphism [40]. In [50], Suzuki extended the definition to general
´etale groupoids.
Kerr introduced the notion of almost finiteness for free actions of groups on com-
pact metric spaces in [27]. This complements Matui's original definition in the sense
that the transformation groupoids arising from such actions are almost finite. The
motivation for the definition in the group case is to provide a dynamical substitute
for the properties of Z-stability from the topological setting and hyperfiniteness from
the measure-theoretic setting.
It is shown in [27] that the definition accomplishes
this goal; almost finite, free, minimal actions of infinite groups on compact metris-
able spaces of finite covering dimension have Z-stable crossed product C ∗-algebras.
Therefore, almost finiteness should be viewed as a dynamical analogue of Z-stability,
at least for amenable groups acting on compact metric spaces.
It then becomes natural to ask to what extent the methods of [27] apply to actions
of groupoids. The present work explores this question, generalising the foundations
behind the classification result in [27] to the ´etale groupoid case and applying the
new machinery to the groupoid arising from an aperiodic, repetitive tiling satisfying
FLC. As part of this development, we extend the Ornstein-Weiss quasitiling theorem
to ´etale groupoids, which could play into a more general Z-stability result for ´etale
groupoids.
Classification of tiling algebras was conjectured in [39], and several subclasses of
tiling algebras were subsequently shown to be classifiable. Our approach extends and
unifies these previous results, which we now outline. The most general classification
result prior to ours classifies all rational tilings, those whose tiles are polyhedra with
all vertices lying on rational coordinates. For such tilings, [45, Lemma 5] implies
that Kellendonk's algebra is isomorphic to C(Ωsq) ⋊ Zd, where Ωsq is the hull of a
tiling where all prototiles are cubes. Such C ∗-algebras were classified by Winter in
[59, Corollary 3.2] using the nuclear dimension estimates of Szab´o [51, Theorem 5.3].
Subsequently, Deeley and Strung [8] proved that the stable and unstable algebras
of Smale spaces containing projections are classifiable using the dynamic asymptotic
dimension of Guentner, Willett and Yu [19]. Such Smale spaces include all substitu-
tion tilings satisfying our standard hypotheses. We note that the work in this paper
was originally inspired by Deeley and Strung's results. We had hoped that Deeley
and Strung's techniques would extend to all aperiodic tilings, however, we were not
able to prove that certain subgroupoids along so called "fault lines" in tilings were
relatively compact. Finally, [22, Example 9.6] uses a Rohklin dimension argument to
prove that all one-dimensional tilings with our standard hypotheses are classifiable.
The paper is organised as follows. In Section 2, we present a few preliminary facts
about groupoids, and establish notation. We then explore groupoid actions, and
introduce a new, but associated, notion of almost finiteness. This generalises Kerr's
almost finiteness from the group case, and links back to Matui's original definition for
groupoids [33]. In section 3, we prove an analogue of the Ornstein-Weiss quasitiling
theorem for ´etale groupoids. In Section 4, we review the theory of crossed products
arising from groupoid actions. Sections 5 and 6 contain a brief introduction to tilings,
and the construction of tiling groupoids and their C ∗-algebras. We show that these
tiling groupoids are almost finite (in the sense of [33]) in Section 7. Following the
results of Section 2, we are then able to show that tiling groupoids arise from almost
4
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
finite groupoid actions, and we spend some time constructing the associated structure
concretely. Section 8 contains our main result on the Z-stability of tiling C ∗-algebras.
We give a direct proof that tiling C ∗-algebras are quasidiagonal in Section 9, along
with some examples of tilings whose C ∗-algebras have unique trace, and so benefit
from the more direct route to classification that this property provides.
We note that the results in this paper made up the bulk of the first author's PhD
thesis [23], which also contains extensive background information on various aspects
of this article.
Acknowledgements. We are grateful to Charles Starling, Gabor Szab´o and Stuart
White for their helpful comments and mathematical insights.
2. Almost finiteness for groupoid actions
For an introduction to ´etale groupoids and their C ∗-algebras, see [48]. We typically
denote groupoids by G, with unit space G(0) and r, s : G → G(0) the range and source
maps of G. A subset V of a groupoid G is said to be a G-set if both the range and
source maps are injective on V . For A ⊂ G and g ∈ G, let
Ag := {ag s(a) = r(g)},
so that if r(g) /∈ s(A), we obtain Ag = ∅. Similarly, for subsets A, B ⊂ G, we denote
AB := {ab a ∈ A, b ∈ B, s(a) = r(b)}.
This will be particularly useful when we consider A ⊂ G and u ∈ G(0), and obtain
the simplified notation
and
Au := A ∩ Gu := {a ∈ A s(a) = u}
uA := A ∩ Gu := {a ∈ A r(a) = u}.
A topological groupoid is ´etale if the range and source maps are local homeomor-
phisms, and is ample if it is ´etale and G(0) is zero dimensional. A bisection is a subset
B of a topological groupoid such that there exists an open set U containing B such
that r : U → r(U) and s : U → s(U) are homeomorphisms. Since the topology
on any ´etale groupoid has a base of open bisections, when G is ´etale and A ⊂ G is
compact, there exists N ∈ N such that Au < N for all u ∈ G(0). We will use this
fact frequently.
We introduce the notion of almost finiteness for actions of groupoids, generalising
the definition for group actions [27, Definition 8.2]. We first make precise what it
means for a groupoid to act on a set. To the best of our knowledge, the following
notion first appears in [9]. The exact formulation presented here appears as [16,
Definition 1.55].
Definition 2.1. Let G be a groupoid and X a set. We say that G acts (on the
left) of X if there is a surjection rX : X → G(0) and a map (g, x) 7→ g · x from
G ∗ X := {(g, x) ∈ G × X s(g) = rX(x)} to X such that
(i) if (h, x) ∈ G ∗ X and (g, h) ∈ G(2), then (g, h · x) ∈ G ∗ X and
(ii) rX(x) · x = x for all x ∈ X.
g · (h · x) = gh · x; and
CLASSIFICATION OF TILING C ∗-ALGEBRAS
5
We will usually drop the dot from the notation and refer to the image of x ∈ X
under g ∈ G by gx. We will preserve the dot in the case that X = G(0), to distinguish
between groupoid multiplication gx, and the action g · x. For W ⊂ X and S ⊂ G,
we denote
S · W := {g · x g ∈ S, x ∈ W and s(g) = rX(x)}.
Again, we may drop the dot to denote this by SW .
We make the following simple observation.
Lemma 2.2. Let G y X be any groupoid action. If (g, x) ∈ G ∗ X, then rX(gx) =
r(g).
Proof. Since (g, x) ∈ G∗X and (g−1, g) ∈ G(2), we have (g−1, gx) ∈ G∗X by condition
(i) in Definition 2.1, so that r(g) = s(g−1) = rX(gx).
(cid:3)
Definition 2.3. Let G be a topological groupoid which acts on a topological space
X.
(i) ([16, Definition 1.60]) We say that the action is continuous if the maps rX :
X → G(0) and (g, x) 7→ gx from G ∗ X → X are continuous.
(ii) We say that the action is free if, for every x ∈ X, gx = x implies g = rX(x).
(iii) We say that the action is minimal if, for every x ∈ X, the subset {gx g ∈ G}
is dense in X.
We remark that when G acts continuously on a compact space X, then G(0) arises
as the continuous image of the compact space X under the surjection rX : X → G(0).
It is therefore necessary that G(0) be compact in this situation.
Lemma 2.4. Let G be a locally compact Hausdorff ´etale groupoid acting continuously
on a locally compact Hausdorff space X. Then the range map rX : X → G(0) is open.
Proof. By [16, Proposition 1.72], G ⋉ X is ´etale. Suppose that U ⊂ X is open. We
identify U ⊂ X with V = {(rX(u), u) u ∈ U} ⊂ (G ⋉ X)(0). Since U is open in
X and X ∼= (G ⋉ X)(0) is open in G ⋉ X, it follows that V ∼= U is open in G ⋉ X.
Observe that rX(U) = πG(V ), where πG is projection to the G-coordinate on G × X.
Since πG is an open map, this shows that rX(U) is open in G.
(cid:3)
The following notion of dynamical comparison is a generalisation of [27, Defini-
tion 3.1] to the groupoid setting.
Definition 2.5. Let G y X be a groupoid action, and let A, B ⊂ X. We write
A ≺ B if, for every closed C ⊂ A, there exist a finite collection U of open subsets of
X which cover C, and a subset SU ⊂ G for each U ∈ U such that rX(U) ⊂ s(SU ) so
that the collection {tU U ∈ U, t ∈ SU } consists of pairwise disjoint subsets of B.
We now present a construction of a groupoid from a groupoid action. Such
groupoids will be vitally important later in the paper.
Definition 2.6. Let G be a groupoid acting on a set X. The associated transfor-
mation groupoid G ⋉ X is the set G ∗ X with the following structure. The set of
composable pairs is
(G ⋉ X)(2) = {((g, x), (h, y)) ∈ (G ⋉ X) × (G ⋉ X) h · y = x}.
The product of such a pair is given by (g, x)(h, y) = (gh, y). The inverse operation
is (g, x)−1 = (g−1, gx).
In the case that G is a topological groupoid acting on a topological space X,
G ⋉ X = G ∗ X inherits the relative topology from the product topology on G × X.
6
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
We can compute the missing structure maps as follows. The range and source of
(g, x) ∈ G ⋉ X are given by
s(g, x) = (g, x)−1(g, x) = (g−1, gx)(g, x) = (g−1g, x) = (s(g), x) = (rX(x), x)
and
r(g, x) = (g, x)(g, x)−1 = (gg−1, gx) = (r(g), gx) = (rX(gx), gx).
Hence, we see that (G ⋉ X)(0) can be naturally identified with X using the map
(rX(x), x) 7→ x. Under this identification, s(g, x) = x and r(g, x) = gx. From all
this, it is easy to see that when G is a group, we recover the usual notion of a
transformation groupoid.
The following observations about the topology of groupoid actions will be useful.
Lemma 2.7. Let G be a locally compact Hausdorff groupoid which acts continuously
on a locally compact Hausdorff space X. Then
(i) if W ⊂ X and S ⊂ G are compact, then S · W is compact in X;
(ii) [16, Proposition 1.72] if G is ´etale, then G ⋉ X is ´etale; and
(iii) if G is ´etale and W ⊂ X and S ⊂ G are open, then S · W is open in X.
Proof. For (i), observe that S · W is the image of the compact subset (S × W ) ∩ (G ∗
X) ⊂ G ∗ X under the continuous action map (g, x) 7→ gx, and is therefore compact
in X.
For (ii), it is shown in [40] that G is ´etale if and only if it admits a Haar system
and the range and source maps are open. In this case, by [16, Proposition 1.72], the
range and source of G ⋉ X are also open and G ⋉ X admits a Haar system, so that
G ⋉ X is ´etale.
For (iii), as above, the range map of G ⋉ X is open. Then S · W = r((S × W ) ∩
(cid:3)
(G ⋉ X)) is open in (G ⋉ X)(0) ∼= X.
The notion of approximate invariance is of central importance to notions of almost
finiteness, and there are many formulations for approximate invariance of subsets
of groups (see [28, Chapter 4], in particular Definition 4.32, and the paragraph im-
mediately following Definition 4.34). All of these formulations are equivalent in the
sense that if a sequence of subsets becomes arbitrarily approximately invariant in one
formulation, then it does so in all of them. We extend this situation to groupoids.
Lemma 2.8. Let G be an ´etale groupoid. Let C, A ⊂ G be compact and nonempty,
and let ǫ > 0. Consider the following conditions.
(i) ([50, Definition 3.1]; c.f. [33, Definition 6.2]) For every u ∈ G(0),
CAu \ Au
Au
< ǫ.
(ii) For every u ∈ G(0),
{a ∈ Au Ca ⊂ Au} > (1 − ǫ)Au.
Then (i) and (ii) are equivalent in the sense that if {An}n∈N is a sequence of subsets
such that for any compact C ⊂ G and ǫ > 0, there exists N ∈ N such that An satisfies
either (i) or (ii) for C and ǫ whenever n ≥ N, then there exists M ∈ N such that An
satisfies the other condition for C and ǫ whenever n ≥ M. If A ⊂ G satisfies either
condition, we will say that A is (C, ǫ)-invariant.
CLASSIFICATION OF TILING C ∗-ALGEBRAS
7
Remark 2.9. It is important to note that (C, ǫ)-invariance of a single set A in the
sense of (i) is not equivalent to (C, ǫ)-invariance of A in the sense of (ii). Indeed,
it is easy to construct examples to show that neither implication holds. Therefore,
our use of this terminology is rather imprecise. This is justified by the fact that for
our purposes we always only care about arbitrary approximate invariance. Still, we
endeavour to make clear which defiition we are using at any time.
Proof of Lemma 2.8. First, notice that if A, C ⊂ G are compact and u ∈ G(0), then
{a ∈ Au Ca ⊂ Au} = Au − {a ∈ Au Ca 6⊂ Au},
so that condition (ii) is equivalent to
{a ∈ Au Ca 6⊂ Au} < ǫAu.
Therefore, it will suffice to find constants c1, c2 > 0 which depend only on C such
that
CAu \ Au ≤ {a ∈ Au Ca 6⊂ Au} ≤ c2CAu \ Au.
1
c1
(i) =⇒ (ii): Fix A, C ⊂ G compact and ǫ > 0, and suppose that A is (C, ǫ)-
invariant in the sense of condition (i), so that
CAu \ Au < ǫAu.
Set c2 = supv∈G(0) vC, which is finite as C is compact. Notice that it is possible
for distinct a1, a2 ∈ {a ∈ Au Ca 6⊂ Au} to admit k1, k2 ∈ C such that k1a1 =
k2a2 ∈ CAu \ Au, and that this requires that k1 6= k2 are distinct, but share a range.
Therefore this situation can occur for at most c2 distinct elements a1, . . . , ac2.
In
other words, any map from {a ∈ Au Ca 6⊂ Au} to CAu \ Au defined by a 7→ kaa,
where ka ∈ C is chosen such that kaa /∈ Au is at most c2-to-one. It follows that
{a ∈ Au Ca 6⊂ Au} ≤ c2CAu \ Au < c2ǫAu,
so that A is (C, c2ǫ)-invariant in the sense of condition (ii).
(ii) =⇒ (i): This time, suppose that A is (C, ǫ)-invariant in the sense of (ii), so
that
{a ∈ Au Ca 6⊂ Au} < ǫAu,
and let c1 = supw∈G(0) Cw > 0, which is again finite since C is compact. Observe
that
CAu \ Au =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
[w∈r(Au) [c∈Cw
≤ Xw∈r(Au) Xc∈Cw
{ca a ∈ wAu, ca /∈ Au}(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
{cau a ∈ wAu, ca /∈ Au}.
Now, for each fixed w ∈ r(Au), the second sum contributes Cw ≤ c1 terms. Each of
these terms corresponds to some c ∈ Cw, and counts the a ∈ wAu for which ca /∈ Au.
Therefore, if a ∈ wAu is such that Ca ⊂ Au, it will never be counted in this sum,
whereas if a ∈ wAu has Ca 6⊂ Au, it can be counted at most Cw ≤ c1 times. Thus,
we obtain
CAu \ Au ≤ Xw∈r(Au) Xc∈Cw
≤ Xw∈r(Au)
{cau a ∈ wAu, ca /∈ Au}
Cw{a ∈ wAu Ca 6⊂ Au}
8
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
≤ c1 Xw∈r(Au)
= c1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Gw∈r(Au)
{a ∈ wAu Ca 6⊂ Au}
{a ∈ wAu Ca 6⊂ Au}(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
= c1{a ∈ Au Ca 6⊂ Au}
< c1ǫAu.
showing that A is (C, c1ǫ)-invariant in the sense of condition (i).
(cid:3)
The following definition is easily seen to generalise the concepts in [27].
Definition 2.10. Suppose that G is a locally compact Hausdorff ´etale groupoid, and
that X is a compact metric space. Let α : G y X be a continuous action. Let C ⊂ G
be compact, and let ǫ > 0.
a nonempty compact open subset S ⊂ G such that S = FN
• A tower of α is a pair (W, S) consisting of a nonempty subset W ⊂ X and
j=1 Sj decomposes
into compact open Sj with s(Sj) = rX(W ) for each j, so that the range and
source maps are injective on each Sj, and such that the sets SjW are pairwise
disjoint for j ∈ {1, . . . , N}. We refer to the set W as the base, S as the shape,
and the sets SjW as the levels of the tower.
FNi
• Consider a finite collection {(W1, S1), . . . , (Wn, Sn)} of towers such that for
each i ∈ {1, . . . , n}, the shape of the i-th tower has decomposition Si =
j=1 Si,j. Such a sequence is called a castle if the collection of all tower levels
{Si,jWi i ∈ {1, . . . , n}, j ∈ {1, . . . , Ni}} is pairwise disjoint. The sets Wi are
called the bases, Si the shapes, and Si,jWi the levels of the castle {(Wi, Si)}n
i=1.
A castle is called a tower decomposition of α if the levels of the castle partition
X.
• A castle {(Wi, Si)}n
i=1 is said to be (C, ǫ)-invariant if all of the shapes S1, . . . , Sn
are (C, ǫ)-invariant subsets of G, in the sense defined by condition (ii) from
Lemma 2.8.
• A castle {(Wi, Si)}n
i=1 is said to be open (respectively closed, clopen) if each
Wi is open (respectively closed, clopen) in X.
• A groupoid action is said to be almost finite if, for every m ∈ N, every compact
C ⊂ G, and every ǫ > 0, there exist
(i) an open castle {(Wi, Si)}n
i=1 whose shapes are (C, ǫ)-invariant, and whose
levels have diameter less than ǫ; and
(ii) sets S′
Siu/m, so that (Wi, S′
i ⊂ Si for each i ∈ {1, . . . , n} such that, for each u ∈ G(0), S′
iu <
i) is a subtower of (Wi, Si), in the sense that if Si =
j=1 Si,j, then there exist Li ∈ N and a subcollection {ji,1, . . . , ji,Li} ⊂
FNi
{1, . . . , Ni} such that S′
i =FLi
nGi=1
In a tower (W, S), where S = FN
X \
SiWi ≺
j=1 Sj, each set Sj should be thought of as corre-
sponding to a single group element from [27, Definition 4.1]. The injectivity condition
of the source and range together with the condition that s(Sj) = rX(W ) ensure that
each element of W has exactly one image under the action of Sj. Aside from this
l=1 Si,ji,l, and such that
S′
iWi.
nGi=1
CLASSIFICATION OF TILING C ∗-ALGEBRAS
9
modification to ensure compatibility with the fibred structure of the groupoid, the
definition is identical in spirit to [27, Definition 8.2].
We show that the diameter condition appearing in the statement is no obstacle in
our case of interest.
In the group setting, closedness of the towers comes with no
loss of generality by [27, Lemma 10.1]. We conjecture that this should also be true
in the groupoid setting, but we do not have a proof, so it appears as an assumption
in the following result. Another lingering question to consider is whether the castles
witnessing almost finiteness can always be chosen to partition X when X is totally
disconnected (c.f. [27, Theorem 10.2]).
Lemma 2.11 (c.f. [27, Theorem 10.2]). Let G be a locally compact Hausdorff ´etale
groupoid, X a totally disconnected compact metric space, and α : G y X a free and
continuous action which admits arbitrarily invariant clopen castles, as in the defini-
tion of almost finiteness, but which do not necessarily satisfy the diameter condition.
Then α is almost finite. That is, we can choose the invariant clopen castles to satisfy
the diameter condition, with no loss of generality.
tower as in the statement of the lemma, so that S =Fm
Proof. Let C ⊂ G be compact, let ǫ > 0, and let (W, S) be a (C, ǫ)-invariant clopen
i=1 Si, where Si is compact and
open in G. We first show that we can refine (W, S) into a clopen castle whose levels
have diameter smaller than ǫ. The bases of the towers in the castle will partition W,
and the shape of each new tower will simply be the subset of S consisting of arrows
which act on its base. Since S is (C, ǫ)-invariant, it will follow that the shape of each
new tower is also (C, ǫ)-invariant.
Consider the first tower level S1 · W. Since X is totally disconnected, it has a basis
of clopen sets. Therefore, since S1 · W is compact and open, there exists a finite
collection of clopen subsets {Un1}n1=1,...,N of X with diameter smaller than ǫ such
j=1 Uj for each n1 = 1, 2, . . . , N
in turn, we may assume without loss of generality that the collection {Un1}n1=1,...,N
is pairwise disjoint.
n1=1 Un1. By replacing Un1 by Un1 \Sn1−1
that S1 · W =SN
For each n1 ∈ {1, . . . , N}, let Vn1 = S−1
1
· Un1, which is a compact open subset
of X by Lemma 2.7. Observe that S1 · Vn1 = Un1 since rX(Un1) ⊂ r(S1) = s(S−1
1 ).
Furthermore, since the collection {Un1}n1=1,...,N is pairwise disjoint and the source
map is injective on S1, the collection {Vn1}n1=1,...,N is pairwise disjoint, and we have
W = S−1
n1=1 Vn1 so that the collection {Vn1}n1=1,...,N
partitions W. Associate to Vn1 the subset Hn1 of S1 which is able to act upon it. That
is, put Hn1 = S1 ∩ s−1(rX(Vn1)), so that Hn1 · Vn1 = S1 · Vn1 = Un1. Observe that
Hn1 is compact and open because rX is continuous and open by Lemma 2.4 and s is
a local homeomorphism.
n1=1 Un1 =FN
1 ·FN
n1=1 S−1
1 ·Un1 =FN
In this manner, we have created N "single-level" towers (Vn1, Hn1) for n1 ∈ {1, . . . , N}
whose levels have diameter smaller than ǫ and partition S1 · W, and such that for each
n1 ∈ {1, . . . , N} and each x ∈ Vn1, we have Hn1rX(x) = S1rX(x).
Now, for each n1 = 1, . . . , N in turn, consider the subset S2 · Vn1 of S2 · W. Arguing
similarly as above, construct a cover of S2 · Vn1 by pairwise disjoint clopen subsets
of X with diameter smaller than ǫ. For each n2 ∈ {1, . . . , Nn1}
{Un1,n2}n2=1,...,Nn1
put Vn1,n2 = S−1
· Un1,n2 and H(n1,n2) = (Hn1 ∪ S2) ∩ s−1(rX(Vn1,n2)) so that Vn1,n2
2
and H(n1,n2) are both compact and open, and so that {Vn1,n2}n2=1,...,Nn1
partitions
Vn1. Define H(n1,n2),1 = S1 ∩ H(n1,n2) and H(n1,n2),2 = S2 ∩ H(n1,n2) so that H(n1,n2) =
i=1 H(n1,n2),i. Observe that H(n1,n2),1 ·Vn1,n2 ⊂ Un1 and H(n1,n2),2 ·Vn1,n2 ⊂ Un1,n2 both
F2
10
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
have diameter smaller than ǫ. In this way, for each n1 ∈ {1, . . . , N} we construct
Nn1 "two-level" towers (Vn1,n2, H(n1,n2)) for n2 ∈ {1, . . . , Nn1}. The collection of all
of the levels of these towers is pairwise disjoint and partitions (S1 ∪ S2) · W, each
level has diameter smaller than ǫ, and for each x ∈ Vn1,n2 we have H(n1,n2)rX(x) =
(S1 ∪ S2)rX(x).
Continue to iterate this procedure, at the k-th step beginning with a collection
of towers (Vn1,n2,...,nk−1, H(n1,n2,...,nk−1)) for n1 ∈ {1, . . . , N}, n2 ∈ {1, . . . , Nn1}, . . . ,
nk−1 ∈ {1, . . . , Nn1,...,nk−2}, and covering Sk · Vn1,n2,...,nk−1 by finitely many pairwise
disjoint clopen subsets {Un1,...,nk}nk=1,...,Nn1,...,nk−1
of X with diameter smaller than ǫ.
Set Vn1,...,nk = S−1
partitions
Vn1,...,nk−1. Also set H(n1,...,nk) = (H(n1,...,nk−1) ∪ Sk) ∩ s−1(rX(Vn1,...,nk)) (with decom-
position H(n1,...,nk),i = Si ∩ H(n1,...,nk) for each i ∈ {1, . . . , k}) to obtain a collection of
"k-level" towers {(Vn1,...,nk, H(n1,...,nk))} indexed by n1 up to nk such that the collec-
tion of all tower levels is pairwise disjoint and partitions (S1 ∪ · · · ∪ Sk) · W , each level
of each tower has diameter smaller than ǫ, and so that for each x ∈ Vn1,...,nk we have
H(n1,...,nk)rX(x) = (S1 ∪ · · · ∪ Sk)rX(x).
k · Un1,...,nk so that the collection {Vn1,...,nk}nk=1,...,Nn1,...,nk−1
The procedure terminates after m steps (where m is the number of levels in
the tower (W, S)) to produce a clopen castle {(Vn1,...,nm, H(n1,...,nm))}, where n1 ∈
{1, . . . , N}, n2 ∈ {1, . . . , Nn1}, . . . , nm ∈ {1, . . . , Nn1,...,nm−1}, such that every level
of every tower has diameter smaller than ǫ. The levels of this castle will partition
(S1 ∪ · · · ∪ Sm) · W = S · W. For each x ∈ Vn1,...,nm = r−1
X (s(H(n1,...,nm))) we have
H(n1,...,nm)rX(x) = (S1 ∪ · · · ∪ Sm)rX(x) = SrX(x), so that for each u ∈ s(H(n1,...,nm))
we can use (C, ǫ)-invariance of S to obtain
CH(n1,...,nm)u \ H(n1,...,nm)u = CSu \ Su < ǫSu = ǫH(n1,...,nm)u,
which shows that H(n1,...,nm) is (C, ǫ)-invariant.
It remains to show that given a clopen castle {(Wj, Sj)}j=1,...,J witnessing condition
(ii) of almost finiteness, the clopen castle {(V (j)
obtained
by applying the procedure above on each tower in the castle satisfies condition (ii) of
almost finiteness. For each j ∈ {1, . . . , J} find sets S′
(n1,...,nmj ))}j,n1,...,nmj
, H (j)
n1,...,nmj
j ⊂ Sj such that
X \
JGj=1
SjWj ≺
S′
jWj.
JGj=1
Observe that for each j we have
Gn1,...,nmj
(S′
j ∩ H (j)
(n1,...,nmj ))V (j)
n1,...,nmj
= S′
jWj
and that the sets appearing in the union on the left-hand side are levels of the
tower (V (j)
ju for each
u ∈ G(0). It follows that
(n1,...,nmj )). Observe also that (S′
(n1,...,nmj ))u ≤ S′
j ∩ H (j)
, H (j)
n1,...,nmj
X \
JGj=1 Gn1,...,nmj
H (j)
(n1,...,nmj )V (j)
n1,...,nmj
= X \
SjWj
JGj=1
≺
JGj=1
S′
jWj
j
i=1 F (i)
1 ;
of K (0) such that
pairwise disjoint collection of clopen subsets nF (i)
(i) F =Fn
(ii) K (0) =Fn
(iv) K =Fn
i=1FNi
i=1F1≤j,k≤Ni V (i)
j,k satisfying r(V (i)
j,k ) = F (i)
K-set V (i)
j,k .
j=1 F (i)
j
;
(iii) for each i ∈ {1, . . . , n} and j, k ∈ {1, . . . , Ni} there is a unique compact open
and s(V (i)
j,k ) = F (i)
k ; and
j
CLASSIFICATION OF TILING C ∗-ALGEBRAS
11
=
JGj=1 Gn1,...,nmj
(S′
j ∩ H (j)
(n1,...,nmj ))V (j)
n1,...,nmj
.
Therefore, the choices (H (j)
finiteness for the new castle.
(n1,...,nmj ))′ = (S′
j∩H (j)
(n1,...,nmj )) witness property (ii) of almost
(cid:3)
We now see how our notion of almost finiteness links back to Matui's original
definition. For convenience, we first present Suzuki's generalisation of Matui's original
notion.
Definition 2.12 ([50, Definition 3.2]). Let K be a compact Hausdorff ´etale groupoid.
We say that a clopen subset F of K (0) is a fundamental domain of K if there exist
a natural number n ∈ N, natural numbers Ni ∈ N for each i ∈ {1, . . . , n}, and a
i ∈ {1, . . . , n}, j ∈ {1, . . . , Ni}o
Definition 2.13 ([50, Definition 3.4]). Let G be a locally compact Hausdorff ´etale
groupoid. A (compact) subgroupoid K of G is called elementary if G(0) ⊂ K and if
K admits a fundamental domain.
Definition 2.14 ([50, Definition 3.6]). Let G be a locally compact Hausdorff ´etale
groupoid. We say that G is almost finite if it satisfies
(i) the union of all compact open G-sets covers G; and
(ii) for any compact subset C ⊂ G and ǫ > 0, there exists a (C, ǫ)-invariant
elementary subgroupoid of G.
Theorem 2.15 (c.f.
[50, Lemma 5.2]). Let G be a locally compact Hausdorff ´etale
groupoid, X a compact metric space, and α : G y X a continuous action. Let C ⊂ G
be a compact subset, and fix ǫ > 0.
Suppose that α admits a (C, ǫ)-invariant clopen tower decomposition. Then the
transformation groupoid G⋉X admits a (C ×X, ǫ)-invariant elementary subgroupoid.
The converse holds in the case that G is ample.
Proof. (⇒): Take a clopen tower decomposition {(Wi, Si)}n
i=1 of α as in the theorem,
j=1 Si,j for each i ∈ {1, . . . , n}, where the range map of G is injective
on each Si,j. Define an elementary subgroupoid
so that Si = FNi
K =
nGi=1 G1≤j,k≤Ni
V (i)
j,k
as follows. We will let V (i)
j,k be a collection of arrows from the k-th to the j-th level
of the i-th tower (Wi, Si), which we will obtain by moving from the k-th level Si,kWi
down to the base along S−1
i,k , and then from the base to the j-th level along Si,j. More
precisely, for each j ∈ {1, . . . , Ni}, consider the collection
V (i)
j,base = (Si,j × Wi) ∩ (G ⋉ X) = {(g, x) g ∈ Si,j, x ∈ Wi, s(g) = rX(x)}.
12
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
Notice that for each x ∈ Wi there is exactly one associated element (g, x) ∈ V (i)
Define
j,base.
V (i)
j,k = V (i)
j,base(V (i)
k,base)−1.
This defines an elementary subgroupoid with F (i)
j = Si,j · Wi.
Now, fix a compact subset C ⊂ G and ǫ > 0 and suppose that the clopen tower
decomposition {(W1, S1), . . . , (Wn, Sn)} is (C, ǫ)-invariant. Construct the associated
elementary subgroupoid
K =
nGi=1 G1≤j,k≤Ni
V (i)
j,k .
j
Fix x ∈ X. By construction, there exists a unique i ∈ {1, . . . , n} and j ∈ {1, . . . , Ni}
such that x ∈ F (i)
. Then the only elements k ∈ K for which s(k) = x lie in the
sets V (i)
for l ∈ {1, . . . , Ni}. From each such set, there exists a unique kl ∈ K with
l,j
s(kl) = x, so Kx = Ni. On the other hand, we assumed that we can partition Si
j=1 Si,j such that each Si,j has the same source, and such that the
source map is injective on each Si,j. For any u ∈ s(Si), it follows that Siu = Ni,
which shows that Kx = Siu.
into subsets Si =FNi
For the choice of x ∈ F (i)
y ∈ Wi. We can compute
j = Si,jWi from above, write x = hy for h ∈ Si,j and
Kx = {(gh−1, x) g ∈ Si,l, s(g) = s(h), l = 1, . . . , Ni}
= {(gh−1, hy) g ∈ Si,l, s(g) = s(h), l = 1, . . . , Ni},
which gives
(C × X)Kx = {(c, z)(gh−1, hy) g ∈ Si,l, s(g) = s(h), l = 1, . . . , Ni, c ∈ C, z ∈ X}.
These elements are composable when s(c) = r(g) and gy = z, producing the product
(cgh−1, hy). Thus, with h and y fixed as above, we see that
(C × X)Kx \ Kx = {(cgh−1, hy) g ∈ Si,l, s(g) = s(h), l = 1, . . . , Ni,
c ∈ C, s(c) = r(g), cg /∈ Si,m for m = 1, . . . , Ni}
= {(cgh−1, hy) g ∈ Si, s(g) = s(h), c ∈ C, s(c) = r(g), cg /∈ Si}.
On the other hand, for u = rX(y) ∈ S(0)
i
, we have
CSiu \ Siu = {cgu g ∈ Si, s(g) = u = rX(y), c ∈ C, s(c) = r(g), cg /∈ Si}
= {cgu g ∈ Si, s(g) = s(h), c ∈ C, s(c) = r(g), cg /∈ Si},
so we see that CSiu \ Siu = (C × X)Kx \ Kx via the bijection (cgh−1, hy) 7→ cgu.
Thus, combining this with our previous observation that Kx = Siu = Ni for each
u ∈ S(0)
, we have obtained
i
(C × X)Kx \ Kx
Kx
=
CSiu \ Siu
Siu
< ǫ,
where the last inequality uses (C, ǫ)-invariance of Si. This shows that the elementary
subgroupoid K is (C × X, ǫ)-invariant.
(⇐): Let G be ample, and let K = Fn
j,k be a (C × X, ǫ)-invariant
elementary subgroupoid of G ⋉ X. Since each V (i)
j,k is a compact subset of G ⋉ X, the
image πG(V (i)
j,k ) under the projection to the G coordinate is also compact and open.
Since G is ample, the topology on G has a base of clopen bisections by [5, Lemma 2.2].
i=1F1≤j,k≤Ni V (i)
Si = πG(s−1(F (i)
1 ) ∩ K) = πG
NiGl=1
V (i)
l,1 ,
CLASSIFICATION OF TILING C ∗-ALGEBRAS
13
Therefore, using compactness, we can write πG(V (i)
j,k ) as a union of finitely many
pairwise disjoint clopen bisections, each of which will be compact in G because they
are closed in the compact subset πG(V (i)
j,k ). We may use these bisections to split V (i)
j,k
into smaller compact open (G ⋉X)-sets, to assume without loss of generality that the
range and source maps are injective on πG(V (i)
j,k ). We note that because the original
V (i)
j,k was a (G ⋉ X)-set, the newly defined subsets V (i)
j,k also yield redefined clopen sets
j ⊂ X such that the collection {F (i)
F (i)
We then form a clopen tower decomposition {(Wi, Si)}i=1,...,n as follows. Set Wi =
1 and
j }i,j is pairwise disjoint.
F (i)
j=1 Si,j, where Si,j = πG(V (i)
with associated decomposition Si = FNi
j,1 ). The levels of
the i-th tower are then precisely the clopen sets F (i)
for j ∈ {1, . . . , Ni}. Using
the properties of the elementary subgroupoid, we see that the base of each tower is
clopen, each Si,j is compact and open, s(Si,j) = rX(Wi), and that the collection of
tower levels is pairwise disjoint and partitions X. Due to the way we modified V (i)
j,k ,
each Si,j is a subset of a bisection, and it follows that the source and range maps are
injective on Si,j. This shows that {(Wi, Si)}i=1,...,n is a clopen tower decomposition,
as we claimed.
j
Next, we show that this clopen tower decomposition is (C, ǫ)-invariant. For each
x ∈ Wi we compute
(C × X)Kx \ Kx = {(cg, x) c ∈ C, (g, x) ∈ Kx, (cg, x) /∈ Kx}
= {cg c ∈ C, g ∈ πG(Kx), cg /∈ πG(Kx)}
= {cg c ∈ C, g ∈ SirX(x), cg /∈ SirX(x)}
= CSirX(x) \ SirX(x).
We also have Kx = πG((s−1(Wi) ∩ K)x) = SirX(x). Putting this all together
gives for each x ∈ s(Si) = Wi that
CSirX(x) \ SirX(x)
SirX(x)
=
(C × X)Kx \ Kx
Kx
< ǫ.
Thus, we have produced a (C, ǫ)-invariant clopen tower decomposition of α.
(cid:3)
Corollary 2.16. Let G be a locally compact Hausdorff ´etale groupoid with compact
unit space, X a totally disconnected compact metric space, and α : G y X a contin-
uous action.
If α is almost finite, with the additional requirement that the open castles in Def-
inition 2.10 can be chosen to be clopen tower decompositions of X, then the trans-
formation groupoid G ⋉ X is almost finite. Conversely, if G is ample and G ⋉ X is
almost finite, then α is almost finite and the castles appearing in the definition can
be chosen to be clopen tower decompositions.
Proof. Fix ǫ > 0 to be used throughout the proof.
First assume that α is almost finite, and that the open castles in the definition can
be chosen to be clopen tower decompositions. Fix a compact subset A ⊂ G ⋉ X. We
aim to produce an (A, ǫ)-invariant elementary subgroupoid of G ⋉ X. Denote by πG :
G ⋉ X → G the projection map to the G-coordinate. Since πG is continuous, πG(A)
14
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
is a compact subset of G. By our assumption on α, we can find a (πG(A), ǫ)-invariant
clopen tower decomposition. Applying Theorem 2.15 produces a (πG(A) × X, ǫ)-
invariant elementary subgroupoid of G ⋉ X. Since A ⊂ πG(A) ⋉ X, this elementary
subgroupoid is also (A, ǫ)-invariant, so we see that G ⋉ X satisfies condition (ii) of
Definition 2.14. Since (G ⋉ X)(0) ∼= X is totally disconnected, G ⋉ X automatically
satisfies condition (i) of Definition 2.14 (see for example [13]), and so is seen to be
almost finite.
Now assume that G is ample and G ⋉ X is almost finite. Fix a compact subset
C ⊂ G. We aim to construct a (C, ǫ)-invariant open castle which satisfies conditions
(i) and (ii) from the final part of Definition 2.10. By almost finiteness of G ⋉ X, since
C ×X is a compact subset of G⋉X, there exists a (C ×X, ǫ) elementary subgroupoid
of G ⋉ X. Since G is ample, we can apply Theorem 2.15 to obtain a (C, ǫ)-invariant
clopen tower decomposition of α. Thus, we have obtained a (C, ǫ)-invariant open
castle which covers X, so that condition (ii) from Definition 2.10 is automatically
satisfied. It only remains to show that we can arrange for the levels of this castle
to have diameter smaller than ǫ, which follows from Lemma 2.11 since X is totally
disconnected.
(cid:3)
3. Ornstein-Weiss quasitiling machinery for groupoids
Following [27], we require a generalisation of the Ornstein-Weiss quasitiling theorem
[37, page 24, Theorem 6] to the groupoid setting. In this section we prove a version
of the Ornstein-Weiss theorem that mimics the alternative formulation appearing as
[28, Theorem 4.36].
We will need the following concepts:
Definition 3.1 ([28, Definition 4.29]). Let (X, µ) be a finite measure space. Let
λ, ǫ ≥ 0. We say that a collection {Ai}i∈I of measurable subsets of X
(i) is a λ-even covering of X if there exists a positive integer M such that
plicity of the λ-even covering.
Pi∈I 1Ai ≤ M andPi∈I µ(Ai) ≥ λMµ(X), in which case M is called a multi-
(ii) λ-covers X if µ(Si∈I Ai) ≥ λµ(X).
(iii) is ǫ-disjoint if, for each i ∈ I, there exists a subset cAi ⊂ Ai such that µ(cAi) ≥
(1 − ǫ)µ(Ai), and such that the collection {cAi}i∈I is pairwise disjoint.
We extend the above definition to apply to collections of compact subsets of ´etale
groupoids by asking for the appropriate concept to hold at each (finite) source bundle
within each subset. One can also write down a similar definition for locally compact
groupoids admitting a Haar system by using the Haar measure at each unit.
In
the ´etale case, the Haar measure is just the counting measure, which we use in the
definition below.
Definition 3.2. Let G be an ´etale groupoid and λ ≥ 0. Let A ⊂ G be compact. We
say that a collection {Ai}i∈I of compact subsets of A
(i) is a λ-even covering of A with multiplicity M if, for every u ∈ s(A), the
collection {Aiu}i∈I is a λ-even covering of Au with multiplicity M.
(ii) λ-covers A if {Aiu}i∈I λ-covers Au for each u ∈ s(A).
(iii) is ǫ-disjoint if {Aiu}i∈I is ǫ-disjoint for each u ∈ s(A).
The following pair of lemmas will be applied repeatedly in the proof of our qua-
sitiling result.
CLASSIFICATION OF TILING C ∗-ALGEBRAS
15
Lemma 3.3 (c.f. [28, Lemma 4.31]). Let A be a compact subset of any locally compact
Hausdorff ´etale groupoid (whose unit space is not necessarily compact). Let 0 ≤ ǫ ≤ 1
2
and 0 < λ ≤ 1, and let {Ai}i∈I be a λ-even covering of A. Then, for each u ∈ s(A),
there is an ǫ-disjoint subcollection {Aju}j∈Ju of {Aiu}i∈I, indexed by Ju ⊂ I, which
ǫλ-covers Au.
Proof. By definition, for each u ∈ s(A), the collection {Aiu}i∈I is a λ-even covering
of the finite measure space Ax (equipped with the counting measure) in the sense
of Definition 3.1. Thus, for each u ∈ s(A), applying [28, Lemma 4.31] yields the
ǫ-disjoint subcollection of {Aiu}i∈I that ǫλ-covers Au that we seek.
(cid:3)
Throughout this section, we will use as standard the notion of approximate invari-
ance given by property (ii) in Lemma 2.8.
Definition 3.4. Let G be an ´etale groupoid. Let C, A ⊂ G be nonempty and
compact, and let ǫ > 0. We denote
and we say that A is (C, ǫ)-invariant if, for each u ∈ s(A), we have
IC(A) := {a ∈ A Ca ⊂ A},
IC(Au) ≥ (1 − ǫ)Au.
Lemma 3.5 (c.f.
[28, Lemma 4.33]). Let G be a locally compact Hausdorff ´etale
groupoid with compact unit space. Let C ⊂ G be a nonempty, compact subset con-
taining G(0). Choose ǫ > 0, and let
δ ≥ max 0, 1 −
inf w∈G(0) Cw
supw∈G(0) wC
(1 − ǫ)! .
Suppose that A ⊂ G is a nonempty, compact subset which is (C, ǫ)-invariant. Then
the collection {Ca a ∈ IC(A)} is a (1 − δ)-even covering of A with multiplicity
supw∈G(0) wC.
Proof. Since A is (C, ǫ)-invariant, for each u ∈ s(A) we have IC(Au) ≥ (1 − ǫ)Au.
Therefore, we see that
Xa∈IC (A)
Cau ≥ IC(Au)
inf
a∈IC(Au)
Cr(a)
≥ (1 − ǫ)Au
inf
w∈G(0)
Cw
= (1 − ǫ)Au supw∈G(0) wC
inf w∈G(0) Cw
supw∈G(0) wC
≥ (1 − δ) supw∈G(0) wCAu,
so that the second condition in the definition of a (1 − δ)-even covering with multi-
plicity supw∈G(0) wC is satisfied.
Next, to have g ∈ Ca1u1 ∩ Ca2u2, we require that u1 = s(g) = u2 = u. So, suppose
that g ∈ Ca1u ∩ Ca2u for distinct a1, a2 ∈ IC(Au). Then there exist c1, c2 ∈ C such
that c1a1 = g = c2a2, which requires that r(c1) = r(g) = r(c2). Since a1 6= a2, we
must also have that c1 6= c2. Extending this argument, we see that for each fixed
u ∈ s(A), the intersection of n ∈ N sets from the collection {Cau a ∈ IC(A)} at a
common element g requires the existence of n distinct elements of r(g)C. This shows
that, for each u ∈ s(A) and g ∈ Au,
1Cau(g) ≤ r(g)C ≤ supw∈G(0) wC.
(cid:3)
Xa∈IC (Au)
16
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
In the group case, one may alternatively characterise approximate invariance by
using a "small boundary" condition. We introduce an analogue of the notion for
groupoids.
Definition 3.6 (c.f.
´etale groupoid, and let C, A be compact subsets of G. The C-boundary of A is
[28, Definition 4.34]). Let G be a locally compact Hausdorff
∂C(A) := {g ∈ G Cg ∩ A 6= ∅ and Cg ∩ Ac 6= ∅}
Notice that ∂C(A) may be infinite. Therefore, to obtain the aforementioned "small
boundary" condition, we need to consider the C-boundary of the source bundle at
each unit, and ask that for every u ∈ s(A) we have
∂C(Au) ≤ ǫAu.
In particular, notice that ∂C(Au) ⊂ C −1Au and so it is always finite since C −1A is
compact.
We now introduce our analogue of quasitilings. When reading the following defini-
tion, notice that cutting down to the finite source bundles Au is still necessary, but
is built into our terminology via Definition 3.2.
Definition 3.7 (c.f.
[28, Definition 4.35]). Let G be a locally compact Hausdorff
´etale groupoid with compact unit space. Let A be a compact subset of G, and fix
ǫ > 0. A finite collection {T1, . . . , Tn} of compact subsets of G is said to ǫ-quasitile
i=1 TiCi ⊂ A and so that the collection of
A if there exist C1, . . . , Cn ⊂ G such thatSn
right translates {Tic i ∈ {1, . . . , n}, c ∈ Ci} is ǫ-disjoint and (1 − ǫ)-covers A.
The subsets T1, . . . , Tn are referred to as the tiles, and C1, . . . , Cn as the centres of
the quasitiling.
The following result is our analogue of the Ornstein-Weiss quasitiling theorem. In
the case that G is an amenable group, tiles T1, . . . , Tn satisfying the conditions of
the theorem automatically exist. We conjecture that the same is true when G is a
(topologically) amenable groupoid. We show that our main examples admit such tiles
in Lemma 8.1.
Theorem 3.8 (c.f. [28, Theorem 4.36]). Let G be a locally compact Hausdorff ´etale
groupoid with compact unit space. Let 0 < ǫ < 1/2, and choose m ∈ N such that
m ≥ 2. Let n be a positive integer such that (1 − ǫ/m)n < ǫ. Suppose that T1 ⊂ T2 ⊂
· · · ⊂ Tn are compact subsets of G which contain G(0) and are such that, for each
i ∈ {1, . . . , n},
(i) 1/m <
inf
sup
w∈G(0) Tiw
w∈G(0) wTi ≤ 1;
inf
w∈G(0) Tiw
w∈G(0) wTi(1 − ǫ) ≥ 1/m; and
(ii)
(iii) ∂Ti−1(Tiw) ≤ (ǫ2/8)Tiw for each w ∈ G(0).
sup
Then every (Tn, ǫ2/4)-invariant compact subset of G is ǫ-quasitiled by {T1, . . . , Tn}.
We briefly describe the ideas behind the proof of Theorem 3.8. The general phi-
losophy will be to start with an (Tn, ǫ2/4)-invariant compact subset A of G, apply
Lemma 3.5 to construct an even covering of A whose tolerance is limited by Tn and
ǫ, and then use Lemma 3.3 to turn our even covering into an ǫ-disjoint collection
which still covers some fixed portion of each source bundle in A. The assumptions of
Theorem 3.8 are designed to ensure that we can iterate this procedure n ∈ N times.
In particular, at each stage, the relative complement in A of everything covered up
CLASSIFICATION OF TILING C ∗-ALGEBRAS
17
to that point will be sufficiently invariant (under some Tk, where the index k shrinks
as the procedure continues) to allow another application of this pair of lemmas. In
addition, the assumptions ensure that when this procedure terminates, the union
of all the ǫ-disjoint subcollections that have been constructed will (1 − ǫ)-cover the
invariant compact set A that we started with.
Proof of Theorem 3.8. Let A be a nonempty (Tn, ǫ2/4)-invariant compact subset of
G. We will recursively construct Cn, . . . , C1 ⊂ G (in that order) such that, for each
i=k TiCi ⊂ A, and so that the collection of translates {Tic i ∈
{k, . . . , n}, c ∈ Ci} is ǫ-disjoint and λk-covers A, where
k ∈ {1, . . . , n}, Sn
λk = min(1 − ǫ, 1 − (1 − ǫ/m)n−k+1).
By our assumption on n it will then follow that {T1, . . . , Tn} ǫ-quasitiles A. Indeed,
we have that (1 − ǫ/m)n < ǫ, so λ1 = 1 − ǫ, as required.
For the base step (when k = n), we apply Lemma 3.5 to the (Tn, ǫ2/4)-invariant
compact subset A to notice that the collection {Tna a ∈ ITn(A)}, where ITn(A) =
{a ∈ A Tna ⊂ A} is as in Definition 3.4, is a β-even covering of A, where
β =
inf w∈G(0) Tnw
supw∈G(0) wTn
(1 − ǫ2/4).
Observe that β ≥ inf Tnw
sup wTn(1 − ǫ), and so, by condition (ii), β ≥ 1/m. This shows
that the collection {Tna a ∈ ITn(A)} is a (1/m)-even covering of A. Then we can
apply Lemma 3.3 to find, for each u ∈ G(0), a subset Cn,u ⊂ ITn(Au) such that the
subcollection {Tnc c ∈ Cn,u} is ǫ-disjoint and (ǫ/m)-covers Au. This allows us to
construct Cn =Fu∈G(0) Cn,u, which satisfies the properties we seek.
Suppose that, for some k ∈ {1, . . . , n−1}, we have constructed Cn, Cn−1, . . . , Ck+1 ⊂
G with the properties we desire. Set A′
ku < ǫAu for every
u ∈ s(A), then we can finish our construction by taking each of Ck, . . . , C1 to be the
empty set, so suppose that the set Zk := {u ∈ s(A) A′
ku ≥ ǫAu} is nonempty.
ku.
Then, for each u ∈ G(0) \ Zk, we will take Cku = ∅, and we define Ak = Su∈Zk A′
We now wish to show that Ak is (Tk, ǫ
k = A \Sn
i=k+1 TiCi. If A′
First, observe that, for each u ∈ G(0), i ∈ {k + 1, . . . , n}, and c ∈ Ciu, we have
∂Tk (Ticu) ≤ ∂Ti−1(Ticu), since Tk ⊂ Ti−1. Notice that the map a 7→ ac−1 is an
injection which maps ∂Ti−1(Tics(c)) into ∂Ti−1(Tir(c)). Using this along with the
boundary condition (iii) on the Ti, we see that
2)-invariant.
∂Tk(Tics(c)) ≤ ∂Ti−1(Tics(c)) ≤ ∂Ti−1(Tir(c)) ≤ (ǫ2/8)Tir(c).
Since the collection {Tics(c) i ∈ {k + 1, . . . , n}, c ∈ Ci} is 1
2 ),
i=k+1 TiCiu ≤ Au ≤ ǫ−1Aku for every u ∈ Zk, we obtain, for each
2-disjoint (since ǫ < 1
and since Sn
u ∈ Zk, that
n[i=k+1 [c∈Ciu
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
∂Tk (Ticu)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
≤
=
≤
Tir(c)
Tic
ǫ2
8
ǫ2
8
nXi=k+1 Xc∈Ciu
nXi=k+1 Xc∈Ciu
TiCiu(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
4 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
n[i=k+1
ǫ2
18
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
(3.1)
≤
ǫ
4
Aku,
where we used 1
1
2-disjointness to obtain the penultimate inequality, as follows. Use
2-disjointness to find, for each i ∈ {k + 1, . . . , n} and c ∈ Ciu, a subset dTi,c ⊂ Ti such
that dTi,cc ≥ 1
2Tic, and such that the collection {dTi,cc i ∈ {k + 1, . . . , n}, c ∈ Ciu}
is pairwise disjoint. Then we have
n[i=k+1
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
TiCiu(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Tic(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
=(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
n[i=k+1 [c∈Ciu
≥(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
nGi=k+1 Gc∈CiudTi,cc(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
nXi=k+1 Xc∈Ciu
dTi,cc
nXi=k+1 Xc∈Ciu
1
2
≥
=
Tic.
Notice that all the sums in the computation of (3.1) are finite. In particular, one
can see that Ciu < ∞ for each i and each u ∈ G(0), because TiCiu ⊂ Au which is
finite, and so ∞ > TiCiu = Pc∈Ciu Tir(c) ≥ Ciu infw∈G(0) Tiw ≥ Ciu because
Since A is (Tn, ǫ2/4)-invariant, and Tk ⊂ Tn, A is also (Tk, ǫ2/4)-invariant. Thus,
for each u ∈ s(A), ITk(Au) = {a ∈ Au Tka ⊂ A} ≥ (1 − ǫ2/4)Au, and so, for
each u ∈ s(Ak) ⊂ Zk, we have
G(0) ⊂ Ti.
ITk(Aku) = {a ∈ Aku Tka ⊂ Aku}
ITk(Au) \
=(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
TiCiu ∪ [c∈Ciu
n[i=k+1
≥ ITk(Au) −(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
−(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
TiCiu(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
n[i=k+1
4! Au − (Au − Aku) −
≥ 1 −
∂Tk (Ticu)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
∂Tk (Ticu)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
n[i=k+1 [c∈Ciu
Aku
ǫ
4
ǫ2
ǫ
4
= Aku −
Aku −
≥(cid:18)1 −
ǫ
2(cid:19) Aku,
ǫ2
4
Au
which shows that Ak is (Tk, ǫ
2)-invariant. In the first equality, we think of the comple-
ment as enforcing additional constraints on the elements of ITk(Au) to get elements
i=k+1 TiCiu, as this set
is the complement of Aku in Au. Furthermore, if an element a ∈ ITk (Au) lies in
∂Tk (Ticu) for some i ∈ {k + 1, . . . , n} and c ∈ Ci, then there exists t ∈ Tk such
i=k+1 TiCiu = Au \ Aku, so such elements cannot lie in the left-
hand set. To obtain the third line, we have used inequality (3.1) and the fact that
i=k+1 TiCiu = Au \ Aku. The final line uses the fact that Au ≤ ǫ−1Aku, since
of the left-hand side. Clearly, such elements cannot lie in Sn
that ta ∈ Ticu ⊂ Sn
Sn
u ∈ Zk.
CLASSIFICATION OF TILING C ∗-ALGEBRAS
19
Thus, we can apply Lemma 3.5 to see that the collection of right translates of Tk
which lie in Ak form a βk-even covering of Ak, where
βk =
inf w∈G(0) Tkw
supw∈G(0) wTk(cid:18)1 −
ǫ
2(cid:19) .
Observe that βk ≥ inf Tkw
sup wTk(1 − ǫ) and so, by condition (ii), we see that βk > 1/m.
Therefore, the collection {Tka a ∈ Ak, Tka ⊂ Ak} is a (1/m)-even covering of Ak.
Use Lemma 3.3 to obtain, for each u ∈ s(Ak), an ǫ-disjoint subcollection {Tkcu c ∈
Ck,u} of these translates which (ǫ/m)-covers Ak, and set Ck =Fu∈s(Ak) Ck,u.
Note that Sn
collections, {Tkc c ∈ Ck} andSn
i=k{Tic c ∈ Ci} is ǫ-disjoint, because it is the union of two ǫ-disjoint
i=k+1{Tic c ∈ Ci}, which are such that the members
of each collection are disjoint from all the members of the other. All that remains
to show is that the new collection λk-covers A. Indeed, it is enough to show that it
λk-covers Au for each u ∈ Zk, since we already know the collection (1 − ǫ)-covers Au
for each u /∈ Zk.
i=k+1 TiCi was a λk+1-cover of A, so, at each unit u ∈ s(A), the
remainder has cardinality at most (1 − λk+1)Au. For each u ∈ Zk, we constructed
i=k TiCiu is
We assumed thatSn
an (ǫ/m)-cover of this remainder. Thus, for each u ∈ Zk, the collection Sn
an αk-cover of Au where
αk = λk+1 +
(1 − λk+1).
ǫ
m
If λk+1 = 1 − ǫ, we obtain αk = 1 − ǫ + ǫ2/m > 1 − ǫ ≥ λk.
If λk+1 = 1 − (1 − ǫ/m)n−k, we obtain
αk = 1 −(cid:18)1 −
= 1 −(cid:18)1 −
= 1 −(cid:18)1 −
≥ λk.
ǫ
m(cid:18)1 −
m(cid:19)
ǫ
ǫ
ǫ
+
m(cid:19)n−k
m(cid:19)n−k(cid:18)1 −
m(cid:19)n−k+1
ǫ
i=k TiCi is a λk-cover.
ǫ
m(cid:19)n−k
(cid:3)
In both cases, αk ≥ λk, so Sn
4. Groupoid crossed products and conditional expectations
In order to keep the paper as self-contained as possible, we now briefly review of the
theory of groupoid crossed products introduced in [41]. There are several excellent
treatments in the literature [16,29], but we take the approach of [36], which strongly
emphasises the bundle structure.
The main ideas for the reader to take away from this section are that we are trying
to get our hands on formal sums which interact in a similar manner as elements of a
group crossed product, but which respect the fibred structure of the groupoid. The
end of the section contains some new material, in which we consider a conditional
expectation on groupoid crossed products and use this to generalise a lemma of [39]
for use in our main result (see pages 24 -- 28).
In order to better focus on our examples of interest, we restrict our development to
the topological setting, where the crossed product arises from the action of a groupoid
on a space.
20
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
Definition 4.1 ([16, Definition 3.12]). Let Y be a locally compact Hausdorff space.
We say that a C ∗-algebra A is a C0(Y )-algebra if there exists a ∗-homomorphism ΦA
from C0(Y ) into the center of the multiplier algebra ZM(A) which is nondegenerate
in the sense that the set
ΦA(C0(Y )) · A := span{ΦA(f )a f ∈ C0(Y ), a ∈ A}
is dense in A.
In the case that Y is compact, we instead say that A is a C(Y )-algebra.
Suppose that G is a groupoid acting continuously on a compact Hausdorff space X,
with the associated continuous surjection rX : X → G(0) (observe that this implies
that G(0) is compact). Then A = C(X) becomes a C(G(0))-algebra under the ∗-
homomorphism ΦA(f ) = f ◦ rX.
For u ∈ G(0), let Iu ⊂ C(X) denote the ideal of functions which vanish on r−1
X (u).
Denote the quotient A(u) := A/Iu, and the image of a ∈ A under the associated
quotient map by a(u). Observe that A(u) ∼= C(r−1
X (u)). Denote the disjoint union
A := Fu∈G(0) A(u), and associate to A the projection map p : A → G(0) sending
A(u) to u. We think of A as a bundle of C ∗-algebras over G(0) and use [57, Theo-
rem C.27] to equip A with a topology so that the map a 7→ (u 7→ a(u)) becomes an
isomorphism from A to the set of continuous sections Γ(G(0), A) := {f : G(0) → A
f is continuous and p(f (u)) = u for every u ∈ G(0)}.
The action G y X induces an action α : G y C(X) via αg(f )(x) = f (g−1x), and
the triple (C(X), G, α) is an example of a groupoid dynamical system [36]. Thus, for
each g ∈ G, we obtain an isomorphism αg : A(s(g)) → A(r(g)) between fibres of the
bundle A.
Now, suppose that G is locally compact and Hausdorff, so that it admits a Haar
system {λu}u∈G(0). Consider the pull-back bundle r∗A of A by the range map r :
G → G(0). This is a bundle over G with fibres r∗A(g) ∼= A(r(g)). We equip the set
of continuous compactly supported sections Γc(G, r∗A) with the operations
The (reduced) groupoid crossed product will be the completion of Γc(G, r∗A) in the
(reduced) norm arising from a certain class of representations, which we now detail.
We remark that as in [36] we could instead immediately define a (non-C ∗) norm
k · kI on Γc(G, r∗A) directly and define the crossed product to be the enveloping
C ∗-algebra. However, we will need to make use of the representations we are about
to define sooner or later, so we give a full exposition here. As in the group case, we
will obtain a representation of Γc(G, r∗A) by combining representations of A and of
G into an integrated form. Due to the fibrations involved in the groupoid case, we
must represent onto bundles of Hilbert spaces.
Definition 4.2 ([16, Definition 3.61]). Suppose H = {H(y)}y∈Y is a collection of
separable nonzero complex Hilbert spaces indexed by an analytic Borel space Y . The
total space is defined to be
Y ∗ H := {(y, h) y ∈ Y, h ∈ H(y)}
and is equipped with the obvious projection map π : Y ∗ H → Y . We say that Y ∗ H
is an analytic Borel Hilbert bundle if it is equipped with a σ-algebra which makes it
an analytic Borel space (see [57, Appendix D]) such that
f ∗ g(h) =ZG
f (t)αt(g(t−1h))dλr(h)(t)
and f ∗(h) = αh(f (h−1)∗).
CLASSIFICATION OF TILING C ∗-ALGEBRAS
21
a) π is measurable; and
b) there is a sequence {fn}n∈N of sections (called a fundamental sequence for
Y ∗ H) such that
(i) the maps ¯fn : Y ∗ H → C given by
¯fn(y, h) = hfn(y), hiH(y)
are measurable (with respect to the Borel σ-algebra on C) for each n;
(ii) for each n and m, the map Y → C given by
y 7→ hfn(y), fm(y)iH(y)
is measurable; and
(iii) the collection of functions {π} ∪ { ¯fn}n∈N separates points of Y ∗ H.
The set of measurable sections of an analytic Borel Hilbert bundle Y ∗ H is denoted
by B(Y ∗ H).
Notation 4.3. Given Hilbert spaces H1 and H2, denote the the collection of unitary
transformations U : H1 → H2 by U(H1, H2).
Definition 4.4. If Y ∗ H is an analytic Borel Hilbert bundle then its isomorphism
groupoid is defined as
Iso(Y ∗ H) := {(x, V, y) V ∈ U(H(y), H(x))}
equipped with the smallest σ-algebra such that the map (x, V, y) 7→ hV f (y), g(x)iH(x)
is measurable for all measurable sections f, g ∈ B(Y ∗ H).
We define the set of composable pairs as
Iso(Y ∗ H)(2) := {((x, V, y), (w, U, z)) ∈ Iso(Y ∗ H) × Iso(Y ∗ H) y = w}
and the operations are given by
(x, V, y)(y, U, z) = (x, V U, z)
and (x, V, y)−1 = (y, V ∗, x).
Definition 4.5. Let G be a locally compact ´etale groupoid. A groupoid represen-
tation of G is a triple (µ, G(0) ∗ H, U) where µ is a (finite) Radon measure on G(0),
G(0) ∗ H is an analytic Borel Hilbert bundle, and U : G → Iso(G(0) ∗ H) is a mea-
surable groupoid homomorphism (with respect to the Borel σ-algebra on G) such
that, for each g ∈ G, there exists a unitary Ug : H(s(g)) → H(r(g)) such that
U(g) = (r(g), Ug, s(g)).
Definition 4.6 ([16, Definition 3.80]). Suppose that Y ∗H is an analytic Borel Hilbert
bundle and µ is a measure on Y . Consider
L2(Y ∗ H, µ) :=(cid:26)f ∈ B(Y ∗ H) (cid:12)(cid:12)(cid:12)(cid:12) ZY
kf (y)k2
H(y)dµ(y) < ∞(cid:27) .
Equip L2(Y ∗H, µ) with the operations of pointwise addition and scalar multiplication
and let L2(Y ∗ H, µ) be the quotient of L2(Y ∗ H, µ) such that sections which agree
µ-almost everywhere are identified. When equipped with the operations inherited
from L2(Y ∗ H, µ) and the inner product
hf, gi =ZY
hf (y), g(y)iH(y)dµ(y),
L2(Y ∗ H, µ) becomes a Hilbert space, which we call the direct integral of H with
respect to µ.
22
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
Definition 4.7 ([16, Definition 3.85]). Let Y ∗ H be an analytic Borel Hilbert bundle
and µ a finite measure on Y . An operator T on L2(Y ∗ H, µ) is called diagonal if
there exists a bounded measurable function φ : Y → C such that
T (h)(y) = φ(y)h(y)
for µ-almost every y ∈ Y . Given such a function φ, the associated diagonal operator
is denoted by Tφ.
Definition 4.8 ([16, Definition 3.98]). Let Y be a second countable locally compact
Hausdorff space, A a C0(Y )-algebra, Y ∗ H an analytic Borel Hilbert bundle, and µ
a finite Borel measure on Y . We say that a representation π : A → B(L2(Y ∗ H, µ))
is C0(Y )-linear if
for every a ∈ A and φ ∈ C0(Y ).
π(φ · a) = Tφπ(a)
Definition 4.9. Suppose Y ∗H is an analytic Borel Hilbert bundle with a fundamental
sequence {fn}. A family of bounded linear operators T (y) : H(y) → H(y) is a Borel
field of operators if the map
is measurable with respect to the Borel σ-algebra on C for all m and n.
y 7→ hT (y)(fn(y)), fm(y)iH(y)
The next two propositions detail the formation of a C0(Y )-linear representation of
A onto L2(Y ∗ H, µ) from representations of each A(y) onto H(y), and the decompo-
sition of such a representation back into its fibrewise representations.
Proposition 4.10 ([16, Proposition 3.93]). Suppose Y is a second countable locally
compact Hausdorff space, A a separable C0(Y )-algebra, Y ∗H an analytic Borel Hilbert
bundle, and µ a σ-finite Borel measure on Y . Given a collection of representations
{πy : A(y) → B(H(y)) y ∈ Y } such that for each a ∈ A the set {πy(a(y)) y ∈ Y }
is a Borel field of operators, we can form a representation
of A on L2(Y ∗ H, µ) called the direct integral, and defined for a ∈ A by
πydµ(y)
Y
π =Z ⊕
π(a) =Z ⊕
Y
πy(a(y))dµ(y).
Proposition 4.11 ([16, Proposition 3.99]). Suppose Y is a second countable locally
compact Hausdorff space, A a separable C0(Y )-algebra, Y ∗H an analytic Borel Hilbert
bundle, and µ a finite Borel measure on Y . Given a C0(Y )-linear representation π :
A → L2(Y ∗ H, µ), there exists for each y ∈ Y a (possibly degenerate) representation
πy : A(y) → B(H(y)) such that for each a ∈ A, the set {πy(a(y))}y∈Y is an essentially
bounded Borel field of operators, and
π =Z ⊕
Y
πydµ(y).
Definition 4.12 ([16, Definition 3.70]). Suppose G is a locally compact Hausdorff
groupoid with Haar system λ = {λu}, where λu is supported on Gu = r−1(u). Denote
the images of these measures under the inverse map of the groupoid by λu = (λu)−1,
so that λu is supported on Gu = s−1(u). Given a Radon measure µ on G(0), we
CLASSIFICATION OF TILING C ∗-ALGEBRAS
23
define the induced measures ν and ν−1 to be the Radon measures on G defined by
the equations
ν(f ) :=ZG(0)ZG
ν−1(f ) :=ZG(0)ZG
f (g)dλu(g)dµ(u),
and
f (g)dλu(g)dµ(u)
for all f ∈ Cc(G). We say that the measure µ is quasi-invariant if ν and ν−1 are
mutually absolutely continuous.
In this case we write ∆ for the Radon-Nikodym
derivative dν/dν−1, and call it the modular function of µ.
Definition 4.13. Suppose (A, G, α) is a separable groupoid dynamical system, and
that µ is a quasi-invariant measure on G(0). A covariant representation (µ, G(0) ∗
H, π, U) of (A, G, α) consists of a unitary representation (µ, G(0) ∗ H, U) of G, and a
C0(G(0))-linear representation π : A → B(L2(G(0) ∗ H, µ)). We require that if {πu}
is a decomposition of π, and ν is the measure induced by µ, there exists a ν-null set
N ⊂ G such that for all g /∈ N we have
for every a ∈ A(s(g)).
Ugπs(g)(a) = πr(g)(αg(a))Ug
Remark 4.14. We will usually simply denote the covariant representation by (π, U)
and take for granted that there is a quasi-invariant measure µ and a Borel Hilbert
bundle G(0) ∗ H in the background.
H, π, U) is a covariant representation. Let π = R ⊕
Definition 4.15. Suppose (A, G, α) is a separable dynamical system and that (µ, G(0)∗
G(0) πudu be a decomposition of π.
Then there is a I-norm decreasing nondegenerate ∗-representation π⋊U of Γc(G, r∗A)
on L2(G(0) ∗ H, µ) called the integrated form of (π, U), and given by
π ⋊ U(f )(h)(u) =ZG
πu(f (g))Ugh(s(g))∆(g)− 1
2 dλu(g)
for f ∈ Γc(G, r∗A), h ∈ L2(G(0) ∗ H, µ) and u ∈ G(0).
With all this theory in hand, we can define the full crossed product as the comple-
tion of Γc(G, r∗A) in the universal norm
kf k := sup{kπ ⋊ U(f )k (π, U) is a covariant representation of (A, G, α)}.
We will instead work with the reduced crossed product, for which we need to de-
fine left-regular representations. Begin with a separable groupoid dynamical system
(A, G, α), and a nondegenerate C0(G(0))-linear representation π : A → B(L2(G(0) ∗
H, µ)) with decomposition
π =Z ⊕
G(0)
πudµ(u)
so that πu : A(u) → B(L2((G(0) ∗ H)u, µ)). As in [16, Example 3.84], form the
pull-back bundle
s∗(G(0) ∗ H) = {(g, h) h ∈ H(s(g))}.
Let ν be the induced measure, as in Definition 4.12. By [57, Theorem I.5], we can
use the source map of G to decompose ν into finite measures νu on Gu. Denote by
νu the image of νu under inversion in G. For each u ∈ G(0) we can obtain a separable
Hilbert space K(u) = L2(s∗(G(0) ∗ H)Gu, νu), by considering the integrable sections
of the bundle s∗(G(0) ∗ H) which are supported on Gu. We use these as fibres to form
24
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
a Borel Hilbert bundle G(0) ∗ K. Denote by ∆ the modular function which arises from
µ. As in [16, Example 3.105], for each g ∈ G, one can obtain a unitary
λg : L2(s∗(G(0) ∗ H)Gs(g), νs(g)) → L2(s∗(G(0) ∗ H)Gr(g), νr(g))
by defining λgf (h) = ∆(g)1/2f (g−1h) for each h ∈ Gr(g).
Also, define π : A → B(L2(G(0) ∗ K, µ)) by
(π(a)f (u))(g) = πs(g)(α−1
g (a(u))) (f (u)(g))
where a ∈ A, f ∈ L2(G(0) ∗ K, µ), u ∈ G(0) and g ∈ Gu.
One can show that (π, λ) defines a covariant representation of (A, G, α). Covariant
representations of this special form are known as left regular representations. We now
discuss the integrated form π⋊λ of such representations, which will be representations
of Γc(G, r∗A) on L2(s∗(G(0) ∗ H), ν−1). In fact, instead of working directly with the
integrated form π ⋊ λ, we work with the following representation, which is unitarily
equivalent to the integrated form [16, Example 3.122 and Remark 3.123]:
Lπ(f )h(g) =ZGr(g)
πs(g)(α−1
g (f (t)))h(t−1g)dλr(g)(t),
where f ∈ Γc(G, r∗A). h ∈ L2(s∗(G(0) ∗ H), ν−1), and g ∈ G.
Due to the equivalence of L and π ⋊ λ, we refer to Lπ as the integrated left regular
representation associated to π. When the choice of C0(G(0))-linear representation is
clear, we will denote this representation simply by L.
Definition 4.16. Suppose (A, G, α) is a separable groupoid dynamical system. We
define the reduced norm on Γc(G, r∗A) by
kf kr := sup{kL(f )k L is an integrated left regular representation of (A, G, α)}.
The completion of Γc(G, r∗A) in this norm is a C ∗-algebra called the reduced groupoid
crossed product of A by G, and is denoted A ⋊α,r G, or just A ⋊r G if the choice of
action is clear.
For the remainder of the section, we extend some results of [39] to the groupoid
setting. We will need the following definition, which is introduced in [7] (see also
[39, Remark 1.2] for this formulation).
Definition 4.17. Let A be a C ∗-algebra and let a, b ∈ A be positive elements. We
say that a is Cuntz subequivalent to b over A, and write a -A b, if there exists a
sequence (rn)n∈N in A with a = limn→∞ rnbr∗
n.
We aim to show (Lemma 4.20) that for any positive element a ∈ C(X) ⋊r G, we
can find a positive element of C(X) which is Cuntz subequivalent to a. This will be
useful to prove Z-stability in Theorem 8.4.
We make a remark on notation. In the sequel, it will be important to distinguish
between a section of the pull-back bundle, f ∈ Γc(G, r∗A), and the images of elements
of G under this section, fg ∈ C(r−1
X (r(g))). To do this, we will denote the section
g 7→ fg for each g ∈ G byPg∈G fgUg. Observe that this notation is highly suggestive
of the image of the section under the integrated form of some covariant representation
(π, U), and, in fact, we can multiply sections as follows:
Xg∈G
agUgXh∈G
bhUh = X(g,h)∈G(2)
agUgbhUh = X(g,h)∈G(2)
agαg(bh)Ugh.
CLASSIFICATION OF TILING C ∗-ALGEBRAS
25
Lemma 4.18. Let f ∈ Γc(G, r∗A), and write f =Pg∈G fgUg. Then the map
Eα(f ) := Xu∈G(0)
fu
extends to a faithful conditional expectation E : C(X) ⋊r G → C(X).
Proof. The proof of most of the properties is routine, except perhaps for faithfulness,
for which the reader is referred to [47, Lemma 1.2.1]. We check that the image of
E is contained within C(X). Given f ∈ Γc(G, r∗A), write f = Pg∈G fgUg. By
[16, Proposition 4.38], there is an isomorphism ι : Cc(G ⋉ X) → Γc(G, r∗A), with
inverse j given by j(f )(g, x) = fg(gx). Since G is ´etale and Hausdorff, G(0) is clopen
in G. This implies that f G(0) ∈ Γc(G, r∗A), so that j(f G(0)) ∈ Cc(G ⋉ X).
In
X (r(g)) unless g ∈ G(0), observe that j(f G(0)) is supported
fact, because (f G(0))g = 0r−1
on G(0) ∗ X, which is homeomorphic to X via Φ(x) = (rX(x), x). Thus, under the
identification of G(0) ∗ X with X, we see that j(f G(0))G(0)∗X ∈ C(X). On the other
hand, for each x ∈ X,
E(f )(x) = frX (x)(x) = j(f G(0))(rX(x), x) = j(f G(0))(Φ(x)),
so E(f ) = j(f G(0)) ◦ Φ, which shows that E(f ) ∈ C(X) too.
(cid:3)
We remark for later that an almost identical argument works to show that for
any clopen bisection B ⊂ G, the sum Pg∈B fg defines an element of C(X). The only
modification that needs to be made is in the final equality, where we must precompose
j(f B) with the homeomorphism (rB∗X)−1 arising from r : G ⋉ X → X, to obtain
fg = j(f B) ◦ (rB∗X)−1 ◦ Φ ∈ C(X).
Xg∈B
Lemma 4.19 (c.f.
[39] Lemma 7.8). Let G be a locally compact Hausdorff ´etale
groupoid acting freely and continuously on a compact Hausdorff space. Let a ∈
Γc(G, r∗A) ⊂ C(X) ⋊ G, and let ǫ > 0. Then there exists f ∈ C(X) such that
0 ≤ f (x) ≤ 1 for every x ∈ X, f a∗af ∈ C(X), and kf a∗af k ≥ kE(a∗a)k − ǫ.
Proof. Write b = a∗a. If kE(b)k ≤ ǫ, then we can simply take f = 0, so suppose
there exists x ∈ X such that E(b)(x) > ǫ. Since a is a compactly supported section,
and the product of compact subsets of G is always compact, there exists a compact
subset K ⊂ G, and bg ∈ C(r−1
Since b is positive, E(b) is too. Let
X (r(g))) for each g ∈ K, such that b = Pg∈K bgUg.
U = {x ∈ X E(b)(x) > kE(b)k − ǫ},
which is a nonempty open subset of X. We wish to find a nonempty open set W ⊂ U
such that the sets in the collection {gW g ∈ K} are pairwise disjoint.
Notice that for g, h ∈ K, gW ∩ hW 6= ∅ if and only if there exist w1, w2 ∈ W
such that gw1 = hw2 if and only if w1 = g−1hw2 if and only if W ∩ K −1KW 6= ∅,
so it is equivalent to ask that whenever g, h ∈ K are such that g−1h /∈ G(0), we have
W ∩ g−1hW = ∅.
Since G is ´etale, the topology on G has a base of open bisections {Bi}i∈I. Therefore,
{Bi}i∈I is an open cover of K −1K, which is the product of compact subsets in an
´etale groupoid, and so is compact itself by [19, Lemma 5.2]. So, there exists a finite
subcover {B1, . . . , BN } of K −1K.
In fact, since G is ´etale and Hausdorff, G(0) is
a clopen bisection. Therefore, by putting B0 = G(0) and Bk = Bk \ G(0) for each
k ∈ {1, . . . , N}, then redefining N and relabelling, we may assume that B1 = G(0)
26
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
and that Bi ∩ G(0) = ∅ for each i ∈ {2, . . . , N}. We will construct an open W ⊂ U
such that W ∩SN
k=2 BkW = ∅.
Now, we wish to find a subset V ⊂ U such that, for each k ∈ {1, . . . , N}, either
rX(V ) ⊂ s(Bk), or rX(V )∩s(Bk) = ∅. In other words, for each k, either every element
of V will be acted upon by Bk, or none of them will be. We will construct a nested
sequence of nonempty open sets U ⊃ V1 ⊃ V2 ⊃ · · · ⊃ VN = V so that, for each
k ∈ {1, . . . , N}, and each 1 ≤ j ≤ k, either rX(Vk) ⊂ s(Bj), or rX(Vk) ∩ s(Bj) = ∅.
Observe that since B1 = G(0), the choice V1 = U works.
Now, suppose we have constructed open sets U = V1 ⊃ V2 ⊃ · · · ⊃ Vk−1 as above
for some 2 ≤ k ≤ N, and proceed in cases as follows.
Case 1: If rX(Vk−1) ∩ s(Bk) = ∅, then set Vk = Vk−1.
Case 2: If rX(Vk−1) ∩ s(Bk) 6= ∅, then set Vk = Vk−1 ∩ r−1
X (s(Bk)). Observe that Vk is
X (s(Bk)) is open, since Bk is open, s is an open map, and rX
open because r−1
is continuous.
Choose any x ∈ V = VN . For each k ∈ {1, . . . , N}, since Bk is a bisection, there
k=1 Bkx ⊂ X is
finite with cardinality at most N. Since the action is free, and since Bk ∩ G(0) = ∅
k=2 BkrX(x) we have gx 6= x. Since X is
Hausdorff, for each such gx = y, we can find disjoint open subsets x ∈ Vy and y ∈ Zy.
is at most one element g ∈ Bk such that s(g) = rX(x), so the set SN
whenever 2 ≤ k ≤ N, for each g ∈ SN
Let V ′ = V ∩(cid:18)Ty∈SN
for each y ∈SN
every g ∈Sk
We now wish to find a second finite sequence of nested open subsets V ′ ⊃ W1 ⊃
W2 ⊃ · · · ⊃ WN ∋ x, such that, for each k ∈ {2, . . . , N}, we have Wk ∩ gWk = ∅ for
j=2 Bj. Then W = WN will be such that gW ∩ hW = ∅ for all g 6= h ∈ K,
k=2 Bkx Vy(cid:19), which is nonempty since x ∈ V ′, and is open in X
because it is the intersection of finitely many open sets. Observe also that V ′ ∩Zy = ∅
as we wanted. Start with W1 = V ′, and proceed inductively as follows.
Suppose that we have constructed V ′ = W1 ⊃ · · · ⊃ Wk−1 ∋ x as above for some
2 ≤ k ≤ N, and construct Wk by proceeding in cases as follows. First, observe that
since Wk−1 ⊂ V , either rX(Wk−1) ⊂ s(Bk), or rX(Wk−1) ∩ s(Bk) = ∅.
k=2 Bkx.
Case 1: Suppose rX(Wk−1) ∩ s(Bk) = ∅. This means that BkWk−1 = ∅, so we have
j=2 Bj we have
j=2 Bj we still have Wk−1∩gWk−1 = ∅.
gWk−1 = ∅ for each g ∈ Bk. We assumed that for each g ∈Sk−1
Wk−1∩gWk−1 = ∅, and so for each g ∈Sk
Thus, we see that the choice Wk = Wk−1 is suitable.
Case 2: Suppose rX(Wk−1) ⊂ s(Bk). Then, since x ∈ Wk−1, there exists g ∈ Bk such
that s(g) = rX(x). Write y = gx. Since Bk is open in G, and since G is ´etale,
we see by Lemma 2.7 that BkWk−1 is an open subset of X. Thus, Zy and
BkWk−1 are open subsets of X which both contain y, so that Zy ∩ BkWk−1 is
a nonempty open set in X. Put Wk = B−1
k (Zy ∩ BkWk−1) ⊂ Wk−1. Observe
that x ∈ Wk, and Wk is open by Lemma 2.7, because B−1
is open in X.
k
All we have to check is that for each g ∈ Bk we have Wk ∩ gWk = ∅ (note
j=2 Bj, because Wk ⊂ Wk−1). Indeed,
that this automatically holds for g ∈Sk−1
gWk ⊂ BkWk ⊂ Zy, which is disjoint from V ′ ⊃ Wk, as required.
So, construct an open set W ⊂ U such that the collection {gW g ∈ K} is pairwise
disjoint. Note in particular that, by construction of K, r(K) ⊂ K, so that r(g)W
and gW are disjoint for each g ∈ K. Choose x0 ∈ W , and let f ∈ C(X) satisfy
CLASSIFICATION OF TILING C ∗-ALGEBRAS
27
0 ≤ f (x) ≤ 1 for every x ∈ X, supp(f ) ⊂ W , and f (x0) = 1. Then
f bf = Xg∈K
f bgUgf = Xg∈K
f bgαg(f )Ug.
Since supp(f ) ⊂ W and supp(αg(f )) ⊂ gW , observe that f bgαg(f ) = bgf αg(f ) is
supported inside W ∩gW . Since gW ⊂ r−1
X (r(g)), we have W ∩gW = r(g)(W ∩gW ) =
r(g)W ∩ gW , and we thus see that f bgαg(f ) is supported within r(g)W ∩ gW . So,
for g ∈ K \ G(0), our construction of W provides f bgαg(f ) = 0. Thus, we have
f bf = Xu∈K∩G(0)
f buαu(f )Uu = Xu∈K∩G(0)
f buf r−1
X (u),
where we have identified the central expression with a complex-valued function on
X. Since supp(bu) ⊂ r−1
X (u), we see that f buf r−1
X (u) = f buf , so that
f bf = f Xu∈G(0)
In addition, we have
bu f = f E(b)f ∈ C(X).
kf bf k ≥ f bf (x0) = f (x0)brX (x0)(x0)f (x0) = brX (x0)(x0) = E(b)(x0) > kE(b)k−ǫ,
where the final inequality follows from the fact that x0 ∈ U.
(cid:3)
[39, Lemma 7.9]). Let G y X be a free continuous action of
Lemma 4.20 (c.f.
a locally compact Hausdorff ´etale groupoid on a compact Hausdorff space. Let B ⊂
C(X) ⋊r G be a unital subalgebra such that
(i) C(X) ⊂ B; and
(ii) B ∩ Γc(G, r∗A) is dense in B.
Let a ∈ B+ \ {0}. Then there exists b ∈ C(X)+ \ {0} such that b -B a, where -B
denotes the relation of Cuntz subequivalence over B.
Proof. Without loss of generality, we may assume kak ≤ 1. Since the conditional
expectation E : C(X) ⋊r G → C(X) is faithful, E(a) ∈ C(X) is a nonzero positive
element. Let ǫ = 1
8kE(a)k and note that ǫ ≤ 1, since kE(a)k ≤ kak. Since B ∩
Γc(G, r∗A) is dense in B, choose c ∈ B ∩ Γc(G, r∗A) such that kc − a1/2k < ǫ.
Without loss of generality, we may choose c such that kck ≤ 1. Notice that
ǫ2 > kc − a1/2k2 = k(c − a1/2)(c − a1/2)∗k
= kcc∗ + a − ca1/2 − c∗a1/2k
≥ kcc∗ − ak − k2a − ca1/2 − c∗a1/2k
≥ kcc∗ − ak − ka1/2kk2a1/2 − c − c∗k,
so that
kcc∗ − ak < ǫ2 + ka1/2kka1/2 − ck + ka1/2kk(a1/2 − c)∗k < 3ǫ.
Similarly, kc∗c − ak < 3ǫ. Since E is norm decreasing, we see that kE(c∗c − a)k < 3ǫ,
which implies that kE(a)k < kE(c∗c)k + 3ǫ. Applying Lemma 4.19 with c in place of
a, and with the ǫ defined above, yields f ∈ C(X) with 0 ≤ f ≤ 1, which satisfies
kf c∗cf k ≥ kE(c∗c)k − ǫ > kE(a)k − 4ǫ = 8ǫ − 4ǫ = 4ǫ.
Therefore, (f c∗cf − 3ǫ)+ is a nonzero positive element of C(X). To conclude the
proof, we apply [39, Lemma 1.4(6)] at the first step, [39, Lemma 1.7] and the fact
28
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
that cf 2c∗ ≤ cc∗ because 0 ≤ f ≤ 1 at the second step, and [39, Lemma 1.4(10)] and
kcc∗ − ak < 3ǫ at the final step, to see that
(f c∗cf − 3ǫ)+ ∼B (cf 2c∗ − 3ǫ)+ -B (cc∗ − 3ǫ)+ -B a.
(cid:3)
5. Tilings and their dynamics
In this section, we define tilings of Rd and introduce the properties on tilings
that are used throughout this paper. These assumptions lead to a compact space of
tilings with a free and minimal d-dimensional flow, giving rise to a dynamical system
of tilings. Following Kellendonk [25], we consider a closed subset of this dynamical
system that lends itself to defining a rather tractable groupoid and its corresponding
C ∗-algebra. A more complete exposition of the material presented here can be found
in [25,26,56].
We begin by describing the basic building blocks of the tilings we consider.
Definition 5.1. A prototile p is a labelled subset of Rd that is homeomorphic to
the closed unit ball. A prototile set is a finite set of prototiles, typically denoted by
P := {p1, . . . , pn}.
Note that prototiles are labelled subsets of Rd. So we may have two identical subsets
with different labels, making them different prototiles. We will abuse notation and
let the symbol p denote both a subset of Rd and a labelled subset of Rd.
Definition 5.2. A tiling T with prototile set P is a set {t1, t2, . . .} of subsets of Rd,
called tiles, such that
(1) for each i ∈ {1, 2, . . .}, there exists xi ∈ Rd and pi ∈ P such that ti = pi + xi;
(3) int(ti) ∩ int(tj) = ∅ for i 6= j.
A patch P in a tiling T is a finite connected subset of tiles. Given a tiling T , x ∈ Rd
and r > 0, we define the patch T ⊓ Br(x) by
T ⊓ Br(x) := {t ∈ T t ∩ Br(x) 6= ∅}.
These types of patches will be useful for putting a metric on tilings, among other
things. Given a tiling T and x ∈ Rd, we define T + x := {t + x t ∈ T } and use this
to define the set of tilings T + Rd := {T + x x ∈ Rd}.
Definition 5.3. Suppose T is a tiling. Given T1, T2 ∈ T + Rd and 0 < ε < 1, we say
that T1 and T2 are ε-close if there exist x1, x2 ∈ Rd such that kx1k, kx2k ≤ ε, and
such that
(T1 + x1) ⊓ Bε−1(0) = (T2 + x2) ⊓ Bε−1(0).
Define d(T1, T2) to be the infimum of the set of ε which satisfy this hypothesis. If no
such ε exists, set d(T1, T2) = 1.
The idea of the tiling metric is to extend the usual product metric in symbolic
dynamics to the continuous situation of tilings. So two tilings are close if they agree
on a large ball about the origin up to a small translation.
Definition 5.4. Suppose T is a tiling. The continuous hull ΩT of T is the completion
of T + Rd in the tiling metric.
We now introduce several properties of a tiling that we use throughout the paper.
i=1 ti = Rd; and
(2) S∞
CLASSIFICATION OF TILING C ∗-ALGEBRAS
29
Definition 5.5. Suppose T is a tiling.
(1) T is said to have finite local complexity (FLC) if, for every R > 0, the set
{T ⊓ BR(x) x ∈ Rd}/Rd is finite.
(2) T is said to be nonperiodic if T + x = T implies x = 0. We say T is aperiodic
if every tiling in the continuous hull ΩT is nonperiodic.
(3) T is said to be repetitive if, for every patch P ⊂ T , there exists R > 0 such
that, for every x ∈ Rd, a translate of P appears in T ⊓ BR(x).
If T has FLC, then ΩT is a compact Hausdorff space. In [25], it is shown that ΩT
is a space whose points are tilings and that Rd acts continuously on ΩT . So we think
of (ΩT , Rd) as a dynamical system. If T is aperiodic, then the action of Rd on ΩT
is free and if T is repetitive, the action is minimal. In this case, the continuous hull
does not depend on the original tiling T used to construct it, in the sense that the
continuous hull of any other tiling in ΩT will also be ΩT . Therefore, we begin to drop
the "T " from the notation ΩT , and simply denote the continuous hull by Ω.
One of Kellendonk's inspired simplifications of tiling spaces is a quantisation of
the continuous hull [25, Section 2.1]. Suppose Ω is the continuous hull of a tiling
with prototile set P. For each prototile p ∈ P, distinguish a point in its interior and
denote it by x(p). We call x(p) a puncture. Now suppose T is a tiling in Ω. Since
each tile t ∈ T satisfies t = p + y for some y ∈ Rd and p ∈ P , we extend the punctures
to tiles by x(t) = x(p) + y. Thus, every tile in every tiling in Ω can be punctured.
We use these punctures to restrict the possible alignments of tilings by forcing the
origin to lie on a puncture.
Definition 5.6 ([25, Section 2.1]). Suppose Ω is the continuous hull of an FLC tiling.
The discrete hull Ωpunc of a tiling consists of all tilings T ∈ Ω for which the origin is
a puncture of some tile t ∈ T .
We now begin to assume that each prototile p ∈ P lies in Rd with its puncture on
the origin. Suppose T is a tiling in Ωpunc. We denote the (unique) tile with puncture
on the origin by T (0). We remark that Ωpunc is a Cantor set, as shown in [26]. A
neighbourhood base for the Cantor set topology is given by sets of the form
U(P, t) := {T ∈ Ωpunc P − x(t) ⊂ T }.
6. Tiling groupoids and Kellendonk's C ∗-algebra of a tiling
We now describe Kellendonk's construction of an ´etale groupoid associated to a
tiling [25].
Suppose Ωpunc is the discrete hull of a repetitive and aperiodic tiling with FLC. An
´etale principal groupoid is defined using translational equivalence on Ωpunc:
Rpunc := {(T, T − x(t)) T ∈ Ωpunc and t ∈ T }.
Then Rpunc is a groupoid with inverse and partially defined product
(T, T ′)−1 = (T ′, T )
and
(T, T ′) · (T ′, T ′′) := (T, T ′′).
The unit space of Rpunc is Ωpunc, and the source and range maps s, r : Rpunc → Ωpunc
are defined by s(T, T ′) := T ′ and r(T, T ′) := T .
There is a metric on Rpunc defined by
dR(cid:16)(T1, T1 − x(t1)), (T2, T2 − x(t2))(cid:17) = d(T1, T2) + kx(t1) − x(t2)k.
30
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
(6.1)
and let
(6.2)
An equivalent topology is given as follows. Let P be a patch of tiles and t, t′ ∈ P , so
that both P − x(t) and P − x(t′) are patches with puncture on the origin. Consider
the set of tilings
V (P, t, t′) := {(cid:16)T − (x(t′) − x(t)), T )(cid:17) P − x(t) ⊂ T }
V := {V (P, t, t′) P is a patch from a tiling in Ωpunc and t, t′ ∈ P }.
The collection V is a neighbourhood base of compact open sets; see for example
[56, Lemma 2.5 and 2.6]. Lemma 2.6 in [56] shows that the range and source maps are
local homeomorphisms satisfying s(V (P, t, t′)) = U(P, t) and r(V (P, t, t′)) = U(P, t′).
This implies that Rpunc is an ´etale groupoid.
In [25,26,56], a groupoid C ∗-algebra Apunc is associated to Rpunc. Rather than go
into the specifics of this construction, we present a set of elements that span a dense
subalgebra of Apunc, and a faithful and nondegenerate representation of Apunc on a
Hilbert space. The details of this specific construction can be found in [25, Section
2.2].
To construct a dense spanning class in Apunc we use the neighbourhood base V from
(6.2) to construct partial isometries. We set e(P, t, t′) = 1V (P,t,t′) to be the indicator
function of the clopen set V (P, t, t′). Explicitly, for a pair (T ′, T ) ∈ Rpunc we have
(6.3)
e(P, t, t′)(T ′, T ) =
The collection
1
0
if P − x(t) ⊂ T and T ′ = T − (x(t′) − x(t))
otherwise.
(6.4)
E := {e(P, t, t′) P is a patch from Ωpunc and t, t′ ∈ P }
generates a dense subalgebra of Cc(Rpunc), and hence of Apunc. The proof of this
appears in [56, Proposition 3.3]. Moreover, these functions have the following prop-
erties.
Proposition 6.1. Suppose Ωpunc is the discrete hull of a repetitive and aperiodic
tiling with FLC, and E is the collection in (6.4). Then
(1) e(P, t, t′)∗ = e(P, t′, t); and
(2) e(P, t, t′) · e(P ′, t′, t′′) = e(P ∪ P ′, t, t′′) if P and P ′ agree where they intersect.
To define a representation of Cc(Rpunc) as bounded operators on a Hilbert space,
we follow the development in [25, Section 2.2]. Suppose T is any tiling in Ωpunc and
let us identify the tile t ∈ T with the tiling T − x(t) ∈ Ωpunc. Then the induced
representation from the unit space πT : Cc(Rpunc) → B(ℓ2(T )) acts on elements of
the spanning family E as follows. For q ∈ T with P − x(t) ⊂ T − x(q), there exists
a tile q′ with puncture x(q′) = x(q) + (x(t′) − x(t)) such that P − x(t′) ⊂ T − x(q′),
and we have
(6.5)
(cid:16)πT (e(P, t, t′))ξ(cid:17)(q) = ξ(q′)
for ξ ∈ ℓ2(T ).
Since T is aperiodic and repetitive, the induced representation extends to a faithful
nondegenerate representation of Apunc on ℓ2(T ), as shown in [32, Section 4]. Restrict-
ing (6.5) to delta functions δt′′ ∈ ℓ2(T ) gives
(6.6)
δt′′′
0
if P − x(t) ⊂ T − x(t′′) and x(t′′′) = x(t′′) + (x(t′) − x(t))
otherwise.
πT (e(P, t, t′))δt′′ =
CLASSIFICATION OF TILING C ∗-ALGEBRAS
31
7. Almost finiteness of tiling groupoids
We now present a viewpoint of tiling groupoids as transformation groupoids as-
sociated to a special sort of groupoid action, and prove that they are almost finite
in the sense of [33, Definition 6.2] and satisfy the diameter condition appearing in
Theorem 2.15. We remark that we could simply apply Lemma 2.11 to obtain the di-
ameter condition, but our direct proof will be instructive, and will allow us to choose
our clopen towers to have a particularly nice form.
Given a free action of Rd on a compact metrisable space Ω, we will restrict the
action to a special subset of Ω of the following type.
Definition 7.1 ([15, Definition 2.1]). Let d ∈ N. Let ϕ be a free action of Rd on a
compact, metrisable space Ω. We call a closed subset X ⊂ Ω a flat Cantor transversal
if the following are satisfied.
(i) X is homeomorphic to a Cantor set.
(ii) For any x ∈ Ω, there exists p ∈ Rd such that ϕp(x) ∈ X.
(iii) There exists a positive real number M > 0 such that
C = {ϕp(x) x ∈ X, p ∈ BM (0)}
is open in Ω, and
X × BM (0) ∋ (x, p) 7→ ϕp(x) ∈ C
is a homeomorphism.
(iv) For any x ∈ X and r > 0 there exists an open neighbourhood U ⊂ X of x
such that {p ∈ Br(0) ϕp(x) ∈ X} = {p ∈ Br(0) ϕp(y) ∈ X} for all y ∈ U.
The following appears as [33, Remark 6.4] (see also the proof of [33, Lemma 6.3]).
Lemma 7.2 ([33, Remark 6.4]). Suppose that ϕ : Rd y Ω is a free action on a
compact metrisable space, and let X ⊂ Ω be a flat Cantor transversal. Construct an
´etale principal groupoid G as in [15] as follows:
G = {(x, ϕp(x)) p ∈ Rd and x, ϕp(x) ∈ X}.
Then G is almost finite.
Notice that when we consider the free action of Rd on the continuous hull of an
aperiodic and repetitive tiling with FLC, and set X = Ωpunc, the groupoid G con-
structed in the lemma above is isomorphic to Rpunc. So, in order to prove that Rpunc
is almost finite, it suffices to prove the following lemma.
Lemma 7.3. Let Ω be the continuous hull of an aperiodic and repetitive tiling with
FLC. Then Ωpunc ⊂ Ω is a flat Cantor transversal.
Proof. As shown in [26], Ωpunc is homeomorphic to a Cantor set. Given T ∈ Ω,
observe that ϕ−x(T (0))(T ) = T − x(T (0)) ∈ Ωpunc, so (ii) is satisfied.
Since T has FLC, the set of punctures in any tiling in Ω forms a Delone set in
Rd, and in particular is uniformly discrete. Furthermore, all the patches in any
tiling in Ωpunc already appear in the initial tiling that we chose, so we can choose
the same constants which witness uniform discreteness and relative density in all of
these Delone sets. Take M = r/2, where r is the chosen constant witnessing uniform
density, so that the distance between any two distinct punctures is smaller than r.
Form C as in Definition 7.1, and take a tiling T ∈ C. By definition, there exists
a unique p ∈ Rd such that kpk < M and ϕ−p(T ) ∈ Ωpunc. Existence is clear by
32
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
the definition of C, and uniqueness follows because if x, y have kxk, kyk < M and
T − x, T − y ∈ Ωpunc, then kx − yk ≤ kxk + kyk < 2M = r, which contradicts our
choice of r since x and y are puncture locations. Take 0 < ǫ < (M − kpk)/2 to
be small enough that ǫ−1 − ǫ > M. Then, when d(T, T ′) < ǫ, we can find vectors
x, y ∈ Rd with kxk, kyk < ǫ such that (T − x) ⊓ Bǫ−1(0) = (T ′ − y) ⊓ Bǫ−1(0). Since
T − x has a puncture at p − x, with
kp − xk ≤ kpk + kxk < kpk + (M − kpk)/2 < M < ǫ−1,
we see that T ′ − y also has a puncture at p − x, so that T ′ has a puncture at p − x + y.
Observe that
kp − x + yk ≤ kpk + kxk + kyk ≤ kpk + 2ǫ < kpk + 2(M − kpk)/2 = M,
and therefore T ′ ∈ C. This proves that C is open.
Since the action of Rd on Ω is continuous, the map defined in (iii) is clearly contin-
uous, and is clearly surjective by definition of C. As in the proof above, our choice
of M guarantees that for each T ∈ C there is a unique T ′ ∈ Ωpunc such that there
exists p ∈ BM (0) with ϕp(T ′) = T , and therefore the map is injective. To conclude
the proof that the map is a homeomorphism, we prove that it is open. It suffices
to show that the images of subsets of the form U(P, t) × Br(x) are open in Ω. So,
take T = ϕp(T ′) for some T ′ ∈ U(P, t) ⊂ Ωpunc and p ∈ Br(x) ⊂ BM (0). Choose
ǫ < (r − kp − xk)/2 small enough that P ⊂ Bǫ−1−ǫ−kpk(x(t)). Observe that Bǫ(T ) is
a subset of the image of U(P, t) × Br(x), which proves that the image is open.
Finally, for property (iv), choose T ∈ Ωpunc and r > 0. Let P ⊂ T be a patch
which contains Br(0) ⊂ T , and set U = U(P, t). Then whenever T ′ ∈ U, we see that
T and T ′ agree on Br(0), and thus the set of punctures in T which lie within Br(0)
is the same subset of Rd as the set of punctures in T ′ which lie within Br(0). This
shows that
{p ∈ Br(0) ϕp(T ) ∈ Ωpunc} = {p ∈ Br(0) ϕp(T ′) ∈ Ωpunc},
as required.
(cid:3)
Lemma 7.4. Let T be an aperiodic and repetitive tiling with FLC. Then the tiling
groupoid Rpunc is almost finite. Furthermore, the subsets F (i)
in the definition of the
j
fundamental domain of the elementary subgroupoid can be chosen to be arbitrarily
small in diameter.
Proof. We have already shown that Rpunc is almost finite in the sense of Defini-
tion 2.14, so all that remains is to show that we can arrange for the small diameter
condition to be satisfied. We make use of the proof of Theorem 2.15 to phrase the
almost finiteness of this groupoid in terms of clopen tower decompositions of Ωpunc
(see Definition 2.10). We will use a similar argument as in the proof of Lemma 2.11
to show that we can iteratively modify any given tower decomposition to obtain the
diameter condition.
Let C ⊂ Rpunc be compact and open, and let ǫ > 0. Obtain a (C, ǫ)-invariant clopen
tower decomposition of Ωpunc. Choose any tower (W, S) in this decomposition. We
aim to modify the tower by splitting it into finitely many towers which are still (C, ǫ)-
invariant, and such that the collection of the levels of the new towers is a partition
of the union of the levels of (W, S), but so that the levels of the new towers all have
diameter smaller than ǫ. Applying this procedure to all of the towers simultaneously
will produce a clopen tower decomposition with the properties we seek.
CLASSIFICATION OF TILING C ∗-ALGEBRAS
33
Recall that S decomposes into finitely many compact open G-sets S = FK
k=1 Sk,
and first consider the G-set S1, which moves the base of the tower to the first level.
Consider the collection of patches {T ⊓BR1(0) T ∈ Ωpunc} for a sufficiently large R1,
to be defined. By FLC, there are finitely many such patches {P 1
N1}. Notice
n (0))}N1
that each patch contains the origin, and that the collection {U(P 1
n=1 is
pairwise disjoint and covers Ωpunc, where P 1
n which
contains the origin.
n (0) denotes the (unique) tile in P 1
1 , . . . , P 1
n , P 1
Since the G-set S1 is compact and open in Rpunc, and the collection of sets V (P, t, t′)
contained in Rpunc is a basis for the topology on Rpunc, we can choose R1 large enough
so that there exists a subcollection {P 1
N1}, and a collection
of tiles {t1
1 , . . . , P 1
nm for each m ∈ {1, . . . , M1} such that
m,l}l=1,...,Lm ⊂ P 1
n1, . . . , P 1
} of {P 1
nM1
S1 =
LmGl=1
Since W = s(S1), this implies that the base of the tower is W =FM1
m=1SLm
and that the first level is S1 · W =SM1
nm(0), t1
M1Gm=1
l=1 U(P 1
nm, P 1
nm, t1
V (P 1
m,l).
m,l).
V (P 1
nm, P 1
nm(0), t1
m,l) ∩ V (P 1
nm, P 1
nm(0), t1
m,l′) = ∅
and
Since S1 is a G-set, and since, for any fixed m ∈ {1, . . . , M1} and l 6= l′, we have
m=1 U(P 1
nm, P 1
nm(0)),
nm, P 1
s(V (P 1
nm(0), t1
m,l)) = U(P 1
nm, P 1
the injectivity of the source map of S1 implies that all the tiles t1
must be equal, to t1
nm, say. Then, in fact, we have S1 =FM1
nm(0)) = s(V (P 1
m ∈ P 1
nm, P 1
m,l′)),
nm(0), t1
m,l for l ∈ {1, . . . , Lm}
m=1 V (P 1
m).
nm(0), t1
nm, P 1
Since, for m 6= m′, we have
nm, P 1
V (P 1
m′) = ∅,
the injectivity of the range map on S1 tells us that the collection
m) ∩ V (P 1
nm′ (0), t1
nm(0), t1
nm′ , P 1
nm, P 1
must be pairwise disjoint.
{r(V (P 1
nm(0), t1
m))}m=1,...,M1 = {U(P 1
nm, t1
m)}m=1,...,M1
By choosing R1 to be large enough, we may assume that we have Bǫ−1(x(t1
m)) ⊂ P 1
nm
for each m ∈ {1, . . . , M1}, so that the diameter of U(P 1
m) is smaller than ǫ. In-
deed, since the vectors of translation associated to elements of S1 are already pre-
scribed above, and since enlarging R1 produces a new finite collection of patches, each
of which contains some patch Pnm from above, increasing R1 corresponds to parti-
tioning each V (P 1
m)}q=1,...,Qm, where
P 1
nm ⊂ P 1
m)) ⊂
P 1
nm,q.
The clopen tower (W, S) now splits into the collection of "one-level" clopen towers
nm(0), t1
m does not depend on q, we simply ensure that Bǫ−1(x(t1
m) into subsets {V (P 1
nm,q. Since t1
nm(0), t1
nm,q, P 1
nm, P 1
nm, t1
{(W 1
m, H 1
m) m = 1, . . . , M1} by taking
W 1
nm, P 1
m := W ∩ U(P 1
nm, P 1
nm(0), t1
to be the base of the m-th tower, and H 1
m) to
be the shape of the m-th tower. The single level of this tower is U(P 1
m), and, by
our choice of R1, it has diameter smaller than ǫ. In addition, as we saw above, the
collection of the tower levels {H 1
m)}m=1,...,M1 is pairwise
m) of the tower (W, S).
nm(0)) = U(P 1
m := S1 ∩ s−1(W 1
m}m=1,...,M1 = {U(P 1
m=1 U(P 1
disjoint, and so partitions the first level S1·W =FM1
nm, P 1
nm, t1
m) = V (P 1
m, t1
nm, t1
m · W 1
nm(0))
34
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
So, from the "one-level" tower (W, S1), we have obtained a collection of towers whose
levels have diameter less than ǫ, and still partition S1 · W .
m , H k−1
n1, . . . , P k
nMk
1 , . . . , P k
} of the finite set of patches {P k
m=1 V (P k
nm(0), tk
nm, P k
of the form T ⊓ BRk (0) such that Sk =FMk
Now, we iterate this procedure. At the k-th step we replace W by the base of
each of the towers (W k−1
m ), for m = 1, . . . , Mk−1, which were constructed in the
(k − 1)-th step, in turn, and replace S1 by Sk. We find Rk ≥ Rk−1 large enough that
there exists a subcollection {P k
Nk}
m) (the fact that there
is just one tile associated to each patch uses the fact that Sk is a G-set, and an
argument as above). We further choose Rk to ensure that Bǫ−1(x(tk
nm for each
m ∈ {1, . . . , Mk}. Then, to form the towers for the k-th step, we take bases of the
form W k
m ⊂
W k−1
m).
That is, we "pick up" all of the arrows that our construction had previously associated
to this subset, and also include the new arrows contained in the subset Sk ⊂ S of the
shape of the tower (W, S) whose sources lie in this subset.
nm(0)). Observe that, for each m ∈ {1, . . . , Mk}, we have W k
m ∪Sk)∩s−1(W k
m for some m ∈ {1, . . . , Mk−1}. We define the shape H k
m = (H k−1
m = U(P k
nm, P k
m)) ⊂ P k
for m ∈ {1, . . . , MK} will have diameter smaller than ǫ, and will partition FK
After finitely many iterations, the process terminates. At the end of the process,
we have obtained an R = RK large enough to work for all of the sets Sk in the
construction above simultaneously. In particular, the levels of each tower (W K
m , H K
m )
k=1(Sk ·
W ). We claim that the shapes of the towers we constructed are still (C, ǫ)-invariant.
Let m ∈ {1, . . . , MK} and let T ∈ s(H K
m ) be arbitrary. By construction, the collection
of arrows from S with source T is equal to the set of arrows from H K
m with source T .
This, combined with the fact that S was (C, ǫ)-invariant, shows that H K
m is (C, ǫ)-
invariant too.
(cid:3)
Remark 7.5. By following the procedure in the proof above, we may assume, without
loss of generality, that each tower (W, S) in a clopen tower decomposition of the
action Rpunc y Ωpunc has W = U(P, t) and S = Ft′∈Q V (P, t, t′) for some patch P ,
some t ∈ P , and some subset Q ⊂ P . Given any ǫ > 0, we may enlarge P to further
assume that Bǫ−1(x(t′)) ⊂ P for each t′ ∈ Q.
8. Z-stability of tiling algebras
This section contains our Z-stability result (Theorem 8.4). Since an application of
Theorem 3.8 will be crucial to our proof, we first present subsets of the tiling groupoid
which are compatible with this theorem.
We caution the reader that for the purposes of this section, we will need to work
with tilings of Rd alongside quasitilings of the tiling groupoid Rpunc. In order to make
this distinction clearer, we will make an attempt to be consistent with the notation
for tilings as elements of the unit space of the groupoid Rpunc. Generally, tilings of
Rd will be denoted by the letters u, x or y; vectors in Rd will be denoted by the letter
z, and quasitilings of Rpunc will be made up of tiles T1, . . . , Tn ⊂ Rpunc and centres
C1, . . . , Cn ⊂ Rpunc (with additional subscripts as necessary).
Lemma 8.1. The groupoid Rpunc associated to an aperiodic and repetitive tiling of
Rd with FLC admits sequences of subsets Tl as in Theorem 3.8 of arbitrary length.
Proof. We directly construct a sequence of subsets {Tl}l∈N as in Theorem 3.8. Let
Ωpunc be the punctured hull of a tiling as in the statement. By FLC, there is a
prototile of minimal volume Vmin, and one of maximal volume Vmax. Find m ∈ N
CLASSIFICATION OF TILING C ∗-ALGEBRAS
35
such that m > max(Vmax/Vmin, 2). For n ∈ N, define Tn ⊂ Rpunc as follows. For
each u ∈ Ωpunc consider the set of punctures in u which lie in Bn(0). For each such
puncture x(t) (where t is a tile in the tiling u of Rd), include the arrow (u − x(t), u) in
Tn. In other words, Tn is the collection of allowable translates by vectors of magnitude
smaller than n
Tn = {(u + z, u) ∈ Rpunc u ∈ Ωpunc, kzk < n}.
Notice that Tn is closed in the metric topology on Rpunc. Indeed, if {(uj + zj, uj)}j∈N
converges to (u + z, u) then both d(u, uj) → 0 and kz − zjk → 0. Eventually, this
convergence forces uj ⊓ Bn(0) = u ⊓ Bn(0). Furthermore, since kzjk < n for each
j, using the uniform discreteness of the puncture set together with the last sentence
and the convergence of zj to z implies that eventually zj = z for all j. Thus we see
that kzk < n and so (u + z, u) ∈ Tn, showing that Tn is closed. In addition, we have
that
Tn ⊂ [u∈Ωpunc [t∈u⊓Bn(0)
V (u ⊓ Bn(0), t, u(0)).
By FLC, there are only finitely many patches u ⊓ Bn(0) involved in the union on the
right, each of which contain finitely many tiles. Thus the right-hand side is a finite
union of compact sets, and so is itself compact. Thus we see that Tn is a closed subset
of a compact set in a metric space, so it is compact. Note that Tn is closed under
taking inverses, and therefore uTn = Tnu for each u ∈ Ωpunc. It follows that
inf u∈Ωpunc Tnu
supu∈Ωpunc uTn
≤ 1.
We look to bound inf u∈Ωpunc Tnu and supu∈Ωpunc uTn = supu∈Ωpunc Tnu. We will
use volume estimates to do this.
First, for u ∈ Ωpunc to maximise Tnu we wish there to be as many tiles of u
intersecting Bn(0) as possible, so that we can assume the punctures of all these tiles
lie in this ball. By FLC there exists a prototile of largest diameter Dmax. Consider
the ball Bn+Dmax(0). If a tile t intersects the complement of this ball then, since its
diameter is at most Dmax, it cannot intersect Bn(0). Therefore, since t intersects
Bn(0) only if it is contained within Bn+Dmax(0), we ask how many tiles can fit within
the latter ball. Clearly a possible bound is given by
supu∈Ωpunc uTn ≤
Vol(Bn+Dmax(0))
Vmin
=
Vd(n + Dmax)d
Vmin
where Vd denotes the volume of the unit ball in Rd.
We argue similarly to establish our second bound. From this point onwards we will
assume that n > Dmax. If a tile intersects Bn−Dmax(0), then its puncture must lie in
Bn(0), so we seek to minimize the number of tiles intersecting this ball. To do so, we
may assume any such tile has volume Vmax and is completely contained within the
ball. Then we see that
inf
u∈Ωpunc
Tnu ≥
Vd(n − Dmax)d
Vmax
.
Combining our bounds yields
inf u∈Ωpunc Tnu
supu∈Ωpunc uTn
≥
VdVmin(n − Dmax)d
VdVmax(n + Dmax)d →
n→∞
Vmin
Vmax
>
1
m
.
36
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
Therefore, for large enough n and small enough 0 < ǫ < 1/2 we have that
inf u∈Ωpunc Tnu
supu∈Ωpunc uTn
(1 − ǫ) ≥
1
m
too.
Now, fix k ∈ N. Since Tn is the set of translates in Rpunc with magnitude smaller
than n, we see that any translate in ∂Tn(Tn+k) must have magnitude larger than
(n + k) − n = k and smaller than (n + k) + n = 2n + k. Thus, to bound ∂Tn(Tn+ku)
above, we wish to maximise the number of punctures which appear in the annulus
with center 0, inner radius k, and outer radius 2n + k. A tile which intersects the
complement of the annulus with center 0, inner radius k − Dmax, and outer radius
2n + k + Dmax cannot have its puncture in the region of interest, so we estimate the
number of tiles contained in this latter annulus as follows
∂Tn(Tn+ku) ≤
Vd((2n + k + Dmax)d − (k − Dmax)d)
Vmin
where the top of the fraction is just the volume of the second annulus. On the other
hand, we know that
Tn+ku ≥
Vd(n + k − Dmax)d
Vmax
.
Cn,k :=
Vmax((2n + k + Dmax)d − (k − Dmax)d)
Vmin(n + k − Dmax)d
Let
so that
∂Tn(Tn+ku) ≤
Vd((2n + k + Dmax)d − (k − Dmax)d)
Vmin
Vd(n + k − Dmax)d
Vmax
= Cn,k
≤ Cn,kTn+ku.
Notice that for each fixed n, we have that Cn,k → 0 as k → ∞. Therefore, for each
n, we can find k(n) ≥ 1 such that for any k ≥ k(n) we have Cn,k ≤ ǫ2/8. Then the
inequality above becomes
∂Tn(Tn+k(n)u) ≤ Cn,k(n)Tn+k(n)u ≤ ǫ2/8Tn+k(n)u.
We now relabel the sequence (Tn) by removing the first terms and shifting index, to
assume that every set in the sequence satisfies properties (i) and (ii) from the state-
ment of Theorem 3.8. We construct a subsequence by choosing indices ni recursively
as follows. Let n1 = 1 and put ni+1 = ni + k(ni). By construction, the sequence
(Tni)i∈N will satisfy all of the requirements of Theorem 3.8.
(cid:3)
To obtain Z-stability, we use the following criterion of Hirshberg-Orovitz, which
was obtained using the breakthrough result of Matui and Sato, where they show that
any nuclear tracially AF-algebra is Z-stable, see [34, Theorem 1.1 and Theorem 5.4].
Hirshberg's and Orovitz's criterion requires us to embed arbitrarily large matrix al-
gebras into the crossed product so that the image approximately commutes with a
given finite set and is large in trace. That this criterion is equivalent to Z-stability
for simple separable unital nuclear C ∗-algebras is recorded in [21, Proposition 2.2 and
Theorem 4.1].
CLASSIFICATION OF TILING C ∗-ALGEBRAS
37
Theorem 8.2 ([21, Definition 2.1]). Let A be a simple separable unital nuclear C ∗-
algebra which is not isomorphic to C. Write - for the relation of Cuntz subequivalence
over A, and given a, b ∈ A write [a, b] for the commutator ab − ba. Suppose that for
every n ∈ N, finite set F ⊂ A, ǫ > 0 and nonzero positive element a ∈ A there exists
an order zero completely positive and contractive map φ : Mn → A such that
(i) 1 − φ(1) - a,
(ii) k[w, φ(B)]k < ǫ for all w ∈ F and B ∈ Mn with kBk = 1.
Then A is Z-stable.
The following lemma will allow us to witness Cuntz subequivalence when checking
condition (i) of Theorem 8.2.
[27, Lemma 12.3]). Let G y X be an action of a groupoid on a
Lemma 8.3 (c.f.
compact metrizable space. Let A be a closed subset of X and B an open subset of X
such that A ≺ B. Let f, g : X → [0, 1] be continuous functions such that f = 0 on
X \ A, and g = 1 on B. Then there exists a v ∈ C(X) ⋊r G such that v∗gv = f .
Proof. Since A ≺ B, and A is a closed subset of A, find a finite collection U1, . . . , UM
of open subsets of X which cover A, and subsets S1, . . . , SM of G as in the definition.
Without loss of generality, by removing any element t ∈ Sm such that s(t) /∈ rX(Um),
we may assume that rX(Um) = s(Sm). Furthermore, by preserving one element from
each source bundle and removing all other elements from Sm if necessary, we may
assume that s : Sm → rX(Um) is injective for each m ∈ {1, . . . , M}.
Using a partition of unity argument, produce, for each m ∈ {1, . . . , M}, a contin-
uous function hm : X → [0, 1] such that hm(x) = 0 for any x ∈ X \ Um, and which
m=1 hm(a) = 1 for any a ∈ A. Put
m=1 hm(x) ≤ 1 for any x ∈ X, and PM
satisfy 0 ≤PM
v =PM
m=1Pt∈Sm Ut(f hm)1/2.
Denote by α the induced action of G on C(X), given by
We compute that
f (g−1x)
0
αg(f )(x) =
(f hm)1/2Ut−1 g
MXm=1 Xt∈Sm
MXm, m=1 Xt∈Sm Xt∈S m
v∗gv =
=
X (r(g))
if x ∈ r−1
otherwise.
MXm=1 Xt∈S m
Ut(f h m)1/2
Ut−1gαt(f )1/2αt(hm)1/2αt(h m)1/2αt(f )1/2Ut.
Note that αt(hm) is supported on tUm, and αt(h m) is supported on tU m, which are
disjoint unless m = m and t = t. Thus, the only nonzero contributions to the above
sum require t = t and m = m, so we obtain
v∗gv =
=
MXm=1 Xt∈Sm
MXm=1 Xt∈Sm
Ut−1gαt(hm)αt(f )Ut
αt−1(g)hmf.
38
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
The term corresponding to a particular m ∈ {1, . . . , M} and t ∈ Sm in the above sum
is supported in Um ∩ r−1
X (s(t)), we have αt−1(g)(x) = g(tx),
where tx ∈ tUm ⊂ B, so we obtain αt−1(g) = 1 wherever this term is supported,
because g(b) = 1 for any b ∈ B by assumption. This gives
X (s(t)). For x ∈ Um ∩ r−1
v∗gv =
MXm=1 Xt∈Sm
(f hm)Um∩r−1
X (s(t))
Since s(Sm) = rX(Um), since s : Sm → rX(Um) was injective for each m, and using
the facts that hm is supported inside Um and f is supported inside A, together with
m=1 hm(a) = 1 for each a ∈ A, we see that
our assumption that PM
v∗gv =
(f hm)Um =
MXm=1
MXm=1
f hm =
MXm=1
f (hm)A = f
MXm=1
(hm)A = f.
(cid:3)
We now turn to our Z-stability result. Our method of proof is by now well-
established in the group case [6,27]. Nevertheless, we wish to provide an outline of the
proof. Our outline will focus on the first half of the proof (pages 39-46), which contains
a large amount of technical setup. For now, we aim to ignore the technicalities for
a sharper focus on the philosophy behind the proof. We also highlight parts of the
argument which are specifically necessary in the groupoid case.
We begin with a tiling C ∗-algebra, Apunc = C ∗(Rpunc). Observe that Rpunc is
isomorphic to the transformation groupoid Rpunc ⋉ Ωpunc associated to the canonical
∼= Rpunc ⋉ Ωpunc is
groupoid action of Rpunc on its unit space Ωpunc. Since Rpunc
almost finite by Lemma 7.4, it follows from Theorem 2.15 that this action is almost
finite. We use [16, Proposition 4.38] to think of C ∗(Rpunc) ∼= C ∗(Rpunc ⋉ Ωpunc) ∼=
C(Ωpunc) ⋊r Rpunc as a groupoid crossed product, and we seek to embed arbitrarily
large matrix algebras into this algebra to witness Z-stability using Theorem 8.2.
We consider compact subsets Ωpunc ⊂ T1 ⊂ · · · ⊂ TL ⊂ Rpunc as in Lemma 8.1,
and we use almost finiteness of the action Rpunc y Ωpunc to obtain a clopen tower
decomposition {(Wk, Sk)} whose shapes are sufficiently invariant to be quasitiled
by the tiles {T1, . . . , TL}, so that the quasitiling of Sk has centres {Ck,1, . . . , Ck,L}.
Compared to the group case, this step requires additional care because it will be
important that the quasitiling encapsulates the structure of the tower. In the group
case, each tower level is obtained by allowing a single group element to act on the
base of the tower, but this is not generally true for groupoids due to the fibred nature
of their actions. Therefore, we must take some care to ensure that, for each tower
level Sk,pWk, each of the sets Ck,l and TlCk,l either contains all of Sk,p, or does not
intersect it at all. This consideration is the main point of difficulty in generalising to
the groupoid setting, and must be accounted for whenever we modify these subsets.
We introduce a finite open cover {U1, . . . , UM } of Ωpunc by sets of sufficiently small
diameter, and use Lemma 2.11 to assume that our tower levels have diameter smaller
than the Lebesgue number of the cover. We partition each of the sets Ck,l into
Ck,l,1, . . . , Ck,l,M ⊂ Ck,l by requiring that the tower levels associated to elements
of Ck,l,m are subordinate to Um. This allows the later use of a uniform continu-
ity argument to obtain approximate centrality of our matrix embedding under a
finite subset of C(Ωpunc). We then choose n equally sized pairwise disjoint subsets
C (1)
k,l,m, . . . , C (n)
k,l,m ⊂ Ck,l,m of cardinality ⌊Ck,l,m/n⌋, and define a system of bijections
between these sets. The matrix unit eij ∈ Mn will be implemented as movement
CLASSIFICATION OF TILING C ∗-ALGEBRAS
39
from the tower levels associated to elements of TlC (j)
corresponding elements of TlC (i)
k,l,m under the appropriate bijection.
k,l,m to those associated to the
We also need to ensure that our matrix embedding is approximately central under
the action of a predefined compact subset of the groupoid. To do so, we use β-
disjointness of the quasitiling {Tlc c ∈ Ck,l} to slightly shrink the tiles to eliminate
as much "stacking" of tower levels as possible. We assume that the tiles are sufficiently
invariant under multiplication by the compact set that we can find a large "core" in
each consisting of the elements that remain within the tile after a large number Q of
such multiplications. We use this "core" to partition the tile, grouping the elements
according to how many multiplications it takes to remove them from the tile. We
then assign each tower level in our matrix embedding an indicator function which is
scaled by a factor q/Q for some q ∈ {1, . . . , Q} based on which set in the partition the
associated groupoid element lies. For example, elements in the "core", which remain
in the tile after Q multiplications, are assigned the value Q/Q = 1, while those which
are removed from the tile after just one multiplication are assigned the value 1/Q.
As a result of this construction, conjugation by an element of the compact set is
guaranteed to perturb the indicator function by no more than 1/Q.
The final step is to ensure that our matrix embedding is large in trace. This is
expressed by comparison to a positive element, which we may assume is a function
in C(Ωpunc) by Lemma 4.20. By using minimality of the action and increasing the
invariance of the shapes of the tower if necessary, we are able to assume that a large
number of the tower levels are contained within the support of this function; whence,
with no loss of generality, we may assume the function takes the constant value 1
on these tower levels. More precisely, since the matrix embedding was constructed
to be supported on a union of tower levels, we may ask that the number of tower
levels on which the matrix embedding is not supported is smaller than the number
of tower levels on which the positive element takes the value 1. By setting up an
injection between the first set of tower levels and the second, we are able to put
the complement of the matrix embedding below the positive element, witnessing the
largeness in trace.
Theorem 8.4. The C ∗-algebra associated to an aperiodic and repetitive tiling with
FLC is Z-stable.
Proof. To simplify notation, denote G = Rpunc and X = Ωpunc. As per usual, we let α
denote the induced action of G on C(X), so that αg : C(X) → C(r−1
X (r(g))) is given
by αg(f )(x) = f (g−1x) for x ∈ r−1
X (r(g)), and we define αg(f )(x) = 0 if rX(x) 6= r(g).
Let n ∈ N, ǫ > 0, and let a ∈ C(X) ⋊r G be a nonzero positive element. Let
F ⊂ C(X) ⋊r G be a finite set. We will show that there exists a map φ : Mn →
C(X) ⋊r G as in Theorem 8.2, from which we will be able to conclude that C(X) ⋊r G
is Z-stable.
Notice that, since Γc(G, r∗A) is dense in A ⋊r G, it will be enough to consider finite
subsets F ⊂ Γc(G, r∗A). Each such subset is generated by a finite subset Υ of the
unit ball of C(X) and a compact subset K ⊂ G which, without loss of generality, we
may assume contains G(0) and is closed under taking inverses. To see that Υ can be
and G is ample, C admits a finite cover of pairwise disjoint compact open bisections
{B1, . . . , BN }. Then f is generated by {Ug g ∈ C} together with the finite collection
chosen to be finite, fix any f =Pg∈C fgUg ∈ F and observe that, since C is compact
of functions {Pg∈Bi fg}i=1,...,N , each of which is in C(X) by the argument following
40
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
the proof of Lemma 4.18. Since C(X) ⋊r Rpunc is generated by sets of this form, it
will be enough to check condition (ii) in Theorem 8.2 for each element w ∈ Υ and
for w =Pg∈K Ug.
Since K is a compact subset of Rpunc, there exists a real number RK > 0 such that
if (u − z, u) ∈ K (for some u ∈ Ωpunc and z ∈ Rd) then z < RK. By enlarging K
we may further assume, without loss of generality, that R−1
K < r/2, where r is the
constant of uniform discreteness for the puncture set in the tiling. This ensures that
whenever u, u′ ∈ Ωpunc have d(u, u′) < R−1
K in the tiling metric, the subsets Ku and
Ku′ ⊂ Rpunc consist of translations by the same set of vectors in Rd. We enlarge K
one last time to assume that K = {(u − z, u) ∈ Rpunc z < RK}. Reasoning as for
the sets Tn in Example 8.1, this choice of K is seen to be compact.
By Lemma 4.20, we may assume that a ∈ C(X). Then, since a is a nonzero
positive element, there exists x0 ∈ X and θ > 0 such that a is strictly positive on the
closed ball of radius 2θ centred at x0. Thus, we may assume that a is a [0, 1]-valued
function which takes the value 1 on all points within distance 2θ from x0, and the
value 0 at all points at distance at least 3θ from x0. Denote by O the open ball of
radius θ centred at x0. Because the action is minimal, there exists a subset D ⊂ G
such that D−1 · O = X and so that there exists m ∈ N such that, for each u ∈ G(0),
D ∩ r−1(u) < m and D ∩ s−1(u) < m. To see this, since G is ´etale it has a base
of open bisections {Bi}i∈I. For each i ∈ I, the set BiO is open in X by Lemma 2.7,
because Bi is open in G and O is open in X. Because Bi is a bisection, for each point
x ∈ O, there exists at most one g ∈ Bi such that s(g) = r−1
X (x), and therefore at most
one image gx ∈ BiO. Since the action is minimal, the collection {BiO}i∈I is an open
cover of X and, since X is compact, there exists a finite subcover {B1O, . . . , BmO}.
Let D =Sm
j=1 B−1
j
, which has the properties we seek.
Let 0 < κ < 1, to be determined on page 52. Choose an integer
Denote by K Q the product of K with itself Q times in G:
Q > (n2 supu∈G(0) uK)/ǫ.
K Q := {kQ · · · k1 (ki+1, ki) ∈ K (2)
for every i = 1, . . . , Q − 1}.
Take β > 0 to be small enough so that if T is a nonempty compact subset of G which
is sufficiently invariant under left-translation by K Q, then, for every T ′ ⊂ T with
T ′u ≥ (1 − nβ)T u at every u ∈ G(0), we have, for every u ∈ G(0), that
(8.1)
≥ (1 − κ)T u.
\g∈K Q
gT ′ ⊔ Gv∈G(0)\r(g)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
vT ′ u(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
To ease thinking, note that since G(0) ⊂ K and K is closed under taking inverses,
the left-hand set is the subset consisting of t ∈ T ′u such that K Qt ⊂ T ′u. Choose
L ∈ N large enough so that (1 − β/2)L < β. By Example 8.1, there exist compact
subsets G(0) ⊂ T1 ⊂ · · · ⊂ TL of G such that ∂Tl−1(Tlu) ≤ (β2/8)Tlu for each
l = 2, . . . , L and u ∈ G(0). We may also enlarge each Tl in turn to assume that it
is sufficiently invariant under left translation by K Q so that we can use the previous
paragraph to ensure that for every subset T ′ ⊂ Tl satisfying T ′u ≥ (1 − nβ)Tlu for
each u ∈ G(0), we have, for each u ∈ G(0), that (8.1) holds for this choice of T ′ and
with T replaced with Tl.
Since TL was constructed as in Example 8.1, there exists RL > 0 such that TL =
{(u − z, u) ∈ Rpunc z < RL}. By the uniform continuity of the functions in Υ ∪ Υ2,
CLASSIFICATION OF TILING C ∗-ALGEBRAS
41
and of the action map TL × X → X, there exists 0 < η < (QRK + RL)−1 so that if
x, y ∈ X, and t, t′ ∈ TL are such that s(t) = x and s(t′) = y, and satisfy d(x, y) < η
and d(t, t′) < η, then f (tx) − f (t′y) < ǫ2/(4n4). Let U = {U1, . . . , UM } be an
open cover of X whose members all have diameter less than η. Let η > η′ > 0 be a
Lebesgue number for U which is no larger than θ.
Let E be a compact subset of G containing TL, and let δ > 0 be such that δ < β2/4.
Since Gu is infinite for each u ∈ G(0), we may enlarge E and shrink δ as necessary
to ensure that, for each nonempty (E, δ)-invariant compact set S ⊂ G and each
u ∈ G(0), we have that
(8.2)
βSu ≥ Mn
LXl=1
maxv∈G(0) Tlv.
tower decomposition of X with levels of diameter less than η′. Write Sk =Fnk
By combining Lemma 7.4 and Theorem 2.15, we see that the action G y X admits
clopen tower decompositions of arbitrary invariance. Therefore, for some N ∈ N,
we can find nonempty clopen sets W1, . . . , WN ⊂ X and nonempty (E, δ)-invariant
compact open sets S1, . . . , SN ⊂ G such that the family {(Wk, Sk)}N
k=1 is a clopen
j=1 Sk,j
for the decomposition of each shape. By Remark 7.5, we may assume, without loss
of generality, that, for each k ∈ {1, . . . , N}, there exists a patch Pk and a tile tk ∈ Pk
such that Wk = U(Pk, tk), and so that, for each j ∈ {1, . . . , nk}, there exists tk,j ∈ Pk
such that Sk,j = V (Pk, tk, tk,j). Note in particular that, for each l ∈ {1, . . . , L}, since
Tl = {(u − z, u) ∈ Rpunc z < Rl} for some 0 < Rl ≤ RL, our choice of η′ ensures
that if x and y are in the same tower level, then Tlx and Tly consist of arrows which
implement exactly the same set of vectors of translation.
SL
Let k ∈ {1, . . . , N}. Since Sk is (TL, β2/4)-invariant, and since the sets T1, . . . , TL
satisfy the assumptions of Theorem 3.8, we can find Ck,1, . . . , Ck,L ⊂ Sk such that
l=1 TlCk,l ⊂ Sk, and so that the collection {Tlc l ∈ {1, . . . , L}, c ∈ Ck,l} is β-disjoint
and (1 − β)-covers Sk. We claim that we may assume that, for each l ∈ {1, . . . , L}
and each p ∈ {1, . . . , nk}, we have either Sk,p ⊂ TlCk,l or Sk,p ∩ TlCk,l = ∅. That is,
TlCk,l either contains all the groupoid elements corresponding to a given level of the
tower, or none of them. We prove this assertion over a number of steps.
To elaborate, recall that Ck,L was constructed from the set I = {c ∈ Sk TLc ⊂ Sk}.
Suppose c ∈ Sk,j = V (Pk, tk, tk,j) is an element of I. Then, for each fixed t ∈ TLr(c),
tc ∈ Sk, and so, in particular, tc ∈ Sk,i = V (Pk, tk, tk,i) for some i. Recall that TL
was the collection of allowable vectors of translation of magnitude smaller than RL.
Thus, if tc ∈ V (Pk, tk, tk,i) ⊂ Sk for some c ∈ Sk,j, then the vector x(tk,i) − x(tk,j) was
associated to t, and was allowable in r(c) ∈ U(Pk, tk,j). Let c′ be any other element
of Sk,j. By the diameter condition on the tower levels, TLr(c′) consists of arrows
which implement the same set of vectors as the arrows in TLr(c). By the argument
at the start of the paragraph, we see that any such arrow t′ ∈ TLr(c′) implements
the vector x(tk,i) − x(tk,j) for some i. Since c′ ∈ Sk,j = V (Pk, tk, tk,j), we see that
t′c′ ∈ V (Pk, tk, tk,i) = Sk,i ⊂ Sk. This shows that c′ ∈ I, so that Sk,j ⊂ I. Hence, we
have shown that, for each j, either Sk,j ∩ I = ∅ or Sk,j ⊂ I.
To obtain Ck,L, we applied Lemma 3.3 to get a subset Ck,L ⊂ I such that the
collection {TLc c ∈ Ck,L} was β-disjoint. In doing so, we claim that we can choose
to either remove or keep each tower level as a whole. Indeed, if we fix any x ∈ Wk and
apply Lemma 3.3 to the collection {TLcx c ∈ I}, then we obtain Ck,Lx ⊂ Ix such
that {TLcx c ∈ Ck,Lx} is β-disjoint. Now, choose any other y ∈ Wk = U(Pk, tk).
42
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
By the previous paragraph, I contains all or none of the elements of each Sk,p, so, by
the structure of the tower (Wk, Sk), Ix and Iy consist of arrows which implement the
same vectors of translation. This allows us to choose Ck,Ly to consist of arrows which
implement the same vectors as those in Ck,Lx. Since I ⊂ Sk, if c, c′ ∈ I implement
the same vector of translation and have s(c) = x and s(c′) = y, then r(c) and r(c′)
are in the same tower level, and thus (as tilings) they agree on Bη−1(0). Therefore,
TLr(c) and TLr(c′) consist of elements which implement the same set of vectors. This
shows that {TLcx c ∈ Ck,Lx} and {TLc′y c′ ∈ Ck,Ly} implement the same set of
vectors. Combining this with the fact that the collection {TLcx c ∈ Ck,Lx} was
β-disjoint, we see that the collection {TLc′y c′ ∈ Ck,Ly} is also β-disjoint. Repeat
this construction for each y ∈ Wk, and set Ck,L =Fy∈Wk Ck,Ly, so that the collection
{TLc c ∈ Ck,L} is β-disjoint. Observe that by construction, for each j, we have
either Sk,j ∩ Ck,l = ∅ or Sk,j ⊂ Ck,l.
By construction, since Ck,L ⊂ I, we have TLCk,L ⊂ Sk. Now, suppose that g ∈
Sk,p = V (Pk, tk, tk,p) is such that g ∈ TLCk,L. Then there exists c ∈ Ck,L such
that c ∈ Sk,j = V (Pk, tk, tk,j) for some j, and t ∈ TL implementing the vector
x(tk,p) − x(tk,j), such that tc = g. Choose any other g′ ∈ Sk,p, and consider the
groupoid element c′ with source s(g′) and range s(g′) + x(tk,j) − x(tk). Observe that
c′ is in Sk,j = V (Pk, tk, tk,j), which is a subset of Ck,L by the previous paragraph, since
c ∈ Sk,j ∩ Ck,L. Since r(c′) is in the same tower level as r(c) = s(t), the arrows in
TLr(c′) implement exactly the same vectors as the arrows in TLs(t), so, in particular,
there is an element t′ ∈ TLr(c′) which implements the same vector x(tk,p) − x(tk,j)
that t does. Then t′c′ implements the vector x(tk,p) − x(tk), to s(t′c′) = s(g′), so we
see that t′c′ = g′, and hence that g′ ∈ TLCk,L. This shows that, for each p, either
Sk,p ∩ TLCk,L = ∅, or Sk,p ⊂ TLCk,L.
We now simply repeat the procedure of the previous three paragraphs for each
l ∈ {L − 1, . . . , 1} in turn. As in the proof of Lemma 3.8, at each step we consider
j=l+1 TjCk,j,
we have removed entire tower levels from consideration, so at each step we are working
with a subtower of (Wk, Sk).
j=l+1 TjCk,j}. Notice that when forming Sk \FL
I = {c ∈ Sk Tlc ⊂ Sk \FL
Since the levels of the tower (Wk, Sk) have diameter less than η′, which is a Lebesgue
number for U, for each l ∈ {1, . . . , L} there is a partition
Ck,l = Ck,l,1 ⊔ Ck,l,2 ⊔ · · · ⊔ Ck,l,M
such that cWk ⊂ Um whenever m ∈ {1, . . . , M} and c ∈ Ck,l,m.
In fact, by the
diameter condition on the tower levels, for each j such that Sk,j ⊂ Ck,l, there exists
m such that Sk,jWk ⊂ Um, whence, for each c ∈ Sk,j, we may put c ∈ Ck,l,m. In this
way, we may assume that, for each j and m, either Sk,j ∩ Ck,l,m = ∅ or Sk,j ⊂ Ck,l,m.
In addition, since TlCk,l,m ⊂ Sk, and making use of the structure of Tl and Sk,j, it is
then automatic that, for each j and m, either Sk,j ∩ TlCk,l,m = ∅ or Sk,j ⊂ TlCk,l,m by
following a similar argument as that for TLCk,L above.
k,l,m, . . . , C (n)
For each l and m, choose pairwise disjoint subsets C (1)
k,l,m of Ck,l,m such
that, for each u ∈ G(0) and i ∈ {1, . . . , n}, we have C (i)
k,l,mu = ⌊Ck,l,mu/n⌋, where
⌊·⌋ denotes the floor function. Further enforce that, if Sk,j ⊂ Ck,l,m and c ∈ Sk,j is
included in C (i)
k,l,m. Since the source map is injective on Sk,j, this
amounts to adding a single element to C (i)
k,l,mu for each u ∈ s(Sk,j) = Wk. In this way,
each C (i)
k,l,m corresponds to a choice of a (1/n)-th of the tower levels associated to Ck,l,m
k,l,m, then Sk,j ⊂ C (i)
CLASSIFICATION OF TILING C ∗-ALGEBRAS
43
(ignoring any remainder), so that, for each j, either Sk,j ∩ C (i)
As in the last paragraph, it is then automatic that either Sk,j ∩ TlC (i)
Sk,j ⊂ TlC (i)
k,l,m for each j.
k,l,m = ∅, or Sk,j ⊂ C (i)
k,l,m.
k,l,m = ∅, or
For each i ∈ {2, . . . , n}, choose a bijection
Λk,i : Gl,m
C (1)
k,l,m → Gl,m
C (i)
k,l,m
k,l,mu to C (i)
which sends C (1)
k,l,mu for every l ∈ {1, . . . , L}, m ∈ {1, . . . , M}, and every
u ∈ G(0). Further enforce that, whenever Sk,j ⊂ C (1)
k,l,m, we have, for each i, that
Λk,i(Sk,j) = Sk,p for some p such that Sk,p ⊂ C (i)
k,l,m, so that Λk,i sends the set of
groupoid elements associated to one particular tower level to the set of groupoid
elements associated to some other tower level. Also, define Λk,1 to be the identity
map on Fl,m C (1)
k,l,m, and denote
Λk,i,j := Λk,i ◦ Λ−1
k,j : Gl,m
C (j)
k,l,m → Gl,m
C (i)
k,l,m.
is pairwise disjoint. Recall that, for each c ∈ Ck,l, we have Tlc ⊂ Sk \FL
Since the collection {Tlc l ∈ {1, . . . , L}, c ∈ Ck,l} is β-disjoint, for every l ∈
{1, . . . , L} and c ∈ Ck,l, we can find a Tk,l,c ⊂ Tl satisfying Tk,l,cu ≥ (1 − β)Tlu
for each u ∈ G(0), which is such that the collection {Tk,l,cc l ∈ {1, . . . , L}, c ∈ Ck,l}
j=l+1 TjCk,j.
Therefore, given c ∈ Ck,l and c′ ∈ Ck′,l′, observe that Tlc and Tl′c′ can only intersect
when k = k′ (otherwise the levels of the k-th and k′-th tower would intersect), and
when l = l′. Therefore, by following a similar argument as in the construction of
Ck,L, we may perform this construction in such a way that, whenever Sk,j ⊂ Ck,l and
c, c′ ∈ Sk,j, we choose Tk,l,cr(c) and Tk,l,c′r(c′) to consist of arrows which implement
In addition, by construction, for every c ∈ C (j)
the same set of vectors. This shows that, for each j, either Sk,j ∩Fc∈Ck,l Tk,l,cc = ∅ or
Sk,j ⊂Fc∈Ck,l Tk,l,cc. We are also free to choose Tk,l,cu = Tlu for every u ∈ G(0)\{r(c)}.
k,l,m and every i ∈ {1, . . . , n}, we have
r(c) ∈ Um, and r(Λk,i,j(c)) ∈ Um. In particular, this means that d(r(c), r(Λk,i,j(c))) <
η in the tiling metric, so the tilings r(c) and r(Λk,i,j(c)) agree on Bη−1(0). There-
fore, for each i, Tlr(c) and Tlr(Λk,i,j(c)) consist of groupoid elements which imple-
ment exactly the same vectors to r(c) and r(Λk,i,j(c)), respectively. Consider the set
gTk,l,c, which we obtain from Tk,l,c by removing the arrows which correspond to the
same vectors as the arrows which are removed from any set of the form Tlr(Λk,i,j(c))
to construct Tk,l,Λk,i,j(c).
In other words, if an arrow corresponding to translating
r(Λk,i,j(c)) by some vector is removed from Tlr(Λk,i,j(c)) when constructing Tk,l,Λk,i,j(c)
for any i, then the arrow corresponding to translating r(c) by the same vector does
In addition,
the sets of vectors of translation associated to the elements of any two of the sets
^Tk,l,Λk,i,j(c)r(Λk,i,j(c)), for i ∈ {1, . . . , n}, are the same. Also, when u 6= r(c), we have
not appear in gTk,l,c. Since we had Tk,l,Λk,i,j(c)r(Λk,i,j(c)) ≥ (1 − β)Tlr(Λk,i,j(c))
for each i ∈ {1, . . . , n}, we see that gTk,l,cr(c) ≥ (1 − nβ)Tlr(c).
gTk,l,cu = Tlu. Observe that, whenever c, c′ ∈ Sk,j, the sets gTk,l,cr(c) and ]Tk,l,c′r(c′)
each j, either Sk,j ∩Fc∈Ck,l gTk,l,cc = ∅ or Sk,j ⊂Fc∈Ck,l gTk,l,cc.
will consist of arrows which implement the same vectors of translation, so that, for
44
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
Now, for each j ∈ {1, . . . , n} and each c ∈ C (j)
k,l,m, consider the set T ′
k,l,c
:=
i=1
Tn
^Tk,l,Λk,i,j(c). Observe that, for each u ∈ G(0),
T ′
k,l,cu ≥ (1 − nβ)Tlu.
Sk,j ∩Fc∈Ck,j T ′
In addition, we still have, for each j, that whenever c, c′ ∈ Sk,j, the sets T ′
k,l,cr(c)
and Tk,l,c′r(c′) implement the same vectors of translation, so that, for each j, either
k,l,cr(Λk,i,j(c))
k,l,cr(c), we
k,l,cr(Λk,i,j(c)) which implements the same vector
consist of arrows which implement the same vectors for each i, given t ∈ T ′
will denote by t(i) the element of T ′
as t.
k,l,cc = ∅ or Sk,j ⊂ Fc∈Ck,j T ′
k,l,cr(c) and T ′
k,l,cc. Since T ′
Set
Bk,l,c,Qu = \g∈K Q
gT ′
k,l,c ⊔ Gv∈G(0)\r(g)
vT ′
k,l,c ,
noting that, by our assumption on the invariance of the sets Tl, and using (8.1), we
have, for each u ∈ G(0), that Bk,l,c,Qu ≥ (1 − κ)Tlu. We can alternatively describe
Bk,l,c,Q as the collection of elements t ∈ T ′
Suppose g ∈ Sk,p is such that g ∈ Sc∈Ck,l Bk,l,c,Qc, so that g = tc, with t ∈ Bk,l,c,Q
for c ∈ Sk,j ⊂ Ck,l, say. Given c′ ∈ Sk,j, define t′ ∈ T ′
k,l,c′ to be the element of
Gr(c′) which implements the same vector of translation that t does. Observe that,
since r(t) and r(t′) are in the same tower level, they agree on Bη−1(0), and so K Qr(t)
and K Qr(t′) consist of arrows implementing the same vectors of translation. Since
k,l,c such that K Qt ⊂ T ′
t ∈ Bk,l,c,Q, it follows that t′ ∈ Bk,l,c′,Q, and therefore t′c′ ∈Fc∈Ck,l Bk,l,c,Qc. Since t′c′
was an arbitrary element of Sk,p, this shows that either Sk,p ∩Fc∈Ck,l Bk,l,c,Qc = ∅, or
Sk,p ⊂Fc∈Ck,l Bk,l,c,Qc.
For each q ∈ {0, . . . , Q − 1}, set
k,l,c.
Bk,l,c,q = K Q−qBk,l,c,Q \ K Q−q−1Bk,l,c,Q.
Since G(0) ⊂ K, we have, for each q ∈ {0, . . . , Q}, that Bk,l,c,q ⊂ (G(0))q(K Q−qBk,l,c,Q\
K Q−q−1Bk,l,c,Q) ⊂ K QBk,l,c,Q, so these sets partition K QBk,l,c,Q. In addition, we have
Bk,l,c,Q ⊂ (G(0))QBk,l,c,Q ⊂ K QBk,l,c,Q, so that, for each u ∈ G(0), K QBk,l,c,Qu ≥
Bk,l,c,Qu ≥ (1 − κ)Tlu.
Since K QBk,l,c,Q is a subset of T ′
k,l,c, we have K QBk,l,c,Qc ⊂ Sk. Also, since
K qx and K qy consist of groupoid elements which implement the same vectors of
translation whenever q ∈ {1, . . . , Q} and d(x, y) < η, and since we have either
Sk,p ⊂ Fc∈Ck,l Bk,l,c,Qc or Sk,p ∩Fc∈Ck,l Bk,l,c,Qc = ∅, we see that, for each k, l, q and
j, we have either Sk,j ⊂Fc∈Ck,l Bk,l,c,qc or Sk,j ∩Fc∈Ck,l Bk,l,c,qc = ∅. In other words,
we may ensure that all the elements of Sk which correspond to any one clopen tower
level are all associated to the same q.
We claim that, for each q ∈ {1, . . . , Q} and i ∈ {1, . . . , n}, whenever t ∈ Bk,l,c,q
we have t(i) ∈ Bk,l,c,q as well. Suppose that s(t) = r(Λk,r,j(c)), and observe that
s(t(i)) = r(Λk,i,j(c)). By construction, s(t) and s(t(i)) are in the same Um, and so
the distance between them is smaller than η, so these tilings agree on BQRK +RL. By
construction, this means that the groupoid elements in K QTL implement the same
set of translations at both of these tilings. We assumed that t ∈ Bk,l,c,q, so we can
find b ∈ Bk,l,c,Q and g ∈ K Q−q such that gb /∈ K Q−q−1Bk,l,c,Q, and such that t = gb.
By construction, since gb ∈ K QTL, the vector which implements b is also allowable
at s(t(i)), and is implemented by b ∈ TL, say, and then the vector which implements
CLASSIFICATION OF TILING C ∗-ALGEBRAS
45
g at r(b) is also allowable at r(b), and is implemented by g ∈ Rpunc, say. Since r(b)
and r(b) agree on BQRK (0), and since g ∈ K Q−q, it follows from our choice of K that
g ∈ K Q−q. Next, we show that b ∈ Bk,l,c,Q. To do so, we prove that K Qb ⊂ T ′
k,l,c.
Indeed, we assumed that b ∈ Bk,l,c,Q, so we know that K Qb ⊂ T ′
k,l,c. Since r(b) and
r(b) agree on BQRK (0), we know that K Qr(b) and K Qr(b) consist of the same vectors
k,l,c, since s(b) and s(b) agree on BQRK +RL(0), the
of translation. By construction of T ′
k,l,cs(b) also implement the same set of vectors. Thus, since b and
sets T ′
b implemented the same vector, we see that, whenever k ∈ K Qr(b) and k ∈ K Qr(b)
implement the same translation, we have kb ∈ T ′
k,l,c. Since
k,l,c, so b ∈ Bk,l,c,Q, as required.
K Qb ⊂ T ′
Finally, we must show that gb /∈ K Q−q−1Bk,l,c,Q. This follows because gb implements
the same vector as gb, which was not an element of K Q−q−1Bk,l,c,Q, and the elements
of the sets K Q−q−1Bk,l,c,Qs(b) and K Q−q−1Bk,l,c,Qs(b) are implemented by precisely
the same set of vectors.
k,l,c by assumption, this shows that K Qb ⊂ T ′
k,l,c if and only if kb ∈ T ′
k,l,cs(b) and T ′
Given g ∈ K, it is clear that
(8.3)
gBk,l,c,Q ⊂ KBk,l,c,Q ⊂ Bk,l,c,Q−1 ∪ Bk,l,c,Q.
For q ∈ {1, . . . , Q − 1}, we have
(8.4)
gBk,l,c,q ⊂ Bk,l,c,q−1 ∪ Bk,l,c,q ∪ Bk,l,c,q+1,
because it is clear that gBk,l,c,q ⊂ K Q−q+1Bk,l,c,Q = FQ
Bk,l,c,q, if we had gh ∈ K Q−q−2Bk,l,c,Q = FQ
j=q−1 Bk,l,c,j, while, given h ∈
j=q+2 Bk,l,c,j then the closure of K under
inverses would provide h ∈ g−1K Q−q−2Bk,l,c,Q ⊂ KK Q−q−2Bk,l,c,Q = K Q−q−1Bk,l,c,Q,
which contradicts the membership of h in Bk,l,c,q.
We obtain a ∗-homomorphism ψ : Mn → C(X) ⋊r G by defining it on the standard
matrix units {eij}n
i,j=1 of Mn by
ψ(eij) =
NXk=1
LXl=1
MXm=1 Xc∈C(j)
k,l,m
QXq=1 Xt∈Bk,l,c,qr(c)
Ut(i)Λk,i,j(c)c−1t−11tcWk
where t(i) ∈ Tk,l,Λk,i,j(c)r(Λk,i,j(c)) is the element constructed earlier, and extending
linearly.
For each k ∈ {1, . . . , N}, let hk : X → [0, 1] denote the indicator function for
the clopen set Wk, and note that, since Wk is clopen, hk ∈ C(X). For each k ∈
{1, . . . , N}, l ∈ {1, . . . , L}, m ∈ {1, . . . , M}, i, j ∈ {1, . . . , n} and c ∈ C (j)
k,l,m, we set
hk,l,c,i,j =
=
QXq=1 Xt∈Bk,l,c,qr(c)
QXq=1 Xt∈Bk,l,c,qr(c)
q
Q
q
Q
Ut(i)Λk,i,j(c)c−1t−1αtc(hk)
Ut(i)Λk,i,j(c)c−1t−11tcWk.
Define a linear map φ : Mn → C(X) ⋊r G by setting
φ(eij) =
NXk=1
LXl=1
MXm=1 Xc∈C(j)
k,l,m
hk,l,c,i,j
46
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
and extending linearly. Set
h =
NXk=1
LXl=1
MXm=1
nXi=1 Xc∈C(i)
k,l,m
hk,l,c,i,i
so that h : X → [0, 1] is a continuous function. We check that h commutes with the
image of ψ, and that
for every b ∈ Mn, which will show that φ is an order-zero c.p.c map.
φ(b) = hψ(b)
q
Q
Ut(r)Λk,r,r(c)c−1t−11tcWk
Ut(i)Λk,i,j(c)c−1t−11tcWk
Ut(r)t−11tcWkUt(i)Λk,i,j(c)c−1t−11tcWk
We have
hψ(eij) =
k,l,m
k,l, m
QXq=1 Xt∈Bk,l,c,qr(c)
NXk=1
LXl=1
MXm=1
nXr=1 Xc∈C(r)
QXq=1 Xt∈Bk,l,c, qr(c)
MXm=1 Xc∈C(j)
LXl=1
NXk=1
nXr=1 Xc∈C(r)
NXk,k=1
MXm, m=1
LXl,l=1
k,l,m Xc∈C(j)
QXq=1 Xt∈Bk,l,c,qr(c)
QXq=1 Xt∈Bk,l,c, qr(c)
nXr=1 Xc∈C(r)
MXm, m=1
LXl,l=1
NXk,k=1
k,l,m Xc∈C(j)
QXq=1 Xt∈Bk,l,c,qr(c)
QXq=1 Xt∈Bk,l,c, qr(c)
q
Q
q
Q
k,l, m
k,l, m
=
=
Ut(r)t−11tcWk1t(i)Λk,i,j (c)Wk
Ut(i)Λk,i,j (c)c−1t−1
k,l,m ⊂ Ck,l and t ∈ T ′
Notice that, since t is associated to c ∈ C (r)
k,l,m, we have t(r) = t. Indeed, let t = (r(c)−
z, r(c)), so that t(r) = (r(Λk,r,r(c))−z, r(Λk,r,r(c))) = (r(c)−z, r(c)) = t. Now, we have
c ∈ C (r)
k,l,c ⊂ Tk,l,c ⊂ Tl, and similarly Λk,i,j(c) ∈ Ck,l and t(i) ∈ Tl,
so that tc ∈ Sk and t(i)Λk,i,j(c) ∈ Sk, so for the supports of the indicator functions in
the above sum to intersect, and hence the corresponding term to be nonzero, we must
have k = k (because levels of different towers are disjoint). Since tc and t(i)Λk,i,j(c)
are both elements of Sk, for the indicators accociated to them to intersect, we must
have tc = t(i)Λk,i,j(c). But tc ∈ Tk,l,cc, and t(i)Λk,i,j(c) ∈ Tk,l,Λk,i,j(c)Λk,i,j(c), and these
sets are disjoint unless l = l and c = Λk,i,j(c) (which implies also that m = m and
r = i). Combining this with the fact that tc = t(i)Λk,i,j(c) = t(i)c shows that t = t(i),
and hence that q = q. Therefore, we only need to sum over k, l, m, c ∈ C (j)
k,l,m, and q
and t ∈ Bk,l,c,qr(c) (because knowing c and t allows us to determine c and t). Putting
this all together yields
hψ(eij) =
NXk=1
LXl=1
MXm=1 Xc∈C(j)
k,l,m
QXq=1 Xt∈Bk,l,c, qr(c)
q
Q
Ut(i)(t(i))−11t(i)Λk,i,j (c)WkUt(i)Λk,i,j (c)c−1t−1
CLASSIFICATION OF TILING C ∗-ALGEBRAS
47
k,l,m
MXm=1 Xc∈C(j)
MXm=1 Xc∈C(j)
k,l,m
QXq=1 Xt∈Bk,l,c, qr(c)
QXq=1 Xt∈Bk,l,c, qr(c)
q
Q
q
Q
Ur(t(i))1t(i)Λk,i,j(c)Wk Ut(i)Λk,i,j(c)c−1t−1
Ut(i)Λk,i,j(c)c−1t−11tcWk
=
=
NXk=1
NXk=1
LXl=1
LXl=1
= φ(eij).
Similarly, we compute that
ψ(eij)h =
QXq=1 Xt∈Bk,l,c, qr(c)
k,l, m
MXm=1 Xc∈C(j)
LXl=1
NXk=1
nXr=1 Xc∈C(r)
MXm=1
LXl=1
NXk=1
nXr=1 Xc∈C(r)
MXm, m=1
LXl,l=1
NXk,k=1
k,l,m Xc∈C(j)
QXq=1 Xt∈Bk,l,c,qr(c)
QXq=1 Xt∈Bk,l,c, qr(c)
k,l, m
k,l,m
=
Ut(i)Λk,i,j(c)c−1t−11tcWk
QXq=1 Xt∈Bk,l,c,qr(c)
q
Q
Ut(r)Λk,r,r(c)c−1t−11tcWk
q
Q
Ut(i)Λk,i,j (c)c−1t−11tcWk
Ur(t)1tcWk.
This time, the only nonzero contribution occurs when k = k and when tc = tc. Using
similar arguments as above, this forces c = c and t = t, and hence q = q, l = l,
m = m, and r = j. Thus, the sum above becomes
ψ(eij)h =
NXk=1
LXl=1
MXm=1 Xc∈C(j)
k,l,m
QXq=1 Xt∈Bk,l,c,qr(c)
q
Q
Ut(i)Λk,i,j(c)c−1t−11tcWk
= φ(eij).
Next, we verify condition (ii) in Theorem 8.2 for the element w = Pg∈K Ug. Let
1 ≤ i, j ≤ n. Using the fact that K is closed under taking inverses in the sum over
K in the second term, we have
whk,l,c,i,j − hk,l,c,i,jw =
Ugt(i)Λk,i,j(c)c−1t−11tcWk
q
Q
QXq=1 Xt∈Bk,l,c,qr(c) Xg∈Kr(t(i))
QXq=1 Xt∈Bk,l,c, qr(c) Xg∈Kr(t)
−
q
Q
Ut(i)Λk,i,j (c)c−1t−11tcWkUg−1
=
QXq,q=1 Xt∈Bk,l,c,qr(c)
t∈Bk,l,c, qr(c)
Xg∈Kr(t(i))
g∈Kr(t)
q
Q
Ugt(i)Λk,i,j(c)c−1t−11tcWk −
q
Q
Ut(i)Λk,i,j(c)c−1(gt)−11gtcWk.
In view of (8.3) and (8.4), we may pair up the elements in the first and second terms
as follows. Because gt(i) ∈ Bk,l,c,q+a for a ∈ {−1, 0, 1}, we want to associate the
element q
QUgt(i)Λk,i,j (c)c−1t−11tcWk to an element corresponding to q = q + a. To find
48
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
an appropriate g and t, observe that by construction there is an element h of K
with source equal to r(t) which implements the same vector of translation as g. Set
t = ht, noting that by our construction, t ∈ Bk,l,c,q+a, and set g = h−1 so that gt = t.
Observe that by construction, we obtain t(i) = gt(i), so the term of the sum which
corresponds to these choices of g, g, q, q and t, t becomes
−a
Q
Ugt(i)Λk,i.j(c)c−1t−11tcWk =
−a
Q
1gt(i)Λk,i,j (c)WkUgt(i)Λk,i.j(c)c−1t−1,
and the norm of this element is no larger than 1/Q < ǫ/(n2 supu∈G(0) uK). Now, all
of the associated indicator functions are supported on tower levels gt(i)Λk,i,j(c)Wk,
and the only time two of these levels can intersect is if they are associated to two
different g ∈ K with the same range. Thus, we obtain
kwhk,l,c,i,j − hk,l,c,i,jwk ≤
supu∈G(0) uK
Q
<
ǫ
n2 .
All of the associated indicator functions in sight here are supported in K QBk,l,c,QcWk.
Since these sets are pairwise disjoint for distinct k, l, and c, we obtain
kwφ(eij) − φ(eij)wk = supk,l,c kwhk,l,c,i,j − hk,l,c,i,jwk ≤
ǫ
n2 .
Hence, for any norm-one b = (bij) ∈ Mn, we have
k[w, φ(b)]k = kwφ(b) − φ(b)wk
≤
nXi,j=1
kwφ(bij) − φ(bij)wk ≤ n2 ·
ǫ
n2 = ǫ.
Next, we check condition (ii) in Theorem 8.2 for the functions in Υ. Let i, j ∈
k,l,m.
{1, . . . , n} and f ∈ Υ ∪ Υ2. Let k ∈ {1, . . . , N} and l ∈ {1, . . . , L}. Let c ∈ C (j)
Since k, i, j are fixed in the calculation below, we write Λ = Λk,i,j for short.
h∗
k,l,c,i,jf hk,l,c,i,j =
q
Q
QXq=1 Xt∈Bk,l,c,qr(c)
QXq=1 Xt∈Bk,l,c, qr(c)
QXq,q=1 Xt∈Bk,l,c,qr(c) Xt∈Bk,l,c, qr(c)
=
∗
f
Ut(i)Λk,i,j(c)c−1t−1αtc(1Wk)
Ut(i)Λk,i,j (c)c−1t−1αtc(1Wk)
q
Q
q q
Q2 1tcWkUtcΛ(c)−1(t(i))−1f Ut(i)Λ(c)c−1t−11tcWk.
To simplify the notation, let gt = tc and ht = t(i)Λ(c). Then we have
h∗
k,l,c,i,jf hk,l,c,i,j =
=
QXq,q=1 Xt∈Bk,l,c,qr(c) Xt∈Bk,l,c, qr(c)
QXq,q=1 Xt∈Bk,l,c,qr(c) Xt∈Bk,l,c, qr(c)
q q
Q2 1gtWkαgth−1
t
(f )Ugth−1
t htgt
1gtWk
q q
Q2 αgth−1
t
(f )1gtWk1gth−1
t htWk
Ugth−1
t htg−1
t
.
Extracting the important part, this term is associated to the product of indicator
functions
1tcWk1tcΛ(c)−1(t(i))−1t(i)Λ(c)Wk .
CLASSIFICATION OF TILING C ∗-ALGEBRAS
49
The groupoid element (t(i))−1t(i) is only defined when r(t(i)) = r(t(i)). We already
know that s(t(i)) = s(t(i)) = r(Λk,i,j(c)), so, by principality of Rpunc, this element is
only defined when t(i) = t(i). Since t(i) and t(i) are associated to the same vectors
in r(Λk,i,j(c)) as t and t are in c, this shows that the only contributions to the sum
occur when t = t, and hence when q = q. Thus, we obtain
(8.5)
h∗
k,l,c,i,jf hk,l,c,i,j =
Similarly, we obtain
QXq=1 Xt∈Bk,l,c,qr(c)
q2
Q2 αtcΛk,i,j (c)−1(t(i))−1(f )1tcWk.
f h∗
k,l,c,i,jhk,l,c,i,j = f
q
Q
QXq=1 Xt∈Bk,l,c,qr(c)
QXq=1 Xt∈Bk,l,c, qr(c)
QXq,q=1 Xt∈Bk,l,c,qr(c) Xt∈Bk,l,c, qr(c)
=
∗
Ut(i)Λk,i,j(c)c−1t−1αtc(1Wk)
Ut(i)Λk,i,j(c)c−1t−1αtc(1Wk)
q
Q
q q
Q2 f 1tcWkUtcΛk,i,j(c)−1(t(i))−1t(i)Λk,i,j (c)c−1t−11tcWk.
As before, this term is only defined when ((t(i))−1, t(i)) ∈ G(2), which forces t = t and
q = q, whence we obtain
(8.6)
f h∗
k,l,c,i,jhk,l,c,i,j =
QXq=1 Xt∈Bk,l,c,qr(c)
q2
Q2 f 1tcWk.
Now, let x ∈ Wk. By definition of Ck,l,m and Λk,i,j, both Λk,i,j(c)x and cx belong
to Um, which had diameter less than η. In addition, we have d(t, t(i)) < η for all t
and i. Therefore, by the definition of η, we obtain that
f (t(i)Λk,i,j(c)x) − f (tcx) <
ǫ2
4n4
and hence
supy∈tcWk(cid:12)(cid:12)(cid:12)(cid:16)αtcΛk,i,j(c)−1(t(i))−1(f ) − f(cid:17) (y)(cid:12)(cid:12)(cid:12)
= supx∈Wk(cid:12)(cid:12)(cid:12)(cid:16)αtcΛk,i,j(c)−1(t(i))−1(f ) − f(cid:17) (tcx)(cid:12)(cid:12)(cid:12)
= supx∈Wk(cid:12)(cid:12)(cid:12)αtcΛk,i,j (c)−1(t(i))−1(f )(tcx) − f (tcx)(cid:12)(cid:12)(cid:12)
= supx∈Wk f (t(i)Λk,i,j(c)x) − f (tcx)
≤
ǫ2
4n4 .
Using this along with (8.5) and (8.6), and the fact that the tower levels tcWk are
disjoint for distinct t, we obtain
k,l,c,i,jf hk,l,c,i,j − f h∗
(8.7)
kh∗
k,l,c,i,jhk,l,c,i,jk
(8.8)
= max
q∈{1,...,Q}
sup
t∈Bk,l,c,q
q2
Q2 k(αtcΛk,i,j(c)−1(t(i))−1(f ) − f )1tcWkk <
ǫ2
3n4
Now, set w = φ(eij) and fix f ∈ Υ. Since the indicator functions associated to
hk,l,c,i,j have pairwise disjoint supports for distinct k ∈ {1, . . . , N}, l ∈ {1, . . . , L},
50
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
m ∈ {1, . . . , M} and c ∈ C (j)
k,l,m, it follows from (8.8) that
kw∗gw − gw∗wk <
ǫ2
3n4
where g is equal to f or f 2 (because in (8.8) we were working with f ∈ Υ ∪ Υ2).
From this, it follows that
kw∗f 2w − f w∗f wk ≤ kw∗f 2w − f 2w∗wk + kf 2w∗w − f w∗f wk
<
≤
<
=
ǫ2
3n4 + kf (f w∗w − w∗f w)k
ǫ2
3n4 + kf kkf w∗w − w∗f wk
ǫ2
3n4 +
2ǫ2
3n4 .
ǫ2
3n4
Hence
so that
kf w − wf k2 = k(f w − wf )∗(f w − wf )k
= kw∗f 2w − f w∗f w + f w∗wf − w∗f wf k
≤ kw∗f 2w − f w∗f wk + k(f w∗w − w∗f w)f )k
ǫ2
3n4
<
=
2ǫ2
3n4 +
ǫ2
n4
kf w − wf k <
ǫ
n2 .
Therefore, for every norm-one b = (bij) ∈ Mn, we have
k[f, φ(b)]k ≤
nXi,j=1
k[f, φ(bij)]k < n2 ·
ǫ
n2 = ǫ.
To complete the proof, we now show that 1 − φ(I) - a. By enlarging E to
include D, and shrinking δ if necessary, we can ensure that the sets S1, . . . , SN are
sufficiently left-invariant under D so that, for each k ∈ {1, . . . , N} and u ∈ G(0), the
set Rku := {s ∈ Sku Ds ⊂ Sk} has cardinality at least Sku/2. Note that, since
D−1 · O = X, we have s(D) = X = G(0). Thus, for any t ∈ G, we have t ∈ D−1Dt,
because Dt 6= ∅.
k. So, for each u ∈ G(0), R′
Fix k ∈ {1, . . . , N}. Let Rk = Fu∈G(0) Rku ⊂ Sk. Let R′
Rk with the property that the collection {Ds s ∈ R′
that if s, t ∈ Rk have Ds ∩ Dt 6= ∅, then s ∈ D−1Dt. Therefore, if R′
such a way that there exists t ∈ Rk such that R′
element of D−1Dt∩Rk in R′
we obtain a larger set which satisfies the disjointness condition on R′
the maximality of R′
each subset of the form D−1Dt for t ∈ Rku. Now, the smallest that we can make R′
k be a maximal subset of
k} is pairwise disjoint. Observe
k is chosen in
k ∩D−1Dt = ∅, then, by including any
k (for instance, the choice t ∈ D−1Dt∩Rk is always valid),
k, contradicting
ku contains at least one element from
ku
CLASSIFICATION OF TILING C ∗-ALGEBRAS
51
is to assume that we choose exactly one element from each such set, and to assume
that all of these sets are subsets of Rku. In this case, we get
R′
ku ≥
Rku
maxt∈Rku D−1Dt
≥
Sku
2 maxu∈G(0) D−1Du
.
Put P = maxu∈G(0) D−1Du, and note that P < ∞ since we assumed that Du and
D−1u = uD were uniformly bounded for u ∈ G(0).
By constructing D using the basis of clopen bisections {V (T ⊓ Br(0), t, t′)}, where
we only consider sufficiently large values of r, we may ensure that the sets Dx and
Dy consist of groupoid elements which implement the same vectors whenever x and
y are sufficiently close. This will allows us to choose R′
k to contain either all of
the groupoid elements (of some Sk,j) which make up a particular level of the tower
(Wk, Sk), or none of them. Combining this with the inequality above, we may assume
that (Wk, R′
k) is a subtower of (Wk, Sk) which contains at least one (2P )-th of the
levels of the tower (Wk, Sk).
Since D−1O = X, for each s ∈ R′
k there exists t ∈ D such that tsWk intersects
O. Indeed, s ∈ Sk, so sWk 6= ∅, and thus we can find t ∈ D and w ∈ O such that
t−1w = r(s) ∈ sWk, which implies that w ∈ tsWk and hence that tsWk ∩ O 6= ∅. By
construction of Rk, we have ts ∈ Sk, so tsWk is a subset of a level of our tower, which
was assumed to have diameter less than η′ ≤ θ. Thus, since O was the ball of radius
θ centred at x0, and since tsWk intersects O and has diameter smaller than θ, the
tower level containing tsWk is contained in the ball of radius 2θ centred at x0, and
thus a takes value 1 on this entire tower level. Furthermore, since the sets Ds for
s ∈ R′
k are disjoint, all the the elements ts ∈ Sk constructed here are distinct. Thus,
we have constructed an injective association s 7→ s of elements s ∈ R′
k to elements
s = ts ∈ Sk, such that a takes the constant value 1 on the tower level which contains
r(s). This shows that, for each u ∈ G(0), the set S♯
ku of elements t ∈ Sku such
that a takes the constant value 1 on the tower level containing r(t) has cardinality
S♯
k, then
all elements s ∈ Sk,j are in S♯
k implies that a takes the constant
value 1 on the tower level Sk,jWk. In other words, (Wk, S♯
k) is a subtower of (Wk, Sk),
and contains at least one (2P )-th of the levels of this tower.
ku ≥ Sku/(2P ). Note that, for any j, if t ∈ Sk,j is an element of S♯
k as well, since t ∈ S♯
ku ≥ R′
Set
S′′
k =
LGl=1
MGm=1
nGi=1 Gc∈C(i)
k,l,m
Bk,l,c,Qc,
noting that (Wk, S′′
v ∈ G(0), that Bk,l,c,Qv ≥ (1 − κ)Tlv, so we obtain
k ) is a subtower of (Wk, Sk), and fix u ∈ G(0). We have, for each
S′′
k u ≥
LXl=1
MXm=1
≥ (1 − κ)
= (1 − κ)
k,l,mu
nXi=1 Xc∈C(i)
MXm=1
LXl=1
MXm=1
LXl=1
Bk,l,c,Qr(c)
k,l,mu
nXi=1 Xc∈C(i)
nXi=1 Xc∈C(i)
k,l,mu
Tlr(c)
Tlc
52
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
≥ (1 − κ)
= (1 − κ)
LXl=1
LXl=1
MXm=1
MXm=1
nXi=1
nXi=1
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Tlc(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
[c∈C(i)
k,l,mu
TlC (i)
k,l,mu.
Recall that Sn
i=1 C (i)
use this to continue the computation above, obtaining
k,l,m ⊂ Ck,l,m contains all except at most n elements of Ck,l,m. We
S′′
k u ≥ (1 − κ)
TlC (i)
k,l,mu
nXi=1
≥ (1 − κ)
≥ (1 − κ)
MXm=1
LXl=1
MXm=1
LXl=1
MXm=1
LXl=1
TlCk,l,mu(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
LXl=1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
M[m=1
LXl=1
≥ (1 − κ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
TlCk,lu(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
L[l=1
≥ (1 − κ)
≥ (1 − κ)
− Mn
≥ (1 − κ) ((1 − β)Sku − βSku)
≥ (1 − κ)(1 − 2β)Sku,
LXl=1
(TlCk,l,mu − n maxv∈G(0) Tlv)
(TlCk,l,mu − n maxv∈G(0) Tlv)
− Mn maxv∈G(0) Tlv!
(TlCk,lu − Mn maxv∈G(0) Tlv)
maxv∈G(0) Tlv!
where we used the fact that {Tlc l ∈ {1, . . . , L}, c ∈ Ck,l} is a (1−β)-cover of Sk, and
the inequality (8.2) together with (E, δ)-invariance of Sk to obtain the second-to-last
line.
So, if κ and β are small enough, we can ensure that, for each u ∈ G(0), we have
Sku
4P
In this case, choose an injection fk : Sk \ S′′
(Sk \ S′′
k )u ≤
≤
.
S♯
ku
2
k which sends (Sk \ S′′
k )u to S♯
ku
for each u ∈ G(0). Further enforce that if t, t′ ∈ Sk,j for some j, and fk(t) ∈ Sk,p
for some p, then fk(t′) ∈ Sk,p as well.
In this way, fk will map the collection of
elements associated to any tower level to a collection of elements associated to some
other tower level. Since {(Wk, Sk)}N
k=1 is a clopen tower decomposition of X, we have
k → S♯
X =FN
k=1 SkWk, and hence
X \
S′′
k Wk =
(Sk \ S′′
k )Wk.
NGk=1
NGk=1
k ) and (Wk, S♯
Since (Wk, Sk \ S′′
and S♯
k =Fj∈J Sk,j,
k =Fj♯∈J ♯ Sk,j♯, where J and J ♯ are finite index sets. Observe that the collection
k) are subtowers of (Wk, Sk), write Sk \ S′′
of subsets Sk,jWk for k ∈ {1, . . . , N} and j ∈ J covers X \FN
k Wk. For each
j ∈ J, and each t ∈ Sk,j, consider the element fk(t)t−1 ∈ G, noting that s(fk(t)) =
s(t) = r(t−1) by construction of fk, so that the multiplication is defined. Then
k=1 S′′
CLASSIFICATION OF TILING C ∗-ALGEBRAS
53
k), so, since
fk was injective and all the levels SkWk of the tower (Wk, Sk) were disjoint for k ∈
Ft∈Sk,j (fk(t)t−1)tWk =Ft∈Sk,j fk(t)Wk is a level Sk,j♯ of the tower (Wk, S♯
{1, . . . , N}, the collection {Ft∈Sk,j fk(t)Wk k ∈ {1, . . . , N}, j ∈ J} of open subsets
of FN
sets which covered X \FN
kWK is pairwise disjoint, and consists of images (under the action map) of
k Wk. Thus, we have shown that
k=1 S♯
k=1 S′′
NGk=1
Since the function 1−φ(I) is supported on X \FN
value 1 on FN
NGk=1
such that v∗av = 1 − φ(I), which shows that 1 − φ(I) - a.
k Wk, and a takes the constant
kWk, it follows by Lemma 8.3 that there exists a v ∈ C(X) ⋊r G
(cid:3)
S♯
kWk.
k Wk ≺
k=1 S♯
X \
S′′
k=1 S′′
9. Quasidiagonality of tiling algebras
In this section, we give a direct proof that the C ∗-algebra Apunc associated to any
aperiodic, repetitive tiling T with finite local complexity is quasidiagonal. Note that
these algebras were already known to be quasidiagonal as a consequence of the general
result obtained by Tikuisis, White and Winter [53, Corollary B]. Our result sacrifices
generality to allow for a less abstract proof specific to tiling C ∗-algebras.
Quasidiagonality does not give us any new information regarding the classifiability
of these algebras, but it does simplify the machinery required for their classification in
certain cases. In particular, under the additional assumption that the algebra has a
unique trace, the main results of [35] and [34] show that tiling C ∗-algebras have finite
decomposition rank, and are therefore classified by their Elliott invariant. This allows
us to obtain classifiability without reference to some of the complicated constructions
culminating in [53]. We also note that [46, Theorem B] provides yet another route to
classification in the monotracial setting, which does not require quasidiagonality.
In the light of the above, for the purposes of this section, we are interested in
tilings whose C ∗-algebras have unique trace. It is shown in [25] that the C ∗-algebras
associated to aperiodic, repetitive substitution tilings with FLC have unique trace.
Recall that such C ∗-algebras were already known to be classifiable in the case that the
substitution system causes prototiles to appear in only finitely many orientations by
combining [59, Corollary 3.2] and [45] (see also [8]). However, this route to classifica-
tion relies on changing the prototile set so that the tling becomes a square tiling with
appropriate matching rules and equipped with a Zd-action, while our classification
result requires no such modification of the tiling.
In the non-substitution case, the class of quasiperiodic tilings is detailed in [43,
Section 8.2], in which it is stated [43, Corollary 8.5] (see also [42, Section 12]) that
the dynamical systems associated to these tilings are uniquely ergodic, and hence
their C ∗-algebras have unique trace.
The concept of quasidiagonality was introduced for sets operators by Halmos in
[20, Page 902], and extended to representations of C ∗-algebras by Thayer [52]. We
will use the following definition for abstract C ∗-algebras discussed in [3, Chapter 7.2].
Definition 9.1. Let H be a separable Hilbert space. A subset A ⊂ B(H) is called
a quasidiagonal set of operators if there exists an increasing sequence of finite rank
projections, P1 ≤ P2 ≤ P3 ≤ · · · , which converge strongly to the identity operator
(that is, for any x ∈ H, kPn(x) − xk → 0 as n → ∞), and are such that, for every
54
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
a ∈ A,
k[a, Pn]k := kaPn − Pnak → 0 as n → ∞.
A C ∗-algebra A is called quasidiagonal if there exists a faithful representation π :
A → B(H) such that π(A) is a quasidiagonal set of operators.
Theorem 9.2. The C ∗-algebra associated to any aperiodic and repetitive tiling with
FLC is quasidiagonal.
Proof. Let Ωpunc be the punctured hull of such a tiling, and fix a tiling T ∈ Ωpunc. In
[40] it is shown that the induced representation from the unit space π := πT described
on the generating set E defined in (6.4) extends to a faithful nondegenerate represen-
tation of Apunc on ℓ2(T ) := ℓ2([T ]). We will prove that π(Apunc) is a quasidiagonal
set of operators. Define a sequence of projections Qn ∈ B(ℓ2(T )) by
Qn(δt) :=
δt
0
if t ∈ T ⊓ Bn(0)
otherwise.
It is clear that (Qn)n∈N is an increasing sequence of projections. Since T has FLC,
Qn has finite rank for each n ∈ N, and (Qn)n∈N converges strongly to the identity by
square-summability of the elements of ℓ2(T ).
We now show that, for each element e(P, t, t′) ∈ E, we have
lim
n→∞
kQnπ(e(P, t, t′)) − π(e(P, t, t′))Qnkop = 0.
This will prove that Apunc is quasidiagonal, since span{E} = Apunc. We first observe
that
lim
n→∞
kQnπ(e(P, t, t′)) − π(e(P, t, t′))Qnkop
(9.1)
= lim
n→∞
sup
t′′∈T
kQnπ(e(P, t, t′))δt′′ − π(e(P, t, t′))Qnδt′′k2.
Set ǫ > 0. Then there exists m ∈ N such that
kQnπ(e(P, t, t′))δt′′ − π(e(P, t, t′))Qnδt′′k2
sup
t′′∈T
(9.2)
≤ ǫ +
sup
kQnπ(e(P, t, t′))δt′′ − π(e(P, t, t′))Qnδt′′k2.
t′′∈T ⊓Bm(0)
Since the supremum in (9.2) is over a finite set, we can use this bound in (9.1) and
then interchange the limit and supremum on the right-hand side to obtain
lim
n→∞
kQnπ(e(P, t, t′)) − π(e(P, t, t′))Qnkop
(9.3)
≤ ǫ +
sup
t′′∈T ⊓Bm(0)
lim
n→∞
kQnπ(e(P, t, t′))δt′′ − π(e(P, t, t′))Qnδt′′k2.
Choose n ∈ N large enough that T ⊓ Bn(0) includes all tiles that intersect the patch
Bm+kx(t′)−x(t)k(0). Then we compute that
sup
kQnπ(e(P, t, t′))δt′′ − π(e(P, t, t′))Qnδt′′k2
t′′∈T ⊓B(0,m)
sup
kQnδt′′′ − π(e(P, t, t′))δt′′k2
if P − x(t) ⊂ T − x(t′′)
t′′∈T ⊓B(0,m)
0
and x(t′′′) = x(t′′) + (x(t′) − x(t))
otherwise
=
CLASSIFICATION OF TILING C ∗-ALGEBRAS
55
sup
kδt′′′ − δt′′′k2
if P − x(t) ⊂ T − x(t′′)
t′′∈T ⊓B(0,m)
and x(t′′′) = x(t′′) + (x(t′) − x(t))
otherwise
=
0
= 0.
Since this holds for all sufficiently large n ∈ N, we see that
(9.4)
sup
t′′∈T ⊓Bm(0)
lim
n→∞
kQnπ(e(P, t, t′))δt′′ − π(e(P, t, t′))Qnδt′′k2 = 0.
Using (9.4) in (9.3), we obtain
lim
n→∞
kQnπ(e(P, t, t′)) − π(e(P, t, t′))Qnkop ≤ ǫ,
which gives the desired result.
(cid:3)
Remark 9.3. In [56], Kellendonk's construction of a tiling C ∗-algebra was extended to
include tilings with infinite rotational symmetry whose prototiles appear in infinitely
many orientations. In this case, a unitary is added to the densely spanning set E
defined in (6.4) to keep track of the rotational deviation of each tile from the standard
orientation of the prototile. This unitary can easily be incorporated into the proof
above to show that the C ∗-algebras of tilings with infinite rotational symmetry are
also quasidiagonal.
References
[1] J. E. Anderson and I. F. Putnam. Topological invariants for substitution tilings and their
associated C ∗-algebras. Ergod. Theory Dyn. Syst., 18(3):509âĂŞ537, 1998.
[2] J. Bellissard. Gap labelling theorems for Schrodinger operators. In From Number Theory to
Physics, Proc. Les Houches meeting, Berlin, 1992. Springer-Verlag.
[3] N. P. Brown and N. Ozawa. C ∗-algebras and Finite-Dimensional Approximations, volume 88
of Graduate Studies in Mathematics. American Mathematical Society, 2008.
[4] J. Castillejos, S. Evington, A. Tikuisis, S. White, and W. Winter. Nuclear dimension of simple
C ∗-algebras. 2019. arXiv:1901.05853v2 [math.OA].
[5] L. O. Clark, C. Farthing, A. Sims, and M. Tomforde. A groupoid generalisation of Leavitt path
algebras. Semigroup Forum, 89(3):501 -- 517, 2014.
[6] C. T. Conley, S. C. Jackson, D. Kerr, A. S. Marks, B. Seward, and R. D. Tucker-Drob. Følner
tilings for actions of amenable groups. Math. Ann., 371(1 -- 2):663 -- 683, 2018.
[7] J. Cuntz. Dimension Functions on Simple C ∗-Algebras. Math. Ann., 233(2):145 -- 153, 1978.
[8] R. J. Deeley and K. R. Strung. Nuclear dimension and classification of C ∗-algebras associated
to Smale spaces. Trans. Amer. Math. Soc., 370(5):3467 -- 3485, 2018.
[9] C. Ehresmann. Gattungen von lokalen Strukturen. Jahresbericht der Deutschen Mathematiker-
Vereinigung, 60:49 -- 77, 1958.
[10] G. A. Elliott. On the classification of inductive limits of sequences of semisimple finite-
dimensional algebras. J. Algebra, 38(1):29 -- 44, 1976.
[11] G. A. Elliott. On the classification of C ∗-algebras of real rank zero. J. Reine Angew. Math.,
443:179 -- 219, 1993.
[12] G. A. Elliott, G. Gong, H. Lin, and Z. Niu. On the classification of simple amenable C ∗-algebras
with finite decomposition rank, II. 2016. arXiv:1507.03437v3 [math.OA].
[13] R. Exel. Reconstructing a totally disconnected groupoid from its ample semigroup. Proc. Amer.
Math. Soc., 138(8):2991 -- 3001, 2010.
[14] A. Forrest, J. Hunton, and J. Kellendonk. Topological invariants for projection method patterns.
Mem. Amer. Math. Soc., 159(758), 2002.
[15] T. Giordano, H. Matui, I. F. Putnam, and C. F. Skau. Orbit equivalence for Cantor minimal
Zd-systems. Invent. Math., 179(1):119 -- 158, 2010.
[16] G. Goehle. Groupoid Crossed Products. PhD thesis, 2009. arXiv:0905.4681v1 [math.OA].
56
LUKE J. ITO, MICHAEL F. WHITTAKER, AND JOACHIM ZACHARIAS
[17] D. Gon¸calves and M. Ramirez-Solano. On the K-theory of C ∗-algebras for substitution tilings
(a pedestrian version). arXiv:1712.09551v2 [math.OA], 2018.
[18] G. Gong, H. Lin, and Z. Niu. Classification of finite simple amenable Z-stable C ∗-algebras.
2015. arXiv:1501.00135v6 [math.OA].
[19] E. Guentner, R. Willett, and G. Yu. Dynamic Asymptotic Dimension: relation to dynamics,
topology, coarse geometry, and C ∗-algebras. Math. Ann., 367(1 -- 2):785 -- 829, 2017.
[20] P. R. Halmos. Ten problems in Hilbert space. Bull. Amer. Math. Soc., 76(5):887 -- 933, 1970.
[21] I. Hirshberg and J. Orovitz. Tracially Z-absorbing C ∗-algebras. J. Funct. Anal., 265(5):765 -- 785,
2013.
[22] I. Hirshberg, G. Szab´o, W. Winter, and J. Wu. Rokhlin dimension for flows. Comm. Math.
Phys., 353:253 -- 316, 2017.
[23] L. J. Ito. Almost finiteness of groupoid actions and Z-stability of C ∗-algebras associated to
tilings. PhD thesis, University of Glasgow, 2019.
[24] X. Jiang and H. Su. On a simple unital projectionless C ∗-algebra. Amer. J. Math., 121(2):359 --
413, 1999.
[25] J. Kellendonk. Noncommutative Geometry of Tilings and Gap Labelling. Rev. Math. Phys.,
7(7):1133 -- 1180, 1995.
[26] J. Kellendonk and I. F. Putnam. Tilings, C ∗-algebras, and K-theory. In Directions in Math-
ematical Quasicrystals, volume 13 of CRM Monogr. Ser., pages 177 -- 206, Providence, 2000.
American Mathematical Society.
[27] D. Kerr. Dimension, comparison, and almost finiteness. J. Eur. Math. Soc. (to appear), 2019.
arXiv:1710.00393v2 [math.DS].
[28] D. Kerr and H. Li. Ergodic Theory: Independence and Dichotomies. Springer, Berlin, 2017.
[29] M. Khoshkam and G. Skandalis. Crossed products of C ∗-algebras by groupoids and inverse
semigroups. J. Operator Theory, 51(2):255 -- 279, 2004.
[30] E. Kirchberg. The classification of purely infinite C ∗-algebras using Kasparov's theory. Preprint.
[31] E. Kirchberg and N. C. Phillips. Embedding of exact C ∗-algebras in the Cuntz algebra O2. J.
Reine Angew. Math., 525:17 -- 54, 2000.
[32] M. Mampusti and M. F. Whittaker. Fractal spectral triples on Kellendonk's C ∗-algebra of a
substitution tiling. J. Geom. Phys., 112:224 -- 239, 2017.
[33] H. Matui. Homology and topological full groups of ´etale groupoids on totally disconnected
spaces. Proc. Lond. Math. Soc., 104(1):27 -- 56, 2012.
[34] H. Matui and Y. Sato. Strict comparison and Z-absorption of nuclear C ∗-algebras. Acta Math.,
209(1):179 -- 196, 2012.
[35] H. Matui and Y. Sato. Decomposition rank of UHF-absorbing C ∗-algebras. Duke Math. J.,
163(14):2687 -- 2708, 2014.
[36] P. Muhly and D. P. Williams. Renault's Equivalence Theorem for Groupoid Crossed Products.
New York Journal of Mathematics Monographs, 3:1 -- 83, 2008.
[37] D. S. Ornstein and B. Weiss. Entropy and isomorphism theorems for actions of amenable groups.
J. Analyse Math., 48:1 -- 141, 1987.
[38] N. C. Phillips. A classification theorem for nuclear purely infinite simple C ∗-algebras. Doc.
Math., 5:49 -- 114, 2000.
[39] N. C. Phillips. Crossed products of the Cantor set by free minimal actions of Zd. Comm. Math.
Phys., 256:1 -- 42, 2005.
[40] J. Renault. A Groupoid Approach to C ∗-Algebras, volume 793 of Lect. Notes in Math. Springer,
Berlin, 1980.
[41] J. Renault. Repr´esentation des produits crois´es d'alg`ebres de groupoıdes. J. Operator Theory,
18(1):67 -- 97, 1987.
[42] E. A. Robinson Jr. The Dynamical Theory of Tilings and Quasicrystallography. In Ergodic
Theory of Zd-Actions, volume 228 of London Math. Soc. Lecture Note Ser., pages 451 -- 474.
Cambridge Univ. Press, 1996.
[43] E. A. Robinson Jr. Symbolic Dynamics and Tilings of Rd. In Symbolic Dynamics and its Ap-
plications, volume 60 of Proc. Sympos. Appl. Math., pages 81 -- 120, Providence, 2004. Amer.
Math. Soc.
[44] M. Rørdam. A simple C ∗-algebra with a finite and an infinite projection. Acta Math.,
191(1):109 -- 142, 2003.
CLASSIFICATION OF TILING C ∗-ALGEBRAS
57
[45] L. Sadun and R. F. Williams. Tiling spaces are Cantor set fiber bundles. Ergod. Theory Dyn.
Syst., 23(1):307 -- 316, 2003.
[46] Y. Sato, S. White, and W. Winter. Nuclear dimension and Z-stability. Invent. Math.,
202(2):893 -- 921, 2015.
[47] A. Sierakowski. Discrete Crossed product C ∗-algebras. PhD thesis, 2009.
[48] A.
Sims. Hausdorff
groupoids
´etale
and
their C ∗-algebras.
Lecture
notes,
arXiv:1710.10897v2 [math.OA], 2018.
[49] C. Starling. Finite symmetry group actions on substitution tiling C ∗-algebras. Munster J.
Math., 7:381 -- 412, 2014.
[50] Y. Suzuki. Almost finiteness for general ´etale groupoids and its applications to stable rank of
crossed products. Int. Math. Res. Notices (to appear), 2018. arXiv:1702.04875v3 [math.OA].
[51] G. Szab´o. The Rokhlin dimension of topological Zm-actions. Proc. Lond. Math. Soc., 110:673 --
694, 2015.
[52] F. J. Thayer. Quasi-diagonal C ∗-algebras. J. Funct. Anal., 25(1):50 -- 57, 1977.
[53] A. Tikuisis, S. White, and W. Winter. Quasidiagonality of nuclear C ∗-algebras. Ann. Math.,
185(1):229 -- 284, 2017.
[54] A. S. Toms. On the independence of K-theory and stable rank for simple C ∗-algebras. J. Reine
Angew. Math., 578:185 -- 199, 2005.
[55] A. S. Toms. On the classification problem for nuclear C ∗-algebras. Ann. Math., 167(3):1029 --
1044, 2008.
[56] M. F. Whittaker. C ∗-algebras of tilings with infinite rotational symmetry. J. Operator Theory,
64(2):299 -- 319, 2010.
[57] D. P. Williams. Crossed products of C ∗-algebras, volume 134 of Mathematical Surveys and
Monographs. American Mathematical Society, 2007.
[58] W. Winter. Nuclear dimension and Z-stability of pure C ∗-algebras. Invent. Math., 187(2):259 --
342, 2012.
[59] W. Winter. Classifying crossed product C ∗-algebras. Amer. J. Math., 138(3):793 -- 820, 2016.
[60] W. Winter and J. Zacharias. The nuclear dimension of C ∗-algebras. Adv. Math., 224(2):461 -- 498,
2010.
Luke J. Ito, Michael F. Whittaker and Joachim Zacharias, School of Mathematics
and Statistics, University of Glasgow, University Place, Glasgow Q12 8QQ, United
Kingdom
E-mail address: [email protected]
E-mail address: [email protected]
E-mail address: [email protected]
|
1101.1694 | 3 | 1101 | 2015-01-11T03:18:45 | Quantum Functions | [
"math.OA"
] | Weaver has recently defined the notion of a quantum relation on a von Neumann algebra. We demonstrate that the corresponding notion of a quantum function between two von Neumann algebras coincides with that of a normal unital $*$-homomorphism in the opposite direction. This is essentially a reformulation of a previously known result from the theory of Hilbert von Neumann modules. | math.OA | math |
QUANTUM FUNCTIONS
ANDRE KORNELL
Abstract. Weaver has recently defined the notion of a quantum relation on
a von Neumann algebra. We demonstrate that the corresponding notion of a
quantum function between two von Neumann algebras coincides with that of
a normal unital ∗-homomorphism in the opposite direction. This is essentially
a reformulation of a previously known result from the theory of Hilbert von
Neumann modules.
A relation between sets X and Y is simply a subset of Y × X. Motivated
by Kuperberg and Weaver's work on quantum metrics [1], Weaver has recently
proposed the following generalization of relations to the noncommutative setting:
Definition (Weaver, [6, Definition 2.1]). Let M ⊆ B(H) and N ⊆ B(K) be von
Neumann algebras. A quantum relation between M and N is an ultraweakly closed
subspace V ⊆ B(H, K) such that N ′VM′ ⊆ V.
This definition reduces to the usual one when M = ℓ∞(X) and N = ℓ∞(Y ).
It has many other virtues. It is simple to state and simple to handle, and many
familiar properties of relations have natural analogs.
Definition (Weaver, [6, Definition 2.4]). Let M0, M1 and M2 be von Neumann
algebras.
• The diagonal quantum relation on M0 is the quantum relation M′
0 between
M0 and M0.
• If V is a quantum relation between M0 and M1, then the inverse of V is
the quantum relation V ∗ between M1 and M0.
• If V0 is a quantum relation between M0 and M1, and V1 is a quantum
relation between M1 and M2, then their composition V1◦V0 is the quantum
relation between M0 and M2 defined by
V1 ◦ V0 = V1V0 = span{v1v0 v1 ∈ V1, v0 ∈ V0}
ultraweak
.
If we interpret inclusion between quantum relations as the proper generaliza-
tion of inclusion between classical relations, we arrive immediately at the following
definitions.
Definition (Weaver, [6, Definition 2.4]). Let V be a quantum relation on a von
Neumann algebra M. Then V is said to be
• reflexive in case M′ ⊆ V,
• symmetric in case V ∗ = V,
• antisymmetric in case V ∩ V ∗ ⊆ M′, and
• transitive in case VV ⊆ V.
The research reported here was supported by National Science Foundation grant DMS-0753228.
1
2
ANDRE KORNELL
Thus, Weaver has generalized a large class of mathematical objects including
orderings, graphs, equivalence relations, etc.
In foundations, a function from a set X to a set Y is typically defined as a
relation F ⊆ Y × X such that for each element x ∈ X, there is exactly one element
y ∈ Y such that (y, x) ∈ F . Denoting the diagonal relations on X and Y by ∆X
and ∆Y respectively, we may restate this condition as a pair of inequalities:
(1) ∆X ⊆ F ∗ ◦ F
(2) F ◦ F ∗ ⊆ ∆Y
It is therefore natural to investigate the quantum relations that satisfy the analogs
of these inequalities, i.e., quantum functions. In fact, we show that they correspond
exactly to the normal unital ∗-homomorphisms:
Definition. Let M ⊆ B(H) and N ⊆ B(K) be von Neumann algebras. A quantum
function from M to N is a quantum relation V between M and N such that
M′ ⊆ V ∗V and VV ∗ ⊆ N ′.
Theorem. Let M ⊆ B(H) and N ⊆ B(K) be von Neumann algebras. There
is a canonical bijective correspondence between normal unital ∗-homomorphisms
N −→ M, and quantum functions from M to N . This correspondence is functorial.
By chance, Debashish Goswami visited Berkeley soon after I posted this paper
to arxiv.org. He explained that, although the motivation and the statement of
theorem are new, the proof is essentially already known. He recommended his book
Quantum Stochastic Processes and Noncommutative Geometry [4] as a reference.
Specifically, the core argument is essentially the proof of [4, Theorem 4.2.7]; the
condition VV ∗ ⊆ N ′ implies that V ∗ is a right Hilbert N ′-module. However, this
theorem cannot be applied directly because Hilbert von Neumann modules are
defined to be closed in the strong operator topology, whereas a quantum relation
is defined to be closed in the ultraweak topology. Later, Alexandru Chirvasitu
pointed out that both theorems can be obtained from [3, Proposition 6.12]. I thank
Alexandru Chirvasitu and Debashish Goswami for their explanations. I also thank
my advisor, Marc Rieffel, for suggesting Nik Weaver's papers to me.
Conventions used in this article. Let H be a Hilbert space.
If ξ ∈ H, then
ξ ∈ B(C, H) is defined by ξ(c) = cξ. If V and W are ultraweakly closed subspaces of
B(H), then VW = span{vw v ∈ V, w ∈ W}
The tensor product of two Hilbert spaces is defined in such a way that H ⊗ C =
H = C ⊗ H. The von Neumann algebra of scalar operators on H is denoted by CH.
, and V⊗W = span{v ⊗ w v ∈ V, w ∈ W}
uw
uw
.
Definition. Let M and N be von Neumann algebras. Then vN(N , M) denotes
the set of normal unital ∗-homomorphisms N −→ M, and qF(M, N ) denotes the
set of quantum functions from M to N .
1. Functions from Homomorphisms
Let M ⊆ B(H) and N ⊆ B(K) be von Neumann algebras, and let π : N −→ M
be a normal unital ∗-homomorphism.
Theorem 1.1 (Dixmier, [5, Theorem IV.5.5]). There is a Hilbert space L and an
isometry w ∈ B(H, K ⊗ L) such that π(b) = w∗(b ⊗ 1)w.
Let L and w be as in Theorem 1.1 above.
QUANTUM FUNCTIONS
3
Proposition 1.2. For all b ∈ N , (b ⊗ 1)w = wπ(b).
Proof. For all b ∈ N ,
ww∗(b∗ ⊗1)(b⊗1)ww∗ = wπ(b∗b)w∗ = wπ(b∗)π(b)w∗ = ww∗(b∗⊗1)ww∗(b⊗1)ww∗,
so ww∗(b∗ ⊗ 1)(1 − ww∗)(b ⊗ 1)ww∗ = 0. Since 1 − ww∗ is a projection, we conclude
that (1 − ww∗)(N ⊗CL)ww∗ = 0, i.e., ww∗ ∈ (N ⊗CL)′. Therefore, for all b ∈ N ,
(b ⊗ 1)w = (b ⊗ 1)ww∗w = ww∗(b ⊗ 1)w = wπ(b).
(cid:3)
Definition 1.3. The set G(π) = {v ∈ B(H, K) ∀b ∈ N bv = vπ(b)} is a quantum
relation between M and N .
Proposition 1.4. M′ ⊆ G(π)∗G(π)
Proof. Choose a basis {eα}α∈I of L. For each α ∈ I, (1 ⊗ e∗
b(1 ⊗ e∗
It follows that
α)(b ⊗ 1)w = (1 ⊗ e∗
α)w = (1 ⊗ e∗
α)wπ(b) for all b ∈ N , i.e., (1 ⊗ e∗
α)w ∈ B(H, K) satisfies
α)w ∈ G(π).
1 = w∗w =Xα∈I
α)w =Xα∈I
w∗(1 ⊗ eαe∗
((1 ⊗ e∗
α)w)∗ ((1 ⊗ e∗
α)w) ,
so CH ⊆ G(π)∗G(π). We conclude that M′ = CHM′ ⊆ G(π)∗G(π)M′ = G(π)∗G(π).
Proposition 1.5. G(π)G(π)∗ ⊆ N ′
(cid:3)
Proof. For all v0, v1 ∈ G(π), and all b ∈ N , bv0v∗
v0v∗
1 b.
1 = v0π(b)v∗
1 = v0(v1π(b∗))∗ =
(cid:3)
Proposition 1.6. Therefore G(π) is a quantum function from M to N .
Thus, we have defined a function G : vN(N , M) −→ qF(M, N ).
2. Homomorphisms from Functions
Let M ⊆ B(H) and N ⊆ B(K) be von Neumann algebras, and V a quantum
function from M to N . The following lemma is a special case of Paschke's structural
theorem for self-dual Hilbert W ∗-modules [2, Theorem 3.12].
Lemma 2.1. There exists a family Lemma {uα}α∈I of partial isometries in V such
that
(1) for all distinct α, β ∈ I, uαu∗
β = 0, and
αuα = 1.
(2) P u∗
Proof. Let F be the collection of all sets S of partial isometries in V such that
uu∗ = 0 whenever u, u ∈ S are distinct. Applying Zorn's Lemma, we obtain a
maximal such set S.
Suppose that Pu∈ S u∗u 6= 1, and let p = 1 −Pu∈ S u∗u. The subspace Vp is
non-zero because 1 ∈ M′ ⊆ V ∗V. Therefore, pick v 6= 0 in Vp ⊆ V. We will now
obtain a partial isometry to add to S from the polar decomposition of v.
Let W ∗(V) be the von Neumann algebra generated by V, which is the ultraweakly
closed subspace of B(H⊕K) generated by finite words in the elements of V and their
conjugates. Since VV ∗V ⊆ N ′V ⊆ V, W ∗(V) is in fact generated by words of one of
the following forms: 1, v0, v∗
0 . We conclude that [K]W ∗(V)[H] = V,
where [H] and [K] denote projections onto H and K respectively.
1 v0, and v1v∗
0, v∗
4
ANDRE KORNELL
Let v = uvv be the polar decomposition of x. Since [K]v[H] = v, [K]uv[H] = uv,
so uv ∈ V. Furthermore, since vp = v, uvp = uv, so uvu∗ = 0 for all u ∈ S. Thus,
S ∪ {uv} is an element of F strictly larger than S, a contradiction.
(cid:3)
Definition 2.2. Let G−1(V) be the normal unital ∗-homomorphism defined by
G−1(V)(b) =Xα∈I
u∗
αbuα = w∗
I (b ⊗ 1)wI ,
uαu∗
αuα = 1 and
β = 0 whenever α 6= β, and the isometry wI ∈ B(H, K ⊗ ℓ2(I)) is defined by
where {uα}α∈I is any family of partial isometries in V such that P u∗
wI ξ =Pα∈I uαξ ⊗ eα.
Proposition 2.3. The normal unital ∗-homomorphism G−1(V) is well defined.
Proof. Apply Lemma 2.1 above to obtain a family {uα}α∈I of partial isometries in
αuα = 1 and uαu∗
B(H, K ⊗ ℓ2(I)). For all ξ ∈ H,
V such that P u∗
kwI (ξ)k2 =Xα,β
huα(ξ)⊗eαuβ(ξ)⊗eβi =Xα
β = 0 whenever α 6= β. Let wI = P uα ⊗ eα ∈
αuα! ξi = kξk2.
huαξuαξi = hξ Xα
u∗
Thus, wI is an isometry, and we may now define a normal unital completely positive
map πI : N −→ B(H) by πI (b) = w∗
αbuα. For all b0, b1 ∈ N ,
because uαu∗
For all b ∈ N , c ∈ M′,
α ∈ VV ∗ ⊆ N ′. We conclude that πI is a normal unital ∗-homomorphism.
cπI (b) = Xα,β∈I
u∗
αuαcu∗
βbuβ = Xα,β∈I
u∗
αbuαcu∗
βuβ = πI (b)c
β ∈ VM′V ∗ ⊆ VV ∗ ⊆ N ′. By the Double Commutant Theorem,
because uαcu∗
πI (N ) ⊆ M, so πI may be viewed as a normal unital ∗-homomorphism N −→ M.
αuα = 1
β = 0 for distinct α, β ∈ J, we may obtain in the same way a normal unital
If {uα}α∈J is another family of partial isometries that satisfies Pα∈J u∗
and uαu∗
∗-homomorphism πJ : N −→ M. However, for all b ∈ N ,
πJ (b) = Xα∈I,β∈J
αuαu∗
u∗
βbuβ = Xα∈I,β∈J
αbuαu∗
u∗
βuβ = πI (b)
because uαu∗
and we may define G−1(V) = πI .
β ∈ VV ∗ ⊆ N ′. Thus, πI is independent of our choice of family {uα}α∈I,
(cid:3)
Thus, we have defined a function G−1 : qF(M, N ) −→ vN(N , M).
I (b ⊗ 1)wI =P u∗
αb0uα!
Xβ
u∗
u∗
αb0uαu∗
αb1uα
u∗
βb1uβ
u∗
αb0b1uαu∗
αuα
u∗
αb0b1uα = πI (b0b1)
πI (b0)πI (b1) = Xα
=Xα
=Xα
=Xα
QUANTUM FUNCTIONS
5
3. G−1 is the Inverse of G
Let M ⊆ B(H) and N ⊆ B(K) be von Neumann algebras.
Proposition 3.1. Let π : N −→ M be a normal unital ∗-homomorphism. Then
G−1(G(π)) = π.
Proof. Let {uα}α∈I be a family of partial isometries in G(π) such thatPα∈I u∗
β = 0 whenever α 6= β. For all b ∈ N ,
1 and uαu∗
αuα =
G−1(G(π))(b) =Xα∈I
u∗
αbuα =Xα∈I
u∗
αuαπ(b) = π(b).
(cid:3)
Proposition 3.2. Let L be a Hilbert space, and let w0, w1 ∈ B(H, K ⊗ L) be
isometries such that πi(b) = w∗
i (b ⊗ 1)wi defines a pair of ∗-homomorphisms N −→
M. If π0 = π1, then w0w∗
1 ∈ (N ⊗CL)′.
Proof. For all b ∈ N , (b ⊗ 1)w0w∗
w0((b∗ ⊗ 1)w1))∗ = w0w∗
1(b ⊗ 1).
1 = w0π0(b)w∗
1 = w0π1(b)w∗
1 = w0(w1π1(b∗))∗ =
(cid:3)
Lemma 3.3. Let V be a quantum function from M to N , and let {uα}u∈I and wI
be as in Definition 2.2. Then (N ⊗Cℓ2(I))′wIM′ = V⊗B(C, ℓ2(I)).
Proof. By definition, wI = P uα ⊗ eα ∈ V⊗B(C, ℓ2(I)), so (N ⊗Cℓ2(I))′wIM′ =
(N ′⊗B(ℓ2(I)))wI M′ ⊆ V⊗B(C, ℓ2(I)).
Let f ∈ ℓ2(I) and v ∈ V. Then for all α ∈ I,
uα ⊗ f =Xβ∈I
so
(1 ⊗ f e∗
α)(uβ ⊗ eβ) = (1 ⊗ f e∗
α)wI ∈ (N ⊗Cℓ2(I))′wIM′,
v ⊗ f =Xα∈I
(v ⊗ f )(u∗
αuα) =Xα∈I
(vu∗
α ⊗ 1)(uα ⊗ f ) ∈ (N ⊗Cℓ2(I))′wIM′
because vu∗
concluding the proof.
α ∈ VV ∗ ⊆ N ′.
It follows that V⊗B(C, ℓ2(I)) ⊆ (N ⊗Cℓ2(I))′wIM′,
(cid:3)
Proposition 3.4. The function G−1 is injective.
Proof. For k ∈ {0, 1}, let Vk ∈ qF(M, N ), and let {uα}α∈Ik be a family of partial
isometries in Vk such that uαu∗
αuα = 1. We
may assume that I0 and I1 have equal, non-zero cardinality by throwing in indexed
instances of the zero partial isometry where necessary. Thus, we may choose a
unitary s ∈ B(ℓ2(I0), ℓ2(I1)).
β = 0 for distinct α, β ∈ Ik, and P u∗
Suppose that G−1(V0) = G−1(V1). For all b ∈ N ,
((1 ⊗ s)wI0 )∗(b ⊗ 1)((1 ⊗ s)wI0 ) = w∗
I0 (b ⊗ 1)wI0 = G−1(V0)(b)
= G−1(V1)(b) = w∗
I1 (b ⊗ 1)wI1
By Proposition 3.2, (1 ⊗ s)wI0 w∗
I1 ∈ (N ⊗Cℓ2(I1))′.
6
ANDRE KORNELL
By Lemma 3.3 above,
V1⊗B(C, ℓ2(I1)) = (N ⊗Cℓ2(I1))′wI1M′
I1 )wI1M′
⊇ (N ⊗Cℓ2(I1))′((1 ⊗ s)wI0 w∗
= (N ⊗Cℓ2(I1))′(1 ⊗ s)wI0M′
= (1 ⊗ s)(N ⊗Cℓ2(I0))′wI0M′
= (1 ⊗ s)(V0⊗B(C,ℓ2(I0)))
= V0⊗B(C, ℓ2(I1)).
Choosing an arbitrary unit vector f ∈ ℓ2(I1), we conclude that
V1 = (1 ⊗ f ∗)(cid:0)V1⊗B(C, ℓ2(I1))(cid:1) ⊇ (1 ⊗ f ∗)(cid:0)V0⊗B(C, ℓ2(I1))(cid:1) = V0.
Similarly, V1 ⊆ V0, so V1 = V0.
(cid:3)
Theorem 3.5. Let M ⊆ B(H) and N ⊆ B(K) be von Neumann algebras. The
function
G : vN(N , M) −→ qF(M, N )
defined by
is a bijection.
G(π) = {v ∈ B(H, K) ∀b ∈ N bv = π(b)v}
Proof. The theorem follows immediately from Propositions 3.1 and 3.4.
(cid:3)
4. Functoriality
Proposition 4.1. Let M0, M1, and M2 be von Neumann algebras.
qF(M0, M1) and V1 ∈ qF(M1, M2), then V1V0 ∈ qF(M0, M2).
If V0 ∈
Proof.
0 ⊆ V ∗
0 V0 ⊆ V ∗
M′
(V1V0)(V1V0)∗ = V1V0V ∗
0 M′
1V0 ⊆ V ∗
0 V ∗
1 V1V0 = (V1V0)∗(V1V0)
1 ⊆ M′
2
1 ⊆ V1V ∗
1V ∗
0 V1 ⊆ V1M′
(cid:3)
Proposition 4.2. Let M0, M1, and M2 be von Neumann algebras. If π1 : M2 −→
M1 and π0 : M1 −→ M0 are normal unital ∗-homomorphisms, then G(π0 ◦ π1) =
G(π1)G(π0).
Proof. Clearly, G(π1)G(π0) ⊆ G(π0 ◦ π1). By Definition 2.2, G−1(G(π1)G(π0)) =
G−1(G(π0 ◦ π1)). By Theorem 3.5, we conclude that G(π1)G(π0) = G(π0 ◦ π1). (cid:3)
Proposition 4.3. Let M be a von Neumann algebra, and let ι : M −→ M be the
identity ∗-homomorphism. Then G(ι) = M′.
Proof. This is an immediate consequence of Definition 1.3.
(cid:3)
Definition 4.4. Let vN be the category whose objects are von Neumann algebras,
and whose morphisms are normal unital ∗-homomorphisms.
Definition 4.5. Let qF be the following category:
• The objects of qF are von Neumann algebras.
• For any two objects M and N , a morphism from M to N is a quantum
function from M to N .
QUANTUM FUNCTIONS
7
• For any two morphisms V0 ∈ qF(M0, M1) and V1 ∈ qF(M1, M2), V1 ◦
V0 = V1V0.
• For any object M, the identity morphism at M is the quantum function
M′ ∈ qF(M, M).
Theorem 4.6. The functor G : vN −→ qF defined by
• for all objects M of vN, G(M) = M, and
• for all morphisms π ∈ vN(N ⊆ B(K), M ⊆ B(H)),
G(π) = {v ∈ B(H, K) ∀b ∈ N bv = vπ(b)},
is a coisomorphism of categories.
Proof. This is a straightforward consequence of Proposition 4.2, Proposition 4.3,
and Theorem 3.5.
(cid:3)
References
[1] G. Kuperberg and N. Weaver, A von Neumann algebra approach to quantum metrics,
arXiv:1005.0353v2.
[2] W. L. Paschke, Inner product modules over B∗-algebras, Transactions of the Americal Math-
ematical Society 182 (1973), 443-468.
[3] M. A. Rieffel, Morita equivalence for C*-algebras and W*-algebras, Journal of Pure and Ap-
plied Algebra 5 (1974), 51-96.
[4] K. B. Sinha and D. Goswami, Quantum Stochastic Processes and Noncommutative Geometry,
Cambridge Tracts in Mathematics, vol. 169, Cambridge University Press, 2007.
[5] M. Takesaki, Theory of Operator Algebras I, Springer, 1979.
[6] N. Weaver, Quantum relations, arXiv:1005.0354v1.
Department of Mathematics, University of California, Berkeley, CA 94720-3840
E-mail address: [email protected]
|
1605.00038 | 1 | 1605 | 2016-04-29T22:53:47 | Anticommutation in the Presentations of Theta-Deformed Spheres | [
"math.OA",
"math.AT"
] | We consider an analogue of the theta-deformed even spheres, modifying the relations demanded of the self-adjoint generator x in the usual presentation. In this analogue, x is given anticommutation relations with all of the other generators, as opposed to being central. Using even K-theory, we show that noncommutative Borsuk-Ulam theorems hold for these algebras. | math.OA | math | Anticommutation in the Presentations of Theta-Deformed Spheres
6
1
0
2
r
p
A
9
2
]
Benjamin Passer∗
Washington University
St. Louis, MO 63130
[email protected]
May 3, 2016
Abstract
.
A
O
h
t
a
m
[
1
v
8
3
0
0
0
.
5
0
6
1
:
v
i
X
r
a
We consider an analogue of the θ-deformed even spheres, modifying the relations demanded of
the self-adjoint generator x in the usual presentation. In this analogue, x is given anticommutation
relations with all of the other generators, as opposed to being central. Using even K-theory, we
show that noncommutative Borsuk-Ulam theorems hold for these algebras.
Keywords: Borsuk-Ulam, anticommutation, noncommutative suspension
MSC2010: 46L85
1 Introduction
When a Rieffel deformation procedure ([11]) is applied to the function algebra C(Sk) of a sphere Sk,
the resulting C ∗-algebra admits a succinct presentation ([7], [4]).
Definition 1.1. Let ρ be an n×n matrix with 1 on the diagonal, all entries unimodular, and ρjk = ρkj .
Then C(S2n−1
ρ ) are given by the following C ∗-presentations.
) and C(S2n
ρ
C(S2n−1
ρ
) ∼= C ∗(z1, . . . , zn zjz ∗
j = z ∗
j zj, zkzj = ρjkzjzk, z1z ∗
1 + . . . + znz ∗
n = 1)
C(S2n
ρ ) ∼= C ∗(z1, . . . , zn, x zjz ∗
z1z ∗
j = z ∗
1 + . . . + znz ∗
j zj, x = x∗, zkzj = ρjkzjzk, xzj = zjx,
n + x2 = 1)
The presentation and notation illustrate the mentality suggested by the Gelfand-Naimark theorem,
in that we view C(Sk
ρ) as a (noncommutative) function algebra with coordinate functions labeled zj
or x. The noncommutativity relations of C(Sk
ρ ) is realized as
a quotient C(S2n+1
n+1i, where ω is such that zn+1 is central and ρ is the upper left n × n
submatrix of ω. However, this type of quotient is reasonable (that is, nondegenerate) even when zn+1
anticommutes with some of the other zj. We consider below the quotients C(S2n+1
n+1i
when anticommutation always occurs.
ρ) vary in a continuous fashion, and C(S2n
)/hzn+1 − z ∗
)/hzn+1 − z ∗
ω
ω
∗Partially supported by National Science Foundation Grants DMS 1300280 and DMS 1363250
1
Definition 1.2. Let ρ be an n×n matrix with 1 on the diagonal, all entries unimodular, and ρjk = ρkj .
Then R2n
ρ
is defined by the following presentation.
R2n
ρ
∼= C ∗(z1, . . . , zn, x zjz ∗
z1z ∗
j = z ∗
1 + . . . + znz ∗
n + x2 = 1)
j zj, x = x∗, zkzj = ρjkzjzk, xzj = −zjx,
(1.3)
We adopt the convention that R2n without a subscript denotes R2n
ρ
algebra, z1, . . . , zn commute with each other and anticommute with x.
for ρ a matrix of all ones. In this
Each R2n
ρ may be realized as a noncommutative unreduced suspension (in the sense of [9], Def-
inition 3.4) of the unital C ∗-algebra C(S2n−1
) given by the antipodal map, which is the order two
homomorphism that negates each generator of the standard presentation. In general, if β generates a
Z2 action on a unital C ∗-algebra A, then
ρ
ΣβA := {f ∈ C([0, 1], A ⋊β Z2) : f (0) ∈ A, f (1) ∈ C ∗(Z2)}
defines the noncommutative unreduced suspension of (A, β, Z2). When β is trivial, ΣβA is isomorphic
to the unreduced suspension ΣA; this was considered in [5] and [9] in pursuit of noncommutative
Borsuk-Ulam theory.
Theorem 1.4 (Borsuk-Ulam). No continuous, odd maps exist from Sk to Sk−1, and every continuous,
odd map on Sk−1 is homotopically nontrivial.
Noncommutative Borsuk-Ulam theorems ([13], [15], [2], [8], [5], [9], [3]) arise when a C ∗-algebraic
theorem generalizes the result of passing the traditional Borsuk-Ulam theorem, or one of its various
restatements and extensions, through Gelfand-Naimark duality.
In short, these theorems describe
when an equivariant map between two related C ∗-algebras cannot exist, or when an equivariant self-
map must behave nontrivially with respect to some invariant. The θ-deformed odd spheres were
considered for this purpose in [8] with Zk rotation actions; the Z2 case is repeated below.
Theorem 1.5. Let α denote the antipodal Z2-action on any C(Sk
of the standard presentation. Then any α-equivariant, unital ∗-homomorphism from C(S2n−1
C(S2n−1
equivariant, unital ∗-homomorphisms from C(Sk
ρ), which negates each generator
) to
) must induce a nontrivial map on K1 ∼= Z. Consequently, for any k, there are no α-
γ) to C(Sk+1
).
ω
ρ
δ
The K-theory of θ-deformed spheres exactly matches that of the commutative case, as Rieffel
deformation preserves K-theory ([12]), but there are more detailed descriptions ([7], [10]).
K0(C(S2n−1
ρ
)) ∼= Z K1(C(S2n−1
ρ
)) ∼= Z
K0(C(S2n
ρ )) ∼= Z ⊕ Z K1(C(S2n
ρ )) ∼= {0}
On the other hand, the anticommutation present in R2n
is not a matter that can be resolved contin-
ρ
uously, as the generator x is self-adjoint. Despite this obstruction, a noncommutative Borsuk-Ulam
theorem does hold.
Theorem 1.6. Let A2n, B2n be C ∗-algebras of the type R2n
ρ ) (the types may differ or use
different matrices ρ), both of which are equipped with an antipodal map that negates each generator.
Both algebras have K0 ∼= Z ⊕ Z, with the first summand generated by the trivial projection 1. Any
equivariant, unital ∗-homomorphism from A2n to B2n must induce a K0 map that is nontrivial on the
second summand.
ρ or C(S2n
2
ω =
ρ
1
1
...
. . . 1
=⇒ C(S2n+1
ω
2 Commutation
As seen in [5], it is known that the even θ-deformed sphere C(S2n
suspension of C(S2n−1
).
ρ
ρ ) may be written as the unreduced
C(S2n
ρ ) ∼= ΣC(S2n−1
ρ
) ∼= {f ∈ C([−1, 1], C(S2n−1
ρ
)) : f (−1) ∈ C, f (1) ∈ C}
(2.1)
One approach is to define a homomorphism C(S2n
and then to see that the realization of C(S2n−1
Sn−1 to the quantum torus C(Tn
C(S2n+1
algebra of functions on a disk, instead of an interval.
ρ ),
) as an algebra of functions from the positive section of
ρ ) ([7]) helps produce an inverse map. Further, when the odd sphere
)/hzn+1 − z ∗
) itself is isomorphic to a similar
) using the presentation of C(S2n
) is such that C(S2n+1
ρ ) → ΣC(S2n−1
n+1i ∼= C(S2n
ρ ), C(S2n+1
ω
ω
ω
ρ
ρ
) ∼= {g ∈ C(D, C(S2n−1
ρ
)) : gS1 is C-valued}
(2.2)
The antipodal maps for these algebras of functions, which are indexed below by dimension, satisfy
the following identities.
α2n(f )[x] = α2n−1(f (−x)), x ∈ [−1, 1]
α2n+1(g)[u] = α2n−1(g(−u)), u ∈ [−1, 1]
To prove noncommutative Borsuk-Ulam theorems for even spheres, we note that the odd dimensional
case relied on the following idea from [8], the backbone of Theorem 1.5.
)) ∼= Z,
Lemma 2.3. If U ∈ U(C(S2n−1
[U ] is an even integer. Because there is a dimension 2n−1 unitary matrix Zρ(n) with odd entries and
[Zρ(n)] = 1, it follows that any unitary matrix W of dimension (2m + 1)2n−1 with odd entries must
have [W ] ∈ 2Z + 1.
)) is a unitary matrix with even entries, then in K1(C(S2n−1
ρ
ρ
The lemma showcases the particularly nice interaction between K1 and the antipodal action for an
odd sphere. The K1(C(S2n−1
)) generator Zρ(n) has ∗-monomial entries (see [7]), which are all odd.
Before considering anticommutation relations, we seek to establish a related identity on even spheres.
ρ )) ∼= Z ⊕ Z, there is a distinguished projection
Motivated by [6], [4], and [10], we see that in K0(C(S2n
ρ
where Zρ(n) is the same formal ∗-monomial matrix described above. When Zρ(n) is viewed with
entries in C(S2n
j )I2n−1 = (1 − x2)I2n−1 , and it also
commutes with xI2n−1 . The projection Pρ(n) on the even sphere, much like its analogous unitary
Zρ(n) on the odd sphere, is set apart by its interaction with the antipodal action; the form
1−x
Zρ(n)
2 I2n−1
Zρ(n)∗
Pρ(n) =(cid:20) 1+x
2 I2n−1 (cid:21) ,
ρ ), it is normal and satisfies Zρ(n)Zρ(n)∗ = (P zjz ∗
2(cid:20) xI2n−1
Zρ(n)∗ −xI2n−1 (cid:21)
Pρ(n) =
Zρ(n)
I2n +
1
2
1
shows that α(Pρ(n)) = I2n − Pρ(n), where the antipodal action α is applied entrywise. Matrices of
particular dimensions satisfying this identity will be distinguished in K0. There is a marked advantage
of odd K-theory for these types of problems, as the group operation of K1 is realized as both direct
sum and matrix multiplication, which allows results like Lemma 2.3 to manipulate the fixed point
subalgebra instead of the (−1)-eigenspace. On the other hand, the group operation of K0 is only
realized as the direct sum, so we glean K0 information on C(S2n
).
ω )) is a projection, then B = 2P −I is a self-adjoint square root of I, or equivalently
B is self-adjoint and unitary. Through (2.1), viewing B as a function of x ∈ [−1, 1] with (self-adjoint
ρ ) from K1 information of C(S2n+1
If P ∈ Mk(C(S2n
ω
3
and unitary) values in Mk(C(S2n−1
introducing an additional coordinate.
ω
)) allows us to form a unitary function eB : D → Mk(C(S2n−1
ω
)) by
eB(x + iy) =p1 − y2B
x
p1 − y2! + iyIk
x
a unitary element of Mk(C(S2n+1
dimension k over C(S2n
) is continuous, meaning it respects
path components, and in addition it is compatible with the direct sum. Also, if P is a trivial projection
p1 − y2! is bounded,
The function eB appears to be only well-defined on D \ {±i}, but the term B
as B has norm 1, and the multiplication by p1 − y2 implies that eB extends continuously to D by
the squeeze theorem. Because B is self-adjoint and unitary, eB is unitary. Moreover, the boundary
conditions on P and B = 2P − I imply that if x2 + y2 = 1, then eB(x + iy) ∈ Mk(C), so eB represents
)) by (2.2). The association P 7→ B 7→ eB between projections of
Ik ⊕ 0l, then the function eB takes values in Uk(C). The domain D is contractible, so eB is connected
this contraction was possible because the range of eB only included scalar-valued matrices, which did
the identity, even though D is contractible.) In other words, if Ωk denotes the function P 7→ eB for
projections P of dimension k, then the various Ωk are compatible and produce a single homomorphism
not conflict with boundary conditions. In general an element in Uk+l(C(S2n+1
)) to a scalar-entried matrix, which is then connected to the identity. (Note that
ω ) to unitaries of dimension k over C(S2n+1
within Uk+l(C(S2n+1
)) is not connected to
ρ
ρ
ρ
ρ
Ω : K0(C(S2n
ω )) → K1(C(S2n+1
ρ
))
on the K-groups, and Ω has all trivial projections Ik ⊕0l in its kernel. Moreover, each Ωk is compatible
with the Z2 antipodal action in the following way.
ω )) is a projection satisfying
α2n(P ) = I − P , then B = 2P − I has α2n(B) = −B, or rather, B is odd. This is reflected in the
If P ∈ Mk(C(S2n
function algebra as α2n(B)[x] = α2n−1(B(−x)) = −B(x), which will help show that Ωk(P ) = eB is
also odd.
−x
α2n+1(eB)[x + iy] = α2n−1(eB(−x − iy))
= α2n−1 p1 − (−y)2B
=p1 − y2"α2n−1 B −x
=p1 − y2 −B
= − p1 − y2B
= −eB(x + iy)
p1 − (−y)2! + i(−y)Ik!
p1 − y2!!# − iyIk
p1 − y2!! − iyIk
p1 − y2! + iyIk!
x
x
Theorem 2.4. If P is a projection matrix over C(S2n
I −P , then the class of P in K0(C(S2n
ω ) of dimension (2m + 1)2n that satisfies α(P ) =
ρ )) ∼= Z⊕Z is not in the subgroup generated by trivial projections.
Proof. Suppose P is in the subgroup generated by trivial projections, and consider V := Ω(2m+1)2n (P )
in U(2m+1)2n (C(S2n+1
)), where ρ is an (n + 1) × (n + 1) parameter matrix with ω in the upper left
and 1 in all other entries. It follows that [V ]K1 = Ω([P ]K0) is the trivial element of K1(C(S2n+1
)).
However, V is odd and of dimension (2m+1)2n, so this contradicts Lemma 2.3 for spheres of dimension
2n + 1 = 2(n + 1) − 1.
ρ
ρ
4
This theorem shows that the Z2 structure of even spheres is reflected in their K-theory. Just as
in the odd case, we now prove a noncommutative Borsuk-Ulam theorem.
Corollary 2.5. Suppose Φ : C(S2n
ω ) is a unital ∗-homomorphism that is equivariant for
the antipodal maps. If K0 ∼= Z ⊕ Z is such that the first summand is generated by trivial projections,
then the K0 map induced by Φ is nontrivial on the second summand.
ρ ) → C(S2n
Proof. Consider the projection Pρ(n) =(cid:20) 1+x
2 I2n−1
Zρ(n)∗
Zρ(n)
2 I2n−1 (cid:21), which is of dimension 2n and sat-
1−x
isfies α(Pρ(n)) = I − Pρ(n). Because Φ is equivariant, the same applies to Φ(Pρ(n)), so Φ(Pρ(n)) is
not in the subgroup generated by trivial projections.
3 Anticommutation
The same argument to establish (2.1), also seen in [9], can show that R2n
unreduced suspension of C(S2n−1
).
ω
ω is a noncommutative
R2n
ω
∼= ΣαC(S2n−1
ω
) := {f ∈ C([0, 1], C(S2n−1
ω
) ⋊α Z2) : f (0) ∈ C(S2n−1
ω
), f (1) ∈ C ∗(Z2)}
(3.1)
In the above, α : C(S2n−1
α(zj) = −zj. The generators z1, . . . , zn, x in the presentation (1.3) of R2n
function forms.
) denotes the antipodal map, the unique homomorphism with
ω then take the following
) → C(S2n−1
ω
ω
x ∼ X(t) = tδ
The computation of even K-theory for R2n
ω will be aided by the following matrix expansion map.
zj ∼ Zj(t) =p1 − t2zj
Definition 3.2. The expansion map Eα : C(S2n−1
∗-homomorphism defined by the following rule.
ω
) ⋊α Z2 → M2(C(S2n−1
ω
)) is the injective, unital
f + gδ 7→(cid:20)
f
α(g) α(f ) (cid:21)
g
See [14], section 2.5 for a similar map. The map Eα is certainly not surjective, as all matrices in
its range satisfy a very visible symmetry condition.
Proposition 3.3. If M = (cid:20)
M =(cid:20) 0 1
1 0 (cid:21)∗
α(M )(cid:20) 0 1
1 0 (cid:21). The K1(C(S2n−1
ω
f
α(g) α(f ) (cid:21) ∈ Ran(Eα), then M satisfies the symmetry condition
g
)) class of M is then represented by an even integer.
Proof. The symmetry condition is apparent, and we may apply [8], Theorem 3.10.
ω
) ⋊α Z2 is isomorphic to the K-theory of the fixed point subalgebra C(S2n−1
The antipodal action α is saturated, so as a result of Morita equivalence, K-theory of the crossed
product C(S2n−1
)α,
the algebra of even elements. This is useful when considering the ideal J of R2n
ω consisting of
functions which vanish at the endpoints 0 and 1, so J ∼= S(C(S2n−1
) ⋊α Z2) and consequently
Ki(J) ∼= Ki(S(C(S2n−1
)α). The fixed point
ω
)α is a Rieffel deformation of C(RP2n−1) ∼= C(S2n−1)α and therefore has iden-
subalgebra C(S2n−1
tical K-theory (K-theory of real projective space is computed in [1], Proposition 2.7.7). Boundary
) ⋊α Z2))) ∼= K1−i(C(S2n−1
) ⋊α Z2) ∼= K1−i(C(S2n−1
ω
ω
ω
ω
ω
5
ω /J ∼= C(S2n−1
information is contained in the quotient R2n
sums, so the six term exact sequence yields the following.
ω
) ⊕ (C ⊕ Cδ), and K-theory respects direct
K0(S(C(S2n−1
ω
) ⋊α Z2))
η
K1(C(S2n−1
ω
)) ⊕ K1(C + Cδ)
γ
λ
K0(R2n
ω )
K1(R2n
ω )
υ
κ
K0(C(S2n−1
ω
)) ⊕ K0(C + Cδ)
σ
K1(S(C(S2n−1
ω
) ⋊α Z2))
(3.4)
We repeat this sequence below with the known isomorphism classes and use it to calculate K0(R2n
ω ).
η
Z
Z
γ
λ
K0(R2n
ω )
K1(R2n
ω )
υ
κ
Z ⊕ Z ⊕ Z
σ
Z ⊕ Z2n−1
(3.5)
Proposition 3.6. The group K0(R2n
are considered as paths from (3.1), the K0 data is obtained from the two ranks Rankδ=±1(P (1)).
ω ) is isomorphic to Z⊕ Z, and if matrix projections P ∈ Mk(R2n
ω )
Proof. The subalgebra C ∗(Z2) = C + Cδ ≤ C(S2n−1
) ⋊α Z2 is two dimensional and isomorphic to
a continuous function algebra on two points, C({±1}), so an evaluation of δ at 1 or −1 can be
used on this subalgebra (however, an evaluation does not make sense on the entire crossed product).
) ⊕ (C + Cδ)) ∼= Z ⊕ Z ⊕ Z is essentially rank data in three instances: a rank for
Moreover, K0(C(S2n−1
)) ∼= Z and two ranks for K0(C({±1})) ∼= Z⊕Z based on evaluations δ = ±1. However, the
K0(C(S2n−1
ω
ω
ω
ω
1 ± δ
2
ω
map υ in (3.5) is not surjective by the following argument. First, the projections
∈ C + Cδ that
generate its K0 group are sent under Eα to rank one projections(cid:20) 1/2 ±1/2
1/2 (cid:21) in M2(C(S2n−1
±1/2
))) ∼= K0(C(S2n−1
)) ∼= Z these are the generator 1. Next, if we consider C(S2n−1
in K0(M2(C(S2n−1
),
whose K0 group is again cyclic and generated by the trivial projection 1, the image Eα(1) = I2 has
doubled in rank, and the same will happen to any projection over C(S2n−1
) when Eα is applied. Now,
any projection matrix over R2n
) ⋊α Z2, so the K0 class
of the images remains constant along the path. Based on the previous computations, if P ∈ Mk(R2n
ω )
is viewed as a path into Mk(C(S2n−1
) ⋊α Z2), then there is the following restriction on the ranks of
P (t).
ω is a path of projection matrices over C(S2n−1
)), so
ω
ω
ω
ω
ω
2 · RankC(S2n−1
ω
)(P (0)) = RankC(S2n−1
= RankC(S2n−1
= Rankδ=1(P (1)) + Rankδ=−1(P (1))
)(Eα(P (0)))
)(Eα(P (1)))
ω
ω
From this equation, we reach constraints on the range of υ in (3.5).
(l, m, n) ∈ Ran(υ) =⇒ 2l = m + n
(3.7)
The first clear projection in Ran(υ) is the identity element, which produces the predictable ranks
(1, 1, 1). The second is based on adjusting coefficients from a projection over θ-deformed even spheres;
in R2n
ω , x anticommutes with each ∗-monomial entry of Zω(n).
P := P ′
ω(n) :=
1
2
I2n +
1
2(cid:20) xI2n−1 Zω(n)
Zω(n)∗ xI2n−1 (cid:21)
(3.8)
6
When viewed as a path, the projection P has P (0) =
ω
)), and P (1) = 1+δ
K0(C(S2n−1
2 I2n , which has rank 2n at δ = 1 and rank 0 at δ = −1. Therefore, P
produces the element (2n−1, 2n, 0) ∈ Ran(υ). The tuples (1, 1, 1) and (2n−1, 2n, 0) are independent,
so their span has free rank 2, but for large n the second tuple is not in reduced form, meaning we
cannot tell if 2l = m + n completely characterizes elements of Ran(υ) (that is, if Ran(υ) is a full or
proper subset of spanZ{(1, 1, 1), (1, 2, 0)}).
1
2(cid:20) I2n−1
Zω(n)∗
Zω(n)
I2n−1 (cid:21), which has rank 2n−1 in
spanZ{(1, 1, 1), 2n−1(1, 2, 0)} ≤ Ran(υ) ≤ spanZ{(1, 1, 1), (1, 2, 0)}
ω
ω
Next, note that K1(C(S2n−1
)), so an index 2 matrix in K1(Ran(Eα)) ∼= K1(C(S2n−1
) ⋊α Z2) ∼= K1(C(RP2n−1)) ∼= Z is generated by Zω(n), which is
again seen through Eα. The expansion Eα(Zω(n)) is of index 2 because it is unitarily equivalent
to Zω(n) ⊕ Zω(n), and by Proposition 3.3, any invertible matrix in Ran(Eα) will correspond to an
) ⋊α Z2) ∼= Z
even integer in K1(C(S2n−1
is a generator. Since Eα is an isomorphism between C(S2n−1
⋊α Z2) and Ran(Eα), this implies
that K1(C(S2n−1
) ⊕
(C + Cδ)) ∼= K1(C(S2n−1
) ⋊α Z2)) from the six term sequence mimics the form
of the isomorphism K1(C(S2n−1
⋊α Z2))) of Bott periodicity, but now
that we know K1(C(S2n−1
) alone, we can conclude
that η is surjective. Therefore γ is the zero map and υ is injective, so K0(R2n
ω ) is isomorphic to
Ran(υ) ∼= Z ⊕ Z.
) ⋊α Z2) is generated by Zω(n). The connecting map η between K1(C(S2n−1
) ⋊α Z2) is generated by a unitary over C(S2n−1
) ⋊α Z2) → K0(S(C(S2n−1
)) and K0(S(C(S2n−1
ω
ω
ω
ω
ω
ω
ω
ω
ω
ω
ω and C(S2n
Even though the K0 groups of R2n
different in that K0 data of R2n
whereas K0(C(S2n
another for a nontrivial vector bundle. Despite all of this, the form of the nontrivial projection P ′
in (3.8) is just a slight sign change from a projection Pω(n) over C(S2n
2.5. The distinguished projection
ω ) are abstractly isomorphic, the generators are
ω is contained in the possibly distinct ranks Rankδ=1 and Rankδ=−1,
ω )) inherits its K-theory from K0(C(S2n)), which has one summand for rank and
ω(n)
ω ) seen in the proof of Corollary
P ′
ω(n) :=
1
2
I2n +
1
2(cid:20) xI2n−1 Zω(n)
Zω(n)∗ xI2n−1 (cid:21)
(3.9)
is also compatible with the antipodal map, in perfect analogy with the projection Pω(n) on C(S2n
Note that the antipodal map on R2n
ω
) is given in path form as
∼= ΣαC(S2n−1
ω
ω ).
a(f )[t] = αbα(f (t)) =bαα(f (t)),
ω
where bα : a + bδ 7→ a − bδ is the dual action on C(S2n−1
) ⋊α Z2. The ranks Rankδ=±1 at t = 1
take different values for P ′
2 I2n . Based on this example, we can
ask if any 2n × 2n projection in R2n
ω satisfying a(P ) = I − P is nontrivial in K0. This result does
hold, in analogy with the θ-deformed even spheres, but the proof is necessarily quite different, as the
suspension argument of the previous section does not pass over cleanly to this case. In order to rectify
this problem, we manipulate a particular non-free action on commutative spheres.
ω(n), as evaluation at t = 1 yields 1+δ
Theorem 3.10. Let the commutative sphere C(Sk) be equipped with a Z2 action γ that fixes a
distinguished real-valued coordinate h but negates the remaining coordinates x1, . . . , xk, and let U be
a unitary matrix over C(Sk) with γ(U ) = U ∗ and U h=±1 = I. Then the following hold.
1. If U represents a trivial element in K1(C(Sk)), then there is a path connecting some stabilization
U ⊕ I to I within the set of unitaries that satisfy γ(M ) = M ∗ and M h=±1 = I.
2. If k is odd, then U corresponds to an even integer in K1(C(Sk)) ∼= Z.
7
Proof. We proceed by induction; the only applicable unitaries over C(S0) are the identity matrices,
so the claim at dimension 0 is automatically satisfied. If the claim holds for k, then suppose U is a
unitary matrix over C(Sk+1) with γ(U ) = U ∗ and U h=±1 = I. Let Sk denote the equator in Sk+1
determined by xk = 0 (noting xk is a coordinate negated by γ), so that the restrictioneγ of γ to C(Sk)
is an action of the same type:
it fixes h and negates x1, . . . , xk−1. Also note that the fixed points
h = ±1 are in this equator Sk. To use the inductive hypothesis, first form a path of unitaries that
"stretches" the equator data of U . Realize C(Sk+1) as the unreduced suspension ΣC(Sk), with xk
representing the path coordinate in [−1, 1]. Then U = U (xk) represents a path of unitary matrices
over C(Sk), and we form a continuous path Ut as follows.
Ut(xk) =
t−2 (xk + 1) − 1(cid:17) if −1 ≤ xk ≤ −t
U(cid:16) −2
U(cid:16) −2
t−2 (xk − 1) + 1(cid:17) if
2 ≤ xk ≤ t
2
t
2 ≤ xk ≤ 1
U (0)
−t
if
2
This path connects U = U0 to V = U1, which has its equator data repeated on a band neighborhood
2 ≤ xk ≤ 1
− 1
t and still assigns the identity on h = ±1, as these
points are in the equator xk = 0. The equator function U (0) = V (0) is trivial in K1(C(Sk)) because
2 . The path Ut also satisfies γ(Ut) = U ∗
Sk sits inside a contractible subset of Sk+1; it also satisfies eγ(U (0)) = U (0)∗ and sends the fixed
W0 = U (0) ⊕ I to W1 = I within the unitaries over C(Sk) that also satisfyeγ(Wt) = W ∗
points at h = ±1 to the identity matrix. By the inductive hypothesis, there is a path Wt connecting
t and assign the
fixed points h = ±1 to the identity matrix. Apply this path to the equator of V ⊕ I while maintaining
the same values on the path for xk ≥ 1
2 .
V (xk) ⊕ I
Vt(xk) =(cid:26) Wφt(xk)
φt(xk) =(cid:26) −2xk + t
0
if
if
if
if
xk ≤ 1
2
1
2 ≤ xk ≤ 1
xk ≤ t
2
t
2 ≤ xk ≤ 1
2
The path Vt connects V0 = V ⊕ I to V1, where the unitaries Vt still satisfy γ(Vt) = V ∗
t , and by
the inductive assumption, the fixed points h = ±1 of γ are still always assigned the identity. Because
V1 also assigns the identity matrix on the entire equator, V1 is the commuting product of two unitary
matrices F and G, where F assigns the identity matrix for xk ≥ 0, G assigns the identity matrix for
xk ≤ 0, and γ(F ∗) = G. How to proceeed is slightly different based on the parity of k + 1.
If k+1 is even, then K1(C(Sk+1)) is the trivial group, so let Ft denote a path of unitaries connecting
F0 = F ⊕ I to I. Moreover, we can insist that Ft always assigns the identity matrix on points with
xk ≥ 0, as this region is contractible to a point and K1(C(Sk+1 \ {pt})) is also trivial. Then the
commuting product Qt = Ft · γ(F ∗
t ) forms a path of unitaries that connects V1 ⊕ I to I, satisfies
γ(Qt) = Q∗
t , and assigns the identity matrix at (at least) h = ±1. Tracing back all of the paths used
so far shows that U ⊕ I is connected to I within the same set of restricted unitaries.
If k + 1 is odd, then K1(C(Sk+1)) ∼= Z, and as γ and the adjoint are both orientation-reversing, F
and G = γ(F ∗) represent the same element in K1(C(Sk+1)). The product V1 = F · γ(F ∗) shows that
the class of V1 (and therefore of U ) in K1 is 2[F ]K1, an even integer. If this integer is zero, then once
again, K1(C(Sk+1 \ {pt})) is isomorphic to K1(C(Sk+1)), so just as in the previous paragraph, the
trivial element F ⊕ I can be connected to I in a path of unitaries Ft that always assign the identity
on points with xk ≥ 0. Finally, the commuting product Qt = Ft · γ(F ∗
t ) connects V1 ⊕ I to I, satisfies
γ(Qt) = Q∗
t , and always send the points h = ±1 to the identity matrix. A composition of paths
establishes the same for U and completes the induction.
In the above theorem, the set of matrices M satisfying γ(M ) = M ∗ is not a C ∗-subalgebra of the
matrix algebra over C(Sk), as the adjoint operation reverses the order of multiplication. As such, some
of the ideas in the proof are motivated by the six term exact sequence, but cannot be implemented
8
this way. Mention of function values at the fixed points h = ±1 appears to be unnecessary on first
glance, but this is crucial even at low dimensions.
Example 3.11. Let C(S1) be generated by the complex coordinate z = x + iy with a Z2 action γ
that negates y but fixes x. Then γ(z) = z ∗, but z is the generator of K1(C(S1)), associated to the
odd integer 1. The fixed points are x = 1 and x = −1, and z assigns these points to ±1. The previous
theorem does not apply, as it only considers matrices which assign the identity matrix on the fixed
points of γ. The same scenario happens in any C(S2n−1) for the standard K1 generator Z(n).
Theorem 3.10 concerns an action γ that is not free, but it will be instrumental in defining an
invariant for unitaries that satisfy α(U ) = U ∗, where α is the antipodal map on C(S2n−1). In turn,
this invariant will show that the relation algebras R2n
ω have a Borsuk-Ulam property for projections
in K0, by first proving the case when z1, . . . , zn all commute with each other.
Definition 3.12. Fix an isomorphism C(Sk) ∼= ΣC(Sk−1), with the path coordinate denoted by
If M ∈ Mp(C(Sk)) is such that M (0) ∈ Mp(C) (i.e., it has constant entries), then let M + ∈
h.
Mp(C(Sk)) ∼= Mp(ΣC(Sk−1)) be defined as follows for h ∈ [−1, 1].
M +(h) = M(cid:18) h + 1
2 (cid:19)
The matrix M + encodes information about M only for points h ≥ 0. Note in particular that M is
also only defined in reference to a particular, fixed coordinate h and a suppressed isomorphism with
an unreduced suspension. We do not expect that any M + that is definable with respect to multiple
coordinate choices has any uniqueness properties of any kind.
Theorem 3.13. Suppose U0, U1 ∈ Uk(C(S2n−1)) are unitary matrices which assign an equator h = 0
to the identity matrix and satisfy α(Uj) = U ∗
j . If there is a path of unitary matrices Ut connecting
t (but have no restriction on the equator), then the classes of U +
U0 and U1 that satisfy α(Ut) = U ∗
0
and U +
1 in K1(C(S2n−1)) ∼= Z differ by an even integer.
Proof. Fix an isomorphism C(S2n−1) ∼= ΣC(S2n−2), with the path coordinate specified by h, and let γ
be a Z2 action on C(S2n−1) that fixes h but applies the antipodal map pointwise on C(S2n−2). Define
Vt ∈ Uk(C(S2n−1)) as follows.
Vt(h) =(cid:26)
U−h(0)
1+t + 1)
Ut( h−1
if −1 ≤ h ≤ −t
if −t ≤ h ≤ 1
Note that Vt(−1) = U1(0) is always the identity matrix and Vt(1) = Ut(1) is a constant matrix,
meeting the necessary boundary conditions to define a unitary over the sphere. Further, V1 is equal
to U +
1 , and V0 contains data of U +
0 with equator information of Ut.
V0(h) =(cid:26) U−h(0)
U0(h)
if −1 ≤ h ≤ 0
if
0 ≤ h ≤ 1
Since U0(0) is the identity matrix, this formula shows that V0 can be written as a commuting product
of two unitaries F · G, as follows.
F (h) =(cid:26) I
U0(h)
if −1 ≤ h ≤ 0
if
0 ≤ h ≤ 1
=⇒
[F ]K1 = [U +
0 ]K1
Now, the path matrices Ut satisfy α(Ut) = U ∗
t , and the matrix G contains equator data from Ut,
so it satisfies γ(G) = G∗ where γ negates every coordinate in C(S2n−1) except h. That is, γ applies
G(h) =(cid:26) U−h(0)
I
if −1 ≤ h ≤ 0
if
0 ≤ h ≤ 1
9
the antipodal map of C(S2n−2) pointwise on ΣC(S2n−2). G also assigns the fixed points of γ, h = ±1,
to the identity matrix, so the class of G in K1(C(S2n−1)) is an even integer by Theorem 3.10. The
formulas V0 = F · G and [F ]K1 = [U +
0 ]K1 , and
[V0]K1 is the same as [V1]K1 = [U +
1 ]K1.
0 ]K1 imply that [V0]K1 is an even integer away from [U +
The above theorem defines a Z2 invariant for any unitary over C(S2n−1) satisfying α(U ) = U ∗
which assigns the identity on a specified equator h = 0: [U +]K1 mod 2. This invariant is preserved by
paths Ut satisfying α(Ut) = U ∗
t where U0 and U1 assign the identity on the equator. It is important
to note that only the endpoints of the path have specified equator data; if the equators of Ut were
also the identity, U +
0 and U +
1 would be equal,
with no need for reducing mod 2. Some examples of unitary paths with α(Ut) = U ∗
t will be necessary
for later calculations.
Example 3.14. Fix an isomorphism C(S2n−1) ∼= ΣC(S2n−2) with path coordinate x1 = Re(z1), and
consider the standard K1 generator Z(n) ∈ U2n−1(C(S2n−1)). Define a unitary V ∈ U2n−1(C(S2n−1))
as follows.
t would be a well-defined path and the K1 classes of U +
V (x1) = −Z(n)[2x1 − 1]
Now, V has V (±1) = −Z(n)[1] = −I and V (0) = −Z(n)[−1] = I, and the antipodal map α on
C(S2n−1) comes from applying x1 7→ −x1 and the pointwise antipodal map eα on C(S2n−2).
The application ofeα to −Z(n)[2x1−1] negates y1 = Im(z1) and z2, . . . , zn, which produces (−Z(n)[2x1−
1])∗ by an inductive argument, soeα(V (x1)) = V (x1)∗ for each x1. Finally, the above equation implies
that α(V ) = V ∗. Since V (0) = I and V + = −Z(n), it follows that [V +]K1 = 1. Moreover, V is
connected to −I2n−1 within the unitaries satisfying α(M ) = M ∗ by considering the following path Vt.
α(V )[x1] =eα(V (−x1)) =eα(V (x1)) =eα(−Z(n)[2x1 − 1])
Now, V0(x1) = V (x1) = V (x1), so V0 = V , and V1(x1) = V (1) = −I2n−1.
Vt(x1) = V ((1 − t)x1 + t)
α(Vt)[x1] =eα(Vt(−x1))
=eα(Vt(x1))
=eα(V ((1 − t)x1 + t))
= V ((1 − t)x1 + t)∗
= Vt(x1)∗
Finally, V is connected to −I2n−1 within the unitary matrices satisfying α(M ) = M ∗.
Example 3.15. Fix an isomorphism C(S2n−1) ∼= ΣC(S2n−2) with path coordinate x1. Suppose
U0 ∈ Uk(C(S2n−1)) is anti self-adjoint (U ∗
0 = −U0) and odd (α(U0) = −U0), which implies that
α(U0) = U ∗
0 . Define a path Ut as follows.
2−t (−x1 − 1) + 1)∗)
eα(U0( 2
(t + 2x1)I +p1 − (t + 2x1)2eα(U0(0)∗)
(t − 2x1)I +p1 − (t − 2x1)2 U0(0)
2−t (x1 − 1) + 1)
U0( 2
if −1 ≤ x1 ≤ − t
2
− t
if
2 ≤ x1 ≤ 0
0 ≤ x1 ≤ t
if
2
t
if
2 ≤ x1 ≤ 1
Note that U0(0) is anti self-adjoint and odd under the antipodal action eα on C(S2n−2), so Ut is
well-defined and unitary with α(Ut) = U ∗
which have α(W ∗
t . Let U1 = W0 and define another path of unitaries Wt
t ) = Wt and also assign the equator to I.
Ut(x1) =
Wt(x1) =
(1 + 2
eα(U0(−2x1 − 1)∗)
1+t x1)I +q1 − (1 + 2
1+t x1)I +q1 − (1 − 2
U0(2x1 − 1)
(1 − 2
1+t x1)2eα(U0(t)∗)
1+t x1)2 U0(t)
if −1 ≤ x1 ≤ − 1+t
2
if − 1+t
2 ≤ x1 ≤ 0
0 ≤ x1 ≤ 1+t
2
1+t
2 ≤ x1 ≤ 1
if
if
10
Note that the formula defining Wt produces a unitary matrix because each U0(t) is anti self-adjoint,
1 ]K0 = 0.
entries for each x1, which implies that [W +
Consider the C ∗-algebra R2n ∼= ΣαC(S2n−1), in which z1, . . . , zn commute with each other. Be-
as is each eα(U0(t)∗). Finally, W1(0) = I, and because U0(1) has scalar entries, W1(x1) has scalar
cause the antipodal map a is implemented as αbα pointwise on C(S2n−1) ⋊α Z2, an element or matrix
F (t) + G(t)δ over R2n
is odd if and only if each F (t) is odd in C(S2n−1) and each G(t) is even in
ρ
C(S2n−1), which implies that G(t) commutes with δ and F (t) anticommutes with δ. An odd matrix
F (t) + G(t)δ is then self-adjoint if and only if F (t) and G(t) are self-adjoint. If F (t) + G(t)δ is also
unitary, then it satisfies (F (t) + G(t)δ)2 = 1, which places tight restrictions on F and G.
(F (t) + G(t)δ)2 = F (t)2 + (G(t)δ)2 + F (t)G(t)δ + G(t)δF (t)
= (F (t)2 + G(t)2) + (F (t)G(t) − G(t)F (t))δ
= 1 + 0δ
(3.16)
This implies that the self-adjoint elements F (t) and G(t) must commute and satisfy F (t)2 + G(t)2 = 1.
In other words, U (t) = G(t) + iF (t) is a unitary matrix over C(S2n−1) for each t, and because G(t)
is even and F (t) is odd, α(U (t)) = U (t)∗.
Theorem 3.17. Let α denote the antipodal action on the commutative sphere C(S2n−1) and consider
If P ∈ M(2m+1)2n (R2n) is a
the path algebra R2n = ΣαC(S2n−1) with antipodal action a = αbα.
projection with a(P ) = I − P , then [P ]K0 is not in the subgroup generated by trivial projections.
2 I + 1
Proof. Write P = 1
2 B where B is an odd, self-adjoint unitary matrix. By the calculation (3.16),
B(t) = F (t) + G(t)δ produces a path of unitaries U (t) = G(t) + iF (t) ∈ U(2m+1)2n(C(S2n−1)) with
α(U (t)) = U (t)∗. Since B(1) is a matrix over C + Cδ, but F (1) must be odd, it follows that F (1) = 0
and B(1) = G(1)δ where G(1) is a self-adjoint matrix with constant entries. Moreover, the boundary
conditions of ΣαC(S2n−1) imply that G(0) = 0, so U (0) = iF (0) is an anti self-adjoint odd unitary. If
[P ]K0 is in the subgroup generated by trivial projections, then the ranks Rankδ=±1(P (1)) are equal,
so U (1) = G(1) is a self-adjoint, unitary matrix over C whose eigenspaces for eigenvalues ±1 are of
equal dimension. It follows that U (1) may be connected within the self-adjoint, unitary matrices over
C to I(2m+1)2n−1 ⊕ −I(2m+1)2n−1, and the matrices in this path satisfy α(M ) = M = M ∗.
Because U (0) is an anti self-adjoint, odd, unitary matrix, by Example 3.15, U (0) is connected
via a path of unitaries satisfying α(M ) = M ∗ to a matrix W with identity on the equator and
∼= 0 mod 2. Similarly, since U (1) is connected via a path of unitaries satisfying α(M ) = M ∗
[W +]K1
to I(2m+1)2n−1 ⊕ −I(2m+1)2n−1, repeated use of Example 3.14 for (2m + 1) summands of −I2n−1
shows U (1) is also connected to a matrix V such that V assigns the identity on the equator and
∼= (2m + 1) ∼= 1 mod 2. This contradicts Theorem 3.13, as V and W assign the identity on the
[V +]K1
equator and are connected via a path of unitaries satisfying α(M ) = M ∗ (with no assumption on the
path's equator data), even though their invariants [V +]K1 mod 2 and [W +]K1 mod 2 are different.
All that remains is to remove the assumption that z1, . . . , zn commute with each other. First,
)) shows that the relation algebra
) ⋊α Z2 → M2(C(S2n−1
ω
note that the expansion map Eα : C(S2n−1
) embeds into M2(C(S2n
R2n
ω = ΣαC(S2n−1
ω
ω
ω )), as follows.
ΣαC(S2n−1
ω
) = {f ∈ C([0, 1], C(S2n−1
ω
) ⋊α Z2) : f (0) ∈ C(S2n−1
ω
), f (1) ∈ C + Cδ}
))) : f (t) =(cid:20)
), f (1) ∈ C + Cδ}
ω
g(t)
α(h(t)) α(g(t)) (cid:21) for all t,
h(t)
∼= {f ∈ C([−1, 1], C(S2n−1
ω
) ⋊α Z2) : f (−1) ∈ C(S2n−1
∼= {f ∈ C([−1, 1], M2(C(S2n−1
ω
h(−1) = 0, and g(1), h(1) ∈ C}
≤ M2(ΣC(S2n−1
ω
)) = M2(C(S2n
ω ))
11
g(t)
Denote this subalgebra of M2(C(S2n
realized as(cid:20)
conjugation by(cid:20) 1
α(h(t)) α(g(t)) (cid:21) 7→(cid:20) α(g(t)) −α(h(t))
0 −1 (cid:21). The spheres C(S2n
−h(t)
ω )) isomorphic to R2n
h(t)
g(t)
0
ω ) are θ-deformations of C(S2n), so if f and g are fixed
(matrices of) smooth elements, the following continuity properties apply, as weak forms of strong
quantization ([11]). Here Cω denotes the collection of parameter matrices ρ which differ from ω only
in one prescribed pair of conjugate entries, (i, j) and (j, i).
ω as Bω. The antipodal action a on Bω is
(cid:21), which is entrywise application of α and
f ·ω g − f ·ρ gρ → 0 as ρ → ω within Cω
f ρ → f ω as ρ → ω within Cω
Any smooth approximations in M2(C(S2n
ω )) to an element of Bω can be made without leaving
Bω. Note that the pointwise action α on ΣC(S2n−1
ω ) which fixes
x but negates every generator zi, so γ commutes with the rotation action of Rn defining the Rieffel
deformation. Moreover, because the antipodal map on Bω is implemented with an entrywise action
ω ) is an action γ on C(S2n
) = C(S2n
ω
and a conjugation, a matrix function(cid:20)
g(t)
α(h(t)) α(g(t)) (cid:21) =(cid:20)
h(t)
g
γ(h) γ(g) (cid:21) in Bω is odd if and only
h
if γ(g) = −g and γ(h) = h. Together, these observations imply that smooth approximations to Bω
elements may be made in Bω while preserving homogeneity properties in the antipodal action. Also,
as the adjoint operations of Rieffel deformations all have the same effect on smooth elements, smooth
approximations can also be made to preserve self-adjointness.
Theorem 3.18. Suppose P ∈ M(2m+1)2n (R2n
in the subgroup generated by the trivial projections.
ω ) is a projection with a(P ) = I − P . Then [P ]K0 is not
Proof. Suppose the theorem fails for P ∈ M(2m+1)2n (R2n
ω ), so there is a path of projections connecting
∼= Bω ≤
P ⊕ 0 ⊕ I to a trivial projection I ⊕ 0. If this path is viewed under the isomorphism R2n
ω
M2(C(S2n
ω )), then any matrix over Bω can be approximated by matrices with smooth entries, and this
approximation can be done without leaving Bω. Moreover, if the original matrix is odd or self-adjoint,
these properties can be preserved in the approximation. Any smooth matrix over Bω ≤ M2(C(S2n−1
))
may then be viewed as a matrix over Bρ ≤ M2(C(S2n−1
)) for any ρ.
ω
ρ
The path of projections connecting P ⊕ 0 ⊕ I to I ⊕ 0 is uniformly continuous, so let P0, . . . , Pk
be finitely many elements of this path with the following properties.
P0 = P ⊕ 0 ⊕ I
Pj − Pj−1ω < ε
Pk = I ⊕ 0
Fix 0 < ε < 1
13 , choose a self-adjoint smooth approximation Q to P such that a(Q) = I − Q and
Q ·ω Q − Qω < ε, and let Q0 = Q ⊕ 0 ⊕ I. It follows that Q0 ·ω Q0 − Q0ω < ε. The matrix
Pk = I ⊕ 0 is already smooth, so let Qk = Pk. Finally, choose smooth approximations Q1, . . . , Qk−1
of P1, . . . , Pk−1 so that these restrictions hold for each j.
Q0 = Q ⊕ 0 ⊕ I
Qj = Q∗
j
Qj ·ω Qj − Qjω < ε
Qj − Qj−1ω < ε
Qk = I ⊕ 0
If ρ is another parameter matrix differing from ω only in a prescribed pair of conjugate entries,
then if ρ is close enough to ω, similar inequalities hold.
Q0 = Q ⊕ 0 ⊕ I
Qj = Q∗
j
Qj ·ρ Qj − Qjρ < ε
Qj − Qj−1ρ < ε
Qk = I ⊕ 0
Let Rt, t ∈ [0, 1], denote the piecewise linear path connecting Q0, . . . , Qk, so each Rt is self-adjoint
and has Rtρ ≤ max {Qjρ : j ∈ {0, . . . , k}} ≤ 1 + ε. Further, for any t there is a j such
that Rt − Qjρ < ε, so liberal use of the triangle inequality and properties of Qj shows that
Rt ·ρ Rt − Rtρ < 3ε + 2ε2.
R0 = Q ⊕ 0 ⊕ I
Rt = R∗
t
Rt ·ρ Rt − Rtρ < 3ε + 2ε2
R1 = I ⊕ 0
12
2 I + 1
Write Rt = 1
2 Ft, so Ft is self-adjoint and Ft ·ρ Ft − 1ρ < 12ε + 8ε2 < 1. This implies
that Ft is invertible, so normalize Ft to a self-adjoint unitary (under ·ρ) to produce Ft ·ρ Ft−1ρ
2 Ft ·ρ Ft−1ρ. Note that this normalization process does nothing to
2 I + 1
(2m + 1)2n-dimensional matrix algebra by similar means. Moreover, since Q = 1
2 F satisfies
a(Q) = I − Q, F is a self-adjoint odd invertible, so F ·ρ F −1ρ is a self-adjoint odd unitary, and
and the projection fRt = 1
R1 and respects the summands of R0: fR0 = eQ ⊕ 0 ⊕ I where eQ is a projection obtained Q in the
a(eQ) = I −eQ. Finally, the pathfRt is a path of projections in Bρ ∼= R2n
ρ connecting eQ ⊕ 0 ⊕ I to I ⊕ 0,
and eQ ∈ M(2m+1)2n (Bρ) = M(2m+1)2n(R2n
ρ ) is in the K0 subgroup generated by trivial projections.
This means that assuming the theorem fails for a single parameter matrix ω implies that the theorem
fails for all ρ that are sufficiently close to ω and differ from ρ in a prescribed pair of conjugate entries.
We may select ρ such that this pair of conjugate entries consists of roots of unity of odd order, and
then repeat the argument starting with ρ and another pair of conjugate entries.
In finitely many
iterations, it will follow that the theorem fails for a parameter matrix µ which has odd order roots of
unity in every entry. This is a contradiction by the following argument.
2 I + 1
Suppose µ is a parameter matrix with an odd order root of unity in each entry such that the
µ ), and consider R2n = ΣαC(S2n−1). There are
µ 7→ M2q+1(R2n)
theorem fails for a projection P ∈ M(2m+1)2n(R2n
unitary matrices V1, . . . , Vn ∈ U2q+1(C) such that VkVj = µjkVjVk, so define B : R2n
as follows.
B : zj 7→ zjVj
B : x 7→ xI2q+1
The ∗-homomorphism B exists because the desired images satisfy the relations defining R2n
µ , as
all noncommutativity information among z1, . . . , zn is pushed to the matrices Vj. Further, B is
equivariant for the antipodal map, as the odd generators are sent to matrices with odd entries. So,
if P ∈ M(2m+1)2n (R2n
µ ) satisfies a(P ) = I − P and [P ]K0 is in the subgroup generated by trivial
projections, then the same properties apply to B(P ) ∈ M(2q+1)(2m+1)2n (R2n), contradicting Theorem
3.17.
Corollary 3.19. Suppose Φ : R2n
is a unital ∗-homomorphism that is equivariant for the
antipodal map. If K0 ∼= Z ⊕ Z is such that the first summand is generated by trivial projections, then
the K0 map induced by Φ is nontrivial on the second summand.
ω → R2n
ρ
Proof. The image Φ(P ′
I − Φ(P ′
ω(n)). Its K0 class is therefore not in the subgroup generated by trivial projections.
ω(n)), where P ′
ω(n) is as in (3.9), is a 2n×2n projection that satisfies α(Φ(P ′
ω(n))) =
Remark. Because the θ-deformed spheres have the same nontriviality statement in K0, the domain or
codomain could be replaced by a θ-deformed sphere of dimension 2n.
A curious aspect of these proofs is that they discuss unitaries satisfying α(M ) = M ∗ over C(S2n−1
),
a sphere one dimension lower than the original, whereas the argument for C(S2n
ω ) worked by pushing
up one dimension and focusing on odd unitaries.
In both cases, the alternative method of proof
appears to have "insurmountable" road blocks, where one switch of sign nullifies an entire argument.
Nevertheless, we have seen that a noncommutative Borsuk-Ulam theorem does hold in both cases.
ω
4 Acknowledgments
This work was partially supported by NSF grants DMS 1300280 and DMS 1363250. Moreover, I am
grateful to my advisors John McCarthy and Xiang Tang at Washington University in St. Louis for
their continued support.
References
[1] M. F. Atiyah. K-Theory. Lecture notes by D. W. Anderson. W. A. Benjamin, Inc., New York-
Amsterdam, 1967.
13
[2] Paul F. Baum, Ludwik D
conjectures. arxiv:1502.05756.
abrowski, and Piotr M. Hajac. Noncommutative Borsuk-Ulam-type
'
[3] Alexandru Chirvasitu and Benjamin Passer. Compact Group Actions on Topological and Non-
commutative Joins. arxiv:1604.02173.
[4] Alain Connes and Giovanni Landi. Noncommutative Manifolds, the Instanton Algebra and
Isospectral Deformations. Comm. Math. Phys., 221(1):141–159, 2001.
[5] Ludwik D
[6] Ludwik D
abrowski. Towards a noncommutative Brouwer fixed-point theorem. arxiv:1504.03588.
'
abrowski and Giovanni Landi. Instanton algebras and quantum 4-spheres. Differential
'
Geom. Appl., 16(3):277–284, 2002.
[7] T. Natsume and C. L. Olsen. Toeplitz Operators on Noncommutative Spheres and an Index
Theorem. Indiana Univ. Math. J., 46(4):1055–1112, 1997.
[8] Benjamin Passer. A Noncommutative Borsuk-Ulam Theorem for Natsume-Olsen Spheres.
arxiv:1503.01822. To appear in Journal of Operator Theory.
[9] Benjamin Passer. Saturated Actions on C ∗-algebra Suspensions and Joins by Finite Cyclic
Groups. arxiv:1510.04100.
[10] Mira A. Peterka. Finitely-Generated Projective Modules over the θ-Deformed 4-Sphere. Comm.
Math. Phys., 321(3):577–603, 2013.
[11] Marc A. Rieffel. Deformation Quantization for Actions of Rd. Mem. Amer. Math. Soc.,
106(506):x+93, 1993.
[12] Marc A. Rieffel. K-groups of C ∗-Algebras Deformed by Actions of Rd. J. Funct. Anal.,
116(1):199–214, 1993.
[13] Ali Taghavi. A Banach Algebraic Approach to the Borsuk-Ulam Theorem. Abstr. Appl. Anal.,
pages Art. ID 729745, 11, 2012.
[14] Dana P. Williams. Crossed Products of C ∗-Algebras, volume 134 of Mathematical Surveys and
Monographs. American Mathematical Society, Providence, RI, 2007.
[15] Makoto Yamashita. Equivariant Comparison of Quantum Homogeneous Spaces. Comm. Math.
Phys., 317(3):593–614, 2013.
14
|
1809.02461 | 1 | 1809 | 2018-09-07T13:26:59 | Quasi-invariant measures for generalized approximately proper equivalence relations | [
"math.OA",
"math-ph",
"math.DS",
"math-ph"
] | We introduce a generalization of the notion of approximately proper equivalence relations studied by Renault and with it we build an \'etale groupoid. Choosing a suitable set of continuous functions to play the role of a potential, we construct a cocycle in that groupoid and discuss the corresponding Radon-Nikodym problem. | math.OA | math |
QUASI-INVARIANT MEASURES FOR
GENERALIZED APPROXIMATELY PROPER
EQUIVALENCE RELATIONS
R. Bissacot∗, R. Exel∗∗, R. Frausino∗ and T. Raszeja∗
We introduce a generalization of the notion of approximately proper equivalence rela-
tions studied by Renault and with it we build an ´etale groupoid. Choosing a suitable
set of continuous functions to play the role of a potential, we construct a cocycle in that
groupoid and discuss the corresponding Radon-Nikodym problem.
1. Introduction.
In [7], Renault introduced the notion of an approximately proper equivalence relation on a
compact topological space X, consisting of an increasing sequence {Rn}n∈ of equivalence
relations on X, each of which is proper in the sense that the corresponding quotient map
is a local homeomorphism. When equipped with the inductive limit topology, the union
R = Sn Rn becomes an ´etale groupoid, and if one is moreover given a suitable sequence
of continuous real valued functions on X, a cocycle1 may be defined on R.
As its title suggest, the main goal of [7] is to study the corresponding Radon-Nikodym
problem, i.e., to find the probability measures on X which are quasi-invariant with Radon-
Nikodym derivative equal to the aforementioned cocycle.
Among other things, the relevance os solving the Radon-Nikodym problem lies in the
fact that the solutions lead to KMS states on the groupoid C*-algebra and hence have a
profound relevance to Statistical Mechanics.
As mentioned in [7: Section 7], approximately proper equivalence relations arise nat-
urally in the study of local homeomorphisms from a compact topological space to itself.
Precisely speaking, given a compact topological space X, and a local homeomorphism
σ : X → X, one lets,
Rn = (cid:8)(x, y) ∈ X × X : σn(x) = σn(y)(cid:9),
for each n ≥ 0, and it is not hard to see that each Rn is a proper equivalence relation so
that {Rn}n∈ is an approximately proper equivalence relation in the sense of [7].
Proeminent examples of local homeomorphisms on compact topological spaces are
given by Markov shifts. On the other hand, in a recent paper [1], we have focused on a
generalization of Markov shifts introduced by M. Laca and the second named author in
[3], which in turn have been shown by Renault [6] to consist essentially of a generalized
shift space, with the notable difference that the shift map is no longer defined on the whole
space, but only on a proper open subset.
∗ Institute of Mathematics and Statistics (IME-USP), University of Sao Paulo, Brazil.
∗∗ Universidade Federal de Santa Catarina and University of Nebraska. Partially supported by CNPq.
1 In this paper the term cocycle will always be taken to mean a one-cocycle.
1
The precise setup of [6] is that of a locally compact space X, an open subset U ⊆ X,
and a local homeomorphism
σ : U → X.
However, if one starts from this data, it is not possible to build an approximately proper
equivalence relation by the procedure indicated above, not least because σn fails to be
defined on the whole space X. If one wants to make sense of the relation
x ∼ y ⇐⇒ σn(x) = σn(y),
one must restrict attention to elements x and y for which σn(x) and σn(y) make sense,
namely elements of the domain of σn, which we shal henceforth denote by Un. We thus
define
which is clearly a proper equivalence relation on Un. If n ≤ m, observe that
Rn = (cid:8)(x, y) ∈ Un × Un : σn(x) = σn(y)(cid:9),
σn(x) = σn(y) ⇒ σm(x) = σm(y),
as long as all of the above terms are defined, i.e, as long as x and y lie in the smaller set
Um. This may be more sucintly expressed by saying that
Rn ∩ (Um × Um) ⊆ Rm.
(1.1)
If one misreads the above inclusion, ignoring the intersection with Um × Um, one will
be left with the impression that the Rn are increasing, just as in [7], although this is
evidently not true given that the sets where these relations are defined in fact decrease.
Another distinctive feature of the Rn is the fact that, still under the hypothesis that
n ≤ m, one has that Um is invariant under Rn, meaning that
which may be expressed by saying that
(cid:0)(x, y) ∈ Rn(cid:1) ∧ (y ∈ Um) ⇒ x ∈ Um,
Rn ∩ (Un × Um) ⊆ Um × Um.
(1.2)
The reader may easily verify that, together, (1.1) and (1.2) are equivalent to
Rn ∩ (Un × Um) ⊆ Rm,
which might not have an immediately intuitive interpretation, but due to its sheer simplic-
ity, is adopted in this work as the main axiom in our generalization of Renault's notion of
approximately proper equivalence relations, given in full detail in (5.1), below, and referred
to as a gap, for short.
The main aim of the present work is to conduct a study of gaps along the lines of
Renault's study of approximately proper equivalence relations. We thus show that the
union R = Sn Rn is an equivalence relation, hence a principal groupoid, which becomes
2
ρ(x) :=
ρ′(x)
Py∼x ρ′(y)
´etale when given the inductive limit topology. A suitable notion of potential is introduced,
leading up to a cocycle relative to which the Radon-Nikodym problem may be investigated.
Since each Rn is assumed to be proper, one has that R is the (not necessarily increas-
ing) union of proper equivalence relations, so it is not surprising that the study of proper
relations is as important here as it is in [7]. Should our Un be compact we would be able to
borrow the results of the first few sections of [7], but the example of infinite state Markov
shifts, our main motivation, requires an understanding of the Radon-Nikodym problem for
proper relations on non-compact spaces. The lack of compactness indeed brings several
complications, most of them steming from the fact that equivalence classes no longer need
to be finite. For example, the normalization achieved in [7: Proposition 3.1.iii] by means
of replacing a potential ρ′ by
needs to be dealt with in a more careful way if equivalence classes are allowed to be infinite.
Once the proper case is taken care of, we apply our results for gaps, showing, among
other things, that quasi-invariant measures may be characterized, much in the same way
as DLR measures, as those which are fixed by a family of conditional expectations. See
section (6), and in particular Corollary (6.8), for full details.
The existence part of the Radon-Nikodym problem, which follows easily from compact-
ness when that property is present, e.g. as in [4: 8.2], turns out to be a delicate question
here.
In fact existence may already fail in the proper case, but it is nevertheless easy
to determine precisely when this happens. The crucial point is to analyze the partition
function
ζ(x) = Xy∼x
ρ(y),
defined in (4.2), which may well return infinite values, should equivalence classes be infinite.
The set of points x for which ζ(x) = ∞, which we denote by Zρ, is a forbidden zone for
finite quasi-invariant measures in the sense that any such measure assigns zero mass to Zρ.
Thus, if ζ is identically infinite, a situation very easy to arrange, there are no nontrivial
solutions for the Radon-Nikodym problem. Excluding this extreme situation, i.e. when ζ is
finite on at least one point, one may easily show the existence of quasi-invariant measures.
See section (4) for more details in the proper case.
Unfortunately we have no definitive answer for the existence question in the most
general situation of gaps treated here, which is perhaps to be expected given that similar
results rely heavily on compactness. However we can offer several partial existence results
which the reader will find in section (7).
The second named author would like to acknowledge the warm hospitality of Rodrigo
Bissacot and his group during a very productive visit to the University of Sao Paulo, when
the bulk of the results presently being reported were developped.
2. Proper equivalence relations.
As mentioned above, we start by analyzing proper equivalence relations, avoiding the
compactness assumption.
3
2.1. Standing Hypothesis. Throughout this notes we will assume that X is a locally
compact, second countable, metrizable space.
We will denote the σ-algebra of Borel measurable subset of X by B(X), and the set
of all Borel measurable functions
f : X → [0, +∞]
by M
+(cid:0)X, B(X)(cid:1).
2.2. Definition. An equivalence relation R ⊆ X × X is said to be proper, provided the
quotient space X/R is Hausdorff, and the quotient map
is a local homeomorphism2.
π : X → X/R
◮ From now on we shall fix a proper equivalence relation R on X.
Given any x in X, we will denote its equivalence class by R(x), in symbols
R(x) = {y ∈ X : (x, y) ∈ R}.
2.3. Definition. For each f in M
+(cid:0)X, B(X)(cid:1), we will let
E(f )
x
f (y).
= Py∈R(x)
E(f )
Observe that the above sum could very well diverge, in which case we of course set
to be ∞. Therefore, like f , one has that E(f ) is a function taking values in [0, ∞].
We will soon prove that E(f ) is B(X)-measurable, but so far we will see it simply
x
as an element of M
respect to which any function is measurable). In other words, E may be seen as a map
+(cid:0)X, P(X)(cid:1), where P(X) is the σ-algebra of all subsets of X (with
E : M
+(cid:0)X, B(X)(cid:1) → M
+(cid:0)X, P(X)(cid:1).
2.4. Proposition. E is σ-additive, in the sense that it is positively homogeneous and
fn(cid:17) =
E(fn),
∞
Pn=1
E(cid:16) ∞
Pn=1
+(cid:0)X, B(X)(cid:1).
for any sequence {fn}n in M
2 A map ϕ : X → Y , between topological spaces X and Y , is said to be a local homeomorphism
provided for every x in X, there are open subsets A ⊆ X, and B ⊆ Y , such that x ∈ A and ϕ is a
homeomorphism from A onto B.
4
Proof. It is evident that E is positively homogeneous. Given any sequence {fn}n in
M
+(cid:0)X, B(X)(cid:1), for every x in X, we have
∞
Pn=1
E(cid:0)
fn(cid:1) x
= Py∈R(x)
∞
Pn=1
fn(y) =
∞
Pn=1 Py∈R(x)
fn(y) =
∞
Pn=1
E(fn)
.
x
(cid:3)
2.5. Proposition. If f is in M
+(cid:0)X, B(X)(cid:1), then so is E(f ).
Proof. Let {Un}n∈ be a countable open cover of X, such that the quotient map
π : X → X/R
is a homeomorphism when restricted to each Un. Also let {ψn}n∈ be a partition of unit
subordinate to this cover.
Given f in M
we then have that
+(cid:0)X, B(X)(cid:1), put fn = f ψn, so that f = Pn fn, pointwise. Using (2.4),
E(f ) = E(cid:16) ∞
Pn=1
fn(cid:17) =
∞
Pn=1
E(fn),
so it suffices to prove that each E(fn) is Borel-measurable. Write
τ : π(Un) → Un
for the inverse of the restriction of π to Un, and let Vn = π−1(cid:0)π(Un)(cid:1). We then claim that
E(fn)
x
= (cid:26) fn(cid:0)τ (π(x))(cid:1),
0,
if x ∈ Vn,
otherwise.
Indeed, when x is not in Vn, then π(x) is not in π(Un). So, while fn vanishes outside
therefore has no
Un, there is no y in Un such that (x, y) ∈ R. The sum defining E(fn)
nonzero terms, and hence E(fn)
= 0.
x
On the other hand, if x is in Vn, then π(x) ∈ π(Un), so π(x) = π(y), for a unique
y in Un, namely y = τ(cid:0)π(x)(cid:1), whence f (y) is the only possibly nonzero term in the
aforementioned sum. Therefore
x
E(fn)
x
= f (y) = fn(cid:0)τ (π(x))(cid:1),
ble on Vn, we have that E(fn) is Borel-measurable on X.
as claimed. Since the correspondence x 7→ fn(cid:0)τ (π(x))(cid:1) is easily seen to be Borel-measura-
+(cid:0)X, B(X)(cid:1)
to itself. We therefore no longer need to consider the σ-algebra P(X), and we shall
henceforth use the simplified notation
Notice that, in view of the above result, E may be viewed as a map from M
(cid:3)
A few other useful properties of E are as follows:
M
+
(X) := M
+(cid:0)X, B(X)(cid:1).
5
2.6. Proposition. Given f, g ∈ M
(i) E(f ) is R-invariant, meaning that if (x, y) ∈ R, then E(f )
(X), one has that:
+
= E(f )
,
y
x
(ii) if g is R-invariant, then E(gf ) = gE(f ),
(iii) if f ≤ g, then E(f ) ≤ E(g),
(iv) if f vanishes outside a subset A ⊆ X, then E(f ) vanishes outside
Orb(A) := {y ∈ X : ∃x ∈ A, (x, y) ∈ R}.
Proof. We prove only (iv). Given x ∈ X, suppose that
0 6= E(f )
x
= Py∈R(x)
f (y).
Then it is easy to see that there exists at least one y such that (x, y) ∈ R, and f (y) 6= 0.
Consequently y ∈ A, whence x ∈ Orb(A). This proves that
E(f )
x
6= 0 ⇒ x ∈ Orb(A),
from where the conclusion follows immediately.
(cid:3)
When multiplying extended real numbers, as in the multiplication "gf " above, we
adopt the convention according to which 0 × ∞ = ∞ × 0 = 0. A trivial, but highly relevant
fact to be noted regarding this convention is that multiplication of positive extended real
numbers is both associative and infinitely distributive, i.e.,
c
∞
Pn=1
an =
can,
∞
Pn=1
for every c and every sequence {an}n in [0, ∞]. Incidentally the above choice for the value
of 0 × ∞ is necessary for the validity of the distributive property, since
0 × ∞ = 0 ×
1
n
∞
Pn=1
=
∞
Pn=1
0 ×
1
n
= 0.
Given ρ, f in M
+
(X), we have that E(ρf ) ∈ M
+
(X). Fixing ρ we may then define
the map
Eρ : f ∈ M
+
(X) 7→ E(ρf ) ∈ M
+
(X),
which is clearly also σ-additive. Therefore, for every measure ν on B(X), we may consider
the measure E∗
ρ(ν) are in
order:
ρ(ν) given by (A.3). Some elementary observations regarding E∗
+
2.7. Proposition. Given a function ρ in M
has that
(i) E∗
(ii) if A is any R-invariant3 Borel-measurable subset of X with ν(A) = 0, then E∗
ρ(ν) is a finite measure if and only if E(ρ) is ν-integrable,
(X) as well as a measure ν on B(X), one
ρ (ν)(A) =
0 as well.
3 A subset A ⊆ X is said to be R-invariant if, whenever x ∈ A, and (x, y) ∈ R, one has that y ∈ A.
6
Proof. The first point follows immediately from
E∗
ρ(ν)(X) = ZX
1 dE∗
ρ(ν) = ZX
Eρ(1) dν = ZX
E(ρ) dν.
Regarding (ii), and denoting the characteristic function of A by 1A, it is obvious that
ρ1A vanishes outside A. Therefore E(ρ1A) vanishes outside Orb(A) by (2.6.iv). However,
since A is invariant, we have that Orb(A) = A, so in fact E(ρ1A) vanishes outside A.
Therefore
E∗
ρ(ν)(A) = ZX
1A dE∗
ρ(ν) = ZX
E(ρ1A) dν = ZX\A
E(ρ1A) dν = 0.
(cid:3)
3. The operator E on Cc(X).
In this section we continue assuming that X satisfies (2.1) and that R ⊆ X × X is a proper
equivalence relation on X. Whenever we speak of R as a topological space, we will be
referring to the topology induced on R by the product topology of X × X.
We will often view R as a groupoid under the multiplication operation according to
which the product (x, y) · (z, w) is defined if and only if y = z, in which case it is set to
be (x, w). The unit space of such a groupoid is therefore the diagonal {(x, x) : x ∈ X},
which we identify with X in the obvious way. The range and source maps are then given
respectively by
r(x, y) = x,
and
s(x, y) = y,
∀ (x, y) ∈ R.
It is well known that R is then a Hausdorff ´etale groupoid.
3.1. Proposition. Given any continuous, complex valued function f on R, suppose that
f vanishes outside a given subset L ⊆ R, such that s(L) is relatively compact. Then:
(i) The expression
g(y) = Xγ:r(γ)=y
f (γ)
gives a well defined and continuous function on X.
(ii) If r(L) is also relatively compact, then g has compact support.
Proof. We first claim that R is closed in X × X. In order to see this let
denote the quotient map and observe that
π : X → X/R
R = {(x, y) ∈ X × X : π(x) = π(y)} = {(x, y) ∈ X × X : (cid:0)π(x), π(y)(cid:1) ∈ ∆},
where ∆ is the diagonal in X/R × X/R. Since X/R is Hausdorff, we have that ∆ is closed,
and hence R is closed in X × X.
7
Given y0 in X, let K be a compact neighborhood of y0, and observe that
r−1(K) ∩ L ⊆ K × s(L),
so r−1(K) ∩ L is relatively compact in X × X, and hence also in the closed subspace R.
For each
γ ∈ r−1(K) ∩ L,
namely the closure of r−1(K) ∩ L within R, let Uγ ⊆ R and Vγ ⊆ X be open sets such
that γ ∈ Uγ, r(γ) ∈ Vγ, and such that the restriction of the range map r to Uγ gives a
homeomorphism onto Vγ.
We shall also insist that, whenever r(γ) 6= y0, the open neighborhood Vγ of r(γ) is
chosen such that y0 /∈ V γ. We therefore get an open cover {Uγ}γ of the compact set
r−1(K) ∩ L, from which one may extract a finite subcover, say {Uγ}γ∈F , where F is some
finite set of γ's.
Splitting F according to whether or not r(γ) = y0, we define
F1 = {α ∈ F : r(α) = y0},
and
F2 = {β ∈ F : r(β) 6= y0}.
We then put
V = int(K) ∩ \α∈F1
Vα ∩ \β∈F2
X \ V β,
and we claim that y0 ∈ V . To see this, notice that y0 lies in int(K) because K is a
neighborhood of y0. Moreover, for every α in F1, we have that
and finally, for every β in F2, we have explicitly chosen Vβ so that y0 /∈ V β.
We next claim that, for every η in R,
y0 = r(α) ∈ Vα,
r(η) ∈ V ∧ f (η) 6= 0 ⇒ η ∈ Uα, for some α ∈ F1.
(3.1.1)
Indeed, given η satisfying the above antecedent, we clearly have that
η ∈ r−1(K) ∩ L,
so there exists some γ in F , such that η ∈ Uγ, and we would now like to decide whether γ
lies in F1 or in F2. The key observation here is that
r(η) ∈ V ∩ r(Uγ) = V ∩ Vγ,
so Vγ has a nonempty intersection with V , and this can only happen when γ ∈ F1, thus
completing the proof of (3.1.1).
Next use the fact that R is Hausdorff to produce a collection of pairwise disjoint open
sets {Wα}α∈F1 such that each α ∈ Wα and finally put
Ω = V ∩ \α∈F1
r(Wα ∩ Uα).
8
Noticing that Wα ∩Uα is open in Uα, we see that r(Wα ∩Uα) is open in r(Uα) = Vα, so
Ω is an open subset of X. Also, since α ∈ Wα ∩ Uα, we have that y0 = r(α) ∈ r(Wα ∩ Uα),
so y0 ∈ Ω.
For each α in F1, denote by tα the inverse of the homeomorphism
r↾Uσ : Uα → Vα,
and, regarding the function g referred to in the statement, we claim that for every y in Ω,
one has that
g(y) = Xα∈F1
f(cid:0)tα(y)(cid:1).
(3.1.2)
To prove this claim it is enough to show that
{γ ∈ R : r(γ) = y, f (γ) 6= 0} = {tα(y) : α ∈ F1, f (tα(y)) 6= 0},
(3.1.3)
and that the tα(y) in the description of the set in the right hand side above are pairwise
distinct.
With respect to this last statement, notice that for each α in F1,
tα(y) ∈ tα(cid:0)r(Wα ∩ Uα)(cid:1) = Wα ∩ Uα ⊆ Wα,
so the tα(y) lie in pairwise disjoint sets and hence are necessarily pairwise distinct.
We next observe that the inclusion "⊇" in (3.1.3) is evident, so we focus on the reverse
inclusion "⊆". For this, pick γ in R such that r(γ) = y, and f (γ) 6= 0, and notice that by
(3.1.1) it follows that γ ∈ Uα for some α ∈ F1. Therefore γ = tα(y), so we see that γ lies
in the set in the right hand side of (3.1.3).
This proves (3.1.3) and hence also (3.1.2), from where it is clear that the sum defining
g has finitely many nonzero terms, so that g is well defined, and moreover that g is
continuous.
In order to prove (ii), it is enough to observe that if g(y) 6= 0, there must be at least
one γ with r(γ) = y, and f (γ) 6= 0, whence γ ∈ L, and then y ∈ r(L). Viewing through
the counterpositive
y /∈ r(L) ⇒ f (γ) = 0, for all γ such that r(γ) = y,
which in turn implies that g(y) = 0. Thus g vanishes outside the relatively compact set
r(L), and hence it is compactly supported.
(cid:3)
3.2. Corollary. Given f ∈ Cc(R), the correspondence
x 7→ Xy∈R(x)
f (x, y)
defines a compactly supported, continuous function on X.
Proof. Follows immediately from (3.1.ii) upon choosing L to be the support of f .
(cid:3)
9
Recall that the operator E defined in (2.3) is only defined for non-negative functions.
This is due to the fact that the summation involved in its definition is not supposed to
converge but, as long as the summands are non-negative, we may always assign a sensible
value to the sum, that value being ∞ in the divergent case. In case of compactly supported
functions the situation is however much better behaved:
3.3. Proposition. Given f in Cc(X), and for every x in X, the sum
Xy∈R(x)
f (y),
has at most finitely many nonzero terms. Moreover, defining
E(f )
x
= Xy∈R(x)
f (y),
∀ x ∈ X,
one has that E(f ) is a continuous function on X.
Proof. Since R is a proper equivalence relation, we have that R(x) is a closed, discrete set
for every x in X. Therefore, if K is the compact support of f , one has that R(x) ∩ K is
finite from where the first assertion follows immediately.
Addressing the last assertion, consider the continuous function
Denoting the support of f by L, notice that g vanishes outside the set
g : (x, y) ∈ R 7→ f (y) ∈ C.
K := (X × L) ∩ R.
Since s(K) ⊆ L, we see that s(K) is relatively compact. We may therefore employ
(3.1.i) to conclude that the function g defined there is continuous, namely
g(x) = Xγ:r(γ)=x
g(γ) = Xy∈R(x)
g(x, y) = Xy∈R(x)
f (y) = E(f )
,
x
concluding the proof.
In view of the above result we get a map
On the other hand, recall that in (2.3) we defined an operator
E : Cc(X) → C(X).
E : M
+
(X) → M
+
(X),
(cid:3)
using the exact same formula as in (3.3). Clearly the two operators referred to above
coincide on the intersection of their domains, so there is no ambiguity in using the same
notation "E" for these maps.
Some of the main properties of E on Cc(X) reflect those listed in (2.6):
10
3.4. Proposition. For every f and g in Cc(X), one has that:
(i) E(f ) is R-invariant,
(ii) if g is R-invariant, then E(gf ) = gE(f ),
(iii) E(f ) is continuous,
(iv) E(f ) is bounded.
Proof. Leaving the easy proofs of (i) and (ii) to the reader, we notice that (iii) was already
proved in (3.3).
Regarding (iv), let K be the compact support of f , and let M be the supremum of
E(f ) on K, which is finite by (iii). We will then prove that E(f ) is bounded by M on
all of X. In order to prove that
for any given x in X, we may evidently assume that E(f )
(cid:12)(cid:12)E(f )
x(cid:12)(cid:12) ≤ M,
(3.4.1)
6= 0. In this case
x
0 6= E(f )
x
= Xy∈R(x)
f (y),
so there exists at least one y in K such that (x, y) ∈ R. Therefore
proving (3.4.1).
(cid:12)(cid:12)E(f )
x(cid:12)(cid:12)
(i)
= (cid:12)(cid:12)E(f )
y(cid:12)(cid:12) ≤ M,
(cid:3)
Complementing (2.7), we may now describe a few other relevant properties of E∗
ρ(ν)
under the extra hypothesis that ρ is finitely valued and continuous.
3.5. Proposition. Let ν be a measure on B(X), and let ρ : X → R, be a non-negative,
continuous function. Setting µ = E∗
(i) if ν(X) < ∞, then µ is a Borel measure (i.e. finite on compact sets),
(ii) if µ′ is any measure on B(X) such that
ρ(ν), one has that:
∞ > ZX
f dµ′ = ZX
E(ρf ) dν,
∀ f ∈ C+
c (X),
then µ = µ′, and in particular the above identity holds for every f in M
+
(X).
Proof. In order to verify (i), and using (B.4), it is enough to prove that every f in C+
c (X)
is µ-integrable. Given such an f , notice that the continuity of ρ implies that ρf lies in
Cc(X), whence E(ρf ) is bounded by (3.4.iv). Therefore
ZX
f dµ = ZX
E(ρf ) dν ≤ ν(X)kE(ρf )k∞ < ∞.
11
Addressing (ii), observe that the hypothesis says that every f in C+
c (X) is µ′-integra-
ble, so another application of (B.4) tells us that µ′ is a Borel measure. By hypothesis we
then have that
f dµ,
∀ f ∈ C+
c (X),
ZX
f dµ′ = ZX
from where we deduce that µ is also a Borel measure. Since any f in Cc(X) may be written
as the linear combination of functions in C+
c (X), we deduce that the identity displayed
above holds for every f in Cc(X), so µ = µ′, by (B.3) and the uniqueness part of the
Riesz-Markov Theorem.
(cid:3)
4. Proper equivalence relations and quasi-invariant measures.
◮ As before, throughout this section we fix a space X satisfying (2.1), as well as a proper
equivalence relation R on X. We will moreover fix a continuous function
ρ : X → R,
which will henceforth be supposed strictly positive, i.e,
ρ(x) > 0,
∀ x ∈ X,
and which will be referred to as the potential for R.
The relevance of ρ is that it leads to a multiplicative cocycle on R via the formula
D(x, y) := ρ(x)/ρ(y),
∀ (x, y) ∈ R,
(4.1)
and the goal of this section is to study quasi-invariant measures relative to this cocycle.
See (4.9) below for the precise definition.
In some applications of our theory, the role of ρ is played by the function ρ(x) = eβh(x),
where β > 0 and h is a continuous, real valued function on X. The assumption that ρ is
strictly positive then holds automatically. Another reason why we need to assume that ρ
is never zero is that, otherwise, the the above definition of D would run into trouble.
4.2. Definition. For each x em X, define
ζ(x) := E(ρ)
x
= Xy∈R(x)
ρ(y).
We will refer to ζ as the partition function for the potencial ρ.
4.3. Proposition. ζ is bounded below by ρ, and consequently
0 < ζ(x) ≤ ∞,
∀ x ∈ X.
Proof. Obvious.
(cid:3)
12
Since ζ is defined to be E(ρ), we have by (2.5) that ζ lies in M
+
(X). We may in fact
prove that ζ satisfies a stronger regularity property:
4.4. Proposition. ζ is lower semi-continuous.
Proof. Let {ϕn}n be as in (B.5). Then
ζ = E(ρ) = E(cid:0) lim
n→∞
ρϕn(cid:1) (A.3.i)
= lim
n→∞
E(ρϕn),
whence ζ is the limit of an increasing sequence of continuous functions by (3.3), from where
the conclusion follows.
(cid:3)
4.5. Corollary. The set
is a Gδ, hence a Borel set.
Proof. Noting that
Zρ = {x ∈ X : ζ(x) = ∞}
Zρ = \k∈
{x ∈ X : ζ(x) > k},
the result is an immediate consequence of (4.4).
(cid:3)
In what follows we will make frequent references to the function ζ −1, so it is worth
discussing it briefly now. Recall from (4.3) that ζ(x) > 0, for all x in X, so we will never
run into the trouble of considering the inverse of zero. On the other hand, when ζ(x) = ∞,
we evidently put ζ −1(x) = 0.
In view of our convention that ∞ × 0 = 0, observe that
ζ(x)ζ(x)−1 = (cid:26) 0,
if x ∈ Zρ,
1, otherwise.
so we have that
ζζ −1 = 1X\Zρ.
(4.6)
(4.7)
In particular we note the following partial-isometric-like property of ζ, to be used
shortly:
ζ −1ζζ −1 = ζ −1.
(4.8)
All things considered, we will see that the somewhat unusual fact that ζζ −1 vanishes
on Zρ will not be so crucial. For example, we will soon encounter expressions such as
ZX
ζζ −1 dµ,
but often the measure µ will also vanishes on Zρ, so the funny behavior of ζζ −1 on Zρ
becomes totally irrelevant.
We next recall the definition of a quasi-invariant measure in the special case of ´etale
groupoids.
13
4.9. Definition. [5: I.3.15] Let G be an ´etale groupoid and let D : G → R
plicative cocycle. A measure µ on G(0) is said to be quasi-invariant relative to D when
∗
+ be a multi-
ZG(0) Xγ∈r−1(x)
f (γ) dµ(x) = ZG(0) Xγ∈s−1(x)
f (γ)D(γ) dµ(x),
for every f in Cc(G).
For the case of our groupoid R, the above quasi-invariance condition becomes
ZX Xy∈R(x)
f (x, y) dµ(x) = ZX Xx∈R(y)
f (x, y)D(x, y) dµ(y),
(4.10)
for every f in Cc(R).
The following result lists several equivalent conditions for a measure to solve the
Radon-Nikodym problem.
4.11. Theorem. Let X be a topological space satisfying (2.1). Also let R be a proper
equivalence relation on X, seen as an ´etale groupoid. Given a continuous, strictly positive
function ρ : X → R, consider the cocycle defined on R by D(x, y) = ρ(x)/ρ(y). Then, for
every finite measure µ on X, the following are equivalent:
(i) µ is D-quasi-invariant,
(ii) ZX
f E(ρg) dµ = ZX
f ζ dµ = ZX
ZX
= ZX
ZX
f dµ
(iii)
(iv)
E(ρf )g dµ,
∀f, g ∈ Cc(X) ,
E(ρf ) dµ,
∀f ∈ C+
c (X) ,
E(f ρζ −1) dµ, ∀f ∈ C+
c (X) ,
(v) there exists a positive measure ν on X, with respect to which ζ is integrable, and
ZX
f dµ
= ZX
E(ρf ) dν,
∀f ∈ Cc(X) .
In addition, if any of the above equivalent conditions hold, then µ(Zρ) = 0.
Proof. (i) ⇒ (ii). Pick f and g in Cc(X), and consider the function F on R given by the
formula
F (x, y) = f (x)g(y)ρ(y).
Plugging F in (4.10) we have
ZX Xy∈R(x)
f (x)g(y)ρ(y) dµ(x) = ZX Xx∈R(y)
f (x)g(y)ρ(x) dµ(y),
14
which translated precisely into (ii).
(ii) ⇒ (iii). Given f in C+
c (X), let {ϕn}n be as in (B.5). Then
= lim
f E(ρϕn) dµ
f E( lim
n→∞
ρϕn) dµ
(A.3.i)
=
E(ρf )ϕn dµ = ZX
E(ρf ) dµ.
f ζ dµ = ZX
ZX
n→∞ZX
f E(ρ) dµ = ZX
n→∞ZX
(ii)
= lim
(iii) ⇒ (iv). We will first prove that µ(Zρ) = 0.
In order to do it suppose by way of
contradiction that µ(Zρ) > 0. Since µ is finite, it is regular by (B.3), so there exists a
compact set K ⊆ Zρ, such that µ(K) > 0. Using Uryhson, take f in C+
c (X), such that
f K = 1, so that
∞ = ∞ × µ(K) = ZK
ζ dµ = ZK
f ζ dµ ≤
≤ ZX
f ζ dµ
(iii)
= ZX
E(ρf ) dµ ≤ kE(ρf )k∞ µ(X)
(3.4.iv)
< ∞.
Arriving at a contradiction we conclude that µ(Zρ) = 0, as desired.
We next claim that (iii) indeed holds for all f in M
+
(X), namely that
ZX
f ζ dµ = ZX
E(ρf ) dµ,
∀ f ∈ M
+
(X).
(4.11.1)
Letting ζµ and µ play the roles of µ′ and ν, respectively, in (3.5.ii), we only need to
prove that every f in C+
c (X) is integrable with respect to ζµ, but this follows from
ZX
f ζ dµ
(iii)
= ZX
Eρ(f ) dµ ≤ µ(X)kEρ(f )k∞
(3.4.iv)
< ∞.
Therefore (4.11.1) is verified so, for any f in C+
c (X), we may plug in f ζ −1 there, obtaining
ZX
E(ρf ζ −1) dµ = ZX
f ζ −1ζ dµ = ZX
f dµ,
where the last equality is justified by (4.6), which says that ζζ −1 = 1 on X \ Zρ, and by
the fact that µ vanishes on Zρ. This proves (iv).
(iv) ⇒ (v). Defining ν := ζ −1µ, and given f in C+
c (X), we have that
ZX
f dµ = ZX
E(ρf ζ −1) dµ
(2.6)
= ZX
E(ρf )ζ −1 dµ = ZX
E(ρf ) dν.
Since Cc(X) is linearly spanned by C+
employing (3.5.ii) once more, one has that
c (X), the last assertion in (v) follows. Furthermore,
ZX
f dµ = ZX
E(ρf ) dν,
∀ f ∈ M
+
(X),
15
so we are allowed to plug f = 1 above, whence
ZX
ζ dν = ZX
E(ρ) dν = ZX
1 dµ = µ(X) < ∞,
hence proving the remaining first assertion of (v).
(v) ⇒ (i). In order to prove that µ is D-quasi-invariant, we need to check (4.10) for every
f in Cc(R). As a notational aid, let us temporarily write
A(x) = Xy∈R(x)
f (x, y),
and B(y) = Xx∈R(y)
f (x, y)D(x, y),
so that our goal is to prove that A and B have the same integral relative to µ. En passant,
notice that A and B lie in Cc(X) by (3.2).
Starting from the left-hand-side of (4.10), observe that
ZX Xy∈R(x)
f (x, y) dµ(x) = ZX
A dµ
(v)
= ZX
E(ρA) dν =
= ZX Xz∈R(x)
ρ(z) Xy∈R(z)
f (z, y) dν(x) = ZX Xz∈R(x) Xy∈R(z)
ρ(z)f (z, y) dν(x).
(4.11.2)
On the other hand, starting from the right-hand-side of (4.10), we have
ZX Xz∈R(y)
f (z, y)D(z, y) dµ(y) = ZX
B dµ
(v)
= ZX
E(ρB) dν =
= ZX Xy∈R(x)
ρ(y) Xz∈R(y)
f (z, y)D(z, y) dν(x) = ZX Xy∈R(x) Xz∈R(y)
ρ(z)f (z, y) dν(x).
(4.11.3)
Notice that the diference between (4.11.2) and (4.11.3) is simply that, in the former,
the sum ranges over all pairs (z, y) such that
while, in the latter, the pairs (y, z) considered are those for which
x ∼
R
z ∼
R
y,
x ∼
R
y ∼
R
z.
Being an equivalence relation, R is transitive, whence in both cases above the sum
ranges over all y and all z in the equivalence class of x, and therefore we see that (4.11.2)
and (4.11.3) coincide. This proves (4.10) and hence that µ is D-quasi-invariant.
(cid:3)
16
The characterization given by (4.11.v) may be used to produce D-quasi-invariant
measures, as we now show:
4.12. Corollary. Given a measure ν on X such that ζ is ν-integrable, there exists a
unique finite, D-quasi-invariant measure µ on X such that
ZX
f dµ = ZX
E(ρf ) dν,
∀ f ∈ Cc(X).
Proof. Given ν, let µ = E∗
ρ(ν), so
µ(X) = ZX
1 dµ = ZX
E(ρ) dν = ZX
ζ dν < ∞,
so µ is indeed a finite measure and it is D-quasi-invariant because it satisfies (4.11.v). The
uniqueness of µ now follows from the uniqueness part of the Riesz-Markov Theorem. (cid:3)
The next result settles the question regarding the existence of nontrivial D-quasi-
invariant measures.
4.13. Corollary. The following are equivalent:
(i) there exists at least one D-quasi-invariant probability measure on X,
(ii) ζ is not identically infinite.
Proof. Recall that Zρ is the set of points where ζ is infinite, so (ii) is equivalent to saying
that Zρ 6= X, or equivalently that X\Zρ is nonempty.
Assuming (i), let µ be a D-quasi-invariant probability measure on X. By the last
sentence in (4.11) we have that µ(Zρ) = 0, and hence that µ(X\Zρ) = 1, so X\Zρ 6= ∅,
proving (ii). Conversely, if X\Zρ is nonempty, it is easy to exhibit a measure ν on X
satisfying
ZX
ζ dν = 1.
Take, for example, any point y0 ∈ X\Zρ and, observing that 0 < ζ(y0) < ∞ by (4.3), it is
enough to choose
ν = ζ(y0)−1δy0 ,
where δy0 is the Dirac measure on y0. Given any such ν, the measure µ built in (4.12) in
terms of ν is a D-quasi-invariant probability measure, proving (i).
(cid:3)
The remainder of this section will be devoted to a closer look at the fourth condition
of (4.11).
4.14. Proposition. Consider the operator Pρ : M
+
(X) → M
+
(X), given by
Pρ(f ) = E(f ρζ −1),
∀ f ∈ M
+
(X).
Then
(i) Pρ(1) = 1X\Zρ,
(ii) P 2
ρ = Pρ,
(iii) the range of Pρ coincides with the set M
+
R,ρ(X), consisting of all R-invariant functions
f in M
+
(X) which vanish on Zρ.
17
Proof. We should first observe that, since ρ and ζ −1 lie in M
a subset of M
(X) by (2.5).
+
In order to prove the first assertion, we compute
+
(X), the range of Pρ is indeed
Pρ(1) = E(ρζ −1)
(2.6.ii)
= E(ρ)ζ −1 = ζζ −1 (4.6)
= 1X\Zρ.
We next claim that:
(a) the range of Pρ is contained in M
+
R,ρ(X), and
(b) Pρ(f ) = f , for every f in M
+
R,ρ(X).
In order to verify (a), pick any f in M
+
(X). Since
Pρ(f ) = E(f ρ)ζ −1
by (2.6.ii), and since ζ −1 vanishes on Zρ, then Pρ(f ) also vanishes on Zρ. The fact that
Pρ(f ) lies in M
R,ρ(X) then follows immediately from (2.6.i).
+
To prove (b), let f ∈ M
+
R,ρ(X). Then
Pρ(f ) = E(f ρζ −1)
(2.6.ii)
= f E(ρ)ζ −1 = f ζζ −1 = f,
where the last step relies on the fact that f vanishes on Zρ. This said, (ii) and (iii) follow
trivially from (a) and (b).
(cid:3)
Among the characterizations of D-quasi-invariant measures given by (4.11), a par-
ticularly useful one is (4.11.iv), given the nice properties of the operator Pρ described in
(4.14). For that reason, and also for future reference, we restate part of the conclusions of
(4.11) in a way as to emphasize the importance of Pρ.
4.15. Corollary. Under the conditions of (4.11) one has that µ is D-quasi-invariant if
and only if P ∗
ρ (µ) = µ.
Some further important facts involving P ∗
ρ are as follows.
ρ (ν)(A) = ν(A\Zρ),
4.16. Proposition. Let ν be any finite measure on X. Then
(i) if A ⊆ X is an invariant Borel subset, then P ∗
(ii) if ν vanishes on an invariant Borel set A ⊆ X, then the same is true for P ∗
(iii) P ∗
(iv) P ∗
(v) if ν is a probability measure vanishing on Zρ, then so is P ∗
(vi) P ∗
(vii) P ∗
ρ (ν) is finite,
ρ (ν) is nonzero if and only if ν(X\Zρ) is nonzero,
ρ (ν)(cid:1) = P ∗
ρ (ν) is D-quasi-invariant.
ρ (ν),
ρ (ν),
ρ(cid:0)P ∗
ρ (ν),
18
Proof. Given A as in (i), we have
P ∗
ρ (ν)(A) = ZX
1A dP ∗
ρ (ν) = ZX
E(1Aρζ −1) dν
(2.6.ii)
= ZX
1AE(ρ)ζ −1 dν =
= ZX
1Aζζ −1 dν
(4.6)
= ZX
1A1X\Zρ dν = ν(A\Zρ).
Points (ii -- v) then follow immediately from (i). Regarding (vi), it is an easy consequence
of (4.14.ii). Finally let us prove (vii). For this, set µ = P ∗
ρ (ν), so we see from (vi) that
P ∗
(cid:3)
ρ (µ) = µ, and the conclusion follows from (4.15).
5. Generalized approximately proper equivalence relations.
As before, throughout this section we assume that X is a locally compact, second countable,
metrizable space.
5.1. Definition. By a generalized approximately proper equivalence relation on X, a gap
for short, we shall mean a pair
R = (cid:16){Un}n∈, {Rn}n∈(cid:17),
where each Un is an open subset of X, and each Rn is a proper equivalence relation on
Un, such that
(i) X = U0 ⊇ U1 ⊇ U2 ⊇ · · ·
(ii) R0 is the identity relation on U0, that is, R0 is the diagonal in U0 × U0,
(iii) if n ≤ m, then Rn ∩ (Un × Um) ⊆ Rm.
Two immediate consequences of the definition are as follows:
5.2. Proposition. If R = (cid:0){Un}n∈, {Rn}n∈(cid:1) is a gap on X then, whenever n ≤ m,
one has that
(i) the restriction of Rn to Um, namely Rn ∩ (Um × Um), is contained in Rm,
(ii) If n ≤ m, then Um is invariant under Rn in the sense that if a point x in Un is
equivalent under Rn to a point y in Um, then x lies in Um.
Proof. We have
Rn ∩ (Um × Um) ⊆ Rn ∩ (Un × Um)
(5.1.iii)
⊆ Rm,
proving (i). If x and y are as in (ii), then
(x, y) ∈ Rn ∩ (Un × Um)
(5.1.iii)
⊆ Rm ⊆ Um × Um,
so x ∈ Um.
(cid:3)
19
It is not hard to see that also (5.2.i -- ii) imply (5.1.iii), so the reader might think of the
latter as subsuming the former, which some may consider a more natural set of conditions.
The main motivation and the main source of examples for gaps is described in detail
in Section (8), below.
◮ From now on we fix a gap R = (cid:0){Un}n∈, {Rn}n∈(cid:1) on X.
5.3. Proposition. Setting
one has that R is an equivalence relation on X.
R = Sn∈
Rn,
Proof. The only slightly nontrivial point regards the transitivity of R. In order to prove
it, suppose that (x, y) and (y, z) lie in R. We may then pick n and m such that (x, y) ∈ Rn
and (y, z) ∈ Rm, and we may assume without loss of generality that n ≤ m. In that case
we have that
Since Rm is transitive we have that
(x, y) ∈ Rn ∩ (Un × Um)
(5.1.iii)
⊆ Rm.
(x, z) ∈ Rm ⊆ R.
(cid:3)
5.4. Lemma. Equipping each Rn with the topology induced from the product topology
on X × X, one has that Rn ∩ Rm is open in Rn, for all n and m in .
Proof. Given (x, y) in Rn ∩ Rm, by the definition of the product topology on Rn we must
prove the existence of open subsets V, W ⊆ X, such that
(x, y) ∈ Rn ∩ (V × W ) ⊆ Rn ∩ Rm.
Assuming that n ≤ m, choose V = Un, and W = Um, and observe that the above
inclusion is then immediately verified thanks to (5.1.iii). On the other hand, proving the
result under the opposite assumption, i.e. that n ≥ m, is equivalent to maintaining the
assumption that n ≤ m (with which the reader must be used to by now) and proving
instead that
(x, y) ∈ Rm ∩ (V × W ) ⊆ Rn ∩ Rm.
(5.4.1)
For each k ∈ , denote by πk the quotient map
πk : Uk → Uk/Rk,
and for each k ∈ {n, m}, let us choose an open set Wk ⊆ Uk, such that y ∈ Wk, and such
that πk restricts to a homeomorphism from Wk to the open set πk(Wk). Replacing both
Wn and Wm by
we may assume that Wn = Wm.
W := Wn ∩ Wm,
20
Notice that πn(x) = πn(y) ∈ πn(W ), so we have that x ∈ π−1
n (cid:0)πn(W )(cid:1), and upon
setting
we see that V is an open subset of X, and we moreover claim that (5.4.1) holds. The first
part, namely that (x, y) ∈ Rm ∩ (V × W ), is evident and, in order to prove that
V := π−1
n (cid:0)πn(W )(cid:1) ∩ Um,
Rm ∩ (V × W ) ⊆ Rn ∩ Rm,
(5.4.2)
let us pick (z, w) in the set appearing in the left-hand side above. It follows that z ∈ V ,
whence πn(z) ∈ πn(W ), so there exists some w′ in W such that πn(z) = πn(w′). Another
way to express this is by saying that (z, w′) ∈ Rn, but since (z, w′) also lies in Um × Um,
we deduce that
(z, w′) ∈ Rn ∩ (Um × Um) ⊆ Rm.
Recall that (z, w) ∈ Rm, as well, so transitivity yields (w, w′) ∈ Rm. Observing that both
w and w′ lie in W , and using that πm is injective on W , we see that w = w′, whence
(z, w) = (z, w′) ∈ Rn.
This finishes the verification of (5.4.2), and hence also of (5.4.1), concluding the proof. (cid:3)
Recal that the inductive limit topology on the union of an increasing sequence of
topological spaces
X0 ⊆ X1 ⊆ X2 ⊆ · · ·
is the topology according to which a subset U ⊆ Sn Xn is open if and only if U ∩ Xn
In our situation, where R = Sn Rn, the Rn do not form
is open in Xn, for every n.
an increasing sequence of subsets but one may nevertheless equipp X with the topology
defined as above, that is, in which a subset U ⊆ R is open if and only if U ∩ Rn is open
in Rn, for every n. Even though this might constitute a slight abuse of the language, we
shall refer to that topology as the inductive limit topology on R.
5.5. Lemma. Equipping R with the inductive limit topology we have that each Rn is
open in R.
Proof. Follows immediately from (5.4).
(cid:3)
The two previous Lemmas form the key to showing the following result, whose other-
wise easy proof we leave for the reader.
5.6. Proposition. Given a generalized approximately proper equivalence relation on X,
one has that R is an ´etale groupoid when equipped with the inductive limit topology.
21
6. Quasi-invariant measures and gaps.
As before, throughout this section we assume that X is a locally compact, second countable,
metrizable space. We will also assume that we are given a gap
on X.
R = (cid:16){Un}n∈, {Rn}n∈(cid:17)
6.1. Definition. By a potential for R we shall mean a collection {kn}n≥1, of continuous
functions
such that for every n ≥ 1, one has that
kn : Un → R,
(x, y) ∈ Rn−1 ∩ (Un × Un) ⇒ kn(x) = kn(y).
(6.1.1)
It is perhaps worth pointing out that a potential involves no k0. On the other hand, the
lowest case of (6.1.1) is tautological, that is, when n = 1, we have that Rn−1, also known
as R0, is the identity relation, and it is no surprise that x = y implies that kn(x) = kn(y).
Regarding (6.1.1), notice that
because
Rn−1 ∩ (Un × Un) = Rn−1 ∩ Rn,
Rn−1 ∩ Rn ⊆ Rn−1 ∩ (Un × Un) ⊆ Rn−1 ∩ (Un−1 × Un)
(5.1.iii)
⊆ Rn−1 ∩ Rn.
An equivalent way to state (6.1.1) is therefore to require that kn(x) = kn(y), for all (x, y)
in Rn−1 ∩ Rn.
◮ From now on we assume that we are given a potential {kn}n≥1 for R.
Our next goal is to use a potential to produce a cocycle on the groupoid R = Sn∈
As a first step we introduce the following notation:
Rn.
6.2. Definition. For all n ≥ 0, let hn : Un → R, be defined recursively by h0 = 0, and
In addition, for all n ≥ 0, we define
hn = hn−1↾Un + kn,
∀ n ≥ 1.
cn : (x, y) ∈ Rn 7→ hn(x) − hn(y) ∈ R.
Of course one may alternatively define hn by
hn =
n
Xi=1
ki↾Un ,
observing that, when n = 0, the usual convention about sums without any summands gives
h0 = 0, as expected.
22
6.3. Proposition. For every n ≥ 1, and all (x, y) in Rn−1 ∩ Rn, one has that
cn−1(x, y) = cn(x, y).
Proof. The difference between cn(x, y) and cn−1(x, y) is precisely kn(x) − kn(y), but since
(x, y) lies in Rn−1 ∩ Rn, condition (6.1.1) applies.
(cid:3)
6.4. Proposition. There exists a (necessarily unique) continuous cocycle c on R, such
that c = cn on each Rn.
Proof. We first claim that, whenever 0 ≤ n ≤ m, one has that
Rn ∩ Rm = Rn ∩ Rn+1 ∩ . . . ∩ Rm−1 ∩ Rm.
(6.4.1)
In order to see this, it is clearly enough to show that Rn ∩ Rm ⊆ Rk, for every k with
n ≤ k ≤ m, and in turn this follows from
Rn ∩ Rm ⊆ Rn ∩ (Um × Um) ⊆ Rn ∩ (Un × Uk)
(5.1.iii)
⊆ Rk.
Given any γ in R, choose n such that γ ∈ Rn, and put
c(γ) = cn(γ).
To see that this is well defined, suppose that γ ∈ Rm, for some other m, and let us prove
that cn(γ) = cm(γ). Assuming without loss of generality that n ≤ m, it follows from
(6.4.1) that γ ∈ Rn ∩ Rn+1 ∩ . . . ∩ Rm−1 ∩ Rm, so we may apply (6.3) to show that
cn(γ) = cn+1(γ) = · · · = cm−1(γ) = cm(γ).
This shows that c is well defined and we leave it as an easy exercise to show that c is
(cid:3)
a continuous cocycle on R.
We shall next present two general results about quasi-invariant measures on ´etale
groupoids, to be used later.
6.5. Proposition. Let G be an ´etale groupoid and suppose that we are given a collection
{Gi}i∈I of open subgroupoids Gi ⊆ G, such that G = Si∈I Gi. Suppose moreover that
∗
D : G → R
+ is a continuous multiplicative cocycle and that µ is a finite measure on
G(0). Then µ is D-quasi-invariant if and only if the restriction (see (A.4) for a discussion
regarding the concept of restricting a measure to a subset) of µ to G(0)
is quasi-invariant
relative to the restriction of D to Gi, for every i.
i
Proof. We prove only the "if" part, leaving the "only if" part to the reader. We must
therefore check (4.9) for every f in Cc(G). By [2: 3.10] (which holds even if G is non-
Hausdorff), we have that f may be written as a finite linear combination of functions fj,
each of which lies in Cc(Uj), for some open bissection Uj. Therefore, since both sides of
(4.9) are clearly linear with respect to f , it suffices to prove (4.9) under the assumption
that f ∈ Cc(U ), for some open bissection U .
23
Letting K be the compact support of f , recall that the hypotheses imply that {U ∩Gi}i
k=1 and a partition of unit
is an open cover for K. Choosing a finite subcover {U ∩ Gik }n
{ϕk}n
k=1 subordinate to it [8: 21.1.5], we may write
f =
n
Xk=1
f ϕk,
observing that f ϕk lies in Cc(Gik ).
The upshot of this argument is that we may further reduce (4.9) by assuming that f is
supported on a single Gi. Under this assumption, observe that the integrands in both sides
of (4.9) vanish whenever x is not in G(0)
, so it suffices to verify a variant of (4.9), namely
where both occurences of G(0) are replaced by G(0)
. The resulting expression is then seen
to hold because the restriction of µ to G(0)
(cid:3)
is D-quasi-invariant by hypothesis.
i
i
i
6.6. Lemma. Let G be an ´etale groupoid with a continuous multiplicative cocycle D :
G → R
(i) If ϕ is a bounded, invariant4, Borel-measurable function on G(0), then ϕµ is also
+, and let µ be a finite, D-quasi-invariant measure on G(0).
∗
D-quasi-invariant.
(ii) If E is an invariant5, Borel subset of G(0), then µE := 1E µ is also D-quasi-invariant.
Proof. In order to prove (i) we pick any f in Cc(G) and we set out to verify (4.9) relative
to the measure ϕµ. Starting from the left-hand side, we have
ZG(0) Xγ∈r−1(x)
f (γ) dϕµ(x) = ZG(0) Xγ∈r−1(x)
= ZG(0) Xγ∈s−1(x)
(4.9)
f (γ)ϕ(cid:0)r(γ)(cid:1) dµ(x)
= ZG(0) Xγ∈r−1(x)
f (γ)ϕ(x) dµ(x) =
f (γ)ϕ(cid:0)s(γ)(cid:1)D(γ) dµ(x) =
= ZG(0) Xγ∈s−1(x)
f (γ)D(γ)ϕ(x) dµ(x) = ZG(0) Xγ∈s−1(x)
f (γ)D(γ) dϕµ(x),
proving (i). Point (ii) now follows from (i) upon taking ϕ to be the characteristic function
of E.
(cid:3)
Returning to the gap we have fixed at the beginning of this section, and assuming we
are given a potential {kn}n≥1, leading up to the cocycle c of (6.4), consider the multiplica-
tive cocycle
D : γ ∈ R 7→ ec(γ) ∈ R
∗
+,
as well as the multiplicative cocycles
We then have the following immediate consequence of (6.5):
Dn : γ ∈ Rn 7→ ecn(γ) ∈ R
∗
+.
4 A function ϕ defined on G(0) is said to be invariant when ϕ(r(γ)) = ϕ(s(γ)), for every γ in G.
5 A subset E ⊆ G(0) is said to be invariant when r(γ) ∈ E ⇔ s(γ) ∈ E, for every γ in G.
24
6.7. Corollary. Let µ be a finite measure on X. Then µ is D-quasi-invariant if and only
if µ↾Un is Dn-quasi-invariant for every n in .
Once the question of the quasi-invariance of a measure µ on X is reduced to the quasi-
invariance of measures on proper equivalence relations, namely the Rn's in the above
Corollary, the results of Section (4) apply. Our goal in what follows is to patch the
conclusions of these results for the various Rn in a meaningful way from the point of view
of R. For each n ∈ , we shall let
and we will henceforth let ζn be the partition function given by (4.2) in terms of ρn.
ρn : x ∈ Un 7→ ehn(x) ∈ R
∗
+,
Alongside ρn, Dn and ζn, all of the other ingredients introduced in Section (4) will
(Un) given
also be relevant here, such as Zρn and Pρn , as well as the operator En on M
by (2.3) relative to the equivalence relation Rn.
+
As a first use of these notations we have the following immediate consequence of (6.7)
and (4.15).
6.8. Corollary. Let µ be a finite measure on X. Then µ is D-quasi-invariant if and only
if
P ∗
ρn (µ↾Un) = µ↾Un,
∀ n ∈ .
Part of the difficulty in simultaneously dealing with so many maps and sets is the fact
that they each refer to a different equivalence relation. Attempting to bring everything to
a common environment we introduce the following:
6.9. Definition. Let n ∈ be given.
(i) For any f in M
+
(Un), we will denote by ιn(f ) the extension of f to the whole of X
obtained by setting it to be zero outside Un. When no confusion is likely to arise, we
shall denote that extension simply by f , by abuse of language.
(ii) We will write Zn and Yn for Zρn and Un\Zρn , respectively, and we will view both Zn
and Yn as subsets of X (which of course they are).
(iii) We will denote by Fn the map from M
+
(X) to itself given, for every f in M
+
(X),
(iv) We will denote by Qn the map from M
+
(X) to itself given, for every f in M
+
(X),
(v) We will say that a given f in M
+
(X) is Rn-invariant if f (x) = f (y), whenever
(x, y) ∈ Rn.
25
and for every x in X, by
Fn(f )
x
and for every x in X, by
Qn(f )
x
=
=
Py∈Rn(x)
0,
f (y),
if x ∈ Un,
otherwise.
f (y)ρn(y)ζn(y)−1,
if x ∈ Un,
Py∈Rn(x)
0,
otherwise.
6.10. Remarks.
(a) Since h0 = 0, we have that ρ0 = 1, and clearly also ζ0 = 1. Therefore Z0 = ∅.
(b) Notice that Fn(f ) and Qn(f ) may be alternatively defined as
Fn(f ) = ιn(cid:0)En(f ↾Un)(cid:1),
and Qn(f ) = ιn(Pρn(cid:0)f ↾Un)(cid:1).
For that reason Fn and Qn should be seen as natural extensions of En and Pρn to
M
(X), respectively. Notice also that
+
Qn(f ) = Fn(cid:0)f ιn(ρnζ −1
n )(cid:1),
∀ f ∈ M
+
(X).
( c ) Observe that the invariance of a function under an equivalence relation is a concept
usually considered when the relation is defined on the whole domain of said function.
However, the fact that Rn is an equivalence relation on Un, rather than on X, does not
prevent us from introducing the invariance notion expressed in (6.9.v). An example of
a function obeying this property is given by ιn(f ), where f is any function in M
(Un)
which is constant on each Rn-equivalence class.
+
Some elementary properties of these extended notions are in order.
6.11. Proposition. Given n ∈ , one has for all f, g ∈ M
+
(X), that
(i) Fn(cid:0)ιn(ρn)(cid:1) = ιn(ζn),
(ii) Qn(1) = 1Yn ,
(iii) Fn(f ) = Fn(1Unf ),
(iv) Qn(f ) = Qn(1Un f ) = Qn(1Ynf ),
(v) Qn(f ) = 1Yn Qn(f ),
(vi) Fn(f ) and Qn(f ) are Rn-invariant and vanish off Un,
(vii) if f is Rn-invariant and vanishes off Un, then f is Rk-invariant for every k ≤ n,
(viii) if g is Rn-invariant, then Fn(gf ) = gFn(f ), and Qn(gf ) = gQn(f ),
(ix) if A is an Rn-invariant subset of Un, then, as a subset of X, A is Rk-invariant for
every k ≤ n,
(x) Fn and Qn are σ-additive.
Proof. Left for the reader.
(cid:3)
Recalling that
and that
Zn = Zρn = {x ∈ Un : ζn(x) = ∞},
Yn = Un\Zρn = {x ∈ Un : ζn(x) < ∞},
we will now study certain relations among these sets, and we begin with the following
auxiliary result.
26
6.12. Lemma. For every 0 ≤ n ≤ m, and for every x in Um, there exists a subset
Λ ⊆ Rm(x), such that x ∈ Λ, and
Rm(x) = Gλ∈Λ
Rn(λ),
the square cup denoting disjoint union.
Proof. We first claim that if Cn and Cm are equivalence classes for Rn and Rm, respectively,
then
Cn ∩ Cm 6= ∅ ⇒ Cn ⊆ Cm.
To see this, choose z ∈ Cn ∩ Cm, and let y ∈ Cn. Then
(y, z) ∈ Rn ∩ (Un × Um)
(5.1.iii)
⊆ Rm,
so y ∈ Cm. This said, we see that the Rm-equivalence class of x splits as the union of
Rn-equivalence classes, whence the conclusion.
(cid:3)
The promissed relations among the Zn and the Yn are in order.
6.13. Proposition.
(a) For every m ≥ 1, and every x in Um, one has that
(b) If 0 ≤ n ≤ m, then Zn ∩ Um ⊆ Zm,
( c ) If 0 ≤ n ≤ m, then Ym ⊆ Yn.
Proof. In order to prove (a) write
Rm(x) = Gλ∈Λ
Rm−1(λ),
ekm(x)ζm−1(x) ≤ ζm(x).
where x ∈ Λ ⊆ Rm(x), by (6.12). So
ζm(x) = Xy∈Rm(x)
ehm(y) = Xλ∈Λ Xy∈Rm−1(λ)
ehm(y) = Xλ∈Λ Xy∈Rm−1(λ)
ehm−1(y)ekm(y) = · · ·
For every λ ∈ Λ, and y ∈ Rm−1(λ), notice that
(y, λ) ∈ Rm−1 ∩ (Um × Um)
(6.1.1)
⇒ km(y) = km(λ),
so the above equals
· · · = Xλ∈Λ
ekm(λ) Xy∈Rm−1(λ)
ehm−1(y) = Xλ∈Λ
ekm(λ)ζm−1(λ) ≥ ekm(x)ζm−1(x).
This proves (a). In order to prove (b), observe that under the hypothesis of (a) we have
that
ζm−1(x) = ∞ ⇒ ζm(x) = ∞,
27
from where we trivially deduce that Zm−1 ∩ Um ⊆ Zm. Assuming now that 0 ≤ n ≤ m,
we will prove (b) by induction on m − n.
In order to do this, notice that the case "m − n = 0" is immediate, while the case
"m − n = 1" has just been proved. When m − n > 1, we then have that
Zn ∩ Um ⊆ Zn ∩ Um−1 ∩ Um ⊆ Zm−1 ∩ Um ⊆ Zm,
taking care of (b). With respect to (c), we have
Ym = Um\Zm = Um ∩ Z c
m
(a)
⊆ Um ∩ (Zn ∩ Um)c ⊆ Um ∩ (Z c
n ∪ U c
m) =
= (Um ∩ Z c
n) ∪ (Um ∩ U c
m) = Um ∩ Z c
n ⊆ Un ∩ Z c
n = Un\Zn = Yn.
(cid:3)
There are many situations in the present context in which not necessarily increasing
sequences satisfy some increasing-like property as we look inside the appropriate Um. For
example, when n ≤ m, there is no comparisson between Rn and Rm, as sets, but when
we restrict Rn to Um, that is, when we consider Rn ∩ (Um × Um), we get a subset of Rm.
Similarly there is no comparission between Zn and Zm, but as seen above, Zn ∩ Um ⊆ Zm.
We next present some crucial properties of the Fn and the Qn.
6.14. Proposition. If 0 ≤ n ≤ m, and if f, g ∈ M
+
(X), then
(i) Fm(cid:0)f Fn(g)(cid:1) = Fm(cid:0)Fn(f )g(cid:1),
(ii) Qm(cid:0)Qn(f )(cid:1) = Qm(f ) = Qn(cid:0)Qm(f )(cid:1).
Proof. Addressing (i), since both sides vanish on X\Um, by definition, it is enough to
prove that they agree on Um. Given x in Um, we have
Fm(cid:0)f Fn(g)(cid:1) x
= Xy∈Rm(x)
f (y) Xz∈Rn(y)
g(z) = · · ·
(6.14.1)
Enploying (6.12) we write
Rm(x) = Gλ∈Λ
Rn(λ),
where Λ ⊆ Rm(x), so (6.14.1) equals
· · · = Xλ∈Λ Xy∈Rn(λ)
f (y) Xz∈Rn(y)
g(z) = Xλ∈Λ Xy∈Rn(λ) Xz∈Rn(λ)
f (y)g(z) =
g(z) Xy∈Rn(z)
= Xλ∈Λ Xz∈Rn(λ)
g(z) Xy∈Rn(z)
This proves (i). In order to prove (ii), recall that hm = Pm
f (y) = Xz∈Rm(x)
f (y) = Fm(cid:0)gFn(f )(cid:1) x
i=1 ki↾Um. So, defining
.
m
ℓ =
Xi=n+1
ki↾Um,
28
we then have that hm = hn + ℓ. We next claim that
(x, y) ∈ Rn ∩ (Um × Um) ⇒ ℓ(x) = ℓ(y).
This is because, for every i = n + 1, . . . , m, we have that
(x, y) ∈ Rn ∩ (Um × Um) ⊆ Rn ∩ (Un × Ui−1) ⊆ Ri−1,
so (x, y) also lies in Ri−1 ∩ (Ui × Ui), and this implies that ki(x) = ki(y), according to
(6.1.1). In other words, ℓ, or rather ιm(ℓ), is Rn-invariant.
To prove (ii) we start with its left-hand-side, taking full advantage of the abuse of
language announced in (6.9.i):
Qm(cid:0)Qn(f )(cid:1) (6.11.viii)
= Fm(cid:0)f ρnζ −1
= Fm(cid:0)Fn(f ρnζ −1
n eℓFn(ρn)(cid:1)ζ −1
n )ρm(cid:1)ζ −1
m = Fm(cid:0)f ρmζ −1
= Fm(cid:0)f ρnζ −1
n ζn(cid:1)ζ −1
(i)
m
m
n Fn(eℓρn)(cid:1)ζ −1
m
(6.11.viii)
=
(4.7)
= Fm(f 1Ynρm)ζ −1
m =
= Qm(f 1Yn)
= Qm(f 1Ym1Yn)
(6.11.iv)
(6.13)
= Qm(f 1Ym) = Qm(f ).
With respect to the second equality in (ii), we have
Qn(cid:0)Qm(f )(cid:1) = Qn(cid:0)1Qm(f )(cid:1)
(6.11.vi,vii&viii)
=
Qn(1)Qm(f )
(6.11.ii)
= 1YnQm(f )
(6.11.v)
=
= 1Yn1YmQm(f )
(6.13)
= 1YmQm(f ) = Qm(f ).
(cid:3)
We next present some useful properties of the dual operators Q∗
n.
n(µ), for every Rn-invariant function g in M
n(gµ) = gQ∗
n(µ) = Q∗
n(µ)↾Un = P ∗
6.15. Proposition. Given n in , and given any finite measure µ on X, one has that
(i) Q∗
(ii) Q∗
(iii) Q∗
(iv) if A is an Rn-invariant, Borel subset of X, then Q∗
(v) if m is another integer with n ≤ m, then Q∗
n(µ)(A) = µ(A ∩ Yn),
n(1Ynµ) = 1Yn Q∗
ρn (µ↾Un ),
n(µ),
(X),
+
m(µ) = Q∗
m(cid:0)Q∗
n(µ)(cid:1) = Q∗
n(cid:0)Q∗
m(µ)(cid:1).
Proof. The first point follows easily from (6.11.viii).
(ii): Pick any f in M
+
(X). Then
ZX
f dQ∗
Qn(f ) dµ
(6.11.v)
= ZX
Qn(f )1Yn dµ =
n(µ) = ZX
= ZX
Qn(f ) d1Ynµ = ZX
f dQ∗
n(1Ynµ),
29
proving the first identity in (ii). As for the second one, we have
ZX
f dQ∗
n(µ) = ZX
Qn(f ) dµ
(6.11.iv)
= ZX
Qn(1Yn f ) dµ = ZX
f 1Yn dQ∗
n(µ),
taking care of (ii).
(iii): Given f ∈ M
+
(Un), we have
ZUn
f dP ∗
ρn (µ↾Un) = ZUn
Pρn (f ) dµ↾Un = ZX
ιn(cid:0)Pρn (f )(cid:1) dµ =
= ZX
(iv): We have
ιn(cid:0)Pρn (ιn(f )↾Un)(cid:1) dµ
= ZX
ιn(f ) dQ∗
n(µ) = ZUn
(6.10.b)
= ZX
Qn(cid:0)ιn(f )(cid:1) dµ =
f dQ∗
n(µ)↾Un,
1A dQ∗
Qn(1A) dµ
(6.11.viii)
=
Q∗
n(µ)(A) = ZX
= ZX
(6.11.viii)
n(µ) = ZX
= ZX
(6.11.ii)
1AQn(1) dµ
1A1Yn dµ = µ(A ∩ Yn).
(v): This is a direct consequence of (6.14.ii).
(cid:3)
We may now state an important quasi-invariance condition for measures on X.
6.16. Theorem. Let µ be a finite measure on X. Then µ is D-quasi-invariant if and
only if, for every n ∈ , one has that
Q∗
n(µ) = 1Un µ.
Proof. We have already seen in (6.7) that µ is D-quasi-invariant if and only each if µ↾Un is
Dn-quasi-invariant. By (4.15) this is in turn equivalent to saying that P ∗
ρn (µ↾Un ) = µ↾Un,
but in view of (6.15.iii), this is now the same as
Q∗
n(µ)↾Un = µ↾Un.
Since we know that Q∗
n(µ) vanishes on X\Un by (6.15.ii), the proof is concluded.
(cid:3)
30
7. Existence of quasi-invariant measures.
Having characterized quasi-invariant measures in a concise way in (6.16), we now discuss
their existence. This is a multi faceted question manifesting itself in different ways on
different parts of X. It is therefore convenient to break X down into simpler pieces, so we
shall henceforth consider the following subsets
Vn := Un\Un+1,
∀ n ∈ ,
V∞ := Tn∈
Un, Z := Sn∈
Zn,
Wn := Vn\Z,
∀ n ∈ ∪ {∞},
which we represent in the following diagram. Please note that each Zn should be thought
of as the largest shaded rectangle possessing the indicated lower-left-hand corner.
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.......................................................................................
...............................................................
................................................
........................................
....................................
..................................
......................................................................................
...............................................................
...............................................
.......................................
...................................
..................................
.........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...............................................................
................................................
........................................
....................................
..................................
...............................................................
...............................................
.......................................
...................................
..................................
...........................................................................................................................................................................................................................................................................................................................................................................................
................................................
........................................
....................................
..................................
...............................................
.......................................
...................................
..................................
................................................
........................................
....................................
..................................
...............................................
.......................................
...................................
..................................
................................................
........................................
....................................
..................................
...............................................
.......................................
...................................
..................................
................................................
........................................
....................................
..................................
...............................................
.......................................
...................................
..................................
................................................
........................................
....................................
..................................
...............................................
.......................................
...................................
..................................
................................................
........................................
....................................
..................................
...............................................
.......................................
...................................
..................................
................................................
........................................
....................................
..................................
...............................................
.......................................
...................................
..................................
................................................
........................................
....................................
..................................
............................................................................................................................................................................................................................................................................................
.......................................
...................................
..................................
........................................
....................................
..................................
.......................................
...................................
..................................
........................................
....................................
..................................
.......................................
...................................
..................................
........................................
....................................
..................................
.......................................
...................................
..................................
........................................
....................................
..................................
.............................................................................................................................................................................................................................................
...................................
..................................
....................................
..................................
...................................
..................................
....................................
..................................
.....................................................................................................................................................................................................................
..................................
..................................
............................................................................................................................................................................................................
................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
W3
W∞
Z3 ր
W2
Z2 ր
W1
Z1 ր
W0
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
V0
V1
V2
V3
V∞
Diagram (7.1)
The all important sets Un's may then be described as
Un = Gn≤m≤∞
Vm,
∀ n ∈ ,
the square cup denoting disjoint union. In particular
X = U0 = Gn∈∪{∞}
Vn,
31
whence also
X = Z ⊔ Gn∈∪{∞}
Wn.
(7.2)
7.3. Proposition. The following sets are R-invariant:
(i) Un, for all n ∈ ,
(ii) Vn and Wn, for all n ∈ ∪ {∞},
(iii) Z.
Proof. Let us prove first that Um is R-invariant, for every m. For this suppose that
(x, y) ∈ R, and y ∈ Um, so we may pick some n such that (x, y) ∈ Rn. Assuming initially
that n ≤ m, we have that
(x, y) ∈ Rn ∩ (Un × Um)
(5.1.iii)
⊆ Rm ⊆ Um × Um,
proving that x ∈ Um, as desired. Assuming now that m ≤ n, the conclusion comes even
easier because
(x, y) ∈ Rn ⊆ Un × Un ⊆ Um × Um,
so again we have that x ∈ Um.
Let us next prove that Z is R-invariant. So we pick (x, y) ∈ R, with y ∈ Z, whence
there are n and m such that (x, y) ∈ Rn, and y ∈ Zm. Assuming initially that n ≤ m, we
have that
(x, y) ∈ Rn ∩ (Un × Um)
⊆ Rm.
(5.1.iii)
Since ζm is known to be Rm-invariant on Um, it follows that ζm(x) = ζm(y) = ∞, whence
x ∈ Zm, as required.
Assuming now that m ≤ n, notice that
y ∈ Zm ∩ Un
(6.13.b)
⊆ Zn,
so the Rn-invariance of ζn implies that
∞ = ζn(y) = ζn(x),
and we conclude that x ∈ Zn ⊆ Z.
Since the R-invariant sets clearly form a complete Boolean algebra, the remaining
(cid:3)
statements follow.
The following result further streamlines the various quasi-invariance conditions and it
will be instrumental in the study of the existence question.
7.4. Theorem. Let µ be a finite measure on X, and for every k ∈ ∪ {∞}, set µk =
1Wk µ. Then µ is D-quasi-invariant if and only if all of the following conditions hold:
(i) µ(Z) = 0,
(ii) Q∗
(iii) Q∗
k(µk) = µk, for every k ≥ 1,
i (µ∞) = µ∞, for every i ≥ 1.
32
Proof. Assuming that µ is D-quasi-invariant, we have by (6.7) that µ↾Uk is Dk-quasi-
invariant, for every k ∈ , whence we deduce from (4.11) that
0 = µ↾Uk (Zk) = µ(Zk),
from where (i) follows.
Given k ≥ 1, recall that Wk is R-invariant, so µk is D-quasi-invariant by (6.6). In
then follows from (6.16), that6
Q∗
k(µk) = 1Uk µk = 1Uk 1Wk µ = 1Wk µ = µk,
proving (ii).
By the reasoning in the first sentence of the paragraph above, we also have that µ∞
is D-quasi-invariant. So, again by (6.16), we have for all k ∈ , that
Q∗
k(µ∞) = 1Uk µ∞ = 1Uk 1W∞ µ = 1W∞ µ = µ∞,
whence (iii).
Conversely, assuming that µ satisfies (i -- iii), we will initially prove that µk is D-quasi-
invariant for all k ∈ ∪ {∞}, via the characterization provided by (6.16). We must
therefore prove that
Q∗
n(µk) = 1Un µk,
∀ n ∈ .
(7.4.1)
When k = ∞, this is provided for by (iii), and the fact that W∞ ⊆ Un, so it remains to
prove (7.4.1) for k ∈ . Assuming first that k < n, we have that
Q∗
n(µk)
(6.15.ii)
= Q∗
n(1Yn µk) = Q∗
n(1Yn1Wk µk) = 0,
because Yn ∩ Wk ⊆ Un ∩ Wk = ∅, and likewise 1Un µk = 0. Assuming now that n ≤ k, we
have
Q∗
n(µk)
(ii)
= Q∗
n(cid:0)Q∗
k(µk)(cid:1) (6.15.i)
= Q∗
k(µk)
(ii)
= µk = 1Uk µk.
Note that the above use of (ii) is not quite correct because it has only been assumed
for k ≥ 1. However, when k = 0, that condition holds trivially because Q∗
0 is the identity
transformation. This concludes the verification of (7.4.1), hence showing that µk is D-
quasi-invariant.
Employing (7.2) we have that
µ = 1Zµ + Pk∈∪{∞}
1Wk µ
(i)
= Pk∈∪{∞}
µk,
which is seen to be a D-quasi-invariant measure since each factor has this property, which
in turn is clearly preserved under sums.
(cid:3)
6 Of course Q∗
n(µk) = 1Un µk, for all n, but so far we only need the case n = k.
33
Observe that any measure µ on X which assigns zero mass to Z satisfies
µ = Pk∈∪{∞}
µk,
(7.5)
where each measure µk lives in Wk (meaning that µk(X\Wk) = 0), namely µk = 1Wk µ.
Conversely, if we are given a collection {µk}k∈∪{∞} of measures on X, such that µk lives
in Wk, then (7.5) may be used to define a measure µ on X which assigns measure zero to
Z. In other words there is a one-to-one correspondence between the µ's and the collections
{µk}k∈∪{∞}. This said, observe that the conditions characterizing a D-quasi-invariant
measure in (7.4) consist of independent conditions on each "coordinate" µk.
In particular, if we fix any k in ∪ {∞}, and if we pick any measure µk living on Wk,
and satisfying the corresponding condition, namely
(a) condition (7.4.ii), in case k ≥ 1,
(b) condition (7.4.iii), in case k = ∞, or
( c ) no condition at all, when k = 0,
then µk, itself, is a D-quasi-invariant measure. En passant, we note that any measure
living in W0 is automatically D-quasi-invariant. In any case, the existence question for
quasi-invariant measures should be split into separate questions regarding the existence of
quasi-invariant measures living in each Wk. In case k is finite, this question has a very
simple answer:
7.6. Proposition. Given an integer n, with 0 ≤ n < ∞, there exists a D-quasi-invariant
probability measure living in Wn if and only if Wn is nonempty.
Proof. Ignoring the blatantly obvious "only if" part, we deal only with the "if" part.
Assuming that Wn is nonempty, choose any probability measure ν living in Wn, e.g. a
Dirac measure based on any point chosen in Wn. Setting µ = Q∗
n(ν), notice that
µ(X) = ZX
1 dQ∗
n(ν) = ZX
Qn(1) dν
(6.11.ii)
= ZX
1Yn dν = ν(Yn) = 1,
where the last equality is a consequence of the fact that
Wn = Vn\Z ⊆ Un\Zn = Yn.
This shows that µ is a probability measure.
Since Wn is Rn-invariant by (7.3), and since ν lives in Wn, then µ also lives in Wn by
(6.15.iv). For that reason we have that µk := 1Wk µ = δnkµ, for every k in . So, in order
to prove that µ is D-quasi-invariant, we need only check (7.4.ii) for k = n, given that all
of the other conditions hold trivially. To do this we compute
Q∗
n(µ) = Q∗
n(ν) = µ,
n(cid:0)Q∗
n(ν)(cid:1) (6.15.i)
= Q∗
proving that µ is D-quasi-invariant.
(cid:3)
34
The existence question for quasi-invariant measures living in W∞ is much more subtle,
not least because (7.4.iii) involves infinitely many conditions. The following result might
have excessively rigid hypotheses, but it is the best existence result we can offer in this
generality:
7.7. Theorem. Suppose that W∞ contains a compact, R-invariant, nonempty subset K.
Suppose also that ζ −1
n ↾K is continuous7 for every n ∈ . Then there exists a D-quasi-
invariant probability measure living in K.
Proof. We begin by observing that W∞ ⊆ Yn, for every n ∈ , because
W∞ = V∞\Z ⊆ Un\Zn = Yn.
Denote by P (X, K) the set of all probability measures on X living in K. This is
clearly a nonempty set given that it contains any Dirac measure δx0 with x0 ∈ K. We
claim that, for every n in , one has that
In fact, if ν is in P (X, K), then
Q∗
n(cid:0)P (X, K)(cid:1) ⊆ P (X, K).
Q∗
n(ν)(X)
(6.15.iv)
= ν(Yn) = ν(Yn ∩ K) = ν(K) = 1,
so we see that Q∗
n(ν) is a probability measure. Moreover
Q∗
n(ν)(X\K)
(6.15.iv)
= ν(cid:0)(X\K) ∩ Yn(cid:1) ≤ ν(X\K) = 0,
n(ν) lives in K. Identifying P (X, K) with the set P (K) of all probability measures
so Q∗
on K via the correspondence
µ ∈ P (X, K) 7→ µ↾K ∈ P (K),
we claim that, for every g in C(K), the function
µ ∈ P (X, K) 7→ ZK
g dQ∗
n(µ) ∈ C
is continuous on P (K) relative to the topology induced by the weak* topology of the dual
of C(K).
To prove it we first use Tietze's extension Theorem to produce a continuous function
f , defined on the whole of X, and whose restriction to X coincides with g. We further
use Uhrysohn's Lemma to obtain a continuous function ϕ on X such that ϕ↾K = 1, and
7 Please note that by saying that ζ−1
n ↾K is continuous we mean that it belongs to C(K). This should
not be confused with the much more stringent requirement that ζ−1
n be continuous at all points of K.
35
whose support is compact and contained in Un. Replacing f by ϕf we may then assume
that the support of our originally chosen f is compact and contained in Un. We then have
ZK
g dQ∗
f dQ∗
n(µ) = ZX
Pρn (f ↾Un) dµ↾Un = ZUn
n(µ) = ZX
= ZUn
= ZUn
(3.4.ii)
En(f ↾Unρn)ζ −1
n dµ↾Un = ZK
En(f ↾Unρn)ζ −1
n dµ.
Qn(f ) dµ = ZX
in(cid:0)Pρn (f ↾Un)(cid:1) dµ =
En(f ↾Unρnζ −1
n ) dµ↾Un
(3.4.ii)
=
Recaling that En(f ↾Unρn) is continuous by (3.3), and that ζ −1
n
hypothesis, the claim follows.
is continuous on K by
We will next prove that the fixed points of Q∗
n in P (X, K) form a closed subset. To
see this let µ be any measure in P (X, K). By the uniqueness part of the Riesz-Markov, to
say that Q∗
n(µ) = µ is to say that
ZK
g dµ = ZK
g dQ∗
n(µ),
∀ g ∈ C(K).
(7.7.1)
Viewed as functions of the variable µ, both sides of the above expression are now
known to be weak*-continuous, so the set of solutions to this system on equations, as g
ranges in C(K), form a closed set, hence proving that the set of fixed points of Q∗
n in
P (X, K) is indeed weak*-closed.
Choosing any ν in P (X, K), we define
µn = Q∗
n(ν),
for every n in . Using Alaoglu's Theorem we may then find a limit point, say
µ∞ ∈ P (X, K),
for the sequence {µn}n. Given two integers n and m, with 0 ≤ n ≤ m, observe that
Q∗
n(µm) = Q∗
n(cid:0)Q∗
m(ν)(cid:1) (6.15.v)
= Q∗
m(ν) = µm,
so we see that all but finitely many µm's are fixed points for Q∗
µ∞, thanks to the closedness of the set fixed points just proved.
n, hence the same holds for
Observing that µ∞ lives in K, and that K ⊆ Un, for every n, we then have that
Q∗
n(µ) = µ = 1Un µ,
so µ∞ is D-quasi-invariant by (6.16).
(cid:3)
36
8. Renault-Deaconu groupoids.
In this section we will describe an example of gap coming from generalized Renault-
Deaconu groupoids [6]. This is in fact the main motivation for introducing and studying
gaps.
We will henceforth fix a space X satisfying (2.1) and we will suppose we are given an
open subset U ⊆ X, and a map
σ : U → X,
which we will assume to be a local homeomorphism.
Let U0 = X, and for each n ≥ 1, put
Un = (cid:8)x ∈ U : σ(x) ∈ Un−1(cid:9).
It is then easy to see that Un is effectively the domain of σn, and that the Un form a
collection of open subsets of X satisfying (5.1.i).
The generalized Renault-Deaconu groupoid Gσ associated to σ was defined by Renault
in [6], and it consists of all triples (x, n, y) in X × Z × X, such that there exist k, l ∈ ,
satisfying n = k − l, x ∈ Uk, y ∈ Ul, and σk(x) = σl(y).
The multiplication of two elements (x, n, y) and (z, m, w) in Gσ is defined only when
y = z, in which case the product is set to be (x, n + m, w). The topology of Gσ is generated
by the collection of subsets
Wk,l,A,B = (cid:8)(x, k − l, y) : x ∈ A, y ∈ B, σk(x) = σl(y)(cid:9),
for all k, l ∈ , and all open subsets A ⊆ Uk, and B ⊆ Ul. With this structure Gσ becomes
an ´etale groupoid and we refer the reader to [6] for more on Gσ.
Let us next consider, for each n ∈ , the subset Rn of Un × Un defined by
Rn = (cid:8)(x, y) ∈ Un × Un : σn(x) = σn(y)(cid:9).
Recalling that σ is a local homeomorphism, it is clear that σn is a local homeomorphism
from Un to X, so Rn is easily seen to be a proper equivalence relation on Un.
8.1. Proposition. One has that
R = (cid:16){Un}n∈, {Rn}n∈(cid:17),
is a gap on X.
Proof. All we need at this point is to verify (5.1.iii). So, suppose that n ≤ m, and let
(x, y) ∈ Rn ∩ (Un × Um). The first conclusion to be drawn is that both x and y lie in Un,
and that σn(x) = σn(y). Besides, y lies in Um, so y is also in the domain of σm. This
implies in particular that σn(y) lies in the domain of σm−n, so the same holds for σn(x).
Consequently
thus showing that (x, y) ∈ Rm.
σm(y) = σm−n(cid:0)σn(y)(cid:1) = σm−n(cid:0)σn(x)(cid:1) = σm(x),
(cid:3)
37
The Renault-Deaconu groupoid admits a continuous cocycle
δ : (x, n, y) ∈ Gσ 7→ n ∈ Z,
whose kernel is therefore an open subgroupoid, which is clearly isomorphic and homeo-
morphic to the gap groupoid R = Sn Rn, via the homeomorphism sending each (x, y) in
R to (x, 0, y).
In order to speak of quasi-invariant measures on Gσ, we need to introduce a real-valued
cocycle. Recall from [1] that if h : U → R is a continuous function, often thought of as a
potential function, one may define
b(x, n − m, y) =
n−1
Pi=0
h(cid:0)σi(x)(cid:1) −
m−1
Pi=0
h(cid:0)σi(y)(cid:1),
(8.2)
whenever σn(x) = σm(y), obtaining in this way a well defined continuous cocycle on Gσ,
taking values in the additive group of real numbers.
The restriction of c to the subgroupoid R is then evidently a continuous cocycle on R,
but we would instead like to introduce it from a different point of view, in line with (6.4).
For each n ≥ 1, let us define
kn(x) = h(cid:0)σn−1(x)(cid:1),
∀ x ∈ Un.
Notice that, for x in Un, the largest integer i for which the expression h(cid:0)σi(x)(cid:1) is guaranteed
to be well defined is i = n − 1. This is because Un = dom(σn), so σn(x) is well defined,
which in turn implies that σn−1(x) is in U , and hence it does makes sense to apply h to
σn−1(x). However there is no reason for σn(x) to lie in U , so h(σn(x)) may not be well
defined.
8.3. Proposition. The collection {kn}n≥1 defined above is a potential for R, in the sense
of (6.1).
Proof. To verify (6.1.1), let n ≥ 1 and choose (x, y) ∈ Rn−1 ∩ (Un × Un). Then σn−1(x) =
σn−1(y), whence kn(x) = kn(y), as needed.
(cid:3)
The associated cocycle is then given on any (x, y) ∈ Rn, by
c(x, y) = cn(x, y) = hn(x) − hn(y) =
kn(x) − kn(y) =
n
Pi=1
=
n
Pi=1
h(cid:0)σi−1(x)(cid:1) − h(cid:0)σi−1(y)(cid:1) =
n−1
Pi=0
h(cid:0)σi(x)(cid:1) − h(cid:0)σi(y)(cid:1),
(8.4)
which happens to be the restriction to R of the cocycle b defined in (8.2).
Observe that both the unit space of Gσ and that of R may be naturally identifyied with
X. So a given finite measure µ on X may be tested for quasi-invariance either relatively
to the cocycle eb on Gσ, or to the cocycle ec on R.
38
8.5. Definition. Let µ be a finite measure on X. We will say that µ is
(i) a conformal measure when it is quasi-invariant relatively to the cocycle eb on Gσ,
(ii) a DLR measure when it is quasi-invariant relatively to the cocycle ec on R.
Since R is a subgroupoid of Gσ, and since ec is the restriction of eb to R, it is immediate
that:
8.6. Proposition. Every conformal measure on X is a DLR measure.
9. Eigenmeasures.
As in the previous section we let X be a space satisfying (2.1), U ⊆ X be an open set, and
σ : U → X be a local homeomorphism. Our goal here is to show that every eigenmeasure
for Ruelle's operator is a DLR measure.
+
For each f in M
(U ), and for every x in X, define
so that L becomes a map
L(f )
x
= Xt∈σ−1(x)
f (t),
L : M
+
(U ) → M
+
(X).
Regarding the expression defining L(f ) above, observe that if x is not in σ(U ), then
σ−1(x) is the empty set, whence there are no summands at all, hence the sum turns out
to be zero. In other words, L(f ) vanishes outside σ(U ).
We also consider the operator
α : M
+
(X) → M
+
(U ),
given by α(f ) = f ◦ σ. We finally define
by
E : M
+
(U ) → M
+
(U ),
E(f )
x
= Xσ(t)=σ(x)
f (t),
∀ x ∈ U.
For f in M
+
(U ) and any x in U , observe that,
α(cid:0)L(f )(cid:1) x
= L(f )
σ(x)
= Xt∈σ−1(σ(x))
f (t) = Xσ(t)=σ(x)
f (t) = E(f )
,
x
so we see that
α ◦ L = E.
39
(9.1)
Anoter useful property is
L(α(g)f ) = gL(f ),
∀ g ∈ M
+
(X),
∀ f ∈ M
+
(U ),
(9.2)
which the reader will have no difficulty in checking.
We shall also fix a continuous function
ρ : U → R,
satisfying ρ(x) > 0, for all x in U . Here ρ is supposed to play the role of eh, where h is
the function used for creating the cocycle b in (8.2). The operator
Lρ : M
+
(U ) → M
+
(X),
defined by the formula Lρ(f ) = L(ρf ), is then the analogue of Ruelle's operator in the
present situation.
9.3. Lemma. Suppose that
(i) µ is a measure on X,
(ii) λ is a nonzero scalar,
such that
Then
ZX
L(ρf ) dµ = λZU
f dµ,
∀ f ∈ M
+
(U ).
ZU
E(ρ)f dµ = ZU
E(ρf ) dµ,
∀ f ∈ M
+
(U ).
(9.3.1)
(9.3.2)
Proof. Given g in M
+
(X), plug f = α(g) in (9.3.1), to get
ZU
α(g) dµ = λ−1ZX
L(cid:0)α(g)ρ(cid:1) dµ
(9.2)
= λ−1ZX
gL(ρ) dµ.
(9.3.3)
Working from the right-hand-side of (9.3.2), we have
ZU
E(ρf ) dµ
(9.1)
= ZU
(9.3.3)
= λ−1ZX
L(ρf )L(ρ) dµ
(9.2)
=
(9.2)
= λ−1ZX
L(cid:0)ρf α(cid:0)L(ρ)(cid:1)(cid:1) dµ
L(cid:0)ρf E(ρ)(cid:1) dµ
(9.3.1)
= ZU
f E(ρ) dµ.
(cid:3)
α(cid:0)L(ρf )(cid:1) dµ
= λ−1ZX
(9.1)
40
For each n ≥ 1, let Un be the domain of σn, so that the map
σn : Un → X,
is a local homeomorphism so it could be used in place of σ in order to define all of the
above ingredients. To be precise these are:
• Ln : M
+
(Un) → M
+
(X), given by Ln(f )
x
= Xt∈σ−n(x)
f (t),
• αn : M
+
(X) → M
+
(Un), given by
αn(f ) = f ◦ σn,
• En : M
+
(Un) → M
+
(Un), given by En(f )
x
= Xσn(t)=σn(x)
f (t).
We shall also let
ρn = ρα(cid:0)ρ(cid:1)α2(ρ) · · · αn−1(ρ).
Regarding the product defining ρn above, observe that each αi(ρ) is a member of
(Ui+1), so they all may be restricted to Un before the multiplication is performed,
+
M
resulting of course in a member of M
+
(Un).
9.4. Proposition. For every n, m ≥ 0, one has that
+
(i) Ln(cid:0)M
(Un+m)(cid:1) ⊆ M
(ii) Lm ◦ Ln = Ln+m.
+
(Um), and
+
Proof. Given f in M
= 0. Arguing
by contradiction, suppose this is not so. Therefore there exists at least one t ∈ σ−n(x),
such that f (t) 6= 0, so it follows that
(Un+m), and x in X\Um, we must prove that Ln(f )
x
x = σn(t) ∈ σn(Un+m) ⊆ Um,
a contradiction, proving (i). In order to prove (ii), pick f in M
then have
+
(Un+m), and x in X. We
Lm(cid:0)Ln(f )(cid:1) x
= Xt∈σ−m(x)
Ln(f )
t
= Xt∈σ−m(x) Xs∈σ−n(t)
f (s) = Xt∈σ−n−m(x)
f (s) = Ln+m(f )
.
x
(cid:3)
9.5. Lemma. Let µ be a measure on X satisfying (9.3.1). Then
ZX
Ln(ρnf ) dµ = λnZUn
f dµ,
∀ f ∈ M
+
(Un).
Proof. For f in M
+
(Un), we have by induction that
Ln−1(cid:0)L(ρα(ρn−1)f )(cid:1) dµ = ZX
Ln−1(cid:0)ρn−1L(ρf )(cid:1) dµ =
(9.3.1)
= λnZU
f dµ = λnZUn
f dµ.
(cid:3)
Ln(ρnf ) dµ = ZX
ZX
= λn−1ZUn−1
L(ρf ) dµ = λn−1ZU
L(ρf ) dµ
41
9.6. Corollary. Let µ be a measure on X satisfying (9.3.1). Then for any n in , one
has that
En(ρnf ) dµ,
∀ f ∈ M
+
(Un).
ZUn
En(ρn)f dµ = ZUn
Proof. Follows immediately by applying (9.3) to σn and ρn.
(cid:3)
Given a continuous function h : U → R, let ρ = eh, and recall from (8.5) that a finite
measure µ on X is said to be a DLR measure for h if it is quasi-invariant relative to the
cocycle ec on R, where c is given in terms of h by (8.4). The cocycle ec is in fact a common
extension of the cocycles ecn defined on each Rn by
ecn(x,y) = exp(cid:16) n−1
Pi=0
h(cid:0)σi(x)(cid:1) − h(cid:0)σi(y)(cid:1)(cid:17) =
ρ(x)ρ(σ(x)) · · · ρ(σn−1(x))
ρ(y)ρ(σ(y)) · · · ρ(σn−1(y))
=
ρn(x)
ρn(y)
.
We then see that, if µ is a finite measure satisfying the conclusions of (9.6), then µ↾Un
satisfies (4.11.iii) relative to ρn, so it is quasi-invariant for ecn by (4.11). Employing (6.7)
it then follows that µ is ec-quasi-invariant, hence a DLR measure. Summarizing we obtain
the following:
9.7. Theorem. Let X be a locally compact metric space, U be an open subset of X, and
σ : U → X be a local homeomorphism. Choosing any continuous potential h : U → R,
let ρ = eh. Then any finite measure on X which is an eigenvalue for the corresponding
Ruelle operator, meaning that it satisfies (9.3.1) with a nonzero eigenvalue λ, is necessarily
a DLR measure.
Proof. Follows immediately by applying (9.6) to σn and ρn.
(cid:3)
It should be noted that Corollary (9.6) may be seen as a generalization of Theorem
(9.7) to measures which are not necessarily finite, as long as we redefine the notion of DLR
measures as those which satisfy the conclusions of (9.6). However, since our theory of DLR
measures was developed only for finite measures, the various equivalent conditions for a
measure to be DLR have not been proved here for infinite measures.
A. Appendix: Elementary remarks about Measure Theory.
In the final two sections of this work we make some elementary remarks, mainly to fix our
notation. By a measurable space we shall mean a pair (X, B), consisting of a nonempty
set X, and a σ-algebra B of subsets of X.
A.1. Definition. Given a measurable space (X, B), we shall denote by M
of all B-measurable functions
+
(X, B) the set
f : X → [0, +∞].
A.2. Definition. Given measurable spaces (X, B) and (Y, C), a positively homogeneous
map
T : M
+
(X, B) → M
+
(Y, C)
42
is said to be σ-additive if for any sequence {fn}n in M
+
(X, B), we have that
T(cid:16)
∞
Xn=1
fn(cid:17) =
∞
Xn=1
T (fn),
all sums being interpreted as pointwise sums.
A.3. Proposition. Let (X, B) and (Y, C) be measurable spaces, and let
T : M
+
(X, B) → M
+
(Y, C)
be a σ-additive map. Then:
(i) If {fn}n is a non-decreasing sequence of functions in M
+
(X, B), then
T(cid:0) lim
n→∞
fn(cid:1) = lim
n→∞
T (fn),
all limits being interpreted as pointwise limits.
(ii) For every measure8 ν on (Y, C), there exists a measure T ∗(ν) on (X, B), such that
ZX
f dT ∗(ν) = ZY
T (f ) dν,
∀ f ∈ M
+
(X, B).
Proof. Regarding (i), define g1 = f1, and for each n ≥ 2, define
gn = fn(x) − fn−1(x),
∀ x ∈ X.
Observe that, in case fn−1(x) = ∞, then necessarily also fn(x) = ∞, in which case we
adopt the convention according to which gn(x) = 0. The functions gn so defined are
then B-measurable and non-negative, and we have that fn−1 + gn = fn. Consequently
fn = Pn
i=1 gi, so
T(cid:0) lim
n→∞
fn(cid:1) = T(cid:16) ∞
Pi=1
gi(cid:17) =
∞
Pi=1
T (gi) = lim
n→∞
n
Pi=1
T (gi) = lim
n→∞
T (fn),
proving (i). In order to prove (ii), for every f in M
+
(X, B), define
I(f ) = ZX
T (f ) dν.
We then claim that, given any sequence {fn}n in M
+
(X, B), one has that
∞
Pn=1
I(cid:0)
fn(cid:1) =
∞
Pn=1
I(fn).
8 All measures in this work are assumed to be σ-additive and positive.
43
This is a consequence of the σ-additivity of T and the monotone convergence Theorem, as
follows:
∞
Pn=1
fn(cid:1) = ZX
I(cid:0)
∞
Pn=1
fn(cid:1) dν = ZX
∞
Pn=1
T(cid:0)
T (fn) dν =
∞
Pn=1ZX
T (fn) dν =
I(fn),
∞
Pn=1
thus proving the claim. Defining µ(E) = I(1E), for every E in B, it then follows that µ is
a σ-additive measure on X, and clearly
I(f ) = ZX
f dµ,
(A.3.1)
for every simple function f in M
+
(X, B).
We then claim that (A.3.1) holds for every f in M
(X, B). To see this, recall that
every such f may be written as the pointwise limit of a non-decreasing sequence {sn}n of
simple functions in M
+
+
(X, B) [8: Section 18.1]. So
I(f ) = ZX
T(cid:0) lim
n→∞
sn(cid:1) dν
(i)
= ZX
lim
n→∞
T (sn) dν
(*)
= lim
n→∞ZX
T (sn) dν =
= lim
n→∞
I(sn) = lim
n→∞ZX
sn dµ = ZX
f dµ.
In time, we observe that (∗), above, is justified by the monotone convergence Theorem and
the easily proven fact that {T (sn)}n is a non-decreasing sequence.
This proves our claim and we then have for every f in M
+
(X, B) that
ZX
f dµ = I(f ) = ZX
T (f ) dν,
so it is enough to put T ∗(ν) := µ.
(cid:3)
Before closing this section let us comment on two related notions which will be used
often.
A.4. Remark. In this work we shall consider two similar, but inequivalent, ways of
restricting a measure on a space X to a Borel subset Y ⊆ X. The first one, officially called
the restriction of µ to Y , consists of the measure denoted µ↾Y , defined on the σ-algebra
B(Y ) of all Borel subsets of Y , by
µ↾Y (A) = µ(A),
∀ A ∈ B(Y ).
The second one, which we will denote by 1Y µ, is nothing but the well known measure
obtained by multiplying the measurable function 1Y by the measure µ. Recall that the
domain of 1Y µ is still B(X), as opposed to B(Y ), and
for all A ∈ B(X).
1Y µ(A) = ZA
1Y dµ = µ(A ∩ Y ),
44
B. Appendix: Elementary remarks about Measure Theory and Topology.
B.1. Proposition. Every space X satisfying (2.1) is σ-compact.
Proof. For every x in X, let Ux be a relatively compact, open neighborhood of x. By
reducing Ux a bit we may assume that it belongs to some previously chosen countable base
B of open sets for X. It then follows that
{U x : x ∈ X}
is a family of compact sets covering X. This family is countable (even though it might be
indexed on an uncountable set) because the Ux's belong to the countable base B.
(cid:3)
The reason for restricting ourselves to (2.1) is to simplify some aspects of measure
theory. In this short section we will explain exactly what we mean by this.
Recall that the Borel σ-algebra, denoted B(X), is the σ-algebra of subsets of X gen-
erated by the closed subsets. On the other hand, the Baire σ-algebra [8: 21.6], denoted
Ba(X), is the smallest σ-algebra of subsets of X for which the functions in Cc(X) are
measurable.
If one is interested in the measurability properties of none other that continuous,
compactly supported functions, the Baire σ-algebra is the most appropriate one to be
considered. The Baire σ-algebra is known to be generated by the compact Gδ subsets of
X [8: Theorem 21.21].
The first advantage of working with (2.1) is as follows:
B.2. Lemma. (See also [8: Theorem 21.20]) Suppose that X is as in (2.1). Then the
Baire and Borel σ-algebras on X coincide.
Proof. It is clear that Ba(X) ⊆ B(X), so we need only wory about the reverse inclusion.
In order to do so it is enough to prove that every closed set F ⊆ X is Baire-measurable.
Temporarily assuming that F is moreover compact, pick any compatible metric d on X
and define
Un = {x ∈ X : d(x, F ) < 1/n}.
Each Un is then open, and F = Tn Un, so we see that F is a compact Gδ, hence Baire-
measurable.
If F is now any closed set, use the fact that X is σ-compact to choose a countable
family {Kn}n of compact subsets of X such that X = Sn Kn. Then F ∩ Kn is a compact
set, and hence Baire-measurable by the first part of this proof. Since F = Sn F ∩ Kn, it
follows that F is Baire-measurable.
(cid:3)
A measure µ defined on B(X) is caled a Borel measure when it assigns finite measure
to every compact set [8: Section 21.3]. If µ is instead defined on Ba(X), it is called a Baire
measure provided it is finite on compact (Baire-measurable) sets [8: Section 21.6].
One of the main applications of Borel measures in Analysis is the Riesz-Markov The-
orem [8: Section 21.6] which states that each positive linear functional on Cc(X) is given
45
by the integration against a unique regular Borel measure (a regular Borel measure is also
called a Radon measure).
Another reason to work under (2.1) is that, in this case, every Baire measure is regular
[8: Theorem 21.27]. Since we now know that the Baire and Borel σ-algebras coincide, we
deduce that:
B.3. Lemma. (See also [8: Theorem 21.20]) Under (2.1), every Borel measure on B(X)
is regular.
B.4. Proposition. Let µ be any measure on B(X). Then µ is a Borel measure if and
only if
ZX
f dµ < ∞,
for all f in C+
c (X).
Proof. We verify only the "if" part. For this, let K be any compact subset of X. Using
local compactness one may find a relatively compact, open set U such that K ⊆ U . By
Uryshon's Theorem let f : X → [0, 1] be a continuous function vanishing off U , and such
that f = 1 on K. If follows that f ∈ C+
c (X), whence
µ(K) = ZX
1K dµ ≤ ZX
f dµ < ∞.
Being finite on compact sets, µ is a Borel measure.
(cid:3)
B.5. Proposition. There exists a sequence {ϕn}n of continuous, compactly supported
functions
ϕn : X → [0, 1],
such that ϕn ≤ ϕn+1, for every n, and lim
n→∞
ϕn = 1, pointwise.
Proof. Let {Kn}n∈ be a sequence of compact subsets of X such that
Kn ⊆ int(Kn+1),
and X = [n
Kn.
Using Uryhson, for each n ∈ , let
ϕn : X → [0, 1]
be a continuous function with ϕn = 1 on Kn and ϕn = 0 on X \ int(Kn+1). It is then easy
to see that ϕn ≤ ϕn+1, and that {ϕn}n converges pointwise to 1 on X.
(cid:3)
References
[1] R. Bissacot, R. Exel, R. Frausino and T. Raszeja, "Conformal measures on generalized Renault-
Deaconu groupoids", preprint, 2018.
[2] R. Exel, "Inverse semigroups and combinatorial C*-algebras", Bull. Braz. Math. Soc. (N.S.), 39
(2008), 191 -- 313.
46
[3] R. Exel and M. Laca, "Cuntz-Krieger algebras for infinite matrices", J. reine angew. Math., 512
(1999), 119 -- 172.
[4] R. Exel and A. Lopes, "C*-algebras, approximately proper equivalence relations, and thermodynamic
formalism", Ergodic Theory Dynam. Systems, 24 (2004), 1051 -- 1082.
[5] J. Renault, "A groupoid approach to C*-algebras", Lecture Notes in Mathematics vol. 793, Springer,
1980.
[6] J. Renault, "Cuntz-like algebras", Proceedings of the 17th International Conference on Operator
Theory (Timisoara 98), The Theta Fondation, 2000.
[7] J. Renault, "The Radon-Nikod´ym problem for appoximately proper equivalence relations", Ergodic
Theory Dynam. Systems, 25 (2005), no. 5, 1643 -- 1672.
[8] H. L. Royden and P. M. Fitzpatrick, "Real Analysis", fourth edition, Pearson Education Asia Ltd.,
2010.
47
|
1705.09104 | 2 | 1705 | 2018-07-12T07:31:27 | A remark on the minimal dilation of the semigroup generated by a normal UCP-map | [
"math.OA"
] | There are known three ways to construct the minimal dilation of the discrete semigroup generated by a normal unital completely positive map on a von Neumann algebra, which are given by Arveson, Bhat-Skeide and Muhly-Solel. In this paper, we clarify the relation of the constructions by Bhat-Skeide and Muhly-Solel. | math.OA | math |
A remark on the minimal dilation of the
semigroup generated by a normal UCP-map
Yusuke Sawada
Abstract: There are known three ways to construct the minimal dilation of
the discrete semigroup generated by a normal unital completely positive map
on a von Neumann algebra, which are given by Arveson, Bhat-Skeide and
Muhly-Solel. In this paper, we clarify the relation of the constructions by
Bhat-Skeide and Muhly-Solel.
Keywords: dilations, completely positive maps, von Neumann algebras, W ∗-
bimodules, relative tensor products
MCD (2010): 45L55 (Primary), 46L08 (Secondary)
1
Introduction
A dynamical transformation in a quantum physical system is described by a
completely positive (CP) map on an operator algebra in a broad sense. We
consider a von Neumann algebra M acting on a Hilbert space H and a normal
unital completely positive (UCP) map T on M. The Stinespring's dilation
theorem ensures the existence of a normal representation (π, K) of M and
an isometry v : H → K such that T (x) = v∗π(x)v for all x ∈ M. When we
consider a time evolution, the n-times transformation T n is important, but
it is difficult to deal with representations {πn}∞
n=1.
Now we consider the minimal dilation of the semigroup {T n} that is a larger
von Neumann algebra N ⊃ M and a ∗-endomorphism α on N such that T n
is represented by αn for each n ∈ N, and it is hoped that (N, α) is minimal.
To be accurate, the notion of minimal dilations is introduced in [5] as the
following.
n=1 associated with {T n}∞
1
Definition 1.1. Let M be a von Neumann algebra and T a normal UCP-
map on M. A triplet (N, α, p) of a von Neumann algebra N ⊃ M, a ∗-
endomorphism α on N and a projection p ∈ N is called a dilation of T if
M = pNp and T n(x) = pαn(x)p for all x ∈ M and n ∈ Z≥0. Moreover,
a dilation (N, α, p) of T is called minimal if N is generated by S∞
n=0 αn(M)
and the central support c(p) of p coincides with 1N .
Dilations for a C ∗-algebra A and those for a continuous semigroup {Tt}t≥0
consisting of CP-maps on A are also defined in a similar way. It is known
that a minimal dilation is unique if it exists, see [5]. That is, for a min-
imal dilation (N, α, p) of a normal UCP-map T : M → M, an operator
pαn1(x1) · · · αnk(xk)p is uniquely determined by T , n1, · · · , nk ∈ Z≥0 and
x1, · · · , xk ∈ N for each k ∈ N. Then the question of the existence of the
minimal dilation arises. Bhat[8] proved the existence of the minimal dilation
in the case when A = B(H) which consists of all bounded operators on a
Hilbert space H, and each Tt is unital. In [9], he generalized a way of the
construction in stages and constructed a minimal dilation on a C ∗-algebra A
under the assumption that A is unital and kTt(1A)k ≤ 1 holds for all t ≥ 0.
These are called the minimal dilation theory for C ∗-algebras.
After that, Bhat-Skeide[10] constructed the minimal dilation on a von
Neumann algebra N ⊃ A in the case when A is a von Neumann algebra and
a semigroup {Tt}t≥0 of normal CP-maps on A has a continuity with respect to
t ≥ 0, by using inductive limits of the tensor products of Hilbert bimodules.
On the other hand, Arveson[1],[2] introduced the product systems and gave a
one-to-one correspondence between product systems and semigroups {αt}t≥0
of ∗-endomorphisms called the E0-semigroups. Consequently, he classified
product systems. But after that, it is understood that Arveson's theory con-
tains the dilation theory substantially, and his idea affected the constructions
of dilations. Muhly-Solel[14] proved the result in [10] for normal UCP-maps
{Tt}t≥0 by the similar way as in [10]. But the constructions are different in
its appearance and no direct relation has not been established yet.
In this paper, we overview the constructions in [10] and [14], of the
minimal dilation in the case when given semigroup is a discrete semigroup
{T n}∞
n=0 generated by a normal UCP-map. We shall make their direct rela-
tionship clear and reveal that these constructions are essentially the same.
The dilation of a discrete semigroup is applicable to the theory of non-
commutative Poisson boundaries as revealed in [13].
In what follows, we assume that all Hilbert spaces are separable, and
2
B(H, K) means the set of all bounded operators from H to K. If K = H, we
denote B(H, K) by B(H). For a set X, the identity map on X is denoted by
idX and F 0 = idX for every map F : X → X. The unit of a unital algebra
A is denoted by 1A.
The author is deeply grateful to Prof. Shigeru Yamagami for insightful
comments and suggestions.
2 Preliminaries
We recall the notion of W ∗-modules and the related notations about them.
Definition 2.1. (1) For von Neumann algebras N and M, a Hilbert space
H with normal ∗-representations of N and the opposite von Neumann
algebra M ◦ of M is a W ∗-N-M-bimodule if their representations com-
mute. When N = C or M = C, we call H a right W ∗-M-module or a
left W ∗-N-module, respectively. We write a W ∗-N-M-bimodule, a right
W ∗-M-module and a left W ∗-N-module by N HM , HM and N H, respec-
tively.
(2) Let N be a von Neumann algebra, XN and YN right W ∗-N-modules,
and N Z and N W left W ∗-N-modules. Hom(XN , YN ) and Hom(N Z, N W )
are the sets of all right and left N-linear bounded maps, respectively.
If X = Y and Z = W , they are denoted by End(XN ) and End(N Z),
respectively.
(3) We denote the standard representation space of a von Neumann algebra
M in [12] by L2(M).
We introduce the notion of Hilbert modules which are tools to construct
the minimal dilation by the ways by Bhat-Skeide and Muhly-Solel. It is a
module over a von Neumann algebra M with an M-valued inner product.
Definition 2.2. Let M be a von Neumann algebra and E be right M-module.
If a map (·, ·) : E × E → M is defined and satisfies the following properties,
then E is called a Hilbert M-module.
(1) (x, αy + βz) = α(x, y) + β(x, z) (x, y, z ∈ E, α, β ∈ C).
(2) (x, ya) = (x, y)a (x, y ∈ E, a ∈ M).
3
(3) (x, y)∗ = (y, x) (x, y ∈ E).
(4) (x, x) ≥ 0 (x ∈ E).
(5) For every x ∈ E, x = 0 if and only if (x, x) = 0.
(6) E is complete with respect to the norm defined by kxk = k(x, x)k
1
2 .
Suppose E and F are Hilbert M-modules. A right module homomorphism
b : E → F is called adjointable if there is a right module homomorphism
b∗ : F → E called the adjoint of b such that (y, bx) = (b∗y, x) holds for
every x ∈ E and a ∈ M. We denote the set of all adjointable right module
homomorphism by Ba(E, F ). Automatically, b ∈ Ba(E, F ) is bounded and
Ba(E) = Ba(E, E) is a C ∗-algebra.
If there is a surjection u ∈ Ba(E, F ) satisfying (ux, uy) = (x, y) for every
x, y ∈ E, it is called an isomorphism or a unitary. Then E and F are said
to be isomorphic and we write E ∼= F .
Definition 2.3. Let M and N be von Neumann algebras and E a Hilbert
N-module. We call E a Hilbert M-N-bimodule when it is an M-N-bimodule
satisfying
(x, ay) = (a∗x, y)
for every x, y ∈ E and a ∈ M.
Definition 2.4. Let M, N and P be von Neumann algebras, E a Hilbert
N-M-bimodule and F a Hilbert M-P -bimodule. Left and right actions of
a ∈ M and c ∈ P on the algebraic tensor product E ⊗alg F are defined by
a(x ⊗ y)c = (ax) ⊗ (yc) for each x ∈ E and y ∈ F . We define that
(x ⊗ y, x′ ⊗ y′) = (y, (x, x′)y′)
for each x, x′ ∈ E and y, y′ ∈ F , and put N = {z ∈ E ⊗alg F (z, z) = 0}.
The tensor product E ⊗M F of E and F is defined by the completion of
(E ⊗alg F )/N with respect to the norm induced from the above inner product.
The left and right actions can be extended on E ⊗M F , thus E ⊗M F becomes
as Hilbert N-P -bimodules.
The tensor product is associative, and for a Hilbert M-M-bimodule E,
we regard as Ba(E) ⊂ Ba(E) ⊗M 1E ⊂ Ba(E ⊗M E).
We introduce the GNS-construction with respect to a normal UCP-map,
see [17] for example.
4
Definition 2.5. Suppose M is a von Neumann algebra and T : M → M
is a normal UCP-map. We define a Hilbert M-M-bimodule E(M, T ) by the
completion of (M ⊗alg M)/N with respect to a norm induced from an inner
product
(a ⊗ b, a′ ⊗ b′)T = b∗T (a∗a′)b′
(a, a′, b, b′ ∈ M),
where N = {z ∈ M ⊗alg M (z, z)T = 0}. If we put ξ = 1M ⊗ 1M + N , then
span(MξM) is dense in E(M, T ) and T (a) = (ξ, aξ) holds for all a ∈ M.
We call the couple (E(M, T ), ξ) the GNS-representation with respect to T .
There is an important identification in Bhat-Skeide's construction as the
following.
Definition 2.6. Let M be a von Neumann algebra acting on a Hilbert space
H and E a Hilbert M-module. Then H and E are a Hilbert M-C-bimodule
and a Hilbert C-M-bimodule, respectively, and hence we can define the tensor
product E ⊗M H as Hilbert bimodules. For ξ ∈ E, we define Lξ : H ∋ h 7→
ξ ⊗ h ∈ E ⊗M H. Then we can identify E as a right M-submodule of
B(H, E ⊗M H) by a map : E ∋ ξ 7→ Lξ ∈ B(H, E ⊗M H). For b ∈ Ba(E), we
can identify that Ba(E) ⊂ B(E ⊗M H) by
b(ξ ⊗ h) = (bξ)h ∈ E ⊗M H (ξ ∈ E, h ∈ H).
If E ⊂ B(H, E ⊗M H) is closed with respect to the strong operator topology,
E is called a von Neumann M-module.
Suppose N is a von Neumann algebra. A von Neumann M-module E is
called a von Neumann N-M-bimodule if E is a Hilbert N-M bimodule, and
a map ρ : N → B(E ⊗M H) defined by
ρ(x)(ξ ⊗ h) = xξ ⊗ h
(ξ ∈ E, h ∈ H)
is normal.
Then Ba(E) ⊂ B(E ⊗M H) is a von Neumann subalgebra; see [17].
A tensor product defined below is used in Muhly-Solele's construction.
Definition 2.7. Let M be a von Neumann algebra acting on a Hilbert space
H and T a normal UCP-map on M. We define a sesquilinear form on the
algebraic tensor product M ⊗alg H by
(x ⊗ ξ, y ⊗ η) = (ξ, T (x∗y)η)
(x, y ∈ M, ξ, η ∈ H).
5
We define the Hilbert space M ⊗T H = (M ⊗alg H)/N, where N = {z ∈
M ⊗alg H (z, z) = 0}.
A representation πT of M on M ⊗T H is defined by
πT (y)(x ⊗ ξ) = yx ⊗ ξ
(x ∈ M, ξ ∈ H).
3 Some isomorphisms between W ∗-bimodules
In this section, some new results on isomorphisms between W ∗-bimodules
are presented as Proposition 3.3–Corollary 3.6. In Subsection 4.4, they will
be used to see a relation between two constructions of the minimal dilation,
which are given by Bhat-Skeide and Muhly-Solel.
First, we introduce notations with respect to W ∗-modules and the relative
tensor products in [11] and [16], and recall the facts about them (cf. [18] and
[6]).
Fact 3.1. (1) Let M be a von Neumann algebra and M H a W ∗-M-module.
For each positive normal functional φ on M, let (πφ, Hφ, ξφ) be the GNS-
representation of M with respect to φ. We denote πφ(x)ξφ by xφ
2 for
each x ∈ M. Since H is decomposable into cyclic representations, there
exists a family of vectors {ξi}i∈I in H such that H = Li∈I Hωi where
ωi(x) = (ξi, xξi). Moreover, if we denote the support of ωi by qi, we have
1
H ∼= M
(L2(M)qi) ∼= (M
L2(M))q
i∈I
i∈I
as W ∗-M-module where q is the diagonal matrix whose diagonal entries
are {qi}i∈I.
(2) For a W ∗-M-N-bimodule M HN , we denote the dual Hilbert space of H
by H∗. For every ξ∗ ∈ H∗, the right action of x ∈ M and the left action
of y ∈ N to ξ∗ are defined by
yξ∗x = (x∗ξy∗)∗ ∈ H∗.
Then H∗ becomes an N-M-bimodule.
(3) For each right W ∗-M-module HM and left W ∗-M module M K, we denote
the relative tensor product of H and K with respect to M by H⊗M K. The
6
relative tensor product is associative. For a faithful semi-finite normal
weight φ, the subspace of sums of the form ξφ− 1
2 η's is dense in H ⊗M K.
Here, the notation ξφ− 1
2 η means that the tensor product of ξ ∈ H and
a φ-bounded vector η ∈ K For details, see [11, Chapter 5, Appendix B].
The relative tensor products have the following property for W ∗-bimodule
N HM and M KP .
H ⊗M L2(M) ∼= H, L2(M) ⊗M K ∼= K,
K ⊗(M ′)◦ K∗ ∼= L2(M), K∗ ⊗M K ∼= L2(M ′)
where these isomorphisms mean as W ∗-bimodules.
(4) We fix a von Neumann algebra M. Let XM be a Hilbert M-module
and HM be a right W ∗-M-module. We can define the right W ∗-module
H(X)M and the Hilbert M-module X(H)M as the following.
H(X)M = (X ⊗M L2(M))M ,
(x ⊗ ξ, y ⊗ η)H(X) = (ξ, (x, y)η)
X(H) = Hom(L2(M)M , HM )M ,
(x, y)X(H) = x∗y ∈ End(L2(M)M ) = M (x, y ∈ X(H)).
(x ⊗ ξ, y ⊗ η ∈ H(X)),
This gives a one-to-one correspondence between Hilbert M-modules and
right W ∗-M-modules.
From now on, we fix a von Neumann algebra M acting on a Hilbert space
H and a normal UCP-map T on M. We see relations between the relative
tensor product ⊗M and the tensor product ⊗T defined in Section 1.
Definition 3.2. Since M acts on the standard space L2(M) of M, we can
define a left W ∗-M-module H(M, T ) = M ⊗T L2(M) (Definition 2.7). We
define a right action of M on H(M, T ) by (x⊗ξ)y = x⊗ξy for each x, y ∈ M
and ξ ∈ L2(M). Then H(M, T ) is a W ∗-M-M-bimodule.
Proposition 3.3. An isomorphism H(M, T ) ⊗M H(M, T ) ∼= M ⊗T (M ⊗T
L2(M)) holds as W ∗-bimodules.
Proof. Let φ be a faithful normal state on M. We define a correspondence
from an each vector
2 )φ− 1
(x ⊗T yφ
1
2 (z ⊗T φ
1
2 w) ∈ (M ⊗T L2(M)) ⊗φ (M ⊗T L2(M))
∼= (M ⊗T L2(M)) ⊗M (M ⊗T L2(M))
= H(M, T ) ⊗M H(M, T )
7
to a vector
x ⊗T ((yz) ⊗T (φ
1
2 w)) ∈ M ⊗T (M ⊗T L2(M)).
Then this correspondence gives a W ∗-bimodule isomorphism.
Proposition 3.4. An isomorphism H(M, T ) ⊗M H ∼= M ⊗T H holds as
W ∗-modules.
✷
Proof. Let φ be a faithful normal state on M. By Fact 3.1 (1) with respect
to the decomposition of H, each vector ξ ∈ H can be represented as Li∈I ξi
for some ξi ∈ L2(M)pi and the projection pi. We define a correspondence
which maps
(x ⊗T yφ
1
2 )φ− 1
2 M
ξi ∈ (M ⊗T L2(M)) ⊗M H
to x ⊗T (Li∈I yξi) ∈ M ⊗T H. This correspondence is a unitary.
✷
i∈I
Now, we have
H(M, T ) ⊗M H(M, T ) ⊗M H(M, T )
= (M ⊗T L2(M)) ⊗M (M ⊗T L2(M)) ⊗M (M ⊗T L2(M))
∼= (M ⊗T L2(M)) ⊗M (M ⊗T (M ⊗T L2(M)))
∼= (M ⊗T (M ⊗T (M ⊗T L2(M)))).
Indeed the first isomorphism is implied from Proposition 3.3 and the third
isomorphism is given by a unitary defined by
(x1 ⊗T x2φ
1
2 )φ− 1
2 (x3 ⊗T (x4 ⊗T φ
1
2 x5)) 7→ x1 ⊗T ((x2x3) ⊗T (x4 ⊗T φ
1
2 x5))
for each x1, x2, x3, x4, x5 ∈ M similarly to the proof of Proposition 3.4. In
the same way, we have
H(M, T ) ⊗M · · · ⊗M H(M, T )
}
We define a W ∗-M-M-bimodule
n times
{z
∼= M (M ⊗T (· · · ⊗T (M
}
n times
{z
⊗T L2(M)) · · · )).
M Hn(M, T )M = M H(M, T ) ⊗M · · · ⊗M H(M, T )
}
n times
{z
M
8
and a W ∗-(M ′)◦-(M ′)◦-bimodule
(M ′)◦H′
n(M, T )(M ′)◦ = (M ′)◦H∗ ⊗M Hn(M, T ) ⊗M H(M ′)◦
for each n ∈ N.
Proposition 3.5. We have an isomorphism
H′
n(M, T ) ∼= H′
as W ∗-bimodules for all n ∈ N.
{z
n times
1(M, T ) ⊗(M ′)◦ H′
1(M, T )
}
Proof. By Fact 3.1 (3), we have isomorphisms
H′
1(M, T )M ′
1(M, T ) ⊗(M ′)◦
H′
= H∗ ⊗M H(M, T ) ⊗M H ⊗(M ′)◦
∼= H∗ ⊗M H(M, T ) ⊗M L2(M) ⊗M H(M, T ) ⊗M H
∼= H∗ ⊗M H(M, T ) ⊗M H(M, T ) ⊗M H
= H′
2(M, T )
H∗ ⊗M H(M, T ) ⊗M H
as W ∗-(M ′)◦-(M ′)◦-bimodules.
✷
Corollary 3.6. We have an isomorphism
Hn(M, T ) ⊗M H ∼= M ⊗T (M ⊗T · · · (M ⊗T (M
}
n times
{z
as W ∗-modules for all n ∈ N.
⊗T H)) · · · ).
Now, we have the following isomorphisms
Hom(M H, M (M ⊗T L2(M)) ⊗M H)
∼= Hom((M ′)◦H∗ ⊗M H, (M ′)◦H∗ ⊗M (M ⊗T L2(M)) ⊗M H)
∼= Hom((M ′)◦L2(M ′), (M ′)◦H∗ ⊗M (M ⊗T L2(M)) ⊗M H)
(∵ Fact 3.1 (3)).
Then Hom((M ′)◦L2(M ′), (M ′)◦H∗ ⊗M (M ⊗T L2(M)) ⊗M H) corresponds to
H∗ ⊗M (M ⊗T L2(M)) ⊗M H = H′
1(M, T ) by Fact 3.1 (4).
9
4 Two constructions of the minimal dilation
In this section, we describe two constructions of the minimal dilation by
Bhat-Skeide[10] and Muhly-Solel[14], and see a relation between these con-
structions. We fix a von Neumann algebra M acting on a Hilbert space H
and a normal UCP-map T on M.
4.1 Bhat-Skeide's construction
Let (E(M, T ), ξ) be the GNS-representation with respect to T . We put
En = E(M, T ) ⊗M · · · ⊗M E(M, T )
}
n times
{z
,
ξn = ξ ⊗ · · · ⊗ ξ
}
n times
{z
Then (En, ξn) is the GNS-representation with respect to T n for each n ∈ N
by the uniqueness of the GNS-representation. Let E be an the inductive
limit of the inductive system ({En}∞
n,m=0). We define Kn =
En ⊗M H for each n ∈ N and K = E ⊗M H. By the identification in Definition
s
n ⊂ B(H, Kn) is a von Neumann M-M-bimodule and
2.6 and [17], each E
⊂ B(H, K) is so, where · s means the closure with respect to the strong
E
operator topology under the embeddings.
n=0, {ξn−m ⊗idEn}∞
s
s
We define an endomorphism θ on Ba(E
) by
θ(b) = b ⊗ idE1
s ∈ Ba(E
s
⊗ E
s
s
1
) ∼= Ba(E
s
)
(b ∈ Ba(E
s
)).
For each a ∈ M, we define j0(a) ∈ Ba(E
s
) by
j0(a)(η) = ξa(ξ, η)
(η ∈ E)
and jn = θn ◦ j0 ∈ Ba(E
s
) for each n ∈ N. Then we have
jm(1M )jn(a)jm(1M ) = jm(T n−m(a))
for all n ≥ m and a ∈ M. We can identify that M = j0(M). Let N be a von
Neumann algebra generated by jZ≥0(M), p be j0(1M ) and α be a restriction
of θ to N. Then the conditions in Definition 1.1 are satisfied.
10
4.2 Muhly-Solel's construction
Put E(0) = M ′. For each n ∈ N, we define Hn = (M ⊗T (· · · ⊗T (M
}
n times
{z
⊗T H) · · · ))
and E(n) = Hom(M H, M Hn). Each E(n) admits an M ′-valued inner product
defined by
(X, Y ) = X ∗Y ∈ M ′
(X, Y ∈ E(n)),
and we can define left and right actions of M ′ on E(n) by
(xX)ξ = (1M ⊗ · · · ⊗ 1M
}
n times
{z
⊗x)Xξ
(x ∈ M ′, X ∈ E(n), ξ ∈ H),
(Xx)ξ = X(xξ)
(x ∈ M ′, X ∈ E(n), ξ ∈ H).
Then E(n) becomes a W ∗-correspondence over M ′ in the sense of [14], and
we identify E(n) ⊗M ′ E(m) with E(n + m) by a map
Un,m : E(n) ⊗M ′ E(m) ∋ Xn ⊗ Xm 7→ (1M ⊗ · · · ⊗ 1M
}
m times
{z
⊗Xn)Xm ∈ E(n + m)
for each n, m ∈ Z≥0.
Now, we put P0 = idE(0) and L0 = H, and for each n ∈ N define a map
Pn : E(n) → B(H) by Pn(X) = i∗ ◦ X for each X ∈ E(n). Let Ln be a
Hilbert space which is given by the completion of E(n) ⊗alg H with respect
to an inner product defined by
(X ⊗ ξ, Y ⊗ η) = (Xξ, Y η)
(X, Y ∈ E(n), ξ, η ∈ H).
For each 0 < m < n, we define isometric operators un,m by
un,m = (Um,n−m ⊗ 1B(H))(idE(m) ⊗ P ∗
un,0 = P ∗
un,n = 1B(Ln) : Ln → Ln,
n : L0 → Ln,
n−m) : Lm → Ln,
where for all Q : E(n) → B(H), a map Q : Ln → Hn is defined by Q(X ⊗ξ) =
Q(X)ξ for each X ∈ E(n) and ξ ∈ H. Let L be the inductive limit of
({Ln}∞
n,m=0) and ιn : Ln → L be the canonical embedding for each
n ∈ Z≥0. For each m ∈ Z≥0 and Xn ∈ E(n), we define Vn(Xn) ∈ B(L) by
n=0, {unm}∞
Vn(Xn)(ιm(Xm ⊗ ξ)) = ιn+m(Un,m(Xn ⊗ Xm) ⊗ ξ)
(Xm ∈ E(m), ξ ∈ H).
11
We put N = V0(M ′)′ and define α(x) = V1(idE(1) ⊗ x) V ∗
Then α is a normal unital ∗-endomorphism on N such that
1 for each x ∈ N.
ι∗
0Nι0 = M,
T n(ι∗
T n(y) = ι∗
0xι0) = ι∗
0αn(x)ι0
0)ι0
0αn(ι0yι∗
(n ∈ Z≥0, x ∈ N),
(n ∈ Z≥0, y ∈ M).
We identify M with ι0Mι∗
have
M ∼= ι0Mι∗
0 and define a projection p = ι0ι∗
0 in N. Then we
0 = ι0ι∗
0Nι0ι∗
0 = pNp ⊂ N.
n=0 is the minimal dilation of the semigroup {T n}∞
Thus the semigroup {αn}∞
n=0
in the sense of [3] and [4]. We have constructed the minimal dilation in the
sense of Definition 1.1.
4.3 The minimal dilation on the standard space
We simplify Muhly-Solel's construction of the minimal dilation when H =
L2(M). When we use the notation in Subsection 4.2, E(0) = M ′ and for
each n ∈ N,
Hn = (M ⊗T (· · · ⊗T (M
}
n times
{z
⊗T L2(M)) · · · )),
E(n) = Hom(M L2(M), M Hn),
Ln = E(n) ⊗ L2(M).
Then for n ∈ Z≥0, the map Un : E(n) ⊗ L2(M) ∋ X ⊗ ξ 7→ Xξ ∈ Hn
∼= Hn as Hilbert spaces. Now, for n ≥ m, we define
gives an isomorphism Ln
an isometry
vn,m = Unun,mU ∗
m : Hm → Hn
n=0, {vnm}∞
where un,m : Lm → Ln is the isometry defined in Subsection 4.2. Then
({Hn}∞
n,m=0) is an inductive system, and let H′ be the inductive
limit of it. Similarly as Subsection 4.2, for each n ∈ Z≥0, let κn : Hn → H′ be
the canonical embedding and we define V ′
n(Xn) ∈ B(H′) for each Xn ∈ E(n)
by
V ′
n(Xn)(κm(x1 ⊗ · · · ⊗ xm ⊗ ξ)) = κn+m(x1 ⊗ · · · ⊗ xm ⊗ Xnξ)
(m ∈ Z≥0, x1 ⊗ · · · ⊗ xm ⊗ ξ ∈ Hm).
12
Then we can prove an analogue of the result in Subsection 4.2 by looking the
proof of the original theorem ([14]) i.e., if we define
0(M ′)′,
R = V ′
β(x) = V ′
1(idE(1) ⊗ x) V ′∗
1
(x ∈ R),
then β is a normal unital ∗-endomorphism on R such that
κ∗
0Rκ0 = M,
T n(κ∗
T n(y) = κ∗
0xκ0) = κ∗
0αn(x)κ0
0)κ0
0αn(κ0yκ∗
(n ∈ Z≥0, x ∈ N),
(n ∈ Z≥0, y ∈ M).
4.4 A relation between the two constructions
In this subsection, we use the notations in Section 3, Subsection 4.1 and 4.2.
By Proposition 3.4,
E(1) = Hom(M H, M M ⊗T H) ∼= Hom(M H, M (M ⊗T L2(M)) ⊗M H)
holds, and hence E(1) corresponds to H∗ ⊗M (M ⊗T L2(M)) ⊗M H. Hence
we get a one-to-one correspondence
E(n) ∼= E(1) ⊗M ′ · · · ⊗M ′ E(1)
}
n times
{z
←→ H∗ ⊗M Hn(M, T ) ⊗M H.
for each n ∈ N.
On the other hand, for each n ∈ N, we can define the tensor product
s
⊗M L2(M) as Definition 2.4 where En is in Subsection 4.1. Then we
⊗M L2(M) ∼= M ⊗T L2(M) = H1 as left W ∗-module.
Indeed a
⊗M L2(M) ∋ (x ⊗T y) ⊗ ξ 7→ x ⊗T yξ ∈ H1 gives an isomorphism
s
s
En
have E1
map : E1
because we have
(x1 ⊗T y1ξ1, x2 ⊗T y2ξ2)H1 = (y1ξ1, T (x∗
1T (x∗
1x2)y2ξ2)L2(M )
1x2)y2ξ2)L2(M )
= (ξ1, y∗
= (ξ1, (x1 ⊗T y1, x2 ⊗T y2)ξ2)L2(M )
= ((x1 ⊗T y1) ⊗ ξ1, (x2 ⊗T y2) ⊗ ξ2)
for all x1, x2, y1, y2 ∈ M and ξ1, ξ2 ∈ L2(M). By induction, we have
Hn(M, T ) ∼= En
s
⊗M L2(M)
13
for each n ∈ N. Thus we have a correspondence
s
En
←→ Hn(M, T ).
This concludes that the constructions of the dilation by Bhat-Skeide and
Muhly-Solel are essentially the same.
References
[1] W. Arveson, Continuous analogues of Fock space, Mem. Amer. Math.
Soc. 80 (1989), no. 409, iv+66 pp.
[2] W. Arveson, Continuous analogues of Fock space. IV, Essential states,
Acta Math. 164 (1990), no. 3-4, 265-300.
[3] W. Arveson, Minimal E0-semigroups, Operator Algebras and their Ap-
plications (Fillmore, P., and Mingo, J., Eds.), Fields Institute Commu-
nications. AMS (1997), 1-12.
[4] W. Arveson, The index of a quantum dynamical semigroup (English
summary), J. Funct. Anal. 146 (1997), no. 2, 557-588.
[5] W. Arveson, Noncommutative dynamics and E-semigroups, Springer
Monographs in Mathematics. Springer-Verlag, New York (2003), x+434
pp.
[6] M. Baillet, Y. Denizeau, J. F. Havet, Indice d'une esp´erance condition-
nelle, Compositio Math. 66 (1988), no. 2, 199-236.
[7] B. V. R. Bhat, Markov dilations of nonconservative quantum dynami-
cal semigroups and a quantum boundary theory, Ph. D. Thesis, Indian
Statistical Institute, New Delhi (1993).
[8] B. V. R. Bhat, An index theory for quantum dynamical semigroups,
Trans. Amer. Math. Soc. 348 (1996), no. 2, 561-583.
[9] B. V. R. Bhat, Minimal dilations of quantum dynamical semigroups to
semigroups of endomorphisms of C ∗-algebras, J. Ramanujan Math. Soc.
14 (1999), no. 2, 109-124.
14
[10] B. V. R. Bhat, M. Skeide, Tensor product systems of Hilbert modules
and dilations of completely positive semigroups, Infin. Dimens. Anal.
Quantum Probab. Relat. Top. 3 (2000), no. 4, 519-575.
[11] A. Connes, Noncommutative Geometry. Academic Press (1994).
[12] U. Haagerup, The standard form of von Neumann algebras. Math.
Scand. 37 (1975), no. 2, 271-283.
[13] M. Izumi, E0-semigroups: around and beyond Arveson's work, J. Oper-
ator Theory 68 (2012), no. 2, 335-363.
[14] P. S. Muhly, B. Solel, Quantum Markov processes (correspondences and
dilations), Internat. J. Math. 13 (2002), no. 8, 863-906.
[15] W. L. Paschke, Inner product modules over B∗-algebras. Trans. Amer.
Math. Soc. 182 (1973), 443-468.
[16] J. L. Sauvageot, Sur le produit tensoriel relatif d'espaces de Hilbert, J.
Operator Theory 9 (1983), no. 2, 237-252.
[17] M. Skeide, Generalised matrix C ∗-algebras and representations of
Hilbert modules, Math. Proc. R. Ir. Acad. 100A (2000), no. 1, 11-38.
[18] M. Takesaki, Theory of operator algebras. II, Encyclopaedia of Mathe-
matical Sciences, 125. Operator Algebras and Non-commutative Geom-
etry, 6. Springer-Verlag, Berlin (2003). xxii+518 pp.
Yusuke Sawada,
Graduate School of Mathematics,
Nagoya University,
Furocho, Chikusa-ku, Nagoya, 464-8602, Japan,
E-mail: [email protected]
15
|
1309.0982 | 1 | 1309 | 2013-09-04T11:33:52 | von Neumann algebra preduals satisfy the linear biholomorphic property | [
"math.OA",
"math.FA"
] | We prove that for every JBW$^*$-triple $E$ of rank $>1$, the symmetric part of its predual reduces to zero. Consequently, the predual of every infinite dimensional von Neumann algebra $A$ satisfies the linear biholomorphic property, that is, the symmetric part of $A_*$ is zero. This solves a problem posed by M. Neal and B. Russo in [Mathematica Scandinavica, to appear] | math.OA | math |
VON NEUMANN ALGEBRA PREDUALS SATISFY THE LINEAR
BIHOLOMORPHIC PROPERTY
ANTONIO M. PERALTA AND LASZLO L. STACH ´O
Abstract. We prove that for every JBW∗-triple E of rank > 1, the sym-
metric part of its predual reduces to zero. Consequently, the predual of every
infinite dimensional von Neumann algebra A satisfies the linear biholomorphic
property, that is, the symmetric part of A∗ is zero. This solves a problem
posed by M. Neal and B. Russo [15, to appear in Mathematica Scandinavica].
1. Introduction
The open unit ball of every complex Banach space satisfies certain holomorphic
properties which determine the global isometric structure of the whole space. An
illustrative example is the following result of W. Kaup and H. Upmeier [13].
Theorem 1.1. [13] Two complex Banach spaces whose open unit balls are biholo-
morphically equivalent are linearly isometric.
(cid:3)
We recall that, given a domain U in a complex Banach space X (i.e. an open,
connected subset), a function f from U to another complex Banach space F is said
to be holomorphic if the Fr´echet derivative of f exists at every point in U . When
f : U → f (U ) is holomorphic and bijective, f (U ) is open in F and f −l : f (U ) → U
is holomorphic, the mapping f is said to be biholomorphic, and the sets U and f (U )
are biholomorphically equivalent. Theorem 1.1 gives an idea of the power of infinite-
dimensional Holomorphy in Functional Analysis. A reviewed proof of Theorem 1.1
was published by J. Arazy in [1].
A consequence of the results established by Kaup and Upmeier in [13] gave raise
to the study of the symmetric part of an arbitrary complex Banach space in the
following sense: Let X be a complex Banach space with open unit ball denoted by
D. Let G = Aut(D) denote the group of all biholomorphic automorphisms of D and
let GO stand for the connected component of the identity in G. Given a holomorphic
function h : D → X, we can define a holomorphic vector field Z = h(z) ∂
∂z , which
is a composition differential operator on the space H(D, X) of all holomorphic
functions from D to X, given by X(f )(z) = (h(z) ∂
∂z )f (z) = f ′(z)(h(z)), (z ∈ D).
It is known that, for each z0 the initial value problem ∂
∂t ϕ(t, z0) = h(ϕ(t, z0)),
ϕ(0; z0) = z0 has a unique solution ϕ(t, z0) : Jz0 → D defined on a maximal open
interval Jz0 ⊆ R containing 0. The holomorphic mapping h is called complete when
Jz0 = R, for every z0 ∈ D. Denoting by aut(D) the Lie algebra of all complete,
Date: August, 2013.
2000 Mathematics Subject Classification. Primary 32N05, 32N15, 17C50; Secondary 17C65,
47L70.
First author partially supported by the Spanish Ministry of Economy and Competitiveness,
D.G.I. project no. MTM2011-23843, and Junta de Andaluc´ıa grant FQM3737.
2
A.M. PERALTA AND L.L. STACH ´O
holomorphic vector fields on D, the symmetric part of D is DS = G(O) = GO(O).
The symmetric part of X, denoted by XS or by S(X), is the orbit of 0 under the
set aut(D) of all complete holomorphic vector fields on D. Furthermore, XS is a
closed, complex subspace of X, DS = XS ∩ D, and hence, DS is the open unit ball
of XS, DS is symmetric in the sense that for each z ∈ DS there exists a symmetry
of D at z, i.e., a mapping sz ∈ Aut(DS) such that sz(z) = z, s2
z = identity, and
s′
[13], [4],
z(z) = −IdE; thus DS = ES ∩ D is a bounded symmetric domain (cf.
and [1]).
A Jordan structure associated with the symmetric part of every complex Banach
space X was also determined by W. Kaup and H. Upmeier in [13]. Namely, for every
a ∈ XS there is a unique symmetric continuous bilinear mapping Qa : X × X → X
such that (a − Qa(z, z)) ∂
∂z is a complete holomorphic vector field on D. A partial
triple product is defined on X × XS × X by the assignment
{., ., .} : X × XS × X → X,
{x, a, y} := Qa(x, y).
It is known (cf. [13] and [4]) that the partial triple product satisfies the following
properties:
(i) {., ., .} is bilinear and symmetric in the outer variables and conjugate linear
in the middle one;
(ii) {XS, XS, XS} ⊆ XS;
(iii) The Jordan identity
{a, b, {x, y, z}} = {{a, b, x} , y, z} − {x, {b, a, y} , z} + {x, y, {a, b, z}} ,
holds for every a, b, y ∈ XS and x, z ∈ X;
(iv) For each a ∈ XS, the mapping L(a, a) : X → X, z 7→ {a, a, z} is a hermitian
operator;
(v) The identity {{x, a, x} , b, x} = {x, a, {x, b, x}} holds for every a, b ∈ XS and
x ∈ X.
It should be remarked here that property (v) appears only implicitly in [4]. A
complete substantiation is included in [16] (compare also [20]).
The extreme possibilities for the symmetric part XS (i.e. XS = X or XS = {0})
define particular and significant classes of complex Banach spaces. The deeply
studied class of JB∗-triples, introduced by W. Kaup in [12], is exactly the class of
those complex Banach spaces X for which XS = X. In the opposite side, we find
the complex Banach spaces satisfying the linear biholomorphic property (LBP, for
short). A complex Banach space X with open unit ball D satisfies the LBP when
its symmetric part is trivial (cf. [1, page 145]).
The symmetric part of some classical Banach spaces was studied and determined
by R. Braun, W. Kaup and H. Upmeier [4], L.L. Stach´o [18], J. Arazy [1], and J.
Arazy and B. Solel [2]. The following list covers the known cases:
(i) For X = Lp(Ω, µ), 1 ≤ p < 1, p 6= 2, and dim(X) ≥ 2, we have XS = 0;
(ii) For X = Hp the classical Hardy spaces with 1 ≤ p < 1, p 6= 2, we have
XS = 0;
(iii) For X = H∞ or the disk algebra, XS = C;
(iv) When X is a uniform algebra A ⊆ C(K), AS = A ∩ A.
(v) When A is a subalgebra of B(H) containing the identity operator I, then AS
is the maximal C∗-subalgebra A ∩ A∗ of A;
THE LINEAR BIHOLOMORPHIC PROPERTY FOR VON NEUMANN ALGEBRA PREDUALS3
(vi) Let X be a complex Banach space with a 1-unconditional basis. Then X = XS
if and only if X is the c0-sum of a sequence of Hilbert spaces. Moreover, if X
is a symmetric sequence space (i.e. the unit vector basis form a 1-symmetric
basis of E) then either XS = {0} or XS = X. In the last case, either X = ℓ1
or X = c0.
In a very recent contribution, M. Neal and B. Russo stated the following problem:
Problem 1.2. [15, Problem 2] Is the symmetric part of the predual of a von Neu-
mann algebra equal to 0? What about the predual of a JBW∗-triple which does not
contain a Hilbert space as a direct summand?
In this note we give a complete answer to the questions posed by Neal and
Russo [15] in the above problem. Our main result proves that for every JBW∗-
triple W which is not a Hilbert space, the symmetric part of its predual reduces to
zero. In particular the symmetric part of the predual of an infinite-dimensional von
Neumann algebra is equal to {0}. Unfortunately, there exist examples of JBW∗-
triples W containing a Hilbert space as a direct summand for which S(W∗) =
(W∗)S = {0}.
2. Computing the symmetric part of a JBW∗-triple predual
We recall that a JB∗-triple is a complex Banach space E satisfying that ES = E.
JB∗-triples were introduced by W. Kaup in [12], where he also gave the following
axiomatic definition of these spaces: A JB∗-triple is a complex Banach space E
equipped with a triple product {·, ·, ·} : E × E × E → E which is linear and
symmetric in the outer variables, conjugate linear in the middle one, satisfies the
axioms (iii) and (iv) in (1) and the following condition:
(vi) k{x, x, x}k = kxk3 for all x ∈ E.
Every C∗-algebra is a complex JB∗-triple with respect to the triple product
2 (xy∗z + zy∗x), and in the same way every JB∗-algebra with respect to
{x, y, z} = 1
{a, b, c} = (a ◦ b∗) ◦ c + (c ◦ b∗) ◦ a − (a ◦ c) ◦ b∗.
Elements a, b in a JB∗-triple E are said to be orthogonal (denoted by a ⊥ b)
whenever L(a, b) = 0. It is known that a ⊥ b ⇔ {a, a, b} = 0 ⇔ {b, b, a} = 0 (cf.
[8, Lemma 1]). The rank, r(E), of a real or complex JB∗-triple E, is the minimal
cardinal number r satisfying card(S) ≤ r whenever S is an orthogonal subset of E,
i.e. 0 /∈ S and x ⊥ y for every x 6= y in S.
We briefly recall that an element e in a JB∗-triple E is said to be a tripotent
whenever {e, e, e} = e. A tripotent e ∈ E is said to be complete whenever a ⊥ e
implies a = 0. When the condition {e, e, a} = a implies that a ∈ Ce, we shall
say that e is a minimal tripotent. The symbol Tri(E) will stand for the set of all
tripotents in E.
The following characterization of complete holomorphic vector fields, which is
originally due to L.L. Stach´o (see [18], [19] and [21]), has been borrowed from [2,
Proposition 2.5].
Proposition 2.1. Let X be a complex Banach space whose open unit ball is denoted
by D and let h : D → X be a holomorphic mapping. Then h ∈ aut(D) if and only
if h extends holomorphically to a neighborhood of D, and, for every z ∈ X, ϕ ∈ X ∗
satisfying kzk = kϕk = 1 = ϕ(z), we have ℜeϕ(h(z)) = 0.
(cid:3)
4
A.M. PERALTA AND L.L. STACH ´O
In order to simplify the arguments, we recall some geometric notions. Elements
s, y in a complex Banach space X are said to be L-orthogonal, denoted by x ⊥L y,
(respectively, M -orthogonal, denoted by x ⊥M y) if kx ± yk = kxk + kyk (respec-
tively, kx ± yk = max{kxk, kyk}). It is known that x ⊥L y if, and only if, for all
real numbers s, t, sx ⊥L ty if, and only if, there exist elements a, b ∈ X ∗ satisfying
a ⊥M b, kxk kak = kxk = a(x), and kyk kbk = kyk = b(y) (see, for example, [10,
Lemma 3.1 and Corollary 4.3]). It is also known that for each pair of elements (a, b)
in a JB∗-triple E, the condition a ⊥ b implies a ⊥M b (cf.
[8, Lemma 1] and [11,
Lemma 1.3(a)]).
We also recall that a JBW∗-triple is a JB∗-triple which is also a dual Banach
space. In this sense, JBW∗-triples play an analogue role to that given to von Neu-
mann algebras in the setting of C∗-algebras. Every JBW∗-triple admits a unique
(isometric) predual and its product is separately weak∗-continuous (see [3]).
We can proceed with a first technical result on the structure of the symmetric
part of a JBW∗-triple predual.
Proposition 2.2. Let W be a JBW∗-triple with predual W∗ = F . Suppose, e1, e2
are two tripotent elements in W , ϕ1, ϕ2 ∈ F with kϕkk = 1, e1 ⊥ e2, and ej(ϕk) =
δjk (j, k = 1, 2). Then e1(φ) = e2(φ) = 0, for every φ in FS .
Proof. Let φ be an element in FS. Since φ ∈ FS, the holomorphic vector field
(cid:2)φ − Qφ(z, z)(cid:3) ∂
∂z is tangent to the unit sphere of F . Thus, by Proposition 2.1,
ℜe(cid:10)e, φ − Qφ(ϕ, ϕ)(cid:11) = 0,
for every ϕ ∈ F , e ∈ W with kϕk = kek = 1 = (cid:10)e, ϕ(cid:11)(cid:0) = e(ϕ)(cid:1).
Since e1 ⊥ e2 implies e1 ⊥M e2, it follows from the hypothesis that ϕ1 ⊥L ϕ2.
In particular, for any weight 0 ≤ λ ≤ 1 and κ1, κ2 ∈ T := {κ ∈ C :
κ = 1},
κ1(1 − λ)ϕ1 + κ2λϕ2 belongs to the unit sphere of F and κ1e1 + κ2e2 is a supporting
functional for it. Therefore,
0 = ℜeDκ1e1 + κ2e2, φ − Qφ(cid:0)κ1(1 − λ)ϕ1 + κ2λϕ2, κ1(1 − λ)ϕ1 + κ2λϕ2(cid:1)E
= ℜe(cid:16)κ1e1(φ) + κ2e2(φ) + κ1(1 − λ)2α1 + κ2λ2α2 + κ1λ(1 − λ)β1 + κ2λ(1 − λ)β2(cid:17)
with the constants αk := (cid:10)ek, Qφ(ϕk, ϕk)(cid:11), βk := 2(cid:10)e3−k, Qφ(ϕk, ϕ3−k)(cid:11). In par-
ticular, with the choice λ = 1 we get
ℜe(cid:0)κ1e1(φ) + κ2e2(φ) + κ2α2(cid:1) = 0
for every κ1, κ2 ∈ T. Replacing κ2 with −κ2 we have ℜe(cid:0)κ1e1(φ)(cid:1) = 0 (κ1 ∈ T),
and hence e1(φ) = 0.
(cid:3)
Before dealing with our main result we shall review some results on JB∗-triples
of rank one. For a JB∗-triple E, the following are equivalent:
(a) E has rank one;
(b) E is a complex Hilbert space equipped with the triple product given by 2{a, b, c} :=
(ab)c + (cb)a, where (..) denotes the inner product of E;
(c) The set of complete tripotents in E is non-zero and every complete tripotent
in E is minimal;
(d) E contains a complete tripotent which is minimal.
THE LINEAR BIHOLOMORPHIC PROPERTY FOR VON NEUMANN ALGEBRA PREDUALS5
The equivalence (a) ⇔ (b) follows, for example, from [7, Proposition 4.5]. The
implications (b) ⇒ (c) and (c) ⇒ (d) are clear. It should be commented here that
a general JB∗-triple might not contain any tripotent. However, since the complete
tripotents of a JB∗-triple E coincide with the real and complex extreme points of its
closed unit ball (cf. [14, Proposition 3.5] and [5, Lemma 4.1]), by the Krein-Milman
theorem, every JBW∗-triple contains an abundant set of (complete) tripotents. In
the setting of JBW∗-triples, a tripotent e is minimal if and only if it cannot be
written as an orthogonal sum of two (non-zero) tripotents (compare the arguments
in [17, Proposition 2.2]). Back to the equivalences, the implication (d) ⇒ (a) is
established in [9, Proposition 3.7 and its proof].
Theorem 2.3. Let W be a JBW∗-triple of rank > 1 and let F denote its predual.
Then FS = {0}, that is, F satisfies the linear biholomorphic property.
Proof. Let φ be an element in FS. According to the Krein-Milman Theorem, the
finite linear combinations of the extreme points of the closed unit ball, D(W ), of
W form a weak∗-dense subset in D(W ). Therefore, it suffices to prove that
(1)
e(φ) = 0 for all e ∈ Ext(cid:0)D(W )(cid:1),
or equivalently, e(φ) = 0 for every complete tripotent e ∈ W .
Let e be a complete tripotent in W . Since W has rank > 1, the comments
preceding this theorem guarantee the existence of two non-zero tripotents e1, e2 in
W such that e1 ⊥ e2 and e = e1 + e2. Let us notice that the JBW∗-subtriple
U of W generated by e1 and e2 coincides with Ce1 L∞ Ce2. We can easily define
two norm-one functionals ψ1, ψ2 in U∗ satisfying ψj(ek) = δjk. By [6, Theorem],
there exists norm-one weak∗-continuous functionals ϕ1, ϕ2 in W∗ which are norm-
preserving extensions of ψ1 and ψ2, respectively. Applying Proposition 2.2 we have
ej(φ) = 0, for every j = 1, 2, and finally e(φ) = e1(φ) + e2(φ) = 0 as we desired. (cid:3)
It is known that a von Neumann algebra, regarded as a JBW∗-triple, has rank
one if and only if it coincides with C. We therefore have:
Corollary 2.4. Let W be a von Neumann algebra of dimension < 1 and let F =
W∗. Then FS = {0}, that is, F satisfies the linear biholomorphic property.
(cid:3)
There is an additional aspect of Problem 1.2 that should be commented. Suppose
H is a complex Hilbert space, W is a non-zero JBW∗-triple, and consider the JBW∗-
triple U = H L∞ W (the orthogonal sum of H and W ). It is clear that U has rank
> 1. Thus, Theorem 2.3 implies that S(U∗) = {0}. In other words, the predual of a
JBW∗-triple which does not contain a Hilbert space as a direct summand satisfies
the linear biholomorphic property but the class of all JBW∗-triples whose preduals
satisfy the linear biholomorphic property is strictly bigger.
References
[1] J. Arazy, An application of infinite-dimensional holomorphy to the geometry of Banach
spaces, in Geometrical aspects of Functional Analysis (1985/86), 122-150, Lecture Notes
in Math., 1267, Springer, Berlin, 1987.
[2] J. Arazy, B. Solel, Isometries of nonselfadjoint operator algebras, J. Funct. Anal. 90, no. 2,
284-305 (1990).
[3] T.J. Barton, R.M. Timoney, Weak∗-continuity of Jordan triple products and its applications.
Math. Scand. 59, 177-191 (1986).
[4] R. Braun, W. Kaup, H. Upmeier, On the automorphisms of circular and Reinhardt domains
in complex Banach spaces, Manuscripta Math. 25, no. 2, 97-133 (1978).
6
A.M. PERALTA AND L.L. STACH ´O
[5] R. Braun, W. Kaup, H. Upmeier, A holomorphic characterization of Jordan C∗-algebras,
Math. Z. 161, 277-290 (1978).
[6] L.J. Bunce, Norm preserving extensions in JBW∗-triple, Quart. J. Math. Oxford 52, No.2,
133-136 (2001).
[7] L.J. Bunce and C.-H. Chu, Compact operations, multipliers and Radon-Nikodym property
in J B∗-triples, Pacific J. Math. 153, 249-265 (1992).
[8] M. Burgos, F.J. Fern´andez-Polo, J. Garc´es, J. Mart´ınez, A.M. Peralta, Orthogonality pre-
servers in C∗-algebras, JB∗-algebras and JB∗-triples, J. Math. Anal. Appl. 348, 220-233
(2008).
[9] M. Burgos, J.J. Garc´es, A.M. Peralta, Automatic continuity of biorthogonality preservers
between weakly compact JB∗-triples and atomic JBW∗-triples, Studia Math. 204 (2) 97-121
(2011).
[10] C.M. Edwards, G.T. Ruttimann, Orthogonal faces of the unit ball in a Banach space, Atti
Sem. Mat. Fis. Univ. Modena 49, 473-493 (2001).
[11] Y. Friedman, B. Russo, Structure of the predual of a JBW∗-triple, J. Reine Angew. Math.
356, 67-89 (1985).
[12] W. Kaup, A Riemann Mapping Theorem for bounded symmentric domains in complex Ba-
nach spaces, Math. Z. 183, 503-529 (1983).
[13] W. Kaup, H. Upmeier, Banach spaces with biholomorphically equivalent unit balls are iso-
morphic, Proc. Amer. Math. Soc. 58, 129-133 (1978).
[14] W. Kaup, H. Upmeier, Jordan algebras and symmetric Siegel domains in Banach spaces,
Math. Z. 157, 179-200 (1977).
[15] M. Neal, B. Russo, A holomorphic characterization of operator algebras, to appear in Math-
ematica Scandinavica. arXiv:1207.7353v1
[16] D. Panou, Uber die Klassifikation der beschrankten bizirkularen Gebiete in Cn, Ph.D. Dis-
sertation, Tubingen, 1988.
[17] A.M. Peralta, L.L. Stach´o, Atomic decomposition of real JBW∗-triples, Quart. J. Math.
Oxford 52, no. 1, 79-87 (2001).
[18] L.L. Stach´o, A short proof of the fact that biholomorphic automorphisms of the unit ball in
certain Lp spaces are linear, Acta Sci. Math. 41, 381-383 (1979).
[19] L.L. Stach´o, A projection principle concerning biholomorphic automorphisms, Acta Sci.
Math. 44, 99-124 (1982).
[20] L.L. Stach´o, On the classification of bounded circular domains, Proc. Roy. Irish Acad. 91,
No.2, 219-238 (1991)
[21] H. Upmeier, Jordan algebras in analysis, operator theory, and quantum mechanics, CBMS,
Regional conference, No. 67 (1987).
E-mail address: [email protected]
Departamento de An´alisis Matem´atico, Facultad de Ciencias, Universidad de Granada,
18071 Granada, Spain.
E-mail address: [email protected]
Bolyai Institute, University of Szeged, 6720 Szeged, HUNGARY
|
1810.05323 | 3 | 1810 | 2019-02-06T07:51:55 | Equilibrium states on higher-rank Toeplitz noncommutative solenoids | [
"math.OA"
] | We consider a family of higher-dimensional noncommutative tori, which are twisted analogues of the algebras of continuous functions on ordinary tori, and their Toeplitz extensions. Just as solenoids are inverse limits of tori, our Toeplitz noncommutative solenoids are direct limits of the Toeplitz extensions of noncommutative tori. We consider natural dynamics on these Toeplitz algebras, and compute the equilibrium states for these dynamics. We find a large simplex of equilibrium states at each positive inverse temperature, parametrised by the probability measures on an (ordinary) solenoid. | math.OA | math |
EQUILIBRIUM STATES ON HIGHER-RANK TOEPLITZ
NONCOMMUTATIVE SOLENOIDS
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
Abstract. We consider a family of higher-dimensional noncommutative tori,
which are twisted analogues of the algebras of continuous functions on ordinary
tori, and their Toeplitz extensions. Just as solenoids are inverse limits of tori, our
Toeplitz noncommutative solenoids are direct limits of the Toeplitz extensions of
noncommutative tori. We consider natural dynamics on these Toeplitz algebras,
and compute the equilibrium states for these dynamics. We find a large simplex
of equilibrium states at each positive inverse temperature, parametrised by the
probability measures on an (ordinary) solenoid.
1. Introduction
Classical solenoids are inverse limits of tori. There are noncommutative ana-
logues of tori, which are the twisted group algebras C ∗(Zn, σ) of the abelian group
Zn. For n = 2, these are the rotation algebras Aθ generated by two unitaries U, V
satisfying the commutation relation UV = e2πiθV U. When θ is irrational, these
are simple C ∗-algebras, and have been extensively studied (see, for example, [10,
Chapter VI]). For θ = 0, we recover the commutative algebra C(T2), and hence
the Aθ are also known as "noncommutative tori." In [24], Latr´emoli`ere and Packer
studied a family of noncommutative solenoids that are direct limits of noncom-
mutative tori.
(The connection is that the commutative algebra of continuous
functions on a solenoid is the direct limit of the algebras of continuous functions
on the approximating tori.)
Following surprising results about phase transitions for the KMS states of the
Toeplitz algebras of the ax + b-semigroup of the natural numbers [21, 19], many
authors have studied the KMS structure of Toeplitz extensions in other settings.
Typically, these Toeplitz extensions exhibit more interesting KMS structure. This
recent work has covered Toeplitz algebras of directed graphs and their higher-
rank analogues [15, 16, 7, 13, 8] (after earlier work in [11]), Toeplitz algebras
arising in number theory [9], the Nica-Toeplitz extensions of Cuntz-Pimsner al-
gebras [19, 17, 18, 1, 4], and Toeplitz algebras associated to self-similar actions
[22, 23]. In [6], Brownlowe, Hawkins and Sims described Toeplitz extensions of
the noncommutative solenoids from [24], and considered a natural dynamics on
this extension. They showed that for each inverse temperature β > 0, the KMSβ
states are parametrised by the probability measures on a commutative solenoid
which is the inverse limit of 1-dimensional tori [6, Theorem 6.6].
Date: February 7, 2019.
2010 Mathematics Subject Classification. 46L05, 46L30, 46L55.
Key words and phrases. KMS states, C ∗-algebras, direct limit, Toeplitz algebra.
This research was supported by the Australian Research Council and the Marsden Fund of the
Royal Society of New Zealand.
1
2
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
Here we consider a family of higher-rank noncommutative solenoids and their
Toeplitz extensions. As for the algebras of higher-rank graphs [16], there is an
obvious gauge action of a torus Td on these algebras, but to get a dynamics one
has to choose an embedding of R in the torus. We fix r ∈ [0, ∞)d, giving an
embedding t 7→ eitr of R in Td, and compose with the gauge action to get a
dynamics αr.
The building blocks in [6] are Toeplitz noncommutative tori in which one gen-
erator U is unitary, the other V is an isometry, the relation is still given by
UV = e2πiθV U, and the dynamics fixes U. Here we fix d, k ∈ N. Our blocks
Bθ are Toeplitz noncommutative tori generated by a unitary representation U of
Zd and a Nica-covariant isometric representation V of Nk, and the commutation
relation is given by UnVp = e2πipT θnVpUn for a fixed k × d matrix θ with entries
in [0, ∞). Then the dynamics αr is given by a vector r ∈ (0, ∞)k; it fixes the
unitaries Un, and multiplies Vp by eitpT r.
We begin by describing the direct system of Toeplitz noncommutative tori whose
limit is the Toeplitz noncommutative solenoid of the title. Everything is defined in
terms of presentations of the blocks: building the connecting maps is in particular
quite complicated, and requires us to be careful with the notation, which we try
to keep consistent throughout the paper. We then discuss the dynamics, which
is again defined using actions on the individual blocks. Then, remarkably, we
have a presentation of the direct limit which allows us to state our main result as
Theorem 2.7. This gives a satisfyingly explicit description of the KMSβ states in
Td. This concrete
terms of measures on a commutative solenoid of the form lim←−
description is new even in the case k = d = 1 studied in [6].
The first step in the proof of our theorem is an analysis of the KMS states of
a building block Bθ, which we do in §3. The description in Proposition 3.7 looks
rather like the descriptions of KMS states on graph algebras in [15, Theorem 3.1]
and [16, Theorem 6.1], and on algebras associated to local homeomorphisms in
[2, Theorem 5.1]: we find a subinvariance relation which identifies the measures
on the torus associated to KMS states, and then describe the solutions of that
relation in terms of a concrete simplex of measures.
In the next section (§4), we show how the subinvariance relations for the build-
ing blocks combine to give one continuously parametrised subinvariance relation
for the direct limit (Theorem 4.1). We then describe the solutions to this new
subinvariance relation in Theorem 5.1, which is the key technical result in the
paper. This solution is very concrete, involving a formula which is reminiscent
of a multi-variable Laplace transform, and is much more direct than the ad hoc
approach used in [6].
In the last section, we give a concrete description of the isomorphism µ 7→ ψµ
Td) of probability measures on the solenoid onto the sim-
of the simplex P (lim←−
plex of KMSβ states on the Toeplitz noncommutative torus. Then by evaluating
these KMS states on generators, we arrive at the explicit values described in The-
orem 2.7.
2. Toeplitz noncommutative solenoids
We define a Toeplitz noncommutative solenoid as the direct limit of a sequence
of blocks, which we call Toeplitz noncommutative tori. So we begin by looking at
KMS STATES
3
these blocks. In the course of this section we will introduce notation which will
be used throughout the paper.
First we fix positive integers d and k. We write AT for the transpose of a
matrix A. We view elements of Rk as column vectors, and write the inner product
of n, p ∈ Rk in matrix notation as pT n. We use similar conventions for Rd.
The pair (Zk, Nk) is a quasi-lattice ordered group in the sense of Nica [25].
Indeed, for every p, q ∈ Nk, the element p ∨ q defined pointwise by
(p ∨ q)j = max{pj, qj} for 1 ≤ j ≤ k
is a least upper bound for p and q, so it is lattice-ordered. An isometric represen-
tation V : Nk → B(H) is Nica-covariant if it satisfies
VpV ∗
p VqV ∗
q = Vp∨qV ∗
p∨q
for all p, q ∈ Nk,
or equivalently [20, (1.4)] if
V ∗
p Vq = V(p∨q)−pV ∗
(p∨q)−q
for all p, q ∈ Nk.
For θ ∈ Mk,d([0, ∞)), we consider the universal C ∗-algebra Bθ generated by a
unitary representation U of Zd and a Nica-covariant isometric representation V of
Nk such that
(2.1)
UnVp = e2πipT θnVpUn
for p, q ∈ Nk and n ∈ Zd.
We then have also
(2.2)
= e−2πipT θnV ∗
p Un.
UnV ∗
p = (VpU−n)∗ =(cid:0)e−2πipT θ(−n)U−nVp(cid:1)∗
Direct calculation shows that for p, q, p′, q′ ∈ Nk and n, n′ ∈ Zd, we have
VpUnV ∗
q Vp′Un′V ∗
q′ = VpUnV(q∨p′)−qV ∗
(q∨p′)−p′Un′V ∗
q′
= e2πi((q∨p′−q)T θn+(q∨p′−p′)T θn′)Vp+(q∨p′)−qUn+n′V ∗
q′+(q∨p′)−p′,
and we deduce that
Bθ = span{VpUnV ∗
q : n ∈ Zd and p, q ∈ Nk}.
We call Bθ a Toeplitz noncommutative torus.
Now we move on to noncommutative solenoids. First we need some more con-
ventions. We write Sd for the compact quotient space Rd/Zd, and view functions
f ∈ C(Sd) as Zd-periodic continuous functions f : Rd → C. We write M(Sd)
for the set of positive measures on Sd, and view measures µ ∈ M(Sd) as positive
0 f dµ on C(Sd). Then kµk := µ(Sd) is the norm of the corre-
sponding functional, and P (Sd) := {µ ∈ M(Sd) : kµk = 1} is the set of probability
measures.
functionals f 7→ R 1
We consider three sequences of matrices {θm} ⊂ Mk,d([0, ∞)), {Dm} ⊂ Mk(N),
and {Em} ⊂ Md(N) such that: each Dm is diagonal with entries larger than 1;
each Em has det Em > 1; and
(2.3)
We choose a sequence {rm} = {(rm
Dmθm+1Em = θm for m ≥ 1.
j )} of vectors in (0, ∞)k satisfying
(2.4)
rm+1 = D−1
m rm for m ≥ 1
Notice that both sequences are determined by the first terms θ1 ∈ Mk,d([0, ∞))
and r1 ∈ [0, ∞)k.
4
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
Example 2.1. We fix N ≥ 2, and set d = k = 1, Dm = Em = N for m ≥ 2,
θ1 ∈ (0, ∞) and θm = N −2(m−1)θ1. Taking the equivalence classes of the θm in
S = R/Z yields an example of the set-up of [6] except that we are insisting that
θm = N 2θm+1 as real numbers, not just as elements of S. This has the consequence
that θm → 0 as m → ∞, which need not happen in the situation of [6]; but see
Remark 2.2 below.
Remark 2.2. Our hypothesis that Dmθm+1Em = θm exactly, and not just modulo
Zd, seems to be crucial in our arguments. Specifically, to assemble the sequences
of KMS states that we will construct on the approximating subalgebras Bm into a
KMS state on B∞, we will need to show that the associated probability measures
νm (see Proposition 3.7(a)) intertwine through the maps induced by the ET
m. We do
this in Lemma 6.2, and we indicate there the step in the first displayed calculation
where it is critical that Dmθm+1Em = θm exactly. This prompted us to review
carefully the arguments of [6] and we believe that those arguments also require
that N 2θm+1 = θm exactly. Specifically, the calculation at the end of the proof of
[6, Theorem 6.9] implicitly treats θj as an element of R (there are many solutions
to N kγ = θj in S). Similarly the formulas in [6, Section 8] that involve setting
rj = β/(N jθj) only make sense if θj is an element of R. In particular, in the final
displayed calculation in the proof of [6, Lemma 8.1], it is critical that N 2θj+1 = θj
exactly.
For each m there is a Toeplitz noncommutative torus Bm := Bθm with generators
Um,n and Vm,p such that: U : n 7→ Um,n is a unitary representation of Zd, V : p 7→
Vm,p is a Nica-covariant isometric representation of Nk, and the pair U, V satisfy
the commutation relation (2.1) for the matrix θm.
Next we use the matrices Dm and Em to build homomorphisms from Bm to
Bm+1.
Proposition 2.3. Suppose that m is a positive integer. Then there is a homo-
morphism πm : Bm → Bm+1 such that πm(Um,n) = Um+1,Emn and πm(Vm,p) =
Vm+1,Dmp.
Proof. We define U : Zd → Bm+1 by Un = Um+1,Emn and V : Nk → Bm+1
by Vp = Vm+1,Dmp. Then since Dm and Em have entries in N, U is a unitary
representation of Zd and V is an isometric representation of Nk.
We claim that V is Nica-covariant. To see this, we take p, q ∈ Nk. Then Nica
covariance of p 7→ Vm+1,p implies that
(2.5)
q = Vm+1,DmpV ∗
VpV ∗
p VqV ∗
m+1,DmpVm+1,DmqV ∗
m+1,Dmq
= Vm+1,(Dmp)∨(Dmq)V ∗
m+1,(Dmp)∨(Dmq).
Now recall that Dm is diagonal1, with diagonal entries dm,j, say. Then for 1 ≤ j ≤
k we have
(cid:0)(Dmp) ∨ (Dmq)(cid:1)j = max{(Dmp)j, (Dmq)j} = max{dm,jpj, dm,jqj}
1This is crucial here. For example, consider
Then De1 = e1, De2 = e1 + e2, e1 ∨ e2 = e1 + e2, and D(e1 ∨ e2) = 2e1 + e2 is not the same as
(De1) ∨ (De2) = e1 + e2.
D =(cid:18)1 1
0 1(cid:19) .
KMS STATES
5
= dm,j max{pj, qj} = dm,j(p ∨ q)j
= (Dm(p ∨ q))j.
Thus
Vm+1,(Dmp)∨(Dmq) = Vm+1,Dm(p∨q) = Vp∨q,
and (2.5) says that V is Nica covariant.
We next claim that U and V satisfy the commutation relation (2.1). We take
n ∈ Zd and p ∈ Nk, and compute using the commutation relation in Bm+1:
UnVp = Um+1,EmnVm+1,Dmp
= e2πi(Dmp)T θm1 EmnVm+1,DmpUm+1,Emn
= e2πipT (Dmθm1 Em)nVm+1,DmpUm+1,Emn
= e2πipT θmnVpUn using (2.3).
Now the universal property of Bm gives the desired homomorphism πm.
(cid:3)
Remark 2.4. Although we don't think we use this anywhere, the homomor-
phisms πm are in fact injective. One way to see this is to use the Nica covariance
of n 7→ Vm,n to get a homomorphism πVm : T (Nk) → Bθm, and interpret (2.1)
as saying that (πVm, Um) is a covariant representation of a dynamical system
(T (Nk), Zd, γm) in the algebra Bθm. Then Bθm has the universal property which
characterises the crossed product T (Nk) ⋊γm Zd, and we can deduce from the
equivariant uniqueness theorem for the crossed product (for example, [26, Corol-
lary 4.3]) that the representation
of T k(Nk) ⋊γm Zd in T k(Nk) ⋊γm+1 Zd is faithful.
πDm,Em := πVm+1◦Dm
⋊ (Um+1 ◦ Em)
We now define our higher-rank Toeplitz noncommutative solenoid to be the
direct limit
(2.6)
B∞ := lim−→
m∈N
(Bm, πm).
We write πm,∞ for the canonical homomorphism of Bm into B∞. To ease notation
we also write Um,n for the image πm,∞(Um,n) in B∞.
Now we use the vectors rm ∈ (0, ∞)k from our set-up to define the dynamics
we propose to study.
Proposition 2.5. There is a dynamics α : R → Aut B∞ such that
(2.7)
αt(cid:0)Vm,pUm,nV ∗
m,q(cid:1) = eit(p−q)T rm
Vm,pUm,nV ∗
m,q.
Proof. Since Um and V ′
and Vm, there is a dynamics αrm : R → Aut Bm such that
m : p 7→ eitpT rmVm,p satisfy the same relations in Bm as Um
αrm
(Vm,pUm,nV ∗
t = αrm+1
t
We claim that πm ◦ αrm
First, for n ∈ Zd we have
m,q) = eit(p−q)T rm
Vm,pUm,nV ∗
m,q.
◦ πm. To see this, we compute on generators.
αrm+1
t
(πm(Um,n)) = αrm+1
t
(Um+1,Emn) = Um+1,Emn
t (Um,n)).
= πm(Um,n) = πm(αrm
6
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
Second, for p ∈ Nk, and using the relation (2.4) at the crucial step to pass from
rm+1 to rm, we have
αrm+1
t
(πm(Vm,p)) = αrm+1
t
(Vm+1,Dmp) = eit(Dmp)T rm+1
Vm+1,Dmp
= eitpT Dmrm+1
= πm(αrm
t (Vm,p)).
πm(Vm,p) = eitpT rm
πm(Vm,p)
Now the universal property of the direct limit implies that for each t ∈ R, there
is an automorphism αt of B∞ such that αt ◦ πm,∞ = πm,∞ ◦ αrm
. The formula (2.7)
(which implicitly involves the homomorphisms πm,∞) implies that t 7→ αt is a
strongly continuous action of R on B∞.
(cid:3)
t
Our goal is to describe the KMS states of the dynamical system (B∞, α). But
first we pause to establish some conventions about probability measures on inverse
limits.
Remark 2.6. All measures in this paper are positive Borel measures. We view
probability measures on a compact space X as states on the C ∗-algebra C(X) of
continuous functions. We write P (X) for the set of probability measures on X.
When (cid:8)hm : Xm+1 → Xm : m ∈ N(cid:9) is an inverse system of compact spaces
with each hm surjective, the inverse limit lim←−(Xm, hm) is also a compact space.
We write hm,∞ for the canonical map of X∞ := lim←−(Xm, hm) onto Xm, so that
we have hm,∞ = hm ◦ hm+1,∞ for all m ∈ N. The maps hm,∞ induce maps hm,∞∗
on measures:
if µ is a probability measure on X∞, then µm := hm,∞∗(µ) is the
measure on Xm such that
ZXm
f dµm =ZX∞
(f ◦ hm,∞) dµ for f ∈ C(Xm).
Conversely, because each hm is surjective, for any sequence of probability measures
{µm ∈ P (Xm) : m ∈ N} such that µm = hm∗(µm+1) for all m there is a probability
measure µ ∈ P (X∞) such that µm = hm,∞∗(µ) for all m (see [5, Lemma 6.1], for
example). Thus the simplices P (lim←− Xm) and lim←− P (Xm) are canonically isomor-
phic.
To state our main result, we need to observe that, because the entries in the
m on Rd maps Zd into Zd and hence induces a
m of Sd = Rd/Zd onto itself. We show that the KMS states are
m), which
m) on the
Em are integers, multiplication by ET
homomorphism ET
parametrised by the probability measures on the inverse limit lim←−(Sd, ET
is an ordinary solenoid. We write ET
m,∞ for the projection of lim←−(Sd, ET
mth copy of Sd, so that we have
ET
m,∞ = ET
m ◦ ET
m+1,∞ for m ∈ N.
The main theorem of this paper is the following; we prove it at the end of the
paper.
corresponding sequence of probability measures on Sd. For m ∈ N and n ∈ Nd, we
define the moment Mm,n(µ) to be the number
m)(cid:1) and β > 0. Let {µm} be the
Theorem 2.7. Suppose that µ ∈ P(cid:0) lim←−(Sd, ET
e2πixT n dµm(x) =Zlim←−(Sd,ET
Mm,n(µ) =ZSd
m)
e2πiET
m,∞(x)T n dµ(x).
Then there is a KMSβ state ψµ on (B∞, α) such that
KMS STATES
7
(2.8)
ψµ(Vm,pUm,nV ∗
m,q) = δp,qe−βpT rm
βrm
j
βrm
j − 2πi(θT
mn)j
Mm,n(µ).
k
Yj=1
The map µ 7→ ψµ is an affine homeomorphism of P(cid:0) lim←−(Sd, ET
KMSβ(B∞, α) of KMSβ states.
m)(cid:1) onto the simplex
Remark 2.8. As a reality check, we take p = q = 0 and n = 0. Then Vm,pUm,nV ∗
m,q
is the identity 1Bm = 1B∞, and our formula collapses to ψµ(1) = 1.
Remark 2.9. It is interesting to set d = k = 1 and compare the formula (2.8)
with the formula (6.4) in Theorem 6.9 of [6], which on the face of it looks different.
The point is that the integral on the right-hand side of [6, (6.4)] is with respect
to the subinvariant measure associated to the probability measure µ, which in our
notation would be νµm. There is no specific description for this measure in [6]:
they get an isomorphism of the simplex P (lim←−
S) onto the simplex of subinvariant
measures by specifying it on the extreme points (see [6, Lemma 8.2]). We reconcile
the two approaches in Remark 5.3.
3. Equilibrium states on a Toeplitz noncommutative torus
In this section, we fix θ ∈ Mk,d([0, ∞)), and investigate the KMS states on the
Toeplitz noncommutative torus Bθ.
For n ∈ Zd, we write gn for the character on Sd given by gn(x) = e2πixT n, and
ι : C(Sd) → C ∗(Zd) ⊂ Bθ for the isomorphism such that ι(gn) = Un. Then we
have
Bθ = span(cid:8)Vpι(f )V ∗
q : f ∈ C(Sd), p, q ∈ Nk(cid:9).
For y ∈ Rd we define Ry : Sd → Sd by Ry(x) = x + y. Later, we will also write R∗
y
for the automorphism of C(Sd) given by R∗
yf = f ◦ Ry, and Ry∗ for the dual map
on measures defined by
ZSd
f dRy∗(µ) =ZSd
R∗
y(f ) dµ =ZSd
f ◦ Ry dµ.
The assignment y 7→ R∗
each Ry∗ is norm-preserving.
y is a strongly continuous action R of Rd on C(Sd), and
Lemma 3.1. For f ∈ C(Sd) and p ∈ Nk we have
V ∗
(3.1)
and
Vpι(f ) = ι(cid:0)f ◦ R−θT p(cid:1)Vp
p ι(f ) = ι(cid:0)f ◦ RθT p(cid:1)V ∗
p .
Proof. Since C(Sd) = span{gn : x 7→ e2πixT n : n ∈ Zd}, it suffices to check (3.1)
for f = gn. Let n ∈ Zd. Then (2.1) gives
Vp ι(gn) = VpUn = e−2πipT θnUnVp = e−2πipT θnι(gn)Vp.
Since
e−2πipT θngn(x) = e−2πipT θne2πixT n = gn(x − θT p) = (gn ◦ R−θT p)(x),
the first equality follows. The second follows from a similar computation us-
ing (2.2).
(cid:3)
8
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
Remark 3.2. The minus sign in the first identity in (3.1) is crucial. As a reality
check, notice that the signs in the two formulas have to be different, because
V ∗
p Vp = 1 means the ±θT p have to cancel. As a corollary, note that VpV ∗
p , which is
a proper projection, commutes with the ι(f ). (To see that VpV ∗
p 6= 1, we can use
the specific representation of Bθ constructed in the proof of Proposition 3.7(b).)
We now fix r ∈ (0, ∞)k. The universal property of Bθ gives a dynamics αr :
R → Aut Bθ such that
(3.2)
αr
t (Un) = Un
and αr
t (Vp) = eitpT rVp
for n ∈ Zd, p ∈ Nk, t ∈ R.
Then αr
t (VpUnV ∗
q ) = eit(p−q)T rVpUnV ∗
q , and hence
{VpUnV ∗
q : n ∈ Zd, p, q ∈ Nk}
is a set of αr-analytic elements spanning an αr-invariant dense subset of Bθ.
To describe the KMSβ states of (Bθ, αr), it was tempting to apply [3, Theo-
rem 6.1] to the Toeplitz algebra of the commuting homeomorphisms hj : x 7→ x+θj
associated to the rows θj of θ. That result is in several ways more general than
we need, but has an unfortunate hypothesis of rational independence on the set
{rj} which we prefer to avoid.
Proposition 3.3. Suppose that β > 0 and φ is a KMSβ state of (Bθ, αr). Then
φ is a KMSβ state of (Bθ, αr) if and only if
(3.3)
φ(VpUnV ∗
q ) = δp,qe−βpT rφ(Un)
for n ∈ Zd and p, q ∈ Nk.
To prove Proposition 3.3 we need two lemmas. The arguments are based on the
proofs of Lemmas 5.2 and 5.3 in [16].
Lemma 3.4. Suppose that β > 0 and φ is a KMSβ state of (Bθ, αr). If p, q ∈ Nk
satisfy pT r = qT r, then
(a) φ(VpUnV ∗
(b) φ(Vp ι(f )V ∗
p ) = φ(VqUnV ∗
q ) for n ∈ Zd; and
q ) ≤ φ(Vp ι(f )V ∗
p ) for positive f ∈ C(Sd).
Proof. For (a), since Vq is an isometry, we have
and since pT r = qT r the KMS condition gives
φ(VpUnV ∗
φ(VpUnV ∗
q Vq)V ∗
p ) = φ(cid:0)VpUn(V ∗
p ) = e−β(p−q)T rφ(cid:0)VqV ∗
p(cid:1) = φ(cid:0)(VpUnV ∗
q )(VqV ∗
p )(cid:1),
q )(cid:1) = φ(VqUnV ∗
q ).
p (VpUnV ∗
For (b), we take a positive function f in C(Sd). By linearity and continuity,
p ). Using the Cauchy -- Schwarz
q ) = φ(Vp ι(f )V ∗
part (a) implies that φ(Vq ι(f )V ∗
inequality at the second step, we calculate:
φ(Vp ι(f )V ∗
q )2 =(cid:12)(cid:12)φ(cid:0)(Vp ι(pf ))(Vq ι(pf ))∗(cid:1)(cid:12)(cid:12)
p )φ(Vq ι(f )V ∗
q )
p )2.
≤ φ(Vp ι(f )V ∗
= φ(Vp ι(f )V ∗
2
Since both sides are the squares of non-negative numbers, we can take square
roots, and we retrieve (b).
(cid:3)
KMS STATES
9
Lemma 3.5. Suppose that β > 0 and φ is a KMSβ state of (Bθ, αr). Suppose that
p, q ∈ Nk satisfy pT r = qT r and that f ∈ C(Sd). Write P := (p ∨ q) − p. Then
(3.4)
φ(cid:0)Vp ι(f )V ∗
q(cid:1) = φ(cid:0)Vp+lP ι(f ◦ RlθT P )V ∗
q+lP(cid:1)
If p 6= q, then φ(Vp ι(f )V ∗
q ) = 0.
for all l ∈ N.
Proof. We prove (3.4) by induction on l. The base case l = 0 is trivial. Now
suppose that (3.4) holds for l ≥ 0. The inductive hypothesis gives
φ(Vp ι(f )V ∗
Since the dynamics αr fixes the element Vq+lP V ∗
that
q+lP , the KMS condition implies
φ(Vpι(f )V ∗
q+lP Vp+lP ι(f ◦ RlθT P )V ∗
q+lP(cid:1)
q+lP Vq+lP V ∗
q ) = φ(cid:0)Vp+lP ι(f ◦ RlθT P )V ∗
= φ(cid:0)Vp+lP ι(f ◦ RlθT P )V ∗
q ) = φ(cid:0)Vq+lP V ∗
q+lP(cid:1).
q+lP(cid:1),
and Nica covariance gives
φ(Vp ι(f )V ∗
q )
((q+lP )∨(p+lP ))−(p+lP ) ι(f ◦ RlθT P )V ∗
q+lP(cid:1).
For c ∈ Nk we have (p + c) ∨ (q + c) = (p ∨ q) + c. Thus
φ(Vp ι(f )V ∗
= φ(cid:0)Vq+lP V((q+lP )∨(p+lP ))−(q+lP )V ∗
q ) = φ(cid:0)Vq+lP V(p∨q)−qV ∗
= φ(cid:0)V(p∨q)+lP V ∗
q+lP(cid:1)
= φ(cid:0)V(p∨q)+lP ι(f ◦ RlθT P ◦ RθT P )V ∗
= φ(cid:0)Vp+(l+1)P ι(f ◦ R(l+1)θT P )V ∗
q+(l+1)P(cid:1)
P ι(f ◦ RlθT P )V ∗
(p∨q)−p ι(f ◦ RlθT P )V ∗
q+lP(cid:1)
q+(l+1)P(cid:1) by Lemma 3.1
because (p ∨ q) + lP = p + (l + 1)P . This completes the inductive step, and hence
the proof of (3.4).
Now suppose that p 6= q. Then at least one of P and (p ∨ q) − q is nonzero. We
argue the case where P 6= 0, and the other case follows by taking adjoints. For
l ∈ N we have
φ(Vp ι(f )V ∗
q ) =(cid:12)(cid:12)φ(cid:0)Vp+lP ι(f ◦ RlθT P )V ∗
q+lP(cid:1)(cid:12)(cid:12)
≤ φ(cid:0)Vp+lP ι(f ◦ RlθT P )V ∗
p+lP(cid:1) by Lemma 3.4(b)
= e−β(p+lP )T rφ(cid:0)V ∗
= e−β(p+lP )T rφ(cid:0)ι(f ◦ RlθT P )(cid:1)
p+lP Vp+lP ι(f ◦ RlθT P )(cid:1)
≤ e−β(p+lP )T rkf k∞.
Since P > 0 and r ∈ (0, ∞)k, we have (p + lP )T r → ∞ as l → ∞, and hence
e−β(p+lP )T rkf k∞ → 0 as l → ∞. Thus φ(Vp ι(f )V ∗
(cid:3)
q ) = 0.
Proof of Proposition 3.3. First suppose that φ is a KMSβ state for (Bθ, αr). For
n ∈ Zd and p, q ∈ Nk, two applications of the KMS condition give
(3.5)
φ(VpUnV ∗
q ) = e−βpT rφ(UnV ∗
q Vp) = e−β(p−q)T rφ(VpUnV ∗
q ).
10
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
It follows immediately that if (p − q)T r 6= 0, then φ(VpUnV ∗
but p 6= q, then Lemma 3.5 gives φ(VpUnV ∗
equality in (3.5) gives
q ) = 0. If (p − q)T r = 0
q ) = 0. This combined with the first
φ(VpUnV ∗
q ) = δp,qe−βpT rφ(UnV ∗
q Vp) = δp,qe−βpT rφ(Un)
because Vp is an isometry. This is the desired formula (3.3).
Now suppose that φ is a state satisfying (3.3). Since the VpUnV ∗
q are analytic
elements spanning a dense αr-invariant subspace of Bθ, it suffices to fix p, q, b, c ∈
Nk and n, n′ ∈ Zd, and show that
(3.6)
φ(VpUnV ∗
q VbUn′V ∗
c ) = e−β(p−q)T rφ(VbUn′V ∗
c VpUnV ∗
q ).
Let P := (q ∨ b) − b and Q := (q ∨ b) − q. Then P, Q ∈ Nk are the unique
elements such that P ∧ Q = 0 and P + b = Q + q, and Nica covariance says that
V ∗
q Vb = VQV ∗
P . Now we calculate, using first the identities (2.1) and (2.2), and
then (at the last step) the assumption (3.3):
c ) = φ(VpUnVQV ∗
q VbUn′V ∗
φ(VpUnV ∗
P Un′V ∗
c )
(3.7)
= e2πiQT θnφ(Vp(VQUn)V ∗
P Un′V ∗
c )
= e2πi(QT θn+P T θn′)φ(VQ+pUn+n′V ∗
= δQ+p,P +ce−β(Q+p)T re2πi(QT θn+P T θn′)φ(Un+n′).
P +c)
Similarly, let M := (c ∨ p) − p and N := (c ∨ p) − c. Then M, N ∈ Nk are the
unique elements such that M ∧ N = 0 and M + p = N + c, and the right-hand
side of (3.6) is
e−β(p−q)T rφ(VbUn′VN V ∗
M UnV ∗
q )
(3.8)
= e−β(p−q)T re2πi(N T θn′+M T θn)φ(Vb+N Un+n′V ∗
= δN +b,M +qe−β(p−q+b+N )T re2πi(N T θn′+M T θn)φ(Un+n′).
q+M )
To see that (3.7) is equal to (3.8), we first show that the two Kronecker deltas
have the same value. For this, observe that by definition of M, N, P, Q, we have
(P + b) + (N + c) = (Q + q) + (M + p),
and consequently (N + b) − (M + q) = (Q + p) − (P + c). Thus δQ+p,P +c = 1 if
and only if δN +b,M +q = 1. So it now suffices to prove that (3.7) equals (3.8) when
Q + p = P + c and N + b = M + q.
We first claim that M = Q and N = P . By assumption, we have M +q = N +b,
and we have P + b = Q + q by definition of P, Q. Subtracting these equations, we
obtain M − Q = N − P , and rearranging gives M − N = Q − P . Since P ∧ Q = 0
and M ∧ N = 0, we deduce that Q = (Q − P ) ∨ 0 = (M − N) ∨ 0 = M, and then
P = N too, as claimed.
We now have
e2πi(QT θn+P T θn′) = e2πi(M T θn+N T θn′),
and so it remains to check that
e−β(p−q+b+N )T r = e−β(p+Q)T r.
KMS STATES
11
For this, we apply N = P , from above, at the second equality and b + P = q + Q,
by definition of Q, P , at the third to get
(p − q) + (b + N) = p + (b + N − q) = p + (b + P − q)
= p + (q + Q − q) = p + Q,
which gives the result. Thus φ is a KMSβ state.
(cid:3)
Lemma 3.6. Write θj for the jth row of θ. Then the series
(3.9)
e−βpT rRθT p∗
Xp∈Nk
converges in the operator norm of B(C(Sd)) to an inverse forQk
Proof. We first need to understand the sum (3.9), which we want to calculate as
an iterated sum. So we interpret (3.9) as a B(C(Sd))-valued integral over Nk with
respect to counting measure σ (for which all functions on Nk are measurable).
Since each RθT p is norm-preserving, we have
j=1(id −e−βrj RθT
j ∗).
By Tonelli's theorem, we have
Xp∈Nk(cid:13)(cid:13)e−βpT rRθT
∞
k
∞
· · ·
j p∗(cid:13)(cid:13) =
(cid:13)(cid:13)e−βpT rRθT p∗(cid:13)(cid:13) = e−βpT r =
Xp1=0(cid:16)
Xp2=0(cid:16)
Xp2=0(cid:16)
Xpk=0
Xpk=0
Xpk=0
Yj=1
Yj=2
Yj=2
· · ·
· · ·
=
=
k
k
∞
∞
∞
∞
e−βpj rj .
k
Yj=1
∞
e−βpj rj(cid:17)
e−βpj rj(cid:17)(cid:16)
e−βp1r1(cid:17)
e−βpj rj(cid:17)(1 − e−βr1)−1.
Xp1=0
Repeating this k − 1 more times gives
Xp∈Nk(cid:13)(cid:13)e−βpT rRθT
j p∗(cid:13)(cid:13) =
(1 − e−βrj )−1.
k
Yj=1
Thus the function p 7→ e−βpT rRθT
j p∗ is integrable with respect to σ, and Fubini's
theorem for functions with values in a Banach space (for example, [12, Theo-
rem II.16.3]) implies that
e−βpT rRθT p∗ =
Xp∈Nk
=
e−βpj rj Rpj θT
j ∗(cid:17)
∞
k
Xpk=0
Yj=1(cid:16)
∞
k
· · ·
∞
Xp1=0(cid:16)
Yj=1
Xpj=0(cid:0)e−βrj RθT
j ∗(cid:1)pj(cid:17).
12
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
Writing the infinite sum as a limit of partial sums shows that
∞
(3.10)
To simplify the product
Xpj=0(cid:0)e−βrj RθT
j ∗(cid:1)pj(cid:0) id −e−βrj RθT
j ∗(cid:1) = id .
(cid:18) k
Yj=1(cid:16)
∞
Xpj=0(cid:0)e−βrj RθT ∗(cid:1)pj(cid:17)(cid:19)(cid:16)
k
Yj=1(cid:0) id −e−βrj RθT
j ∗(cid:1)(cid:17),
we write the left-hand product from j = k to j = 1, and the right-hand one from
j = 1 to j = k. Now k applications of (3.10) show that the product telescopes to
the identity id of B(C(Sd)).
(cid:3)
The next proposition is an analogue of [16, Theorem 6.1] and [3, Theorem 6.1].
Proposition 3.7. Fix β ∈ (0, ∞).
(a) Suppose that φ is a KMSβ state for (Bθ, αr), and let ν ∈ P (Sd) be the
measure such that
φ(ι(f )) =ZSd
f dν
for f ∈ C(Sd).
Suppose that F ⊂ Nk is a finite set such that p 6= q ∈ F implies p ∧ q = 0.
Then the measure ν satisfies the subinvariance relation
(3.11)
Yp∈F(cid:0) id −e−βpT rRθT p∗(cid:1)(ν) ≥ 0.
(b) Define yβ :=Pp∈Nk e−βpT r, and suppose that κ is a positive measure on Sd
β . Write θj for the jth row of θ. Then
with total mass y−1
k
ν = νκ :=
Yj=1(cid:0) id −e−βrj Rθj ∗(cid:1)−1(κ)
is a subinvariant probability measure, and there is a KMSβ state φν of
(Bθ, αr) such that
(c) The map κ 7→ φνκ is an affine isomorphism of the simplex
(3.12)
φν(cid:0)Vp ι(f )V ∗
f dν
q(cid:1) = δp,qe−βpT rZSd
Σβ,r =(cid:8)positive measures κ : kκk = y−1
β (cid:9)
onto the simplex of KMSβ states of (Bθ, αr).
for p, q ∈ Nk and f ∈ C(Sd).
Proof. (a) We take a positive function f ∈ C(Sd)+ and compute
(3.13)
ZSd
f d(cid:16)Yp∈F
(id −e−βpT rRθT p∗)(ν)(cid:17) =ZSd
(−1)S(cid:16)Yp∈S
=ZSd XS⊂F
f ◦(cid:16)Yp∈F
e−βpT r(cid:17)(cid:16)f ◦Yp∈S
RθT p(cid:17) dν.
(id −e−βpT rRθT p)(cid:17) dν
KMS STATES
13
We write pS :=Pp∈S p, and observe that Qp∈S e−βpT r = e−βpT
RθT pS . Thus
S r and Qp∈S RθT p =
(3.13) =ZSd XS⊂F
= φ(cid:16)XS⊂F
= φ(cid:16)XS⊂F
= φ(cid:16)XS⊂F
(−1)Se−βpT
(−1)Se−βpT
S r(cid:0)f ◦ RθT pS(cid:1) dν
S rι(cid:0)f ◦ RθT pS(cid:1)V ∗
(−1)SVpS ι(cid:0)f ◦ RθT pS(cid:1)V ∗
(−1)SVpS V ∗
pS ι(f )(cid:17) by (3.1).
pS VpS = 1
pS VpS(cid:17) since V ∗
pS(cid:17) by the KMS condition
Because the set F has the property that p ∧ q = 0 for p 6= q ∈ F , Nica covariance
gives VpV ∗
p+q for p 6= q ∈ F . Thus for each S ⊂ F , we
have VpS V ∗
p∨q = Vp+qV ∗
p , and
q = Vp∨qV ∗
p VqV ∗
pS =Qp∈S VpV ∗
XS⊂F
(−1)SVpS V ∗
(1 − VpV ∗
p ).
pS = Yp∈F
The latter product is a projection, and it is fixed by the action α. Hence another
application of the KMS condition gives
ZSd
f d(cid:16)Yp∈F
(id −e−βpT rRθT p∗)(ν)(cid:17) = φ(cid:16)Yp∈F
p )(cid:17)2
ι(f )(cid:17)
p )ι(f )Yp∈F
= φ(cid:16)(cid:16)Yp∈F
= φ(cid:16)Yp∈F
(1 − VpV ∗
(1 − VpV ∗
(1 − VpV ∗
p )ι(f )(cid:17)
(1 − VpV ∗
p )(cid:17).
This last term is positive because the argument of φ is a positive element of Bθ,
and this proves (a).
(b) We have
so ν is subinvariant. By Lemma 3.6 we have
(3.14)
ZSd
e−βpT
Yj=1(cid:0) id −e−βrj Rθj ∗(cid:1)(ν) = κ ≥ 0,
1 dν =ZSd
RθT p∗(κ)(cid:17)
= Xp∈Nk
= Xp∈Nk
1 d(cid:16)Xp∈Nk
e−βpT rZSd
e−βpT rkκk = yβkκk = 1,
1 ◦ RθT p dκ
and hence ν is a probability measure.
We will build a KMSβ state using a representation of Bθ on ℓ2(Nk) ⊗ L2(Sd, κ).
Recall that we write gn for the trigonometric polynomial gn(x) = e2πixT n. Then
the formula Wnf := gnf defines a unitary representation W of Zd on L2(Sd, κ).
14
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
Write {δp : p ∈ Nk} for the orthonormal basis of point masses for ℓ2(Nk), and let
Dn be the bounded operator such that Dnδp := e2πipT θnδp. Then D is a unitary
representation of Zd on ℓ2(Nk), and hence D ⊗ W is a unitary representation of
Zd on ℓ2(Nk) ⊗ L2(Sd, κ).
Let T be the usual Toeplitz representation of Nk by isometries on ℓ2(Nk). Then
we have
and
(Tp ⊗ 1)(Dn ⊗ Wn)(δq ⊗ f ) = e2πiqT θn(δp+q ⊗ Wnf ),
(Dn ⊗ Wn)(Tp ⊗ 1)(δq ⊗ f ) = e2πi(p+q)T θn(δp+q ⊗ Wnf )
= e2πipT θn(Tp ⊗ 1)(Dn ⊗ Wn).
Hence the universal property of Bθ gives a representation
π : Bθ → B(ℓ2(Nk) ⊗ L2(Sd, κ))
such that π(Un) = (Dn ⊗ Wn) and π(Vp) = Tp ⊗ 1.
SincePp∈Nk e−βpT r is convergent, there is a positive linear functional φν : Bθ →
C such that
φν(a) := Xp∈Nk
e−βpT r(cid:0)π(a)(δp ⊗ 1) δp ⊗ 1(cid:1).
Then (3.14) implies that φν(1) = 1, and φν is a state. To see that φν is a KMSβ
state, we take p, q ∈ Nk, n ∈ Zd, and calculate:
(3.15)
φν(Vpι(gn)V ∗
q ) = φν(VpUnV ∗
q )
q δb ⊗ 1) T ∗
= Xb∈Nk
= Xb≥p∨q
= δp,qXb≥p
= δp,q(cid:16)Xb∈Nk
φν(ι(gn)) = φν(Un) =(cid:16)Xb∈Nk
e−βbT r(cid:0)(Dn ⊗ Wn)(T ∗
p δb ⊗ 1(cid:1)
e−βbT r(cid:0)e2πi(b−q)T θnδb−q ⊗ gn δb−p ⊗ 1(cid:1)
e−βbT re2πi(b−p)T θn(cid:0)gn 1(cid:1)
e−β(b+p)T re2πibT θn(cid:17)ZSd
e−βbT re2πibT θn(cid:17)ZSd
gn dκ.
gn dκ.
φν(VpUnV ∗
q ) = δp,qe−βpT rφν(Un),
In particular,
(3.16)
Thus
and φν is a KMSβ state by Proposition 3.3.
From (3.15), we have
φν(cid:0)Un) = Xb∈Nk
= Xb∈Nk
e−βbT rZSd
e−βbT rZSd
e2πi(x+bT θ)T n dκ(x)
gn ◦ RθT b dκ
KMS STATES
15
gn d(cid:16)Xb∈Nk
which by Lemma 3.6 is RSd gn dν. Thus
=ZSd
φν(Vpι(gn)V ∗
e−βbT rRθT b∗(κ)(cid:17),
q ) = δp,qe−βpT rφν(ι(gn)) = δp,qe−βpT rZSd
gn dν.
Since C(Sd) = span{gn : n ∈ Zd}, (3.12) follows from (3.16) and the linearity and
continuity of φν.
(c) We first observe that both maps κ 7→ νκ and ν 7→ φν are affine, and hence
so is the composition. To see that the composition is surjective, we take a KMSβ
state φ, restrict it to the range of ι to get a measure ν, and take
k
κ =
Yj=1(cid:0) id −e−βrj Rθj ∗(cid:1)(ν).
Then the formula (3.3) implies that φ and φνκ agree on the elements Vpι(f )V ∗
q ,
and hence by linearity and continuity on all of Bθ. Thus φ = φνκ. The procedure
which sends φ to κ is weak* continuous and inverts κ 7→ φνκ. Thus it is a con-
tinuous bijection of one compact Hausdorff space onto another, and is therefore a
homeomorphism. Thus so is the inverse κ 7→ φνκ.
(cid:3)
4. The subinvariance relation for the direct limit
We now return to the set-up in which the dynamics α on the direct limit B∞ is
given by a sequence {rm}.
Suppose that φ is a KMSβ state of (B∞, α) and νm are the measures on Sd that
implement the restrictions of φ ◦ πm,∞ to C(Sd) ⊂ Bm. Since the embeddings πm
are all unital, so are the πm,∞. Thus for each m, the restriction φ ◦ πm,∞ is a
KMS1 state of (Bm, αrm), and hence is given by a probability measure νm which
satisfies the subinvariance relations for θ = θm in (3.11) parametrised by subsets
F of {1, . . . , k}. But here, since φ ◦ πm,∞ = φ ◦ πm+l,∞ ◦ πm,m+l for l ∈ N, the
measure νm satisfies a sequence of subinvariance relations parametrised by l as
well as F . Our first main result says that these can be combined into one master
subinvariance relation with real parameters s ∈ [0, ∞)k.
We now describe our continuously parametrised subinvariance relation. For
k = 1 this follows from [6, Definition 6.7 and Theorem 6.9].
Theorem 4.1. Suppose that φ is a KMSβ state on (B∞, α) and m ∈ N. We write
ιm for the inclusion of C(Sd) in Bm, and then
Let νm be the probability measure on Sd such that
ιm(C(Sd)) = span{Um,n : n ∈ Nd}.
(4.1)
φ ◦ πm,∞(ιm(f )) =ZSd
f dνm for f ∈ C(Sd).
Write θm,j for the jth row of the matrix θm. Then for every s ∈ [0, ∞)k, we have
(4.2)
k
Yj=1(cid:0) id −e−βsj rm
j Rsj θT
m,j ∗(cid:1)(νm) ≥ 0.
16
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
We prove Theorem 4.1 at the end of this section. We first need two preliminary
results.
The homomorphism πm : Bm → Bm+1 maps ιm(C(Sd)) into ιm+1(C(Sd)). When
we view ιm(C(Sd)) as span{Um,n}, the homomorphism πm is characterised by
πm(Um,n) = Um+1,Emn;
m : Sd → Sd, and hence we have πm(ιm(f )) = ιm+1(f ◦ ET
when we view ιm(C(Sd)) as {ιm(f ) : f ∈ C(Sd)}, πm is induced by the covering
map ET
m). In particular,
πmC(Sd) is (ET
m. The corresponding map on measures is given by
ET
m)∗ : f 7→ f ◦ ET
m∗:
ZSd
f dπm(ν) =ZSd
f dET
m∗(ν) =ZSd
(f ◦ ET
m) dν.
Lemma 4.2. Suppose that φ is a KMSβ state on (B∞, α). For m ∈ N, let νm be
the probability measure on Sd satisfying (4.1). Then for every finite subset F of
Nk such that p ∧ q = 0 for all p 6= q ∈ F , we have
m p)T rm
Yp∈F(cid:0) id −e−β(D−1
RθT
mD−1
m p∗(cid:1)(νm) ≥ 0.
Proof. We apply Proposition 3.7(a) to the state φ ◦ πm+1,∞ of (Bm+1, αrm+1). We
deduce that
(4.3)
Yp∈F(cid:0) id −e−βpT rm+1
RθT
m+1p∗(cid:1)(νm+1) ≥ 0.
To convert this to a statement about νm, we want to apply ET
side. We first observe that
m∗ to the left-hand
(4.4)
ET
m ◦ RθT
m+1p(x) = ET
= ET
m+1p
mx − ET
mx − θT
mθT
mD−1
m p
m p ◦ ET
m(x).
mD−1
= RθT
using (2.3)
Since ET
(4.3) implies that
m∗ preserves positivity and h 7→ h∗ is covariant with respect to composition,
RθT
mD−1
m∗(νm+1) using (4.4)
0 ≤ ET
m∗(cid:16)Yp∈F(cid:0) id −e−βpT rm+1
=(cid:16)Yp∈F(cid:0) id −e−βpT rm+1
= Yp∈F(cid:0) id −e−βpT (Dm)−1rm
= Yp∈F(cid:0) id −e−β(D−1
m p)T rm
RθT
m+1p∗(cid:1)(νm+1)(cid:17)
m p∗(cid:1)(cid:17) ◦ ET
m p∗(cid:1)(νm)
m p∗(cid:1)(νm).
mD−1
RθT
RθT
mD−1
using (2.4)
(cid:3)
For a positive integer l, we can apply the argument of Lemma 4.2 to the em-
bedding πm,m+l of Bm in Bm+l. This amounts to replacing the matrix Dm with
Dm,m+l := Dm+l−1Dm+l−2 · · · Dm+1Dm, Em with a similarly defined Em,m+l, θm+1
with θm+l, and rm+1 with rm+l. We obtain:
KMS STATES
17
Corollary 4.3. Suppose that φ is a KMSβ state on (B∞, α), and νm is the prob-
ability measure satisfying (4.1). Then for every positive integer l and for every
finite subset F of Nk such that p ∧ q = 0 for all p 6= q ∈ F , we have
m,m+lp)T rm
Yp∈F(cid:0) id −e−β(D−1
RθT
mD−1
m,m+lp∗(cid:1)(νm) ≥ 0.
Proof of Theorem 4.1. For each l ≥ 0 and p ∈ Nk, we can apply Corollary 4.3 to
the finite subset Fp := {pjej : 1 ≤ j ≤ k} of Nk. This gives us the subinvariance
relation
(4.5)
k
Yj=1(cid:0) id −e−β(D−1
m,m+lpj ej)T rm
RθT
mD−1
m,m+lpj ej ∗(cid:1)(νm) ≥ 0.
Each factor in the left-hand side L of (4.5) has the form id −e−sRv∗. Since
(e−sRv∗)(e−tRw∗) = e−(s+t)R(v+w)∗, the product (id −e−sRv∗)(id −e−tRw∗) of two
such terms collapses to
id −e−sRv∗ − e−tRw∗ + e−(s+t)R(v+w)∗.
Thus we can expand
L = id + X∅6=G⊂{1,...,k}
(−1)Ge−β(D−1
m,m+lpG)T rm
RθT
mD−1
m,m+lpG∗(νm),
where pG :=Pj∈G pjej.
For each fixed f ∈ C(Sd) and ν ∈ P (Sd), the function s 7→ R Rs(f ) dν on Rk
is continuous, being the composition of the norm-continuous map s 7→ Rs(f ) and
the bounded functional given by integration against ν. We now consider a positive
function f in C(Sd): we write f ∈ C(Sd)+. For s ∈ [0, ∞)k and G ⊂ {1, . . . , k},
we write sG =Pj∈G sjej. Then
gG : s 7→Z f d(e−βsT
Grm
RθT
msG∗)(νm)
is continuous, and so is the linear combination
L(s) :=Z f d(cid:16) XG⊂{1,...,k}
(−1)Ge−βsT
Grm
RθT
msG∗(cid:17)(νm).
The subinvariance relation (4.5) says that L(s) ≥ 0 for all s of the form Dm,m+lp
for l ≥ 0 and p ∈ Nk.
Since each of the matrices Dm is diagonal with entries dm,j, say, at least 2, we
have
D−1
m,m+lpjej =(cid:16)
l−1
Yn=0
d−1
m+n,j(cid:17)pjej.
are dense in [0, ∞). Thus the vectors s for which L(s) ≥ 0 form a dense subset
of [0, ∞)k, and the continuity of L implies that L(s) ≥ 0 for all s ∈ [0, ∞)k. A
Since dn,j ≥ 2 for all n and j, the rational numbers of the form (cid:0)Ql−1
m+n,j(cid:1)pj
measure ν which hasR f dν ≥ 0 for all f ∈ C(Sd)+ is a positive measure, and this
is what we had to prove.
n=0 d−1
(cid:3)
18
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
5. The solution of the subinvariance relation
We now describe the solutions to the subinvariance relation (4.2). We observe
that the formula on the right of (5.1) below is the Laplace transform of a periodic
function, and as such is given by an integral over a finite rectangle. This obser-
vation motivated our calculations, but in the end we found it easier to work with
the trigonometric polynomials x 7→ e2πinx.
Theorem 5.1. Let θ ∈ Mk,d(Sd), β ∈ (0, ∞) and r = (rj) ∈ (0, ∞)k. Denote the
jth row of θ by θj.
(a) For each µ ∈ M(Sd), there is a nonnegative measure νµ ∈ M(Sd) such that
f dνµ =Z[0,∞)k
e−βwT rZSd
ZSd
and νµ has total mass kµkQk
Yj=1(cid:0) id −e−βsj rj Rsj θT
subinvariance relation
k
f (x + θT w) dµ(x) dw for f ∈ C(Sd),
j=1(βrj)−1. The measure ν = νµ satisfies the
for s ∈ [0, ∞)k.
j ∗(cid:1)(ν) ≥ 0
(5.1)
(5.2)
(5.3)
(b) For each ν satisfying the subinvariance relation (5.2), there is a measure
µν ∈ M(Sd) such that
· · ·
1
lim
s1→0+
f dµν = lim
sk→0+
ZSd
f d(cid:0)Qk
for f ∈ C(Sd), and µν has total mass kνkQk
sk · · · s1 ZSd
j=1(βrj).
j=1(id −e−βsj rj Rsj θT
j ∗)(cid:1)(ν)
(c) The map µ 7→ νµ is an affine homeomorphism of M(Sd) onto the simplex of
measures satisfying the subinvariance relation (5.2), and the inverse takes
ν to µν.
satisfies the analogous relation involving Qj∈J(id −e−βsj rj Rsj θT
e−βsj rj Ry∗. Since the numbers −βrj are negative, the series P∞
Remark 5.2. A measure ν that satisfies the subinvariance relation (5.2) also
j ∗) for any subset
J of {1, . . . , k}. To see this, observe that for any vector y ∈ [0, ∞)d, Ry is an
isometric positivity-preserving linear operator on C(Sd). Hence so are Ry∗ and
n=0 e−βsj rj nRn
y∗
converges in norm in the Banach space of bounded linear operators on M(Sd) to
an inverse for id −e−βsj rj Ry∗. Hence applying this inverse allows us to remove
factors from the subinvariance relation without losing positivity.
Remark 5.3 (Reality check). We reassure ourselves that the description of subin-
variant measures in Theorem 5.1 is consistent with the description in [6, Theo-
rem 7.1]. There d = k = 1, and they describe the simplex of subinvariant proba-
bility measures by specifying the extreme points of the simplex.
We recall that the matrices Dm ∈ M1(N) = N and Em ∈ M1(N) are all the
same integer N ≥ 2, and the sequence θm then satisfies N 2θm+1 = θm. In terms
of our generators, the dynamics α : R → Aut Bm in [6] is given by
αt(Vm,pUm,nV ∗
m,q) = eit(p−q)N −m
Vm,pUm,nV ∗
m,q
KMS STATES
19
(see [6, Proposition 6.3]), which is our αrm with rm = N −m. We are interested
in KMSβ states, so the subinvariant probability measures for (Bm, α) are those in
the set denoted Ωr
m (see [6, Notation 6.8]2).
sub for r = βN −mθ−1
m = βrmθ−1
Since the calculation in [6] is about extreme points, we start with a point mass
δy ∈ P (lim←−
m)∗δy is the point mass µm = δym, where ym is obtained
by realising y as a sequence {ym} satisfying Nym+1 = ym. Then the measure νµm
in Theorem 5.1(a) is given by
S). Then (ET
Z f dνµm =Z ∞
=Z ∞
0
0
e−βwrmZ 1
0
f (x + θmw) dµm(x) dw
e−βwrm
f (ym + θmw) dw.
For f (x) = e2πinx, we get
Z f dνµm = e2πinymZ ∞
0
e−βwrm
e2πinθmw dw,
and a change of variables gives
Z f dνµm = e2πinymZ ∞
m Z ∞
= e2πinymθ−1
0
0
e−βθ−1
m vrm
e2πinvθ−1
m dv
e−(βrmθ−1
m )ve2πinv dv.
Now we recognise the integral as the Laplace transform of the periodic function
x 7→ e2πinx, and hence
(5.4)
Z f dνµm = e2πinymθ−1
m
1
m Z 1
0
1 − e−βrmθ−1
e−(βrmθ−1
m )ve2πinv dv.
In the notation of [6], we set r := βrmθ−1
m , and rewrite (5.4) as
Z f dνµm = e2πinymβ−1N mZ 1
0
r
1 − e−r e−rve2πinv dv
= β−1N mZ 1
0
e2πinv d(Rym)∗(mr)(v).
This shows that the measure νµm is a multiple of the measure (Rym)∗(mr) ap-
pearing in [6, Theorem 7.1]. We are off by the scalar β−1N m because that theorem
is about the simplex of subinvariant probability measures, and the measures νµ in
Theorem 5.1 have total mass (βrm)−1 = β−1N m.
For the proof of Theorem 5.1(a), we need the following lemma, which is known to
probabilists as the inclusion-exclusion principle. We couldn't find a good reference
for this measure-theoretic version, but fortunately it is relatively easy to prove by
induction on the number k of subsets.
2The displayed equation there is meant to say this, as opposed to r = βN −mθm, which is the
way we first read it.
20
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
Lemma 5.4. Suppose that λ is a finite measure on a space X and {Sj : 1 ≤ j ≤ k}
is a finite collection of measurable subsets of X. For each subset G of {1, . . . , k},
we set SG :=Tj∈G Sj. Then
λ(cid:0)Sk
j=1 Sj(cid:1) = X∅6=G⊂{1,...,k}
(−1)G−1λ(SG).
Proof of Theorem 5.1(a). We first claim that there is a positive functional I on
C(Sd) such that I(f ) is given by the right-hand side of (5.1). Indeed, the estimate
e−βwT rZSd
(cid:12)(cid:12)(cid:12)Z[0,∞)k
f (x + θT w) dµ(x) dw(cid:12)(cid:12)(cid:12)
≤Z[0,∞)k
e−βwT rZSd
kf k∞dµ(x)dw,
shows that the right-hand side of (5.1) determines a bounded function I : C(Sd) →
C. This function I is linear because the integral is linear, and f ≥ 0 implies
I(f ) ≥ 0 because all the integrands in (5.1) are non-negative. Thus there is a
finite nonnegative measure νµ satisfying (5.1). The norm of the integral is given
by the total mass of the measure νµ, which is
ZSd
1 dνµ =Z[0,∞)k
e−βwT rkµk dw.
To compute the exact value of the integral, observe that
e−βwT r = e−β Pk
j=1 wj rj =
e−βwj rj .
k
Yj=1
Yj=1Z ∞
0
k
Thus
kνµk = kµkZ[0,∞)k
k
Yj=1
e−βwj rj dw = kµk
e−βwj rj dwj = kµk
(βrj)−1.
k
Yj=1
This proves the assertions in the first sentence of part (a).
For the subinvariance relation, we fix f ∈ C(Sd)+, and aim to prove that
k
ZSd
f d(cid:16)
Yj=1(cid:0) id −e−βsj rj Rsj θj ∗(cid:1)(νµ)(cid:17) ≥ 0.
As in the proof of Theorem 4.1, we write
k
Yj=1(cid:0) id −e−βsj rj Rsj θj∗(cid:1) = id + X∅6=G⊂{1≤j≤k}
(−1)Ge−βsT
GrRθT sG∗,
with sG := Pj∈G sjej. For j ≤ k we define Sj = {v ∈ [0, ∞) : vj ≥ sj}, and
SG :=Tj∈G Sj. Then
f d(cid:0)e−βsT
ZSd
e−βsT
(5.5)
e−βwT re−βsT
f (x + θT w + θT sG) dµ(x) dw
Gr(cid:0)f ◦ RθT sG(cid:1) dνµ
GrZSd
GrRθT sG∗(cid:1)(νµ) =ZSd
=Z[0,∞)k
=ZSG
e−βvT rZSd
f (x + θT v) dµ(x) dv.
Since f is fixed, we can define a measure m on [0, ∞)k by
KMS STATES
21
Now (5.5) says that
Z[0,∞)k
g dm =Z[0,∞)k
ZSd
f d(cid:0)e−βsT
k
Thus
ZSd
f d(cid:16)
g(v)e−βvT rZSd
f (x + θT v) dµ(x) dv.
GrRθT sG∗(cid:1)(νµ) = m(SG).
By the inclusion-exclusion principle, this is
Yj=1(cid:0) id −e−βsj rj Rsj θj∗(cid:1)(νµ)(cid:17) = m([0, ∞)k) + X∅6=G⊂{1≤j≤k}
j=1[0, sj)(cid:1) ≥ 0.
m([0, ∞)k) − m(cid:0)Sk
j=1 Sj(cid:1) = m(cid:0)Qk
We now move towards a proof of part (b), and for that the first problem is
to prove that the iterated limit in (5.3) exists. We will work with l satisfying
1 ≤ l ≤ k, and show by induction on l that the iterated limit
(−1)Gm(SG).
(cid:3)
lim
sl→0+
· · ·
lim
s1→0+
exists. We will be doing some calculus, so we often assume that our test functions
f belong to the dense subalgebra C ∞(Sd) of C(Sd) consisting of smooth functions
all of whose derivatives are also periodic.
We begin by establishing that, even after dividing by the numbers which are
going to 0, the norms of the measures remain uniformly bounded.
Lemma 5.5. Suppose that ν is a finite positive measure on Sd satisfying the
subinvariance relation (5.2). Then for each s ∈ (0, ∞)k,
λs :=
1
sksk−1 · · · s1
is a positive measure with total mass
k
Yj=1(cid:0) id −e−βsj rj Rsj θT
j ∗(cid:1)(ν)
(5.6)
kλsk ≤(cid:16)
k
(βrj)(cid:17)kνk.
Yj=1
Proof. The subinvariance relation implies that the measure is positive. For the
estimate on the total mass of λs, we deal with the variables si separately. So for
1 ≤ l ≤ k, we set
k
which by Remark 5.2 are all positive measures. We have
k
ZSd
1 d(cid:16)
Yj=1(cid:0) id −e−βsj rj Rsj θT
σl :=
j ∗(cid:1)(ν),
Yj=l(cid:0) id −e−βsj rj Rsj θT
j ∗(cid:1)(ν)(cid:17) =ZSd
=ZSd
1 dσ1
1 d(cid:0)(id −e−βs1r1Rs1θT
1 ∗)(σ2)(cid:1)
22
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
=ZSd
=ZSd
j ∗(cid:1)(ν)(cid:17) =ZSd
1 ◦(cid:0) id −e−βs1r1Rs1θT
(1 − e−βs1r1) dσ2.
1(cid:1) dσ2
1 − e−βs1r1
s1
dσ2.
So for all s1 > 0,
(5.7)
1
s1 ZSd
The integrand here is
k
1 d(cid:16)
Yj=1(cid:0) id −e−βsj rj Rsj θT
1 − e−βs1r1
s1
=
f (0) − f (s1)
s1
for f (s1) := e−βs1r1.
Hence for each fixed s1 > 0, the mean value theorem implies that there exists
c ∈ (0, s1) such that
1 − e−βs1r1
s1
= −f ′(c) = −(−βr1e−βcr1),
which is a positive number less than βr1. Thus (5.7) is at most βr1kσ2k.
Now we repeat this argument, first to see that
1
s2 ZSd
1 d(cid:0) id −e−βs2r2Rs2θT
2 ∗(cid:1)(σ3)
has mass at most βr2kσ3k. After k − 2 more steps, we arrive at the estimate (5.6).
(cid:3)
Lemma 5.6. Suppose that 1 ≤ j ≤ k and that λ ∈ M(Sd) satisfies
Then for all f ∈ C ∞(Sd), we have
(cid:0) id −e−βsrj RsθT
(5.8)
f d(id −e−βsrj RsθT
lim
s→0+(cid:16)1
sZSd
for all s > 0.
j ∗(cid:1)(λ) ≥ 0
j ∗)(λ)(cid:17) = βrjZSd
f dλ −ZSd
θT
j (∇f ) dλ.
Proof. Let g : Sd × R → C be the function g(x, s) = e−βsrj f (x + sθT
on the left of (5.8) can be rewritten as
j ). The term
1
sZSd(cid:0)f (x) − e−βsrj f (x + sθT
j )(cid:1) dλ(x) =
1
sZSd
(g(x, 0) − g(x, s)) dλ(x).
So we want to show that the function G defined by G(s) := RSd g(x, s) dλ(x) is
differentiable at 0 with −G′(0) equal to the right-hand side of (5.8).
We compute
∂g
∂s
(x, s) = −βrje−βsrj f (x + sθT
j ) + e−βsrj θT
j ∇f (x + sθT
j ).
The Cauchy-Schwarz inequality for the inner product θT
gives
j (∇f ) = (θj ∇f ) then
(5.9)
≤ βrjkf k∞ + kθT
j k2k∇f (x + sθT
j )k2.
The right-hand side is uniformly bounded on Sd, and hence there is an integrable
function on Sd that dominates the right-hand side for all s ∈ [0, 1], say. Thus we
∂g
∂s
(cid:12)(cid:12)(cid:12)
(x, s)(cid:12)(cid:12)(cid:12)
KMS STATES
23
can differentiate under the integral sign, using Theorem 2.27 of [14], for example.
We deduce that G is differentiable on [0, 1] with derivative
G′(s) =ZSd(cid:0) − βrje−βsrj f (x + sθT
j ) + e−βsrj θT
j ∇f (x + sθT
j )(cid:1) dλ(x).
Taking s = 0 gives the negative of the right-hand side of (5.8), as required.
(cid:3)
Our next step is the inductive argument, which is quite a complicated one. As
a point of notation, for each tuple I = {i1, . . . , im} with entries in {1, 2, . . . , k},
and for f ∈ C ∞(Sd), we write I := m and DIf for the partial derivative
DIf :=
∂mf
∂xi1∂xi2 · · · ∂xim
.
Lemma 5.7. Suppose that ν is a positive measure on Sd satisfying the subinvari-
ance relation (5.2). Let 1 ≤ l ≤ k.
(a) The iterated limit
lim
sl→0+
· · ·
lim
s1→0+
exists for all f ∈ C(Sd).
1
sl · · · s1 ZSd
f d(cid:0)Ql
j=1(id − e−βsj rj Rsj θT
j ∗)(cid:1)(ν)
(b) Write
n=0{1, . . . , k}n.
Σl :=Sl
Then there are real scalars {K l
j=1(βrj),
and: for every f ∈ C ∞(Sd) and for every measure ν on Sd satisfying the
subinvariance relation (5.2), the limit in (a) is
I : I ∈ Σl} such that K l
∅ = Ql
(5.10)
ZSd(cid:16)XI∈Σl
K l
IDIf(cid:17) dν.
Proof. We prove by induction on l that the limit in (a) exists for every f ∈ C ∞(Sd),
and that there exist the scalars K l
I. Then, since we know from Lemma 5.5 that
the measures λs are norm-bounded by(cid:0)Qk
in C(Sd), we get convergence in (a) also for f ∈ C(Sd).
j=1(βrj)(cid:1)kνk and C ∞(Sd) is norm dense
When l = 1, the index set Σ1 consists of the empty set ∅ and the one-point sets
{j}. Lemma 5.6 implies that K 1
∅ = βr1 and K 1
{j} = θT
1 ej = θ1j.
We fix l between 1 and k − 1, and suppose as our inductive hypothesis that for
every measure λ such that
l
(5.11)
Yj=1(cid:0) id −e−βsj rj Rsj θT
j ∗(cid:1)(λ) ≥ 0 for all s ∈ [0, ∞)k,
I} parametrised by I ∈ Σl. We now have to start with a
we have such scalars {K l
measure κ that satisfies
l+1
(5.12)
and find suitable scalars K l+1
.
I
Yj=1(cid:0) id −e−βsj rj RθT
j ∗(cid:1)(κ) ≥ 0 for all s ∈ [0, ∞)k,
24
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
We define
Remark 5.2 reassures us that λ is another positive measure, and (5.12) implies
that it satisfies (5.11). The induction hypothesis gives
λ :=(cid:0) id −e−βsl+1rl+1Rsl+1θT
l+1∗(cid:1)(κ).
L(sl+1) := lim
sl→0+
· · ·
lim
s1→0+
1
sl+1 · · · s1 ZSd
f d(cid:16)
Lemma 5.6 implies that
l+1
=
1
K l
Yj=1(cid:0) id −e−βsj rj Rsj θT
sl+1(cid:18)ZSd(cid:16)XI∈Σl
l+1∇(cid:16)XI∈Σl
Xi=1
j ∗(cid:1)(cid:17)(κ)
IDIf(cid:17) dλ(cid:19).
IDIf(cid:17) dλ
∂xi (cid:17) dλ(x).
Iθl+1,i
∂DI f
K l
K l
θT
d
lim
sl+1→0+
K l
L(sl+1) = βrl+1ZSd(cid:16)XI∈Σl
= βrl+1ZSd(cid:16)XI∈Σl
IDIf(cid:17) dλ −ZSd
IDIf(cid:17) dλ −ZSd(cid:16)XI∈Σl
K l
To finish off the inductive step, we set K l+1
set
∅ = βrl+1K l
∅, and for I ′ = (I, iI+1) we
K l+1
I ′ =(K l
Iθ(l+1)il+1
βrl+1K l
I ′ − K l
Iθl+1,iI+1
if I = l
if I < l.
This completes the inductive step, and hence the proof.
(cid:3)
Proof of Theorem 5.1(b). Lemma 5.7 shows that the limit exists for all f ∈ C(Sd),
and for f ∈ C ∞(Sd) gives us a formula for the limit. The limit is linear in f , posi-
tive when f is, and is bounded by kf k∞(cid:0)Qk
nite positive measure µν. Since the total mass of the measure is integration against
the constant function 1, and since 1 is smooth, the total mass is given by (5.10).
But since all derivatives of 1 are zero, the only nonzero terms are the ones on
which I = ∅. Now the formula for K k
j=1(βrj)(cid:1)kνk. Thus it is given by a fi-
∅ implies that kµνk =(cid:0)Qk
j=1(βrj)(cid:1)kνk. (cid:3)
We now work towards the proof of Theorem 5.1(c). To prove that N : µ 7→ νµ
is a bijection of the measures arising from KMSβ states onto the subinvariant
measures, we prove that N is one-to-one and that M : ν 7→ µν satisfies N ◦M(ν) =
ν for all subinvariant measures ν. We then have N ◦(M ◦N)(µ) = (N ◦M)◦N(µ) =
N(µ), and injectivity of N implies (M ◦ N)(µ) = µ. Thus Theorem 5.1(c) follows
from the following proposition.
Proposition 5.8. Suppose that ν is a measure on Sd satisfying the subinvariance
relation (5.2) and with total mass Qk
Suppose that ν is a subinvariant measure and f ∈ C ∞(Sd). We need to show
that the functional defined by integrating against νµν , which is defined in parts
(a) and (b) of the theorem as
(5.13)
j=1(βrj)−1. Then ν = νµν .
Z[0,∞)k
e−βwT r lim
s→0+
1
sk · · · s1 ZSd
f (x + θT w) d(cid:16)
k
Yj=1(cid:0) id −e−βsj rj Rsj θT
j ∗(cid:1)(cid:17)(ν)(x)dw,
KMS STATES
25
is in fact implemented by ν. We will do this by peeling off the iterated limit one
variable at a time. For this, the next lemma is crucial.
Lemma 5.9. Consider a positive measure λ on Sd, b ∈ (0, ∞) and v ∈ Sd. For
f ∈ C ∞(Sd), we have
Z ∞
0
e−bt lim
s→0+
1
sZSd
f (y + tv) d(cid:0)(id −e−bsRsv∗)λ(cid:1)(y) dt
s→0+Z ∞
= lim
e−bt
0
s ZSd
f (y + tv) d(cid:0)(id −e−bsRsv∗)λ(cid:1))(y) dt.
sZSd(cid:0)f (y +tv)−e−bsf (y +tv +sv)(cid:1) dλ(y).
1
Proof. For s > 0, we have
1
We write this last integrand as
sZSd
f (y +tv) d(cid:0)(id −e−bsRsv∗)λ(cid:1)(y) =
K(y,s, t) = s−1(cid:0)f (y + tv) − e−bsf (y + tv + sv)(cid:1)
= s−1(cid:0)f (y + tv) − e−bsf (y + tv) + e−bsf (y + tv) − e−bsf (y + tv + sv)(cid:1)
f (y + tv) + e−bs f (y + tv) − f (y + tv + sv)
1 − e−bs
=
.
s
s
We estimate the first summand using the mean value theorem on e−bs, and the
second summand using the same theorem on f , to find
K(y, s, t) ≤ bkf k∞ + kvT (∇f )k∞.
Thus we have
e−bt
s ZSd
(cid:12)(cid:12)(cid:12)
K(y, s, t) dλ(y)(cid:12)(cid:12)(cid:12)
≤ e−bt(cid:0)bkf k∞ + kvT (∇f )k∞(cid:1)kλk.
Now the result follows from the dominated convergence theorem for Lebesgue
measure on [0, ∞) (modulo the trick of observing that it suffices to work with
sequences sn → 0+ -- see the proof of [14, Theorem 2.27]).
(cid:3)
Proof of Proposition 5.8. As in the proof of Theorem 5.1(b), Lemma 5.7 implies
that there is a positive measure η on Sd such that for g ∈ C ∞(Sd), we have
ZSd
g dη =
lim
sk,...,s2→0+
1
sk · · · s2 ZSd
g d(cid:16)
k
Yj=2(cid:0) id −e−βsj rj Rsj θT
j ∗(cid:1)(cid:17)(ν).
Since the operators id −e−βsj rj Rsj θT
j ∗ commute with each other,
(5.14)
ZSd
g d(cid:0) id −e−βs1r1Rs1θT
1 ∗(cid:1)(η)
sk · · · s2 ZSd
1
=
lim
sk,...,s2→0+
k
g d(cid:16)
Yj=1(cid:0) id −e−βsj rj Rsj θT
j ∗(cid:1)(cid:17)(ν).
Now we need some complicated notation to implement the peeling process. First
of all, we fix f ∈ C ∞(Sd). In an attempt to avoid an overdose of subscripts, we
write s = (s1, s), w = (w1, w) and r = (r1, r). We also write θ for the k − 1 × d
26
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
matrix obtained from θ by deleting the first row: thus θT has block form (θT
1
With the new notation, (5.13) becomes
θT ).
e−βr1w1×
Z[0,∞)k−1
0
e−β wT rZ ∞
s1 ZSd
× lim
s1→0+
1
Now we can apply Lemma 5.9 to the inside integrals, which gives
f (x + θT w + w1θT
1 ) d(cid:0) id −e−βr1w1Rs1θT
1 ∗(cid:1)(η)(x)dw1d w.
(5.15)
ZSd
f dνµν =Z[0,∞)k−1
×ZSd
e−β wT r
lim
s1→0+
1
s1 Z ∞
0
e−βr1w1×
f (x + θT w + w1θT
1 ) d(cid:0) id −e−βr1w1Rs1θT
1 ∗(cid:1)(η)(x)dw1d w.
We now consider the function g on (0, ∞) defined by
1
s1 Z ∞
e−βr1w1ZSd
0
g(s1) :=
f (x + θT w + w1θT
1 ∗(cid:1)(η)(x)dw1.
We aim to prove that g(s1) → RSd f (x + θT w) dη(x) as s1 → 0+. To this end, we
1 ) d(cid:0) id −e−βr1w1Rs1θT
compute:
g(s1) =
1
s1 Z ∞
0
Changing the variable in the second integral to get an integral over [s1, ∞) gives
1
−
e−βr1w1ZSd
s1 Z ∞
s1 Z s1
1
0
0
g(s1) =
f(cid:0)x + θT w + w1θT
e−βr1(w1+s1)ZSd
1(cid:1) dη(x)dw1
f(cid:0)x + θT w + (w1 + s1)θT
1(cid:1) dη(x)dw1.
f(cid:0)x + θT w + w1θT
e−βr1w1ZSd
1(cid:1) dη(x)dw1.
0
1
f (x + θT w + w1θT
s1 Z s1
e−βr1w1ZSd
s1 Z s1
0 ZSd(cid:0)e−βr1w1f (x + θT w + w1θT
1 ) dη(x)dw1 −ZSd
1 ) − f (x + θT w)(cid:1) dη(x)dw1.
1
f (x + θT w) dη(x)
Since y 7→ e−βyT rf (x + θT w + θT y) is uniformly continuous, there exists δ such
that
So for 0 ≤ s1 < δ we have
1 ) − f (x + θT w)(cid:12)(cid:12) <
0 ≤ w1 < δ =⇒(cid:12)(cid:12)e−βr1w1f (x + θT w + w1θT
s1 Z s1
0 ZSd
g(s1) −ZSd
(cid:12)(cid:12)(cid:12)
Thus g(s1) →RSd f (x + θT w) dη(x), as we wanted.
f (x + θT w) dη(x)(cid:12)(cid:12)(cid:12)
kηk
≤
1
ǫ
dη(x)dw1 = ǫ.
ǫ
kηk
.
f (x + θT w) dη(x)
Now we have
g(s1)−ZSd
=
=
Putting the formula for lims1→0+ g(s1) in (5.15) gives
KMS STATES
27
ZSd
f dνµν =Z[0,∞)k−1
e−β wT rZSd
f (x + θT w) dη(x) d w,
which is the right-hand side of (5.13) with one lims→0+ and one R ∞
Repeating the argument k − 1 times gives
0
removed.
(cid:3)
as required.
ZSd
f dνµν =ZSd
f dν,
As described before Proposition 5.8, this completes the proof of Theorem 5.1.
6. A parametrisation of the equilibrium states
We are now ready to describe the KMS states of our system. At the end of the
section, we will use the following theorem to prove our main result.
Theorem 6.1. Consider our standard set-up, and suppose that β > 0.
(a) Suppose that µ ∈ P(cid:0) lim←−(Sd, ET
m,∞∗(µ) and take νµm to be the subinvariant measure on Sd obtained by
ET
applying Theorem 5.1 to the measure µm. Then there is a KMSβ state ψµ
of (B∞, α) such that
m)(cid:1). Define measures µm ∈ P (Sd) by µm =
(6.1)
ψµ(Vm,pUm,nV ∗
m,q) = δp,qe−βpT rm
k
Yj=1
(βrm
j )ZSd
e2πixT n dνµm(x).
(b) The map µ 7→ ψµ is an affine homeomorphism of P(cid:0) lim←−(Sd, ET
KMSβ(B∞, α).
m)(cid:1) onto
To prove the theorem, we first build some maps between the spaces of subin-
variant measures. We will make use of Theorem 5.1, but the measures described
there are not all normalised. To ensure we are dealing with probability measures,
we introduce the numbers
k
k
cm :=
Yj=1
(βrm
j ) = βk(cid:0)
Yj=1
rm
j (cid:1)
and
dm := det Dm.
Because Dm is diagonal, Equation 2.4 shows that the two sets of numbers are
related by dmcm+1 = cm.
In particular, the functions fβ,rm from Remark ?? have constant value cm, and
m fβ,rm. Thus with Σβ,r from Proposition 3.7(c), we can define
so fβ,rm+1 = d−1
σm : Σβ,rm+1 → Σβ,rm by
σm(ν) = d−1
m ET
m∗(ν).
Lemma 6.2. Suppose that µ ∈ P(cid:0) lim←−(Sd, ET
m)(cid:1), and define µm := ET
m ≥ 1. Then the measures νµm given by Theorem 5.1 satisfy σm(νµm+1) = νµm.
m,∞∗(µ) for
Proof. We take f ∈ C(Sd), and compute using (5.1):
ZSd
f dσm(νµm+1) = d−1
m ZSd
f ◦ ET
m dνµm+1
28
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
= d−1
= d−1
m Z[0,∞)k
m Z[0,∞)k
m Z[0,∞)k
e−βwT rm+1ZSd
e−βwT rm+1ZSd
m rmZSd
e−βwT D−1
= d−1
(f ◦ ET
m)(x + θT
m+1w) dµm+1(x) dw
f (ET
mx + ET
mθT
m+1w) dµm+1(x) dw
f (ET
mx + θT
mD−1
m w) dµm+1(x) dw,
where at the last step we used3 both (2.4) and (2.3). Now substituting v = D−1
in the outside integral gives
m w
(6.2)
ZSd
f dσm(νµm+1) =Z[0,∞)k
e−βvT rm+1ZSd
f (ET
mx + θT
mv) dµm+1(x) dv.
We write s := θT
by τs(f )(x) = f (x + s). Then the inside integral on the right of (6.2) is
mv and consider the translation automorphism τs of C(Sd) defined
f (ET
mx + θT
ZSd
mv) dµm+1(x) =ZSd
=ZSd
=ZSd
(τs(f ) ◦ ET
m) dµm+1
τs(f ) d(ET
m)∗(µm+1)
f (x + θT
mv) dµm(x).
Putting this back into the double integral in (6.2) gives the right-hand side of (5.1)
for the measure µm, and we deduce from (5.1) that
ZSd
f dσm(νµm+1) =ZSd
f dνµm for all f ∈ C(Sd),
(cid:3)
as required.
Sd → Sd
are surjective, each µm is a probability measure on Sd. Thus we deduce from
Proof of Theorem 6.1(a). Since the maps
ET
m,∞ : lim←−
Theorem 5.1 that νm := (cid:0)Qk
j )(cid:1)νµm is a probability measure satisfying the
(Bm, αrm) such that ψm(Um,n) = RSd e2πixT n dνm(x). We now need to check that
ψm+1 ◦ πm = ψm so that we can deduce from [6, Proposition 3.1] that the ψm
combine to give a KMSβ state of (B∞, α).
subinvariance relation (5.2). Thus Proposition 3.7(b) gives a KMS state ψm of
j=1(βrm
Since we are viewing measures as functionals on C(Sd), the map ET
m∗ on M(Sd)
mx on Sd. Then for f ∈ C(Sd)
is induced by the continuous function ET
m : x 7→ ET
(6.3)
ZSd
f ◦ ET
m(x) dνm+1(x) =ZSd
f (x) dET
m∗(νm+1)(x).
For the functions gn ∈ C(Sd) given by gn(x) = e2πinT x (so that ιm(gn) = Um,n ∈
Bm), we have
gn ◦ ET
m(x) = e2πi(ET
mx)T n = e2πixT Emn = gEmn(x).
3This also uses that Dmθm+1Em = θm on the nail, i.e. as opposed to modulo Z. Otherwise the
difference would appear in the last formula multiplied by the real variable w.
Substituting this on the left-hand side of (6.3) gives
KMS STATES
29
(6.4)
ZSd
gEmn(x) dνm+1(x) =ZSd
Using again dm = det Dm and cm = Qk
Lemma 6.2 gives
gn(x) dET
m∗(νm+1)(x).
j=1(βrm
j ) and the relation dmcm+1 = cm,
(6.5)
ET
m∗(νm+1) = σm(dmνm+1) = dmcm+1σm(νµm+1)
= dmcm+1νµm = cmνµm = νm.
Using (6.4) at the third step, and (6.5) at the fourth step, we now calculate:
ψm+1(πm(Un,m)) = ψm+1(Um+1,Emn) =ZSd
m∗(νm+1) =ZSd
=ZSd
gn dET
gEmn dνm+1
gn dνm = ψm(Um,n).
Thus the states ψm give an element (ψm) of the inverse limit lim←− KMSβ(Bm, αrm),
and surjectivity of the isomorphism in [6, Proposition 3.1] gives a KMSβ state ψµ
of (B∞, α) such that ψm = ψµ ◦ πm,∞ for m ≥ 1.
(cid:3)
Remark 6.3. We observe that the KMSβ state of Theorem 6.1(a) is given on
B∞ = span{Vm,pUm,nVm,q∗} by
ψµ(Vm,pUm,nV ∗
m,q) = δp,qe−βpT rmZSd
gn d(νET
m,∞∗(µ)).
Proof of Theorem 6.1(b). We first prove that every KMSβ state has the form ψµ.
So suppose that φ is a KMSβ state of (B∞, α). Then for each m ≥ 1, φ ◦ πm,∞ is a
KMSβ state of (Bm, αrm), and hence there are probability measures νm such that
φ ◦ πm,∞(f ) =ZSd
f dνm for all f ∈ C(Sd)
m∗(νm+1) = νm for all m ≥ 1. Theorem 4.1 implies that each νm satisfies
m (Sd) and
m (Sd) for the set of measures and the set of probability measures satisfying (4.2).
and ET
the corresponding subinvariance relation. More specifically, we write M sub
P sub
Then we have νm ∈ P sub
m (Sd).
Once more using dm = det Dm and cm = Qk
orem 5.1(a) gives a function µ 7→ cmνµ from M(Sd) to the simplex M sub
Lemma 6.2 gives commutative diagrams
j ), the construction of The-
m (Sd).
j=1(βrm
µm
M(Sd)
cmνµm
M sub
m (Sd)
ET
m∗
ET
m∗
M(Sd)
µm+1
M sub
m+1(Sd)
cm+1νµm+1,
30
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
and Theorem 5.1 implies that the vertical arrows are bijections. A simple set-
theoretic argument then implies that we also have commutative diagrams
c−1
m µνm
P (Sd)
νm
P sub
m (Sd)
ET
m∗
ET
m∗
P (Sd)
c−1
m+1µνm+1
P sub
m+1(Sd)
νm+1.
Thus the sequence (µm) := (c−1
and hence is given by a probability measure µ ∈ P(cid:0) lim←−(Sd, ET
that φ = ψµ. Since both are states, it suffices to check that they agree on ele-
ments Vm,pUm,nV ∗
m,q. Since φ is a KMSβ state and the measure νm implements φ
on C(Sd) = span{Um,n}, we have
m µνm) belongs to the inverse limit lim←−(cid:0)P (Sd), ET
m∗(cid:1),
m)(cid:1). We want to show
e2πixT n dνm(x).
φ(cid:0)Vm,pUm,nV ∗
Since ν = νµν for all ν and µνm = cmµm, we have νm = cmνµm and
m,q(cid:1) = δp,qe−βpT rmZSd
cmZSd
m,q(cid:1) = δp,qe−βpT rm
This is precisely the formula for ψµ(cid:0)Vm,pUm,nV ∗
m,q(cid:1) in (6.1). Thus φ = ψµ.
Since each ψµ is a state, it follows from the formula (6.1) that µ 7→ ψµ is
affine and weak* continuous from M(Sd) = C(Sd)∗ to the state space of B∞. The
formula (6.1) also implies that µ 7→ ψµ is injective, and since we have just shown
that it is surjective, we deduce that it is a homeomorphism of the compact space
(cid:3)
φ(cid:0)Vm,pUm,nV ∗
e2πixT n dνµm(x).
P(cid:0) lim←−(Sd, ET
n )(cid:1) onto the simplex of KMSβ states of (B∞, α).
Proof of Theorem 2.7. According to (6.1) in Theorem 6.1, we have to compute
Z[0,∞)k
e2πixT n dνµm(x),
which by Theorem 5.1 is
(6.6)
Z[0,∞)k
e−βwT rmZSd
e2πi(x+θT
mw)T n dµm(x)dw
e−βwT rm
e2πiwT θmnZSd
e2πixT n dµm(x)dw
e−βwT rm
e2πiwT θmnMm,n(µ) dw.
=Z[0,∞)k
=Z[0,∞)k
We can rewrite the integrand as
e−βwT rm
e2πiwT θmn = ePk
j=1 wj(−βrm
j +2πi(θmn)j ) =
ewj(−βrm
j +2πi(θmn)j ).
k
Yj=1
KMS STATES
31
When we view R[0,∞)k dw as an iterated integral, we find that
k
(6.6) =
Since β > 0 and each rm
Thus
(6.6) =
k
Yj=1
and the result follows from (6.1).
j +2πi(θmn)j)Mm,n(µ) dwj(cid:17).
j ) → 0 as wj → ∞.
0
ewj (−βrm
j > 0, we have
Yj=1(cid:16)Z ∞
j +2πi(θmn)j )(cid:12)(cid:12) = ewj(−βrm
Mm,n(µ)(cid:12)(cid:12)(cid:12)(cid:12)
∞
=
0
(cid:12)(cid:12)ewj(−βrm
ewj (−βrm
−βrm
j +2πi(θmn)j )
j + 2πi(θmn)j
1
βrm
j − 2πi(θmn)j
Mm,n(µ),
k
Yj=1
(cid:3)
References
[1] Z. Afsar, N. Brownlowe, N.S. Larsen and N. Stammeier, Equilibrium states on right LCM
semigroup C ∗-algebras, Int. Math. Res. Not. IMRN, to appear; doi:10.1093/imrn/rnx162.
[2] Z. Afsar, A. an Huef and I. Raeburn, KMS states on C ∗-algebras associated to local home-
omorphisms, Internat. J. Math. 25 (2014), article no. 1450066, 28 pp.
[3] Z. Afsar, A. an Huef and I. Raeburn, KMS states on C ∗-algebras associated to a family of
∗-commuting local homeomorphisms, J. Math. Anal. Appl. 464 (2018), 965 -- 1009.
[4] Z. Afsar, N.S. Larsen and S. Neshveyev, KMS states on Nica-Toeplitz algebras, preprint,
arXiv:1807.05822.
[5] L.W. Baggett, N.S. Larsen, J.A. Packer, I. Raeburn and A. Ramsay, Direct limits, multires-
olution analyses, and wavelets, J. Funct. Anal. 258 (2010), 2714 -- 2738.
[6] N. Brownlowe, M. Hawkins and A. Sims, The Toeplitz noncommutative solenoid and its
Kubo-Martin-Schwinger states, Ergod. Theory Dyn. Syst. 39 (2019), 105 -- 131.
[7] J. Christensen, Symmetries of the KMS simplex, Comm. Math. Phys. 364 (2018), 357 -- 383.
[8] J. Christensen, KMS states on the Toeplitz algebras of higher-rank graphs, preprint, arXiv:-
1805.09010.
[9] J. Cuntz, C. Deninger and M. Laca, C ∗-algebras of Toeplitz type associated with algebraic
number fields, Math. Ann. 355 (2013), 1383 -- 1423.
[10] K.R. Davidson, C ∗-Algebras by Example, Fields Inst. Monographs, vol. 6, Amer. Math.
Soc., Providence, 1996.
[11] R. Exel and M. Laca, Partial dynamical systems and the KMS condition, Comm. Math.
Phys. 232 (2003), 223 -- 277.
[12] J.M.G. Fell and R.S. Doran, Representations of ∗-Algebras, Locally Compact Groups, and
Banach ∗-Algebraic Bundles, Vol. 1, Academic Press, San Diego, 1988.
[13] J. Fletcher, A. an Huef and I. Raeburn, A program for finding all KMS states on the Toeplitz
algebra of a higher-rank graph, J. Operator Theory, to appear; arXiv:1801.03189.
[14] G.B. Folland, Real Analysis: Modern Techniques and their Applications, second ed., Wiley,
New York, 1999.
[15] A. an Huef, M. Laca, I. Raeburn and A. Sims, KMS states on the C ∗-algebras of finite
graphs, J. Math. Anal. Appl. 405 (2013), 388 -- 399.
[16] A. an Huef, M. Laca, I. Raeburn and A. Sims, KMS states on C ∗-algebras associated to
higher-rank graphs, J. Funct. Anal. 266 (2014), 265 -- 283.
[17] E.T.A. Kakariadis, On Nica-Pimsner algebras of C ∗-dynamical systems over Zn
+, Int. Math.
Res. Not. IMRN, 2017, 1013 -- 1065.
[18] E.T.A. Kakariadis, Equilibrium states and entropy theory for Nica-Pimsner algebras,
preprint, arXiv:1806.02443.
[19] M. Laca and S. Neshveyev, Type III1 equilibrium states of the Toeplitz algebra of the affine
semigroup over the natural numbers, J. Funct. Anal. 261 (2011), 169 -- 187.
32
ZAHRA AFSAR, ASTRID AN HUEF, IAIN RAEBURN, AND AIDAN SIMS
[20] M. Laca and I. Raeburn, Semigroup crossed products and the Toeplitz algebras of non-
abelian groups, J. Funct. Anal. 139 (1996), 415 -- 440.
[21] M. Laca and I. Raeburn, Phase transition on the Toeplitz algebra of the affine semigroup
over the natural numbers, Adv. Math. 225 (2010), 643 -- 688.
[22] M. Laca, I. Raeburn, J. Ramagge and M.F. Whittaker, Equilibrium states on the Cuntz-
Pimsner algebras of self-similar actions, J. Funct. Anal. 266 (2014), 6619 -- 6661.
[23] M. Laca, I. Raeburn, J. Ramagge and M.F. Whittaker, Equilibrium states on operator
algebras associated to self-similar actions of groupoids on graphs, Adv. Math. 331 (2018),
268 -- 325.
[24] F. Latr´emoli`ere and J.A. Packer, Noncommutative solenoids, New York J. Math. 24a (2018),
155 -- 191. arXiv:1110.6227.
[25] A. Nica, C ∗-algebras generated by isometries and Wiener-Hopf operators, J. Operator The-
ory 27 (1992), 17 -- 52.
[26] I. Raeburn, On crossed products by coactions and their representation theory, Proc. London
Math. Soc. 64 (1992), 625 -- 652.
Zahra Afsar, School of Mathematics and Statistics, University of Sydney,
NSW 2006, Australia
Astrid an Huef and Iain Raeburn, School of Mathematics and Statistics, Vic-
toria University of Wellington, PO Box 600, Wellington 6140, New Zealand
Aidan Sims, School of Mathematics and Applied Statistics, University of Wol-
longong, NSW 2522, Australia
|
1803.08733 | 1 | 1803 | 2018-03-23T10:57:19 | Dimension groups for self-similar maps and matrix representations of the core of the associated C*-algebras | [
"math.OA"
] | We introduce a dimension group for a self-similar map as the ${\rm K}_0$-group of the core of the $C^*$-algebra associated with the self-similar map together with the canonical endomorphism. The key step for the computation is an explicit description of the core as the inductive limit using their matrix representations over the coefficient algebra, which can be described explicitly by the singularity structure of branched points. We compute that the dimension group for the tent map is isomorphic to the countably generated free abelian group ${\mathbb Z}^{\infty}\cong {\mathbb Z}[t]$ together with the unilatral shift, i.e. the multiplication map by $t$ as an abstract group. Thus the canonical endomorphisms on the ${\rm K}_0$-groups are not automorphisms in geneal. This is a different point compared with dimension groups for topological Markov shifts. We can count the singularity structure in the dimension groups. | math.OA | math |
DIMENSION GROUPS FOR SELF-SIMILAR MAPS AND MATRIX
REPRESENTATIONS OF THE CORES OF THE ASSOCIATED
C∗-ALGEBRAS
TSUYOSHI KAJIWARA AND YASUO WATATANI
Abstract. We introduce a dimension group for a self-similar map as the K0-group of the
core of the C∗-algebra associated with the self-similar map together with the canonical
endomorphism. The key step for the computation is an explicit description of the core as
the inductive limit using their matrix representations over the coefficient algebra, which can
be described explicitly by the singularity structure of branched points. We compute that
the dimension group for the tent map is isomorphic to the countably generated free abelian
group Z∞ ∼= Z[t] together with the unilatral shift, i.e. the multiplication map by t as an
abstract group. Thus the canonical endomorphisms on the K0-groups are not automorphisms
in geneal. This is a different point compared with dimension groups for topological Markov
shifts. We can count the singularity structure in the dimension groups.
1. Introduction
Dimension groups for topological Markov shifts were introduced and studied by Krieger
in [11] and [12] motivated by the K-groups for the AF-subalgebras of the associated Cuntz-
Krieger algebras [2]. The dimension group is an ordered abelian group with a canonical
automorphism. The dimension group for topological Markov shifts is very important, because
it is a complete invariant up to shift equivalence. Matsumoto introduced a class of C∗-algebras
associated with general subshifts in [13] and studied dimension groups for subshifts in [14].
In our paper we shall introduce dimension groups for self-similar maps as the K0-groups of
the cores of C∗-algebras associated with self-similar maps, where the core is the fixed point
subalgebra under the gauge action of the C∗-algebra associated with a self-similar map. But
canonical automorphisms on dimension goups can be generalized as only endomorphisms in
the case of self-similar maps. The key step for the computation is an explicit description
of the cores as the inductive limit using matrix representations over the coefficient algebra,
which heavily depends on the structure of branched points.
Self-similar maps on compact metric spaces produce many interesting self-similar sets like
Sierpinski Gasket, see for example [4], [10]. We constructed C∗-algebras for self-similar maps
using C∗-correspondence and Pimsner construction ([6]). C∗-algebras associated with self-
similar maps are simple, purely infinite and in UCT class, and are classifiable by their K-
groups ([6]). The algebra of the continuous functions on the self-similar set is shown to be
maximal abelian in [9].
In [7], [8], we classified traces and ideals of the core of the C∗-algebras associated with
self-similar maps under some conditions, i.e. Assumption A and Assumption B, concerning
If there exists no branched points,then the core of the C∗-algebra
branched point sets.
associated with a self-similar map is simple and has a unique tracial state. On the other hand,
if there exists a branched point, then the core has many finite discrete traces. Moreover, in
1
[8], we constructed the matrix representation of the core over the coefficient algebra, which
makes the classification of ideals of the core possible.
In the paper, we present an explicit form of matrix representations over the coefficient
algebra of the finite cores of the C∗-algebra associated with self-similar maps under the
Assumption B. Here the Assumption B roughly means that the set of branched points is a
finite set and every orbit has no branched point or one branched point at most. Many typical
examples like a tent map satisfies the Assumption B. We express the K groups of the cores of
C∗-algebras associated with tent map and Sierpinski Gasket respectively as inductive limit.
We compute that the K0-group of the core of the C∗-algebras associated with the tent map is
isomorphic to the countably generated free abelian group Z∞ as an abstract group. Moreover,
we can prove that the canonical endomorphism of the dimension group for tent map case is
the unilatral shift on Z∞, which is not surjective. On the otherhand, if a self-similar map
has no branched points, then the canonical endomorphism of the dimension group becomes
an automorphism.
We refer other interesting approach on dimension groups for interval maps by F. Shultz
[19],[20] and Deaconu-Shultz [3].
The contents of the paper is as follows: Section 1 is an introduction. In Section 2, we recall
preliminary results on self-similar maps and C∗-correspondences. In Section 3, we summarize
the construction of the matrix representation over the coefficient algebra of the core following
[8]. In Section 4, we present explicit description of the matrix representations of the finite
cores for the case of tent map and Sierpinski Gasket.
In Section 5, we present explicit
calculation of K-groups of the cores for the case of tent map and Sierpinski Gasket using
explicit description of the finite core. In Section 6, we define the cnanonical endomorphism
on the dimension group in general.
If the the Hilbert module is finitely generated, then
the canonical endomorphism on the dimension group is induced by the dual action of the
gauge action. In the case of the tent map, we show that the canonical endomorphism is not
surjective by the explicit calculation.
This work was supported by JSPS KAKENHI Grant Number JP16K05178, JP23540242,
and JP25287019.
2. Self-similar maps and their C∗-correspondences
Let (Ω, d) be a (separable) complete metric space. A map f : Ω → Ω is said to be
a proper contraction if there exists a constant c and c′ with 0 < c′ ≤ c < 1 such that
0 < c′d(x, y) ≤ d(f (x), f (y)) ≤ cd(x, y) for any x, y ∈ Ω.
We consider a family γ = (γ0, . . . , γN−1) of N proper contractions on Ω. We always assume
that N ≥ 2. Then there exists a unique non-empty compact set K ⊂ Ω which is self-similar
in the sense that K =SN−1
i=0 γi(K). See Falconer [4] and Kigami [10] for more on fractal sets.
In this paper, we usually forget an ambient space Ω as in [6] and start with the following:
Let (K, d) be a compact metric set and γ = (γ0, . . . , γN−1) be a family of N proper contrac-
tions on K. We say that γ is a self-similar map on K if K = SN−1
i=0 γi(K). Throughout the
V of K such that γj(V ) ∩ γk(V ) = φ for j 6= k and SN−1
Definition 1. We say that γ satisfies the open set condition if there exists an open subset
i=0 γi(V ) ⊂ V . See a book [4] by
paper we assume that γ is a self-similar map on K.
Falconer. .
2
Let Σ = { 0, . . . , N − 1}. For k ≥ 1, we put Σk = { 0, . . . , N − 1}k. For a self-similar map
γ on a compact metric space K, we introduce the following subsets of K:
Bγ ={ b ∈ K b = γi(a) = γj(a), for some a ∈ K and i 6= j },
Cγ ={ a ∈ K γi(a) = γj(a), for some a ∈ K and i 6= j } = { a ∈ K γj(a) ∈ Bγ for some j },
Pγ ={ a ∈ K ∃k ≥ 1, ∃(j1, . . . , jk) ∈ Σk such that γj1 ◦ ··· ◦ γjk (a) ∈ Bγ}.
Any element b in Bγ is called a branched point and any element c in Cγ is called a branched
value. We call Bγ the branch set of γ, Cγ the branch value set of γ, and Pγ the postcritical
set of γ.
Throughout the paper, we assume that a self-similar map γ on K satisfies the following
Assumption B:
each j.
(1) There exists a continuous map h of K to K which satisfies h(γj (y)) = y (y ∈ K) for
(2) The set Bγ is a finite set.
(3) Bγ ∩ Pγ = ∅.
The conditon (3) means that every orbit has no branched point or one branched point at
most.
If (2) is replaced by a stronger condition
(2)' The sets Bγ and Pγ are finite sets,
then it is exactly the assumption A studied in [7].
Therefore Bγ is the set of b ∈ K such that h is not locally homeomorphism at b, that is,
Bγ is the set of the branched points of h in the usual sense. Moreover Cγ = h(Cγ) is the
branch value set of h and ∩Pγ is the postcritical set of h in the usual sense.
Many important examples including tent map and Sierpinski gasket satisfy the assumption
B above. If we assume that γ satisfies the assumption B, then K does not have any isolated
points and K is not countable.
Notation. For fixed b ∈ Bγ, we denote by eb the number of j such that b = γj(h(b)). We
call eb the branch index at b. Then eb the branch index at b for h in the usual sense. We
label these indices j so that
{j ∈ Σ b = γj(h(b))} = {j(b, 0), j(b, 1), . . . , j(b, eb − 1)}
satisfying j(b, 0) < j(b, 1) < ··· < j(b, eb − 1).
Example 2.1. [6] (tent map) Define a family γ = (γ0, γ1) of proper contractions on K = [0, 1]
by γ0(y) = (1/2)y and γ1(y) = 1 − (1/2)y. Then γ is a self-similar map. We see that
Bγ = { 1/2}, Cγ = { 1}. We define the ordinary tent map h on [0, 1] by
h(x) =(2x
−2x + 2
0 ≤ x ≤ 1/2
1/2 ≤ x ≤ 1
Then γ satisfies Assumption B. γ = (γ0, γ1) is the inverse branches of the tent map h. We
see that Bγ and Cγ are finite sets. We note that h(1/2) = 1, h(1) = 0, h(0) = 0. Hence
Pγ = { 0, 1} and Bγ ∩ Pγ = ∅.
Example 2.2. [6] (Sierpinski gasket) Consider a regular triangle T on R2 with three vertices
P = (1/2,√ 3/2), Q = (0, 0) and R = (1, 0). Let S = (1/4,√ 3/4), T = (1/2, 0) and
3
U = (3/4,√ 3/4). Define proper contractions γ0, γ1 and γ2 on the triangle T by
2(cid:19) .
γ2(x, y) =(cid:18) x
γ1(x, y) =(cid:16) x
γ0(x, y) =(cid:18) x
2
y(cid:19) ,
y
2(cid:17) ,
1
2
,
+
1
4
,
1
2
,
2
+
y
2
Let αθ be a rotation by angle θ. We define γ0 = γ0, γ1 = α−2π/3◦γ1, γ2 = α2π/3◦γ2. We denote
by ∆ABC the regular triangle whose vertices are A, B and C. Then γ0(∆PQR) = ∆PSU,
γ1(∆PQR) = ∆TSQ and γ2(∆PQR) = ∆TRU. Let
K =
∞
\n=1 \(j1,...,jn)∈Σn
(γj1 ◦ ··· ◦ γjn)(T ).
The compact set K is called a Sierpinski gasket. Then γ = (γ0, γ1, γ2) is a self-similar map
on K. Moreover Bγ = { S, T, U}, Cγ = { P, Q, R}. Define a continuous map h by
h(x, y) =
γ−1
0 (x, y)
γ−1
1 (x, y)
γ−1
2 (x, y)
(x, y) ∈ ∆PSU ∩ K,
(x, y) ∈ ∆TSQ ∩ K,
(x, y) ∈ ∆TRU ∩ K.
Then we have that h(S) = Q, h(U) = R, h(T) = P, h(P) = P, h(Q) = R and h(R) = Q.
Hence Pγ = { P, Q, R} and the self-similar map γ on K satisies the assumption B.
We recall a construction of a C∗-correspondence (or Hilbert C∗-bimodule) for the self-
similar map γ = (γ0, . . . , γN−1) in [6]. Let A = C(K) be a coefficient algebra. Consider a
cograph Cγ = { (γj (y), y) j ∈ Σ, y ∈ K }. Let Xγ = C(Cγ). Define left and right A-module
actions and an A-valued inner product on Xγ by, for f , g ∈ Xγ and a, b ∈ A,
(a · f · b)(γj(y), y) = a(γj(y))f (γj(y), y)b(y)
y ∈ K,
j = 0, . . . , N − 1,
(fg)A(y) =
f (γj(y), y)g(γj (y), y).
N−1
Xj=0
Recall that the "compact operators" K(Xγ) is the C∗-algebra generated by the rank one
operators {θx,y x, y ∈ Xγ}, where θx,y(z) = x(yz)A for z ∈ Xγ. When we do stress the role
of the module X, we write θx,y = θX
x,y. We denote by K(Xγ) the set of "compact" operators
on Xγ, and by L(Xγ) the set of adjointable operators on Xγ. We define an ∗-homomorphism
φ of A to L(Xγ) by φ(a)f = a · f .
Lemma 2. [6] Let γ = (γ0, . . . , γN−1) be a self-similar map on a compact set K. Then Xγ
is an A-A correspondence and full as a right Hilbert module.
We write by Oγ = Oγ(K) the Cuntz-Pimsner C∗-algebra ([16]) associated with the C∗-
correspondence Xγ. We call it the Cuntz-Pimsner algebra Oγ associated with a self-similar
map γ. We usually identify i(a) with a in A. We also identify Sξ with ξ ∈ X and simply
write ξ instead of Sξ. The gauge action α : R → Aut Oγ is defined by αt(Sξ) = eitSξ for
ξ ∈ Xγ and αt(a) = a for a ∈ A.
Theorem 3. [6] Let γ be a self-similar map on a compact metric space K. If (K, γ) satisfies
the open set condition, then the associated Cuntz-Pimsner algebra Oγ is simple and purely
infinite.
4
For example, if γ is a self-similar map associated with the tent map or the Sierpinski gasket,
then Oγ is simple and purely infinite.
of A on X⊗n
For the n-times inner tensor product X⊗n
γ
. Define
F (n) = A ⊗ I + K(Xγ ) ⊗ I + K(X⊗2
γ
of Xγ, we denote by φn the left module action
γ ) ⊗ I + ··· + K(X⊗n
γ
) ⊂ L(X⊗n
γ
)
Then F (n) is embedded to F (n+1) by T 7→ T ⊗ I for T ∈ F (n). Let F (∞) =S∞n=0 F (n). It is
important to recall that Pimsner [16] shows that we can identify F (n) with the C∗-subalgebra
of Oγ generated by A and SxS∗y for x, y ∈ X⊗k, k = 1, . . . , n under identifying SxS∗y with
θx,y. Therefore we see that the inductive limit algebra F (∞) can be identified with the fixed
point subalgebra OT
γ of Oγ under the gauge action. We call it the core of Oγ. We also call
F (n) the finite core or the finite n-th core of Oγ.
3. Matrix representation over the coefficient algebra of cores
For a Hilbert right A-module X, we denote by K0(X) the set of finite sums of "rank one
" operators θx,y for x, y ∈ X. If X is finitely generated projectile module, then K0(X) =
K(X) = L(X).
If a self-similar map γ = (γ0, . . . , γN−1) has a branched point, then the
Hilbert right A-module Xγ is not finitely generated projectile module and K(Xγ) 6= L(Xγ).
But if the self-similar map γ satisfies Assumption B, it is shown in [8] that Xγ is near to
a finitely generated projectile module in the following sense: K0(Xγ) = K(Xγ ). Moreover
K(Xγ ) is realized as a subalgebra of the full matrix algebra MN (A) = C(K, MN (C) over
A = C(K) consisting of matrix valued functions f on K such that their scalar matrices f (c)
live in certain subalgebras Qγ(c)MN (C)Qγ(c) cutted down by projections Qγ(c) for each c
in the finite set Cγ and live in the full matrix algebra MN (C) for other c /∈ Cγ. Therefore
K(Xγ ) has certain informations on the sigularity structure of branched points.
We present only the outline of the matrix representation over the coefficient algebra of the
finite core of Oγ under Assumption B. We refer the detail to the section two of [8].
Let Yγ := AN be a free module over A = C(K). Then L(Yγ) is isomorphic to MN (A).
Recall that left and right A-module actions and an A-inner product on Yγ := AN are given
as follows:
(a · f · b)i(y) = a(γi(y))fi(y)b(y),
(fg)A(y) =
fi(y)gi(y),
N−1
Xi=0
where f = (f0, . . . , fN−1), g = (g0, . . . , gN−1) ∈ Yγ and a, b ∈ A. Thus Yγ is an A-A
correspondence and Yγ is a finitely generated projective right module over A.
Let A be a subalgebra of L(Yγ) = MN (A) = C(K, MN (C)). For z ∈ K, the fiber A(z) of
A at z is defined by
A(z) = {T (z) ∈ MN (C) T ∈ A ⊂ C(K, MN (C))}.
We define a subset Zγ of Yγ by
Zγ := { f = (f0, . . . , fN−1) ∈ AN
for any c ∈ Cγ, b ∈ Bγ with h(b) = c, fj(b,k)(c) = fj(b,k′)(c)
5
0 ≤ k, k′ ≤ eb − 1}.
Thus i-th component fi(c) of the vector (f1(c), . . . , fN (c)) ∈ CN is equal to the i′-th
component fi′(c) of it for any i, i′ in the same index subset
{j ∈ Σ b = γj(c)} = {j(b, 0), j(b, 1), . . . , j(b, eb − 1)}
In order to get the idea of rather complicated but important notations below, we consider
for each b ∈ Bγ.
a simplified example: Assume that N = 5, c ∈ Cγ and h−1(c) = {b1, b2} ⊂ Bγ,
b1 = γ0(c) = γ1(c), b2 = γ2(c) = γ3(c) = γ4(c).
that is,
b1
γ0,γ1⇐= c
γ2,γ3,γ4=⇒ b2.
Consider the following degenerated subalgebra A of a full matrix algebra M5(C):
A = {a = (aij) ∈ M5(C) a0j = a1j, ai0 = ai1, a2j = a3j = a4j, ai2 = ai3 = ai4}.
Then
b
b
c d d d
c d d d
c d d d
is isomorphic to M2(C). Consider the subspace
A = {
a a b
a a b
c
c
c
b
b
a, b, c, d ∈ C}
W = {(x, x, y, y, y) ∈ C5 x ∈ C, y ∈ C}
of C5. Let u1 = 1√2
basis of W . Let Qγ(c) be the projection onto the the subspace W . Then
t(1, 1, 0, 0, 0) ∈ W and u2 = 1√3
t(0, 0, 1, 1, 1) ∈ W . Then {u1, u2} is a
Moreover {θW
ui,uj}i,j=1,2 is a matrix unit of A and
A = Qγ(c)M5(C)Qγ(c) ∼= M2(C).
2
A = {
Xi,j=1
aijθW
ui,uj aij ∈ C} = L(W ).
In general, wc := dim(Zγ(c)) is equal to the cardinality of h−1(c) without counting mul-
tiplicity. We can take the following basis { uc
i }i=1,...,wc of the fiber W := Zγ(c) ⊂ CN :
1√ebi
Rename h−1(c) = {b1, . . . , bwc}. Then the j-th component of the vector uc
if j ∈ {j ∈ Σ bi = γj(h(bi))} = {j(bi, 0), j(bi, 1), . . . , j(bi, ebi − 1)} and is equal to 0 if j is
otherwise.
i is equal to
Let Qγ(c) be the projection onto the the subspace W . Then
Then Zγ is also an A-A correspondence with the A-bimodule structure and the A-valued
L(W ) = Qγ(c)MN (C)Qγ(c) ∼= Mwc(C).
inner product inherited from Yγ. For ξ ∈ Xγ, we define ϕ(ξ) ∈ Zγ by
y ∈ K.
ϕ(ξ)(y) = (ξ(γ0(y), y), . . . , ξ(γN−1(y), y))
By Lemma 5 in [8], the C∗-correspondences Xγ and Zγ are isomorphic under ϕ. Identifying
through the map ϕ, we may regard K(Xγ ) as a closed subset of L(Yγ) = L(AN ) = C(MN (C)).
6
For a Hilbert A-module W , we denote by K0(W ) the set of "finite rank operators" (i,e,
finite sum of rank one operators) on W , that is,
n
Xi=1
K0(W ) = {
θW
xi,yi n ∈ N, xi, yi ∈ W}.
As in [8], we define the following subalgebra Dγ of MN (C(K)) = C(K, MN (C)):
Dγ ={ a = [aij]i,j ∈ MN (A) = C(K, MN (C)) for c ∈ Cγ, b ∈ Bγ with h(b) = c,
aj(b,k),i(c) = aj(b,k′),i(c)
ai,j(b,k)(c) = ai,j(b,k′)(c)
0 ≤ k, k′ ≤ eb − 1, 0 ≤ i ≤ N − 1
0 ≤ k, k′ ≤ eb − 1, 0 ≤ i ≤ N − 1},
The algebra Dγ is a closed *subalgebra of MN (A) = K(Yγ) and its fiber Dγ(c) at c is
described as
Dγ(c) = Qγ(c)MN (C)Qγ(c) ∼= Mwc(C),
where Qγ(c) is the projection of CN onto the subspace W = Zγ(c) and wc = dim Zγ(c).
For f = (f0, . . . , fN−1) ∈ Zγ and g = (g0, . . . , gN−1) ∈ Zγ, the rank one operator θYγ
f,g ∈
L(Yγ) is in Dγ and represented by the operator matrix
θYγ
f,g = [figj]ij ∈ MN (A),
by Lemma 7 in [8]. Let K(Zγ ⊂ Yγ) ⊂ L(Yγ) be the norm closure of
n
Xi=1
K0(Zγ ⊂ Yγ) := {
θYγ
xi,yi ∈ L(Yγ) n ∈ N, xi, yi ∈ Zγ}.
The following is Lemma 9 in [8], and essential for matrix representation of finite cores.
Lemma 4. Let γ be a self-similar map on a compact metric space K and satisfies Assumption
B. Then K0(Xγ) = K(Xγ ), K0(Zγ) = K(Zγ ) and K0(Zγ ⊂ Yγ) = K(Zγ ⊂ Yγ) = Dγ ⊂
MN (A).
For any T ∈ K(Zγ ⊂ Yγ), we have T (Zγ) ⊂ Zγ and by Lemma 8 in [8], the restriction map
is an onto *isomorphism such that
δ : K(Zγ ⊂ Yγ) ∋ T → TZγ ∈ K(Zγ)
Then we have an isometric *-homomorphism δ−1 from K(Zγ) to K(Zγ ⊂ Yγ) satisfying
θYγ
xi,yi) =
θZγ
xi,yi.
δ(
n
Xi=1
n
Xi=1
n
Xi=1
n
Xi=1
δ−1(
θZγ
xi,yi) =
θYγ
xi,yi.
We can define an injective *-homomorphism Ω(1) of K(Xγ) to K(Yγ) such that
Ω(1)(θXγ
f,g ) = θYγ
ϕ(f ),ϕ(g).
Then by Lemma 4, the range of Ω(1) is Dγ.
7
Fix a natural number n. We shall describe the matrix representation of K(X⊗n
) for
n-times tensor product X⊗n
. We consider the composition of self-similar maps γ. Put
γn = {γin ◦ ··· ◦ γi1}(i1,...,in)∈Σn. Then γn is a self-similar map on the same compact metric
space K. For the description of the branched points for γn, the following Lemmas are given
in [8].
γ
γ
Lemma 5. [8] Let γ be a self-similar map on a compact metric space K and satisfies As-
sumption B. Then Cγn and Bγn are finite sets and Cγn ⊂ Cγn+1 for each n = 1, 2, 3, . . . . The
set of branched points Bγn is given by
Bγn = { (γjk ◦ ··· ◦ γj1)(b) b ∈ Bγ, (j1, . . . , jk) ∈ Σk, 0 ≤ k ≤ n − 1}.
Moreover, if γin ◦···◦ γi1(c) = γjn ◦···◦ γj1(c) and (i1, . . . , in) 6= (j1, . . . , jn), then there exists
unique 1 ≤ s ≤ n such that is 6= js and ip = jp for p 6= s.
And the following lemma holds for Cγn:
Lemma 6. [8] Let a ∈ K. Then a ∈ Cγn if and only if there exist 0 ≤ p ≤ n − 1 and
(i1, i2, . . . , ip) ∈ Σp such that γip◦···◦γi1
(a) ∈ Cγ.
We denote by Xγn the A-A correspondence for γn. By Proposition 2.2 in [6], X⊗n
Xγn are isomorphic as A-A correspondence. Using it, we can get similar results for X⊗n
Z⊗n
γ . For example, X⊗n
For γn, we define a subset Dγn of MN n(A) as in the case of γ. We also consider Cγn instead
of Cγ. We use the same notation eb for b ∈ Bγn with hn(b) = c and { j(b, k) 0 ≤ k ≤ eb − 1}
for γn as in γ if there occurs no trouble. Let
is isomorphic to a closed submodule Zγn of AN n
and
and
γ
γ
γ
.
Dγn
= { [aij ]ij ∈ MN n(A) for c ∈ Cγn, b ∈ Bγn with hn(b) = c,
aj(b,k),i(c) = aj(b,k′),i(c), ai,j(b,k)(c) = ai,j(b,k′)(c)
for all 0 ≤ k, k′ ≤ eb − 1, 0 ≤ i ≤ N n − 1 }.
Then K(X⊗n
γ
subalgebra Dγn
) and K(Z⊗n
of Yγn. As Lemma 4, we have the following Proposition:
) are isomorphic, and K(Z⊗n
γ
γ
) is isometrically extended to the
Propositoin 7. [8] Let γ be a self-similar map on a compact metric space K and satis-
fies Assumption B. Then K0(X⊗n
) and is isomorphic to the closed
*subalgebra Dγn of MN n(A).
) coincides with K(X⊗n
γ
γ
Let 0 ≤ r ≤ n. As in the case r = 1, there exists a matrix representation Ω(r) of K(X⊗r
γ )
to MN r (A) satisfying
Ω(r)(θ
X ⊗r
x,y ) = θ
γ
Y ⊗r
γ
ϕ(x),ϕ(y)
γ , where ϕ is an isomorphism from X⊗r
γ
for x, y ∈ X⊗r
0 ≤ r ≤ n. Then there exist a representation Ω(n,r) from K(X⊗r
that
We shall give a matrix representation of the finite core F (n) in MN n(A). Let n ≥ 0. Fix
) = MN n(A) such
γ ) to L(Y ⊗n
to Z⊗r
γ .
γ
for T ∈ K(X⊗r
γ ).
Ω(n,r)(T ) = Ω(r)(T ) ⊗ In−r
8
We introduce some useful notations. We fix N . For a sequence of indices (i1, . . . , is) ∈ Σs,
we define a non negative integer i1 . . . is(N,s) by N -adic expansion
s
We shall use this N -adic expansion as suffixes to refer to the entries in a big matrix.
i1 . . . is(N,s) =
ijN s−j.
Xj=1
Let n be an integer, p and q be non negative integers with n = p + q. For T ∈ MN q (C)
and 0 ≤ i ≤ N p − 1, define πq,i(T ) ∈ MN n(C) by
(πq,i(T ))a,b =( Ta−iN q,b−iN q
0 (otherwise)
(iN q ≤ a, b ≤ (i + 1)N q − 1)
.
Thus (πq,i(T ))a,b is a block diagonal matrix such that the only i-th block is T and the other
blocks are 0.
For mutually different integers 0 ≤ ik ≤ N p − 1 (k = 1, . . . , r), define a block diagonal
matrix such that the i1, . . . , ir-th blocks are the same T and the other blocks are 0 by
r
Let 0 ≤ i, j ≤ N p − 1, i 6= j. We also define σq,i,j(T ) ∈ MN n(C) by
πq,i1,...,ir (T ) =
πq,ik(T ).
Xk=1
(σq,i,j(T ))a,b =
Ta−iN q,b−iN q
Ta−iN q,b−jN q
Ta−jN q,b−iN q
Ta−jN q,b−jN q
0 otherwise.
iN q ≤ a, b ≤ (i + 1)N q − 1
iN q ≤ a ≤ (i + 1)N q − 1, jN q ≤ b ≤ (j + 1)N q − 1
jN q ≤ a ≤ (j + 1)N q − 1, iN q ≤ b ≤ (i + 1)N q − 1
jN q ≤ a, b ≤ (j + 1)N q − 1
same T and the other blocks are 0.
Thus (σq,i,j(T ))a,b is a a block matrix such that the i− i, i− j, j − i, j − j-th blocks are the
By Lemma 15 in [8], the natural embedding
L(Y ⊗r
γ
) ∋ T 7→ T ⊗ In−r ∈ L(Y ⊗n
γ
) = C(K, MN n (C))
has a matrix representation as a block diagonal matrix for each fiber at P ∈ K:
(T ⊗ In−r)(P) = X(i1,...,in−r)∈Σn−r
πr,i1...in−r(N,n−r)(T ((γin−r ◦ ··· ◦ γi1)(P))),
where T ((γin−r ◦ ··· ◦ γi1)(P)) ∈ MN r (C) is a fiber of T at (γin−r ◦ ··· ◦ γi1)(P) . Therefore
(T ⊗ In−r)(P) is a block diagonal matrix by these fibers.
Therefore we can describe the form of
Ω(n,r) : K(X⊗r
γ ) → L(Y ⊗n
γ
) = MN n(A).
For T ∈ K(X⊗r
γ ) (0 ≤ r ≤ n − 1), Ω(n,r)(T ) is expressed as:
Ω(n,r)(T )(P) = X(i1,...,in−r)∈Σn−r
πr,i1...in−r(N,n−r)(Ω(r)(T )((γin−r ◦ ··· ◦ γi1)(P))
for each P ∈ K.
9
As in [8], we have the following Theorem:
Theorem 8. [8] (matrix representation of the finite core) Let γ be a self-similar map
on a compact metric space K and satisfies Assumption B. Then there exists an isometric
r=0 Tr ⊗ In−r ∈ F (n) with
∗-homomorphism Π(n) : F (n) → MN n(A) such that, for T = Pn
Tr ∈ K(X⊗r
γ ),
n
Π(n)(T ) =
and if we identify X⊗r
γ with Z⊗r
γ , then
Ω(n,r)(Tr),
Xr=0
The image Π(n)(T ) is independent of the expression of T =Pn
Moreover the following diagram commutes:
r=0 Tr ⊗ In−r ∈ F (n).
Ω(n,r)(θ
Zγr
x,y ) = θ
Yγr
x,y ⊗ In−r.
F (n)
Π(n)
−−−−→ MN n(A)
y
F (n+1) Π(n+1)
−−−−→ MN n+1(A).
y
In particular the core F (∞) is represented in the MN ∞ ⊗ A as a C∗-subalgebra.
We note that for T ∈ K(X⊗r) it holds that
Π(n)(T )(P) = X(i1,...,in−r)∈Σn−r
πr,i1...in−r(N,n−r)(Π(r)(T ))((γin−r ◦ ··· ◦ γi1)(P)).
Hence Π(n)(T )(P) is a block diagonal matrix consisting of Π(r)(T ))((γin−r ◦ ··· ◦ γi1)(P).
4. Explicit form of finite cores in terms of singularity
4.1. General calculation. Let γ = (γ0, . . . , γN−1) be a self-similar map on K satisfying the
assumption B. We shall describe the structure of Π(n)(F (n))(P) for P ∈ Pγ. Let 0 ≤ q ≤ n,
and p = n − q. The elements of Π(n)(K(X⊗q))(P) are represented as block diagonal matrices
with blocks which consists of matrices in MN q (C).
Lemma 9. Let P ∈ K. Assume that γik ◦ ··· ◦ γi1(P) /∈ Bγ (for any k = 1, . . . , p − 1). If
γip−1 ◦ γip−2 ◦ ··· ◦ γi1(P) = γjp−1 ◦ γjp−2 ◦ ··· ◦ γj1(P),
then it holds (i1, i2, . . . , ip−1) = (j1, i2, . . . , jp−1).
Proof. It is clear fronm the definition of Bγ.
Definition 10. For a point R ∈ K and non-negative integers p and q, the p − q orbit
Orbitp−q(R) through R is defined by
γi2→ R2 . . .
γip+q→ Rp+q (R0, . . . , Rp+q) ∈ K p+q+1,
Orbitp−q(R) := {R0
γi1→ R1
(cid:3)
(i1, i2, . . . , ip+q) ∈ Σp+q, γip(R0) = R}
10
The following proposition shows that p − q orbit Orbitp−q(R) through R representes the
branched points structure. And this will make the explicit computation of the finite core
possible.
Propositoin 11. Suppose the Assmption B . For non-negative integers p and q, if R is
a branched point, then any element in the p − q orbit Orbitp−q(R) through R satisfies the
following:
The points R0, R1, . . . , Rp−1 and a finite word (i1, . . . , ip−1) are uniquely determined, and ip
is j(R, 0), j(R, 1), . . . or j(R, eR − 1). Moreover the number of Orbitp−q(R) is exactly eRN q.
Proof. By the Assmption B, Rk = hk(R) for k = 0, 1, . . . , p − 1 and (i1, . . . , ip−1) is uniquely
determined by Lemma 9. Then the rest is clear.
(cid:3)
We write the above situation by the following picture:
P = R0
γi1→ R1
γi2→ R2 . . .
γip−1→ Rp−1
γij(R,0)
,...,γij(R,eR−1)
,γij(R,1)
=⇒
Rp = R ր...
ց
ր...
ց
Let P be a point of K and R be a branched point. Fix a natural number n. Then the p− q
orbit Orbitp−q(R) through R suggests us an important subalgebra C(P, R, p) of Π(n)(F (n))(P),
which represents a building block of the finite core.
Definition 12. Let P be a point of K and R be a branched point. Fix a natural number n.
Let p be a non-negative integer with hp(R) = P and q = n− p. Define a subalgebra C(P, R, p)
of Π(n)(F (n))(P) by
C(P, R, p)
={ πq,i1...ip−1j(R,0)(N,p), i1...ip−1j(R,1)(N,p), ..., i1...ip−1j(R,eR−1)(N,p)(A) A ∈ MN q (C)}
={ X0≤l≤eR−1
πq,i1...ip−1j(R,l)(N,p)(A) A ∈ MN q (C)}.
Any operator T = P0≤l≤eR−1 πq,i1...ip−1j(R,l)(N,p)(A) in the subalgebra C(P, R, p) is repre-
sented as a block diagonal matrix consisting of A ∈ MN q (C) such that A appears with multi-
plicity eR in the diagonal positions of i1 . . . ip−1ip(N,p) with ip = j(R, 0), j(R, 1), . . . , j(R, eR − 1).
Therefore the subalgebra C(P, R, p) can be read off from the singularity structure of branched
points.
Moreover C(P, R, p) is isomorphic to MN q (C) as a C∗-algebra with q = n − p.
If we fix n, then the algebra C(P, R, p) is uniquely determined by P ∈ K, R ∈ Bγ and p.
The following Proposition characterizes matrices in Π(m)(K(X⊗m))(P).
Propositoin 13. Let m ≥ 1 be an integer. If P /∈ Cγm, then Π(m)(K(X⊗m))(P) = MN m(C).
If P ∈ Cγm, then a matrix T ∈ MN m(C) is contained in Π(m)(K(X⊗m))(P) if and only if T
satisfies the following:
For any 1 ≤ p ≤ m and R ∈ Bγ with hp(R) = P, take (i1, . . . , ip−1) ∈ Σp−1, which is uniquely
determined as above.
Ti1...ip−1j(R,t)ip+1...im(N,m),j1j2...jm(N,m)
Ti1i2...im(N,m),i1...ip−1j(R,s)jp+1...jm(N,m)
= Ti1...ip−1j(R,t′)ip+1...im(N,m),j1j2...jm(N,m)
,
= Ti1i2...im(N,m),i1...ip−1j(R,s′)jp+1...jm(N,m)
11
for arbitrary 0 ≤ t, t′, s, s′ ≤ eR − 1 and arbitrary (ip+1, . . . ,im), (jp+1, . . . , jm) ∈ Σm−p,
(i1, . . . , im), (j1, . . . , jm) ∈ Σm.
Moreover
Π(m)(K(X⊗m))(P) = Qγm(P)MN m (C)Qγm(P),
where Qγm(P) ∈ MN m(C) is the projection onto the subspace Zγm(P) ⊂ CN m
Proof. It follows from the form of Dγm
Lemma 14. Let P ∈ Pγ and n = p + q. Assume that (i1, i2, . . . , ip) satisfies γik ◦···◦γi1(P) /∈
Bγ for any k = 1, . . . , p. Then it holds that
and the description of Bγm
(cid:3)
.
.
πq,i1...ip(N,p)(Π(q)(K(X⊗q))(γip ◦ ··· ◦ γi1(P))) ⊂ Π(n)(K(X⊗n))(P).
Proof. We shall show the left hand side is contained in the right hand side. Put Q = γip ◦···◦
γi1(P). Then Q /∈ Bγ. Π(q)(K(X⊗q))(Q) is a subalgebra of MN q (C). A matrix T ∈ MN q (C)
is contained in Π(q)(K(X⊗q))(Q) if and only if the following holds:
For any r and R with 1 ≤ r ≤ q and R = γip+r ◦ ··· γp+1(Q) ∈ Bγ,
Tiip+1 ...ip+r−1j(R,t)ip+r+1...in(N,q),jp+1...jn(N,q)
= Tiip+1 ...ip+r−1j(R,t′)ip+r+1...in(N,q),jp+1...jn(N,q)
,
(1)
and
Tip+1...in(N,q),ip+1...ip+r−1j(R,s)jr+1...jn(N,q)
(2)
for each 0 ≤ t, t′, s, s′ ≤ eR − 1 and for each (ip+r+1, . . . ,in), (jp+r+1, . . . , jn) ∈ Σq−r ,
(i1, . . . , iq), (j1, . . . , jq) ∈ Σq.
= Tip+1...in(N,q),ip+1...ip+r−1j(R,s′)jr+1...jn(N,q)
Now assume that T ∈ MN q (C) satisfies the above condition of compact algebra. Put
,
S = πq,i1...ip(N,p)(T ) ∈ MN n(C).
We add (i1, . . . , ip) to the row and the column of the condition (1), then we get the following:
Si1...ipip+1...ip+r−1j(R,t)ip+r+1...in(N,n),i1...ipjp+1...jn(N,n)
=Si1...ipip+1...ip+r−1j(R,t′)ip+r+1...in(N,n),i1...ipjp+1...jn(N,n)
.
To show that S is in Π(n)(K(X⊗n))(P), it is sufficient to show that the equality holds for
arbitrary column. For (j1, . . . , jp) 6= (i1, . . . , ip), both
Si1...ipip+1...ip+r−1j(R,t)ip+r+1...in(N,n),j1...jpjp+1...jn(N,n)
,
and
Si1...ipip+1...ip+r−1j(R,t′)ip+r+1...in(N,n),j1...jpjp+1...jn(N,n)
are 0, and the equality holds. Then if R ∈ Bγ, hp+r(R) = P, and h(R) = γip+r−1 ◦ ··· γi1(P),
it holds
Si1...ipip+1...ip+r−1j(R,t)ip+r+1...in(N,n),j1...jpjp+1...jn(N,n)
=Si1...ipip+1...ip+r−1j(R,t′)ip+r+1...tn(N,n),j1...jpjp+1...jn(N,n)
,
for each 0 ≤ t, t′ ≤ eR − 1, for each (ip+r+1, . . . ,in) ∈ Σq−r, and for each (j1, . . . , jn) ∈ Σn.
The argument also holds if we interchange columns and rows.
12
Therefore
S = πq,i1...ip(N,p)(T ) ∈ Π(n)(K(X⊗n)).
(cid:3)
If a block matrix T with the following form
A 0 A
0
0
T =
A 0 A
0
is known to be block diagonal, then T must be zero. The following lemma holds by a similar
reason.
Lemma 15. Let P ∈ Pγ. Assume that R = γip ◦ ··· ◦ γi1(P) ∈ Bγ for some 1 ≤ p ≤ n. Then
we have that
C(P, R, p) ∩ Π(n)(K(X⊗n))(P) = { 0}.
Proof. We chose T ∈ C(P, R, p) ⊂ MN n(C). We show that if T ∈ Π(n)(K(X⊗n))(P), then
T = 0. It suffices to show that (i1 . . . ip−1j(R, l)(N,p))-th element of the block diagonal matrix
T is zero, i.e. it is sufficient to show that
Ti1...,ip−1j(R,l)ip+1...in(N,n),i1...ip−1j(R,l)jp+1...jn(N,n)
is 0 for each (ip+1, . . . ,in), (jp+1, . . . , jn) ∈ Σn−p. Since T ∈ Π(n)(K(X⊗n)(P), it holds
Ti1...,ip−1j(R,l)ip+1...in(N,n),i1...ip−1j(R,l)jp+1...jn(N,n)
=Ti1...,ip−1j(R,l′)ip+1...in(N,n),i1...ip−1j(R,l)jp+1...jn(N,n)
.
(cid:3)
The right hand side is 0 because T is a block diagonal matrix.
Lemma 16. Assume 1 ≤ p, p′ ≤ n, hp(R) = hp′
and R 6= R′, then we have that
C(P, R, p) ∩ C(P, R′, p′) = { 0}.
(R′) = P, R, R′ ∈ Bγ. If p 6= p′, or p = p′
Proof. Assume h(R) = γip−1 ◦ ··· ◦ γi1(P), h(R′) = γip′−1 ◦ ··· ◦ γi1
(i1, . . . , ip−1) and (i1, . . . ,ip′−1) are uniquely determined by R, R′, p, p′.
First assume p = p′ and R 6= R′. Then it holds then
{ (i1, . . . , ip−1, j(R, s)) 0 ≤ s ≤ eR − 1} ∩ { (i1, . . . ,ip−1, j(R′, s′)) 0 ≤ s′ ≤ eR′ − 1} = ∅.
Therefore the conclusion follows.
Next assume that p′ ≤ p − 1.
If (i1, . . . , ip′−1) 6= (i1, . . . ,ip′−1), it holds
(P). We note that
C(P, R, p) ⊂πn−p′+1,i1...ip′−1(N,p′−1)
C(P, R′, p′) ⊂πn−p′+1,i1...ip′−1(N,p′−1)
(MN n−p′+1(C))
(MN n−p′+1(C)).
If (i1, . . . , ip′−1) = (i1, . . . ,ip′−1), for 0 ≤ l ≤ eR′ − 1 it holds ip′ 6= j(R′, l). Then it holds
C(P, R, p) ⊂πn−p′,i1...ip′−1ip′(N,p′)
C(P, R′, p) ⊂ X0≤l≤eR′−1
(MN n−p′ (C))
πn−p′,i1...ip′−1j(R′,l)(N,p′)
13
(MN n−p′ (C)).
Since the non-zero positions are different, the conclusion follows.
Lemma 17. Let P ∈ Pγ. Assume 1 ≤ p′ < p and R′ ∈ Bγ satisfy hp′
(R′) = P. We note that
(i1, . . . , ip′−1) is uniquely determined by h(R′) = γip′−1 ◦ ··· γi1(P) ∈ Cγ. Then we have that
(cid:3)
X
πq,i1...,ip′−1j(R′,l)ip′+1...ip
(ip′+1,...,ip)∈Σp−p′ , 0≤l≤eR′−1
(Π(q)(K(X⊗q)(γip ◦ ··· ◦ γip′+1 ◦ γj(R′,l) ◦ γip′−1 ◦ ··· ◦ γi1(P))) ⊂ C(P, R′, p′).
Proof. C(P, R′, p′) in the right hand side is
C(P, R′, p′) = { πq′,i1...ip′−1j(R′,0)(N,p′),...,i1...ip′−1j(R′,eR′−1)(N,p′)
(A) A ∈ MN n−p′ (C)}.
The left hand side is expressed as follows:
X(ip′+1,...,ip)∈Σp−p′ X0≤l≤eR′−1
πn−p′,i1...ip′−1j(R′,l)(N,p′)
Then
(πq,ip′+1...ip(N,p−p′)(Π(q)(K(X⊗q)((γip ◦ ··· ◦ γip′+1 ◦ γj(R′,l) ◦ γip′−1 ◦ ··· ◦ γi1)(P)).
πn−p′,i1...ip′−1j(R′,l)(N,p′)
X0≤l≤eR′−1
(πq,ip′+1...ip(N,p−p′)(Π(q)(K(X⊗q)((γip ◦ ··· ◦ γip′+1 ◦ γj(R′,l) ◦ γip′−1 ◦ ··· ◦ γi1)(P))
is contained in the right hand side for each (ip′+1, . . . , ip) ∈ Σp−p′
.
(cid:3)
The following thorem shows that the matrix representation over the coefficient algebra of
the n-th core is described by the fibres at P . They consist of two parts, that is, "the compact
operators" and the sum of block diagonal algebras C(P, R, p), which remind the pictures of
orbits through a branched point. The key point of the proof is to understand the two effects
of the branched points to "the compact operators" and the inclusion F (k) ⊂ F (k+1). For
example, these two are described by the following typical forms of matrices
0
A 0 A
0
0
A 0 A
and
B 0
0
0
0
0
0
0 B
Therefore it is easy to see that any block diagonal algebra C(P, R, p) protrudes from "the
compact operators" Π(n)(K(X⊗n))(P). The following theorem means that they are all which
protrude from "the compact operators" Π(n)(K(X⊗n))(P)..
Theorem 18. Let P ∈ Pγ. Then the fiber of Π(n)(F (n)) ⊂ C([0, 1], M2n (C)) at P is given by
Π(n)(F (n))(P) =Π(n)(K(X⊗n))(P) +
M
1≤p≤n, R∈Bγ , hp(R)=P
C(P, R, p)
.
The first term and the second term is the sum as vector spaces and not as *-algebras.
14
Proof. We shall show the left hand side is contained in the right hand side. Let 0 ≤ q ≤ n.
It is enough to show that Π(n)(K(X⊗q)(P) are contained in the right hand side because
n
Xq=0
Fix 0 ≤ q ≤ n and put p = n − q. Then it holds
Π(n)(F (n))(P) =
Π(n)(K(X⊗q))(P).
Π(n)(K(X⊗q))(P) = X(i1,...,ip)∈Σp
πq,i1...ip(N,p)(K(X⊗q)((γip ◦ ··· ◦ γi1)(P))).
We divide Σp into three groups as follow:
Ap ={ (i1, . . . , ip) ∈ Σp γik ◦ ··· ◦ γi1(P) /∈ Bγ
Bp ={ (i1, . . . , ip) ∈ Σp γip ◦ ··· ◦ γi1(P) ∈ Bγ },
C p ={ (i1, . . . , ip) ∈ Σp γip′ ◦ γip′−1 ◦ ··· ◦ γi1(P) ∈ Bγ,
Let Q = γip ◦ ··· ◦ γi1(P). Then Ap corresponds to the case that there exists no branched
point before Q. Bp corresponds to the case that Q is a branched point. C p corresponds to
the case that there exists a branched point before Q.
for some p′ (1 ≤ p′ ≤ p − 1)}.
for any k (1 ≤ k ≤ p)},
First we investigate the case Ap. By Lemma 14, it holds
X(i1,...,ip)∈Ap
πq,i1...ip(N,p)Π(n)(K(X⊗q)(γip ◦ ··· ◦ γi1(P))) ⊂ Π(n)(K(X⊗n))(P).
Next we investigate the case Bp. We assume that R1, . . . , Rm ∈ Bγ satisfy hp(Rk) = P
for k = 1, . . . , m. Then it holds
h(Rk) = (γik
1 ◦ ··· ◦ γik
p−1
)(P) k = 1, . . . , m.
1, . . . , ik
(ik
as:
p−1) (k = 1, . . . , m) are uniquely determined. Then the set of indices Bp is described
Bp =
Then it holds
m
[k=1
{ (ik
1, . . . , ik
p−1, j(Rk, l)) l = 0, . . . , eRk − 1}.
πq,i1...ip(N,p)Π(K(X⊗q)(γip ◦ ··· ◦ γi1(P)))
πq,ik
1 ...ik
p−1j(Rk,l)(Π(q)(K(X⊗q)(Rk))
πq,ik
1...ik
p−1j(Rk,l)(Ak) Ak ∈ MN q (C)}
m
=
eRk−1
X(i1,...,ip)∈Bp
Xk=1
Xl=0
Xl=0
Xk=1
={
Xk=1
eRk−1
=
m
m
C(P, Rk, p).
15
Last we investigate the case C p. It holds
πq,i1...ip(N,p)Π(K(X⊗q)(γip ◦ ··· ◦ γi1(P)))
X(i1,...,ip)∈C p
= X1≤p′≤p−1, XR∈Bγ ,hp′ (R)=P, X0≤l≤eR−1, X(ip′+1,...,ip )∈Σp−p′
πq,i1...ip′−1j(R,l)ip′+1...ip(N,p)
⊂ X1≤p′≤p−1, XR∈Bγ ,hp′ (R)=P
(Π(q)(K(X⊗q))(γip ◦ ··· ◦ γip′+1(R)))
C(P, R, p′).
The last inclusion follows form Lemma 17.
The right hand side is contained in the left hand side, because the equality holds for Bp. (cid:3)
thogonal, we shall modify C(P, R, p) to make them orthogonal. We define the modification
C(P, R, p) by
Definition 19. Since Π(n)(K(X⊗n))(P) and (cid:16)L1≤p≤n, R∈Bγ , hp(R)=P
C(P, R, p) := (I − Qγn(P)) C(P, R, p) 6= 0,
C(P, R, p)(cid:17) are not or-
where Qγn(P) ∈ MN n(C) is the projection onto the subspace Zγn(P) ⊂ CN n
. Since the sub-
space Zγn(P) is invariant under any operator in C(P, R, p), the projection Qγn(P) commutes
with any operator in C(P, R, p). Therefore C(P, R, p) is a C∗-algebra and is isomorphic to
MN q (C). We may write it as follows:
πq,i1...ip−1j(R,l)(N,p)(A)
C(P, R, p) = { X0≤l≤eR−1
eR X0≤m,n≤eR−1
−
1
σq,i1...ip−1j(R,m)(N,p),i1...ip−1j(R,n)(N,p)(A) A ∈ MN q (C)}.
Theorem 20. Let P ∈ Pγ. Then the fiber of Π(n)(F (n)) ⊂ C([0, 1], M2n (C)) at P is given by
Π(n)(F (n))(P) =Π(n)(K(X⊗n))(P) ⊕
M
1≤p≤n, R∈Bγ , hp(R)=P
and the right hand side is a direct sum as *-algebras.
Proof. For any T ∈ C(P, R, p),
C(P, R, p)
,
T = Qγn(P)T + (I − Qγn(P))T,
and Qγn(P)T = Qγn(P)T Qγn(P) is in Π(n)(K(X⊗n))(P). These imply the conclusion.
(cid:3)
The following Proposition will be used later to describe the model traces in terms of matrix
representations.
Propositoin 21. In the situation of the above Theorem, for q = n − p with q + 1 ≤ s ≤ n,
C(P, R, p) is orthogonal to Π(n)(K(X⊗s) ⊗ I)(P).
16
Proof. Put Q = γin−s ◦ ··· ◦ γi1(P). Consider C(Q, R, s − q) in Π(s)(F (s))(Q). Then we see
that C(Q, R, s − q) is orthogonal to Π(s)(K(X⊗s)(Q).
Since there exists no branched points,
C(P, R, p) = πs,i1...in−s(N,n−s)(C(Q, R, s − q)).
Keep the i1 . . . in−s(N,n−s)-th block-diagonal component of Π(n)(K(X⊗s) ⊗ I)(P) and make
all the other components zero. Then we get the matrix
πs,i1...in−s(N,n−s)Πs(K(X⊗s)(Q).
Therefore C(P, R, p) is orthogonal to πs,i1...in−s(N,n−s)Πs(K(X⊗s)(Q). Moreover it is clear
that C(P, R, p) is orthogonal to any other block-diagonal component of Π(n)(K(X⊗s)⊗ I)(P).
Therefore C(P, R, p) is orthogonal to Π(n)(K(X⊗s) ⊗ I)(P).
4.2. Tent map. We describe the matrix representation for the case of tent map. Since proofs
are given in general situation, we only present the results.
(cid:3)
(Case of the fiber at x = 1 ):
We fix n and consider the fiber Π(n)(F (n))(1) of Π(n)(F (n)) at x = 1. We investigate p − q
orbit through the branched point 1
2 .
1
γ0,γ1=⇒
1
2
ր...
ց
ր...
ց
The following is the condition that a matrix is contained in the image of a compact algebra.
Π(n)(K(X⊗n))(1)
={T ∈ M2n(C) T0i2i3...in,j1j2...jn = T1i2i3...in,j1j2...jn,
Ti1i2i3...in,0j2...jn = Ti1i2i3...in,1j2...jn
=(cid:26)(cid:18)B B
Since γ0(1) = γ1(1) = 1/2 ∈ Bγ, only C(1, 1/2, 1) appears for x = 1. Put
B B(cid:19) B ∈ M2n−1 (C)(cid:27) .
for each (i2, i3, . . . , in) and (j1, j2, j3, . . . , jn)
for each (i1, i2, i3, . . . , in) and (j2, j3, . . . , jn) }
(3)
C(1, 1/2, 1) = { πn−1,0(2,1),1(2,1)(A) A ∈ M2n−1 (C)} =(cid:26)(cid:18)A O
O A(cid:19) A ∈ M2n−1(C)(cid:27) ,
and put
C(1, 1/2, 1) = { πn−1,0(2,1),1(2,1)(A) −
=(cid:26)(cid:18) (1/2)A −(1/2)A
−(1/2)A (1/2)A (cid:19) A ∈ M2n−1 (C)(cid:27) .
1
2
σn−1,0(2,1),1(2,1)(A) A ∈ M2n−1 (C)}
Then we get the following matrix representation at x = 1:
Π(n)(F (n)(1))
=Π(n)(K(X⊗n))(1) ⊕ C(1, 1/2, 1)
=(cid:26)(cid:18)(1/2)A (1/2)A
(1/2)A (1/2)A(cid:19) A ∈ M2n−1 (C)(cid:27) ⊕(cid:26)(cid:18) (1/2)A −(1/2)A
−(1/2)A (1/2)A (cid:19) A ∈ M2n−1 (C)(cid:27)
17
(Case of the fiber at x = 0 ).
We investigate p − q orbit through the branched point 1
2 .
0
γ0→ 0
γ0→ 0 . . . 0
γ1→ 1
γ0,γ1=⇒
1
2
ր...
ց
ր...
ց
We consider Π(n)(K(X⊗n))(0). The point 0 is contained in Pγ and is not contained in Cγ.
The range of the compact algebra Π(n)(K(X⊗n))(0) is the set of matrices in M2n(C) which
satisfy the following conditions for rows and columns:
T0···010,j1···jn = T0···011,j1···jn, Ti1···in,0···010, = Ti1···in,0···011,
T0···010in,j1···jn = T0···011in,j1···jn, Ti1···in,0···010jn, = Ti1···in,0···011jn,
. . .
T0···010ik···in,j1···jn = T0···011ik···in,j1···jn, Ti1···in,0···010jk···jn, = Ti1···in,0···011jk···jn,
. . .
T010i3···in,j1···jn = T011i3···in,j1···jn,
T10i2···in,j1···jn = T11i2···in,j1···jn, Ti1···in,10j2···jn, = Ti1···in,11j2···jn.
Ti1···in,010j3···jn, = Ti1···in,011j3···jn,
(0, 0, . . . , 0, 1).
We note that for 2 ≤ p ≤ n γip−1◦γip−2◦···◦γi1(0) ∈ Cγ if and only if (i1, i2, . . . , ip−2, ip−1) =
For each 2 ≤ p ≤ n, we define
C(0, 1/2, p) = { πn−p,0...010(2,p),0...011(2,p)(A) A ∈ M2n−p(C)},
and also define
C(0, 1/2, p) ={πn−p,0...010(2,p),0...011(2,p)(A) − σn−p,0...010(2,p),0...011(2,p)((1/2)A) A ∈ M2n−p(C)}
O
O
O
O
O (1/2)A −(1/2)A O
O −(1/2)A (1/2)A O
O
O
O
O
=
A ∈ M2n−p(C)
.
(Matrix representation):
The expression of the total algebra is as follows:
0 < x < 1,
Π(n)(F (n))(x) = M2n(C)
Π(n)(F (n))(1) = Π(n)(K(X⊗n))(1) ⊕ C(1, 1/2, 1) ≃ M2n−1 (C) ⊕ M2n−1 (C), (n ≥ 1)
Π(n)(F (n))(0) = Π(n)(K(X⊗n))(0) ⊕
M2p(C).
C(0, 1/2, p) ≃ M2n−1+1(C) ⊕
n−2
n
Mp=0
Mp=2
18
Moreover we have the following:
0 ≤ x ≤ 1,
0 ≤ x < 1,
Π(0)(F (0))(x) = Π(0)(A)(x) ≃ C
Π(1)(F (1))(x) ≃ M2(C)
Π(1)(F (1))(1) ≃ C ⊕ C,
Π(2)(F (2))(0) ≃ M3(C) ⊕ C
Π(2)(F (2))(1) ≃ M2(C) ⊕ M2(C)
Π(3)(F (3))(0) ≃ M5(C) ⊕ C ⊕ M2(C)
Π(3)(F (3))(1) ≃ M4(C) ⊕ M4(C)
4.3. Sierpinski Gasket case. Recall the set Bγ = { S, T, U} of branched points and the set
Cγ = { P, Q, R} of branch values. We list the image of branched points and branch values
under the contractions (γ0, γ1, γ2) for the convenience of computation. .
γ0(P) = P,
γ0(Q) = S,
γ0(R) = U,
γ1(P) = T,
γ1(Q) = S,
γ1(R) = Q,
γ2(P) = T,
γ2(Q) = R,
γ2(R) = U.
(Picture of the fiber at P ).
We investigate p − q orbits through the branched point T .
γ1,γ2=⇒ T ր...
ց
γ0→ P . . .
γ0→ P
γ0→ P
P
The beginning γ0 may be omitted.
ր...
ց
(Picture of the fiber at Q ).
We investigate p − q orbits through the branched point S.
Q
γ2→ R
γ1→ Q
γ2→ R
γ1→ Q . . .
γ2→ R
γ1→ Q
γ0,γ1=⇒ S ր...
ց
ր...
ց
We investigate p − q orbits through the branched point U .
Q
γ2→ R
γ1→ Q
γ2→ R
γ1→ Q . . .
γ1→ R
γ0,γ2=⇒ U ր...
ց
ր...
ց
satisfying the following all conditions:
The picture of the fiber at R is similar with the picture of the fiber at Q.
Let 1 ≤ p ≤ n. Elements in Π(n)(K(X⊗n))(P) are described as elements in M3n(C)
• (1, i2, . . . , in)th row and (2, i2, . . . , in)th row are equal, and (1, i2, . . . , in)th column
• (0, 1, i3, . . . , in)th row and (0, 2, i3, . . . , in)th row are equal and (0, 1, i3, . . . , in)th col-
and (2, i2, . . . , in)th column are equal.
umn and (0, 2, i3, . . . , in)th column are equal.
19
• (0, . . . , 0, 1, ik+1, . . . , in)th row and (0, . . . , 0, 2, ik+1, . . . , in)th row are equal and
(0, . . . , 0, 1, ik+1, . . . , in)th column and (0, . . . , 0, 2, ik+1, . . . , in)th column are equal for
(1 ≤ k ≤ n − 1).
(0, . . . , 0, 2)th column are equal.
• (0, . . . , 0, 1)th row and (0, . . . , 0, 2)th row are equal and (0, . . . , 0, 1)th column and
The condition of compactness at Q and R are similar, and the dimension of the image under
matrix representation of the compact algebras are equal for each branch value. We note that
γip−1 ◦ γip−2 ◦ ··· ◦ γi1(P) ∈ Cγ if and only if (i1, . . . , ip−2, ip−1) = (0, . . . , 0, 0). Put
C(P, T, p) = { πn−p,00...1(3,p),00...2(3,p)(A) A ∈ M3n−p (C)}.
We also put
C(P, T, p) =n πn−p,00...1(3,p),00...2(3,p)(A) − (1/2)σn−p,00...1(3,p),00...2(3,p)(A) A ∈ M3n−p (C)o .
Then by the definition of C(P, T, p) it holds
Π(n)(K(X⊗n))(P) + C(P, T, p) = Π(n)(K(X⊗n))(P) ⊕ C(P, T, p).
Then it holds
Π(n)(F (n))(P) = Π(n)(K(X⊗n))(P) ⊕
C(P, T, p)
n−1
Mp=0
Next we consider the cases Q and R. Since Q and R are symmetric, we only consider the case
Q. Let 1 ≤ p ≤ n−1. Then γip−1◦···◦γi1(Q) ∈ Cγ if and only if (i1, . . . , ip−1) = (2, 1, 2, 1, . . . )
We note that the form changes according to the the parity of p. For the case p is even, put
C(Q, U, p) = { πn−p,2121...20(3,p),2121...22(3,p)(A) A ∈ M3n−p(C)}.
For the case p is odd, put
C(Q, S, p) = { πn−p,2121...10(3,p),2121...11(3,p)(A) A ∈ M3n−p(C)}.
Moreover, for p even, put
C(Q, U, p) = { πn−p,2121...20(3,p),2121...22(3,p)(A)−σn−p,2121...20(3,p),2121...22(3,p)((1/2)A) A ∈ M3n−p(C)},
and for p odd, put
C(Q, S, p) = { πn−p,2121...20(3,p),2121...22(3,p)(A)−σn−p,2121...10(3,p),2121...11(3,p)((1/2)A) A ∈ M3n−p(C)}.
We put
Then also by Theorem 18,
Hp =( C(Q, U, p)
C(Q, S, p)
(p is even ),
(p is odd ).
n
Π(n)(F (n))(Q) = Π(n)(K(X⊗n))(Q) ⊕
Hp.
Mp=1
We consider the case of R. For p even, put
C(R, S, p) = { πn−p,1212...10(3,p),1212...11(3,p)(A)−σn−p,1212...10(3,p),1212...11(3,p)((1/2)A) A ∈ M3n−p(C)},
and for p odd, put
C(R, U, p) = { πn−p,1212...20(3,p),1212...22(3,p)(A)−σn−p,1212...20(3,p),1212...22(3,p)((1/2)A) A ∈ M3n−p(C)}.
20
Similarly we put
Ip =( C(R, S, p)
C(R, U, p)
(p is even ),
(p is odd ).
Then by Theorem 18, it holds that
n
Π(n)(F (n))(R) = Π(n)(K(X⊗n))(R) ⊕
Mp=1
Therefore we get the matrix representation of F (n) as follows:
Π(n)(F (n))(X) ≃M3n(C), X 6= P, Q, R
Mp=1
Π(n)(F (n))(P) =Π(n)(K(X⊗n))(P) ⊕
Mp=1
Mp=1
Π(n)(F (n))(Q) =Π(n)(K(X⊗n))(Q) ⊕
Π(n)(F (n))(R) =Π(n)(K(X⊗n))(R) ⊕
n
n
n
Ip.
n
n
Mp=1
Mp=1
Mp=1
n
C(P, T, p) ≃ M(1/2)(3n +1)(C) ⊕
M3p−1(C),
Hn−p+1 ≃ M(1/2)(3n +1)(C) ⊕
M3p−1(C),
In−p+1 ≃ M(1/2)(3n +1)(C) ⊕
M3p−1(C).
Three algebras are isomorphic as C∗-algebras. But the realizations of them in M3n(C) are
different.
5. Calculation of K-groups of the cores
5.1. Tent map case. From the explicit calculation of matrix representation, we have that
Π(n)(F (n))(x) = M2n(C),
0 < x < 1,
D(1) :=Π(n)(F (n))(1) = Π(n)(K(X⊗n))(1) ⊕ C(1, 1/2, 1) ≃ M2n−1 (C) ⊕ M2n−1 (C).
D(0) := Π(n)(F (n))(0) = Π(n)(K(X⊗n))(0) ⊕
We introduce the following notation:
n−2
Mp=0
C(0, 1/2, p) ≃ M2n−1+1(C) ⊕
M2p(C).
n−2
Mp=0
J ={ T ∈ C([0, 1], M2n (C)) T (0) = O, T (1) = O },
B ={ T ∈ C([0, 1], M2n (C)) T (0) ∈ D(0), T (1) ∈ D(1)} = Π(n)(F (n)),
C =D(0) ⊕ D(1).
Then we have the following exact sequence:
From this, we have the following 6-term exact sequence of K-groups:
{0} → J → B → C → {0}.
K0(J) −−−−→ K0(F (n)) −−−−→ K0(C)
indx
K1(C) ←−−−− K1(F (n)) ←−−−− K1(J).
21
exp
y
Substituting known K-groups, we have that
{0} −−−−→ K0(F (n)) −−−−→ Z2 ⊕ Zn
indx
{0} ←−−−− K1(F (n)) ←−−−− K1(J).
exp
y
We calculate the exponential map (exp map) from K0(C) to K1(J). Use 12.2 in Rodam et al.
[17], for example. Let p1 be a minimal projection in Π(n)(K(X⊗n))(1) ≃ M2n−1(C). There
exists an Hermite element T ∈ M2n(C[0, 1]) with T (1) = p1, T (0) = 0. Then δ(([p1], 0)) =
−[exp(2πiT )] = −1. Thus −1 is the image of [p1] in K1(J) under the exponential map. On the
other hand, let p2 be a minimal projection in C(1, 1/2, 1) ≃ M2n−1(C). Then There exists an
Hermite element T ∈ M2n(C[0, 1]) with T (1) = 0, T (0) = p2. Thus δ((0, [p2])) = 1 ∈ K1(J).
For a positive element in Z2 ⊕ Zn, exp map is described as
exp((m1, m2, r1, . . . , rn)) = −m1 − m2 + r1 + ··· + rn.
K1(J) ≃ Z and the exp map is surjective. Then for each n, it holds that K1(F (n)) = {0}.
By the extension of exp map by homomorphism property, it is shown that exp is given on
the whole part of Z2 ⊕ Zn.
On the other hand, it holds that
K0(F (n)) ≃ ker(exp) ≃ { ((m1, m2), (r1, . . . , rn)) ∈ Z2⊕ Zn m1 + m2 = r1 +··· + rn } ≃ Zn+1.
1 be a minimal projection
Next, we calculate the embeddings of { K0(F (n))}n=0,1,2,.... Let p1
of C(1, 1/2, 1) and p1
2 be a minimal projection of Π(n)(K(X))(1). Then it holds that
2] m1, m2 ∈ Z} ≃ { (m1, m2) m1, m2 ∈ Z}.
1] + m2[p1
K0(D(1)) = { m1[p1
i be a minimal projection of C(0, 1/2, i + 1) for 1 ≤ i ≤ n − 1 and p0
n be a minimal
Let p0
projection of Π(n)(K(X⊗n))(0). Then it holds
1] + ··· + rn[p0
K0(D(0)) = { r1[p0
n] ri ∈ Z} ≃ { (r1, . . . , rn) ri ∈ Z}.
We denote by rn
Let 1 ≤ i ≤ n, and we calculate the embedding of ((1, 0), (0, . . . , 1i, . . . , 0)) ∈ Z2 ⊕ Zn. We
write as Ψn,n+1 the embedding map from K0(F (n)) to K0(F (n+1)).
1 a minimal projection of C(1, 1/2, 1), and by qn
i a minimal projection of
C(0, 1/2, n − i + 1). We take T ∈ F (n) such that T (0) = qn
1 . We calculate the
forms of Π(n+1)(T )(0) and Π(n+1)(T )(1). It holds that
Π(n+1)(T )(0) = diag(Π(n)(T )(0), Π(n)(T )(1)), Π(n+1)(T )(1) = diag(Π(n)(T )(1/2), Π(n)(T )(1/2)).
i , T (1) = rn
Then it holds that
Ψ(n,n+1)((1, 0), (0, . . . , 1i, . . . , 0)) = ((1, 1), (11, . . . , 1i+1, . . . , 0)).
Similarly, it holds that
Ψ(n,n+1)((0, 1), (0, . . . , 1i, . . . , 0)) =((1, 1), (0, . . . , 1i+1, . . . , 1n+1))
Ψ(n,n+1)((1, 0), (0, . . . , 0, . . . , 1n)) =((1, 1), (11, . . . , 0, . . . , 1n+1))
Ψ(n,n+1)((0, 1), (0, . . . , 0, . . . , 1n)) =((1, 1), (0, . . . , 0, . . . , 2n+1)).
The algebra K0(F (∞)) can be described as the inductive limit algebra under the embedding
given by Ψn,n+1.
22
As an basis in K0(F (n)), we can take
en
i = ((1, 0), (0, . . . , 0, 1i, 0, . . . , 0))
We note that
(i = 1, . . . , n),
en
n+1 = ((0, 1), (1, 0, . . . , 0, 0)).
((0, 1), (0, . . . , 0, 1i, 0, . . . , 0)) = en
Then the action of Ψ(n,n+1) on the bases is expressed as
n+1 + en
i − en
1 .
Ψ(n,n+1)(en
i ) =((1, 1), (11, 0, . . . , 0, 1i+1, 0, . . . , 0)) = en+1
i+1 + en+1
n+2
(i = 1, . . . , n),
Ψ(n,n+1)(en
n+1) =((1, 1), (0, . . . , 2n+1))
=((1, 0), (0, . . . , 0, 1n+1)) + ((0, 1), (11, . . . , 0)) − ((1, 0), (11, . . . , 0))
+ ((1, 0), (0, . . . , 1n+1)) = −en+1
1 + 2en+1
n+1 + en+1
n+2.
The matrix representation A(n,n+1) of Ψn,n+1using the bases is expressed as
A(n,n+1) =
0 0 0 ···
1 0 0 ···
0 1 0 ···
...
. . .
0 0 0 ···
1 1 1 ···
...
...
0 −1
0
0
0
0
...
...
2
1
1
1
.
The matrix A(n,n+1) is an (n + 2) × (n + 1) matrix whose entries are integers.
Propositoin 22. The K group of the core of the C∗-algebra associated with the tent map is
given as follows:
n→∞(cid:18)Zn+1 A(n,n+1)
→ Zn+2(cid:19)
K0(F (∞)) = lim
K1(F (∞)) ={ 0}.
Lemma 23. The map Φ(n,n+1) is injective for each n.
Proof. Since the matrix A(n,n+1) is of rank n, Φ(n,n+1) is injective.
Theorem 24. Let F (∞) be the core of the C∗-algebra associated with tent map, then K0(F (∞))
is isomorphic to the countably generated free abelian group Z∞ ∼= Z[t] as an abstract abelian
group. Moreover for the tracial state τ (∞) on F (∞) corresponding to the Hutchinson measure,
we have that τ (∞)∗(K0(F (∞))) = Z[ 1
2 ].
Proof. Since each n Φ(n,n+1) is injective, the canonical map of K0(F (n)) to K0(F (∞)) is
[17]. By Theorem 7.8. of [7] we
injective, for example, see Exercises 6.7 in Rodam et al.
know that the extreme tracial states on the core of the C∗-algebra associated with the tent
map are exactly the model traces
(cid:3)
{τ (∞)}[{τ (1/2,r) r = 0, 1, 2, . . . }.
where 1/2 is the unique branched point of the tent map. Put τ (r) = τ (1/2,r). Then they
are described in Example 7.1. of [7] as follows: We define discrete measures µ(r)
i on [0, 1] for
23
0 ≤ i ≤ r by
µ(r)
i (f ) =
1
2r−i
X(j1,...,jr−i)∈{ 1,2 }r−i
f (γj1 ◦ ··· ◦ γjr−i(1/2)),
(f ∈ C[0, 1]).
It is enough to define traces τ (r) on F (∞) for r ≥ 0 by the restriction τ (r)
i = τ (r)K(X ⊗i
γ ) as
τ (r)
i =( πi(µ(r)
0
(0 ≤ i ≤ r)
i )
(r + 1 ≤ i)
γ ) corresponding to µ(r)
i using Morita equiva-
for each i, where πi(µ(r)
lence. We can also define a trace τ (∞) on F (∞) by
i ) is the trace on K(X⊗i
τ (∞)
i = πi(µ∞),
for each i where µ∞ = µH is the normalized Lebesgue measure on [0, 1] and coincides with
the Hutchinson measure. The set of extreme traces on the core of C∗-algebras associated
with the tent map on [0, 1] is exactly { τ (0), τ (1), . . . , τ (∞) }.
Fix an integer n ≥ 0 for a while. We need to describe the restriction τ (r)F (n) of these
model traces τ (r) on F (n) in termes of matrix representations. It is clear that τ (r)K(X ⊗n
γ ) = 0
for n ≥ r + 1 by definition. We denote by tr the normalized trace on the full matrix algebra.
Let n = 0. Then F (0) = A and τ (0)F (0)(a) = a( 1
Let n = 1. Then F (1) = A ⊗ I + K(Xγ) and Π(1)(F (1)) ⊂ M2(A). And
2 ) for a ∈ A.
τ (1)F (1)(T ) = tr(Π(1)(T )(
1
2
))
for T ∈ F (1). Let E(1) be the projection of the fiber
Π(1)(F (1))(1) = Π(1)(K(Xγ))(1) ⊕ C(1,
1
2
, 1)
onto C(1, 1
2 , 1) = C. Then
τ (0)F (1)(T ) = E(1)(Π(1)(T )(1)).
In fact, if T = a ⊗ I, then E(1)(Π(1)(T )(1)) = a(γ0(1)) = a( 1
2 ).
τ (0)(T ) = 0 = E(1)(Π(1)(T )(1)) by definition.
If T is in K(Xγ), then
Let n ≥ 2 in general. By induction, we can show the following: Recall that Π(n)(F (n)) ⊂
M2n(A) . Then
τ (n)F (n)(T ) = tr(Π(n)(T )(
1
2
))
for T ∈ F (n). Let E(n−1) be the projection of the fiber
Π(n)(F (n))(1) = Π(n)(K(X⊗n
γ
))(1) ⊕ C(1,
onto the direct sum component C(1, 1
2 , 1) = M2n−1 (C). Then
1
2
, 1)
τ (n−1)F (n)(T ) = tr(E(n−1)(Π(n)(T )(1))),
24
for T ∈ F (n). Take r = n − p for some p with 2 ≤ p ≤ n. Let E(n−p) be the projection of the
fiber
onto the direct sum component C(0, 1
Π(n)(F (n))(0) = Π(n)(K(X⊗n
))(0) ⊕ ⊕n
2 , p) = M2n−p(C). Then
γ
p=2C(0,
, p)
1
2
τ (n−p)F (n)(T ) = tr(E(n−p)(Π(n)(T )(0))).
for T ∈ F (n). In fact, if T = S⊗I for some S ∈ L(X⊗n−p
Π(n−p)(S)( 1
γ
2 ), since p − (n − p) orbit through the branched point 1
ր...
ց
γ0→ 0 . . . 0
ր...
ց
γ0→ 0
γ1→ 1
γ0,γ1=⇒
1
2
0
2 is
)∩F (n−p), then E(n−p)(Π(n)(T )(0)) =
if T = S ⊗ I for some S ∈ K(X⊗k
because Π(n)(T )(0) and C(0, 1
γ ) ∩ F (k) with k ≥ n − p, then E(n−p)(Π(n)(T )(0)) = 0,
2 , p) are orthogonal by Proposition 21. Therefore
τ (n−p)F (n)(T ) = tr(E(n−p)(Π(n)(T )(0))).
for T ∈ F (n).
We define a map
ϕ : K0(F (∞)) →
R
∞
Yn=0
by ϕ([p]) = (τ (n)(p))n for projections p in F (∞). Define projections p0 = I ∈ A and pn ∈
K(X⊗n
) for n = 1, 2, 3, . . . by constant matrix function of Jn such that for any x ∈ [0, 1]
Π(n)(pn)(x) is the rank-one projection Jn, where any entry of Jn is 1
γ
2n . Put
cn := ϕ([pn]) = (0, 0, . . . , 0,
1
2n ,
1
2n ,
1
2n , . . . ).
We shall show that ϕ(K0(F (∞))) is isomorphic to the countably generated free abelian group
Z∞.
In fact it is enough to show that ϕ(K0(F (∞))) is generated by these Z-independent
elements {cn ∈ Q∞n=0
R n = 0, 1, 2, . . . }. We note that ϕ(K0(F (∞))) contains cn − 2cn+1 =
2n , 0, . . . ) and
(0, 0, . . . , 0, 1
τ (n)
∗
(K0(F (∞))) = {
m
2n m ∈ Z}.
Consider the canonical inclusion in : F (n) → F (∞), then
ϕ(K0(F (∞))) = ∪∞n=0ϕ(in ∗ (K0(F (n)))).
We shall also show that ϕ is one to one. Therefore it is enough to show that ϕ(in ∗
(K0(F (n)))) is contained in the subgroup generated by {cm ∈ Q∞n=0
R m = 0, 1, 2, . . . }
and ϕ ◦ in∗ : K0(F (n)) → Q∞n=0
R is one to one for each n by Proposition 6.2.5 in [17]. Let
n+1} be a basis of K0(F (n)) chosen before. Then for k = 1, 2, . . . , n + 1, we have
{en
2 , . . . , en
1 , en
τ (0)(en
k ) = δk,2, τ (1)(en
k ) =
δk,3,
1
2
τ (2)(en
k ) =
1
22 δk,4,
. . . , τ (n−2)(en
k ) =
25
1
2n−2 δk,n, τ (n−1)(en
k ) =
1
2n−1 δk,n+1,
and
Moreover for any m ≥ n + 1,
τ (n)(en
k ) =
τ (m)(en
k ) =
1
2n .
1
2n .
Hence the image of the basis {en
K0(F (n)) → Q∞n=0
subgroup generated by generated by { cm ∈ Q∞n=0
is isomorphic to Z∞. It is easy to see that τ∞∗(K0(F∞) = Z[ 1
2 ].
n+1} under ϕ is independent over Z. Thus ϕ◦ in∗ :
R is one to one. We also have that ϕ(in ∗ (K0(F (n)))) is contained in the
R m = 0, 1, 2, . . . }. Therefore K0(F (∞))
2 , . . . , en
1 , en
(cid:3)
Remark 5.1. Compare the tent map with the following self-similar map. Let K = [0, 1],
γ0(y) = (1/2)y, γ1(y) = (1/2)y + 1/2. Then γ = (γ0, γ1) has no branched points. For this
γ, Oγ ≃ O2, and F (∞) is the UHF C∗-algebra M2∞ (C), and it is well known that K0(F (∞))
is the group Z[ 1
2 ] is not isomorphic to
Z∞. In Z[ 1
5.2. Sierpinski Gasket case. We calculate the K-group of the core of the C∗-algebra asso-
ciated with the self similar map which gives Sierpinski Gasket S.
2 ], for any a, 2x = a has a solution x. But in Z∞, 2x = c1 has no solution x.
2 ] of 2-adic integers and totally ordered. Note that Z[ 1
Lemma 25. Let S be the Sierpinski Gasket S. Then we have that
K0(C(S)) = Z, K1(C(S)) = Z∞,
Proof. The K group of the Sierpinski Gasket S is calculated by the inductive limit construc-
tion. The inductive limit of the figures Sn is obtained by cutting off 3 open discs from the
unit disc successively is homeomorphic to S. Since K0(C(Sn)) is isomorphic to Z, we have
that K0(C(S)) is also isomorphic to Z. Since the number of free generators of K1 increases
by 3, K1(C(S)) is isomorphic to Z∞.
(cid:3)
We calculate the K-groups of the finite cores of the C∗-algebra associated with the Sier-
pinski Gasket using explicit calculation of exp maps. For the Sierpinski Gasket S , K1(C(S))
contains many elements in contrast to the tent map case.
We denote the following algebras:
n−1
n−1
M3p(C),
n−1
M3p(C)),
n−1
Mp=0
Mp=0
Mp=0
n−1
D(Q) =Π(n)(K(X⊗n))(Q) ⊕
D(P) =Π(n)(K(X⊗n))(P) ⊕
C(P, T, p) ≃ M(1/2)(3n +1)(C) ⊕
H n−p+1 ≃ M(1/2)(3n +1)(C) ⊕
Mp=0
Mp=0
Mp=0
Jn ={ T ∈ C(S, MN n(C)) T (P) = O, T (Q) = O, T (R) = O },
B =(F (n) =){ T ∈ C(S, MN n(C)) T (P) ∈ D(P), T (Q) ∈ D(Q), T (R) ∈ D(R)},
C =D(P) ⊕ D(Q) ⊕ D(R).
I n−p+1 ≃ M(1/2)(3n +1)(C) ⊕
D(R) =Π(n)(K(X⊗n))(R) ⊕
n−1
M3p(C),
26
indx
indx
exp
y
exp
y
We have the exact sequence:
Using this, it holds the following 6-term exact sequence:
{0} → Jn → B → C → {0}.
K0(Jn) −−−−→ K0(F (n)) −−−−→ K0(C)
K1(C) ←−−−− K1(F (n)) ←−−−− K1(Jn).
We need to consider Jn, which is the adjoining unit C∗-algebra of Jn, when we calculate
K1(Jn). We put J = J0. Since Jn = J0 ⊗ M3n(C), the K1-group of Jn is identical to the
K1-group of J.
To compute K1( J ), we use the following 6 term exact sequence of K-groups:
K0(J) −−−−→ K0(C(S)) −−−−→ K0(C3)
K1(C3) ←−−−− K1(C(S)) ←−−−− K1(J).
Then it holds that
K0(J) −−−−→ Z −−−−→ Z3
indx
{0} ←−−−− Z∞ ←−−−− K1( J).
exp
y
Since the identity I in C(S) is mapped to (1, 1, 1) in C3, the map from K0(C(S)) to K0(C3)
is injective. It hollows that K0(J) is equal to {0}. Then the following long exact sequence
{0} → Z → Z3 → K1( J ) → Z∞ → {0}
holds. Then it holds that
We substitute known K group:
K1( J) ≃(cid:0)Z3/Z(cid:1) ⊕ Z∞ ≃ Z2 ⊕ Z∞.
{0} −−−−→ K0(F (n)) −−−−→ Zn+1 ⊕ Zn+1 ⊕ Zn+1
indx
{0} ←−−−− K1(F (n)) ←−−−−
exp
y
K1(Jn).
Since C has a unit, we can write exp map explicitly using Proposition 12.2.2 (ii) of Rordam et
al. [17]. Since the Sierpinski Gasket is connected, for projections p1, p2, p3, if their dimensions
are not equal, there does not exist an self-adjoint element T ∈ F (n) such that T (P) = p1,
T (Q) = p2 and T (R) = p3, and if their dimensions are equal there exists an Hermit element
T ∈ F (n) such that T (P) = p1, T (Q) = p2 and T (R) = p3.
27
1, . . . , m1
By putting m1 = (m1
2 + ··· + mi
mi
K0(F (n)) ≃ The group generated by { (m1, m2, m3) ∈ Zn+1
n), m2 = (m2
n, it holds that
1, . . . , m2
n+1
={ ((m1, . . . , mn+1), (r1, . . . , rn+1), (s1, . . . , sn+1)) ∈ Zn+1 ⊕ Zn+1 ⊕ Zn+1
Xi=1
Xi=1
Xi=1
si }.
mi =
ri =
n+1
n+1
n), m3 = (m3
1, . . . , m3
n), mi = mi
1 +
+ ⊕ Zn+1
+ ⊕ Zn+1
+ m1 = m2 = m3}
We calculate the embedding of K groups of the finite cores derived from the embedding
from F (n) to F (n+1). The generators of K0(F (n)) are given by paths of projections in F (n), and
they are also paths of projections in F (n+1). The embedding is calculated using the matrix
representation. For the calculation, we present the generators of K0-groups of D(P), D(Q)
and D(R). Let pP
i be a minimal projection of
Hn−i and Π(n)(K(X⊗n))(Q), pR
n+1 be a
minimal projection in Π(n)(K(X⊗n))(P), pQ
n+1 be a a minimal projection in Π(n)(K(X⊗n))(Q)
and pR
i be a minimal projection of In−i for 1 ≤ i ≤ n. Let pP
i be a minimal projection of C(P, T, i + 1), pQ
n+1 be a minimal projection in Π(n)(K(X⊗n))(R). Then it holds
K0(D(P)) ={ m1
K0(D(Q)) ={ m2
K0(D(R)) ={ m3
i ∈ Z} ≃ { (m1
i ∈ Z} ≃ { (m2
i ∈ Z} ≃ { (m3
1 ] + ··· + m1
1 ] + ··· + m2
1 ] + ··· + m3
n+1] m1
n+1] m2
n+1] m3
n+1[pP
n+1[pQ
n+1[pR
1, . . . , m1
1, . . . , m2
1, . . . , m1
1[pP
1[pQ
1[pR
n+1) m1
n+1) m2
n+1) m3
i ∈ Z}
i ∈ Z}
i ∈ Z}.
We denote by Ψn,n+1 the map of embedding from K0(F (n)) to K0(F (n+1)).
Ψn,n+1((0, . . . , 1iP , . . . , 0), (0, . . . , 1iQ , . . . , 0), (0, . . . , 1iR, . . . , 0))
=((0, . . . , 1iP+1, . . . , 0), (0, . . . , 1iR+1, . . . , 0), (0, . . . , 1iQ+1, . . . , 0))
+ ((11, . . . , 0, . . . , 1n+2), (11, . . . , 0, . . . , 1n+2), (11, . . . , 0, . . . , 1n+2)).
Then K0(F (∞)) is the inductive limit group of the above inclusion maps {Ψ(n,n+1)}n=0,1,2,....
We take a basis of K0(F (∞)) ≃ Z3n+1 as follows:
an
i =((0, . . . , 0, 1i, 0, . . . , 0), (0, . . . , 0, 1n+1), (0, 0, . . . , 1n+1))
bn
i =((11, 0, . . . , 0), (0, . . . , 0, 1i, 0, . . . , 0), (0, . . . , 0, 1n+1))
cn
i =((11, 0, . . . , 0), (0, . . . , 0, 1n+1), (0, . . . , 0, 1i, 0, . . . , 0))
i = 1, . . . , n,
i = 1, . . . , n.
i = 1, . . . , n + 1,
We do the following preliminary calculation:
((11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2))
=((0, . . . , 0, 1n+2), (0, . . . , 0, 1n+2), (0, . . . , 0, 1n+2)) + ((11, 0, . . . , 0), (11, 0, . . . , 0), (11, 0, . . . , 0))
=an+1
=an+1
n+2 + ((11, . . . , 0), (11, . . . , 0), (0, . . . , 1n+1)) + ((0, . . . , 0), (0, . . . , 0), (11, . . . ,−1n+1))
n+2 + bn+1
1 + (cn+1
1 + cn+1
1
.
1 − an+1
1
) = −an+1
n+2 + bn+1
1 + an+1
28
We calculate the action of Φ(n,n+1) to each basis:
Φ(n,n+1)(an
i ) =((0, . . . , 0, 1i+1, 0, . . . , 0), (0, . . . , 0, 1n+1), (0, . . . , 0, 1n+1))
+ ((11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2))
i+1 + (−an+1
1 + an+1
1 + cn+1
)
1 + cn+1
1 + an+1
i+1 + an+1
n+2 + bn+1
n+2 + bn+1
1
1
.
=an+1
= − an+1
Φ(n,n+1)(bn
i ) =((0, 11, 0, . . . , 0), (0, . . . , 0, 1i+1, 0, . . . , 0), (0, . . . , 0, in+2))
+ ((11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2))
=((0, 11, 0, . . . , 0), (0, . . . , 0, 1n+2), (0, . . . , 0, in+2))
+ ((0, . . . , 0), (0, . . . , 0, 1i+1, 0, . . . ,−1n+2), (0, . . . , 0))
+ ((11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2))
2 + (bn+1
1 + cn+1
)
1
i+1 − an+1
1
) + (−an+1
n+2 + bn+1
1 + an+1
1 + bn+1
n+2 + bn+1
i+1 + cn+1
.
1
1 + an+1
2 + an+1
=an+1
= − 2an+1
Φ(n,n+1)(cn
i ) =((0, 11, 0, . . . , 0), (0, . . . , 0, 1n+2), (0, . . . , 0, 1i+1, 0, . . . , 0))
+ ((11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2))
=((0, 11, 0, . . . , 0), (0, . . . , 0, 1n+2), (0, . . . , 0, 1n+1))
+ ((0, . . . , 0), (0, . . . , 0), (0, . . . , 0, 1i+1, 0, . . . ,−1n+2))
+ ((11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2), (11, 0, . . . , 0, 1n+2))
2 + (cn+1
) + (−an+1
n+2 + bn+1
1 + an+1
1 + cn+1
1 + cn+1
n+2 + bn+1
1 + cn+1
i+1 .
1 + an+1
2 + an+1
i+1 − an+1
)
1
1
=an+1
= − 2an+1
For the map Ψ(n,n+1), we can define a matrix B(n,n+1) using the above bases.
For n = 3, we have
B(n,n+1) =
To state the form for general n, we prepare some notations. For a ∈ Z, am denotes the row
vector whose entries are all a and am denotes the column vector whose entries are all a. Em
denotes the identity matrix of dimension m, Om denotes the zero matrix of dimension m and
Fm denotes the m dimensional matrix such that (Fm)1,j = 1 and (Fm)i,j = 0 for 2 ≤ i ≤ m
29
−1 −1 −1 −1 −2 −2 −2 −2 −2 −2
1
1
0
0
0
0
1
1
1
1
0
0
0
0
0
0
1
1
0
0
0
0
0
1
0
1
0
1
1
0
0
0
1
0
0
0
0
0
1
1
1
0
0
0
1
0
0
0
1
0
0
1
1
0
0
1
1
0
0
0
0
0
0
2
1
0
0
0
1
0
0
0
1
0
0
1
1
0
0
0
1
0
1
0
1
0
0
1
1
0
0
0
1
1
0
0
1
0
0
1
1
0
1
0
1
0
0
0
1
0
0
1
1
1
0
0
1
0
0
0
.
and 1 ≤ j ≤ m. For general n, we have
B(n,n+1) =
(−1)n −1 (−2)n
Fn
En
1n
1n
1n
1n
En
On
1n
1n
On
On
0n
2
1
0n
1
0n
(−2)n
Fn
1n
1n
On
1n
En
(4)
At last, we consider the K1 group of F (n). The image of the exp map is Z2. We write a
generators of the range of exp map explicitly. Let S be the set obtained by identifying P, Q,
R from the Sierpinski gasket S. A continuous map hPQ from S to T whose winding number
is equal 1 around the circle from P to R, and 0 around the circle from P to S and R to S.
Then [hPQ] − [hPR] is a generator of the image of exponential map. Another generator is
constructed analogously. Then K1(F (n)) is isomorphic to Z∞ and the inclusion map from
K1(F (n)) to K1(F (n+1)) is identity map by the canonical identification. Then K1(F∞) is
isomorphic to K1(S) ≃ Z∞.
Propositoin 26. Let F (∞) be the core of the C∗-algebra associated with the Sierpinski Gasket.
Then we have that
n→∞(cid:18)Z3n+1 B(n,n+1)
→ Z3n+4(cid:19)
K0(F (∞)) ≃ lim
K1(F (∞)) ≃Z∞.
Lemma 27. Each map Ψ(n,n+1) is injective.
Proof. By the expression (4), we can show that the rank of the matrix B(n,n+1) is 3n + 1. We
show it for the case n = 3. By row basic deformation, the matrix B(3,4) is transformd to the
following:
0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 1 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 1
.
The general n case is similar. Then B(n,n+1) is injective as linear map from R3n+1 to R3n+4.
(cid:3)
30
6. Dimension groups
The dimension groups for topological Markov shifts were introduced and studied by Krieger
in [11] and [12] motivated by the K-groups for the AF-subalgebras of the associated Cuntz-
Krieger algebras [2]. The dimension group is a complete invariant for topological Markov
shifts up to shift equivalence. K. Matsumoto [14] studied the dimension groups for the C∗-
algebras OΛ associated with subshifts Λ. He showed that the dual action α of the gauge
action α on the C∗-algebra OΛ induces an automorphism β on the K-group K0(FΛ) of the
fixed point algebra FΛ through Morita equivalence between the fixed point algebra and the
crossed product algebra by the gauge action. A system (K0(FΛ), K0(FΛ)+, β) is called the
dimension group and an important invariant for symbolic dynamical systems.
In the case of self-similar maps without branched points, the same method works. But in
the case of self-similar maps with branched points, it does not work, since the fixed point
algebra and the crossed product algebra by the gauge action are not Morita equivalent in
general. However we shall show that it is possible to introduce a certain canonical endomor-
phism (similar to the automorphism above) on the K-group of the core using isometries in
the one-spectral subspace directly.
Let G be a compact abelian group. Consider a C∗-dynamical system (A, G, α), that is, A is
a C∗-algebra and α is a homomorphism from G to the automorphism group Aut(A) of A. We
assume that A is unital. For a, b ∈ A and g ∈ G, put fa,b(g) = aαg(b). Then fa,b ∈ C(G, A) is
an element in the crossed product A ⋊α G. We sometimes write it asRG fa,b(z)λzdz. Define a
constant function e ∈ C(G, A) by e(g) = 1 for g ∈ G. The e is a projection and we sometimes
write it as
We can also identify fa,b with "rank one operator" aeb on the Hilbert Aα-module A.
There exists an isomorphism of the fixed point algebra Aα onto the hereditary subalgebra
e =ZG
1λzdz ∈ A ⋊α G ,
e(A ⋊α G)e such that a 7→ fa,1 = ae = ea for a ∈ Aα [18].
Definition 28. (Rieffel, see Phillips [15]) Let G be a compact abelian group and (A, G, α) a
C∗-dynamical system. Then the action α is called saturated if the linear span of {fa,b a, b ∈
A} is dense in A ⋊α G. The closed linear span of {fa,b a, b ∈ A} is the ideal generated by
e. Therefore the action α is saturated if e is a full projection in A ⋊α G.
We can associate left AG- right A ⋊α G-module Y as follows: We put Y = A as a linear
space. For a ∈ AG, f ∈ C(G, A) and x ∈ A, we define module actions by
a · x = ax,
x · f =ZG
αg−1(xf (g)) dg.
For x, y ∈ A, we define two inner products by
Aα(xy) =ZG
αg(xy∗) dg,
(xy)A⋊αG = "g 7→ x∗αg(y)".
Then the action α is saturated if and only if AG-A ⋊α G bimodule Y is a Morita equivalence
module.
For example, the gauge action on a Cuntz-Krieger algebra is saturated. Matsumoto [14]
showed that the gauge action on the C∗-algebra OΛ associated with a subshifts Λ is saturated.
31
J. Jeong [5] showed that the gauge action on the graph C∗-algebra associated with a locally
finite directed graph with no sources or sinks is saturated.
The following theorem is a generalization to the case of Cuntz-Pimsner algebras.
Propositoin 29. Let A be a unital C∗-algebra. Let X be an A-A correspondence. We assume
that X is full, the left module action of A on X is injective and there exits x0 ∈ X such that
(x0x0)A = I. If X has a finite basis as a Hilbert right A-module, then the gauge action α of
T on OX is saturated.
Proof. Let B = OX and G = T. We shall consider a hat map of B = OX into OX ⋊α T: For
b ∈ B, a constant function b = (z ∈ T 7→ b) gives the desired element in OX ⋊α T. For a,
b ∈ B, put fa,b(z) = aαz(b). Put e = 1. Then e is a projection. We identify
fa,b(z)λzdz
1λzdz ∈ OX ⋊α T,
b = be and fa,b =ZT
e =ZT
as elements in the crossed product OX ⋊α T. We can also identify fa,b with "rank one
operator" aeb on the Hilbert Oα
X -module OX .
Since OX ⋊α T is the closure of the span by
{ z 7→ f (z)T f ∈ C(T), T ∈ OX },
it is enough to show that for each n ∈ Z, x ∈ X⊗m, y ∈ X⊗k and c ∈ A,
"z ∈ T 7→ znSxS∗y " ∈ Span{ fa,b a, b ∈ OX },
and
"z ∈ T 7→ znc" ∈ Span{ fa,b a, b ∈ OX }.
Firstly consider the case that n > (x − y). Put l = n − (x − y) > 0. Put v =
x0 ⊗ ··· ⊗ x0 ∈ X⊗l. Then S∗v Sv = 1, and it holds
znSxS∗y =S∗v znSvSxS∗y
=S∗v αz(SvSxS∗y )
=fa,b(z),
where a = S∗v ∈ B, b = SvSxS∗y ∈ B. Secondly we consider the case that n − (x − y) = 0.
Then it holds that
znSxS∗y = 1αz(SxS∗y ) = fa,b(z),
where a = 1 ∈ B and b = SxS∗y ∈ B. Thirdly we consider the case that n − (x − y) < 0,
and put l = (x − y) − n > 0. Let { u1, . . . , up } ∈ X be a finite basis. Then
is a finite bases of X⊗n. Hence we have that
{ ui1 ⊗ ··· ⊗ uin (i1, . . . , in) ∈ {1, . . . , p}n }
Sui1 ··· Suin S∗uin ··· S∗ui1
SxS∗y
Sui1 ··· Suin S∗uin ··· S∗iil−1
αz(S∗uil ··· S∗ui1
SxS∗y )
znSxS∗y =zn X(i1,...,in)∈{1,...,p}n
= X(i1,...,in)∈{1,...,p}n
= X(i1,...,in)∈{1,...,p}n
fai1 ,...,in ,bi1,...,in (z),
32
where ai1,...,in = Sui1 ··· Suin S∗uin ··· S∗uil+1
and bi1,...,in = S∗uil ··· S∗ui1
SxS∗y .
(cid:3)
The proof in Proposition 29 does not work if X does not have a finite basis.
Theorem 30. Let A be a unital C∗-algebra. Let X be an A-A correspondence. We assume
that X is full, left module action of A on X is injective and there exits x0 ∈ X such that
(x0x0)A = I. Suppose that X has a finite basis as a Hilbert right A-module. For the gauge
action (OX , T, α) and its dual action (OX ⋊ T, T, α), consider the following diagram:
K0(OX ⋊α T)
α∗−−−−→ K0(OX ⋊α T)
ϕ∗x
ϕ−1
∗
y
β
−−−−→ K0(Oα
X )
K0(Oα
X )
X . Then e is a full projection in OX ⋊α T and ϕ∗ gives an
X ) onto K0(OX ⋊α T). Moreover if we put β := ϕ−1
∗ α∗ϕ∗, then the above
X , any partial isometry S in the one-
X of OX (i.e., αt(S) = eitS ) with P ≤ S∗S, we have β([P ]) = [SP S∗]
Where ϕ(b) = b = be for b ∈ Oα
isomorphism of K0(Oα
diagram is commutative. For each projection P ∈ Oα
spectral subspace O(1)
in the K0-group K0(Oα
Proof. By Proposition 29, e is a full projection in OX ⋊α T, and e commutes with the Oα
and Oα
X ) onto K0(OX ⋊α T).
Since we put β = ϕ−1
X be a projection, S a partial isometry in the one-spectral subspace of OX such
X . Since eP = P e and eSP S∗ = SP S∗e,
X e = e(OX ⋊α T)e. Therefore ϕ∗ gives an isomorphism of K0(Oα
∗ α∗ϕ∗, β gives an automorphism on the K-group K0(Oα
X ).
that P ≤ S∗S. Then SP S∗ is also a projection in Oα
P e is a projection and SP e is a partial isometry. In fact
Let P ∈ Oα
X ) of the core. .
X
(SP e)∗(SP e) = eP S∗SP e = eP,
(SP e)(SP e)∗ = SP eP S∗ = SP eS∗.
Therefore two projections P e and SP eS∗ are von Neumann equivalent in OX ⋊α T. It is
enough to show that
We see that α(Se) = eS. In fact,
α∗ϕ∗([P ]) = [α(P e)] = [SP S∗e].
1λzdz)S =ZT
αz(S)λzdz =ZT
eS = (ZT
zSλzdz = α(Se).
Hence we have α(eS∗) = S∗e. Therefore
α(P e)] = [α(SP eS∗)] = [α(SP )α(eS∗)] = [SP S∗e]
(cid:3)
We note that [SP S∗] does not depend on the choice of such S.
The following is due to Rieffel as in Phillips [15] 7.1.15.
Theorem 31. (Rieffel) Let (A, G, α) be a C∗-dynamical system with a compact abelian
group G. For τ ∈ G, define the spectral subspace Aτ by
Aτ = { a ∈ A αg(a) = τ (g)a}.
Then α is saturated if and only if A∗τ Aτ = Aα for each τ ∈ G.
33
Propositoin 32. Let A be a unital C∗-algebra. Let X be an A-A correspondence. We
assume that X is full, left module action of A on X is injective and there exits x0 ∈ X such
that (x0x0)A = I. Suppose that X does not have a finite basis as a Hilbert right A-module.
Then the gauge action (T,OX , α) is not saturated.
Proof. The spectral subspace O(−1)
is defined by
O(−1)
X = { T ∈ OX αt(T ) = e−itT }.
X
On the contrary, suppose that the gauge action (OX , T, α) is saturated. Then (O(−1)
X . Since S∗x0Sx0 = I, for any T ∈ O(−1)
OT
for sufficiently large n such that
X )∗O(−1)
X =
and ε > 0 there exist a family {Bk}k=1,...,p ⊂ F (n)
X
p
Since I is in OT
sufficiently large n such that
X , for any ε > 0, there exist families {Bi
k′}i′=1,...,s ⊂ F (n) for
k}i=1,...,s, {C i
k′!k < ε
S∗x0C i
kT −
Xk=1
S∗x0Bkk < ε/2
kI −
kI −
S∗x0Bi
k!∗ q
Xk′=1
k∗Sx0S∗x0C i
Bi
k′k < ε.
s
Xi=1 p
Xk=1
Xi=1Xk,k′
s
Therefore I can be approximated by elements of K(X⊗n). This contradicts that XA does not
have a finite basis. Hence the gauge action is not saturated.
(cid:3)
The fixed point algebra and the crossed product algebra by the gauge action are not Morita
equivalent in general. However it is possible to introduce an endomorphism (similar to the
automorphism as above) on the K-group of the core using isometries in the one-spectral
subspace directly.
Propositoin 33. Let A be a unital C∗-algebra. Let X be an A-A correspondence. We assume
that X is full, left module action of A on X is injective and there exits x0 ∈ X such that
(x0x0)A = I. Then there exists an endomorphism β on the K0-group (K0(Oα
X )+)
of the core satisfying the following: For any projection P ∈ Oα
X, any isometry S in the one-
spectral subspace O(1)
X of OX we have β([P ]) = [SP S∗] in the K0-group K0(Oα
X ) of the core.
The endomorphism β does not depend on the choice of such isometry S.
X ), K0(Oα
X gives an *-endomorphism δS on the fixed point algebra Oα
Proof. For any isometry S in the one-spectral subspace O(1)
T ∈ Oα
endomorphism β on the K0-group (K0(Oα
for any projection P ∈ Oα
OX . Then
X of OX , δS(T ) = ST S∗ for
X . Therefore δS induces an
X )+) of the core such that β([P ]) = [SP S∗]
X of
X . Take another isometry W in the one-spectral subspace O(1)
X ), K0(Oα
(W P S∗)∗(W P S∗) = SP W ∗W P S∗ = SP S∗,
(W P S∗)(W P S∗)∗ = W P W ∗.
Therefore SP S∗ and W P W ∗ are von Neumann equivalent and [SP S∗] = [W P W ∗].
(cid:3)
34
Definition 34. In the above situation, we call the system (K0(Oα
X )+, β) the di-
mension group for the bimodule X. Moreover,
let γ = (γ0, . . . , γN−1) be a self-similar
map on a compact set K and X = Xγ is the associated bimodule. Then we also call
the system (K0(Oα
X )+, β) the dimension group for the original self-similar map
γ = (γ0, . . . , γN−1).
X ), K0(Oα
X ), K0(Oα
We shall calculate this endomorphism on the K-group of the core for some cases that the
C∗-correspondence which does not have a finite basis, and show that the endomorphism is
not surjective in general.
Theorem 35. Let (K, γ) be the self-similar map given by the tent map. Then the dimension
Z ∼= Z[t] as an
abstract group and the canonical endomorphism β can be identified with the unilateral shift
on it i.e. the multiplication map by t .The canonical endomorphism β of the dimension group
is not surjective.
group is isomorphic to the countably generated free abelian group Z∞ =`∞n=0
Proof. Let S1 = (1/√ 2)1Cγ . Then the endomorphism β is given by β([P ]) = [S1P S∗1]. For
each projection P in F (n), β([P ]) is expressed as [P ⊗ q] where
q =
1
2(cid:18)1 1
1 1(cid:19) .
We identify the dimension group with the countably generated free abelian group Z∞
generated by cn = ϕ([pn]) in Theorem 24. Since
we see that the endomorphism β on the dimension group can be identified with the unilateral
shift on it.
(cid:3)
β([pn]) = [pn ⊗ q] = [pn+1]
References
[1] G. Castro, C∗-algebras associated with iterated function systems, Contemporary Math.503 (2010), 27-38,
Operator structures and dynamical systems.
[2] J. Cuntz and W. Krieger, A class of C ∗-algebras and topological Markov chains, Invent. Math. 56 (1980),
251-268.
[3] V. Deaconu and F. Shultz, C ∗-algebras associated with interval maps, Trans. Amer. Math. Soc. 359
(2007), 1889-1924.
[4] K. J. Falconer, Fractal Geometry, Wiley, 1997.
[5] J. Jeong, Saturated actions by finite-dimensional Hopf*-algebras on C ∗-algebras, Internat. J. Math. 19
(2008), 125-144.
[6] T. Kajiwara and Y. Watatani, C∗-algebras associated with self-similar maps, J. Oper. Th.56(2006), 225 --
247.
[7] T. Kajiwara and Y. Watatani, Traces on core of the C∗-algebras constructed from self-similar maps,
Ergod. Th. Dyn. Sys. 34(2014), 1964 -- 1989.
[8] T. Kajiwara and Y. Watatani, Ideals of the core of C∗-algebras associated with self-similar maps J. Oper.
Th. 75(2016), 1225 -- 1255.
[9] T. Kajiwara and Y. Watatani, Maximal abelian subalgebras of C ∗-algebras associated with complex dy-
namical systems and self-similar maps, J.Math.Anal. Appl. 455 (2017), 1383-1400.
[10] J. Kigami, Analysis on fractals, Cambridge University Press, 2001
[11] W. Krieger, On dimension for a class of homeomorphism groups, Math. Ann. 252 (1980), 87-95.
[12] W. Krieger, On dimension functions and topological Markov chains, Invent. Math. 56 (1980), 239-250.
[13] K. Matsumoto,On C ∗-algebras associated with subshifts, Internat. J. Math. 8 (1997), 357-374.
[14] K. Matsumoto, K-theory for C∗-algebras associated with subshifts, Math. Scand. 82(1998), 237 -- 255.
35
[15] C. Phillips, Equivariant K-theory and freeness of group actions on C∗-algebra, Springer Lecture notes in
Math. 1274, Springer-Verlag, Berlin, 1987.
[16] M. Pimsner, A class of C∗-algebras generating both Cuntz-Krieger algebras and crossed product by Z, Free
probability theory, AMS, (1997), 189 -- 212.
[17] M. Rordam, F. Larsen F. and N.J. Laustsen, An introduction to K-theory for C∗-algebras, London Math-
ematical Society, Student Texts 49,
[18] J. Rosenberg, Appendix to O. Bratteli's paper on crossed products of UHF algebras, Duke Math.J. 46
(1979), 25-26.
[19] F. Shultz, Dimension groups for inteval maps, New York J. Math. 11 (2005), 477-517.
[20] F. Shultz, Dimension groups for inteval maps II, Ergod. Th. Dynam. Sys. 27 (2017), 1287-1321.
(Tsuyoshi Kajiwara) Department of Environmental and Mathematical Sciences, Okayama Uni-
versity, Tsushima, Okayama, 700-8530, Japan
(Yasuo Watatani) Department of Mathematical Sciences, Kyushu University, Motooka, Fukuoka,
819-0395, Japan
36
|
1704.00290 | 5 | 1704 | 2018-03-02T14:17:28 | Modelling questions for quantum permutations | [
"math.OA",
"math.QA"
] | Given a quantum permutation group $G\subset S_N^+$, with orbits having the same size $K$, we construct a universal matrix model $\pi:C(G)\to M_K(C(X))$, having the property that the images of the standard coordinates $u_{ij}\in C(G)$ are projections of rank $\leq 1$. Our conjecture is that this model is inner faithful under suitable algebraic assumptions, and is in addition stationary under suitable analytic assumptions. We prove this conjecture for the classical groups, and for several key families of group duals. | math.OA | math | MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
TEODOR BANICA AND AMAURY FRESLON
Abstract. Given a quantum permutation group G ⊂ S +
N , with orbits having the same
size K, we construct a universal matrix model π : C(G) → MK(C(X)), having the
property that the images of the standard coordinates uij ∈ C(G) are projections of
rank ≤ 1. Our conjecture is that this model is inner faithful under suitable algebraic
assumptions, and is in addition stationary under suitable analytic assumptions. We
prove this conjecture for the classical groups, and for several key families of group duals.
.
A
O
h
t
a
m
[
5
v
0
9
2
0
0
.
4
0
7
1
:
v
i
X
r
a
Introduction
The present paper is motivated by some questions in quantum algebra. Wang intro-
duced in [20] the free analogue S+
N of the quantum permutation group SN . While many
things are known about S+
N ) is still subject
to investigation. One key open problem, slightly stronger than the Connes embedding
property, is whether C(S+
N , the analytic structure of the algebra C(S+
N ) has an inner faithful matrix model. See [5], [8], [10].
A Grassmannian model approach to this question was proposed in [6]. The idea is that
the magic bases of CN form a real algebraic manifold XN , and the problem is whether
the corresponding representation πN : C(S+
N ) → MN (C(XN )) is inner faithful or not. In
order to solve this question, some methods are available from [5], [21], but their application
would require the construction of a measure on XN . An idea here, proposed in [6], is that
of using the push-forward of the Haar measure on U N
N , via a Sinkhorn type algorithm [17].
But the convergence of the algorithm is not established yet.
A perhaps simpler question, with many degrees of freedom, is that of looking first at the
various quantum subgroups G ⊂ S+
N . As explained in [1], the matrix model construction is
available in this setting, with the model space XG ⊂ XN being obtained by imposing the
abstract Tannakian conditions which define G. However, in the non-transitive case the
model space collapses to the null space. We will fix here this issue with a new construction,
the idea being to allow 0 entries in our magic basis when the orbits of G are non-trivial.
To be more precise, we will assume that G is quasi-transitive, in the sense that its orbits
have the same size K ∈ N, with KN, and we will construct a universal matrix model
π : C(G) → MK(C(X)), having the property that the images of the standard coordinates
uij ∈ C(G) are projections of rank ≤ 1.
2010 Mathematics Subject Classification. 46L54 (81R50).
Key words and phrases. Quantum permutation, Matrix model.
1
2
TEODOR BANICA AND AMAURY FRESLON
One important source of examples when trying to understand properties of compact
quantum groups are duals of discrete groups. This is where our construction is interesting.
Indeed, the only transitive group duals are cyclic groups, while there are plenty of quasi-
transitive examples coming from free products of cyclic groups. We can therefore do
computations and give explicit examples of inner faithful models in this enlarged setting.
Our conjecture is that the quasi-flat model is inner faithful under suitable uniformity
assumptions on G, and is in addition stationary under suitable analytic assumptions on
G. We will discuss this conjecture for the classical groups G ⊂ SN , and we will investigate
N itself,
N . The general case, including that of G = S+
it as well for the group dualsbΓ ⊂ S+
and of other transitive subgroups G ⊂ S+
The paper is organized as follows: 1-2 contain preliminaries on quasi-transitive quantum
groups, in 3-4 we construct the universal models and we formulate the conjectures, in 5-6
we perform some basic work on these conjectures, in the classical group and in the group
dual cases, and in 7-8 we discuss in detail the group dual case.
N , remains an open problem.
Acknowledgments. We would like to thank A. Chirvasitu for useful discussions.
1. Quantum permutations
We are interested in the quantum analogues of the permutation groups G ⊂ SN . In
order to introduce these objects, let us recall that a magic unitary is a square matrix over
a C ∗-algebra, u ∈ MN (A), whose entries are projections (p2 = p∗ = p), summing up to 1
on each row and each column. The following key definition is due to Wang [20]:
Definition 1.1. C(S+
magic unitary matrix u = (uij), with the morphisms given by
N ) is the universal C ∗-algebra generated by the entries of a N × N
∆(uij) =Xk
as comultiplication, counit and antipode.
uik ⊗ ukj
,
ε(uij) = δij
, S(uij) = uji
This algebra satisfies Woronowicz' axioms in [23], [24], and the underlying space S+
N is
therefore a compact quantum group, called quantum permutation group.
Observe that any magic unitary v ∈ MN (A) produces a representation π : C(S+
N ) → A,
given by π(uij) = vij. In particular, we have a representation as follows:
π : C(S+
N ) → C(SN )
:
The corresponding embedding SN ⊂ S+
N ≥ 4, where S+
any S+
N is infinite. Moreover, it is known that we have S+
N with N ≥ 4 has the same fusion semiring as SO3. See [4], [20].
The orbit decomposition theory for the subgroups G ⊂ S+
present here an alternative approach, based on the following simple fact:
N was developed in [7]. We
uij → χ(cid:0)σ ∈ SN(cid:12)(cid:12)σ(j) = i(cid:1)
N is an isomorphism at N = 2, 3, but not at
3 , and that
4 ≃ SO−1
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
3
Proposition 1.2. Given a quantum group G ⊂ S+
uij ∈ C(G), the following defines an equivalence relation on {1, . . . , N}:
N , with standard coordinates denoted
In the classical case, G ⊂ SN , this is the orbit equivalence coming from the action of G.
Proof. We first check the fact that we have indeed an equivalence relation:
i ∼ j when uij 6= 0
(1) i ∼ i follows from ε(uij) = δij, which gives ε(uii) = 1, and so uii 6= 0, for any i.
(2) i ∼ j =⇒ j ∼ i follows from S(uij) = uji, which gives uij 6= 0 =⇒ uji 6= 0.
(3) i ∼ j, j ∼ k =⇒ i ∼ k follows from ∆(uik) =Pj uij ⊗ ujk. Indeed, in this formula,
the right-hand side is a sum of projections, so assuming uij 6= 0, ujk 6= 0 for a certain
index j, we have uij ⊗ ujk > 0, and so ∆(uik) > 0, which gives uik 6= 0, as desired.
In the classical case now, G ⊂ SN , the standard coordinates are the characteristic
functions uij = χ(σ ∈ Gσ(j) = i). Thus the condition uij 6= 0 is equivalent to the
existence of an element σ ∈ G such that σ(j) = i, and this means precisely that i, j must
be in the same orbit under the action of G, as claimed.
(cid:3)
Summarizing, we have a quantum analogue of the orbit decomposition from the classical
case. It is convenient to introduce a few more related objects, as follows:
Definition 1.3. Associated to a quantum group G ⊂ S+
lence relation on {1, . . . , N} given by i ∼ j when uij 6= 0, are as well:
N , producing as above the equiva-
(1) The partition π ∈ P (N) having as blocks the equivalence classes under ∼.
(2) The binary matrix ε ∈ MN (0, 1) given by εij = δuij,0.
Observe that each of the objects ∼, π, ε determines the other two ones. We will often
N come in increasing order, in
assume, without mentioning it, that the orbits of G ⊂ S+
the sense that the corresponding partition is as follows:
π = {1, . . . , K1}, . . . , {K1 + . . . + KM −1 + 1, . . . , K1 + . . . + KM }
Indeed, at least for the questions that we are interested in here, we can always assume
that it is so, simply by conjugating everything by a suitable permutation σ ∈ SN .
In analogy with the classical case, we have as well the following notion:
Definition 1.4. We call G ⊂ S+
N transitive when uij 6= 0 for any i, j. Equivalently:
(1) ∼ must be trivial, i ∼ j for any i, j.
(2) π must be the 1-block partition.
(3) ε must be the all-1 matrix.
Let us discuss now the quantum analogue of the fact that given a subgroup G ⊂ SN ,
with orbits of lenghts K1, . . . , KM , we have an inclusion as follows:
G ⊂ SK1 × . . . × SKM
Given two quantum permutation groups G ⊂ S+
L , with magic corepresenta-
tions denoted u, v, we can consider the algebra A = C(G)∗C(H), together with the magic
K, H ⊂ S+
4
TEODOR BANICA AND AMAURY FRESLON
matrix w = diag(u, v). The pair (A, w) satisfies Woronowicz's axioms, and we therefore
obtain a quantum permutation group, denoted G ∗ H ⊂ S+
K+L. See [19].
With this notion in hand, we have the following result:
Proposition 1.5. Given a quantum group G ⊂ S+
partition π ∈ P (N), having blocks of length K1, . . . , KM , we have an inclusion
N , with associated orbit decomposition
G ⊂ S+
K1
∗ . . . ∗ S+
KM
where the product on the right is constructed with respect to the blocks of π. In the classical
case, G ⊂ SN , we obtain in this way the usual inclusion G ⊂ SK1 × . . . × SKM .
Proof. Since the standard coordinates uij ∈ C(G) satisfy uij = 0 for i 6∼ j, the algebra
C(G) appears as quotient of the following algebra:
C(S+
N ). huij = 0, ∀i 6∼ ji = C(S+
= C(S+
K1
K1) ∗ . . . ∗ C(S+
KM
)
∗ . . . ∗ S+
KM
)
Thus, we have an inclusion of quantum groups, as in the statement. Finally, observe
that the classical version of the quantum group S+
K1
∗ . . . ∗ S+
KM
is given by:
(S+
K1
∗ . . . ∗ S+
KM
)class = (SK1 × . . . × SKM )class
= SK1 × . . . × SKM
Thus in the classical case we obtain G ⊂ SK1 × . . . × SKM , as claimed.
Let us discuss now what happens in the group dual case, where the situation is non-
(cid:3)
trivial. Following the work of Bichon in [7], we have the following result:
Proposition 1.6. Given a decomposition N = K1 + . . . + KM , and a quotient group
ZK1 ∗ . . . ∗ ZKM → Γ, we have an embedding, as follows:
bΓ ⊂ ZK1 ∗ . . . ∗ ZKM ⊂ S+
K1
∗ . . . ∗ S+
KM
⊂ S+
N
Moreover, modulo the action of SN × SN on the magic unitaries, obtained by permuting
the rows and columns, we obtain in this way all the group dual subgroupsbΓ ⊂ S+
Proof. Given a quotient group Γ as in the statement, by composing a number of standard
embeddings and identifications, we obtain indeed an embedding, as follows:
N .
bΓ ⊂ \ZK1 ∗ . . . ∗ ZKM =bZK1 ∗ . . . ∗bZKM ≃ ZK1 ∗ . . . ∗ ZKM
⊂ SK1∗ . . . ∗ SKM ⊂ S+
K1
∗ . . . ∗ S+
KM
K1+...+KM
⊂ S+
Regarding now the last assertion, this basically follows by letting N = K1 + . . . + KM
(cid:3)
be the decomposition coming from the orbit structure ofbΓ ⊂ S+
Let us now consider the case where the decomposition N = K1 +. . .+KM is "minimal",
in the sense that the quotient map ZK1 ∗ . . . ∗ ZKM → Γ is faithful on each ZKi. With
this assumption made, we have:
N . See [7].
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
5
from Propositions 1.5 and 1.6 coincide.
N appears as follows:
N comes from a quotient group ZK1 ∗ . . . ∗ ZKM → Γ
with K1 + . . . + KM = N , such that the quotient map is faithful on each ZKi.
K1
∗ . . . ∗ S+
KM
(1) The associated orbit decomposition is N = K1 + . . . + KM .
Theorem 1.7. Assume that bΓ ⊂ S+
(2) The inclusionsbΓ ⊂ S+
Proof. We recall from Proposition 1.6 that the subgroupbΓ ⊂ S+
(1) By construction of bΓ ⊂ S+
must belong to the same orbit under the action ofbΓ, and we are done.
For more on the group dualsbΓ ⊂ S+
bΓ ⊂ ZK1 ∗ . . . ∗ ZKM ⊂ S+
quantum groups, under the extra assumption K1 = . . . = KM .
(2) This is just an observation, which is clear from (1) above.
∗ . . . ∗ S+
KM
⊂ S+
N
K1
2. Quasi-transitivity
N , the orbit decomposition for this quantum group must
appear via a refinement of N = K1+. . .+KM . On the other hand, since ZK1∗. . .∗ZKM → Γ
is faithful on each ZKi, the elements (K1 + . . . + Ki−1) + 1, . . . , (K1 + . . . + Ki−1) + Ki
N , we refer to [7]. We will come back later to these
(cid:3)
We discuss now an extension of the notion of transitivity, that we call quasi-transitivity.
We will see later on that the universal flat matrix model construction from [1], [6], which
works well in the transitive case, adapts to the quasi-transitive case.
In terms of the objects ∼, π, ε introduced above, we have:
Definition 2.1. A quantum permutation group G ⊂ S+
all its orbits have the same size. Equivalently:
(1) ∼ has equivalence classes of same size.
(2) π has all the blocks of equal length.
(3) ε is block-diagonal with blocks the flat matrix of size K.
N is called quasi-transitive when
As a first example, if G is transitive then it is quasi-transitive. In general now, if we
denote by K ∈ N the common size of the blocks, and by M ∈ N their multiplicity, then
we must have N = KM. We have the following result:
Proposition 2.2. Assuming that G ⊂ S+
G ⊂ S+
N is quasi-transitive, we must have
K ∗ . . . ∗ S+
K
where K ∈ N is the common size of the orbits, and M ∈ N is their number.
Proof. This simply follows from Proposition 1.5 above, because, with the notations there,
in the quasi-transitive case we must have K1 = . . . = KM = K. Observe that in the
classical case, we obtain in this way the usual embedding G ⊂ SK × . . . × SK
(cid:3)
.
M terms
{z
}
M terms
{z
}
6
TEODOR BANICA AND AMAURY FRESLON
Let us discuss now the examples. Assume that G ⊂ S+
K ∗ . . . ∗ S+
K.
N ), C(S+
K), consider the quotient map πi : C(S+
If u, v are the
N ) →
fundamental corepresentations of C(S+
C(S+
K) constructed as follows:
u → diag(1K, . . . , 1K,
, 1K, . . . , 1K)
We can then set C(Gi) = πi(C(G)), and we have the following result:
Proposition 2.3. If Gi is transitive for all i, then G is quasi-transitive.
Proof. We have embeddings as follows:
i−th term
v{z}
G1 × . . . × GM ⊂ G ⊂ S+
K ∗ . . . ∗ S+
K
M terms
{z
}
It follows that the size of any orbit of G is at least K (it contains G1 × . . . × GM ) and
(cid:3)
at most K (it is contained in S+
K). Thus, G is quasi-transitive.
K ∗ . . . ∗ S+
We call the quasi-transitive subgroups appearing as above "of product type". Observe
that there are quasi-transitive groups which are not of product type, as for instance the
group G = S2 ⊂ S2 × S2 ⊂ S4 obtained by using the embedding σ → (σ, σ). Indeed, the
quasi-transitivity is clear, say by letting G act on the vertices of a square. On the other
hand, since we have G1 = G2 = {1}, this group is not of product type.
In general, we can construct examples by using various product operations:
Proposition 2.4. Given transitive subgroups G1, . . . , GM ⊂ S+
tions produce quasi-transitive subgroups G ⊂ S+
K ∗ . . . ∗ S+
K
, of product type:
K, the following construc-
(1) The usual product: G = G1 × . . . × GM .
(2) The dual free product: G = G1 ∗ . . . ∗ GM .
M terms
{z
}
Proof. All these assertions are clear from definitions, because in each case, the quantum
groups Gi ⊂ S+
(cid:3)
K constructed in Proposition 2.3 are those in the statement.
In the group dual case, we have the following result:
Proposition 2.5. The group duals bΓ ⊂ S+
K ∗ . . . ∗ S+
K
M terms
precisely those appearing from intermediate groups of the following type:
ZK ∗ . . . ∗ ZK
→ Γ → ZK × . . . × ZK
which are of product type are
{z
}
{z
}
M terms
M terms
{z
}
Proof. It is clear that any intermediate quotient Γ as in the statement produces a quantum
N which is of product type. Conversely, given a group dual
N , coming from a quotient group Z∗M
permutation group bΓ ⊂ S+
bΓ ⊂ S+
K → Γ, the subgroups Gi ⊂bΓ constructed in
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
7
Proposition 2.3 must be group duals as well, Gi =bΓi, for certain quotient groups Γ → Γi.
Now ifbΓ is of product type,bΓi ⊂ S+
K must be transitive, and hence equal tobZK. We then
conclude that we havedZM
In order to give now some other classes of examples, we use the notion of normality for
K ⊂bΓ, and so Γ → ZM
K , as in the proof of Proposition 2.3. (cid:3)
compact quantum groups, from [11], [21]. This notion is introduced as follows:
Definition 2.6. Given a quantum subgroup H ⊂ G, coming from a quotient map π :
C(G) → C(H), the following are equivalent:
(1) A = {a ∈ C(G)(id ⊗ π)∆(a) = a ⊗ 1} satisfies ∆(A) ⊂ A ⊗ A.
(2) B = {a ∈ C(G)(π ⊗ id)∆(a) = 1 ⊗ a} satisfies ∆(B) ⊂ B ⊗ B.
(3) We have A = B, as subalgebras of C(G).
If these conditions are satisfied, we say that H ⊂ G is a normal subgroup.
As explained in [11], in the classical case we obtain the usual normality notion for the
subgroups. Also, in the group dual case the normality of any subgroup, which must be a
group dual subgroup, is automatic. Now with this notion in hand, we have:
Theorem 2.7. Assuming that G ⊂ S+
follows to be quasi-transitive.
N is transitive, and that H ⊂ G is normal, H ⊂ S+
N
Proof. Consider the quotient map π : C(G) → C(H), as in Definition 2.6, given at the
level of standard coordinates by uij 7→ vij. Consider two orbits O1, O2 of H and set:
xi = Xj∈O1
uij
,
yi = Xj∈O2
uij
These two elements are orthogonal projections in C(G) and they are nonzero, because
they are sums of nonzero projections by transitivity of G. We have:
(id ⊗ π)∆(xi) =Xk Xj∈O1
uik ⊗ vkj = Xk∈O1Xj∈O1
uik ⊗ vkj = Xk∈O1
uik ⊗ 1 = xi ⊗ 1
Thus by normality of H we have (π ⊗ id)∆(xi) = 1 ⊗ xi. On the other hand, assuming
that we have i ∈ O2, we obtain:
(π ⊗ id)∆(xi) =Xk Xj∈O1
vik ⊗ ukj = Xk∈O2
vik ⊗ xk
Multiplying this by vik ⊗ 1 with k ∈ O2 yields vik ⊗ xk = vik ⊗ xi, that is to say xk = xi.
In other words, xi only depends on the orbit of i. The same is of course true for yi.
By using this observation, we can compute the following element:
z = Xk∈O2Xj∈O1
ukj = Xk∈O2
xk = O2xi
8
TEODOR BANICA AND AMAURY FRESLON
On the other hand, by applying the antipode, we have as well:
S(z) = Xk∈O2Xj∈O1
ujk = Xj∈O1
yj = O1yj
We therefore obtain the following formula:
S(xi) =
O1
O2
yj
Now since both xi and yj have norm one, we conclude that the two orbits have the
(cid:3)
same size, and this finishes the proof.
Some additional interesting transitivity questions appear in the graph context. See [9].
3. Matrix models
Given a quantum permutation group G ⊂ S+
N , we will be interested in what follows in
the matrix models of type π : C(G) → MK(C(X)), with X being a compact space. There
are many examples of such models, and the "simplest" ones are as follows:
Definition 3.1. Given a subgroup G ⊂ S+
N , a random matrix model of type
is called quasi-flat when the fibers P x
ij = π(uij)(x) all have rank ≤ 1.
π : C(G) → MK(C(X))
As a first observation, the functions x 7→ rx
ij) are locally constant over X,
so they are constant over the connected components of X. Thus, when X is connected,
our assumption is that we have rx
ij = rij ∈ {0, 1}, for any x ∈ X, and any i, j.
ij = rank(P x
Observe that in the case K = N these questions disappear, because we must have
rx
ij = 1 for any i, j, and any x ∈ X. In this case the model is called flat. See [6].
Proposition 3.2. Assume that we have a quasi-flat model π : C(G) → MK(C(X)),
mapping uij 7→ Pij, and consider the matrix rij = rank(Pij).
(1) r is bistochastic, with sums K.
(2) We have rij ≤ εij, for any i, j.
(3) If G is quasi-transitive, with orbits of size K, then rij = εij for any i, j.
(4) If π is assumed to be flat, then G must be transitive.
Proof. These results are all elementary, the proof being as follows:
(1) This is clear from the fact that each P x = (P x
(2) This simply comes from uij = 0 =⇒ Pij = 0.
(3) The matrices r = (rij) and ε = (εij) are both bistochastic, with sums K, and they
ij) is bistochastic, with sums 1.
satisfy rij ≤ εij, for any i, j. Thus, these matrices must be equal, as stated.
(4) This is clear, because rank(Pij) = 1 implies uij 6= 0, for any i, j.
(cid:3)
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
9
In order to construct now universal quasi-flat models, we use the following standard
result from [1], which is a reformulation of Woronowicz's Tannakian duality [24]:
Proposition 3.3. Given an inclusion G ⊂ S+
resentations denoted w 7→ u, we have the following formula:
N , with the corresponding fundamental corep-
C(G) = C(S+
N ).(cid:16)T ∈ Hom(w⊗k, w⊗l), ∀k, l ∈ N, ∀T ∈ Hom(u⊗k, u⊗l)(cid:17)
with the Hom-spaces at left being taken in a formal sense.
Proof. We recall that for a Hopf algebra corepresentation v = (vij), the intertwining
condition T ∈ Hom(v⊗k, v⊗l) means by definition that we have T v⊗k = v⊗lT , the tensor
powers of v = (vij) being the corepresentations v⊗r = (vi1...ir,j1...jr).
We can formally use these notions for any square matrix over any C ∗-algebra, and
in particular, for the fundamental corepresentation of C(S+
N ). Thus, the collection of
relations T ∈ Hom(w⊗k, w⊗l), one for each choice of an intertwiner T ∈ Hom(u⊗k, u⊗l),
produce an ideal of C(S+
N ), and the algebra in the statement is well-defined.
This latter algebra is isomorphic to C(G), due to Woronowicz's Tannakian results in
(cid:3)
[24]. For a short, recent proof here, using basic Hopf algebra theory, see [14].
Now back to our modelling questions, it is convenient to identify the rank one projec-
tions in MN (C) with the elements of the complex projective space P N −1
C
.
We first have the following observation, which goes back to [6]:
Proposition 3.4. The algebra C(S+
N ) has a universal flat model, given by
πN : C(S+
N ) → MN (C(XN ))
,
πN (uij) = [P 7→ Pij]
where XN is the set of matrices P ∈ MN (P N −1
Proof. This is clear from definitions, because any flat model C(S+
the magic corepresentation u = (uij) into a matrix P = (Pij) belonging to XN .
) which are bistochastic with sums 1.
N ) → MN (C) must map
(cid:3)
C
Regarding now the general quasi-transitive case, we have here:
Theorem 3.5. Given a quasi-transitive subgroup G ⊂ S+
a universal quasi-flat model π : C(G) → MK(C(X)), constructed as follows:
N , with orbits of size K, we have
(1) For G = S+
K ∗ . . . ∗ S+
K
with N = KM , the model space is XN,K = XK × . . . × XK
,
M terms
{z
}
M terms
ij) = [(P 1, . . . , P M ) 7→ P r
ij].
{z
}
and with u = diag(u1, . . . , uM ) the map is πN,K(ur
(2) In general, the model space is the submanifold XG ⊂ XN,K obtained via the Tan-
nakian relations defining G.
Proof. This result is known since [1], [6] in the flat case, the idea being to use Proposition
3.3 and Proposition 3.4. In general, the proof is similar:
(1) This follows from Proposition 3.4, by using Proposition 3.2 (3) above, which tells
us that the 0 entries of the model must appear exactly where u = (uij) has 0 entries.
10
TEODOR BANICA AND AMAURY FRESLON
(2) Assume that G ⊂ S+
K ∗ . . . ∗ S+
inclusion G ⊂ S+
N is quasi-transitive, with orbits of size K. We have then an
, and in order to construct the universal quasi-flat model for
K
C(G), we need a universal solution to the following factorization problem:
M terms
{z
}
C(S+
K ∗ . . . ∗ S+
K
) → MK(C(XN,K))
M terms
{z
↓
}
↓
But, the solution to this latter question is given by the following construction, with the
C(G)
→ MK(C(XG))
Hom-spaces at left being taken as usual in a formal sense:
C(XG) = C(XN,K).(cid:16)T ∈ Hom(P ⊗k, P ⊗l), ∀k, l ∈ N, ∀T ∈ Hom(u⊗k, u⊗l)(cid:17)
With this result in hand, the Gelfand spectrum of the algebra on the left is then an
(cid:3)
algebraic submanifold XG ⊂ XN,K, having the desired universality property.
Observe that talking about quasi-flat models for quantum groups which are not nec-
essarily quasi-transitive perfectly makes sense. The universal model spaces can be con-
structed as above, and this was in fact already discussed in [1], but no one guarantees
that in the non-quasi-transitive case, the model spaces are non-empty. So, we prefer to
restrict the attention to the quasi-transitive case, and state Theorem 3.5 as it is.
4. Inner faithfulness
We formulate in what follows a number of conjectures. We first review the notions of
inner faithfulness and stationarity, from [1], [2], [5]. Following [2], we first have:
Definition 4.1. Let π : C(G) → MK(C(X)) be a matrix model.
(1) The Hopf image of π is the smallest quotient Hopf C ∗-algebra C(G) → C(H)
producing a factorization of type π : C(G) → C(H) → MK(C(X)).
(2) When the inclusion H ⊂ G is an isomorphism, i.e. when there is no non-trivial
factorization as above, we say that π is inner faithful.
Observe that when G =bΓ is a group dual, π must come from a group representation
ρ : Γ → C(X, UK), and the above factorization is the one obtained by taking the image,
ρ : Γ → Γ′ ⊂ C(X, UK). Thus π is inner faithful when Γ ⊂ C(X, UK).
Also, given a compact group G, and elements g1, . . . , gK ∈ G, we have a representation
π : C(G) → CK, given by f → (f (g1), . . . , f (gK)). The minimal factorization of π is then
via C(G′), with G′ = < g1, . . . , gK >, and π is inner faithful when G = G′.
In practice, X is often a compact Lie group, or a compact homogeneous space, or a
more general compact probability space. And here, we have the following result:
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
11
Proposition 4.2. Given an inner faithful model π : C(G) → MK(C(X)), with X being
assumed to be a compact probability space, we have
ZG
= lim
k→∞
1
k
kXr=1Z r
G
whereR r
G = (ϕ ◦ π)∗r, with ϕ = tr ⊗RX being the random matrix trace.
Proof. This was proved in [5] in the case X = {·}, using idempotent state theory from
[12]. The general case was recently established in [22].
(cid:3)
The above result can be used as a criterion for detecting the inner faithfulness. To be
more precise, π is inner faithful precisely when the above formula holds. See [5].
Following [1], we call a matrix model stationary when the Ces`aro limiting convergence
in Proposition 4.2 is stationary. In other words, we have the following definition:
Definition 4.3. A matrix model π : C(G) → MK(C(X)), with X assumed to be a
compact probability space, is called stationary when:
ZG
=(cid:18)tr ⊗ZX(cid:19) π
ZG
In the general case, where X is only assumed to be a compact space, π will be called
stationary if it is stationary with respect to some probability measure on X.
There are many interesting examples of such models, see [1]. However, the stationarity
condition is a very strong assumption, and we have the following result, from [1]:
Proposition 4.4. Let π : C(G) → MK(C(X)) be a stationary model. Then,
(1) π is faithful.
(2) C(G) is a type I C*-algebra, hence the discrete dual Γ = bG is amenable.
Proof. We use the basic theory of amenability for discrete quantum groups, from [15].
Assuming that π is stationary, for any x ∈ C(G) we have:
ZG
x = 0 =⇒ (cid:18)tr ⊗ZX(cid:19) π(x) = 0
In particular, with x = yy∗, and by using the fact thatRX is by definition faithful, X
being a compact probability space, we obtain that for any y ∈ C(G) we have:
yy∗ = 0 =⇒ π(yy∗) = 0 =⇒ π(y) = 0
Now since the elements satisfying RG yy∗ = 0 are precisely those in the kernel of the
quotient map λ : C(G) → C(G)red, we obtain a factorization of π, as follows:
π : C(G) → C(G)red → MK(C(X))
12
TEODOR BANICA AND AMAURY FRESLON
Our claim now is that the map on the right, say ρ, is an inclusion. Indeed, let x ∈ ker(ρ),
and let us pick a lift y ∈ C(G) of this element x ∈ C(G)red. We have then:
ρ(x) = 0 =⇒ π(y) = 0 =⇒ π(yy∗) = 0 =⇒ ZG
yy∗ = 0 =⇒ λ(y) = 0 =⇒ x = 0
Thus we have an inclusion C(G)red ⊂ MK(C(X)), and so π factorizes as follows:
π : C(G) → C(G)red ⊂ MK(C(X))
Now since C(G)red must be of type I, and therefore nuclear, the quantum group G must
(cid:3)
be co-amenable, and so π must be faithful, and we are done with both (1,2).
We refer to [1], [2], [5], [10], [12], [22] for more theory and examples, of algebraic and
analytic nature, regarding the notions of inner faithfulness and stationarity.
We recall from [23] that any finitely generated group Γ =< g1, . . . , gN > produces
N , with fundamental corepresentation uij = δijgi, and that
N all appear in this way, modulo a conjugation of the
fundamental corepresentation u = (uij) by a unitary U ∈ UN . Here is now a first result
about stationarity, which is essentially a reformulation of Thoma's theorem [18]:
a closed subgroup bΓ ⊂ U +
the group dual subgroups bΓ ⊂ U +
Theorem 4.5 (Thoma). Given a group dual G =bΓ ⊂ U +
(1) C(G) is of type I.
(2) C(G) has a stationary model.
(3) C(G) has a stationary model, over an homogeneous space.
(4) Γ is virtually abelian.
N , the following are equivalent:
Proof. Here (1) =⇒ (4) is Thoma's theorem [18] and (3) =⇒ (2) =⇒ (1) are trivial
implications. We therefore only have to prove (4) =⇒ (3).
Let Λ < Γ be an abelian subgroup of finite index and let K = [Λ : Γ]. We define a
matrix model π : C ∗(Γ) → MK(C(bΛ)) by:
π(γ)(χ) = IndΓ
Λ(χ)(γ)
To see that this model is faithful, take γ ∈ Γ and recall that the character ψ of IndΓ
Λ(χ)
is given by:
Here T r is the usual (non-normalized) trace on MK(C). Thus:
ψ(γ) = T r(IndΓ
δx−1γx∈Λχ(x−1γx)
Λ(χ)) = Xx∈Γ/Λ
K Xx∈Γ/Λ
1
δx−1γx∈ΛZ bΛ
(cid:18)tr ⊗Z bΛ(cid:19) π(γ) =
χ(x−1γx)dχ
Since the integral over all characters is the indicator function of the trivial element, the
(cid:3)
expression above equals δγ,e and the model is stationary.
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
13
Let us formulate now our main two conjectures, regarding the universal quasi-flat mod-
els for the quantum permutation groups. We first have:
Conjecture 4.6. Assuming that G ⊂ S+
N is quasi-transitive, with orbits of size K, and
that Γ = bG satisfies a suitable "virtual abelianity" condition, we have:
(1) The universal quasi-flat model space XG is an homogeneous space.
(2) The corresponding model π : C(G) → MK(C(XG)) is stationary.
The evidence here comes from Thoma's theorem, in its conjectural stronger form pre-
sented above, as well from a number of explicit verifications, to be performed below, and
notably from a verification in the case where G is classical.
We do not know yet what the "virtual abelianity" condition should mean. When Γ
is classical, as in Thoma's theorem, this condition states that we must have an abelian
subgroup Λ < Γ such that F = Γ/Λ is finite.
Regarding now the notion of inner faithfulness, we have here:
Conjecture 4.7. Assuming that G ⊂ S+
suitable "uniformity" condition, the universal quasi-flat model is inner faithful.
N is quasi-transitive, and satisfies in addition a
We do not know yet what the precise "uniformity" condition should be. We believe
that all this is related to the notion of easiness [13], [16], and this will be confirmed by
some of the verifications performed below, but in general, we have no results.
Finally, let us mention that in the transitive case, the very first question here concerns
N itself, and the problem here is difficult, and open. Indeed, assuming that the
N would follow to be inner linear (in a parametric sense) and we would
N ) has the Connes embedding property. Thus, we will have
G = S+
conjecture holds, S+
therefore obtain that L∞(S+
here a solution to an old open problem. For some comments here, see [5], [6], [8].
5. The classical case
In this section we discuss the classical case, G ⊂ SN . Our question is as follows:
assuming that G is quasi-transitive, with orbits of size K, when do we have a stationary
model π : C(G) → MK(C(X)), for some compact probability space X?
We will use the following notion:
Definition 5.1. A "sparse Latin square" is a square matrix L ∈ MN (∗, 1, . . . , K) having
the property that each of its rows and columns consists of a permutation of the numbers
1, . . . , K, completed with ∗ entries.
In the case K = N, where there are no ∗ symbols, we recover the usual Latin squares.
In general, however, the combinatorics of these matrices seems to be more complicated
14
TEODOR BANICA AND AMAURY FRESLON
than that of the usual Latin squares. Here are a few examples of such matrices:
1 2 ∗
2 ∗ 1
∗ 1 2 ,
1 2 ∗ ∗
2 ∗ 1 ∗
∗ 1 ∗ 2
∗ ∗ 2 1
,
1 2 ∗ ∗
2 1 ∗ ∗
∗ ∗ 1 2
∗ ∗ 2 1
,
With this notion in hand, the result that we need is as follows:
1 2 ∗ ∗ ∗
2 ∗ 1 ∗ ∗
∗ 1 2 ∗ ∗
∗ ∗ ∗ 1 2
∗ ∗ ∗ 2 1
Proposition 5.2. The quasi-flat representations π : C(SN ) → MK(C) appear as
uij 7→ PLij
where P1, . . . , PK ∈ MK(C) are rank 1 projections, summing up to 1, and where L ∈
MN (∗, 1, . . . , K) is a sparse Latin square, with the convention P∗ = 0.
Proof. Assuming that π : C(SN ) → MK(C) is quasi-flat, the elements Pij = π(uij) are
projections of rank ≤ 1, which pairwise commute, and form a magic unitary.
Let P1, . . . , PK ∈ MK(C) be the rank one projections appearing in the first row of
P = (Pij). Since these projections form a partition of unity with rank one projections, any
rank one projection Q ∈ MK(C) commuting with all of them satisfies Q ∈ {P1, . . . , PK}.
In particular we have Pij ∈ {P1, . . . , PK} for any i, j such that Pij 6= 0. Thus we can write
uij 7→ PLij , for a certain matrix L ∈ MN (∗, 1, . . . , K), with the convention P∗ = 0.
In order to finish, the remark is that uij 7→ PLij defines a representation π : C(SN ) →
MK(C) precisely when the matrix P = (PLij )ij is magic. But this condition tells us
precisely that L must be a sparse Latin square, in the sense of Definition 5.1.
(cid:3)
Our task now is to compute the associated Hopf image. We have here:
Proposition 5.3. Given a sparse Latin square L ∈ MN (∗, 1, . . . , K), consider the per-
mutations σ1, . . . , σK ∈ SN given by:
σx(j) = i ⇐⇒ Lij = x
The Hopf image associated to a representation π : C(SN ) → MK(C), uij 7→ PLij as above
is then the algebra C(GL), where GL =< σ1, . . . , σK >⊂ SN .
Proof. We use a method from [3]. The image of π being generated by P1, . . . , PK, we have
an isomorphism of algebras α : Im(π) ≃ C(1, . . . , K) given by Pi 7→ δi. Consider the
following diagram:
C(SN )
π
Im(π)
MK(C)
▼
▼
▼
▼
▼
▼
▼
ϕ
▼
▼
▼
α
▼
&▼
C(1, . . . , K)
/
/
&
/
/
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
15
Here the map on the right is the canonical inclusion and ϕ = απ. Since the Hopf image
of π coincides with the one of ϕ, it is enough to compute the latter. We know that ϕ
is given by ϕ(uij) = δLij , with the convention δ∗ = 0. By Gelfand duality, ϕ must come
from a certain map σ : {1, . . . , K} → SN , via the transposition formula ϕ(f )(x) = f (σx).
With the choice f = uij, we obtain:
δLij (x) = uij(σx)
Now observe that these two quantities are by definition given by:
δLij (x) =(1
0
if Lij = x
otherwise
,
uij(σx) =(1 if σx(j) = i
0 otherwise
We conclude that σx is the permutation in the statement. Summarizing, we have shown
that ϕ comes by transposing the map x → σx, with σx being as in the statement. By
using now the general theory in [2], the Hopf image of ϕ is the algebra C(GL), with
GL =< σ1, . . . , σK >, and this finishes the proof.
(cid:3)
Let us discuss the construction of the universal model space, and then Conjecture 4.6,
in the classical case. We agree to identify the rank one projections in MK(C) with the
corresponding elements of the projective space P K−1
.
C
With these conventions, we first have the following result:
Proposition 5.4. Assuming that G ⊂ SN is quasi-transitive, with orbits of size K, the
universal quasi-flat model space for G is given by XG = EK × LG
N,K, where:
EK =n(P1, . . . , PK) ∈ (P K−1
N,K =nL ∈ MN (∗, 1, . . . , K) sparse Latin square(cid:12)(cid:12)(cid:12)GL ⊂ Go
)K(cid:12)(cid:12)(cid:12)Pi ⊥ Pj, ∀i 6= jo
C
LG
In particular, XG has a canonical probability measure, obtained as the homogeneous space
measure on EK times the the normalized counting measure on LG
N,K.
Proof. The first assertion follows by combining Proposition 5.2 and Proposition 5.3 above,
and the second assertion is clear from the definitions.
(cid:3)
Note that the model above is empty if there is no sparse Latin square L such that
GL ⊂ G. The existence of such a sparse Latin square is a strong condition on G, and here
is an intrinsic characterization of the groups satisfying this condition:
Proposition 5.5. Let G ⊂ SN be a classical group. The following are equivalent:
N,K 6= ∅.
(1) LG
(2) There exist K elements σ1, . . . , σK ∈ G such that σ1(i), . . . , σK(i) are pairwise
distinct for all 1 ≤ i ≤ N .
16
TEODOR BANICA AND AMAURY FRESLON
Proof. (1) =⇒ (2) If σ1, . . . , σK are the permutations associated to L, then σx(i) = σy(i)
means by definition that x = Liσx(i) = Liσy(i) = y.
(2) =⇒ (1) Consider such permutations σ1, . . . , σK.
If i is fixed then for each j
there is at most one index x such that σx(i) = j. We set Lij = x in that case and
Lij = ∗ otherwise. Then, L is a sparse Latin square and the associated permutations are
1 , . . . , σ−1
σ−1
K , which belong to G.
(cid:3)
It turns out that as soon as LG
N,K 6= ∅, the universal quasi-flat model is stationary. To
prove this, let us first give another basic observation:
Proposition 5.6. Given G ⊂ SN , we have an action G y LG
N,K, given by
(Lτ )ij = Lτ −1(i)j
Proof. Given a sparse Latin square L ∈ MN (∗, 1, . . . , K), the matrix Lτ ∈ MN (∗, 1, . . . , K)
constructed in the statement is a sparse Latin square too and τ → (L 7→ Lτ ) is a group
morphism, so it remains to check that GL ⊂ G implies GLτ ⊂ G.
To do this, let us write GL =< σ1, . . . , σK > and GLτ =< σ′
1, . . . , σ′
K >. Then:
x(j) = i ⇐⇒ (Lτ )ij = x ⇐⇒ Lτ −1(i)j = x
σ′
⇐⇒ σx(j) = τ −1(i) ⇐⇒ τ σx(j) = i
Thus, σ′
x = τ σx and:
GL ⊂ G ⇐⇒ σ1, . . . , σK ∈ G
⇐⇒ τ σ1, . . . , τ σK ∈ G
⇐⇒ σ′
1, . . . , σ′
⇐⇒ GLτ ⊂ G
K ∈ G
Thus GL ⊂ G implies GLτ ⊂ G, and this finishes the proof.
(cid:3)
We can now verify Conjecture 4.6 in the classical group case:
Theorem 5.7. If the space LG
transitive subgroup G ⊂ SN ,
N,K is not empty, then the universal flat model for a quasi-
π : C(G) → MK(C(XG))
is stationary with respect to ν ⊗ m where m is the normalized counting measure on LG
and ν is any probability measure on EK.
Proof. Write MK(C(XG)) = MK(C) ⊗ C(EK) ⊗ C(LG
following map:
N,K) and consider, for P ∈ EK, the
N,K
πP = (id ⊗ evP ⊗ id) : C(G) → MK(C(LG
N,K))
Recall that for τ ∈ G and f ∈ C(G), τ.f denotes the map h 7→ f (τ −1h). This
corresponds to the regular action of G on itself. Moreover:
πP (uij)(τ −1L) = πP (uτ (i)j) = πP (τ.uij)
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
17
Since the normalized counting measure m on LG
again m for the integration with respect to m on C(LG
state on C(G), i.e.
constant function equal toRG and the result follows.
it is RG. Summarizing, we have proved that (tr ⊗ id ⊗ m)π is the
N,K is G-invariant, it follows (writing
N,K)) that (tr⊗m)πP is a G-invariant
(cid:3)
6. Group duals
In what follows we discuss universal quasi-flat models in the group dual case. Let us
start with Γ = ZN , where the class of 1 will be denoted by g, and called the canonical
generator. We have:
Proposition 6.1. We have an isomorphism of Hopf algebras
C ∗(ZN ) = C(S+
N ).Duij = ukl(cid:12)(cid:12)(cid:12)j − i = l − k (mod N)E
j=1 wj−iuij, for any i, where w = e2πi/N .
Proof. The quotient algebra A in the statement being generated by the entries of the first
which is such that g =PN
row of u = (uij), it is commutative. If we identify the elements of bZN with the powers
isomorphism between A and the algebra C ∗(ZN ) = C(bZN ). Moreover:
of w, and the elements of ZN with the functions w 7→ wk, then π(uij) = δwj−i defines an
δwk−i ⊗ δwj−k = ∆bZK
(δwj−i)
(π ⊗ π)∆(uij) =Xk
Thus we have an isomorphism of Hopf algebras, as stated. The formula for g in the
(cid:3)
statement then follows from the formula for π.
This translates into a convenient description of the universal flat matrix model.
Proposition 6.2. The universal flat model space for C ∗(ZN ) is the space
EN =n(P1, . . . , PN ) ∈ (P N −1
with the model map given by π(g)(P1, . . . , PN ) =PN
C
)N(cid:12)(cid:12)(cid:12)Pi ⊥ Pj, ∀i 6= jo
j=1 wjPj.
Proof. According to Proposition 6.1 above, π(uij)(P1, . . . , PN ) = Pj−i is a flat matrix
model map, which gives the formula in the statement for π(g).
Let X be a flat matrix model space for C ∗(ZN ), with model map πX. For each x ∈ X
the matrix πX (g)(x) is circulant, so this matrix is completely determined by its first row,
which is an element of EN . Let us set P x
N j, and let P x be the circulant matrix with
first row (P x
N ). The map Φ : X → EN given by Φ(x) = P x being a continuous
embedding, we get a surjective ∗-homomorphism Ψ : MN (C(EN )) → MN (C(X)) through
which πX factors by construction. Thus, we obtain the universality of EN , as claimed. (cid:3)
1 , . . . , P x
j = P x
Regarding now the general group dual case, we have here the following result:
18
TEODOR BANICA AND AMAURY FRESLON
Theorem 6.3. Given a quotient group Z∗M
C ∗(Γ) → MK(C(YΓ)) appears as follows:
K → Γ, the universal quasi-flat model π :
(1) For Γ = Z∗M
K , the model space is YΓ = YN,K = EK × . . . × EK
, and the model map
M terms
{z
}
is given by the formula in Proposition 6.2, on each of the components.
(2) In general, the model space is the subspace YΓ ⊂ YN,K consisting of the elements
x such that π(γ)(x) = Id for any γ ∈ ker(Z∗M
K → Γ).
Proof. The first assertion is clear from Proposition 6.2. Regarding now the second asser-
tion, let us go back to Theorem 3.5 (2) above. The statement there tells us that the model
via the Tannakian conditions
space forbΓ appears from the model space for S+
defining bΓ. But these Tannakian conditions, when expressed in terms of the generators
K are precisely the relations γ = 1 with γ ∈ ker(Z∗M
K → Γ), as stated.
{z
K ∗ . . . ∗ S+
K
of Z∗M
M terms
}
(cid:3)
7. Maximal tori
We will now prove Conjecture 4.7 for several families of duals of discrete groups. The
proofs will be more convenient to write by using a lift of the quasi-flat models in the
following sense:
Proposition 7.1. The affine lift of the universal quasi-flat model for C ∗(Z∗M
C ∗(Z∗M
K )) given on the canonical generator gi of the i-th factor by
K ) → MK(C(U M
K ) is π :
π(gi)(U 1, . . . , U M ) =Xj
wjPU i
j
where U i
j is the j-th column of U i and Pξ denotes the orthogonal projection onto Cξ.
Proof. There is indeed a canonical quotient map UK → EK, obtained by parametrizing
the orthonormal bases of CK by the unitary group UK, and this gives the result.
(cid:3)
According to the results of [7] explained in section 1 above, the maximal group dual
N , which can be regarded as being "maximal tori", are the free products
of type ZK1 ∗ . . . ∗ ZKM with K1 + . . . + KM = N. In the quasi-transitive case, where
K1 = . . . = KM = K with KN, we have the following result:
subgroupsbΓ ⊂ S+
Theorem 7.2. The universal quasi-flat model for Z∗M
K is inner faithful.
Proof. It is enough to prove that the affine lift of the model is inner faithful. Let us
consider a reduced word γ ∈ Z∗M
in , with it 6= it+1, and with
K , and write it as γ = gk1
i1 . . . gkn
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
19
1 ≤ kt ≤ K − 1. Then:
π(γ)(U 1, . . . , U M ) =
=
KXj1...jn=1
KXj1...jn=1
wk1j1PU
i1
j1
. . . wknjnPU in
jn
w<k,j>PU
. . . PU in
jn
i1
j1
Our aim is to prove that there is at least one tuple (U 1, . . . , U M ) for which the matrix
above is not the identity. Recall the following formula valid for any vectors in a Hilbert
space ξ1, . . . , ξl ∈ H, with the scalar product being linear on the left:
Pξ1 . . . Pξl(x) =< x, ξl >< ξl, ξl−1 > . . . . . . < ξ2, ξ1 > ξ1
To compute the trace of this operator, one can consider any orthonormal basis contain-
ing ξl, yielding < ξ1, ξl >< ξl, ξl−1 > . . . . . . < ξ2, ξ1 >. Applying this to π(γ) and using
the equality < Ui, Vj >=Pl Uki ¯Vkj = (V ∗U)ji, we get:
tr ◦ π(γ) =
w<k,j> < U i1
j1 , U in
jn >< U in
jn , U in−1
jn−1 > . . . < U i2
j2 , U i1
j1 >
1
K
KXj1...jn=1
KXj1...jn=1
=
1
K
Xj1
w<k,j>(U in∗U i1)jnj1(U in−1∗U in)jn−1jn . . . (U i1∗U i2)j1j2
Denoting by W the diagonal matrix given by Wij = δijwi, we have:
wk1j1Ujnj1U ∗
Ujnj1W k1
j1lU ∗
lj2 = (UW k1U ∗)jnj2
j1j2 =Xj1l
Applying this n times in the above formula for tr ◦ π(γ) yields:
tr ◦ π(γ) = tr(cid:0)U in∗U i1W k1U i1∗U i2W k2 . . . W kn−1U in−1∗U inW kn(cid:1)
= tr(cid:0)U i1W k1U i1∗ . . . U inW knU in∗(cid:1)
Assume now that π(γ)(U 1, . . . , U M ) = Id for all tuples of unitary matrices. The trace
of a unitary matrix can only be equal to 1 if it is the identity, hence:
In other words, the following noncommutative polynomial vanishes on U M
K :
U ipW kpU ip∗ = Id
P =
X ipW kpX ip∗ − 1
nYp=1
nYp=1
20
TEODOR BANICA AND AMAURY FRESLON
But this is impossible if kt 6= 0 mod K for all t, hence π(γ) is not always the identity
(cid:3)
and π is inner faithful.
The above result can be extended by allowing arbitrary direct products as components
of the free product. More precisely, assume that M = M1 + . . . + Mn. Then, the free
product ZM1
K which is still quasi-transitive with orbits of
size K and we prove that its universal quasi-flat matrix model is still inner faithful.
K is a quotient of Z∗M
K ∗ . . . ∗ ZMn
Proposition 7.3. The universal quasi-flat model for Γ = ZM1
K is inner faithful.
Proof. We first need a picture of the affine lift of the model for the direct product ZM
K .
Note that if two complete families of rank one orthogonal projections commute, then they
are a permutation of one another. We may therefore consider the space UK × SM
K as the
affine lift of our model. If g1, . . . , gM are the canonical generators of the direct product,
their action is then given by:
K ∗ . . .∗ ZMn
π(gi)(U, σ1, . . . , σM ) =
wjPU
=
σ
−1
i
(j)
KXj=1
wσi(j)PUj
KXj=1
This yields the following formula for a general element:
π(gk1
1 . . . gkM
M )(U, σ1, . . . , σM ) =
wk1σ1(j)+...+kM σM (j)PUj
KXj=1
To derive a contradiction, we sum the above equation over all permutations, getting:
1
(K!)M Xσ1,...,σMi
∈SK
wk1(t)σ1(i)+...+kMt (t)σMt (i) =
1
(K!)M
MtYs=1 Xσs∈SK
wks(t)σs(i)!
Let g1(i), . . . , gMi(i) be the generators of ZMi
K , and let consider a reduced word:
(i1)kMi1
γ =(cid:16)g1(i1)k1(1) . . . gMi1
(1)(cid:17) . . .(cid:16)g1(in)k1(n) . . . gMin (in)kMin
introduction of the matrices W . Here we have to replace W kt by QMt
The computation of tr ◦ π(γ) is similar to the one in the proof of Theorem 7.2 until the
, where
Mt)1≤t≤n is the element to which we are applying π(γ) and (Wσ)ij = δijwσ(i).
Assuming that π(γ) = 1, we can apply the same strategy as before: we have a polyno-
mial which must vanish on all tuples of unitary matrices and this is impossible unless all
the matrices appearing in the polynomial are the identity. We therefore get the condition
(n)(cid:17)
s=1 W ks(t)
1, . . . , σt
(U t, σt
σt
s
= Id for all t, which translates into:
s=1 W ks(t)
σt
s
QMt
wσt
s(i)ks(t) = 1
MtYs=1
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
21
For any i′, there are (K − 1)! permutations σ such that σs(i) = i′. This leads to:
Xσs∈SK
Gathering everything yields:
wks(t)σs(i) =
(K − 1)!wks(t)i′
KXi′=1
= K!δks(t),0
Thus ks(t) = 0 for all t and all s, a contradiction.
δks(t),0 = 1
MiYs=1
(cid:3)
8. Further results
In this section we give two other extensions of Theorem 7.2. Before that, let us give an
example showing that the universal model is not always inner faithful for group duals:
Proposition 8.1. The group Γ = (ZK ∗ ZK) × ZK has no inner faithful quasi-flat model.
Proof. Let π : C ∗(Γ) → MK(C(X)) be a matrix model. For all x, π(g1)(x) and π(g3)(x)
are commuting diagonalizable operators with all eigenvalues of multiplicity one. The same
holds for π(g2)(x) and π(g3)(x) so that π(g1)(x) and π(g2)(x) must commute. Since this
holds for any x ∈ X, we conclude that π factors through the quotient Z3
K, so that the
model is not inner faithful.
(cid:3)
We extend now Theorem 7.2 by allowing the free product to be amalgamated over a
subgroup isomorphic to ZL. Assume that we have N = KM as before, and assume in
addition that we have K = LR. We can then consider the following quotient:
Γ = Z∗M
i = gR
K .DgR
j(cid:12)(cid:12)(cid:12)∀i, jE
It is enough to provide an explicit model which is inner faithful and we define it through
its affine lift. Let (ei)1≤i≤N be the canonical basis of CK and consider for 1 ≤ t ≤ L the
subspace Vt ⊂ CK spanned by the vectors ei with (t − 1)R + 1 ≤ i ≤ tR, so that:
CK = V1 ⊕ . . . ⊕ VL
This gives a block-diagonal embedding U L
R ⊂ UK, which is the affine lift of the model
space. However, for computations it is simpler to use a permutation of this model:
Definition 8.2. We denote by XK,L the set of unitary matrices U ⊂ UK such that
(Ut, Ut+L, . . . , Ut+(R−1)L) is a basis of Vt for all 1 ≤ t ≤ L.
Defining π(gi) as usual, we see that the following element does not depend on i:
π(gi)R(U 1, . . . , U M ) =
wjRPU i
j
=
KXj=1
LXt=1
wtR R−1Xs=0
t+sL! =
PU i
wtRPVt
LXt=1
22
TEODOR BANICA AND AMAURY FRESLON
Thus, we get a representation π : C ∗(Γ) → MN (C(X M
K,L)). We have:
Theorem 8.3. The quasi-flat model π : C ∗(Γ) → MK(C(X M
free product Γ = Z∗M
K,L)) for the amalgamated
Proof. As usual, we consider an arbitrary element γ ∈ Γ. Writing h = gR
that γ = hlgk1
in with 0 < l < L and 0 < ki < R for all i. Thus:
i1 . . . gkn
i , we can assume
i = gR
K (cid:14)(cid:10)gR
j(cid:12)(cid:12)∀i, j(cid:11) is inner faithful.
π(γ)(U 1, . . . , U M ) = π(hl)(U 1, . . . , U M ) Xj1...jn
wtlRPVt! Xj1...jn
= LXt=1
LXt=1 Xj1...jn
=
wtlRw<k,j>PVtPU
. . . PU in
jn
i1
j1
w<k,j>PU
i1
j1
. . . PU in
w<k,j>PU i1
. . . PU in
j1
jn!
jn!
Let us consider the term corresponding to a fixed t. The product of projections vanishes
unless Uj1 ∈ Vt. This forces Ui2 ∈ Vt and by induction all the terms must be in Vt. This
means that there are s1, . . . , sn such that jm = t + smL for all 1 ≤ m ≤ n. Thus:
π(γ)(U 1, . . . , U M ) =
wtlRwt(k1+...+kn)
LXt=1
R−1Xs1,...,sn=0
wL<k,s>PU
i1
t+s1L
. . . PU in
t+snL
Let us denote by U i(t) the unitary operator on Vt obtained from the appropriate columns
of U i. The trace of π(γ) can be expressed using these operators. To do this, simply note
that the normalized trace tr can be written as L−1trR, where trR is the normalized trace
on UR. A computation similar to that of Theorem 7.2 then yields:
tr ◦ π(γ) =
1
L
LXt=1
wt(lR+k1+...+kn)trR nYp=1
(U ip(t)W LkpU ip∗(t))!
Assuming that π(γ) = Id, we can now derive a contradiction. Indeed, this forces:
wt(lR+k1+...+kn)
1
L
LXt=1
(U ip(t)W LkpU ip(t)) = Id
nYp=1
Because U i(t) only acts on Vt and the decomposition is orthogonal, this equation is
equivalent to the system formed by the equations for each fixed t. As before, this system
cannot always be satisfied unless kp = 0 for all p. In that case, we are left with wtlR = 1
for all t, implying l = 0 and the proof is complete.
(cid:3)
MODELLING QUESTIONS FOR QUANTUM PERMUTATIONS
23
We end with another construction. This time, we do not identify the copies of ZL but
simply make them commute. More precisely, we set:
j = gR
j gR
Γ = Z∗M
i gR
K (cid:14)(cid:10)gR
This is in a sense a mix between the free product of direct products and the amal-
L understood as tu-
t+(r−1)L) is a basis of
gamated free products. Thus, the model space is XΓ = X M
ples (U 1, . . . , U M , σ1, . . . , σM ) where U i satisfies that (U i
Vσ−1
(t) for all t. Still using the usual formula for the canonical generators, we have:
K,L × SM
t+L, . . . , U i
t , U i
i
π(gi)R(U 1, . . . , U M , σ1, . . . , σM ) =
wσi(t)RPVt
This element commutes with π(gj)R so that we indeed have a universal representation
π : C ∗(Γ) → MK(C(XΓ)). With this convention, we have:
K (cid:14)(cid:10)gR
Proposition 8.4. The quasi-flat model π : C ∗(Γ) → MK(C(XΓ)) for the group Γ =
Z∗M
i gR
j = gR
j gR
Proof. As usual, we consider a non-trivial word γ = gk1
each i, let ki = k′
i + aiR be the euclidean division of ki by R. Then:
i1 . . . gkn
in in a particular form. For
i(cid:12)(cid:12)∀i, j(cid:11)
LXt=1
π(γ)(U 1, . . . , U M , σ1, . . . , σM ) =
i(cid:12)(cid:12)∀i, j(cid:11) is inner faithful.
nYs=1 LXts=1
LXt1,...,tn=1
=
wk′
sjsPU is
js!
wσis (ts)asRPVts! KXjs=1
KXj1,...,js=1
wR<σ(t),a>w<k′,j>PVt1
P i1
Uj1
. . . PVtn P in
Ujn
For the product to be nonzero we need all the Vt's to be the same so that the first sum
reduces to only one index. Moreover, this forces as before jm = t + smL, so we get:
LXt=1
R−1Xs1,...,sn=0
wR(a1σi1 (t)+...+anσin (t))wt(k′
1+...+k′
n)wL<k′,s>PU
i1
t+s1L
. . . PU in
t+snL
As in the proof of Theorem 8.1, we conclude that k′
p = 0 for all p and that:
Summing over SM
wR(a1σi1 (t)+...+anσin (t)) = 1
L then yieldsQ δai,0 = 1 and the proof is complete.
References
(cid:3)
[1] T. Banica, Quantum groups from stationary matrix models, Colloq. Math. 148 (2017), 247–267.
[2] T. Banica and J. Bichon, Hopf images and inner faithful representations, Glasg. Math. J. 52 (2010),
677–703.
[3] T. Banica, J. Bichon and J.-M. Schlenker, Representations of quantum permutation algebras, J.
Funct. Anal. 257 (2009), 2864–2910.
24
TEODOR BANICA AND AMAURY FRESLON
[4] T. Banica and B. Collins, Integration over quantum permutation groups, J. Funct. Anal. 242 (2007),
641–657.
[5] T. Banica, U. Franz and A. Skalski, Idempotent states and the inner linearity property, Bull. Pol.
Acad. Sci. Math. 60 (2012), 123–132.
[6] T. Banica and I. Nechita, Flat matrix models for quantum permutation groups, Adv. Appl. Math.
83 (2017), 24–46.
[7] J. Bichon, Algebraic quantum permutation groups, Asian-Eur. J. Math. 1 (2008), 1–13.
[8] M. Brannan, B. Collins and R. Vergnioux, The Connes embedding property for quantum group von
Neumann algebras, Trans. Amer. Math. Soc. 369 (2017), 3799–3819.
[9] A. Chassaniol, Quantum automorphism group of the lexicographic product of finite regular graphs,
J. Algebra 456 (2016), 23–45.
[10] A. Chirvasitu, Residually finite quantum group algebras, J. Funct. Anal. 268 (2015), 3508–3533.
[11] L.S. Cirio, A. D'Andrea, C. Pinzari and S. Rossi, Connected components of compact matrix quantum
groups and finiteness conditions, J. Funct. Anal. 267 (2014), 3154–3204.
[12] U. Franz and A. Skalski, On idempotent states on quantum groups, J. Algebra 322 (2009), 1774–
1802.
[13] A. Freslon, On the partition approach to Schur-Weyl duality and free quantum groups, Transform.
Groups 22 (2017), 707–751.
[14] S. Malacarne, Woronowicz's Tannaka-Krein duality and free orthogonal quantum groups, Math.
Scand. 122 (2018), 151–160.
[15] S. Neshveyev and L. Tuset, Compact quantum groups and their representation categories, SMF
(2013).
[16] S. Raum and M. Weber, The full classification of orthogonal easy quantum groups, Comm. Math.
Phys. 341 (2016), 751–779.
[17] R. Sinkhorn, A relationship between arbitrary positive matrices and doubly stochastic matrices,
Ann. Math. Statist. 35 (1964), 876–879.
[18] E. Thoma, Uber unitare Darstellungen abzahlbarer, diskreter Gruppen, Math. Ann. 153 (1964),
111–138.
[19] S. Wang, Free products of compact quantum groups, Comm. Math. Phys. 167 (1995), 671–692.
[20] S. Wang, Quantum symmetry groups of finite spaces, Comm. Math. Phys. 195 (1998), 195–211.
[21] S. Wang, Equivalent notions of normal quantum subgroups, compact quantum groups with properties
F and FD, and other applications, J. Algebra 397 (2014), 515–534.
[22] S. Wang, Lp-improving convolution operators on finite quantum groups, Indiana Univ. Math. J. 65
(2016), 1609–1637.
[23] S.L. Woronowicz, Compact matrix pseudogroups, Comm. Math. Phys. 111 (1987), 613–665.
[24] S.L. Woronowicz, Tannaka-Krein duality for compact matrix pseudogroups. Twisted SU(N) groups,
Invent. Math. 93 (1988), 35–76.
T.B.: Department of Mathematics, University of Cergy-Pontoise, F-95000 Cergy-
Pontoise, France. [email protected]
A.F.: Department of Mathematics, Paris-Sud University, F-91405 Orsay Cedex, France.
[email protected]
|
1806.02683 | 1 | 1806 | 2018-06-06T12:28:04 | A note on the injectivity of action by compact quantum groups on a class of $C^{\ast}$-algebras | [
"math.OA",
"math.QA"
] | We give some sufficient conditions for the injectivity of actions of compact quantum groups on $C^{\ast}$-algebra. As an application, we prove that any faithful smooth action by a compact quantum group on a compact smooth (not necessarily connected) manifold is injective. A similar result is proved for actions on $C^{\ast}$- algebras obtained by Rieffel-deformation of compact, smooth manifolds. | math.OA | math |
A note on the injectivity of action by compact quantum groups
on a class of C ∗-algebras
Debashish Goswami1 and Soumalya Joardar 2
Indian Statistical Institute
203, B. T. Road, Kolkata 700108
Email: [email protected],
Phone: 0091 33 25753420, Fax: 0091 33 25773071
Dedicated to Paul F. Baum on the occasion of his eightieth birthday.
Abstract
We give some sufficient conditions for the injectivity of actions of compact quantum
groups on C ∗-algebra. As an application, we prove that any faithful smooth action by a
compact quantum group on a compact smooth (not necessarily connected) manifold is injec-
tive. A similar result is proved for actions on C ∗- algebras obtained by Rieffel-deformation
of compact, smooth manifolds.
Subject classification : 81R50, 81R60, 20G42, 58B34.
Keywords: Compact quantum group, Riemannian manifold, smooth action.
1
Introduction
Quantum groups are natural generalization of groups and they are used as 'generalized symmetry
objects' in mathematics and physics. The pioneering work by Drinfeld and Jimbo ([4], [3],
[11], [12]) and others gave the formulation of quantum groups in the algebraic setting as Hopf
algebras typically obtained by deformations of the universal enveloping algebras of semisimple
Lie algebras. This led to a deep and successful theory having connections with physics, knot
theory, number theory, representation theory etc. On the other hand, S.L. Woronowicz (see,
e.g.
[22]) approached it from a point of view of harmonic analysis on locally compact groups
and came up with a set of axioms for defining compact quantum groups (CQG for short) as
the generalization of compact topological groups.
In this note, we will restrict ourselves to
the framework of compact quantum groups only. It is natural to define quantum analogue of
group action on spaces. This can be done in different ways: for the purely algebraic approach,
In the analytic theory, there are C ∗ and von
this is defined as a co-action of Hopf algebra.
Neumann algebraic notions of action. We'll be concerned with Podles' formulation of C ∗-action
of compact quantum groups on C ∗- algebras. A subtle point about this definition is that it does
not assume the injectivity of the action. We mention here that some authors (e.g. [1]) indeed
prefer to include injectivity in the definition of a C ∗-action, but this is not a universal practice.
In fact, in [16], Soltan discussed examples of non-injective C ∗-actions. On the other hand, group
actions on spaces are always injective. Injectivity also follows in the algebraic setting for Hopf
algebra co-actions as well as for von Neumann algebraic notion of actions of (von Neumann
algebraic) quantum groups. Thus, it is an interesting and important problem to give sufficient
conditions for injectivity of C ∗-action of compact quantum groups in the sense of Podles. This is
the aim of this short note. We'll consider CQG actions on C(M ) and their Rieffel deformations
where M is a compact smooth manifold. Under a smoothness condition on the action on C(M )
( in the sense of [7]) we can prove injectivity. For the Rieffel-deformation of classical manifolds,
we prove injectivity under a natural analogue of smoothness and compatibility of the action
with the canonical toral action.
1Partially supported by J C Bose Fellowship from D.S.T. (Govt. of India) and also acknowledges the Fields
Institute, Toronto for providing hospitality for a brief stay when a small part of this work was done.
2Acknowledges support from CSIR.
1
2 Preliminaries
′
In this paper all the Hilbert spaces are over C unless mentioned otherwise. For a vector space V ,
stands for its algebraic dual. ⊕ and ⊗alg will denote the algebraic direct sum and algebraic
V
tensor product respectively. On the other hand, the minimal C ∗- algebra tensor product and
tensor product of Hilbert spaces as well as Hilbert modules will be denoted by ⊗. In particular,
we consider Hilbert modules of the form H ⊗ C where C is a C ∗ algebra. We shall denote the C ∗-
algebra of bounded operators on a Hilbert space H by B(H) and the C ∗- algebra of compact
operators on H by B0(H). Sp, Sp stand for the linear span and closed linear span of elements
of a vector space respectively, whereas Im(A) denotes the image of a linear map A. Given a
group action γ on a locally convex space Z we denote the fixed point subspace by Z γ.
We call a locally convex space Fr´echet if the family of seminorms is countable (hence the
space is metrizable) and the space is complete with respect to the metric given by the family of
seminorms. There are many ways to equip the algebraic tensor product of two locally convex
spaces with a locally convex topology. Let E1, E2 be two locally convex spaces with family of
seminorms {.1,i} and {.2,j } respectively. Then one wants a family {.i,j} of seminorms
for E1 ⊗alg E2 such that e1 ⊗ e2i,j = e11,ie22,j. The problem is that such a choice is far
from unique and there is a maximal and a minimal choice giving the projective and injective
tensor product respectively. Let us denote the projective tensor product by E1 ⊗E2. A Fr´echet
locally convex space is called nuclear if its projective and injective tensor products with any other
Fr´echet space coincide as a locally convex space. It is known that closed subspaces and quotients
by closed subspaces of a nuclear Fr´echet space are again nuclear. We do not go into further
details of this topic here but refer the reader to [18] for a comprehensive discussion. Furthermore
if the space is a ∗ algebra then we demand that its ∗ algebraic structure is compatible with
its locally convex topology i.e.
the involution ∗ is continuous and multiplication is jointly
continuous with respect to the topology. Projective and injective tensor product of two such
topological ∗ algebras are again topological ∗ algebra. We shall mostly use unital ∗ algebras.
Henceforth all the topological ∗-algebras will be unital unless otherwise mentioned. Consider a
locally convex algebra A for which each of the defining seminorms, say k · ki, satisfies
kxyki ≤ Cikxkikyki
(1)
for some constant Ci and all x, y ∈ A. Then it is easy to see from the definition of the projective
tensor product that the algebra multiplication map (say m) lifts to a continuous map from A ⊗A
to A.
We mainly need a particular class of nuclear locally convex ∗-algebra, which is C ∞(M ),
where M is any compact smooth manifold. The natural Fr´echet topology on C ∞(M ) is given
by the seminorms of the form pU,K,α,
pU,K,α(f ) = supx∈K∂α(f )(x),
where K is a compact subset contained in the domain of some coordinate chart (U, (x1, ..., xn)),
α = (i1, ..., ik) a multi index and ∂α = ∂
, ij ∈ {1, . . . , n}. We can similarly define a
∂xi1
Fr´echet topology on C ∞(M, E), the space of smooth E-valued functions on M for any Fr´echet
space E. We refer the reader to [7] for more details. One can verify condition (1) for the family
of seminorms defining the Fr´echet algebra C ∞(M, A) where A is a Banach algebra, by Leibniz
rule.
... ∂
∂xi
k
2.1 Compact quantum groups and C ∗-actions
Definition 2.1 A compact quantum group (CQG for short) is a unital C ∗-algebra Q with a
coassociative coproduct (see [19], [22]) ∆ from Q to Q ⊗ Q such that each of the linear spans
of ∆(Q)(Q ⊗ 1) and that of ∆(Q)(1 ⊗ Q) is norm-dense in Q ⊗ Q.
2
A unitary representation of a CQG (Q, ∆) on a Hilbert space H is a unitary U ∈ L(H ⊗ Q)
such that the C-linear map V from H to the Hilbert module H ⊗ Q given by V (ξ) = U (ξ ⊗ 1)
satisfies (V ⊗ id) ◦ V = (id ⊗ ∆) ◦ V. Here, the map (V ⊗ id) denotes the extension of V ⊗ id to
the completed tensor product H ⊗ Q which exists as V is an isometry.
Every CQG Q contains a canonical dense unital ∗-subalgebra Q0 of Q on which linear
maps κ and ǫ (called the antipode and the counit respectively) are defined making the above
subalgebra a Hopf ∗ algebra. In fact, this is the algebra generated by the 'matrix coefficients'
of the (finite dimensional) irreducible non degenerate representations (see [19] ) of the CQG.
The antipode is an anti-homomorphism and also satisfies κ(a∗) = (κ−1(a))∗ for a ∈ Q0.
It is known that there is a unique state h on a CQG Q (called the Haar state) which is bi
invariant in the sense that (id⊗h)◦∆(a) = (h⊗id)◦∆(a) = h(a)1 for all a. The Haar state need
not be faithful in general, though it is always faithful on Q0 at least. Given the Hopf ∗-algebra
Q0, there can be several CQG's which have this ∗-algebra as the Hopf ∗-algebra generated by
the matrix elements of finite dimensional representations. We need two of such CQG's: the
reduced and the universal one. By definition, the reduced CQG Qr is the image of Q in the
GNS representation of h, i.e. Qr = πr(Q), πr : Q → B(L2(h)) is the GNS representation.
There also exists a largest such CQG Qu, called the universal CQG corresponding to Q0. It
is obtained as the universal enveloping C ∗-algebra of Q0. We also say that a CQG Q is universal
if Q = Qu. Given two CQG's (Q1, ∆1) and (Q2, ∆2), a ∗ homomorphism π : Q1 → Q2 is said
to be a CQG morphism if (π ⊗ π) ◦ ∆1 = ∆2 ◦ π on Q1. In case π is surjective, Q2 is said to be
a quantum subgroup of Q1 and denoted by Q2 ≤ Q1.
Definition 2.2 We say that a CQG (Q, ∆) (co)-acts on a (unital) C ∗-algebra C if there is a
unital ∗-homomorphism α : C → C ⊗ Q such that (α ⊗ id) ◦ α = (id ⊗ ∆) ◦ α, and the linear span
of α(C)(1 ⊗ Q) is norm-dense in C ⊗ Q.
In Woronowicz theory, it is customary to drop 'co', and call the above co-action simply 'action'
of the CQG on the C ∗-algebra. Let us adopt this convention for the rest of the note. An action
α on C is called faithful if the ∗-subalgebra generated by {(ω ⊗ id)(α(b))}, where b ∈ C and ω
varying over the set of bounded linear functionals on C, is dense in Q.
Given an action α, we define αr = (id ⊗ πr) ◦ α and call it the reduced action. If the Haar
state is faithful on Q, we have α = αr.
Definition 2.3 We call an action α of a CQG Q on a unital C ∗-algebra C to be implemented
by a unitary representation U of Q in H, say, if there is a faithful representation π : C → B(H)
such that U (π(x) ⊗ 1)U ∗ = (π ⊗ id)(α(x)) for all x ∈ C.
It is clear that if an action is implemented by a unitary representation then it is one-to-one.
In fact, as (id ⊗ πr)(U ) gives a unitary representation of Qr in H and the 'reduced action'
αr := (id ⊗ πr) ◦ α of Qr is also implemented by a unitary representation, it follows that even
αr is one-to-one. We see below that this is actually equivalent to implementability by unitary
representation:
Lemma 2.4 Given an action α of Q on a unital separable C ∗-algebra C, the following are
equivalent:
(a) There is a faithful positive functional φ on C which is invariant w.r.t. α, i.e. (φ⊗id)(α(x)) =
φ(x)1Q for all x ∈ C.
(b) The action is implemented by some unitary representation.
(c) The reduced action αr of Qr is injective.
Proof:
If (a) holds, we consider H to be the GNS space of the faithful positive functional φ. The GNS
representation π is faithful, and the linear map V defined by V (x) := α(x) from C ⊂ H =
3
L2(C, φ) to H ⊗ Q is an isometry by the invariance of φ. Thus V extends to H and it is easy to
check that it induces a unitary representation U , given by U (ξ ⊗ q) = V (ξ)q, which implements
α.
We have already argued (b) ⇒ (c), and finally, if (c) holds, we choose any faithful state say
τ on the separable C ∗ algebra C and take φ(x) = (τ ⊗ h)(αr(x)), which is faithful as h is faithful
on Qr and αr is injective. It can easily be verified that φ is α-invariant on the dense subalgebra
C0 mentioned before, and hence on the whole of C.✷
We'll also need the following facts about actions on commutative C ∗-algebras.
Proposition 2.5 If a CQG Q acts faithfully on C(X) for some compact metrizable space X,
then Q is separable and it is also of Kac type, i.e. κ is norm-bounded on Qr, κ2 = id and the
Haar state is tracial.
Proof
Note that X is second countable and hence C(X) is separable. Choose a countable dense set
of points {xi, i = 1, 2, . . .} and a countable norm-dense subset {fn, n = 1, 2, . . .} of C(X).
It follows from faithfulness of the action α (say) that Q is generated as a C ∗-algebra by the
countable set {α(fn)(xi), i, n = 1, 2, . . .}, hence it is separable.
For the proof of the Kac conditions, see [10].✷
3 Smooth actions are injective
Let M be a compact smooth manifold. Let us recall the definition of smooth CQG action on it
from [7].
Definition 3.1 We say that an action α of a CQG Q on C(M ) is smooth if α maps C ∞(M )
into C ∞(M, Q) and Sp α(C ∞(M ))(1 ⊗ Q) is dense in C ∞(M, Q) in the Frechet topology.
Theorem 3.2 If Q has a faithful smooth action α on C ∞(M ), where M is compact manifold,
then for every fixed x ∈ M there is a well-defined, ∗-homomorphic extension ǫx of the counit map
ǫ of Q0 to the unital ∗-subalgebra Q∞
x := {αr(f )(x) : f ∈ C ∞(M )} satisfying ǫx(αr(f )(x)) =
f (x), where αr is the reduced action discussed earlier.
Proof:
Adapting the arguments of [14] and [1], we can get a the Fr´echet dense subalgebra C0 of C ∞(M )
on which α restricts to an algebraic co-action of Q0. For example, C0 may be chosen as the
Peter-Weyl subalgebra in the sense of [1]. Replacing Q by Qr we can assume without loss of
generality that Q has faithful Haar state and α = αr. In this case Q will have bounded antipode
κ (by Proposition 2.5). Let αx : C ∞(M ) → Q∞
x be the map defined by αx(f ) := α(f )(x). It
is clearly continuous w.r.t. the Fr´echet topology of C ∞(M ) and hence the kernel Ix (say) is
a closed ideal, so that the quotient, which is isomorphic to Q∞
x , is a nuclear space. Let us
consider Q∞
x with this topology and then by nuclearity, the projective and injective tensor
products with Q (viewed as a Banach space, which is separable by Proposition 2.5) coincide.
The multiplication map m : Q∞
x ⊗alg Q → Q extends to a continuous map (to be denoted by
x ⊗Q. It follows from the relation (id ⊗ ∆) ◦ α = (α ⊗ id) ◦ α that ∆(αx(f )) =
m again) on Q∞
x ⊗Q. Thus, the composite
x to the (αx ⊗idQ)(C ∞(M ) ⊗Q) ⊆ Q∞
(αx ⊗id)(α(f )), i.e. ∆ maps Q∞
map β := m ◦ (id ⊗ κ) ◦ ∆ : Q∞
x → Q is continuous. Clearly, this map coincides with ǫ(·)1Q on
the Fr´echet-dense subalgebra of Q∞
x spanned by elements of the form α(f )(x), with f varying
in the Fr´echet-dense subalgebra C0 of C ∞(M ). By continuity of β, it follows that the range of
β is C1Q. Hence we can define ǫx by setting ǫx(·)1Q = β(·). This completes the proof of the
lemma.✷
Corollary 3.3 For any smooth action α on C ∞(M ), the conditions of Theorem 2.4 are satis-
fied, hence the reduced action is injective on C(M ).
4
Proof:
Replacing Q by the Woronowicz subalgebra generated by {α(f )(x), f ∈ C(M ), x ∈ M } we
may assume that α is faithful. If αr(f ) = 0 for f ∈ C ∞(M ) then by Lemma 3.2 applying the
extended ǫ we conclude f = 0. Now, consider any positive Borel measure µ of full support on
M , with φµ being the positive functional obtained by integration w.r.t µ. Let ψ := (φµ ⊗ h) ◦ αr
be the positive functional which is clearly αr-invariant and faithful on C ∞(M ), i.e. ψ(f ) =
0, f ∈ C ∞(M ) and f nonnegative implies f = 0. But by Riesz Representation Theorem there is
a positive Borel measure ν such that ψ(f ) = RM f dν. We claim that ν has full support, hence
ψ is faithful also on C(M ). Indeed, for any nonempty open subset U of M there is a nonzero
positive f ∈ C ∞(M ), with 0 ≤ f ≤ 1, and support of f is contained in U . By faithfulness of ψ
on C ∞(M ) we get 0 < ψ(f ) = RU f dν ≤ ν(U ).✷
Remark 3.4 In a recent work [9], Goswami has proved that any CQG which admits a faithful
smooth action on a compact connected smooth manifold must be isomorphic to C(G) for some
group G and the CQG action becomes G-action. Hence it is in particular injective. However,
the result of the present note is applicable to a possibly disconnected manifold. Moreover, the
proof of injectivity given here is rather short and direct.
4 Smooth action on Rieffel deformation
Let us now consider CQG actions on noncommutative C ∗-algebras. Rieffel deformation (see [15])
is a well-known and very useful procedure to obtain interesting noncommutative C ∗-algebras
from the commutative ones. In particular, given a smooth compact Riemannian manifold M
equipped with a toral subgroup T ∼= Tn ⊆ ISO(M ), one can construct a family of (typically
noncommutative) C ∗-algebras C(M )θ indexed by n × n skew symmetric matrices θ. There is a
similar procedure (see [21]) for deforming a CQG with some toral quantum subgroup of rank
n inducing an action of torus of rank 2n combining left and right action by the elements of
the n-toral subgroup. In this case, one gets a CQG by retaining the same co-algebra structure
as the original one but changing the algebra structure. This will be called the Rieffel-Wang
deformation of G.
Let Aθ be the noncommutative n-torus, which is the universal C ∗-algebra generated by
unitaries U1, . . . Un satisfying the commutation relations Uj Uk = exp(2πiθjk)UkUj, where θ =
((θjk)). Given a unital C ∗-algebra C with a Tn-action given by βz (say), the deformed C ∗-algebra
Cθ can be described in two alternative ways: either in the original picture of Rieffel where one
defines a new, twisted multiplication on the spectral algebra for the toral action and then
considers appropriate C ∗-completion, or as in [6], identifying Cθ with the fixed point subalgebra
(C ⊗ Aθ)β⊗v−1
where vz denotes the canonical toral action on Aθ satisfying vz(Ui) = ziUi for
all i. We have a 'dual' T -action on Cθ which is the restriction of (id ⊗ v) on C ⊗ Aθ.
Given a CQG Q and a quantum subgroup of Q isomorphic with T = Tn, with the corre-
sponding surjective CQG morphism π : Q → C(T ), we can define a left and a right Tn-action,
say χl
z = ((evz ◦ π) ⊗ id) ◦ ∆. Using
this, we have a T2n-action χz,w = χl
w on Q and the corresponding deformed CQG is the
C ∗-algebra Qθ⊕(−θ).
z = (id ⊗ (evz ◦ π)) ◦ ∆ and χr
zχr
z, χr
z respectively, by setting χl
We have the following from Theorem 3.11 of [2].
Lemma 4.1 Let C be a unital C ∗-algebra equipped with a Tn-action given by ∗-automorphism
βz, Q be a reduced CQG with an action α of Q on C and a quantum subgroup of Q isomorphic
with Tn as above (with the corresponding morphism π) satisfying βz := (id ⊗ (evz ◦ π)) ◦ α. Then
we have an action αθ of Qθ (θ = θ ⊕ (−θ)) on Cθ. Here the deformation of C is taken w.r.t. the
action β.
Consider now C = C(M ), M being a compact smooth Riemannian manifold equipped with
Tn action, which also induces a Tn action (say β) on C(M ). Let γ = β ⊗ v−1 as before
5
and let us call the subalgebra C ∞(M, Aθ)γ ⊂ C(M, Aθ)γ ≡ (C(M ) ⊗ Aθ)γ = C(M )θ the
'smooth subalgebra' and call an action α on C(M )θ by a CQG Q to be smooth if it maps
the above smooth subalgebra into C ∞(M, Aθ ⊗ Q)(γ⊗id) and the linear span of α(F )(1 ⊗ Q),
F ∈ C ∞(M, Aθ)γ is dense in C ∞(M, Aθ ⊗ Q)(γ⊗id). Now, we can state and prove the following:
Theorem 4.2 Let M be as above and let α be a smooth action of a CQG on C(M )θ in the
above sense. Moreover, assume that there is a quantum subgroup of Q isomorphic with T = Tn,
given by a surjective CQG morphism π : Q → C(T ) such that (id ⊗ (evz ◦ π)) ◦ α coincides with
the canonical 'dual' T -action on C(M )θ. Then the action (in fact, the reduced one) is injective.
Proof:
We only very briefly sketch the proof. As before, assume without loss of generality that the
CQG is reduced. It follows from the proof of Theorem 3.11 of [2] that the action α−θ of Qθ on
∼= C(M ) is smooth. We note that the word 'smooth' in the statement of Theorem
(C(M )θ)−θ
3.11 of [2] is used in a sense weaker than ours:
it only means the invariance of the smooth
algebra there. However, the dense subalgebra (e.g. Peter-Weyl subalgebra)of C(M )θ for the
action α, on which Q0 (co)acts algebraically, can be identified as a vector space with a dense
subalgebra of C ∞(M ) on which the deformed action (which is the same as α a linear map on this
space) α−θ is algebraic. From this, the Podles-type density condition follows, i.e. α−θ is smooth
in our sense. Hence it is injective by Corollary 3.3. Moreover, by that corollary and Theorem
2.4, we get a unitary representation of Qθ which implements α−θ. But by the generalities of
Rieffel-Wang (or, more general cocycle-twisted) deformation of CQG as in the Chapter 7 of [8],
we conclude that α = (α−θ)θ is unitarily implemented too, where the corresponding Hilbert
space and unitary essentially remain the same. In particular, α is injective.✷
Ackowledgement:
The first author would like to recall his fond memories of interaction with Prof. Paul Baum on
several occasions: ICTP (Trieste), TIFR (India) and also in the conference in the Fields Institute
in 2016. Although there was no discussion on the particular topic of this note, the first author
is grateful to him for encouragement and suggestions regarding Baum-Connes' conjecture for
quantum groups and other problems related to quantum symmetry on operator algebras. The
authors wish him a long, healthy and creative life.
References
[1] Baum, Paul F.; De Commer, Kenny; Hajac, Piotr M.: Free actions of compact quantum
groups on unital C ∗-algebras. Doc. Math. 22 (2017), 825-849.
[2] Bhowmick, Jyotishman; Goswami, Debashish: Quantum isometry groups: examples and
computations. Comm. Math. Phys. 285 (2009), no. 2, 421-444.
[3] Drinfeld, V.G.: Quantum Groups, Proceedings of the International Congress of Mathe-
maticians, Berkeley, 1986.
[4] Drinfeld, V. G.: Quasi-Hopf algebras, Leningrad Math. J. 1 (1990), 1419-1457.
[5] Skandalis, G. and Baaj, S.:
locally compact quantum groups,
Mathematisches- Forschungsinstitut Oberwolfach, Tagungsbericht 46/1991, C ∗ -algebren,
20.10 26.10.1991, p. 20.
Duality for
[6] Connes A. and Michel Dubois-Violette, Noncommutative finite-dimensional manifolds. I.
Spherical manifolds and related examples. Comm. Math. Phys., 230(3):539-579, 2002.
[7] Goswami, D. and Joardar, S: Non-existence of faithful isometric action of compact quan-
tum groups on compact, connected Riemannian manifolds. Geom. Funct. Anal. 28 (2018),
no. 1, 146-178.
6
[8] Goswami, D. and Bhowmick, J.: "Quantum Isometry Groups", Infosys Series, Springer
India, 2016.
[9] Goswami, D.: Non-existence of genuine (compact) quantum symmetries of compact, con-
nected smooth manifolds, preprint.
[10] Huang, H.: Faithful compact quantum group actions on connected compact metrizable
spaces, Journal of Geometry and Physics, Volume 70, August 2013, 232-236.
[11] M. Jimbo: Quantum R-matrix for the generalized Toda system, Commun, Math, Phys.
10 (1985), 63 - 69.
[12] Jimbo, M.: A q-difference analogue of U(g) and the Yang-Baxter equation, Lett. Math.
Phys. 10 (1985), no. 1, 63 - 69.
[13] Pietsch, A.: "Nuclear locally convex spaces", Springer-Verlag 1972.
[14] Podles, P.: Symmetries of Quantum Spaces, subgroups and quotient spaces of SU (2) and
SO(3) groups, Comm. Math. Phys., 170(1):1995, 1-20.
[15] Rieffel , Mark A. : Deformation Quantization for actions of Rd , Memoirs of the American
Mathematical Society , November 1993 . Volume 106 . Number 506 .
[16] Soltan, P.: On actions of compact quantum groups, Illinois J. Math. Volume 55, Number
3 (2011), 719-1266.
[17] Takesaki, M: "Theory of Operator Algebra I", Springer (2002).
[18] Treves, F.: "Topological Vector Spaces, Distributions and Kernels", Academic Press, New
York- London (1967).
[19] Maes, A. and Van Daele, A.: Notes on compact quantum groups, Nieuw Arch. Wisk. (4)
16 (1998), no. 1-2, 73-112.
[20] Wang, S.: Quantum symmetry groups of finite spaces, Comm. Math. Phys., 195(1998),
195-211.
[21] Wang, S.: Deformation of Compact Quantum groups via Rieffel's Quantization, Comm.
Math. Phys. 178( 1996 ), 747 - 764.
[22] Woronowicz, S.L.: Compact Matrix Pseudogroups, Comm. Math. Phys., 111(1987), 613-
665.
[23] Woronowicz, S.L., Zakrzewski : Quantum 'ax+b' group, Reviews in Math. Phys., 14, Nos
7 & 8 (2002), 797-828.
7
|
1903.07143 | 3 | 1903 | 2019-05-02T15:44:17 | Rigidity results for von Neumann algebras arising from mixing extensions of profinite actions of groups on probability spaces | [
"math.OA"
] | Motivated by Popa's seminal work \cite{Po04}, in this paper, we provide a fairly large class of examples of group actions $\Gamma \curvearrowright X$ satisfying the extended Neshveyev-St{\o}rmer rigidity phenomenon \cite{NS03}: whenever $\Lambda \curvearrowright Y$ is a free ergodic pmp action and there is a $\ast$-isomorphism $\Theta:L^\infty(X)\rtimes \Gamma \rightarrow L^\infty(Y)\rtimes \Lambda$ such that $\Theta(L(\Gamma))=L(\Lambda)$ then the actions $\Gamma\curvearrowright X$ and $\Lambda \curvearrowright Y$ are conjugate (in a way compatible with $\Theta$). We also obtain a complete description of the intermediate subalgebras of all (possibly non-free) compact extensions of group actions in the same spirit as the recent results of Suzuki \cite{Suzuki}. This yields new consequences to the study of rigidity for crossed product von Neumann algebras and to the classification of subfactors of finite Jones index. | math.OA | math |
Rigidity results for von Neumann algebras arising from mixing
extensions of profinite actions of groups on probability spaces
Ionut Chifan and Sayan Das
Abstract
Motivated by Popa's seminal work [Po04], in this paper, we provide a fairly large class of
examples of group actions Γ y X satisfying the extended Neshveyev-Størmer rigidity phe-
nomenon [NS03]: whenever Λ y Y is a free ergodic pmp action and there is a ∗-isomorphism
Θ : L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ such that Θ(L(Γ)) = L(Λ) then the actions Γ y X and Λ y Y
are conjugate (in a way compatible with Θ). We also obtain a complete description of the
intermediate subalgebras of all (possibly non-free) compact extensions of group actions in the
same spirit as the recent results of Suzuki [Su18]. This yields new consequences to the study
of rigidity for crossed product von Neumann algebras and to the classification of subfactors of
finite Jones index.
1
Introduction
In the mid thirties Murray and von Neumann found a natural way to associate a von Neumann
algebra to any measure preserving action Γ y X of a countable group Γ on a probability space X.
This is called the group measure space von Neumann algebra, denoted by L∞(X) ⋊ Γ. The most
interesting case for study is when the initial action Γ y X is free and ergodic, in which case the
group measure space construction is in fact a type II1 factor. When X is a singleton the group
measure space construction yields just the group von Neumann algebra that will be denoted by
L(Γ). The latter is a II1 factor specifically when all nontrivial conjugacy classes of Γ are infinite
(henceforth abbreviated as the icc property).
A problem of central importance in von Neumann algebras is to determine how much informa-
tion about the action Γ y X can be recovered from the isomorphism class of L∞(X) ⋊ Γ. An
unprecedented progress in this direction emerged over the last decade from Popa's influential de-
formation/rigidity theory [Po06]. A remarkable achievement of this theory was the discovery of
first classes of examples of actions that are entirely remembered by their von Neumann algebras;
for some examples see [Po06, Po07, Io08b, Pe09, PV09, Io10, CP10, Va10b, HPV10, FV10, CS11,
CSU11, PV11, PV12, Io12, Bo12, CIK13, CK15, Dr16, GITD16]. We refer the reader to the surveys
[Va10a, Io18] for an overview of the recent developments.
There are two distinguished subalgebras of L∞(X) ⋊ Γ: the coefficient (or Cartan) subalgebra
L∞(X) ⊂ L∞(X)⋊Γ and the group von Neumann subalgebra L(Γ) ⊂ L∞(X)⋊Γ. The classification
of group measure space von Neumann algebra is closely related to the study of these two inclusions
of von Neumann algebras. For instance, in [Si55] Singer observed that the study of the inclusion
L∞(X) ⊂ L∞(X) ⋊ Γ amounts to the study of the equivalence relation induced by the orbits of
1
Γ y X. Thus reconstructing the action Γ y X from the inclusion L∞(X) ⊂ L∞(X) ⋊ Γ relies
upon the reconstruction from its orbits. This theme in contemporary ergodic theory is known as
orbit equivalence rigidity. The study of orbit equivalence rigidity has received a lot of attention
over the last couple of decades and has major consequences to the classification of von Neumann
algebras in general, and the structure of the crossed product algebras in particular; for instance see
[Fu99a, Ga09, MS04, Ki06, Io08b, CK15, GITD16].
Deriving information about the action Γ y X from the other inclusion L(Γ) ⊂ L∞(X)⋊Γ is another
topic which is implicit in many core rigidity results in von Neumann algebras [NS03, Po03, Po04,
OP07]. When Γ is abelian L∞(X) ⋊ Γ = R is the hyperfinite II1 factor and each of L∞(X) and
L(Γ) is a maximal abelian subalgebra of R (henceforth abbreviated as MASA). In their study on
structural aspects of these MASAs in [NS03] Neshveyev and Størmer discovered that the positions
of these two MASAs inside R completely determines the action. More precisely, they showed the
following: Let Γ be an infinite abelian group, Γ y X be a weak mixing action and Λ y Y be any
action. If there is a ∗-isomorphism Θ : L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ satisfying Θ(L(Γ)) = L(Λ) and
Θ(L∞(X)) is inner conjugate to L∞(Y ) then Γ y X is conjugate with Λ y Y (in a way compatible
to Θ). They also conjectured the same statement holds without the inner conjugacy of the Cartan
subalgebras condition. In other words the inclusion L(Γ) ⊂ L∞(X) ⋊ Γ alone completely captures
the entire crossed product structure of L∞(X) ⋊ Γ.
The first examples of actions satisfying the full statement of Neshveyev-Størmer conjecture emerged
from the impressive work of Popa on the classification of von Neumann algebras associated with
Bernoulli actions, [Po03, Po04]. Specifically, using his influential deformation/rigidity theory Popa
was able to show that this is the case for all clustering (e.g. Bernoulli) actions Γ y X [Po04,
Theorem 0.7]. Remarkably, this holds even when Γ is nonabelian. These significant initial advances
strongly suggest that the Neshveyev-Stormer conjecture could hold in a much larger generality that
supersedes the amenable regime (e.g. Γ is abelian). Motivated by this and the implicit relevance to
the study of rigidity aspects for crossed products it is natural to investigate the following extended
version of the Neshveyev-Størmer rigidity question:
Question 1.1 (Extended Neshveyev-Stormer rigidity question). Let Γ and Λ be icc countable dis-
crete groups and let Γ y X and Λ y Y be free, ergodic, pmp actions. Assume that there is a
∗-isomorphism Θ : L∞(X) ⋊ Γ → L∞(Y ) ⋊ Λ such that Θ(L(Γ)) = L(Λ). Under what conditions
on Γ y X are the actions Γ y X and Λ y Y conjugate?
Besides Popa's examples at this time there are several other families of specific actions Γ y X
for which Question 1.1 has a solution. These arise mostly from decade-long developments in the
classification of von Neumann algebras via Popa's deformation/rigidity program. For instance, this is
the case for all W ∗-superrigid actions (see [Io18] for a survey on W ∗ superrigidity and the references
therein). Also, using [Po04, Theorem 5.2] one can easily see that the rigidity phenomenon in Question
1.1 is also satisfied by any weak mixing action Γ y X for which, up to unitary conjugacy, L∞(X)
is the unique group measure space Cartan subalgebra of L∞(X) ⋊ Γ. This way one can get more
examples using the recent results on uniqueness of Cartan subalgebras, see [OP07, PV11, PV12,
Io12, CIK13, CK15] for example. However not much was known beyond these classes of examples
and it remained open to find a more intrinsic approach to Question 1.1 which does not rely on
uniqueness of Cartan subalgebras results from deformation/rigidity theory.
In this article we develop new technical aspects that enables us to partially answer Question 1.1.
In particular we are able to describe a fairly large family of actions which covers many new ex-
amples beyond all the aforementioned classes, e.g. all nontrivial mixing extensions of free compact
2
actions, satisfying the extended Neshveyev-Størmer rigidity phenomenon. More generally, we have
the following result.
Theorem 1.2. Let Γ be an icc group and let Γ yσ X be an action whose distal quotient Γ y Xd
is free and the extension π : X → Xd is (nontrivial) mixing. Let Λ yα Y be any action. Assume
that Θ : L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ is a ∗-isomorphism such that Θ(L(Γ)) = L(Λ). Then there
exist a unitary x ∈ L(Λ), a character ω : Γ → T, and a group isomorphism δ : Γ → Λ such that
xΘ(L∞(X))x∗ = L∞(Y ) and for all a ∈ L∞(X), γ ∈ Γ we have
Θ(auγ) = ω(γ)Θ(a)x∗vδ(γ)x.
In particular, we have xΘ(σγ (a))x∗ = αδ(γ)(xΘ(a)x∗) and hence Γ y X and Λ y Y are conjugate.
Here {uγ}γ∈Γ and {vλ}λ∈Λ are the canonical group unitaries implementing the actions in L∞(X)⋊Γ
and L∞(Y ) ⋊ Λ, respectively.
In particular the theorem implies that if Γ is any icc group then any action Γ y X which admits a
free profinite quotient Γ y Xd with (nontrivial) mixing extension π : X → Xd satisfies the extended
Neshveyev-Størmer rigidity question. As a concrete example let Γ be any icc residually finite group
and let ···⊳Γn⊳···⊳Γ2 ⊳Γ1⊳Γ be a resolution of finite index normal subgroups satisfying ∩nΓn = 1.
Consider the action Γ y (Γ/Γn, cn) by left multiplication of Γ on the left cosets Γ/Γn seen as a finite
probability space with the counting measure cn and let Γ y (Z, µ) = lim←−(Γ/Γn, cn) be the inverse
limit of these actions. In addition let π : Γ y O(H) be any mixing orthogonal representation and let
Γ y (Y π, νπ) be the corresponding Gaussian action. Then the diagonal action Γ y (Y π × Z, νπ × µ)
is profinite-by-(nontrivial) mixing, and hence by Theorem 1.2 the rigidity Question 1.1 has a positive
solution in this case.
Theorem 1.2 is obtained by heavily exploiting, at the von Neumann algebraic level, the natural
tension that occurs between mixing and compactness properties for actions. Briefly, let Γ y X
and Λ y Y be actions as in Theorem 1.2 so that L∞(X) ⋊ Γ = L∞(Y ) ⋊ Λ with L(Γ) = L(Λ).
First we use the description of compactness via quasinormalizers from [CP11, Io08a] to identify the
von Neumann algebras of their distal parts, i.e. L∞(Xd) ⋊ Γ = L∞(Yd) ⋊ Λ. In turn this is used
to show that the mixing property of the extension L∞(Xd) ⊆ L∞(X) is transferred through von
Neumann equivalence to the extension L∞(Yd) ⊆ L∞(Y ) (Theorem 2.8). Once these are established,
some basic adaptations of Popa's intertwining techniques from [Po03] further show that the Cartan
subalgebras L∞(X) and L∞(Y ) are in fact unitarily conjugate. Then the desired result is derived
from a general principle which states that for any free ergodic actions Γ y X, Λ y Y of icc groups
Γ and Λ, inner conjugacy of L∞(X) and L∞(Y ) together with L(Γ) = L(Λ) imply conjugacy of
Γ y X and Λ y Y (Theorem 4.5). This criterion for conjugacy of group actions generalizes the
earlier works [NS03, Po04] and is obtained using the notion of height of elements with respect to
groups from [IPV10]. Specifically, using Dye's theorem and an averaging argument we show that Γ
has large height with respect to Λ inside L(Λ) (Theorem 4.4). By [IPV10, Theorem 3.1] this further
implies Γ is unitarily conjugate to Λ. Further exploiting the icc condition we deduce conjugacy of
the actions (Theorem 4.5).
While Theorem 1.2 settles the extended Neshveyev-Størmer rigidity question for nontrivial exten-
sions, two natural extreme situations, namely, when Γ y X is either mixing or compact (even
profinite) remain open. We believe that in both of these cases one should still get a positive answer
and we formulate a few sub problems in this direction; see for instance Problem 4.12. However,
3
in order to successfully tackle these questions, significant new technical advancements are needed.
Specifically, if one pursues an approach similar to Theorem 1.2 the key step is to establish the inner
conjugacy of L∞(X) and L∞(Y ). In the presence of mixing this would follow if one can show there
exist free factors Γ y X0 of Γ y X and Λ y Y0 of Λ y Y whose von Neumann algebras coincide,
i.e. L∞(X0) ⋊ Γ = L∞(Y0) ⋊ Λ; see Corollary 4.9. In turn this highlights the importance of studying
intermediate subalgebras in the inclusion L(Γ) ⊂ L∞(X) ⋊ Γ. In addition this seems relevant even
to the study of Question 1.1 for profinite actions.
Note that when Γ is icc, and Γ y X is free, ergodic and pmp, the inclusion L(Γ) ⊂ L∞(X) ⋊ Γ is an
irreducible inclusion of II1 factors. In his seminal paper [Jo81] Jones pioneered the study of inclusions
of type II1 factors, or subfactors. Subfactor theory has had a number of striking applications over the
years in various diverse branches of mathematics and mathematical physics, including Knot theory
and Conformal Field theory, [Jo90, Jo91, Jo09]. A major motivating question in Subfactor theory is
the classification of all intermediate subalgebras. Pursuing this perspective, we were able to classify
all the intermediate subalgebras in compact extensions in the same spirit as Suzuki's recent results
from [Su18]. To properly introduce our result we briefly recall some terminology. Given two actions
Γ yβ X0 and Γ yα X we say that α is an extension of β if there is a Γ-equivariant factor map
π : X→X0. At the von Neumann algebra level this induces an inclusion L∞(X0) ⊆ L∞(X) on which
Γ acts naturally via αγ(f ) = f ◦αγ−1 when f ∈ L∞(X). An intermediate extension for π (or between
Γ y X0 and Γ y X) is an action Γ y Z for which there exist Γ-equivariant factor maps π1 : X→Z
and π2 : Z→X0 such that π2 ◦ π1 = π. Note that the intermediate extensions of π are in bijective
correspondence with the Γ-invariant intermediate subalgebras of L∞(X0) ⊆ L∞(X). We show that
there is a bijective correspondence between intermediate von Neumann algebras in crossed products
and intermediate extensions of dynamical systems. More precisely, we have the following
Theorem 1.3. Let Γ be an icc group and let Γ yβ X0 be a pmp action. Let Γ y X be an ergodic
compact extension of β, [Fur77]. Consider the corresponding group measure space von Neumann
algebras and note that we have the following inclusion L∞(X0) ⋊ Γ ⊆ L∞(X) ⋊ Γ. Then for any
intermediate von Neumann subalgebra L∞(X0) ⋊ Γ ⊆ N ⊆ L∞(X) ⋊ Γ there exists an intermediate
extension Γ y Z between Γ y X and Γ y X0 satisfying N = L∞(Z) ⋊ Γ.
In many respects this theorem complements the results from [Su18]; for instance, it covers various
examples of non-free extensions, most notably, when X0 is a singleton. In this situation our result
provides a complete description of all intermediate von Neumann subalgebras in the inclusion L(Γ) ⊆
L∞(X) ⋊ Γ for any compact ergodic action Γ y X of any icc group Γ. This in turn yields new
interesting consequences towards the classification of finite index subfactors. For example, combining
Theorem 1.3 with the characterization of compactness via quasinormalizers from [Io08a, Theorem
6.10], for any icc group Γ and any free ergodic action Γ y X, we are able to classify all the
intermediate subfactors L(Γ) ⊆ N ⊆ L∞(X)⋊Γ with finite Jones index [N : L(Γ)] < ∞. Specifically
we show that all such N could arise only from the transitive finite factors of Γ y X (see part 2. in
Corollary 1.4); in particular, this entails that the Jones index [N : L(Γ)] is always a positive integer.
This should be compared with the similar statement [Po85, Corollary 2.4] for the intermediate
subfactors of the Cartan inclusion L∞(X) ⊂ N ⊆ L∞(X) ⋊ Γ with [L∞(X) ⋊ Γ : N ] < ∞.
Corollary 1.4. Let Γ be an icc group and let Γ y X be a free ergodic pmp action. If M = L∞(X)⋊Γ
is the corresponding group measure space construction then the following hold:
1. For any intermediate von Neumann algebra L(Γ) ⊆ N ⊆ L∞(X)⋊Γ satisfying N ⊆ QN M (L(Γ))′′
there exists a factor Γ y X0 of Γ y X such that N = L∞(X0) ⋊ Γ.
4
2. If L(Γ) ⊆ N ⊆ L∞(X) ⋊ Γ is an intermediate subfactor with [N : L(Γ)] < ∞ then there
is a finite, transitive factor Γ y X0 of Γ y X such that N = L∞(X0) ⋊ Γ; in particular,
[N : L(Γ)] ∈ N. Thus for any subfactors L(Γ) ⊆ N1 ⊆ N2 ⊆ L∞(X) ⋊ Γ, with either
[N1 : L(Γ)] < ∞ or Γ y X compact, we have [N2 : N1] ∈ N ∪ {∞}.
In particular, part 2. implies that for any icc group Γ with no proper finite index subgroups and any
free ergodic action Γ y X there are no nontrivial intermediate subfactors L(Γ) ⊆ N ⊆ L∞(X) ⋊ Γ
of finite index [N : L(Γ)] < ∞. For example this is the case for all Γ infinite simple groups, e.g.
Tarski's monsters, Burger-Mozes groups [BM01], Camm's groups [Ca53], or Bhattacharjee's groups
[Bh94], just to enumerate a few.
We point out in passing that Theorem 1.3 actually holds in a more general setting, namely, for
actions of groups on compact extensions of possibly non-abelian von Neumann algebras; this notion
is highlighted in Definition 3.9. In this generality our result yields a twisted version of Ge's splitting
theorem for tensor products (see Corollary 3.13) in the same spirit as [Su18, Example 4.14].
The classification of the intermediate subalgebras in Theorem 1.3 is achieved through a new mix of
analytic and algebraic techniques that combines factoriality arguments together with a general alge-
braic criterion outlined in Theorem 3.2. We also note the same criterion can be used in conjunction
with various soft analytical arguments to successfully recover, in the finite von Neumann algebra
case, several well-known results such as Ge's tensor splitting theorem [Ge96, Theorem 3.1] or the
Galois correspondence for group actions [Ch78]. These applications are presented in Corollary 3.3
and Theorems 3.4 and 3.7.
Finally, Theorem 1.3 in combination with methods from Popa's deformation/rigidity theory and
Jones' finite index subfactor theory provide new insight towards rigidity aspects for II1 factors arising
from profinite actions Γ y X of icc property (T) groups Γ. While Ioana has already established in
[Io08b] that such actions are completely reconstructible from their orbits, significantly less is known
about their rigid behavior at the von Neumann algebraic level. When Γ is in addition properly
proximal, Boutonnet, Ioana and Peterson showed in [BIP18] using boundary techniques [BC14] that
all compact Cartan subalgebras in L∞(X) ⋊ Γ are unitarily conjugate to L∞(X). (For Γ direct
products of nonamenable biexact groups this already follows from the earlier works [CS11, CSU11].)
Consequently, this combined with [Io08b] yields that for any non-commensurable groups Γ and Λ
and any free ergodic profinite actions Γ y X and Λ y Y the von Neumann algebras L∞(X) ⋊ Γ
and L∞(Y ) ⋊ Λ are not isomorphic; remarkably, this is the case for lattices Γ = PSLn(Z) and
Λ = PSLm(Z) for all n 6= m. However, without these additional assumption on Γ, the study of
von Neumann algebraic rigidity aspects for profinite (or compact) actions Γ y X remains an wide
open problem. For example, even establishing strong rigidity results similar to the ones obtained
in [Po04] by Popa for Bernoulli actions of rigid groups seems elusive at this time. While it is very
plausible that such results should hold true, we only have the following partial result at this time in
this direction.
Theorem 1.5. Let Γ and Λ be icc property (T) groups. Let Γ y X = lim←− Xn be a free ergodic
profinite action and let Λ y Y be a free ergodic compact action. Assume that Θ : L∞(X) ⋊
Γ→L∞(Y ) ⋊ Λ is a ∗-isomorphism. Then Λ y Y = lim←− Yn is also a profinite action. Moreover,
there exist l ∈ N and a unitary w ∈ L∞(Y ) ⋊ Λ such that Θ(L∞(Xk+l) ⋊ Γ) = w(L∞(Yk+1) ⋊ Λ)w∗
for every integer k ≥ 0.
This should be compared with Popa's work on inductive limits of II1 factors [Po12]. Finally, the same
strategy used in the proof of Theorem 1.2 can be successfully used in combination with Theorem
5
1.5 to provide a purely von Neumann algebraic approach to a version of Ioana's orbit equivalence
superrigidity theorem from [Io08b]; see the proof of Theorem 5.3.
2 Some preliminaries and technical results
2.1 Popa's intertwining techniques
Over a decade ago, Popa introduced in [Po03, Theorem 2.1 and Corollary 2.3] a powerful analytic
criterion for identifying intertwiners between arbitrary subalgebras of tracial von Neumann algebras.
This is now termed Popa's intertwining-by-bimodules technique.
Theorem 2.1. [Po03] Let (M, τ ) be a separable tracial von Neumann algebra and let P, Q ⊆ M be
(not necessarily unital) von Neumann subalgebras. Then the following are equivalent:
1. There exist p ∈ P(P ), q ∈ P(Q), a ∗-homomorphism θ : pP p → qQq and a partial isometry
0 6= v ∈ qM p such that θ(x)v = vx, for all x ∈ pP p.
2. For any group G ⊂ U(P ) such that G′′ = P there is no sequence (un)n ⊂ G satisfying
kEQ(xuny)k2 → 0, for all x, y ∈ M .
If one of the two equivalent conditions from Theorem 2.1 holds then we say that a corner of P
embeds into Q inside M , and write P ≺M Q.
2.2 Quasinormalizers of von Neumann subalgebras
Given an inclusion N ⊆ M , the quasi-normalizer QN M (N ) is the ∗-subalgebra of M consisting of all
elements x ∈ M such that there exist x1, x2, ..., xk ∈ M satisfying N x ⊆Pi xiN and xN ⊆Pi N xi,
[Po99]. The von Neumann algebra QN M (N )′′ is called the quasi-normalizing algebra of N inside M .
This is an extension of normalization and it is precisely the von Neumann algebraic counterpart of the
notion of commensurator in group theory. As usual, NM (N ) = {u ∈ U(M ) : uN u∗ = N} denotes
the normalizing group and NM (N )′′ denotes the normalizing algebra of N in M . We obviously have
N ⊆ N ∨ N ′ ∩ M ⊆ NM (N )′′ ⊆ QN M (N )′′ ⊆ M . In general the quaisnormalizing algebra is (much)
larger than the normalizer but there are natural instances when they coincide; e.g. when N ⊆ M is a
MASA it was shown in [Po01] that QN M (A)′′ = NM (A)′′. Quasinormalizers play an important role
in the classification of von Neumann algebras and over the last decade there have been a sustained
effort towards computing these algebras in various situations [Po01].
In this subsection we highlight some new computations of quasinormalizers of subalgebras in crossed
products from [CP11] that are essential to deriving our main results from Section 4. If Γ yσ X is a
free ergodic action and M = L∞(X)⋊Γ then QN M (L(Γ))′′ was computed in the following situations.
When Γ is infinite abelian and σ is weak mixing Nielsen observed that L(Γ) is a singular MASA
in M [Ni70]. Later Packer was able to show that the normalizer (and hence the quasinormalizer)
depends only on the discrete spectrum of σ; more precisely one has QN M (L(Γ))′′ = L∞(Xc) ⋊ Γ,
where Γ y Xc is the maximal compact factor of Γ y X [Pa01]. More recently Ioana obtained a
far-reaching generalization of Packer's result by showing that the same holds for every Γ and any
ergodic action σ, [Io08a, Section 6]. In [CP11] this analysis was completed at the entire level of the
distal tower of Γ y X using iterated quasinormalizers.
6
An action Γ y X is called distal if it is the last element of an increasing finite or transfinite sequence
Γ y Xβ of factors β ≤ α, such that Γ y Xo is the trivial factor, each extension π : Xβ+1→Xβ
is maximal compact, and for every limit ordinal β ≤ α the action Γ y Xβ is the inverse limit of
the preceding factors. The sequence {Γ y Xβ}β≤α of factors is also called the Furstenberg-Zimmer
tower of Γ y X. Furstenberg [Fur77] and Zimmer [Z76] independently obtained the following
structure theorem
Theorem 2.2. Let Γ y X be any action. Then there exists an ordinal α and a unique distal tower
{Γ y Xβ}β≤α such that the extension π : X→Xα is weak mixing.
In [CP11] Peterson and the first author obtained a purely von Neumann algebraic way of describing
Furstenberg-Zimmer distal tower of factors for an action, namely as towers of quasinormalizers.
Theorem 2.3. Let Γ yσ X be an ergodic action and let {Γ y Xβ}β≤α be the corresponding
Furstenberg-Zimmer tower. Let M = L∞(X) ⋊ Γ and for all β ≤ α let Mβ = L∞(Xβ) ⋊ Γ be the
corresponding cross-products von Neumann algebras. Then the following hold:
1. for all β ≤ β′ ≤ α we have the following inclusions of von Neumann algebras L(Γ) = Mo ⊆
Mβ ⊆ Mβ′ ⊆ Mα ⊆ M ;
2. for all β ≤ α we have QN M (Mβ)′′ = Mβ+1;
3. for every limit ordinal β ≤ α we have ∪γ<βL∞(Xγ)
Mβ;
W OT
= L∞(Xβ) and also ∪γ<βMγ
W OT
=
4. There exists an infinite sequence (γn)n ⊂ Γ such that for every x, y ∈ L∞(X) ⊖ L∞(Xα) we
have that limn→∞ kEL∞(Xα)(xσγn (y))k2 = 0.
2.3 Finite index inclusions of II1 factors
A trace-preserving action Γ y A on a finite von Neumann algebra is called transitive if A is abelian
and there exist finitely many minimal projections F ⊂ P(A) such that spanF = A and for every
p, q ∈ F there is γ ∈ Γ such that σγ(p) = q. Throughout the paper the set F will be denoted by
At(A) and will be called the atoms of A. In particular all atoms of A have same trace, i.e. dim(A)−1.
Lemma 2.4. Let A be an abelian von Neumann algebra and let Γ yσ A be a trace preserving action.
Assume that the inclusion L(Γ) ⊆ A ⋊ Γ admits a finite Pimsner-Popa basis. Then A is completely
atomic. Moreover, if Γ y A is ergodic then Γ y A is transitive.
Proof. By assumption there exist m1, ..., mk ∈ A ⋊ Γ = M , with EL(Γ)(mim∗
j ) = δi,j pi where
pi ∈ P(M ), such that for all x ∈ M we have x = Pk
i )mi. Thus, for all x ∈ M we
have kxk2
2. Approximating mi ∈ M using their Fourier decompositions and
doing some basic calculations this further implies the following: for every ε > 0 one can find aj ∈ A
with 1 ≤ j ≤ l and c > 0 so that for all x ∈ (M )1 we have
i=1 kEL(Γ)(xm∗
i=1 EL(Γ)(xm∗
2 =Pk
i )k2
kEL(Γ)(xai)k2
2.
(2.3.1)
kxk2
2 ≤ ε + c
l
Xi=1
7
2 ≤ ε + cPl
i=1 kEL(Γ)(unai)k2
2 = ε + cPl
i=1 τ (unaip)2 = 0 we get that τ (p) ≤ ε. Letting ε ց 0 we get p = 0, a contradiction.
Assume for the sake of contradiction that A has a diffuse corner, i.e. there is 0 6= p ∈ A so that
Ap is diffuse. Hence one can find a sequence of unitaries un ∈ U(Ap) so that for all x ∈ Ap we
have τ (unx)→0, as n→∞. Since ai ∈ A we have EL(Γ)(unai) = τ (unai) = τ (unaip). Thus using
(2.3.1) we get that τ (p) = kunk2
i=1 τ (unaip)2 and since
limn→∞Pl
To see the moreover part let 0 6= q ∈ A be a minimal projection of maximal trace. Thus for all
γ ∈ Γ either qσγ(q) = 0 or q = σγ(q). Thus the orbit F = {σγ(q) γ ∈ Γ} is necessarily a finite set
of (orthogonal) minimal projections of A. Let t =Pq∈F q and notice that 0 6= t ∈ A is a projection
satisfying σγ(t) = t for all γ ∈ Γ. Since Γ y A is ergodic it follows that t = 1. Since A is completely
atomic this entails that A = spanF . Thus Γ y A is transitive.
Proposition 2.5. Let Γ, Λ be icc groups and let Γ y A, Λ y B be transitive actions so that
A ⋊ Γ and B ⋊ Λ are II1 factors. Assume that θ : A ⋊ Γ→B ⋊ Λ is a ∗-isomorphism such that
θ(L(Γ)) = L(Λ). Then dim(A) = dim(B) and for every a ∈ At(A), b ∈ At(b) there is a unitary
u ∈ L(Λ) so that θ(L(StabΓ(a)) = u∗L(StabΛ(b))u. In addition, if there exists a ∈ At(A) such that
StabΓ(a) is normal in Γ then for every b ∈ At(B) then StabΛ(b) is also normal in Λ; moreover,
Γ/StabΓ(a) ∼= Λ/StabΛ(b).
Proof. To simplify the presentation, we assume that A⋊Γ = B⋊Λ and L(Γ) = L(Λ). Let n = dim(A)
and fix a ∈ At(A). Notice τ (a) = 1/n and hence EL(Γ)(a) = τ (a)1 = 1/n. Also for each x ∈ L(Γ),
using its Fourier decomposition, we have axa = Pγ∈Γ τ (xu−1
γ )aσγ(a)uγ =
Pγ∈StabΓ(a) τ (xuγ−1)uγa = EL(StabΓ(a))(x)a. Since clearly ∗-alg{a, L(Γ)} = A ⋊ Γ then, altogether,
the above relations show that A ⋊ Γ is the basic construction of the inclusion L(StabΓ(a)) ⊆ L(Γ)
and also [Γ : StabΓ(a)] = [A ⋊ Γ : L(Γ)] = n. A similar statement holds for L(Λ) ⊆ B ⋊ Λ.
Since by assumption [A ⋊ Γ : L(Γ)] = [B ⋊ Λ : L(Λ)] it follows that dim(A) = dim(B) = n. To
show the remaining part of the statement fix b ∈ At(B). By the factoriality assumption, since
τ (a) = τ (b) = 1/n, there is a unitary u ∈ A ⋊ Γ so that
b = uau∗.
γ )auγa = Pγ∈Γ τ (xu−1
(2.3.2)
Since a ∈ A ⋊ Γ is the Jones projection for inclusion L(StabΓ(a)) ⊆ L(Γ), by pull-down lemma
there exists m ∈ L(Γ) such that b = uau∗ = mam∗. Thus one can check that 1/n = τ (b) =
EL(Λ)(b) = EL(Γ)(b) = EL(Γ)(mam∗) = mEL(Γ)(a)m∗ = τ (a)mm∗ = (1/n)mm∗. Hence mm∗ = 1
which implies that m ∈ L(Λ) is a unitary. Thus in equation (2.3.2) we can assume wlog that the
unitary u belongs to L(Γ). Hence using (2.3.2) we further have that L(StabΛ(b)) = {b}′ ∩ L(Λ) =
{uau∗}′ ∩ L(Γ) = {uau∗}′ ∩ uL(Γ)u∗ = uL(StabΓ(a))u∗, as desired. Since StabΓ(a) is normal in
Γ it follows from the above relation that uuγu∗ ∈ NL(Λ)(L(StabΛ(b))) for every γ ∈ Γ. Since Λ is
icc and [Λ : StabΛ(b)] < ∞ then L(StabΛ(b)) ⊆ L(Λ) is a irreducible inclusion of II1 factors. Thus
using [SWW09, Corollary 5.3] we have that for every γ ∈ Γ there exist a unitary x ∈ L(Stabγ(b))
and λ ∈ Λ such that uuγu∗ = xvλ. In particular this implies that StabΛ(b) is normal in Λ and also
Γ/StabΓ(a) ∼= Λ/StabΛ(b).
2.4 Mixing extensions
Let B ⊆ A be an inclusion of von Neumann algebras and assume that Γ yσ A is an action that
leaves the subalgebra B invariant. Throughout the paper we call such a system an extension and
8
we denote it by Γ y (B ⊆ A). When A is endowed with a state φ preserved by σ the extension is
said to be φ-preserving and will be denoted by Γ y (B ⊂ A, φ). When A is a finite von Neumann
algebra and φ is a faithful normal trace then Γ y (B ⊂ A, φ) is called a trace-preserving extension.
Definition 2.6. A trace-preserving extension Γ y (B ⊆ A, τ ) is called mixing if for every t, z ∈
A ⊖ B we have limγ→∞ kEB(tσγ(z))k2 = 0.
Lemma 2.7. Let Γ y (B ⊆ A, τ ) be a trace-preserving mixing extension. Then for every t, z ∈
(A ⋊ Γ) ⊖ (B ⋊ Γ) and every sequence (xn)n ⊂ (L(Γ))1 that converges to 0 weakly, we have
limn→∞ kEB⋊Γ(txnz)k2 = 0.
Proof. Fix t, z ∈ (A ⋊ Γ) ⊖ (B ⋊ Γ). Consider the Fourier decompositions t =Pγ EA(tuγ−1)uγ and
z =Pγ uγ−1EA(uγz) and notice that EB(tuγ) = EB(uγz) = 0 for all γ ∈ Γ. Fix ε > 0. Using these
decompositions and basic k · k2-estimates one can find finite subsets F, G ⊂ Γ such that
kEB⋊Γ(txnz)k2 ≤
ε
2
+ Xδ∈F,λ∈G
kEB⋊Γ(EA(tuδ−1)uδxnuλ−1 EA(uλz))k2.
(2.4.1)
Also fix a, b ∈ A and x ∈ L(Γ). Using the Fourier decomposition of x ∈ L(Γ) we see that
EB⋊Γ(axb) =Pγ τ (xuγ−1 )EB⋊Γ(auγb) =Pγ τ (xuγ−1 )EB(aσγ(b))uγ. Thus we have the formula
(2.4.2)
kEB⋊Γ(axb)k2
τ (xuγ−1 )2kEB(aσγ(b))k2
2.
2 =Xγ
τ (xnuλ−1γ−1δ)2kEB(EA(tuδ−1)σγ(EA(uλz))))k2
Since Γ y (B ⊆ A, τ ) is mixing and F, G are finite one can find a finite subset H ⊂ Γ so
that kEB(EA(tuδ−1 )σγ(EA(uλz))))k2 ≤ ε/(√8FG) for all γ ∈ Γ \ H, δ ∈ F and λ ∈ G.
Also since xn→0 weakly, and F, G, H are finite there is an integer n0 such that τ (xnuγ−1) ≤
ε/(p8HGFktk∞kzk∞) for all γ ∈ G−1H −1F and n ≥ n0. Using these basic estimates in
combination with formula (2.4.2) we see that for all n ≥ n0 we have
Xδ∈F,λ∈G
= Xδ∈F,λ∈G
+ Xγ∈Γ\H
≤ Xδ∈F,λ∈G
This combined with (2.4.1) show that for every ε > 0 there exits n0 such that for all n ≥ n0 we have
kEB(txnz)k2 ≤ ε, as desired.
Theorem 2.8. Let Γ y (B ⊆ A, τ ) be a trace-preserving mixing extension. Also let Λ yα (D ⊆
C, τ ) be a trace-preserving extension for which there exists a ∗-isomorphism θ : A ⋊ Γ→C ⋊ Λ
satisfying θ(B ⋊ Γ) = D ⋊ Λ and θ(L(Γ)) = L(Λ). Then Λ y (D ⊆ C, τ ) is a mixing extension.
kEB⋊Γ(EA(tuδ−1)uδxnuλ−1EA(uλz))k2 =
(Xγ∈H
τ (xnuλ−1γ−1δ)2kEB(EA(tuδ−1)σγ (EA(uλz))))k2
2)
Xγ∈H
ε2
2
8F2G2 kxnk2
8F2G2Hktk2
∞kzk2
∞kEB(tσγ(z))k2
≤(cid:18) ε2
8
ε
2
.
1
2
ε2
8 (cid:19)
+
=
2
1
2
2 +
ε2
1
2
9
Proof. Suppressing θ from the notation we assume that A ⋊ Γ = C ⋊ Λ, B ⋊ Γ = D ⋊ Λ and
L(Γ) = L(Λ). Fix t, z ∈ C ⊖ D. We now show that for any infinite sequence (λn)n ⊆ Λ we have that
(2.4.3)
n→∞kED(tαλn (z))k2 = 0.
lim
Since B ⋊ Γ = D ⋊ Λ we note that
ED(zvλ−1 )vλ = Xλ∈Λ
EB⋊Γ(z) = ED⋊Λ(z) = Xλ∈Λ
Similarly we have EB⋊Γ(t) = 0. Since t, z ∈ C and D ⋊ Λ = B ⋊ Γ we see that
n k2
kED(tαλn (z))k2 = kED⋊Λ(tαλn (z))k2 = kED⋊Λ(tvλn z)vλ−1
ED(EC (zvλ−1 ))vλ = Xλ∈Λ
ED(zEC(vλ−1 ))vλ = ED(z) = 0.
= kED⋊Λ(tvλn z)k2 = kEB⋊Γ(tvλn z)k2.
(2.4.4)
Since (λn)n is infinite the sequence (vλn )n ⊂ L(Λ) = L(Γ) converges weakly to 0. Thus applying
Lemma 2.7 we get limn→∞ kEB⋊Γ(tvλn z)k2 = 0 and hence (2.4.3) follows from (2.4.4).
For further use we recall the following technical variation of [Po03, Theorem 3.1]. The proof is
essentially the same with the one presented in [Po03] and will be left to the reader.
Theorem 2.9. Let Let Γ y (B ⊆ A, τ ) be a trace-preserving mixing extension. Denote by M =
A ⋊ Γ ⊃ B ⋊ Γ = N the corresponding inclusion of crossed product von Neumann algebras. Then
for every von Neumann subalgebra C ⊆ N satisfying C ⊀N B we have QN M (C)′′ ⊆ N .
3 Extensions satisfying the intermediate subalgebra prop-
erty
Let Γ y (P0 ⊆ P ) be an extension of tracial von Neumann algebras and consider the corresponding
inclusion P0 ⋊ Γ ⊆ P ⋊ Γ of von Neumann algebras. Suzuki discovered in [Su18] that if P0, P are
abelian and Γ y P0 is free then the extension Γ y (P0 ⊆ P ) satisfies the intermediate subalgebra
property, i.e. every intermediate subalgebra P0 ⋊ Γ ⊆ N ⊆ P ⋊ Γ arises as N = Q ⋊ Γ for some
Γ-invariant intermediate subalgebra P0 ⊆ Q ⊆ P .
In this section we establish the intermediate
subalgebra property for new classes of extensions (e.g. compact) for icc groups Γ (see Theorem
3.10).
In many respects these results complement Suzuki's as they cover many examples of non
free extensions, for instance when P0 = C1. As a consequence, for all free ergodic pmp actions
on probability spaces Γ y X of icc groups Γ, we are able to completely describe all intermediate
subfactors L(Γ) ⊆ N ⊆ L∞(X) ⋊ Γ with finite index [N : L(Γ)] < ∞ (see Theorem 3.14). Our
strategy also enables us to recover some well-known older results on intermediate subalgebras (see
Corollary 3.3 and Theorems 3.4, 3.7).
We briefly introduce a few preliminaries. The first result describes the algebraic structure of fixed
point subspaces associated with u.c.p. maps and it is essentially [BJKW00, Lemma 3.4]. For reader's
convenience we also include a short proof.
Lemma 3.1. Let M be a von Neumann algebra, and let ϕ be a faithful, normal state on M . Let
Ψ : M → M be a normal, u.c.p. map. Define Har(Ψ) = {m ∈ M : Ψ(m) = m} to be the fixed
points of Ψ. If ϕ ◦ Ψ = ϕ then Har(Ψ) is a von Neumann subalgebra of M.
10
Proof. From the definition it is clear that Har(Ψ) is closed under sum and taking adjoint. Also since
Ψ is normal, Har(Ψ) is closed in the weak-operator topology. Thus, to finish the proof we only need
to show that Har(Ψ) is closed under product. Using the polarization identity, it suffices to show
that whenever x ∈ Har(Ψ) we have that x∗x ∈ Har(Ψ) as well. By Kadison-Schwarz inequality we
have that Ψ(x∗x) ≥ Ψ(x)∗Ψ(x) = x∗x, where the last equality follows because x ∈ Har(Ψ); thus
Ψ(x∗x) − x∗x ≥ 0. Since ϕ ◦ Ψ = ϕ we also have ϕ(Ψ(x∗x) − x∗x) = 0. Since ϕ is faithful, we get
that Ψ(x∗x) = x∗x, thereby proving that Har(Ψ) is an algebra.
Theorem 3.2. Let Γ y (P, τ ) be a trace preserving action on a finite von Neumann algebra P and
consider the corresponding crossed product von Neumann algebra P ⋊Γ. Let P0 ⊆ P be a Γ-invariant
subalgebra. Assume that P0 ⋊ Γ ⊆ N ⊆ P ⋊ Γ is an intermediate von Neumann subalgebra. Then
there is a Γ-invariant subalgebra P0 ⊆ Q ⊆ P so that N = Q ⋊ Γ if and only if EN (P ) ⊆ P .
Proof. Denote by M = P ⋊ Γ and let EP : M→P and EN : M→N be the canonical conditional
expectations onto P and N , respectively. To see the direct implication, fix a ∈ P . Since N = Q ⋊ Γ
and L(Γ) ⊆ N we have
EN (a) =Xγ
=Xγ
=Xγ
EQ(EN (auγ−1))uγ
EQ(EN (a)uγ−1)uγ =Xγ
EQ(auγ−1)uγ =Xγ
EQ(aEP (uγ−1))uγ = EQ(a) ∈ P.
EQ(EP (auγ−1))uγ
EN ◦ EP = EP ◦ EN = EN ∩P .
Next we show the reverse implication. Let eP : L2(M )→L2(P ) and eN : L2(M )→L2(N ) be the
canonical orthogonal projections. Since EN (P ) ⊆ P then EP (EN (a)) = EN (a) for all a ∈ P .
Therefore EP ◦ EN ◦ EP = EN ◦ EP and hence eP eN eP = eN eP . Taking adjoints we obtain
eN eP = eP eN and since (eP eN eP )n converges to eN ∧ eP in the strong-operator topology, as n tends
to infty, we conclude that eP eN = eN eP = eN ∧ eP . This also entails that eN ∧ eP = eN ∩P and thus
(3.0.1)
Alternatively, one can show (3.0.1) just by using Lemma 3.1. Indeed since EN (P ) ⊆ P then from
assumptions EN P : P → P is a u.c.p. map which preserves τ , a normal, faithful, tracial state.
Letting Ψ = EN P we can easily see that N ∩ P ⊆ Har(Ψ) ⊆ EN (P ). Since we canonically have
EN (P ) ⊆ N ∩ P we conclude that Har(Ψ) = EN (P ) = P ∩ N . The last equality gives (3.0.1).
Notice that from assumptions Q := N ∩ P ⊆ P is a Γ-invariant von Neumann subalgebra of P
containing P0. So to finish the proof of our implication we only need to show that N = Q ⋊ Γ.
Since Q ⋊ Γ ⊆ N canonically, we will only argue for the reverse inclusion. To see this fix x ∈ N and
consider its Fourier decomposition (in M ) x = Pγ xγuγ where xγ ∈ P . Since L(Γ) ⊆ N we have
Pγ xγuγ = x = EN (x) = EN (Pγ xγuγ) =Pγ EN (xγ)uγ. By (3.0.1) we have EN (xγ) = EQ(xγ) ∈
Q and hence xγ = EQ(xγ) ∈ Q for all γ ∈ Γ. Thus x =Pγ EQ(xγ)uγ ∈ Q ⋊ Γ, as desired.
The conditional expectation property presented in the previous theorem can be used effectively to
describe all the intermediate subalgebras for many inclusions arising from canonical constructions in
von Neumann algebras. In the remaining part of the section we highlight several situations when this
is indeed the case. For instance it provides a very fast approach to Ge's well known tensor-splitting
theorem [Ge96, Theorem 3.1] for finite von Neumann algebras.
11
Corollary 3.3. ([Ge, Theorem 3.1]) Let P1 be a factor and let P2, N be von Neumann algebras
such that P1 ⊗ 1 ⊆ N ⊆ P1 ¯⊗P2. Assume there exist faithful normal states ϕ1 on P1 and ϕ2 on P2,
and a faithful, normal conditional expectation EN : P1 ¯⊗P2 → N preserving ϕ := ϕ1 ⊗ ϕ2. Then
N = P1 ¯⊗Q for some (von Neumann) subalgebra Q ⊆ P2.
Proof. We first claim that EN (1 ⊗ P2) ⊆ 1 ⊗ P2. To see this fix p2 ∈ P2 and p1 ∈ P1. Since
N ⊃ P1 ⊗ 1 we have (p1 ⊗ 1)EN (1⊗ p2) = EN (p1 ⊗ p2) = EN ((1⊗ p2)(p1 ⊗ 1)) = EN (1⊗ p2)(p1 ⊗ 1).
This implies that EN (1⊗ p2) ∈ (P1 ⊗ 1)′ ∩ (P1 ¯⊗P2) = 1⊗ P2, thereby proving the claim. So we have
that EN : 1⊗ P2 → 1⊗ P2 is a u.c.p. map, preserving ϕ, a faithful, normal state. So by Lemma 3.1,
EN (1 ⊗ P2) is a subalgebra of 1 ⊗ P2, which we can identify as a von Neumann subalgebra Q ⊆ P2.
Under this identification we have that 1 ⊗ Q = EN (1 ⊗ P2) = N ∩ (1 ⊗ P2). Hence P1 ¯⊗Q ⊆ N .
To show the reverse containment, we first claim that (P1⊗alg P2)∩N is WOT-dense in N . Let n ∈ N .
By Kaplansky's density theorem, we can find a bounded net (xλ) ⊂ P1 ⊗alg P2 such that xλ → n in
WOT. Since EN is normal, we get that EN (xλ) → n. If xλ =Pi pi ⊗ qi, with pi ∈ P1 and qi ∈ P2
then EN (xλ) = EN (Pi pi ⊗ qi) =Pi(pi ⊗ 1)EN ((1 ⊗ qi)) ∈ (P1 ⊗alg P2) ∩ N , thereby establishing
the claim. Now fix n ∈ (P1 ⊗alg P2) ∩ N. Then, there exist p1, ..., pk ∈ P1 and q1, ..., qk ∈ P2 so that
n =Pi pi ⊗ qi. Now, since n ∈ N and P1 ⊗ 1 ⊆ N we have
EN (pi ⊗ qi) =
(pi ⊗ 1)EN (1 ⊗ qi).
n = EN (n) = EN (
pi ⊗ qi) =
(3.0.2)
k
Xi=1
k
Xi=1
k
Xi=1
Since EN (1 ⊗ qi) ∈ Q then (3.0.2) implies that (P1 ⊗alg P2) ∩ N ⊆ P1 ⊗alg Q. As (P1 ⊗alg P2) ∩ N
is WOT-dense in N we get N = P1 ¯⊗Q.
We also record a twisted version of the above theorem.
assume that qi are orthogonal with respect to τQ (by using the Gram-Schimdt process, and using the
Theorem 3.4. Let P be a II1 factor and let Q be a finite separable von Neumann algebra, equipped
with a trace preserving action of Γ. Assume that Γ yσ P is an outer action. Then for any
intermediate von Neumann subalgebra P ⋊ Γ ⊆ N ⊆ (P ¯⊗Q) ⋊ Γ there is a von Neumann subalgebra
Q0 ⊆ Q such that N = (P ¯⊗Q0) ⋊ Γ.
Proof. Using Theorem 3.2 we only need to show that EN (Q) ⊆ Q. Naturally, we have that EN (Q) ⊆
P ′ ∩ (P ⊗ Q) ⋊ Γ. We shall now briefly argue that P ′ ∩ (P ¯⊗Q) ⋊ Γ ⊆ Q, which will prove our claim.
To see this fix Pγ aγuγ ∈ P ′ ∩ (P ¯⊗Q) ⋊ Γ, where aγ ∈ P ¯⊗Q. Thus for every γ ∈ Γ and p ∈ P we
have that paγ = aγσγ(p). Fix e 6= γ ∈ Γ. Let aγ = Pi pi ⊗ qi, with pi ∈ P and qi ∈ Q. We may
separability of Q). Thus we havePi(ppi)⊗qi = p(Pi pi⊗qi) = (Pi pi⊗qi)σγ(p) =Pi(piσγ(p))⊗qi.
As qi's are orthogonal we further get ppi = piσγ(p) for all i and p ∈ P . Since Γ y P is outer, this
implies pi = 0 for all i and hence aγ = 0. Thus, P ′ ∩ (P ¯⊗Q) ⋊ Γ ⊆ P ′ ∩ (P ¯⊗Q) = Q. Hence
N = Q0 ⋊ Γ where Q0 = EN (Q).
If P is a II1 factor then an action Γ y P is called centrally free if the induced action Γ y P ′ ∩ P ω is
properly outer (see [Su18, Definition 4.3]). Theorem 3.4 was first obtained by Y. Suzuki under the
assumption that the Γ y P is centrally free, [Su18, Example 4.14]. In general the centrally freeness
assumption introduces certain limitations. For instance, if P = L(F2) then P ′ ∩ P ω = C and hence
no nontrivial group admits a centrally free action on P . However, when P is the hyperfinite II1
factor, then requiring the Γ y P to be outer is the same as requiring the Γ y P to be centrally
12
free. This surprising result is a consequence of Ocneanu's central freedom lemma ([EK98, Lemma
15.25]). The reader may also consult [CD18] for another recent application of the central freedom
lemma.
Theorem 3.5. Let R denote the hyperfinite type II1 factor and let Γ be a discrete group acting on
R. Then Γ yσ R is outer if and only if Γ yσ R is centrally free.
Proof. Let Γ yσ R be an outer action. Let a ∈ R′ ∩ Rω and γ ∈ Γ be such that σg−1 (x)a = ax for
all x ∈ R′ ∩ Rω. This clearly implies that uγa ∈ (R′ ∩ Rω)′ ∩ (R ⋊ Γ)ω. Now, by Ocneanu's central
freedom lemma we get that (R′ ∩Rω)′ ∩ (R ⋊ Γ)ω = R∨ (R′ ∩R ⋊ Γ)ω = R (where the last equality
holds because Γ y R is outer). Thus uγ ∈ R which implies that γ = e. Hence Γ yσ R′ ∩ Rω is
outer.
Conversely, assume that Γ yσ R′ ∩ Rω is outer. We will show that R′ ∩ R ⋊ Γ = C, which shall
establish that Γ yσ R is outer. Let x ∈ R′ ∩ R ⋊ Γ, and consider its Fourier decomposition
x = Pγ xγuγ, where xγ ∈ R. Now x ∈ R′ ∩ R ⋊ Γ implies that xγuγ ∈ R′ ∩ R ⋊ Γ for all γ ∈ Γ.
Hence xγuγ ∈ R ∨ (R′ ∩ R ⋊ Γ)ω = (R′ ∩ Rω)′ ∩ (R ⋊ Γ)ω (where the last equality follows from
Ocneanu's central freedom lemma). Thus we get xγuγx = xxγuγ for all x ∈ R′∩Rω which gives that
γ)x = xER′∩Rω (xγ x∗
xγσγ(x) = xxγ, implying σγ(x)xγ x∗
γ),
for all x ∈ R′ ∩ Rω.
Since Γ y R′ ∩ Rω is outer, we get that ER′∩Rω (xγx∗
γ) = 0 for all γ 6= e. Since ER′∩Rω is faithful,
this further implies that xγ = 0 for all γ 6= e. This implies that x ∈ R′∩R = C, thereby establishing
that R′ ∩ R ⋊ Γ = C, which implies that Γ yσ R is outer.
Theorem 3.4 leads to new examples of subalgebras in (P ¯⊗Q) ⋊ Γ that are amenable relative to
P ⋊ Γ, [OP07, Definition 2.2]. Note that for von Neumann algebra inclusions N ⊆ M , the existence
of a maximal amenable subalgebra P in M relative to N follows from [DHI16, Lemma 2.7]. We
remark that very similar methods were used in [JS19, Theorem 3.4] to provide examples of maximal
Haagerup subalgebras arising from extremely rigid actions of an icc group.
γx, which implies ER′∩Rω (xγx∗
γ = xγx∗
Corollary 3.6. Let P be a type II1 factor, let Q be a finite von Neumann algebra, and let Γ be
an amenable group acting outerly on P, Q (the actions are assumed to be trace preserving). Let
Q0 ⊆ Q be a maximal amenable subalgebra. Then (P ¯⊗Q0) ⋊ Γ is a maximal amenable subalgebra
in (P ¯⊗Q) ⋊ Γ relative to P ⋊ Γ. In particular, if R is the hyperfinite II1 factor, and Γ y R is an
outer trace preserving action, then (R ¯⊗Q0) ⋊ Γ is maximal amenable in (R ¯⊗Q) ⋊ Γ.
Proof. Let (P ¯⊗Q0) ⋊ Γ ⊆ N ⊆ (P ¯⊗Q) ⋊ Γ. Then by Theorem 3.4 N = (P ¯⊗Q1) ⋊ Γ, with
Q0 ⊆ Q1 ⊆ Q. If N is amenable relative to P ⋊ Γ, then Q1 is amenable. By maximal amenability
of Q0 we obtain that Q1 = Q0 thereby establishing the result.
The next theorem re-establishes a well known Galois correspondence for group actions.
Theorem 3.7. Let Γ be a group, let Λ ⊳ Γ be a normal subgroup, and let (P, τ ) be a tracial von
Neumann algebra. Assume that Γ acts on P via trace preserving automorphisms such that (P ⋊ Λ)′∩
(P ⋊ Γ) = C. Then for any intermediate subfactor P ⋊ Λ ⊆ N ⊆ P ⋊ Γ there exists an intermediate
subgroup Λ 6 K 6 Γ such that N = P ⋊ K.
Proof. Let K = {γ ∈ Γ : uγ ∈ N}. Clearly, K is a group satisfying Λ 6 K 6 Γ. Also P ⊆ P ⋊Λ ⊆ N
and hence P ⋊ K ⊆ N ⊆ P ⋊ Γ. Next we show that N ⊆ P ⋊ K.
13
First we claim for every γ ∈ Γ there is cγ ∈ C so that EN (uγ) = cγuγ. Fix γ ∈ Γ and let
ψ(x) = uγxu∗
γ, for all x ∈ L(Γ). Since Λ is normal in Γ, ψ restricts to an automorphism of
P ⋊ Λ. Thus for all x ∈ P ⋊ Λ we have ψ(x)uγ = uγx and hence ψ(x)EN (uγ) = EN (uγ)x. This
implies that EN (uγ)∗ψ(x) = xEN (uγ)∗ and hence EN (uγ)EN (uγ)∗ ∈ (P ⋊ Λ)′ ∩ (P ⋊ Γ) = C. Let
d = EN (uγ)EN (uγ)∗. Note that 0 ≤ d ≤ 1. If d 6= 0, we get that (d−1/2EN (uγ))(d−1/2EN (uγ))∗ = 1,
implying that d−1/2EN (uγ) ∈ U(N ). Next consider u = u∗
γEN (d−1/2uγ) ∈ U(M ). For every
x ∈ P ⋊ Λ we can check that
uxu∗ = d−1u∗
γEN (uγ)xEN (uγ)∗uγ = d−1u∗
γEN (uγ)EN (uγ)∗ψ(x)uγ = u∗
γψ(x)uγ = x.
Hence d−1/2u∗
γEN (uγ) = u ∈ (P ⋊ Λ)′ ∩ (P ⋊ Γ) = C. Thus EN (uγ) = cγuγ for some cγ ∈ C.
The claim shows that for any γ ∈ Γ, either EN (uγ) = 0 or uγ ∈ N . Finally, if N ∋ n =Pγ∈Γ nγuγ
is its Fourier decomposition in P ⋊ Γ, then applying EN , we see that n =Pγ∈Γ nγEN (uγ) ∈ P ⋊ K,
as desired.
Below we highlight a few special cases of the above theorem, which are well known in the literature.
Corollary 3.8.
1. ([Ch78],[ILP98, Theorem 3.13]) Let M be a II1 factor, and let Γ be a discrete
group with an outer action on M . Let N be an intermediate subalgebra, i.e. M ⊆ N ⊆ M ⋊ Γ.
Then there exists a subgroup K of Γ such that N = M ⋊ K.
2. Let Γ be an icc group, and let Λ ⊳ Γ be a normal subgroup such that L(Λ)′ ∩ L(Γ) = C.
Then for any intermediate subfactor L(Λ) ⊆ N ⊆ L(Γ) there exists an intermediate subgroup
Λ 6 K 6 Γ such that N = L(K).
Proof. Since Γ y M is outer, M ′ ∩ M ⋊ Γ = C. Taking Λ = {e}, and appealing to Theorem 3.7
yields the first statement. Taking P = C in Theorem 3.7 yields the second statement.
In the remaining part of the section we show that the strategy presented in Theorem 3.2 can be
successfully used to classify all intermediate subalgebras for inclusion of von Neumann algebras
arising from compact extensions. This covers a new situation which complements the case of free
extensions discovered in [Su18, Main Theorem]. To be able to properly introduce our result we first
recall the following notion of compact extension of actions on von Neumann algebras:
Definition 3.9. Let Γ y (P0 ⊆ P ) be an extension of tracial von Neumann algebras. One says that
Γ y (P0 ⊆ P ) is a compact extension if there exists F ⊆ P satisfying the following properties:
k·k2 = L2(P );
1. spanF
2. for every f ∈ F and ε > 0 there exist ξ1, ξ2, ..., ξn ∈ L2(P ) such that for every γ ∈ Γ one can
find κi(γ) ∈ P0, with i = 1, n satisfying sup1≤i≤n,γ∈Γ kκi(γ)k∞ < ∞ and
kσγ(f ) −
n
Xi=1
κi(γ)ξik2 ≤ ε.
When P0 = C1 we simply say that the action Γ y P is compact.
14
Examples. Assume that Γ y X is an ergodic pmp action on a probability space X and let
Γ y X0 be a factor such that the extension π : X→X0 is compact in the usual sense [Fur77, Z76].
Then it is a routine exercise to show that the corresponding von Neumann algebraic extension
Γ y (L∞(X0) ⊆ L∞(X)) automatically satisfies the definition above.
In particular whenever
Γ y X is an ergodic compact pmp action then Γ y L∞(X) is compact in the above sense.
With this definition at hand we can now introduce the main result of this section.
Theorem 3.10. Let Γ be an icc group and let Γ y (P0 ⊆ P ) be a compact extension of tracial
von Neumann algebras as in Definition 3.9. Let P0 ⋊ Γ ⊆ P ⋊ Γ be the corresponding inclusion
of crossed product von Neumann algebras. Then for any intermediate von Neumann subalgebra
P0 ⋊ Γ ⊆ N ⊆ P ⋊ Γ there exists an intermediate von Neumann subalgebra P0 ⊆ Q ⊆ P such that
N = Q ⋊ Γ.
Proof. Let M = P ⋊ Γ. Denote by EN : M → N the canonical trace preserving conditional
expectation and note that it extends to a map from L2(M ) → L2(M ) by EN ( m) = \EN (m). Sim-
ilarly, let E : M → P be the trace preserving conditional expectation. E also extends to a map
E : L2(M ) → L2(M ). For every ξ ∈ L2(M ) let ξ = EN (ξ) − E ◦ EN (ξ). With these notations at
hand we prove the following
Claim 3.11. for every ξ ∈ F and every ε > 0 there exists a finite set K ⊂ Γ\{e} and η1, η2, ..., ηn ∈
spanP K such that for every γ ∈ Γ there exist κi(γ) ∈ P0 with supγ∈Γ kκi(γ)k∞ < ∞ such that
kσγ( ξ) −Xi
κi(γ)ηik2 ≤ ε.
(3.0.3)
Proof of Claim 3.11. First notice that since L(Γ) ⊆ N and P is Γ-invariant then for all ξ ∈ L2(M )
and γ ∈ Γ we have
EN (uγξu∗
γ) = uγEN (ξ)u∗
γ, and E(uγξu∗
γ) = uγE(ξ)u∗
γ.
(3.0.4)
Fix ξ ∈ F and ε > 0. Since Γ yσ P0 ⊆ P is a compact extension there is a finite set ξ1, ξ2, ..., ξn ∈
L2(P ) such that for every γ ∈ Γ there exist κi(γ) ∈ P0 with supγ∈Γ kκi(γ)k∞ < ∞ so that
kσγ(ξ) −Xi
κi(γ)ξik2 ≤
ε
3
.
(3.0.5)
Using (3.0.4) in combination with (3.0.5) and the basic inequalities kEN (m)k2,kE ◦ EN (m)k2 ≤
kmk2, for all m ∈ L2(M ) we get that
kσγ(EN (ξ)) −Xi
κi(γ)EN (ξi)k2 ≤
ε
3
, and kσγ(E ◦ EN (ξ)) −Xi
κi(γ)E ◦ EN (ξi)k2 ≤
ε
3
Subtracting these relations and using the triangle inequality we conclude that
kσγ( ξ)) −Xi
κi(γ) ξi)k2 ≤
2ε
3
.
15
(3.0.6)
Approximating the ξi's one can find a finite set F ⊂ Γ\{e} so that k ξi−ηik2 ≤ ε/(3n supγ∈Γ kκi(γ)k∞)
for all 1 ≤ i ≤ n. Thus kPi κi(γ) ξi − κi(γ)ηik2 ≤ ε/3 and combining it with (3.0.6) we get the
desired conclusion.
Next we prove the following
Claim 3.12. For every ξ ∈ F we have ξ = 0.
Proof of the Claim 3.12. Fix ε > 0 and ξ ∈ F . Approximating ξ there exists a finite set K ⊆ Γ\{e}
and r ∈ spanP K such that
(3.0.7)
Also by Claim 3.11 there exists a finite set G ⊂ Γ \ {e} and η1, η2, ..., ηn ∈ spanPG such that for
every γ ∈ Γ there exist κi(γ) ∈ P0 with supγ∈Γ kκi(γ)k∞ < ∞ such that
k ξ − rk2 ≤ ε.
(cid:4)
kσγ( ξ) −Xi
κi(γ)ηik2 ≤ ε.
(3.0.8)
Since Γ is icc and G, K ⊂ Γ\ {e} are finite by [CSU13, Proposition 3.4] there exists λ ∈ Γ such that
λKλ−1 ∩ G = ∅; in particular, we have
huλruλ−1 ,Xi
κi(λ)ηii = 0.
(3.0.9)
Using (3.0.7) in combination with Cauchy-Schwarz inequality, (3.0.8), and (3.0.9) we see that
k ξk2
λk2
κi(λ)ηii
2 = huλ ξuλ−1, uλ ξuλ−1i
2 = kuλ ξu∗
= εk ξk2 + huλruλ−1 , uλ ξuλ−1i ≤ εk ξk2 + εkrk2 + huλruλ−1 ,Xi
≤ εk ξk2 + ε(k ξk2 + ε).
Letting ε ց 0 we get ξ = 0, as desired.
Claim 3.12 implies that EN (ξ) = E ◦ EN (ξ) for all ξ ∈ F . Since spanF is dense in L2(P ), these
two maps agree on L2(P ) ⊇ P . Appealing to Theorem 3.2 we conclude that N = Q ⋊ Γ, for some
subalgebra P0 ⊆ Q ⊆ P .
Remarks. After the first draft of the paper appeared on the ArXiv, we were kindly informed by Y.
Jiang and A. Skalski that they had subsequently obtained the same characterization of intermediate
subfactors N satisfying L(Γ) ⊆ N ⊆ L∞(X) ⋊ Γ, with Γ y X a profinite action, in an independent
manner (see [JS19, Corollary 3.11]).
(cid:4)
Corollary 3.13. Let Γ be an icc group and let Γ y A and Γ y B be trace preserving actions with
Γ y B compact, and A is a II1 factor. Consider the diagonal action Γ y A ¯⊗B and let (A ¯⊗B) ⋊ Γ
be the corresponding crossed product von Neumann algebra. Then for any von Neumann subalgebra
A ⋊ Γ ⊆ N ⊆ (A ¯⊗B) ⋊ Γ one can find a Γ-invariant von Neumann subalgebra C ⊆ B so that
N = (A ¯⊗C) ⋊ Γ.
Proof. Since Γ y B is compact one can see that Γ y (A ⊆ A ⋊ B) is a compact extension and
hence the conclusion follows from Theorem 3.10.
16
We end this section with an immediate application of Theorem 3.10 to the study of finite index
subfactors. More specifically, we show that Theorem 3.10 can be used effectively to completely
describe all intermediate subfactors L(Γ) ⊆ N ⊆ L∞(X) ⋊ Γ with [N : L(Γ)] < ∞ for any free
ergodic action Γ y X of any icc group Γ.
Corollary 3.14. Let Γ be an icc group and let Γ y X be a free ergodic action. Let M = L∞(X) ⋊ Γ
denote the corresponding group measure space von Neumann algebra. Then the following hold:
1. Suppose L(Γ) ⊆ N ⊆ L∞(X) ⋊ Γ is an intermediate von Neumann subalgebra so that N ⊆
QN M (L(Γ))′′. Then there exists a factor Γ y X0 of Γ y X such that N = L∞(X0) ⋊ Γ
2. For any intermediate subfactor L(Γ) ⊆ N ⊆ L∞(X) ⋊ Γ with [N : L(Γ)] < ∞ there is a finite,
transitive factor Γ y X0 of Γ y X such that N = L∞(X0) ⋊ Γ; in particular [N : L(Γ)] ∈ N.
Thus for any subfactors L(Γ) ⊆ N1 ⊆ N2 ⊆ L∞(X) ⋊ Γ, with either [N1 : L(Γ)] < ∞ or
Γ y X compact, we have [N2 : N1] ∈ N ∪ {∞}.
3. If Γ has no proper finite index subgroups (e.g. Γ is simple) then there are no nontrivial inter-
mediate subfactors L(Γ) ⊆ N ⊆ L∞(X) ⋊ Γ with [N : L(Γ)] < ∞.
Proof. 1. Let Γ y Xc be a maximal compact factor of Γ y X and using [Io08a, Theorem 6.9] we
have that QN M (L(Γ))′′ = L∞(Xc) ⋊ Γ. Altogether these show that L(Γ) ⊆ N ⊆ L∞(Xc) ⋊ Γ. Then
the desired conclusion follows directly from Theorem 3.10.
2. Since [N : L(Γ)] < ∞ then N admits a finite left (and also a finite right) Pimsner-Popa basis
over L(Γ) and hence N ⊆ QN M (L(Γ))′′. By part 1. there is a factor Γ y X0 of Γ y X such that
N = L∞(X0) ⋊ Γ. As Γ is icc and N is a factor we also have that Γ y L∞(X0) is ergodic. Since N
admits a finite Pimsner-Popa basis over L(Γ) then by Proposition 2.4 it follows that Γ y L∞(X0)
is a transitive action. In particular X0 is a finite probability space and Γ y X0 is transitive. If
Γx 6 Γ is the stabilizer of an x ∈ X0 one can also check that [N : L(Γ)] = X0 = Γ/Γx ∈ N. The
rest of the statement follows easily.
3. Assume that [N : L(Γ)] < ∞. From the proof of part 2. we have N = L∞(X0) ⋊ Γ where Γ y X0
is an action on a finite set X0 and also [N : L(Γ)] = X0 = Γ/Γx where Γx is the stabilizer of
x ∈ X0. Since Γ has no nontrivial finite index subgroups then Γ = Γx and hence N = L(Γ).
Final remarks. The previous corollary also holds for intermediate subalgebras L∞(X) ⋊ Γ ⊆
N ⊂ L∞(Y ) ⋊ Γ with [N : L∞(X) ⋊ Γ] < ∞ for von Neumann algebras arising from extensions
Γ y L∞(X) ⊆ L∞(Y ) of icc groups Γ. The proof is essentially the same as the one presented in
Corollary 3.14 with the only difference that we use Theorem 2.3 instead of [Io08a, Theorem 6.9].
Also, parts 1. and 2. hold for any von Neumann algebra N which admits a finite Pimsner-Popa basis
over L(Γ), if we use the Pimsner-Popa index [PP86] instead of Jones index [Jo81].
In connection with the previous problems one may attempt to describe the subfactors of group von
Neumann algebras N ⊆ L(Γ) that are normalized by the Γ itself, i.e. Γ ⊂ NL(Γ)(N ). Very recently
this problem was considered in [AB19] where a complete description was obtained for Γ lattices in
higher rank simple Lie groups via a noncommutative version of Margulis' normal subgroup theorem;
in turn this was obtained using character rigidity techniques introduced [Pe14, CP13]. In this work
we make further progress on this question for many new families of groups Γ complementary to the
ones from [AB19]. In particular, we show that under additional conditions on the relative commutant
N ′ ∩ L(Γ) (e.g. finite dimensional) these subfactors are always "commensurable" with von Neumann
17
algebras arising from the normal subgroups of Γ (Theorem 3.15). Moreover, in the case of all exact
acylindrically hyperbolic groups [DGO11, O16], all nonamenable groups with positive first L2-Betti
number, and all lattices in product of trees the same holds without any a priori assumptions on
N ′ ∩ L(Γ) (see Theorem 3.16, Corollary 3.17, and part 3 in Theorem 3.15).
Theorem 3.15. Let Γ be a countable discrete group and let N ⊂ L(Γ) be a subfactor such that
Γ ⊂ NL(Γ)(N ). Then there exists a normal subgroup Λ ⊳ Γ such that N ⊆ L(Λ) ⊆ N ∨ (N ′ ∩ L(Γ)).
Moreover, we have the following
1) If N ′ ∩ L(Γ) is finite dimensional then the inclusions N ⊆ L(Λ) ⊆ N ∨ N ′ ∩ L(Γ) have finite
index; in particular, when N ′ ∩ L(Γ) = C1 then N = L(Λ).
2) If L(Γ) is solid1 then either N is an amenable factor or the inclusions N ⊆ L(Λ) ⊆ N ∨ N ′ ∩
L(Γ) have finite index. Moreover if L(Γ) is strongly solid2 then either N is finite dimensional
or the inclusions N ⊆ L(Λ) ⊆ N ∨ N ′ ∩ L(Λ) have finite index.
3) Γ be a simple group such that L(Γ) is a prime factor, e.g. Burger-Mozes group [BM01], Camm's
group [Ca53] or Bhattacharjee's group [Bh94] (see [CdSS18]). Then N is either finite dimen-
sional or [L(Γ) : N ] < ∞.
Proof. Denote by Σ the set of all γ ∈ Γ for which there is y ∈ U(N ) such that τ (yuγ) 6= 0. Note
that Σ coincides with the set of all γ ∈ Γ such that EN (uγ) 6= 0.
Fix γ ∈ Σ and denote by φγ : N→N the automorphism given by φγ(x) = uγxuγ−1 for all x ∈ N .
Thus φγ(x)uγ = uγx and applying the expectation EN we also have φγ(x)EN (uγ) = EN (uγ)x for
all x ∈ N . These two relations give that φγ(x)EN (uγ)uγ−1 = EN (uγ)xuγ−1 = EN (uγ)uγ−1φγ(x)
for all x ∈ N ; in particular, aγ := EN (uγ)uγ−1 ∈ N ′ ∩ L(Γ). Thus
EN (uγ) = aγuγ.
(3.0.10)
Thus EN (uγ)EN (uγ−1) = aγa∗
γ. Applying the expectation EN and using EN ◦ EN ′∩L(Γ) = τ (since
N is a factor) we get EN (uγ)EN (uγ−1) = τ (aγa∗
γ)1. As aγ 6= 0 one can find a unitary bγ ∈ N so that
EN (uγ) = kaγk2bγ. Combining with (3.0.10) we get kaγk2bγ = aγuγ and hence uγ = kaγk2a∗
γbg. In
γ ∈ U(N ′ ∩ L(Γ)) and hence uγ ∈ U(N )U(N ′ ∩ L(Γ)) ⊆ N ∨ (N ′ ∩ L(Γ)).
particular we have kaγk2a∗
Let Λ be the set of all γ ∈ Γ such that uγ = xγyγ, where xγ ∈ U(N ) and yγ ∈ U(N ′ ∩ L(Γ)).
Observe that Λ ⊳ Γ is in fact a normal subgroup. The previous relations show that Σ ⊆ Λ and by
the definition of Σ we have that N ⊆ L(Λ). Since L(Λ) ⊆ N ∨ (N ′ ∩ L(Γ)) canonically, the first part
of the conclusion follows.
Since N ′ ∩ L(Γ) is finite dimensional then N ∨ N ′ ∩ L(Γ) admits left (and right) finite Pimsner-Popa
basis over N and 1) follows.
If N is nonamenable, then N ′∩L(Γ) is finite dimensional, as L(Γ) is solid. The rest of 2) follows easily
from 1). If Γ is simple, then Λ = Γ, as Λ is a normal subgroup of Γ; hence, N ∨ N ′ ∩ L(Γ) = L(Γ).
Since L(Γ) is prime, this further implies that either N or N ′ ∩ L(Γ) is finite dimensional, and thus
3) follows from 1).
Next we show that whenever Γ is a "negatively curved" group then all subfactors N ⊆ L(Γ) nor-
malized by Γ are commensurable to subalgebras L(Λ) arising from normal subgroups Λ ⊳ Γ. Our
1For every diffuse A ⊆ L(Γ) the relative commutant A′ ∩ L(Γ) is amenable
2For every diffuse amenable A ⊆ L(Γ) the normalizer NL(Γ)(A)′′ is amenable
18
proof relies heavily on the deformation/rigidity techniques for array/quasi-cocycles on groups that
were introduced and studied in [CS11, CSU11, CSU13, CKP15]. We advise the reader to consult
these references beforehand.
Let π : Γ→O(H) be an orthogonal representation. Let QH1
as(Γ, π) be the set of all unbounded
quasicocycles into π, i.e. unbounded maps q : Γ→H so that d(q) := supγ,λ∈Γ kq(γλ) − q(γ) −
πγ(q(λ))k < ∞. When the defect d(q) = 0 the set QH1
as(Γ, π) is nothing but the first cohomology
group H 1(Γ, π).
Theorem 3.16. Let π : Γ → O(H) be an orthogonal mixing representation that is weakly con-
tained in the left regular representation of Γ. Assume one of the following holds: a) Γ is exact and
QH1
as(Γ, π) 6= ∅, or b) H 1(Γ, π) 6= 0. Let N ⊆ L(Γ) be a subfactor satisfying Γ ⊂ NL(Γ)(N ). Then
there is a normal subgroup Λ ⊳ Γ so that N ⊆ L(Λ) ⊆ N ∨ N ′ ∩ L(Λ) and one of the following holds:
1. N is finite dimensional, or
2. Λ is infinite amenable, or
3. [L(Λ) : N ] < ∞.
Proof. Let M = L(Γ). By Theorem 3.15 there is Λ ⊳ Γ, such that N ⊆ L(Λ) ⊆ N ∨ (N ′ ∩ M ) and
moreover from its proof it follows that for every γ ∈ Λ there are unitaries aγ ∈ N and bγ ∈ N ′∩L(Λ)
so that
(3.0.11)
uγ = aγbγ.
Also since N is a factor, using Ge's tensor splitting result (Theorem 3.3) we also get that
L(Λ) = N ∨ (N ′ ∩ L(Λ)).
(3.0.12)
Λ
as(Λ, π⊕∞
Assume that Λ is nonamenable. Let q ∈ QH1
as(Γ, π) and consider the restriction qΛ. One can
easily see that the representation π⊕∞
is still mixing and is weakly contained in ℓ2(Λ). Moreover
since Λ ⊳ Γ is normal and the representation is mixing it follows that qΛ is unbounded and hence
qΛ ∈ QH1
Λ ). Thus by [CKP15, Corollary 7.2] it follows that the finite conjugacy radical
F C(Λ) of Λ is finite and hence Z(L(Λ)) is finite dimensional.
Assume that N ′ ∩ L(Λ) is amenable. If it is finite dimensional then (3.0.12) already implies 3. If
not then there is a projection 0 6= z ∈ Z(N ′ ∩ L(Λ)) = Z(Λ) such that (N ′ ∩ L(Λ))z is isomorphic
to the hyperfinite factor. Since Λ is nonamenable N is also nonamenable. Thus L(Λ)z has property
(Gamma) and there is a sequence of (un)n of unitaries in (N ′ ∩ L(Λ))z such that uω := (un)n ∈
(L(Λ)′ ∩ L(Λ)ω)z and uω ⊥ L(Λ); here ω is a free ultrafilter on N. On the other hand using [CSU13,
Theorem 4.1] we get that L(Λ)′ ∩ L(Λ)ω ⊆ L(Λ). Thus u ⊥ u, which is a contradiction.
Now assume that N ′ ∩ L(Λ) is nonamenable. If N is amenable then a similar argument as before
shows that N is finite dimensional leading to 1. Thus for the rest of the proof we assume that N
and N ′ ∩ L(Λ) are nonamenable and we will show this leads to a contradiction.
Let P = L(Λ). Following [CS11, Section 2.3] consider Vt : L2( P ) → L2( P ) be the Gaussian
Λ ) where the supralgebra P ⊂ P
deformation corresponding to the quasicocycle qΛ ∈ QH 1
is the Gaussian dilation. Let eP : P → P denote the orthogonal projection. Since N ′ ∩ L(Λ) is
nonamenable there exists a nonzero projection 0 6= p ∈ N ′ ∩ L(Λ) such that (N ′ ∩ L(Λ))p has
as(Λ, π⊕∞
19
no amenable direct summand. Thus applying a spectral gap argument a la Popa (see for instance
[CS11, Theorem 3.2]), we obtain that
t→0 sup
x∈(N )1p ke⊥
P Vt(x)k2! = 0,
lim
and
t→0
lim
sup
x∈(N ′∩L(Λ)1 ke⊥
P Vt(x)k2! = 0.
(3.0.13)
Fix ε > 0. Thus, using the transversality property from [CS11, Lemma 2.8], relations (3.0.13) and
a simple calculation show that there exist C, D > 0 satisfying
kPBC (x) − xk2 < ε for all x ∈ U(N )p,
kPBD (y) − yk2 < ε for all y ∈ U(N ′ ∩ P ).
(3.0.14)
Here for every constant C ≥ 0 we denoted by BC = {λ ∈ Λ : kq(λ)kH ≤ C} and by PBC the
orthogonal projection onto the Hilbert subspace of L2(Λ) spanned by BC . Since by (3.0.11) we have
uγ = aγbγ then (3.0.14) imply
kPBC (uγb∗
γp) − uγb∗
γpk2 < ε
and
kPBD (b∗
γp) − b∗
γpk2 < ε for allγ ∈ Λ.
Thus using triangle inequality, for all γ ∈ Λ, we also have
(3.0.15)
(3.0.16)
γpk2 < 2ε.
kPBC (uγb∗
γp) − uγPBD (b∗
γp)k2 ≤ kPBC (uγpb∗
γ) − uγpbγk2 + kPBD (b∗
γp) − b∗
Since qΛ is unbounded, there exists γ0 /∈ BC+D+3d(qΛ). Also the quasicocycle relation and the trian-
D ⊆ BC+D+3d(qΛ) and thus γ0 /∈ BC B−1
gle inequality show that BC B−1
D . Hence hPBC (ξ), uγ0 PBD (η)i =
0 for all ξ, η ∈ L2(Λ). Thus using inequalities 3.0.16 for γ = γ0 and (3.0.15) we see that 4ε2 ≥
γ0)k2
2 + kuγ0PBD (pb∗
γp)k2
kPBC (uγ0 b∗
2 −
2 ≤ 3ε2, which contradicts p 6= 0 when ε→0. This completes the proof of
2ε2 = 2kpk2
the first part of the theorem in the the case when q is a quasicocycle with d(q) 6= 0. When d(q) = 0
i.e. q is a cocycle the same proof works with the only difference that to derive the convergence 3.0.14,
instead of using [CS11, Theorem 3.2] (which requires exactness of Γ) one can use the spectral gap
arguments as in [Pe09] or [Va10b].
γ0p) − uγ0PBD (b∗
2− 2ε. Thus kpk2
2 = kPBC (uγ0 b∗
2 ≥ kuγ0b∗
2 + kb∗
γ0p)k2
γ0pk2
γ0pk2
When combined with results in geometric group theory the previous result leads to the following
Corollary 3.17. Let Γ be a nonamenable group that is either exact and acylindrically hyperbolic
or has positive first L2-Betti number. Let N ⊆ L(Γ) be a subfactor such that Γ ⊂ NL(Γ)(N ). Then
there is a nonamenable normal subgroup Λ ⊳ Γ so that N ⊆ L(Λ) ⊆ N ∨ N ′ ∩ L(Λ) and one of the
following holds:
1. N is finite dimensional, or
2. [L(Λ) : N ] < ∞.
as(Γ, ℓ2(Γ)) 6= ∅. Hence
Proof. From [PT11] and [HO11] it follows that these families always have QH 1
the result follows directly from the previous theorem as both classes of nonamenable acylindrically
hyperbolic groups and nonamenable groups with positive first L2-Betti number have finite amenable
radical.
20
4 Actions that satisfy Neshveyev-Størmer rigidity
If Γ, Λ are abelian (or more generally amenable) groups, and Γ y X, Λ y Y are free, ergodic, pmp
actions, then L∞(X) ⋊ Γ and L∞(Y ) ⋊ Λ are isomorphic to the hyperfinite II1 factor R. However,
Neshveyev and Størmer proved that if we assume that Θ : L∞(X) ⋊ Γ → L∞(Y ) ⋊ Λ is an ∗-
isomorphism such that Θ(L∞(X)) is unitarily conjugate to L∞(Y ) and Θ(L(Γ)) = L(Λ) then the
actions Γ y X and Λ y Y are conjugate [NS03, Theorem 4.1]. Motivated by this group action
conjugacy criterion, they further conjectured the following: if Γ, Λ are abelian groups, Γ y X, Λ y Y
are free, weak mixing, pmp actions and Θ : L∞(X) ⋊ Γ → L∞(Y ) ⋊ Λ is a ∗-isomorphism satisfying
Θ(L(Γ)) = L(Λ) then Γ y X is conjugate to Λ y Y [NS03, Conjecture]. Shortly after, using his
influential deformation/rigidity theory Popa was able to prove the following striking result: if Γ, Λ are
any countable groups, Γ yσ X, Λ yρ Y are free, ergodic actions, with σ Bernoulli (or more generally
clustering), and Θ : L∞(X) ⋊ Γ → L∞(Y ) ⋊ Λ is an ∗-isomorphism such that Θ(L(Γ)) = L(Γ) then
Γ y X is conjugate to Λ y Y , [Po04, Theorem 5.2]. In particular, this settled Neshveyev-Størmer
conjecture for Bernoulli actions. Popa also showed that the study of the Neshveyev-Størmer rigidity
question in the context of icc property (T) groups eventually leads to his remarkable proof of the
group measure space version of Connes' rigidity conjecture, [Po04, Theorem 0.1]. All these results
motivate the study of the following generalized Neshveyev-Størmer rigidity question.
Question 4.1. Let Γ and Λ be icc countable discrete groups and let Γ y X and Λ y Y be free,
ergodic, pmp actions. Assume that there is a ∗-isomorphism Θ : L∞(X) ⋊ Γ → L∞(Y ) ⋊ Λ so that
Θ(L(Γ)) = L(Λ). Under what conditions on Γ y X can we conclude that Γ y X and Λ y Y are
conjugate?
Informally, the generalized Neshveyev-Størmer rigidity question asks, under what conditions can
Γ y X be completely recovered from the irreducible subfactor inclusion L(Γ) ⊂ L∞(X) ⋊ Γ.
Using existing literature, one can that the generalized Neshveyev-Størmer rigidity phenomenon holds
for the following classes of actions: all Bernoulli actions of icc groups, [Po04]; all W ∗-superrigid
actions, [Pe09, PV09, CP10, Va10b, Io10, HPV10, CS11, CSU11, PV11, PV12, Bo12, CIK13, CK15,
Dr16]; all weak mixing Cgms-superrigid actions, [Po04, Theorem 5.1]; and all mixing Gaussian actions
[Bo12, Corollary 3.9].
In this section we provide new classes of actions satisfying the generalized Neshveyev-Størmer ques-
tion, most notably, all actions that appear as (nontrivial) mixing extension of free distal actions (see
Theorem 4.8).
4.1 A criterion for conjugacy of group actions
Within the class of icc group, we further generalize Neshveyev-Størmer's aforementioned criterion
for conjugacy of group actions on probability spaces by completely removing the weak mixing as-
sumption of Γ y X (see Theorem 4.5). In this context our result also generalizes [Po04, Theorem
0.7] as it covers many new actions (e.g. compact) that were not previously analyzed in this context.
Our proof relies on the usage of the notion of height of elements in group von Neumann algebras
introduced in [IPV10].
In order to prove our result we need to establish first a few preliminary
technical results on height of elements in group von Neumann algebras, [IPV10, Definition 3.1].
Definition 4.2. A trace preserving action Γ yσ A on a finite von Neumann algebra A is called
properly outer over the the center of A if for every γ 6= 1 and every 0 6= z ∈ Z(A) such that σγ(z) = z
21
the automorphism σγ : Az→Az is not inner. When A is abelian this amounts to the usual freeness
of the action Γ y A.
The following lemma is a basic generalization of Dye's famous result in the case of group measure
space von Neumann algebras. For readers' convenience we include a short proof.
Lemma 4.3. Let Γ yσ A and Λ yα B be properly outer actions. Also let Θ : A ⋊ Γ→B ⋊ Λ
be a ∗-isomorphism such that Θ(A) = B. Fix γ ∈ Γ and let Θ(uγ) = Pλ∈Λ aλvλ be the Fourier
decomposition of Θ(ug) in B ⋊ Λ. Then there are mutually orthogonal projections {eλ}λ∈Λ ⊂ Z(B)
and unitaries {xλ}λ∈Λ ⊂ B so that aλ = eλxλ for all λ ∈ Λ. Also, Pλ∈Λ eλ = 1.
Proof. To ease our presentation we assume that A = B. Thus, M := A ⋊ Γ = A ⋊ Λ. Fix γ ∈ Γ and
let uγ = Pλ∈Λ aλvλ. Since σγ(a)uγ = uγa for all a ∈ A then σγ(a)Pλ∈Λ aλvλ = Pλ∈Λ aλvλa =
Pλ∈Λ aλαλ(a)vλ. Thus σγ(a)aλ = aλαλ(a) for all a ∈ A and λ ∈ Λ. If aλ = wλaλ is the polar
decomposition of aλ this further implies that for all a ∈ A and λ ∈ Λ we have
σγ(a)wλ = wλαλ(a).
(4.1.1)
λeλeµ = xµαµ(a)x∗
µeλeµ. Letting a = αµ−1 (b) we get αλµ−1 (b)eλeµ = x∗
λxµbeλeµx∗
λeλ and σγ(a)eµ = xµαµ(a)x∗
Hence eλ = wλw∗
λ ∈ Z(B). Let xλ ∈ U(A) such that wλ = xλeλ. Fix λ 6= µ ∈ Λ. Using (4.1.1),
for all a ∈ A we have σγ(a)eλ = xλαλ(a)x∗
µeµ. Thus σγ(a)eλeµ =
xλαλ(a)x∗
µxλ for
all b ∈ A. Also one can easily check that αλµ−1 (eλeµ) = eλeµ. Since Λ yα A is properly outer and
λµ−1 6= 1, we get eλeµ = 0; thus for all λ 6= µ we have eλeµ = 0.
As uγ ∈ U(M ) we have 1 =Pλ a∗
λaλ = aλ2 ≤Pλ eλ ≤ 1. Thus aλ2 = eλ and hence aλ = eλ for
all λ ∈ Λ; moreover Pλ eλ = 1.
With this result at hand we are now ready to prove the first technical result needed in the proof of
Theorem 4.5.
Theorem 4.4. Let Γ yσ A and Λ yα B be properly outer actions. Assume that Θ : A ⋊ Γ→B ⋊ Λ
is a ∗-isomorphism satisfying the following conditions:
i) Θ(A) = B, and
ii) there exist 1 > ε > 0 and a finite subset K ⊆ B ⋊ Λ such that for every γ ∈ Γ we have
kΘ(uγ) − Xa,b,c,d∈K
aEL(Λ)(b(Θ(uγ))c)dk2 ≤ ε.
(4.1.2)
Then one can find D > 0 and finite subset F ⊆ B such that for every γ ∈ Γ there exists λ ∈ Λ
satisfying maxb,c∈F τ (b∗Θ(uγ)cvλ) ≥ D > 0.
Proof. As before assume that A = B and notice that M = A ⋊ Γ = A ⋊ Λ. Let 1 > ε > 0 and
K ⊆ M a finite subset such that for all g ∈ Γ we have
kuγ − Xz,t,w,r∈K
zEL(Λ)(tuγw)rk2 ≤ ε.
(4.1.3)
22
Approximating the elements of K via Kaplansky's density theorem we can assume there are finite
subsets F ⊆ A, G ⊆ Λ (some elements could be repeated finitely many times!) so that for all γ ∈ Γ
we have
ε ≥ kuγ − Xa,b,c,d∈F
λ1,λ2,λ3,λ4∈G
avλ1 EL(Λ)(vλ−1
2
b∗uγcvλ3 )vλ−1
4
d∗k2
2.
For the simplicity of writing we convene for the rest of the proof thatP a,b,c,d∈F
for all γ ∈ Γ we have
λ1,λ2,λ3,λ4∈G
=PF,G. Thus
aτ (vλ1λ−1
2
ε ≥ kuγ −Xλ∈ΛXF,G
≥ Xλ∈Λ
(kEA(uγvλ−1 ) −XF,G
b∗uγcvλ3λ−1
4 λ−1 )vλd∗k2
2
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1 )aαλ(d∗)k2
2).
(4.1.4)
By the previous lemma 4.3 we have that EA(uγvλ) = eλxλ with xλ ∈ U(A) and eλ ∈ Z(A). Then
using kf − gk2 ≤ kf − gk2 for f, g ∈ A we see that the last quantity in (4.1.4) is larger than
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1 )aαλ(d∗)k2
2)
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1 )aαλ(d∗)k2
2−
(4.1.5)
(keλ − XF,G
2 + kXF,G
(keλk2
≥ Xλ∈Λ
= Xλ∈Λ
− 2Reτ (eλXF,G
2 = 1. Combining this with
4 λ−1)aαλ(d∗))).
From Lemma 4.3 we also have that Pλ∈Λ eλ = 1 and hence Pλ∈Λ keλk2
(4.1.4) and (4.1.5) we get
b∗uγcvλ3λ−1
τ (vλ1λ−1
2
Xλ∈Λ
εkeλk2
2 ≥Xλ
2 + kXF,G
keλk2
−2Reτ (eλXF,G
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1)aαλ(d∗)k2
2
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1 )aαλ(d∗))).
Hence, for every γ ∈ Γ there exists λ ∈ Λ such that eλ 6= 0 satisfies
εkeλk2
2 ≥ keλk2
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1 )aαλ(d∗)k2
2−
2 + kXF,G
− 2Reτ (eλXF,G
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1 )aαλ(d∗)).
(4.1.6)
23
Using (4.1.6) and the operatorial inequality (Pn
(2 − 2√ε)keλk2
τ (vλ1λ−1
i=1 xi)∗(Pn
b∗uγcvλ3λ−1
i=1 xi) ≤ 2n−1Pn
4 λ−1 )aαλ(d∗)k2
2
2
i=1 x∗
i xi we get
b∗uγcvλ3λ−1
4 λ−1)aαλ(d∗))
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1 )2aαλ(d∗)2)1/2)
(4.1.7)
2
τ (vλ1λ−1
2 + kXF,G
2 ≤ (1 − ε)keλk2
≤ 2Reτ (eλXF,G
≤ 2F 4G4+1Reτ (eλ(XF,G
≤ 2F 4G4+1Reτ ((XF,G
≤ 2F 4G4+1(max
τ (vλ1λ−1
2
b∗uγcvλ3λ−1
4 λ−1 )2kak2
∞kdk2
∞)1/2eλ)
a∈F kak∞)2G4F4
µ∈GG−1λ−1GG−1,b,c∈F τ (b∗uγcvµ)keλk2
2.
max
Letting 0 < D0 := 2F 4G4+1(maxa∈F kak∞)2G4F4 and using that keλk2 6= 0, the previous
equation gives that maxµ∈GG−1λ−1GG−1,b,c∈F τ (b∗uγcvµ) ≥
> 0, which finishes the proof.
1 − √ε
D0
The previous technical result on height can be successfully exploited in combination with some soft
analysis arising from icc property for groups in order to derive the conjugacy criterion for actions.
Theorem 4.5. Let Γ y X and Λ y Y be free ergodic actions where Γ is icc. Assume that
Θ : L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ is a ∗-isomorphism such that Θ(L∞(X)) = L∞(Y ) and there exists a
unitary u ∈ L∞(Y ) ⋊ Λ such that Θ(L(Γ)) = uL(Λ)u∗. Then one can find x ∈ NL∞(Y )⋊Λ(L∞(Y )),
a character η : Γ→T, and a group isomorphism δ : Γ→Λ such that xu ∈ U(L(Λ)) and for all
a ∈ L∞(X), γ ∈ Γ we have
(4.1.8)
Here {uγ}γ∈Γ and {vλ}λ∈Λ are the canonical group unitaries implementing the actions in L∞(X)⋊Γ
and L∞(Y ) ⋊ Λ, respectively.
In particular, it follows that Γ y X is conjugate to Λ y Y .
Θ(auγ) = η(γ)Θ(a)x∗vδ(γ)x.
Proof. For the ease of presentation we first introduce some notations. After suppressing Θ from the
notation we assume that A = L∞(X) = L∞(Y ) and hence M = A ⋊σ Γ = A ⋊α Λ. Also letting
C = uAu∗ and Λ1 = uΛu∗ we also have M = C ⋊α′ Λ1 and L(Γ) = L(Λ1). Throughout the proof
we denote by tλ = uvλu∗ and α′
λ(c) = tλctλ−1 for all c ∈ C. Note that the condition ii) in Theorem
4.4 is automatically satisfied and hence by the conclusion of Theorem 4.4 there exists a D > 0 and
a finite subset F ⊂ A so that for every γ ∈ Γ, there is λ ∈ Λ such that
D ≤ max
b,c∈F τ (b∗uγcvλ) = max
b,c∈F τ ((ub∗)uγ(cu∗)uvλu∗) = max
b,c∈F τ ((ub∗)uγ(cu∗)tλ).
(4.1.9)
Approximating b∗u, u∗c in (4.1.9) via Kaplansky's theorem with elements in C ⋊alg Λ1 and then
diminishing D if necessary, we can in fact assume the following: there exists D > 0 and K ⊂ C
finite, such that for every γ ∈ Γ, there exists λ ∈ Λ1 satisfying
d,e∈K τ (duγetλ) ≥ D.
max
(4.1.10)
24
On the other hand since EC (x) = τ (x)1 for all x ∈ L(Λ1) we can see that
d,e∈K τ (duγ etλ) = max
max
= max
d,e∈K τ (duγ tλα′
d,e∈K τ (dEA(uγtλ)α′
d∈K kdk2
∞τ (uγtλ).
≤ max
λ−1 (e)) = max
d,e∈K τ (EC (duγtλα′
λ−1 (e)))
λ−1 (e)) = max
d,e∈K τ (uγvλ)τ (dα′
λ−1 (e))
Combining this with (4.1.10), for every γ ∈ Γ there exists λ1 ∈ Λ1 such that τ (uγtλ1 ) ≥
maxd∈K kdk2
0. Since L(Γ) = L(Λ1) and hΛ1 (Γ) > 0 then by [IPV10, Theorem 3.1] there is w ∈ U(L(Λ1)), a
character η : Γ→T, and a group isomorphism δ1 : Γ→Λ1 satisfying wuγw∗ = η(γ)tδ1(γ). Since
tλ = uvλu∗, letting x = u∗w, we further get that there is a group isomorphism δ : Γ→Λ satisfying
(4.1.11)
∞
xuγx∗ = η(γ)vδ(γ), for all γ ∈ Γ.
D
>
As vλAvλ−1 = A, using (4.1.11) we get xuhx∗Axuh−1 x∗ = A for all h ∈ Γ. Fix arbitrary
a ∈ A and note uhx∗axuh−1 = x∗EA(xuhx∗axuh−1x∗)x. Applying the expectation we also have
uhEA(x∗ax)uh−1 = EA(x∗EA(xuhx∗axuh−1x∗)x). Subtracting these relations, for every h ∈ Γ we
have
uh(x∗ax − EA(x∗ax))uh−1 = x∗EA(xuhx∗axuh−1 x∗)x − EA(x∗EA(xuhx∗axuh−1x∗)x).
(4.1.12)
Fix ε > 0. By Kaplansky Density Theorem there exist finite subsets K ⊂ Γ \ {1}, L ⊂ Γ and
elements yK ∈ spanAK and xL ∈ spanAL such that
kyKk∞ ≤ 2,
kx∗ax − EA(x∗ax) − yKk2 ≤ ε,
kxLk∞ ≤ 1
kx − xLk2 ≤ ε
(4.1.13)
Using (4.1.12) and (4.1.13) together with basic calculations we see that for every h ∈ Γ we have
2 = kuh(x∗ax − EA(x∗ax))uh−1k2
kx∗ax − EA(x∗ax)k2
= huh(x∗ax − EA(x∗ax))uh−1 , x∗EA(xuhx∗axuh−1x∗)x − EA(x∗EA(xuhx∗axuh−1 x∗)x)i
≤ 2ε + huhyKuh−1, x∗EA(xuhx∗axuh−1x∗)x − EA(x∗EA(xuhx∗axuh−1x∗)x)i
≤ 2ε + huhyKuh−1, x∗EA(xuhx∗axuh−1x∗)xi
≤ 4ε + huhyKuh−1, xL
∗EA(xuhx∗axuh−1x∗)xLi
2
(4.1.14)
Since Γ is icc and K ⊂ Γ \ {1}, L ⊂ Γ are finite then by [CSU13, Proposition 2.4] there is h ∈ Γ
so that hKh−1 ∩ L−1L = ∅. Hence huhyKuh−1, xL
∗EA(xuhx∗axuh−1x∗)xLi = 0 and using (4.1.14)
we conclude that kx∗ax − EA(x∗ax)k2 ≤ 4ε. Since this holds for all ε > 0 then x∗ax = EA(x∗ax)
for all a ∈ A. Therefore x∗Ax ⊆ A and since A is a MASA we obtain x∗Ax = A; thus x ∈ NM (A).
This together with (4.1.11) give (4.1.8). In addition, for every a ∈ A and γ ∈ Γ we have xσγ(a)x∗ =
xuγauγ−1x∗ = vδ(γ)xax∗vδ(γ)−1 = αδ(γ)(xax∗); in particular Γ y X and Λ y Y are conjugate.
25
Remarks. The Theorem 4.5 actually holds in a greater generality, namely, for all actions Γ y A,
Λ y B that are properly outer over the center. The proof is essentially the same with the one
presented above. We highlighted only the more particular case of free ergodic actions solely because
this is what we will mainly use to derive the main results of this section.
4.2 Applications to the generalized Neshveyev-Størmer rigidity question
In this subsection we show that large families of group actions verify the conjugacy criterion presented
in Theorem 4.5 and therefore will satisfy the generalized Neshveyev-Størmer rigidity question. Our
examples appear as mixing extensions of free distal actions. Our method of proof rely on combining
the persistence of mixing through von Neumann equivalence from Section 2.4 and the von Neumann
algebraic description of compactness using quasinormalizers from [Ni70, Pa01, NS03, Io08a, CP11].
Theorem 4.6. Let Γ y X be a ergodic pmp action whose distal quotient Γ y Xd is free and the
extension π : X → Xd is nontrivial and mixing. Let Λ y Y be an ergodic pmp action whose distal
quotient Λ y Yd is also free. Assume that Θ : L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ is a ∗-isomorphism such that
Θ(L(Γ)) = L(Λ). Then there exists a unitary u ∈ L∞(Yd) ⋊ Λ such that Θ(L∞(Xd)) = uL∞(Yd)u∗
and Θ(L∞(X)) = uL∞(Y )u∗.
Proof. To ease our presentation we assume that M := L∞(X) ⋊ Γ = L∞(Y ) ⋊ Λ with P = L(Γ) =
L(Λ). Using Theorem 2.3 it follows that N := L∞(Xd) ⋊ Γ = L∞(Yd) ⋊ Λ. Next we argue that
L∞(Yd) ≺N L∞(Xd). Indeed, if we assume L∞(Yd) ⊀N L∞(Xd), since the extension π : X→Xd
is assumed to be mixing, by Theorem 2.9 we have that QN M (L∞(Yd))′′ ⊆ N . However since
QN M (L∞(Yd))′′ = M it would imply that M ⊆ N which is a contradiction.
Since Γ y Xd and Λ y Yd are free and L∞(Yd) ≺N L∞(Xd) then by [Po01, Appendix A] one can
find a unitary u ∈ N so that L∞(Xd) = uL∞(Yd)u∗. Passing to relative commutants and using
freeness of Γ y Xd, Λ y Yd again we also get L∞(X) = uL∞(Y )u∗, as desired.
Theorem 4.7. Let Γ be an icc group and let Γ y X be an ergodic pmp action whose distal quotient
Γ y Xd is free and the extension π : X → Xd is nontrivial and mixing. Let Λ y Y be any
free ergodic pmp action. Assume that Θ : L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ is a ∗-isomorphism such that
Θ(L(Γ)) = L(Λ). Then there exists a unitary u ∈ L∞(Yd) ⋊ Λ such that Θ(L∞(Xd)) = uL∞(Yd)u∗
and Θ(L∞(X)) = uL∞(Y )u∗.
Proof. As before we assume that M := L∞(X)⋊ Γ = L∞(Y )⋊ Λ with L(Γ) = L(Λ). Using Theorem
2.3 it follows that N := L∞(Xd) ⋊ Γ = L∞(Yd) ⋊ Λ. Next we argue that L∞(Xd) ≺N L∞(Yd).
First notice since π : X→Xd is mixing it follows from Theorem 2.8 that π : Y →Yd is also mixing.
If we would have L∞(Yd) ⊀N L∞(Xd), since the extension π : Y →Yd is mixing, then Theorem 2.9
would imply that QN M (L∞(Xd))′′ ⊆ N . However since QN M (L∞(Xd))′′ = M it would imply that
M ⊆ N which is a contradiction.
Notice that since Γ is icc and L(Γ) = L(Λ) it follows that Λ is icc as well. Also since L∞(Xd) ≺N
L∞(Yd) and L∞(X) is a Cartan subalgebra in M it follows from [OP07, Lemma 4.1] that Λ y Yd
is free and then the desired conclusion follows from Theorem 4.6.
Remarks. 1) If in the statements of Theorems 4.6 and 4.7 one only requires that the distal factor
Γ y Xd is actually compact, then in the proof of Theorem 4.6 we don't need to use Theorem 2.3.
Instead one can just directly apply [Io08a, Proposition 6.10].
26
2) If in the statement of Theorem 4.7 one requires that the first element Γ y X0 of the distal
tower Γ y Xd is free profinite then one can show the action Λ y Yd is free without appealing to
[OP07, Lemma 4.1]. Briefly, using the mixing we have L∞(X0) ≺M L∞(Y ) and employing some
basic intertwining properties one can further show that L∞(X0) ≺L∞(X0)⋊Γ L∞(Y0) and hence
L∞(Y0)′ ∩ (L∞(Y0) ⋊ Λ) ≺L∞(Y0)⋊Λ L∞(X0) (∗). However using the same calculations from the
proof of part 2. in Theorem 4.12 we have L∞(Y0)′ ∩ (L∞(Y0) ⋊ Λ) = L∞(Y0) ⋊ Σ for some normal
subgroup Σ ⊳ Λ. However since L(Σ) ⊆ L(Γ) the the intertwining (∗) implies that Σ is finite and
since Λ is icc we further have Σ = 1; hence Λ y Y0 must be free.
Combining the previous theorems with Theorem 4.5 we obtain the following
Theorem 4.8. Let Γ be an icc group and let Γ y X be a free, ergodic pmp action whose distal
quotient Γ y Xd is free and the extension π : X → Xd is nontrivial and mixing. Let Λ y Y be any
free ergodic pmp action. Assume that Θ : L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ is a ∗-isomorphism such that
Θ(L(Γ)) = L(Λ). Then there exists y ∈ U(L(Λ), ω : Γ → T a character, and δ : Γ → Λ a group
isomorphism such that yΘ(L∞(X))y∗ = L∞(Y ), and for all a ∈ L∞(X), γ ∈ Γ, we have
Θ(auγ) = ω(γ)Θ(a)y∗vδ(γ)y.
In particular, we have yΘ(σγ(a))y∗ = αδ(γ)(yΘ(a)y∗) and hence Γ y X and Λ y Y are conjugate.
Here {uγ}γ∈Γ and {vλ}λ∈Λ are the canonical group unitaries implementing the actions in L∞(X)⋊Γ
and L∞(Y ) ⋊ Λ, respectively.
Proof. To ease our presentation, we assume, as before, that L∞(X) ⋊ Γ = L∞(Y ) ⋊ Λ, and L(Γ) =
L(Λ). Theorem 4.7 yields that there is a unitary u ∈ L∞(Yd) ⋊ Λ such that L∞(X) = uL∞(Y )u∗.
This is equivalent to assuming that L∞(X)⋊Γ = L∞(Y )⋊Λ, L(Γ) = uL(Λ)u∗ and L∞(X) = L∞(Y ).
We are now exactly in the set up of Theorem 4.5, which yields the desired conclusions.
Examples. Theorem 4.8 implies that if Γ be an icc group and Γ y X is any ergodic pmp action
that admits a free profinite quotient Γ y Xd and the extension π : X → Xd is nontrivial and mixing
then Γ y X satisfies Neshveyev-Størmer rigidity question. For instance if Γ is icc residually finite
then this is the case for any diagonal action Γ y Z × T where Γ y Z is a Gaussian action associated
to a mixing orthogonal representation of Γ and Γ y T is any free ergodic profinite action.
Corollary 4.9. Let Γ be an icc group, let Γ y X be a free, mixing pmp action and let Λ y Y be
any free ergodic pmp action. Also let Γ y X0 be a free factor of Γ y X and Λ y Y0 be a factor of
Λ y Y . Assume that Θ : L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ is a ∗-isomorphism such that Θ(L(Γ)) = L(Λ)
and Θ(L∞(X0) ⋊ Γ) = L∞(Y0) ⋊ Λ. Then there exists y ∈ U(L(Λ), ω : Γ → T a character, and
δ : Γ → Λ a group isomorphism such that yΘ(L∞(X))y∗ = L∞(Y ), and for all a ∈ L∞(X), γ ∈ Γ,
we have
Θ(auγ) = ω(γ)Θ(a)y∗vδ(γ)y.
In particular, we have yΘ(σγ(a))y∗ = αδ(γ)(yΘ(a)y∗) and hence Γ y X and Λ y Y are conjugate.
Here {uγ}γ∈Γ and {vλ}λ∈Λ are the canonical group unitaries implementing the actions in L∞(X)⋊Γ
and L∞(Y ) ⋊ Λ, respectively.
Proof. Since Γ y X is mixing then by so is Λ y Y . In particular the extensions Γ y (L∞(X0) ⊂
L∞(X)) and Γ y (L∞(X0) ⊂ L∞(X)) are mixing. Since Θ(L∞(X0) ⋊ Γ) = L∞(Y0) ⋊ Λ the
conclusion follows using the same arguments as in the proof of Theorem 4.8.
27
Following the terminology from [PV09] a free ergodic action Γ y X is called Cgms-superrigid if up
to unitary conjugacy L∞(X) ⊂ L∞(X) ⋊ Γ = M is the only group measure space Cartan subalgebra
of M . Over the last decade many classes of examples of such actions have been discovered via
deformation/rigidity theory. For some concrete examples the reader is referred to [OP07, CS11,
CSU11, PV11, PV12, Io12, CIK13] and the survey [Io18]. An immediate consequence of [Po04,
Theorem 5.1] is that all weakly mixing Cgms-superrigid actions satisfy the statement of Theorem 4.8.
Using our Theorem 4.5 we obtain the following generalization
Corollary 4.10. Any Cgms-superrigid action Γ y X of any icc group Γ satisfies the statement of
Theorem 4.8.
In particular the generalized Neshveyev-Størmer rigidity holds for all action Γ y X of icc groups
Γ that are: hyperbolic groups, [PV12], free products [Io12] or finite step extensions of such groups
[CIK13].
At this point it is increasingly evident that all the above Neshveyev-Størmer type rigidity results
were achieved by heavily exploiting, at the von Neumann algebra level, the natural tension between
mixing and compactness properties for action. It would be interesting to understand whether such
results could still be obtained only in the compact regime. Specifically, we would like to propose for
study the following
Problem 4.11. If Γ is icc does every free ergodic profinite action Γ y X satisfy the statement of
Theorem 4.8?
While providing a complete answer to this question seems hard at the moment, one can show there
are many aspects of Γ y X that are shared by Λ y Y through this equivalence (e.g. compactness,
profiniteness, etc). In fact we have the following result.
Theorem 4.12. Let Γ y X be a free ergodic action and let Λ y Y be any action. Let Θ :
L∞(X) ⋊ Γ→L∞(Y ) ⋊ Λ be a ∗-isomorphism such that Θ(L(Γ)) = L(Λ). Then the following hold
1. If Γ y X is (weakly) compact then Λ y Y is also (weakly) compact.
2. If Γ is icc and Γ y X is profinite then Λ y Y is also ergodic and profinite. Specifically, if
Γ y X is the inverse limit of Γ y Xn with Xn finite then Λ y Y is the inverse limit of
Γ y Yn with Yn finite so that for every n we have Θ(L∞(Xn) ⋊ Γ) = L∞(Yn) ⋊ Λ. In addition,
the stabilizer StabΛ(Yn) ⊳ Λ is normal and we have that Γ/StabΓ(Xn) ∼= Λ/StabΛ(Yn) for all
n. Finally, there exists a normal subgroup Σ ⊳ Λ so that L∞(Y )′ ∩ L∞(Y ) ⋊ Λ = L∞(Y ) ⋊ Σ.
Proof. 1. As before we assume that L∞(X) ⋊ Γ = L∞(Y ) ⋊ Λ = M and L(Γ) = L(Λ). Since Γ y X
is compact, [Io08a, Theorem 6.10] implies that the quasinormalizer algebra satisfies QN M (L(Γ))′′ =
M . Since canonically QN M (L(Γ))′′ = QN M (L(Λ))′′ then QN M (L(Λ))′′ = M which by [Io08a,
Theorem 6.10] again implies that Λ y Y is also compact. The statement on weak compactness
follows from [OP07, Proposition 3.2].
2. Since Γ is icc then Λ is also icc. Hence Λ y Y is ergodic (otherwise M will not be a factor).
Next we show that Λ y Y is profinite. As Γ y X is profinite, it is the inverse limit of ergodic
actions Γ y Xn on finite spaces. Thus An = L∞(Xn) form a tower of finite dimensional abelian
Γ-invariant subalgebras A0 ⊂ ... ⊂ An ⊂ An+1 ⊂ ... ⊂ L∞(X) such that ∪nAn
= L∞(X).
Moreover Γ y An is transitive for every n. Since L∞(X0) ⋊ Γ = L∞(Y0) ⋊ Λ and L(Γ) = L(Λ)
SOT
28
SOT
i 1 ≤ i ≤ kn} = At(Bn). Since StabΓ(q) is assumed normal in Γ for every q ∈ At(An)
i ) is normal in Λ for every i. Moreover, since the
i ) = StabΛ(Bn) for all
using Theorem 3.10 for every n one can find a Λ-invariant subalgebra Bn ⊂ L∞(Y0) such that
An ⋊ Γ = Bn ⋊ Λ. Factoriality of An ⋊ Γ and Λ being icc imply that the action Λ y Bn is ergodic.
Since L(Γ) ⊆ An ⋊ Γ is a finite index inclusion of II1 factors so is L(Λ) ⊆ Bn ⋊ Λ. Using Lemma
2.4 we get that Bn is finite dimensional and the action Λ y Bn is transitive. One can easily check
that B0 ⊂ ... ⊂ Bn ⊂ Bn+1 ⊂ ... ⊂ L∞(Y0) and also ∪nBn
= L∞(Y0). Thus there exist factors
Λ y Yn of Λ y Y with Yn finite such that Λ y Y is the inverse limit of Λ y Yn.
Denote by {pn
it follows from Proposition 2.5 that StabΛ(pn
action Λ y Bn is transitive one can easily see that we actually have StabΛ(pn
1 ≤ i ≤ kn. Finally, by Proposition 2.5 we also have that Γ/StabΓ(Xn) ∼= Λ/StabΛ(Yn) for all n.
In the remaining part we describe the relative commutant L∞(Y )′ ∩ M . So fix b ∈ L∞(Y )′ ∩ M
and consider its Fourier decomposition b =Pλ∈Λ bλvλ. Since b commutes with L∞(Y ) we get that
ybλ = αλ(y)bλ for all λ ∈ Λ and y ∈ L∞(Y ). Letting eλ be the support projection of bλb∗
λ this
further implies that for all y ∈ L∞(Y0) and λ ∈ Λ we have
yeλ = αλ(y)eλ.
Fix λ such that bλ 6= 0 (and hence eλ 6= 0). Denote by en
expectation EBn in (4.2.1), for all y ∈ Bn we have
λ := EBn (eλ) and applying the conditional
(4.2.1)
yen
λ 6= 0 and hence there is pn
λ = αλ(y)en
λ.
i ∈ At(Bn) satisfying en
λpn
(4.2.2)
i we get ypn
Since eλ 6= 0 then en
i for some scalar c > 0.
i ) = pn
Multiplying (4.2.2) by c−1pn
i
and hence λ ∈ StabΛ(pn
i ) = StabΛ(Bn). Altogether, we have shown that for every λ with bλ 6= 0
we have λ ∈ StabΛ(Bn). Applying this for every n we conclude that λ ∈ ∩nStabΛ(Bn) =: Σ. In
particular b ∈ L∞(Y0) ⋊ Σ and hence L∞(Y )′ ∩ M ⊆ L∞(Y ) ⋊ Σ. Since the reverse containment
canonically holds we get L∞(Y )′ ∩ M = L∞(Y ) ⋊ Σ. As StabΛ(Bn)'s are normal in Λ then Σ is also
normal in Λ.
for all y ∈ Bn. This entails that αλ(pn
i = αλ(y)pn
i
i = cpn
These results can be used to produce additional examples of actions satisfying the statement of
Theorem 4.8. For example part 1. of the previous theorem in combination with Theorem 4.5 and
[BIP18, Theorem 4.16], [CPS12, Theorem 5.1] shows that any free ergodic weakly compact action
Γ y X satisfies the generalized Neshveyev-Størmer rigidity whenever Γ is an icc group in one of the
following classes
1. Γ is any properly proximal group [BIP18, Definition4.1], in particular when Γ = P SLn(Z),
n ≥ 2 or any Γ that admits a proper array into a nonamenable representation (see [CS11,
Definition 2.1]). In fact the latter also follows by using the results in [CS11, CSU11];
2. Γ = H ≀ G is a wreath product where H is nontrivial abelian and G nonamenable [CPS12].
5 Some applications to strong rigidity results in von Neu-
mann algebras and orbit equivalence
Theorem 5.1. Let Γ and Λ be icc property (T) groups. Let Γ y X = lim Xn be a free ergodic
profinite action and let Λ y Y be a free ergodic compact action. Assume that Θ : L∞(X) ⋊
29
Γ→L∞(Y ) ⋊ Λ is a ∗-isomorphism. Then Λ y Y = lim Yn is a profinite action. Moreover there
exists l ∈ N and a unitary w ∈ L∞(Y ) ⋊ Λ such that Θ(L∞(Xk+l) ⋊ Γ) = w(L∞(Yk+1) ⋊ Λ)w∗ for
every positive integer k.
.
SOT
Proof. To simplify the notations let A = L∞(X), B = L∞(Y ) and notice that M = A ⋊ Γ = B ⋊ Λ.
Moreover if L∞(Xn) = An. then An ⊆ An+1 and A = ∪nAn
. Also if Mn = An ⋊ Γ then
Mn ⊆ Mn+1 and M = ∪nMn
Since L(Λ) ⊂ M is rigid subalgebra and M has Haagerup property relative to L(Γ) it follows that
L(Λ) ≺M L(Γ). Hence one can find nonzero projections p ∈ L(Λ), q ∈ L(Γ) a nonzero partial
isometry v ∈ M and an injective ∗-homomorphism φ : pL(Λ)p→rL(Γ)r satisfying φ(x)v = vx for
all x ∈ pL(Λ)p. Since L(Λ) ⊂ M is a irreducible subfactor we have that v∗v = p. Denoting by
Q = φ(pL(Λ)p) we also have that r = vv∗ ∈ Q′ ∩ qM q. Letting u ∈ M a unitary such that uv∗v = v
we have that
(5.0.1)
upL(Λ)pu∗ = Qr.
SOT
Next we prove the following
Claim 5.2. Q ⊆ qL(Γ)q is a finite index subfactor.
Proof of Claim 5.2. Since L(Γ) ⊂ M is a rigid subalgebra and M has Haagerup's property relative
to L(Λ) we also have that L(Γ) ≺M L(Λ). Since L(Λ) is a factor this further entails that L(Γ) ≺M
upL(Λ)pu∗ = Qr. Hence by Popa's intertwining techniques there exist finitely many xj , yj ∈ M
and c > 0 such that Pj kEQr(xj uyj)k2
2 ≥ c for all u ∈ U(L(Γ)). Since EQ(r) = τq(r)q then we
have EQr(x) = EQ(r)−1EQ(qxq)r = τq(r)−1EQ(qxq)r for all x ∈ L(Γ). Using this formula in the
previous inequality we further get that Pj kEQ(qxjuyjq)k2
2 ≥ cτq(r) > 0. Approximating xj and yj
with their Fourier decompositions one can find finitely many ai, bi ∈ A and γi, δi ∈ Γ such that for all
u ∈ U(L(Γ)) we havePi kEQ(quγiaiubiuδi q)k2
. Using this together with EQ = EQ◦EqL(Γ)q
we see that for all γ ∈ Γ we have
2 ≥ cτq(r)
2
cτq(r)
2 ≤
kEQ(quγiaiuγbiuδiq)k2
2 =
s
Xi=1
kEQ(quγiaiσγ(bi)uγδiq)k2
2
s
s
s
Xi=1
Xi=1
i=1,s kaik2
≤ (max
=
kEQ(qEL(Γ)(uγiaiσγ(bi)uγδi)q)k2
2 =
τ (aiσγ(bi))2kEQ(quγiγδiq)k2
2
Xi=1
∞kbik2
∞)(Xi
kEQ(quγiγδiq)k2
2).
cτq(r)
∞kbik2
∞
2 maxi=1,s kaik2
we have thatPi kEQ(quγiγδiq)k2
2 ≥ d > 0 for all γ ∈ G. Hence
Thus letting d =
by Theorem 2.1 we get L(Γ) ≺L(Γ) Q and since L(Γ) is a II1 factor we actually have qL(Γ)q ≺L(Γ) Q
and hence qL(Γ)q ≺qL(Γ)q Q. In particular this entails [qL(Γ)q : Q] < ∞ and the claim follows. (cid:4)
Combining the Claim 5.2 with [Po02, Lemma 3.1] it follows the inclusion qL(Γ)q′∩ qM q ⊆ Q′∩ qM q
has finite Pimsner-Popa probabilistic index. Since Γ is icc and Γ y X is free it follows that
L(Γ)′ ∩ M = C and thus qL(Γ)q′ ∩ qM q = Cq. Combining with the above we conclude that
Q′ ∩ qM q is a finite dimensional von Neumann algebra. Since Q ⊆ qL(Γ)q ⊂ qM1q ⊂ ... ⊂ qMnq ⊂
one can check that Q′ ∩ qM1q ⊂ ... ⊂ Q′ ∩ qMnq ⊂
qMn+1q ⊂ ... ⊂ qM q and qM q = ∪nqMnq
SOT
30
SOT
Q′ ∩ qMn+1q ⊂ ... ⊂ Q′ ∩ qM q and also Q′ ∩ qM q = ∪nQ′ ∩ qMnq
. Since Q′ ∩ qM q is finite
dimensional there must be a minimal integer l so that Q′ ∩ qMlq = Q′ ∩ qM q. In particular, we have
r ∈ qMnq and by (5.0.1) we obtain upL(Λ)pu∗ ⊆ Ml. As Ml is a factor one can find w ∈ U(M ) so
that wL(Λ)w∗ ⊆ Ml.
Since the action Λ y B is compact the using Theorem 3.10 there is a Λ-invariant von Neumann
subalgebra B1 ⊂ B satisfying w(B1 ⋊ Λ)w∗ = Al ⋊ Γ = Ml. Since L(Γ) has property (T) and
[Ml : L(Γ)] < ∞ it follows that Ml has property (T). Thus B1 ⋊ Λ is a factor with property (T) and
as B1 ⋊ Λ has Haagerup property relative to L(Λ) we conclude that B1 ⋊ Λ ≺ L(Λ) and hence by
[CD18, Proposition 2.3] we have [B1 ⋊ Λ : L(Λ)] < ∞. Hence by Lemma 2.4 B1 is finite dimensional
and the action Λ y B1 is transitive. Finally, using Theorem 3.10 successively there exist a tower of
Λ-invariant finite dimensional abelian von Neumann subalgebras B1 ⊂ ... ⊂ Bn ⊂ Bn+1 ⊂ ... ⊂ B
= B and also w(Bk+1 ⋊ Λ)w∗ = Ak+l ⋊ Γ for all k ≥ 0. Thus there exists a
such that ∪n≥1Bn
sequence of factors Λ y Yn of Λ y Y into finite probability spaces Yn such that L∞(Yn) = Bn for
all n ≥ 1. From the previous relations one can check Λ y Y is the inverse limit of Λ y Yn which
gives the desired statement.
SOT
The von Neumann algebraic methods developed in the previous sections can be used effectively to
derive the following version of Ioana's OE-superrigidity theorem [Io08b, Theorem A] for profinite
actions of icc groups.
Theorem 5.3. Let Γ y X be a profinite free ergodic action of an icc property (T) group Γ and let
Λ y Y be an arbitrary free ergodic action of an icc group Λ. Assume that Θ : L∞(X)⋊Γ→L∞(Y )⋊Λ
is a ∗-isomorphism such that Θ(L∞(X)) = L∞(Y ). Then there exist projections p ∈ L∞(X) and
q ∈ L∞(Y ), a unitary u ∈ NL∞(Y )⋊Λ(L∞(Y )) with uΘ(p)u∗ = q, normal subgroups Γ′ ⊳ Γ, Λ′ ⊳ Λ
with [Γ : Γ′] = [Λ : Λ′] < ∞, a character η : Γ′→T and a group isomorphism δ : Γ′→Λ′ such that for
all γ ∈ Γ′ and a ∈ A we have
Θ(apuγ) = η(γ)Θ(ap)u∗vδ(γ)u.
In particular the actions Γ y X and Λ y Y are virtually conjugate.
SOT
SOT
= A = ∪nBn
Proof. Suppressing Θ we can assume L∞(X) = L∞(Y ) = A and A ⋊ Γ = A ⋊ Λ = M . Since
property (T) is an OE-invariant [Fu99a, Corollary 1.4] it follows that Λ is also a property (T) icc
group. Since Γ y A is profinite then it is weakly compact in the sense of Ozawa-Popa and by
[OP07, Proposition 3.4] it follows that Λ y B is also weakly compact. Since Λ has property (T)
then [Io08b, Remark 6.4] implies that Λ y B is compact. Thus using Theorem 5.1 there exist
increasing towers of Γ-invariant, finite dimensional algebras (An)n ⊆ A and (Bn)n ⊆ A such that
∪nAn
. Also there is a unitary w ∈ M and an integer s such that for all k we
have
(5.0.2)
Since As ⋊ Γ ⊆ As+k ⋊ Γ is a finite index inclusion of II1 factors then so is B0 ⋊ Λ ⊆ Bk ⋊ Λ. Thus
by Proposition 2.5 it follows that Bk is finite dimensional. Since Γk := StabΓ(As+k) ⊳ Γ is a finite
index normal subgroup so is Λk := StabΛ(Bk) ⊳ Λ. Since w ∈ M = ∪kAk ⋊ Γ
is a unitary there
exists a sequence wk ∈ U(Ak ⋊ Γ) such that kw − wkk2→0 as k→∞.
For the remaining part of the proof for every m ≥ k we will keep in mind the following diagram of
inclusions
w(As+k ⋊ Γ)w∗ = Bk ⋊ Λ.
SOT
31
ww∗
k(As+m ⋊ Γ)wkw∗ = Bm ⋊ Λ
ww∗
k(As+k ⋊ Γ)wkw∗ = Bk ⋊ Λ
(5.0.3)
∪
∪
∪
∪
w(As ⋊ Γ)w∗
= B0 ⋊ Λ
l
Pick k large enough such that k1 − ww∗
At(Bl) = {bj
(hence dim Bk ≤ dim As+k for all k); in particular we have τ (bj
k)k2 +k(1− ww∗
k(1− ww∗
kai
k)ai
k(1 − ww∗
kk2kai
: 1 ≤ i ≤ rl} and
: 1 ≤ i ≤ tl}. Also we can assume without any loss of generality that dim B0 ≤ dim As
l). Fix 1 ≤ i ≤ rk such that
k)k2 +
kk2 ≤ 10−9. Denote by At(As+l) = {ai
kk2. Hence if we denote by δi
kk2 ≤ 2kai
then we have that
kkk1− ww∗
k(1− ww∗
l ) ≥ τ (ai
k = kai
kk−1
k)ai
2
l
δi
k ≤ 2k1 − ww∗
kk2 < 10−8.
(5.0.4)
With this notations at hand we show that
Claim 5.4. There is a unique 1 ≤ j ≤ tk such that for every γ ∈ Γk one can find λ, λ′ ∈ Λ such
that
(1 − 2δi
(1 − 3δi
k)kbj
k)kbj
kk2
kk2
2 ≤ Reτ (ai
2 ≤ Reτ (ww∗
kai
kuγvλbj
k), and
kuγwkw∗vλ′ bj
k).
Proof of Claim 5.4. Fix γ ∈ Γk. By triangle inequality we have kai
kai
kuγ − ww∗
kk2. Applying the conditional expectation and using (5.0.3) we also have kEBk⋊Λ(ai
kkai
δi
kai
ww∗
kuγwkw∗k2 ≤ δi
kk2. Then the triangle inequality further gives
kkai
kuγwkw∗k2 ≤
kuγ) −
(5.0.6)
(5.0.5)
kai
kai
kuγ − EBk ⋊Λ(ai
kuγ)k2 ≤ 2δi
kuγ − EBk ⋊Λ(ww∗
kai
kkai
kk2
kuγwkw∗)k2 ≤ 3δi
kkai
kk2.
(5.0.7)
(5.0.8)
4(δi
kai
k)2kai
kuγ)k2
k)2kai
keλk2
2.
2 = kai
keλEBk (ai
2 ≤ 4(δi
kuγvλ−1 )k2
keλ − EBk (ai
kuγ − EBk⋊Λ(ai
This combined with (5.0.7) yield
keλ 6= 0 and kai
k)kai
By Lemma 4.3 there exist orthogonal projections eλ ∈ A so that Pλ eλ = 1 and uγ = Pλ∈Λ eλvλ.
Xλ∈Λ
Thus one can find λ ∈ Λ so that ai
ity and basic calculations show that 2(1− 2δi
2Reτ (ai
kuγvλ−1 )); thus
(1 − 2δi
2 ≤ Reτ (ai
k = 1 and EBk (x) =Pj τ (xbj
keλbj
k)τ (bj
kuγvλ−1 bj
keλbj
k 6= 0 and (1−2δi
k)kai
k) 6= 0 this further gives
(1 − 2δi
2 ≤ Reτ (ai
k, relation (5.0.9) implies that (1 −
k)−1. Hence there is a j (that at this point
kk2
k)−1.
Using the formulas Pj bj
2 ≤ RePj τ (ai
k)Pj kai
kk2
2δi
may depend on γ!) so that ai
keλbj
2 = τ (ai
kk2
Simplifying kai
2 = Xλ∈Λ
kk2
kkai
2 +kEBk (ai
keλk2. This inequal-
2 ≤
kuγvλ−1 )k2 ≤ 2δi
keλk2
keλ−EBk (ai
keλk2
2 ≤ (1− 4(δi
kuγvλ−1 )k2
2 ≤ Reτ (ai
keλbj
keλbj
kuγvλ−1 bj
keλEBk (ai
kuγv−1
λ bj
k)2)kai
k)kbj
kk2
k)kai
keλk2
k)τ (bj
k)−1bj
kuγvλ−1 )).
k)τ (bj
k)τ (ai
keλbj
keλbj
k)τ (ai
(5.0.10)
(5.0.9)
k).
32
Proceeding in a similar manner inequality (5.0.8) implies there exist λ′ ∈ Λ and 1 ≤ j′ ≤ rk such
that
(5.0.11)
kuγwkw∗vλ′ bj′
k ).
(1 − 3δi
k)kbj′
k k2
kai
2 ≤ Reτ (ww∗
To finish the proof it suffices to argue that j = j′ and j is unique (hence does not depend on
kuγvλ−1 bj
γ). Since τ (ai
2 =
τ (bj
k − 2bj
k) ≤ τ (bj
kai
k) then (5.0.10) implies kbj
k) and hence
k − ai
k + ai
keλbj
kk2
k) = τ (ai
k) + τ (ai
kEA(uγvλ−1 )bj
k) − 2τ (bj
keλai
kbj
k − ai
By triangle inequality this also yields
k) = τ (ai
k) ≤ 4δi
kk2 ≤ 2(δi
k)
kτ (bj
2kbj
kk2.
1
(5.0.12)
(5.0.13)
Then Cauchy-Schwarz inequality in combination with (5.0.11) and (5.0.13) show that
kbj
k − ww∗
kai
kwkw∗k2 ≤ (2(δi
k)
1
2 + δi
k)kbj
kk2.
kbj′
k bj
kk2 ≥ Reτ (bj
kww∗
≥ Reτ (ww∗
kai
k − 2(δi
≥ (1 − 4δi
k)
kuγwkw∗vλ′ bj′
k )
kuγwkw∗vλ′ bj′
k ) − kbj
kk2
2.
2 )kbj
1
kwkw∗k2kbj′
kai
k − ww∗
k k2
k's are orthogonal this forces that j = j′. Uniqueness of j (hence independence of γ) follows
(cid:4)
As bj
from (5.0.12).
Next we show the following
Claim 5.5. There exist 1 ≤ j ≤ tk and a unitary sj
pj
k and
k ∈ Bk ⋊ Λ such that sj
k(Bk ⋊ Λ)cj
k)∗ = cj
kww∗
kai
kwkw∗(sj
k)∗ = cj
k ≤
kai
sj
kww∗
kwkw∗(sj
k(As+k ⋊ Γ)ai
and for every γ ∈ Γk there is λ′ ∈ Λk satisfying
2 ≤ Reτ (sj
4 )kbj
kk2
(1 − 3δi
kwkw∗) ≤ τ (pj
Proof of Claim 5.5. As τ (ai
kai
k ∈ Bk ⋊ Λ of
bj
kai
k that is equivalent (in Bk ⋊ Λ) to ww∗
kwkw∗. By [Co76, Lemma 4.1] one can find a unitary
k ∈ Bk ⋊ Λ satisfying sj
kwkw∗(sj
sj
k, [si
k)∗ = ci
k − 1 ≤
kai
3ww∗
Using (5.0.13) we see that
k − 36(δi
k)
k) = τ (ww∗
k) there is a subprojection cj
k] = 0 and sj
kwkw∗ − ci
k.
kwkw∗ − ci
kuγwkw∗(sj
kai
k,ww∗
k)∗vλ′ bj
kww∗
(5.0.14)
(5.0.15)
kww∗
kai
kai
k).
k
1
kai
kww∗
kwkw∗ − ci
kk2
1
kk2
2 ≤ 2(2(δi
k)
= (2(2(δi
k)
≤ ((2(2(δi
k)
2 + δi
2 + δi
2 + δi
2 + 2kbj
kk2
k − cj
k)2kbj
kk2
2 = 2(2(δi
k)
k)2 + 2)kbj
kk2
2 − 2kai
k)2 + 2) − 2(1 − 2(δi
k)
2
1
1
1
Combining with the previous inequality we get ksj
Cauchy-Schwarz inequality and (5.0.6) show that for every γ ∈ Γk there is λ′ ∈ Λ such that
1
2 )2)kbj
2 ≤ 18(δi
kk2
k)
4kbj
k − 1k2 ≤ 18(δi
kk2. In turn this together with
k)
2kbj
kk2
2.
1
1
2 + δi
k)2kbj
kk2
2 + 2τ (bj
k − cj
k)
33
Reτ (sj
kww∗
kuγai
kw∗
kw(sj
k)∗vλ′ bj
1
kwvλ′ bj
k) − 2k1 − sj
kuγai
kw∗
4 )kbj
k − 36(δi
kk2
k)
2.
k)∗vλ′ bj
kw(sj
kw∗
kuγai
kww∗
kk2kbj
kk2
kvλ′ bj
k) ≥ Reτ (ww∗
≥ (1 − 3δi
kk2 ≥ Reτ (sj
k(Bk ⋊Λ)bj
kcj
k ⊆ L(Λk)bj
k = L(Λk)bj
k(As+k ⋊Γ)ai
kL(Γk)ai
kwkw∗(sj
k = L(Γk)ai
k)∗ = cj
k and bj
kL(Λk)bj
k) the above inequality
(cid:4)
This shows (5.0.15). Also since kcj
also shows that λ′ ∈ Λk. The rest of the statement follows from the previous relations.
Since ai
k then (5.0.14) of Claim 5.5 implies that
sj
kww∗
k. Since Λk and Γk are icc groups we see that
the conditions (1) and (2) in [KV15, Theorem 4.1] are satisfied, where G = sj
k)∗.
Also (5.0.15) shows that (3) in [KV15, Theorem 4.1] is also satisfied. Therefore using the conclusion
of that theorem we get that cj
In conclusion we have that sj
kwkw∗(si
Γk)ai
that z∗z = ai
(A ⋊ Λk)bj
the algebra (A ⋊ Λk)bj
k(A ⋊
k. Let z ∈ NM (A) such
k)∗ one can check that y is a unitary in
k. Thus applying Theorem 4.5 (working with
k. Notice we also have si
k = (A ⋊ Γk)bj
k(A ⋊ Γ)ai
k and denote by y = bj
kwkw∗(si
k) we get the desired conclusion by letting p = ai
kww∗
kai
k)∗ = si
k and z∗z = bj
k = bj
k.
kL(Γk)ai
kww∗
k)∗ = L(Λk)bj
k(A ⋊ Γ)bj
k satisfying y(sj
kwkw∗(sj
k)∗)y∗ = Abj
kzwkw∗(sj
kwkw∗(sj
kwkw∗(sj
kww∗
kΓkai
k q = bj
k etc.
kww∗
kAai
k)∗ = bj
kww∗
Final remarks. We notice that (5.0.15) can be used directly to show that Γk is isomorphic to
finite index subgroup of Λk. It is plausible that one can exploit this further and show the conclusion
directly, without appealing to the results in [IPV10, KV15].
Acknowledgments
The authors are grateful to Adrian Ioana and Jesse Peterson for many helpful discussions related to
this project. The authors are extremely grateful to Yuhei Suzuki for carefully reading a first draft of
this paper, for his helpful comments and suggestions, and for correcting numerous typos and minor
inaccuracies. The authors are also grateful to Rahel Brugger for her helpful comments on our paper,
and for correcting a few typos. The second author would like to thank Vaughan Jones for several
suggestions and comments regarding the results of this paper. The second author would also like
to thank Krishnendu Khan and Pieter Spaas for stimulating conversations regarding the contents of
this paper. The first author was partially supported by NSF Grant DMS #1600688.
References
[AB19] V. Alekseev and R. Brugger, A rigidity result for normalized subfactors, Preprint arXiv:
1903.04895.
[BJKW00] O. Bratteli, P.E.T. Jørgensen, A. Kishimoto and R. F. Werner, Pure states on Od, J.
Operator Theory 43 (2000),97 -- 143.
[Bh94] M. Bhattacharjee, Constructing finitely presented infinite nearly simple groups, Comm.
Algebra 22 (1994), 4561 -- 4589.
34
[BC14] R. Boutonnet and A. Carderi, Maximal amenable von Neumann subalgebras arising from
maximal amenable subgroups, Geom. Funct. Anal. 25 (2015), 1688 -- 1705.
[BIP18] R. Boutonnet, A. Ioana, and J. Peterson, Properly proximal groups and their von Neumann
algebras, Preprint arXiv:1809.01881.
[Bo12]
R. Boutonnet, W ∗-superrigidity of mixing Gaussian actions of rigid groups, Adv. Math.
244 (2013) 69 -- 90.
[BM01] M. Burger and S. Mozes, Lattices in products of trees, Inst. Hautes `Etudes Sci. Pub. S´er.
I Math. 92 (2001), 151 -- 194.
[Ca53]
R. Camm, Simple free products, J. London Math. Soc. 28 (1953), 66 -- 76.
[CD18]
I. Chifan and S. Das, A remark on the ultrapower algebra of the hyperfinite factor, Proc.
Amer. Math. Soc., 146(2018), 5289 -- 5294.
[CdSS18] I. Chifan, R. de Santiago and W. Sucpikarnon, Tensor product decompositions of II1
factors arising from extensions of amalgamated free product groups , Comm. Math. Phy.,
364(218) issue 3, 1163 -- 1194.
[CK15]
I. Chifan and Y. Kida, OE and W ∗ superrigidity results for actions by surface braid groups,
Proc. of London Math. Soc. 111 no. 6 (2015), 1431 -- 1470.
[CIK13]
I. Chifan, A. Ioana, and Y. Kida, W ∗-superrigidity for arbitrary actions of central quo-
tients of braid groups, Math. Ann., 361(2015), no.3-4, 563 -- 582.
[CKP15] I. Chifan, Y. Kida and S. Pant, Primeness results for von Neumann algebas associated
with surface braid groups, Int. Math. Res. Not. 16 (2016), 4807 -- 4848.
[CP10]
[CP11]
I. Chifan and J. Peterson, Some unique group measure space decomposition results, Duke
Math. J. 162 (2013), no.11, 1923 -- 1966.
I. Chifan and J. Peterson, On approximation properties for probability measure preserving
actions, Preprint 2011.
[CPS12]
I. Chifan, S. Popa, and O. Sizemore, Some OE- and W ∗-rigidity results for actions by
wreath product groups, J. Funct. Anal. 263(2012), no. 11, 3422 -- 3448.
[CS11]
I. Chifan and T. Sinclair, On the structural theory of II1factors of negatively curved groups,
Ann. Sci. ´Ec. Norm. Sup. 46 (2013), 1 -- 33.
[CSU11]
I. Chifan, T. Sinclair, and B. Udrea, On the structural theory of II1 factors of negatively
curved groups, II. Actions by product groups, Adv. Math. 245 (2013), 208 -- 236.
[CSU13]
I. Chifan, T. Sinclair and B. Udrea, Inner amenability for groups and central sequences in
factors, Ergodic Theory Dynam. Systems 36 (2016), 1106 -- 1029.
[CP13] D. Creutz, and J. Peterson, Character rigidity for lattices and commensurators, preprint
arXiv: 1311.4513.
35
[Ch78] H. Choda, A Galois correspondence in a von Neumann algebra, Tohuku Math. J., 30(1978),
491-504.
[Co76]
A. Connes, Classification of injective factors, Ann. Math. 104 (1976) 73-115.
[DGO11] , F. Dahmani, V. Guirardel, and D. Osin, Hyperbolically embedded subgroups and rotating
families in groups acting on HYperbolic spaces, Memoirs of the Amer. Math. Soc., 245,
no 1156, 2017.
[DHI16] D. Drimbe, D. Hoff, and A. Ioana, Prime II1 factors arising from irreducible lattices in
products of rank one simple Lie groups, J. Reine. Angew. Math. (to appear), preprint,
arXiv:1611.02209.
[Dr16]
D. Drimbe, W ∗ superrigidity for coinduced actions, International Journal of Math.
29(2018).
[EK98] D. Evans, and Y. Kawahigashi, Quantum symmetries on operator algebras. Oxford Math-
ematical Monographs. Oxford Science Publications. The Clarendon Press, Oxford Univer-
sity Press, New York, 1998. xvi+829 pp. ISBN: 0-19-851175-2.
[Fu99a] A. Furman, Gromov's measure equivalence and rigidity of higher rank lattices, Ann. of
Math. (2) 150 (1999), 1059 -- 1081.
[Fur77] H. Furstenberg, Ergodic behaviour of diagonal measures and a theorem of Szemeredi on
arithmetic progressions, J. Analyse Math. 31(1977), 204 -- 256.
[FV10]
P. Fima and S. Vaes, HNN extensions and unique group measure space decomposition of
II1 factors, Trans. Amer. Math. Soc. 354 (2012) 2601 -- 2617.
[Ga09] D. Gaboriau, Orbit equivalence and measured group theory, Proceedings of the Interna-
tional Congress of Mathematicians (Hyderabad, India, 2010), Vol. III, 1501-1527, Hindus-
tan Book Agency, New Delhi, 2010.
[Ge96]
L. Ge, On maximal injective subalgebras of factors, Advances in Mathematics, 118(1996),
34-70.
[GITD16] D. Gaboriau, A. Ioana and R. Tucker-Drob, Cocycle superrigidity for translation actions
of product groups, Amer. Journal of Math. (to appear), preprint, arXiv:1603.07616.
[GK96] L. Ge and R. Kadison, On tensor products of von Neumann algebras, Inventiones mathe-
maticae, 123(1996), 453-466.
[HO11] M. Hull and D. Osin, Induced quasi-cocycles on groups with hyperbolically embedded sub-
groups Alg. and Geom. Topology 13 (2013), 2635 -- 2665.
[HPV10] C. Houdayer, S. Popa and S. Vaes, A class of groups for which every action is W ∗-
superrigid, Groups Geom. Dyn. 7 (2013), 577 -- 590.
[Io08a] A. Ioana, Cocycle superrigidity for profinite actions of property (T) groups, ArXiv preprint
version, arXiv 0805.2998v1.
36
[Io08b] A. Ioana, Cocycle superrigidity for profinite actions of property (T) groups, Duke Math J.
157 (2011), 337 -- 367.
[Io10]
[Io12]
A. Ioana, W ∗-superrigidity for Bernoulli actions of property (T) groups, J. Amer. Math.
Soc. 24 (2011), 1175 -- 1226.
A. Ioana, Cartan subalgebras of amalgamated free product II1 factors, Ann. Sci. ´Ec. Norm.
Sup., 48(2015), 71 -- 130.
[Ioana12] A. Ioana, Uniqueness of the group measure space decomposition for Popa's HT factors,
Geom. Funct. Anal., 22(2012), no. 3, 699 -- 732.
[Io18]
A. Ioana, Rigidity for von Neumann algebras, Submitted to Proceedings ICM 2018.
Preprint arXiv 1712.00151v1.
[IPV10] A. Ioana, S. Popa, and S. Vaes, A class of superrigid group von Neumann algebras, Ann.
of Math. (2) 178 (2013), 231 -- 286.
[ILP98] M. Izumi, R. Longo, and S. Popa, Automorphisms of von Neumann Algebras with a Gen-
eralization to Kac Algebras, Journal of Functional Analysis, volume 155, issue 1(1998),
25-63.
[Jo81]
V.F.R. Jones, Index for subfactors, Invent. Math. 72 (1983), 1 -- 25.
[Jo90]
[Jo91]
[Jo09]
[JS19]
[Ki06]
V.F.R. Jones, von Neumann algebras in mathematics and physics, Proceedings of the
International Congress of Mathematicians, vol I, II (Kyoto 1990), 121 -- 138, Math Soc.
Japan, Tokyo (1991).
V.F.R. Jones, Subfactor and Knots, CBMS Regional Conference Series in Mathematics
80, AMS (1991).
V.F.R. Jones, On the origin and development of subfactors and quantum topology, Bull.
Amer. Math. Soc., 46(2009), 309 -- 326.
Y. Jiang and A. Skalski, Maximal subgroups and von Neumann subalgebras with the
Haagerup property, arXiv preprint, 1903.08190v3, 2019.
Y. Kida, Measure equivalence rigidity of the mapping class group, Ann. of Math. (2) 171
(2010), 1851 -- 1901.
[KV15] A. Krogager and S. Vaes, A class of II1 factors with exactly two group measure space
decompositions, Journal de Mathmatiques Pures et Appliques, 108(2017), 88 -- 110.
[MS04] N. Monod and Y. Shalom, Orbit equivalence rigidity and bounded cohomology, Ann. of
Math. (2) 164 (2006), 825 -- 878.
[NS03]
[Ni70]
S. Neshveyev and E. Størmer, Ergodic theory and maximal abelian subalgebras of the hy-
perfinite factor, Journal of Functional Analysis, 195(2002), no. 2, 239 -- 261.
O.A. Nielsen, Maximal abelian subalgebras of hyperfinite factors II, J. Funct. Anal. 6(1970),
192 -- 202.
37
[O16]
D. Osin, Acylindrically Hyperbolic groups, Trans. Amer. Math. Soc., 368(2016), 851 -- 888.
[OP07] N. Ozawa and S. Popa, On a class of II1 factors with at most one Cartan subalgebra, Ann.
of Math. 172 (2010), 713-749.
[OP010] N. Ozawa and S. Popa, On a class of II1 factors with at most one Cartan subalgebra II,
Amer. Journal of Math. 132(2010), no.3, 841 -- 866.
[Pa01]
[Pe09]
[Pe14]
J. Packer, Point spectrum of ergodic abelian group actions and the corresponding group-
measure space factors, Pacific J. Math, 201(2001), 421 -- 428.
J. Peterson, Examples of group actions which are virtually W ∗-superrigid, preprint,
arXiv:1002.1745.
J. Peterson, Character rigidity for lattices in higher-rank groups, preprint, www. math.
vanderbilt. edu/ peters10/ rigidity. pdf (2014).
[PT11]
J. Peterson and A. Thom, Group cocycles and the ring of affiliated operators, Invent. Math.
185 (2011) 561 -- 592.
[PP86] M. Pimsner and S. Popa, Entropy and index for subfactors, Ann. Sci. ´Ecole Norm. Sup.
19 (1986), 57 -- 106.
[Po85]
[Po99]
[Po01]
S. Popa, Notes on Cartan subalgebras in type II1 factors, Math. Scand. 57 (1985), 171 --
188.
S. Popa, Some properties of the symmetric enveloping algebra of a factor, with applications
to amenability and property (T), 4 Doc. Math. (1999), 665 -- 744.
S. Popa, On a class of type II1 factors with Betti numbers invariants, Ann. of Math. 163
(2006), 809-899.
[Po02]
S. Popa, Universal construction of subfactors. J. Reine Angew. Math. 543 (2002), 3981.
[Po03]
[Po04]
[Po06]
[Po07]
[Po12]
S. Popa, Strong rigidity of II1 factors arising from malleable actions of w-rigid groups I,
Invent. Math. 165 (2006), 369 -- 408.
S. Popa, Strong rigidity of II1 factors arising from malleable actions of w-rigid groups II,
Invent. Math. 165 (2006), 409 -- 451.
S. Popa, Deformation and rigidity for group actions and von Neumann algebras, Interna-
tional Congress of Mathematicians. Vol. I, 445 -- 477, Eur. Math. Soc., Zurich, 2007.
S. Popa, On the superrigidity of malleable actions with spectral gap, J. Amer. Math. Soc.
21 (2008), 981 -- 1000.
S. Popa, On the inductive limits of II1 factors with spectral gap, Trans. Amer. Math. Soc.
364 (2012), 2987 -- 3000.
[PV09]
S. Popa and S. Vaes, Group measure space decomposition of II1 factors and W ∗-
superrigidity, Invent. Math. 182 (2010), 371 -- 417.
38
[PV11]
S. Popa and S. Vaes, Unique Cartan decomposition for II1 factors arising from arbitrary
actions of free groups, Acta Math. 212 (2014), 141 -- 198.
[PV12]
S. Popa and S. Vaes, Unique Cartan decomposition for II1 factors arising from arbitrary
actions of hyperbolic groups, J. Reine Angew. Math. 694 (2014), 215 -- 239.
[Si55]
I. M. Singer, Automorphisms of finite factors. Amer. J. Math., 77 (1955), 117 -- 133.
[Su18]
Y. Suzuki, Complete descriptions of intermediate operator algebras by intermediate ex-
tensions of dynamical systems, To appear in Comm. Math. Phy., ArXiv Preprint,
https://arxiv.org/pdf/1805.02077.pdf
[SWW09] R. Smith, S. White, A. Wiggins, Normalizers of irreducibe subfactors J. Math. Anal.
Appl. 352 (2009), 684 -- 695.
[Va10a] S. Vaes, Rigidity for von Neumann algebras and their invariants, Proceedings of the In-
ternational Congress of Mathematicians(Hyderabad, India, 2010) Vol III, Hindustan Book
Agency, 2010, 1624 -- 1650.
[Va10b] S. Vaes, One-cohomology and the uniqueness of the group measure space decomposition of
a II1 factor, Math. Ann. 355 (2013), 661-696.
[Z76]
R. J. Zimmer, Ergodic actions with generalized discrete spectrum, Illinois J. Math.,
20(1976), no. 4, 555 -- 588.
Department of Mathematics, The University of Iowa, 14 MacLean Hall, Iowa City,
IA 52242, U.S.A.
Email: [email protected]
Email: [email protected]
39
|
1609.01225 | 1 | 1609 | 2016-09-05T17:19:47 | KMS quantum symmetric states | [
"math.OA"
] | Let $A$ be a unital C$^*$-algebra and let $\sigma$ be a one-parameter automorphism group of $A$. We consider $\operatorname{QSS}_\sigma(A)$, the set of all quantum symmetric states on $*_1^\infty A$ that are also KMS states (for a fixed inverse temperature, for specificity taken to be $-1$) for the free product automorphism group $*_1^\infty\sigma$. We characterize the elements of $\operatorname{QSS}_\sigma(A)$, we show that $\operatorname{QSS}_\sigma(A)$ is a Choquet simplex whenever it is nonempty and we characterize its extreme points. | math.OA | math | KMS QUANTUM SYMMETRIC STATES
KEN DYKEMA† AND KUNAL MUKHERJEE
Abstract. Let A be a unital C∗-algebra and let σ be a one-parameter automor-
phism group of A. We consider QSSσ(A), the set of all quantum symmetric states
on ∗∞
1 A that are also KMS states (for a fixed inverse temperature, for specificity
taken to be −1) for the free product automorphism group ∗∞
1 σ. We characterize
the elements of QSSσ(A), we show that QSSσ(A) is a Choquet simplex whenever
it is nonempty and we characterize its extreme points.
6
1
0
2
p
e
S
5
]
.
A
O
h
t
a
m
[
1
v
5
2
2
1
0
.
9
0
6
1
:
v
i
X
r
a
1. Introduction
For a unital C∗-algebra A, the set QSS(A) of quantum symmetric states was intro-
duced and studied in [4]. These are the states on the universal, unital free product
C∗-algebra ∗∞
1 A of countably infinitely many copies of A, that are invariant under
the quantum permutation groups of S. Wang [13]. In [4], a version of Kostler and
Speicher's noncommutative de Finetti theorem [8] was proved and used to charac-
terize the quantum symmetric states in terms of reduced amalgamated free products
of C∗-algebras, and the extreme points of QSS(A) were characterized.
In [3], the
set TQSS(A) of tracial quantum symmetric states was studied and was shown to
be a Choquet simplex and, moreover, the Poulsen simplex. The extreme points of
TQSS(A) were also characterized in [3].
In this paper, given a pointwise continouous one-parameter automorphism group
σ of A, we study the set QSSσ(A), of quantum symmetric states on ∗∞
1 A that are
also KMS states for the free product automorphism group ∗∞
1 A. KMS states
arose in the study of quantum statistical mechanics and are fundamental in the
Tomita-Takasaki theory of von Neumann algebras (see the introduction of [1] for
more, and Sec. 2 below for definitions). Our main result (Thm. 6.2) is that QSSσ(A)
is a Choquet simplex; we also give (Thm. 4.5) a characterization of the elements
of QSSσ(A) in terms of amalgamated free products and automorphism groups, and
(also in Thm. 6.2) a characterization of the extreme points of QSSσ(A).
1 σ of ∗∞
Let us mention two results that we use in the proofs of the above described results.
Thm. 3.2 shows how KMS states arise naturally using the reduced amalgmated free
product construction, and Thm. 5.2 characterizes the center of an amalgamated free
product of infinitely many copies of a given pair consisting of a von Neumann algebra
and a faithful conditional expectation. It is a generalization of Thm. 3.3 of [3], which
was in the tracial setting.
Date: Septeber 5, 2016.
2000 Mathematics Subject Classification. 46L54 (46L53).
Key words and phrases. quantum symmetric state, KMS state, amalgamated free product.
†Research supported in part by NSF grant DMS-1202660.
1
2
DYKEMA AND MUKHERJEE
Here is a brief summary of the rest of the paper: Sec. 2 gives background on
and recalls some results about KMS states. Sec. 3 proves the result about KMS
states of amalgamated free product C∗-algebras. Sec. 4 proves the characterization
of KMS quantum symmetric states. Sec. 5 describes the center of the amalgamated
free product of infinitely many copies of a von Neumann algebra. Sec. 6 proves that
QSSσ(A) is a Choquet simplex, whenever it is nonempty.
2. KMS states
In this section, we recall the definition of KMS states and some related facts. All
of these are well known, but we provide short proofs of some of them for convenience.
Consider a C∗-algebra A and a one-parameter, pointwise-norm continous group
(σt)t∈R of automorphisms of A. The continuity condition is that for each x ∈ A, the
function t 7→ σt(x) is norm continuous. An element a ∈ A is said to be entire analytic
if there is a function C ∋ z 7→ σz(a) ∈ A that is differentiable with respect to the
norm topology in A and such that, as suggested by the notation, the restriction to
the real axis of this function is given by the automorphism group acting on a. (See,
for example, Prop. 2.5.21 of [1].) The facts in the following lemma are well known
and are easy to verify, using that every A-valued analytic function is determined by
its restriction to the real line.
Lemma 2.1. Let a, b ∈ A, be entire analytic elements and let z, w ∈ C. Then
(i) a + b is an entire analytic element and σz(a + b) = σz(a) + σz(b);
(ii) ab is an entire analytic element and σz(ab) = σz(a)σz(b);
(iii) a∗ is an entire analytic element and σz(a∗) = σz(a)∗;
(iv) σw(a) is an entire analytic element and σz(cid:0)σw(a)(cid:1) = σz+w(a).
Let Aσ be the set of all entire analytic elements of A. From the above, it follows
that Aσ is a ∗-subalgebra of A. By Prop. 2.5.22 of [1], this subalgebra is dense in A.
Lemma 2.2. Let B be a C∗-subalgebra of a C∗-algebra A and suppose E : A → B
is a conditional expectation. Suppose σ = (σt)t∈R is a one-parameter automorphism
group of A such that E ◦ σt = σt ◦ E for all t ∈ R. Suppose a ∈ A is entire analytic.
Then E(a) is entire analytic, and σz(E(a)) = E(σz(a)) for all z ∈ C.
Proof. Since E has norm 1, the map z 7→ E(σz(a)) is B-valued holomorphic. When
z = t ∈ R, then we have E(σt(a)) = σt(E(a)).
(cid:3)
A proof of the next result is contained in the proof of Prop. 5.3.7 of [2].
Proposition 2.3. Let β ∈ R\{0} and let φ be a state of A. Then the following are
equivalent:
(a) There is a σ-invariant dense ∗-subalgebra B of Aσ, such that for all a ∈ A and
all b ∈ B, φ(aσβi(b)) = φ(ba).
(b) For all a ∈ A and all b ∈ Aσ, φ(aσβi(b)) = φ(ba).
(c) If β > 0, let
and if β < 0, let
Dβ = {z ∈ C 0 < ℑz < β}
Dβ = {z ∈ C β < ℑz < 0}.
KMS QUANTUM SYMMETRIC STATES
3
Let Dβ denote the closure of Dβ. Then for all a, b ∈ A, there is a bounded and
continuous function fa,b : Dβ → C that is holomorphic on Dβ and such that for
all t ∈ R,
fa,b(t) = φ(cid:0)aσt(b)(cid:1),
fa,b(t + iβ) = φ(cid:0)σt(b)a(cid:1).
Definition 2.4. A state φ of A is said to satisfy the Kubo-Martin-Schwinger or
KMS condition with respect to σ at value β ∈ R if it satisfies the above equivalent
conditions.
Note that, if a state satisfies the KMS-condition at value β 6= 0 for the automor-
phism group σ, then it satisfies the KMS-condition at value β = −1 for an aumor-
phism group obtained from σ by reparameterization. Thus, we confine ourselves to
the case β = −1, and we call such states σ-KMS states, or, if σ is clear from context,
simply KMS states. (See Def. 5.3.1 of [2].)
Remark 2.5. By Prop. 5.3.3 of [2], every σ-KMS state φ is σ-invariant, namely,
φ(σt(a)) = φ(a) for every a ∈ A and t ∈ R. It then follows, when a is an entire
analytic element, that φ(σz(a)) = φ(a) for all z ∈ C.
Remark 2.6. From condition (c) of Prop. 2.3, we see that if B is a unital C∗-algebra
with one-parameter automorphism group γ = (γt)t∈R and if θ : B → A is a ∗-
homomorphism such that σt ◦ θ = θ ◦ γt for all t ∈ R, then φ ◦ θ is a γ -- KMS state of
B.
The following easy result will be useful in proving that certain states are KMS
states.
Lemma 2.7. Let σ be a one-parameter automorphism group of a C∗-algebra A. Let
B = {b ∈ Aσ ∀a ∈ A, φ(aσ−i(b)) = φ(ba)}.
Then B is a subalgebra of Aσ.
Proof. Clearly, B is a vector subspace of Aσ. We need only show that B is closed
under multiplication. Let b1, b2 ∈ B and take a ∈ A. Then we have
φ(cid:0)aσ−i(b1b2)(cid:1) = φ(cid:0)aσ−i(b1)σ−i(b2)(cid:1) = φ(cid:0)b2aσ−i(b1)(cid:1) = φ(b1b2a).
Thus, b1b2 ∈ B.
(cid:3)
The following result is well known (as are all in this section). For convenience, we
provide a proof.
Lemma 2.8. Let φ be a σ-KMS state of a C∗-algebra A and suppose that the Gelfand-
Naimark-Segal (GNS) representation πφ of φ is faithful on A. Then φ is faithful
on A. Moreover, the unique normal extension φ of φ to the von Neumann algebra
M = πφ(A)′′ generated by the image of the GNS representation of A, is faithful on
M.
Proof. Using the Cauchy-Schwarz inequality, we easily see that faithfulness of φ is
equivalent to the existence, for every nonzero a ∈ A, of an element b ∈ A such that
φ(ba) 6= 0.
4
DYKEMA AND MUKHERJEE
Let a ∈ A, a 6= 0. Since the GNS representation of φ is faithful, there exist
x, y ∈ A such that φ(y∗ax) 6= 0. Thus, φ(y∗aσt(x)) 6= 0 for all real values t in some
neighborhood of 0. Let f = fy∗a,x be the bounded, continuous function on D−1 as
described in (c) of Prop. 2.3. By the Schwarz reflection principle and uniqueness of
analytic continuation, it follows that we cannot have that f (t − i) = 0 for all t ∈ R
(for more details, see, e.g., Prop. 5.3.6 of [2]). Thus, there exists t ∈ R such that
0 6= f (t − i) = φ(σt(x)y∗a). Taking b = σt(x)y∗, this completes the proof that φ is
faithful.
We now show faithfulness of φ. Note that the automorphism group σ of A ex-
tends to a unitarily implimented, pointwise strong-operator-topology continuous,
one-parameter automorphism group σ = (σt)t∈R of M. Thus, the state, φ is σ-
KMS, according to Defn. 5.3.1 of [2]. Now, using the part of Prop 5.3.7 of [2] that
applies to W∗-algebras, the proof proceeds as before. In particular, for any nonzero
a ∈ M, there exist x, y ∈ M such that φ(y∗ax) 6= 0. Then there is a function f that
is continuous on D−1 and analytic in D−1 so that
∀t ∈ R
f (t) = φ(y∗aσt(x)),
f (t − i) = φ(σt(x)y∗a).
As before, this ensures that φ(σt(x)y∗a) 6= 0 for some t ∈ R, and this implies φ(a∗a) 6=
0.
(cid:3)
3. KMS states and amalgamated free products
The next result is that free products of KMS states are KMS states.
Theorem 3.1. Let I be a nonempty set and for all ι ∈ I let Aι be a unital C∗-algebra
with one-parameter automorphism group σ(ι) and let φι be a σ(ι)-KMS state on Aι.
Let
(A, φ) = ∗
ι∈I
(Aι, φι)
be the reduced free product of C∗-algebras. Since each automorphism group σ(ι) leaves
φι invariant, we have the free product automorphism group σ on A given by σt =
∗ι∈I σ(ι)
t
for all t ∈ R. Then φ is a σ-KMS state on A.
The proof of the above result is straightforward. The result is also a special case
of the following one. We will give a proof of the following result directly, using C∗-
algebra concepts. However, it could also be proved by using Ueda's description [12]
of the modular automorphism group in an amalgamated free product and the fact,
(see [10], Thm. VIII.1.2) that, for a given faithful normal state on a von Neumann
algebra, the state is KMS for the modular automorphism group of the state and for
no other one-parameter automorphism group.
Theorem 3.2. Let B be a unital C∗-algebra. Let I be a nonempty set and for all
ι ∈ I let Aι be a unital C∗-algebra containing B as a unital C∗-subalgebra and let
Eι : Aι → B be a conditional expectation having faithful GNS representation. Let
(A, E) = (∗B)ι∈I(Aι, Eι)
be the reduced amalgamated free product of C∗-algebras.
KMS QUANTUM SYMMETRIC STATES
5
t ◦ Eι = Eι ◦ σ(ι)
Suppose, for each ι ∈ I, σ(ι) is a one-parameter automorphism group of Aι and
for all t ∈ R. Suppose also, for all t ∈ R, that the
to B is the same for all ι ∈ I. There
for all t ∈ R, which
suppose σ(ι)
automorphism of B obtained by restricting σ(ι)
t
is a one-parameter automorphism group σ given by σt = ∗ι∈I σ(ι)
t
we call the free product automorphism group of the σ(ι) for ι ∈ I.
t
Suppose ρ is a state on B such that, for every ι ∈ I, ρ ◦ Eι is a σ(ι)-KMS state of
Aι. Then ρ ◦ E is a σ-KMS state of A.
Proof. The existence of the free product automorphism σt for each t is standard and
is not difficult to verify. These clearly form an automorphism group. The required
pointwise norm-continuity of this group is easy to verify by considering actions on
the dense subalgebra of A generated by Si∈I Aι.
Let φ = ρ ◦ E. Let A(ι) denote the ∗-algebra of elements in Aι that are analytic for
the auotomorphism group σ(ι) and let A denote the ∗-algebra of elements of A that
are analytic for the automorphism group σ. Let
B = {b ∈ A ∀a ∈ A, φ(aσ−i(b)) = φ(ba)}.
By Prop. 2.3, it will suffice to show that B is dense in A. By Lemma 2.7, it will
suffice to show that A(ι) ⊆ B for all ι ∈ I. Indeed, each A(ι) is dense in Aι, so the
algebra they generate will be dense in A.
Fix ι ∈ I and b ∈ A(ι). Note that σ−i(b) ∈ Aι (see Lemma 2.2). In order to show
that b ∈ B, it will suffice to show that the equality
φ(aσ−i(b)) = φ(ba)
(1)
holds (i) for all a ∈ Aι and (ii) for all a of the form a = a1a2 · · · an for n ≥ 1
and aj ∈ Aι(j) ∩ ker Ei(j), for some ι(1), . . . , ι(n) ∈ I with ι(j) 6= ι(j + 1) for all
1 ≤ j ≤ n − 1, and with either n ≥ 2 or with ι(1) 6= ι. Indeed, the set of all such
elements a densely spans the algebra A. If a ∈ Aι as in (i) above, then the identity (1)
holds because the restriction of φ to Aι equals ρ ◦ Eι, which is KMS by hypothesis.
If a is as in (ii) above, then by freeness, we see that E(ba) and E(aσ−i(b)) are both
zero, so both sides of (1) vanish.
(cid:3)
4. KMS quantum symmetric states
Let A be a unital C∗-algebra and let ∗∞
1 A denote the universal unital free product
C∗-algebra of A with itself infinitely many times. Recall that QSS(A) denotes the set
of states of ∗∞
1 A that are quantum symmetric, and Thm. 7.3 of [4] gives a character-
ization of these in terms of amalgamated free products. In particular, it provides a
bijection
Ψ : V(A) → QSS(A),
(2)
where V(A) is the set of all quintuples (B, D, F, θ, ρ) with D a unital C∗-subalgebra
of B, F : B → D a conditional expectation, θ : A → B a unital ∗-homomorphism
and ρ a state on D, subject to some other conditions (see Defn. 7.1 of [4] for the full
description); the state Ψ((B, D, F, θ, ρ)) arises as the composition ρ◦G◦(∗∞
1 θ), where
(Y, G) = (∗D)∞
1 (B, F ) is the reduced amalgamated free product of C∗-algebras.
6
DYKEMA AND MUKHERJEE
Definition 4.1. Let A be a unital C∗-algebra and suppose σ = (σt)t∈R is a contin-
uous one-parameter automorphism group of A. Let ∗∞
1 σ denote the one-parameter
automorphism group of ∗∞
1 σt. Let QSSσ(A) denote the set
of all ψ ∈ QSS(A) that are (∗∞
1 σ)-KMS states. To be short, we will call these the
KMS quantum symmetric states of (A, σ).
1 A given by (∗∞
1 σ)t = ∗∞
In this section, we describe the restriction of the bijection (2) that characterizes the
KMS quantum symmetric states. The KMS condition imposes further restrictions on
elements of V(A) that result in simplifications as compared to Defn. 7.1 of [4]. This
makes it appropriate to consider the following new definition.
Definition 4.2. Let A be a unital C∗-algebra and let σ = (σt)t∈R be a one-parameter
group of automorphisms of A. Let Vσ(A) be the set of all equivalence classes of
sextuples (B, D, F, θ, γ, ρ), such that
(i) B is a unital C ∗-algebra,
(ii) D is a unital C ∗-subalgebra of B,
(iii) F : B → D is a faithful conditional expectation onto D,
(iv) θ : A → B is a unital ∗-homomorphism,
(v) θ(A) ∪ D generates B as a C ∗-algebra,
(vi) D is the smallest unital C ∗-subalgebra of B that satisfies
F(cid:0)x0θ(a1)x1 · · · θ(an)xn(cid:1) ∈ D,
whenever n ∈ N, x0, . . . , xn ∈ D and a1, . . . , an ∈ A,
(vii) γ = (γt)t∈R is a one-parameter automorphism group of B,
(viii) for all t ∈ R, γt ◦ θ = θ ◦ σt and γt ◦ F = F ◦ γt,
(3)
(ix) ρ is a faithful state on D and ρ ◦ F is a γ-KMS state of B,
and where sextuples (B, D, F, θ, γ, ρ) and (B′, D′, F ′, θ′, γ′, ρ′) are defined to be equiv-
alent if and only if there is a ∗-isomorphism π : B → B′ sending D onto D′ and so
that π ◦ F = F ′ ◦ π, π ◦ θ = θ′, ρ′ ◦ π↾D = ρ and γ′
t ◦ π = π ◦ γt for all t ∈ R.
Remark 4.3. (a) Just as explained in Rem. 7.2(a) of [4], for any given A it is possible
to select a Hilbert space H having a basis of sufficiently large cardinality so that
every sextuple (B, D, F, θ, γ, ρ) in Defn. 4.2 is equivalent to (B′, D′, F ′, θ′, γ′, ρ′),
where B′ is a C∗-algebra in B(H). In order to avoid set theoretic difficulties, we
should restrict to sextuples involving C∗-subalgebras of B(H), but we will use
sloppy language and refer to "all sextuples."
(b) Just as explained in Rem. 7.2(b) of [4], we will suppress the notation for equiva-
lence classes and just write (B, D, F, θ, γ, ρ) ∈ Vσ(A).
Lemma 4.4. Suppose (B, D, F, θ) satisfy properties (i) -- (vi) of Defn. 4.2 and α is
an automorphism of A. Then there is at most one automorphism γ of B such that
γ ◦ θ = θ ◦ α and γ ◦ F = F ◦ γ.
Proof. Using property (vi), we see D = S∞
n ≥ 1, Dn is defined recursively to be the algebra generated by
n=1 Dn, where D0 = C1 and where, for
(cid:8)F(cid:0)θ(a1)d1 · · · θ(ak−1)dk−1θ(ak)(cid:1) k ∈ N, a1, . . . , ak ∈ A, d1, . . . , dk−1 ∈ Dn−1(cid:9).
KMS QUANTUM SYMMETRIC STATES
7
We see by induction on n that γ is uniquely determined on Dn. Indeed, the case
n = 0 is clear, while the induction step follows from the identity
γ(cid:0)F(cid:0)θ(a1)d1 · · · θ(ak−1)dk−1θ(ak)(cid:1)(cid:1)
= F(cid:0)θ(α(a1))γ(d1) · · · θ(α(ak−1))γ(dk−1)θ(α(ak))(cid:1).
This implies that γ is uniquely determined on D. It is clearly uniquely determined
on θ(A) and, therefore also on C ∗(θ(A) ∪ D) = B.
(cid:3)
Theorem 4.5. There is a bijection
Ψσ : Vσ(A) → QSSσ(A)
defined by
where Ψ is the bijection in (2) from [4].
Ψσ(B, D, F, θ, γ, ρ) = Ψ(B, D, F, θ, ρ),
(4)
it is clear that the map Vσ(A) ∋ (B, D, F, θ, γ, ρ) 7→
Proof. From Lemma 4.4,
(B, D, F, θ, ρ) that forgets γ is injective. Moreover, the faithfulness of F and ρ,
assumed in Defn. 4.2, guarantees that (B, D, F, θ, ρ) ∈ V(A) (see (2)). Thus, the
definition (4) of Ψσ makes sense and defines an injection from Vσ(A) into QSS(A).
To show that Ψσ maps Vσ(A) into QSSσ(A), suppose ψ = Ψσ(B, D, F, θ, γ, ρ)
and let us show that ψ is a (∗∞
1 σ)-KMS state. By Thm. 7.3 of [4], the state ψ =
Ψ(B, D, F, θ, ρ) is constructed as ψ = ρ ◦ G ◦ (∗∞
1 (B, F ) is
the amalgamated free product of C∗-algebras. Considering the assumptions (vii) -- (ix)
of Defn. 4.2 and applying Thm. 3.2, we see that, with respect to the one-parameter
1 γt)t∈R, the state ρ ◦ G of the C∗-algebra Y is γ -- KMS.
automorphism group γ := (∗∞
1 θ) = (∗∞
1 γt) ◦ (∗∞
Now noting that (∗∞
1 σt) and applying the observation made
in Rem. 2.6, we conclude that ψ is a (∗∞
1 θ), where (Y, G) = (∗D)∞
1 θ) ◦ (∗∞
1 σ)-KMS state.
In order to show that Ψσ maps Vσ(A) onto QSSσ(A), let ψ ∈ QSSσ(A) and, by
virtue of the bijection (2), let (B, D, F, θ, ρ) ∈ V(A) be such that ψ = Ψ(B, D, F, θ, ρ).
We will construct a one-parameter automorphism group γ = (γt)t∈R of B so that
(B, D, F, θ, γ, ρ) ∈ Vσ(A), which will complete the proof. For this, we will work with
the construction of (B, D, F, θ, ρ) from ψ, found in Subsection 5.1 of [4]. Write σt for
the automorphism ∗∞
1 A and σ for the one parameter automorphism group
(σt)t∈R. Since ψ is σ-KMS, we have ψ ◦ σt = ψ for all t. Recall that, for i ∈ N,
λi : A → ∗∞
1 A denotes the ∗-homomorphism sending A onto the i-th copy in the free
product construction, that Mψ denotes the von Neumann algebra generated by the
image of ∗∞
1 A under the GNS representation πψ of ψ and that πi = πψ ◦ λi. Since σt
leaves ψ invariant and is point-norm continuous in t, it yields a unitarily implemented
∗-automorphism σt of Mψ satisfying σt ◦ πψ = πψ ◦ σt and, thus,
1 σt of ∗∞
σt ◦ πi = πi ◦ σt, t ∈ R,
(5)
for every i ∈ N. The implementing unitary is Ut defined by x 7→ (σt(x)) for x ∈ ∗∞
1 A,
and is strongly continuous in t (and this choice is unique on fixing the image of the
standard vacuum vector).
8
DYKEMA AND MUKHERJEE
Recall that the tail algebra is
Tψ =
∞
\
N =1
W ∗(cid:0)
∞
[
i=N
πi(A)(cid:1).
It is, clearly, globally invariant under σt. Thus, Q0, which is the ∗-algebra generated
i=1 πi(A), is also globally invariant under σt and, therefore, so is its closure
by Tψ ∪ S∞
Qψ. Lemma 5.1.9 of [4] constructed a ∗-automorphism α of Qψ that restricts to the
identity on Tψ and satisfies, for every i ∈ N, α ◦ πi = πi+1, and Lemma 5.1.10 of [4]
constructs a conditional expectation Fψ from Qψ onto Tψ by setting
Fψ(x) = WOT -- lim
n→∞
αn(x),
where "WOT" means weak operator topology. We have
α ◦ σt(πi(a)) = α ◦ πi ◦ σt(a) = πi+1 ◦ σt(a) = σt ◦ πi+1(a) = σt ◦ α(πi(a)), a ∈ A.
This implies that α commutes with σt. Since σt is WOT to WOT continuous, this
yields Fψ ◦ σt = σt ◦Fψ. By definition (see Observation 5.3.3 of [4]) the tail C∗-algebra
is
D =
∞
[
n=0
Dψ,n ⊆ Tψ,
where Dψ,0 = C1 and where Dψ,n is defined recursively to be the algebra generated
by
(cid:8)Fψ(cid:0)πi1(a1)d1 · · · πik−1(ak−1)dk−1πik(ak)(cid:1) (cid:12)(cid:12)
k ∈ N, i1, . . . , ik ∈ N, a1, . . . , ak ∈ A, d1, . . . , dk−1 ∈ Dψ,n−1(cid:9).
Now using (5), we easily see by induction on n that
• σt(Dψ,n) ⊆ Dψ,n for every t ∈ R,
• σs ◦ σt(d) = σs+t(d), for every s, t ∈ R and d ∈ Dψ,n,
• the maping t 7→ σt(d) is norm continuous, for every d ∈ Dψ,n.
Thus, the restriction of σt to D yields a continuous one-parameter automorphism
group of D. The algebra B is by definition the C∗-algebra generated by D ∪ π1(A),
the ∗-homomorphism θ : A → B is just π1 and the conditional expectation F : B → D
is just the restriction of Fψ to B. We let γt be the restriction of σt to B and we have
that γ = (γt)t∈R is a continuous, one-parameter automorphism group of B satisfying
γt ◦ θ = θ ◦ σt and γt ◦ F = F ◦ γt.
We will now show that ρ ◦ F is a γ-KMS state of B. Even more, letting (Y, G) =
(∗D)∞
1 (B, F ) be the amalgamated free product of infinitely many copies of (B, F )
and letting γ be the one-parameter automorphism group of Y given by γt = ∗∞
1 γt,
we will show that ρ ◦ G is a γ-KMS state of Y . As in [4], we let ψ denote the
vector state on Mψ for the vector coming from the identity element of ∗∞
1 A, so that
ψ ◦ πψ = ψ. Since ψ is a σ-KMS state on ∗∞
1 A, it follows that ψ is a σ-KMS state
and, in particular (see Prop. 5.3.7 of [2], cf. Prop. 2.3(c)), for every x, y ∈ Mψ, there
KMS QUANTUM SYMMETRIC STATES
9
is a continuous function fx,y : D−1 → C that is holomorphic on D−1 and such that
for all t ∈ R,
fx,y(t) = ψ(cid:0)xσt(y)(cid:1),
fx,y(t − i) = ψ(cid:0)σt(y)x(cid:1).
Thus, the restriction of ψ to any σ-invariant C∗-subalgebra of Mψ is KMS. By
Thm. 7.3 of [4], the C∗-algebra Y is identified with the subalgebra πψ(∗∞
1 A) of Mψ,
where ρ ◦ G is identified with the restriction of ψ and where γ is identified with the
restriction of σ. It follows that ρ ◦ G is γ-KMS.
By Rem. 7.2(c) of [4], the GNS representation of ρ ◦ G is faithful. Therefore, by
Lemma 2.8, ρ ◦ G is faithful. Thus, ρ ◦ F is faithful, so both ρ and F are faithful.
Finally, since ρ◦G is γ-KMS, we get that ρ◦F is γ-KMS. It follows that all conditions
of Defn. 4.2 are satisfied and (B, D, F, θ, γ, ρ) ∈ Vσ(A).
(cid:3)
5. Centers of certain amalgamated free product von Neumann
algebras
In this section, we return to a topic visited in Sec. 3 of [3] and we prove results in
greater generality. This problem is to describe the center of a von Neumann algebra
realized as the amalgamated free product over a von Neumann algebra D of infinitely
many copies of (B, E), where B is a von Neumann algebra and E is a normal, faithful
conditional expectation onto a unital copy D ⊆ B. The answer, unsurprisingly, is
that the center is precisely the intersection of the center of B with D; in [3], we proved
this under the hypothesis that there exists a tracial state τD such that τD ◦ E is a
faithful trace on B, but here we will only assume that E is faithful and that D has a
normal, faithful state.
Throughout this section, we let D ⊆ B be a unital inclusion of von Neumann
algebras with a normal, faithful, conditional expectation E : B → D.
For an element x of a von Neumann algebra, we will let [x] denote the range
projection of x. Thus, [x] is the orthogonal projection onto the closure of the range
of x, considered as a Hilbert space operator, and it belongs to the von Neumann
algebra generated by x. The notation Z(A) means the center of A.
The following is a generalization of Lemma 3.1 of [3]. In fact, though that lemma
formally assumed the existence of a trace as described above, it only depended on the
faithfulness of the conditional expectation E. Thus, the main (and, actually, quite
mild) innovation in the following version is the use of a dense subalgebra A, which
will be used in the proof of Thm. 5.2 below.
Lemma 5.1. Let E : B → D be a faithful, normal, conditional expectation as de-
scribed above. Suppose A is a strong-operator-topology dense ∗-subalgebra of B such
that E(A) ⊆ A. Let
qA = _ {[E(b∗b)] b ∈ A ∩ ker E}.
Then qA is independent of the choice of A, and, writing q for this projection, we have
q ∈ D ∩ Z(B) and (1 − q)B = (1 − q)D.
Proof. We only need to show
qA = _ {[E(b∗b)] b ∈ ker E},
(6)
10
DYKEMA AND MUKHERJEE
because then the proof of Lemma 3.1 of [3] shows that q has the desired properties.
Let q(E) denote the right-hand-side of (6). We clearly have qA ≤ q(E). We may
assume that B is represented as a von Neumann subalgebra of bounded operators on
a Hilbert space H. Suppose ξ ∈ q(E)H. Then for every ǫ > 0, there is n ∈ N and
there are b1, . . . , bn ∈ ker E and ζ1, . . . , ζn ∈ H such that
ξ −
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
n
X
j=1
E(b∗
j bj)ζj
< ǫ.
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2
By applying Kaplansky's density theorem to the real and imaginary parts of bj in-
dividually, we find that, for each j, there is a bounded sequence (aj,k)∞
k=1 in A that
converges in strong operator topology to bj and such that a∗
j,k converges in strong
operator topology to b∗
j,kaj,k) converges in strong operator topology
to E(b∗
j bj). Since E is continuous with respect to strong operator topology and
E(bj) = 0, we may without loss of generality assume E(aj,k) = 0 for each j and k.
This implies that ξ has distance no more than ǫ to an element of qAH. This shows
q(E) ≤ qA, as required.
(cid:3)
j . Thus, E(a∗
Theorem 5.2. As described above, let E : B → D be a faithful normal conditional
expectation of von Neumann algebras. Let
(M, F ) = (∗D)∞
1 (B, E)
(7)
be the von Neumann algebra free product with amalgamation over D of infinitely
many copies of (B, E). Then the center of M is contained in D and is, in particular
Z(B) ∩ D.
Proof. Let x ∈ Z(M) and let us show F (x) = x. We may without loss of generality
assume that B is countably generated. Indeed, the ∗-subalgebra of M spanned by
the words in the various copies of B is strong-operator-topology dense in M, so x is
the limit (in strong-operator-topology) of a sequence, each term of which is a finite
sum of such words. Thus, we may replace B by some countably generated subalgebra
of it, and it will suffice to show F (x) = x under these hypothesis. Note that, then,
D is also countably generated.
It was proved by E. Ricard that F is faithful. (See Thm. 2.1 [7], where this proof
appears in print.) Since D is countably generated, it has a faithful, normal state φ.
Then φ ◦ F is a normal, faithful state of M. We will let Bi denote the i-th copy of
B in M, arising from the amalgamated free product construction and Ei : Bi → D
the corresponding copy of the conditional expectation E. We will let Hi denote the
Hilbert D-bimodule L2(Bi, Ei), which contains an orthocomplemented copy of D, and
we will let Ho
i denote the orthocomplement of D in Hi. Recall that M is constructed
as follows: on the Hilbert D-bimodule
H = D ⊕ M
Ho
i1 ⊗D · · · ⊗D Ho
ik
,
k≥1
i1,...,ik≥1
ij 6=ij+1
KMS QUANTUM SYMMETRIC STATES
11
we have the usual left actions of Bi, i ∈ N. Letting πD denote the GNS representation
of D on L2(D, φ), the left actions of Bi on H induce left actions on the Hilbert space
H ⊗πD L2(D, φ) = L2(D, φ) ⊕ M
Ho
i1 ⊗D · · · ⊗D Ho
ik
⊗πD L2(D, φ),
k≥1
i1,...,ik≥1
ij 6=ij+1
which is in a natural way identified with L2(M, φ ◦ F ).
We will review some elements of Tomita -- Takesaki theory that we need; see, for
example [10]. For x ∈ M, we will write x for the corresponding element of L2(M, φ ◦
F ). We let λ denote the usual left action of M on L2(M, φ◦F ). Thus, λ(x)y = (xy),
y ∈ M. Consider the modular automorphism group σ = (σt)t∈R of M associated to
the faithful state φ ◦ F . Note that σ leave D and each Bi globally invariant. Let Sφ
denote the densely defined, closed, conjugate linear operator defined by Sφ(x) = (x∗)
and, as usual, write Sφ = Jφ∆1/2
for its polar decomposition. Then Jφ is a conjugate
φ
linear, surjective isometry. For x ∈ M, let r(x) denote the densely defined "right
multiplication" operator r(x)y = (yx), y ∈ M. It need not be bounded. However,
(see, for example equations (4)-(6) of [6]), if x ∈ M is entire analytic with respect to
σ, then r(x) = Jφλ(σ−i/2(x∗))Jφ and r(x) is bounded. Note, also, that if x ∈ M is
entire analytic, then so is F (x).
Let x ∈ Z(M) and let η = x − F (x). We will show that η = 0, and this will prove
the theorem. Let ηN denote the orthogonal projection of η onto the subspace
H(N ) = M
Ho
i1 ⊗D · · · ⊗D Ho
ik ⊗πD L2(D, φ),
k≥1
1≤i1,...,ik≤N
ij 6=ij+1
of L2(M, φ ◦ F ). Then limN→∞ kη − ηN k = 0. Suppose b1 ∈ B1 ∩ ker E1 is entire
analytic with respect to the restriction of σ to B1. Let bN be the copy of b1 in BN .
Since x is central, we have
0 = (bN x−xbN ) = (λ(bN )−r(bN ))x = (λ(bN )−r(bN ))(ηN −1+F (x)+η−ηN −1). (8)
By considering the dense set of elements in H(N −1) that are of the form y for y ∈ M,
we see that the three subspaces
λ(bN )H(N −1),
r(bN )H(N −1)
and λ(bN )L2(D, φ) + r(bN )L2(D, φ)
are mutually orthogonal. Therefore, from (8), we get
kλ(bN )ηN −1k2
2 + kr(bN )ηN −1k2
= k(λ(bN ) − r(bN ))(η − ηN −1)k2
2 + k(λ(bN ) − r(bN ))F (x)k2
2
2 ≤ kλ(bN ) − r(bN )k2kη − ηN −1k2
2.
(9)
For every permutation ρ of N, there is an F -preserving automorphism of M that
sends the copy Bi identically to the copy Bρ(i). Thus, we see that kλ(bN ) − r(bN )k
does not depend on N. So the right-hand-side of (9) tends to 0 as N → ∞, and we
N bN )1/2
get limN→∞ kλ(bN )ηN −1k2 = 0. Letting d = E1(b∗
for every N. Moreover, again by considering the dense set of elements in H(N −1) that
1b1)1/2, we have d = EN (b∗
12
DYKEMA AND MUKHERJEE
are of the form y for y ∈ M, we have kλ(bN )vk2 = kλ(d)vk2 for every v ∈ H(N −1).
So we get
kλ(d)ηk2 = lim
N→∞
kλ(d)ηN −1k2 = lim
N→∞
kλ(bN )ηN −1k2 = 0.
Since EN (b∗
N bN ) = d2 for all N, we have
λ(EN (b∗
N bN ))η = 0.
(10)
Let A ⊆ B be the ∗-subalgebra of elements that are entire analytic with respect to
the restriction to B1 of the modular automorphism group σ (where we have identified
B and B1). Then A is strong-operator-topology dense in B (see, for example, Prop.
2.5.22 of [1] or Lemma 2.3 of [10]). Let
q = _ {[E(b∗b)] b ∈ A ∩ ker E}.
By Lemma 5.1, q ∈ Z(B) ∩ D and (1 − q)B = (1 − q)D. All copies of q in the various
BN are, therefore identified with each other, and q lies in the center of M. Thus,
(1 − q)BN = (1 − q)D for all N, and we have (1 − q)M = (1 − q)D. From (10),
we conclude that qη = 0. Since φ ◦ F is faithful, this yields q(x − F (x)) = 0. So
x − F (x) = (1 − q)(x − F (x)) ∈ (1 − q)D. However, x − F (x) ⊥ D, so we have
x − F (x) = 0, as required.
(cid:3)
6. The simplex of KMS quantum symmetric states
Turning back to the notation described in Sec. 4, we let K(∗∞
1 σ-KMS states on the universal unital free product C∗-algebra ∗∞
1 A, ∗∞
1 σ) denote
the set of all ∗∞
1 A.
1 σ) is a compact affine set in the weak∗-topology and is known
This set K(∗∞
to be a Choquet simplex whose extreme points are those φ ∈ K(∗∞
1 σ), the von
Neumann algebras generated by images of whose GNS-representations are factors (cf.
[5], [9] and [11]).
1 A, ∗∞
1 A, ∗∞
Recall that we defined QSSσ(A) = K(∗∞
1 A, ∗∞
1 σ) ∩ QSS(A).
Remark 6.1. We easily see that QSSσ(A) is nonempty if and only if A possesses a
σ-KMS state. Indeed, if φ is a σ-KMS state on A, then we always have at least two
elements of Vσ(A), which are distinct so long as A 6= C:
(i) taking (B, D, F, θ, γ, ρ) = (A, C, φ, idA, σ, idC), we get the free product state
Ψσ(B, D, F, θ, γ, ρ) = ∗∞
1 φ ∈ QSSσ(A);
(ii) taking (B, D, F, θ, γ, ρ) = (A, A, idA, idA, σ, φ), we get the state
Ψσ(B, D, F, θ, γ, ρ) = φ ◦ (cid:0) ∗∞
that arises from the ∗-homomorphism ∗∞
tically to itself, followed by the state φ.
1 idA(cid:1) ∈ QSSσ(A)
1 A → A sending each copy of A iden-
The paper [14] addresses and answers the question of existence of a σ-KMS state of
A.
The following theorem follows easily, just like Thm. 5.3 of [3], since we have already
proved the main ingredients.
KMS QUANTUM SYMMETRIC STATES
13
1 A, ∗∞
Theorem 6.2. If QSSσ(A) is nonempty, then it is a Choquet simplex and a face of
K(∗∞
1 σ). Moreover, given ψ ∈ QSSσ(A), let (B, D, F, θ, γ, ρ) ∈ Vσ(A) be the
element that maps to ψ under the bijection Ψσ from Thm. 4.5. Let B be the von
Neumann algebra generated by the image of B under the GNS representation πρ◦F of
the state ρ ◦ F and let D be the von Neumann subalgebra generated by πρ◦F (D). Then
the following are equivalent:
(i) ψ is an extreme point of QSSσ(A),
(ii) ψ is an extreme point of K(∗∞
(iii) Z(B) ∩ D = C1, where Z(B) is the center of B.
1 A, ∗∞
1 σ),
Proof. The implication (i) =⇒ (ii), when proved, will imply that QSSσ(A) is a face
of K(∗∞
1 σ) and, thus, that it is a Choquet simplex.
1 A, ∗∞
The implication (ii) =⇒ (i) is clearly true.
We now show the implication (iii) =⇒ (ii). By construction, the state ρ ◦ F of
B is γ-KMS and faithful. By Lemma 2.8, this implies that the normal extension
(ρ ◦ F ) of ρ ◦ F to B is faithful. Moreover, the conditional expectation F induces
a projection from L2(B, ρ ◦ F ) onto the subspace L2(D, ρ) and compression by this
projection induces a (ρ ◦ F )-preserving, normal, conditional expectation F : B → D,
that extends F . Letting ρ denote the unique normal extension of ρ to D, we have
(ρ ◦ F ) = ρ ◦ F . The image πψ(∗∞
1 A) of the GNS representation of ψ is, by the
construction in Thm. 7.3 of [4], isomorphic to the reduced amalgamated free product
C∗-algebra
(Y, G) = (∗D)∞
i=1(B, F ),
of infinitely many copies of (B, F ), with amalgamation over D. Moreover, the GNS
representation is itself the defining representation of the reduced free product C∗-
algebra on the Hilbert space H = L2(Y, G) ⊗πρ L2(D, ρ), where L2(Y, G) is the free
product of infinitely many copies of the Hilbert D-bimodule L2(B, F ).
The von Neumann algebra Mψ := πψ(∗∞
1 A)′′ generated by the image of the GNS
representation of ψ is, therefore, isomorphic to the amalgamated free product of von
Neumann algebras,
(∗D)∞
i=1(B, F ),
of infinitely many copies of (B, F ), with amalgamation over D. By Thm. 5.2 and the
hypothesis (iii), it follows that Mψ is a factor. Therefore, ψ is an extreme point of
K(∗∞
1 A, ∗∞
1 σ).
We will now show (i) =⇒ (iii), which will finish the proof. Suppose that (iii) fails
to hold. Then there is a projection p ∈ Z(B) ∩ D that is neither 0, nor 1. Thus,
t := ρ(p) ∈ (0, 1). Note that p lies in the center of Mψ (Thm. 5.2). Consider the
states ρ0 and ρ1 of D, defined, for d ∈ D, by
ρ0(d) = t−1 ρ(πρ(d)p),
ρ1(d) = (1 − t)−1 ρ(πρ(d)p).
We see that ρ0 6= ρ1, by considering a bounded sequence (xn)∞
n=1 in D such that
πρ(xn) converges in strong operator topology to p ∈ D. Consider, for each i ∈ {0, 1},
the state ψi = ρi ◦ F ◦ πψ of ∗∞
1 A. By Thm. 4.1 of [3], each is a quantum symmetric
state of A. Using that p lies in the center of Mψ, we have that each ψi is also
14
DYKEMA AND MUKHERJEE
(∗∞
Therefore, (i) fails to hold. This finishes the proof.
1 σ)-KMS. Thus, ψi ∈ QSSσ(A). But we have ψ = tψ0 + (1 − t)ψ1, and ψ0 6= ψ1.
(cid:3)
References
[1] O. Bratteli and D. W. Robinson, Operator algebras and quantum statistical mechanics. 1, 2nd
ed., Texts and Monographs in Physics, Springer-Verlag, New York, 1987. C ∗- and W ∗-algebras,
symmetry groups, decomposition of states.
[2]
, Operator algebras and quantum statistical mechanics. 2, 2nd ed., Texts and Monographs
in Physics, Springer-Verlag, Berlin, 1997. Equilibrium states. Models in quantum statistical
mechanics.
[3] Y. Dabrowski, K. Dykema, and K. Mukherjee, The simplex of tracial quantum symmetric states,
Studia Math. 255 (2014), 203 -- 218.
[4] K. Dykema, C. Kostler, and J. Williams, Quantum symmetric states on free product C*-algebras,
Trans. Amer. Math. Soc., to appear, available at http://arxiv.org/abs/1305.7293.
[5] G. G. Emch, H. J. F. Knops, and E. J. Verboven, Breaking of Euclidean symmetry with an
application to the theory of crystallization, J. Mathematical Phys. 11 (1970), 1655 -- 1668.
[6] T. Falcone, L2-von Neumann modules, their relative tensor products and the spatial derivative,
Illinois J. Math. 44 (2000), 407 -- 437.
[7] N. A. Ivanov, On the structure of some reduced amalgamated free product C ∗-algebras, Internat.
J. Math. 22 (2011), 281 -- 306.
[8] C. Kostler and R. Speicher, A noncommutative de Finetti theorem: invariance under quantum
permutations is equivalent to freeness with amalgamation, Comm. Math. Phys. 291 (2009),
473 -- 490.
[9] N. Riedel, Disintegration of KMS-states and reduction of standard von Neumann algebras,
Pacific J. Math. 111 (1984), no. 2, 415 -- 431.
[10] M. Takesaki, Theory of operator algebras. II, Encyclopaedia of Mathematical Sciences, vol. 125,
Springer-Verlag, Berlin, 2003. Operator Algebras and Non-commutative Geometry, 6.
[11] M. Takesaki and M. Winnink, Local normality in quantum statistical mechanics, Comm. Math.
Phys. 30 (1973), 129 -- 152.
[12] Y. Ueda, Amalgamated free product over Cartan subalgebra, Pacific J. Math. 191 (1999), 359 --
392.
[13] S. Wang, Quantum symmetry groups of finite spaces, Comm. Math. Phys. 195 (1998), 195 -- 211.
[14] S. L. Woronowicz, On the existence of KMS states, Lett. Math. Phys. 10 (1985), 29 -- 31.
K. Dykema, Department of Mathematics, Texas A&M University, College Station,
TX 77843-3368, USA
E-mail address: [email protected]
K. Mukherjee, Department of Mathematics, Indian Institute of Technology Mad-
ras, Chennai -- 600 036, India
E-mail address: [email protected]
|
1705.00357 | 1 | 1705 | 2017-04-30T18:36:29 | On Controlled Frames in Hilbert $C^*$-modules | [
"math.OA",
"math.FA"
] | In this paper, we introduce controlled frames in Hilbert $C^*$-modules and we show that they share many useful properties with their corresponding notions in Hilbert space. Next, we give a characterization of controlled frames in Hilbert $C^*$-module. Also multiplier operators for controlled frames in Hilbert $C^*$-modules will be defined and some of its properties will be shown. Finally, we investigate weighted frames in Hilbert $C^*$-modules and verify their relations to controlled frames and multiplier operators. | math.OA | math |
ON CONTROLLED FRAMES IN HILBERT C∗-MODULES
M. RASHIDI-KOUCHI, A. RAHIMI
Abstract. In this paper, we introduce controlled frames in Hilbert C ∗-
modules and we show that they share many useful properties with their
corresponding notions in Hilbert space. Next, we give a characterization
of controlled frames in Hilbert C ∗-module. Also multiplier operators for
controlled frames in Hilbert C ∗-modules will be defined and some of
its properties will be shown. Finally, we investigate weighted frames in
Hilbert C ∗-modules and verify their relations to controlled frames and
multiplier operators.
1. Introduction
Frames for Hilbert spaces were first introduced in 1952 by Duffin and
Schaeffer [11] for study of nonharmonic Fourier series. They were reintro-
duced and development in 1986 by Daubechies, Grossmann and Meyer[10],
and popularized from then on. For basic results on frames, see [8].
Hilbert C∗-modules form a wide category between Hilbert spaces and
Banach spaces. Their structure was first used by Kaplansky [20] in 1952.
They are an often used tool in operator theory and in operator algebra
theory. They serve as a major class of examples in operator C∗-module
theory.
The notions of frames in Hilbert C∗-modules were introduced and inves-
tigated in [14]. Frank and Larson [14, 15] defined the standard frames in
Hilbert C∗-modules in 1998 and got a series of result for standard frames in
finitely or countably generated Hilbert C∗-modules over unital C∗-algebras.
Extending the results to this more general framework is not a routine gener-
alization, as there are essential differences between Hilbert C∗-modules and
Hilbert spaces. For example, any closed subspace in a Hilbert space has
an orthogonal complement, but this fails in Hilbert C∗-module. Also there
is no explicit analogue of the Riesz representation theorem of continuous
functionals in Hilbert C∗-modules. We refer the readers to [24] and [19] for
more details on Hilbert C∗-modules and to [15, 28, 30, 29] for a discussion
of basic properties of frame in Hilbert C∗-modules and their generalizations.
2010 Mathematics Subject Classification. Primary 42C15; Secondary 46L08, 42C40,
47A05.
Key words and phrases. Frame, Controlled frame, Weighted frame, Hilbert C ∗-module,
Multiplier operator.
1
2
M. RASHIDI-KOUCHI, A. RAHIMI
Controlled frames have been introduced to improve the numerical effi-
ciency of iterative algorithms for inverting the frame operator on abstract
Hilbert spaces [5], however they are used earlier in [7] for spherical wavelets.
In this paper, we introduce controlled frames in Hilbert C∗-modules and
we show that they share many useful properties with their corresponding
notions in Hilbert space. Next, we give a characterization of controlled
frames in Hilbert C∗-module. Also multiplier operators for controlled frames
in Hilbert C∗-modules will be defined and some of its properties will be
shown. Finally, we investigate weighted frames in Hilbert C∗-modules and
verify their relation to controlled frames and multiplier operators.
The paper is organized as follows. In section 2, we review the concept
Hilbert C∗-modules, frames and multiplier operators in Hilbert C∗-modules.
Also the analysis, synthesis, frame operator and dual frames be reviewed.
In section 3, we introduce controlled frames in Hilbert C∗-modules and
characterize them. In section 4, we investigate weighted frames in Hilbert
C∗-modules and verify their relation to controlled frames and multiplier
operators.
2. Preliminaries
In this section, we collect the basic notations and some preliminary results.
We denote by I the identity operator on H. Let B(H1,H2) be the set of all
bounded linear operators from H1 to H2. This set is a Banach space for the
operator norm kAk = supkxkH1≤1kAxkH2 . The adjoint of the operator A is
denoted by A∗ and the spectrum of A by σ(A). We define GL(H1,H2) as
the set of all bounded linear operators with a bounded inverse, and similarly
for GL(H). Our standard reference for Hilbert space and operator theory is
[9].
Controlled frames introduced in [5] as follows. Let C ∈ GL(H). A frame
controlled by the operator C or C-controlled frame is a family of vectors
Ψ = {ψj ∈ H : j ∈ J}, such that there exist two constants m > 0 and
M < ∞ satisfying
mkfk2 ≤Xj
hf, ψjihCψj , fi ≤ Mkfk2,
for all f ∈ H.
Also weighted frames are defined as follows. Let Ψ = {ψj ∈ H : j ∈ J}
be a sequence of elements in H and {ωj : j ∈ J} ⊆ R+a sequence of positive
weights. This pair is called a weighted frame of H if there exist constants
m > 0 and M < ∞ such that
mkfk2 ≤Xj
ωjhf, ψji2 ≤ Mkfk2,
for all f ∈ H.
Hilbert C∗-modules form a wide category between Hilbert spaces and
Banach spaces. Hilbert C∗-modules are generalizations of Hilbert spaces by
ON CONTROLLED FRAMES IN HILBERT C ∗-MODULES
3
allowing the inner product to take values in a C∗-algebra rather than in the
field of complex numbers.
Let A be a C∗-algebra with involution ∗. An inner product A-module (or
pre Hilbert A-module) is a complex linear space H which is a left A-module
with an inner product map h., .i : H × H → A which satisfies the following
properties:
(1) hαf + βg, hi = αhf, hi + βhg, hi for all f, g, h ∈ H and α, β ∈ C;
(2) haf, gi = ahf, gi for all f, g ∈ H and a ∈ A;
(3) hf, gi = hg, fi∗ for all f, g ∈ H;
(4) hf, fi ≥ 0 for all f ∈ H and hf, fi = 0 iff f = 0.
For f ∈ H, we define a norm on H by kfkH = khf, fik1/2
A . If H is complete
with this norm, it is called a (left) Hilbert C∗-module over A or a (left)
Hilbert A-module.
An element a of a C∗-algebra A is positive if a∗ = a and its spectrum
In this case, we write a ≥ 0. By
is a subset of positive real numbers.
condition (4) in the definition hf, fi ≥ 0 for every f ∈ H, hence we define
f = hf, fi1/2. We call Z(A) = {a ∈ A : ab = ba,∀b ∈ A}, the center of A.
If a ∈ Z(A), then a∗ ∈ Z(A), and if a is an invertible element of Z(A), then
a−1 ∈ Z(A), also if a is a positive element of Z(A), then a
2 ∈ Z(A). Let
HomA(M, N ) denotes the set of all A-linear operators from M to N .
1
Let
with inner product
{aj} ⊆ A :Xj∈J
ℓ2(A) =
h{aj},{bj}i =Xj∈J
a∗j aj converges ink.k
a∗j bj,
{aj},{bj} ∈ ℓ2(A)
and
it was shown that [33], ℓ2(A) is Hilbert A-module.
k{aj}k :=qkX a∗j ajk,
Note that in Hilbert C∗-modules the Cauchy-Schwartz inequality is valid.
Let f, g ∈ H, where H is a Hilbert C∗-module , then
khf, gik2 ≤ khf, fik × khg, gik.
We are focusing in finitely and countably generated Hilbert C∗- modules
over unital C∗-algebra A. A Hilbert A-module H is finitely generated if
there exists a finite set {x1, x2, ..., xn} ⊆ H such that every x ∈ H can be
expressed as x = Pn
i=1 aixi, ai ∈ A. A Hilbert A-module H is countably
generated if there exits a countable set of generators.
The notion of (standard) frames in Hilbert C∗-modules is first defined by
Frank and Larson [15]. Basic properties of frames in Hilbert C∗-modules
are discussed in [16, 17, 21, 22].
4
M. RASHIDI-KOUCHI, A. RAHIMI
Let H be a Hilbert C∗-module, and J a set which is finite or countable.
A system {fj : j ∈ J} ⊆ H is called a frame for H if there exist constants
C, D > 0 such that
(2.1)
Chf, fi ≤Xj∈J
hf, fjihfj, fi ≤ Dhf, fi
If
for all f ∈ H. The constants C and D are called the frame bounds.
C = D it called a tight frame and in the case C = D = 1, it called Parseval
frame. It is called a Bessel sequence if the second inequality in (2.1) holds.
Unlike Banach spaces, it is known [15] that every finitely generated or
countably generated Hilbert C∗- modules admits a frame.
The following characterization of frames in Hilbert C∗- modules, which
was obtained independently in [1] and [18], enables us to verify whether a
sequence is a frames in Hilbert C∗- modules in terms of norms. It also allows
us to characterize frames in Hilbert C∗- modules from the operator theory
point of view.
Theorem 2.1. Let H be a finitely or countably generated Hilbert A-module
over a unital C∗-algebra A and {fj : j ∈ J} ⊆ H a sequence. Then {fj : j ∈
J} is a frame for H if and only if there exist constants C, D > 0 such that
Let {fj : j ∈ J} be a frame in Hilbert A-module H and {gj : j ∈ J} be a
sequence of H. Then {gj : j ∈ J} is called a dual sequence of {fj : j ∈ J} if
≤ Dkfk2,
f ∈ H.
hf, fjihfj , fi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Ckfk2 ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Xj∈J
f =Xj∈J
hf, gjifj
for all f ∈ H. The sequences {fj : j ∈ J} and {gj : j ∈ J} are called a dual
frame pair when {gj : j ∈ J} is also a frame.
For the frame {fj : j ∈ J} in Hilbert A-module H, the operator S defined
by
Sf =Xj∈J
hf, fjifj , f ∈ H
is called the frame operator. It is proved that [15], S is invertible, positive,
adjointable and self-adjoint. Since
hSf, fi = hXj∈J
hf, fjifj , fi =Xj∈J
hf, fjihfj , fi,
f ∈ H
it follows that
Chf, fi ≤ hSf, fi ≤ Dhf, fi,
and the following reconstruction formula holds
f ∈ H
hS−1f, fjifj =Xj∈J
hf, S−1fjifj
f = SS−1f = S−1Sf =Xj∈J
ON CONTROLLED FRAMES IN HILBERT C ∗-MODULES
5
for all f ∈ H.
Let fj = S−1fj, then
f =Xj∈J
hf, fjifj =Xj∈J
hf, fji fj,
for any f ∈ H. The sequence { fj : j ∈ J} is also a frame for H which is
called the canonical dual frame of {fj : j ∈ J}.
In [31], R. Schatten provided a detailed study of ideals of compact oper-
ators using their singular decomposition. He investigated the operators of
the orthonormal families were replaced with Bessel and frame sequences to
define Bessel and frame multipliers.
the formPj λj ϕj ⊗ ψj where (φj) and (ψj) are orthonormal families. In [4],
Definition 2.2. Let H1 and H2 be Hilbert spaces, let (ψj) ⊆ H1 and
(φj) ⊆ H2 be Bessel sequences. Fix m = (mj) ∈ l∞. The operator
defined by
Mm,(φj ),(ψj ) : H1 → H2
Mm,(φj ),(ψj )(f ) =Xj∈J
mjhf, ψjiφj
for all f ∈ H1 is called the Bessel multiplier for the Bessel sequences (ψj ) and
(φj). The sequence m is called the symbol of M. For frames, this operator
is called frame multiplier, for Riesz sequences a Riesz multiplier.
Basic properties and some applications of this operator for Bessel se-
quences, frames and Riesz basis have been proved by Peter Balazs in his
Ph.D habilation [3]. Recently, the concept of multipliers extended and in-
troduced for continuous frames [6], fusion frames [2], p-Bessel sequences
[26], generalized frames [25], controlled frames [27], Banach frames [12, 13],
Hilbert C∗-modules [23] and etc.
Definition 2.3. Let A be a unital C∗-algebra, J be a finite or countable
index set and {fj : j ∈ J} and {gj : j ∈ J} be Hilbert C∗-modules Bessel
sequences for H. For m = {mj}j∈J ∈ ℓ∞(A) with mj ∈ Z(A), for each
j ∈ J, the operator Mm,{fj},{gj} : H → H defined by
Mm,{fj},{gj}f :=Xj∈J
mjhf, fjigj ,
f ∈ H
called the multiplier operator of {fj : j ∈ J} and {gj : j ∈ J}. The
sequence m = {mj}j ∈ J called the symbol of Mm,{fj},{gj}.
The symbol of m has important role in the studying of multiplier oper-
ators. In this paper m is always a sequence m = {mj}j∈J ∈ ℓ∞(A) with
mj ∈ Z(A), for each j ∈ J.
6
M. RASHIDI-KOUCHI, A. RAHIMI
3. Controlled Frames In Hilbert C∗-modules
In this section, we introduce controlled frames in Hilbert C∗-modules and
we show that they share many useful properties with their corresponding
notions in Hilbert space. We also give a characterization of controlled frames
in Hilbert C∗-module.
Definition 3.1. Let H be a Hilbert C∗-module and C ∈ GL(H). A frame
controlled by the operator C or C-controlled frame in Hilbert C*-module
H is a family of vectors Ψ = {ψj ∈ H : j ∈ J}, such that there exist two
constants m > 0 and M < ∞ satisfying
mhf, fi ≤Xj∈J
hf, ψjihCψj , fi ≤ Mhf, fi,
for all f ∈ H.
with bound M if there exists M < ∞ such that
Likewise, Ψ = {ψj ∈ H : j ∈ J} is called a C-controlled Bessel sequence
hf, ψjihCψj , fi ≤ Mhf, fi,
Xj∈J
for every f ∈ H.
By using Lemma 3.2 we have the following definition.
Definition 3.3. Let H be a Hilbert C∗-module and C ∈ GL(H). Assume
the sequence Ψ = {ψj ∈ H : j ∈ J} is C-controlled Bessel sequence in
Hilbert C*-module H. The operator
SC f =Xj∈J
hf, ψjiCψj
is called C-controlled frame operator.
for every f ∈ H, where the sum in the inequality is convergent in norm.
if m = M = 1 it is called a Parseval C-controlled frame.
If m = M , we call this C-controlled frame a tight C-controlled frame, and
Every frame is a I-controlled frame. Hence controlled frames are gener-
alizations of frames.
The proof of the following lemma is straightforward.
Lemma 3.2. Let H be a Hilbert C∗-module and C ∈ GL(H). A sequence
Ψ = {ψj ∈ H : j ∈ J} is C-controlled Bessel sequence in Hilbert C*-module
H if and only if the operator
is well defined and there exists constant M < ∞ such that
SC f =Xj∈J
hf, ψjiCψj
hf, ψjihCψj , fi ≤ Mhf, fi,
Xj∈J
ON CONTROLLED FRAMES IN HILBERT C ∗-MODULES
7
According to the following proposition, the main properties of controlled
frame operators in Hilbert C∗-modules are the same as controlled frame
operators in Hilbert spaces.
Proposition 3.4. Let H be a Hilbert C∗-module on C∗-algebra A and C ∈
End∗A(U, V ). Assume {ψj : j ∈ J} is a C-controlled frame in Hilbert C∗-
module H with bounds m, M > 0 Then C-controlled frame operator SC is
invertible, positive, adjointable and self-adjoint.
Proof. Let f, g ∈ H and Sf =Pj∈Jhf, ψjiψj be frame operator of {ψj : j ∈
J}, then
hSC f, gi =*Xj∈J
=Xj∈J
hf, ψjiCψj , g+ =Xj∈J
hf, ψjihψj , C∗gi = hSf, C∗gi = hf, SC∗gi.
hf, ψjihCψj , gi
Therefore controlled frame operator SC is adjointable and S∗C = SC∗.
Since
hSC f, fi =*Xj∈J
hf, ψjiCψj, f+ =Xj∈J
hf, ψjihCψj , fi.
It follows that
and
mhf, fi ≤ hSC f, fi ≤ Mhf, fi,
mIdH ≤ SC ≤ M IdH.
So SC is positive and invertible. By regard to this fact that every positive
operator in Banach space is self-adjoint, C-controlled frame operator SC is
self-adjoint.
(cid:3)
Now, by using the following lemma in [1], we give a characterization of
controlled frames.
Lemma 3.5. [1] Let A be a C∗-algebra, U and V two Hilbert A-modules,
and T ∈ End∗A(U, V ). Then the following statements are equivalent:
(1) T is surjective;
(2) T ∗ is bounded below with respect to norm, that is, there is m > 0
(3) T ∗ is bounded below with respect to the inner product, that is, there
such that kT ∗fk ≥ mkfk for all f ∈ U ;
is m′ > 0 such that hT ∗f, T ∗fi ≥ m′hf, fi for all f ∈ U .
Theorem 3.6. Let H be a Hilbert C∗-module and C ∈ GL(H). A sequence
Ψ = {ψj ∈ H : j ∈ J} is C-controlled frame in Hilbert C∗-module H if and
only if there exists constants m > 0 and M < ∞ such that
8
M. RASHIDI-KOUCHI, A. RAHIMI
(3.1)
for all f ∈ H.
mkfk2 ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Xj∈J
hf, ψjihCψj , fi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
≤ Mkfk2,
Proof. Let the sequence Ψ = {ψj ∈ H : j ∈ J} is C-controlled frame in
Hilbert C∗-module H. By the definition of C-controlled frame inequality
(3.1) holds.
Now suppose that the inequality (3.1) holds. From Proposition 3.4 C-
controlled frame operator SC is positive, self-adjoint and invertible, hence
hS
kS
stants C, D > 0 such that
C fi = hSC f, fi = Pj∈Jhf, ψjihCψj , fi. So we have √mkfk ≤
C fk ≤ √Mkfk for any f ∈ H. According to Lemma 3.5 there are con-
C f, S
1
2
2
1
1
2
Ckfk2 ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Xj∈J
hf, ψjihCψj , fi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
≤ Dkfk2,
which implies that Ψ = {ψj ∈ H : j ∈ J} is C-controlled frame in Hilbert
C∗-module H.
(cid:3)
We prove the following theorems to show that every controlled frame is a
classical frame.
Theorem 3.7. Let H be a Hilbert C∗-module on C∗-algebra A and {ψj}j∈J
be a sequence in H. Assume S is the frame operator associated with {ψj}j∈J
i.e. Sf =Pj∈Jhf, ψjiψj. Then the following conditions are equivalent:
(1) {ψj}j∈J is a frame with frame bounds C and D;
(2) We have CIdH ≤ S ≤ DIdH.
Proof. (1) ⇒ (2). This is implied by proof of Proposition 3.4.
S = T ∗T and hence kTk2 = kSk for any f ∈ H we obtain
(2) ⇒ (1). Let T be the analysis operator associated with {ψj}j∈J . Since
hf, ψjihψj , fik = khSf, fik ≤ kSfkkfk2 ≤ Dkfk2.
kXj∈J
Also, for all f ∈ H,
Therefor for all f ∈ H
khSf, fik ≥ khCf, fik = Ckfk2.
Ckfk2 ≤ kXj∈J
hf, ψjihψj , fik ≤ Dkfk2.
Now, by Theorem 2.1 proof is complete.
Theorem 3.8. [8] Let X be a Banach space, U : X → X a bounded operator
and kI − Uk < 1. Then U is invertible.
(cid:3)
ON CONTROLLED FRAMES IN HILBERT C ∗-MODULES
9
Theorem 3.9. Let T : H → H be a linear operator. Then the following
conditions are equivalent:
(1) There exist m > 0 and M < ∞ such that mI ≤ T ≤ M I;
(2) T ∈ GL+(H) i.e.T is invertible and positive.
Proof. (1) ⇒ (2) Since mI ≤ T , then 0 < (T − mI), therefore
0 < h(T − mI)f, fi = hT f, fi − hmf, fi
. Hence
therefore T is positive.
0 < mhf, fi ≤ hT f, fi,
Since mI ≤ T ≤ M I, then 0 ≤ I − M−1T ≤ (cid:0) M−m
M (cid:1). Therefore kI −
M−1Tk ≤(cid:0) M−m
M (cid:1) < 1. Now by Theorem 3.8 the operator T is invertible.
(1) ⇒ (2) Let kTk = M . Since kT hk ≤ kTkkhk for every h ∈ H, then
kT hk ≤ Mkhk for every h ∈ H, therefore T < M I. Also,
hence
khk = kT −1T hk ≤ kT −1kkT hk,
kT −1k−1khk ≤ kT hk
for every h ∈ H. Suppose that m = kT −1k−1. Since T is positive, mI ≤
T .
(cid:3)
The following proposition shows that any C-controlled frame is a con-
trolled frame.
Proposition 3.10. Let sequence Ψ = {ψj ∈ H : j ∈ J} be C-controlled
frame in Hilbert C*-module H for C ∈ GL(H). Then Ψ is a frame in
Hilbert C*-module H. Furthermore CS = SC∗ and so
hf, Cψjiψj .
Xj∈J
hf, ψjiCψj =Xj∈J
Proof. Define S := C−1SC. Then for every f ∈ H
Sf = C−1SC f = C−1Xj∈J
hf, ψjiCψj =Xj∈J
hf, ψjiψj .
The operator S : H → H is positive, invertible and self-adjoint. Now by
Theorem 3.7 and Theorem 3.9 Ψ is a frame.
Since the operator SC is self-adjoint and SC = CS, then CS = SC =
(cid:3)
S∗C = S∗C∗ = SC∗.
According to the following proposition, if the operator C is self-adjoint,
then C-controlled frames are equivalent to classical frames. this is a gener-
alization of Proposition 3.3 in [5] for Hilbert C∗-module setting.
Proposition 3.11. Let H be a Hilbert C∗-module and C ∈ GL(H) be self-
adjoint. Then {ψj ∈ H : j ∈ J} is a C- controlled frame for H if and only
if Ψ is a frame for H and C is positive and commutes with frame operator
S .
10
M. RASHIDI-KOUCHI, A. RAHIMI
Proof. Let {ψj : j ∈ J} be a C-controlled frame for H. Then from Proposi-
tion 3.10 {ψj : j ∈ J} is a frame H and C is commutes with frame operator
S. Therefore C = SC C−1 is positive.
For the converse, we note that, if {ψj : j ∈ J} is a frame, then the frame
operator S is positive and invertible. Therefore CS = SC is positive and
invertible. Now, by Theorem 3.9 {ψj : j ∈ J} is a C-controlled frame for
H.
4. Multipliers of controlled frames and weighted frames in
(cid:3)
Hilbert C∗-modules
In this section, we generalize the concept of multipliers of frames for con-
trolled frames in Hilbert C∗-module. Then we investigate weighted frames
in Hilbert C∗-modules and verify their relation to controlled frames and
multiplier operators.
Proposition 4.1. Let H be a Hilbert C∗-module and C ∈ GL(H). Assume
Φ = {φj ∈ H : j ∈ J} and Ψ = {ψj ∈ H : j ∈ J} are C-controlled Bessel
sequence for H. Then the operator
Mm,Φ,Ψ : H → H
defined by
Mm,Φ,Ψf =Xj∈J
is a well defined bounded operator.
mjhf, ψjiCφj
Proof. Let Φ = {φj ∈ H : j ∈ J} and Ψ = {ψj ∈ H : j ∈ J} be C-controlled
Bessel sequences for H with bounds D and D′, respectively. For any f, g ∈ H
and finite subset I ⊂ J,
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Xi∈I
mihf, ψiiCφi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
1
mihf, ψiihCφi, gi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
= supg∈H,kgk=1(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Xi∈I
≤ supg∈H,kgk=1(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
2 Xi∈I
mi2hf, ψii2!
Xi∈I
≤ supg∈H,kgk=1kmk∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
2 Xi∈I
hf, ψii2!
Xi∈I
√DD′kfk
≤ kmk∞
1
1
2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
hCφi, gi2!
2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
hCφi, gi2!
1
This show that Mm,Φ,Ψ is well defined and
kMm,Φ,Ψk ≤ kmk∞
√BB′.
(cid:3)
Above lemma is a motivation to define the following definition.
ON CONTROLLED FRAMES IN HILBERT C ∗-MODULES
11
Definition 4.2. Let H be a Hilbert C∗-module and C ∈ GL(H). Assume
Ψ = {ψj ∈ H : j ∈ J} and Φ = {φj ∈ H : j ∈ J} are C-controlled Bessel
sequences for H. Then the operator
defined by
Mm,Φ,Ψ : H → H
Mm,Φ,Ψf =Xj∈J
mjhf, ψjiCφj
is called the C-controlled multiplier operator with symbol m.
The following definition is a generalization of weighted frames in Hilbert
space to Hilbert C∗-module.
Definition 4.3. Let Ψ = {ψj ∈ H : j ∈ J} be a sequence of elements in
Hilbert C∗-module H and {ωj}j∈J ⊆ Z(A) a sequence of positive weights.
This pair is called a w-frame of Hilbert C∗-module H if there exist constants
C, D > 0 and such that
Chf, fi ≤Xj∈J
ωjhf, ψjihψj , fi ≤ Dhf, fi.
for all f ∈ H.
A sequence{cj : j ∈ J} ∈ Z(A) is called semi-normalized if there are
The following proposition gives a relation between controlled frames,
bounds b ≥ a > 0, such that a ≤ cn ≤ b.
weighted frames and multiplier operators.
Proposition 4.4. Let H be a Hilbert C∗-module and C ∈ GL(H) be self-
adjoint and diagonal on Ψ = {ψj ∈ H : j ∈ J} and assume it generates a
controlled frame. Then the sequence W = {ωj}j∈J ⊆ Z(A), which verifies
the relations Cψn = ωnψn, is semi-normalized and positive. Furthermore
C = MW, Ψ,Ψ.
Proof. By Theorem 3.9, we get the following result for C 1/2:
mkfk2 ≤ kC 1/2fk2 ≤ Mkfk2.
As Cψj = ωjψj, clearly C 1/2ψj = √ωjψj. Applying the above inequalities
to the elements of the sequence, we get 0 < m ≤ ωj ≤ M .
Clearly, for any f ∈ H
Cf = C
Xj∈J
=Xj∈J
hf, ψjiωj ψj = MW, Ψ,Ψf.
=Xj∈J
hf, ψjiψj
hf, ψjiCψj
(cid:3)
12
M. RASHIDI-KOUCHI, A. RAHIMI
As a to the first part of Proposition 4.4 a frame weighted by semi-
normalized sequence is always a frame. Indeed, we have the following lemma.
Lemma 4.5. Let {ωj : j ∈ J} be a semi-normalized sequence with bounds a
and b. If {ψj : j ∈ J} is a frame with bounds C and D in Hilbert C∗-module
H, then {ωj ψj : j ∈ J} is also a frame with bounds a2C and b2D. The
sequence {ω−1
Proof. Since for any f ∈ H hf, ωjψji2 = ωj2hf, ψji2, we get
ψj : j ∈ J} is a dual frame of {ωjψj : j ∈ J}.
j
∆ :=Xj∈J
hf, ωjψji2 =Xj∈J
ωj2hf, ψji2.
Thus ∆ ≤ b2Pj∈J hf, ωjψji2 ≤ b2Dkfk2. In addition,
hf, ψji2 ≥ a2Ckfk2.
∆ ≥ a2Xj∈J
ψj =Pj∈Jhf.ψji ψj = f , these two sequences are dual.
ψj : j ∈ J} is a Bessel sequence dual to a frame.
j
(cid:3)
AsPj∈Jhf, ωjψjiω−1
Since ω−1
Therefore, it is a dual frame of {ωjψj : j ∈ J}.
is bounded, {ω−1
j
j
The following results give a connection between weighted frames and
frame multipliers.
Lemma 4.6. Let Ψ = {ψj ∈ H : j ∈ J} be a frame for Hilbert C∗-module
H. Let m = {mj}j∈J be a positive and semi-normalized sequence. Then the
multiplier Mm,Ψ is the frame operator of the frame {√mj ψj : j ∈ J} and
therefore it is positive, self-adjoint and invertible. If {mj}j∈J is negative
and semi-normalized, then Mm,Ψ is negative, self-adjoint and invertible.
Proof.
Mm,Ψf =Xj∈J
mjhf, ψjiψj =Xj∈J
hf,√mjψji√mj ψj.
positive and invertible.
By Lemma 4.5, {√mj ψj : j ∈ J} is a frame. Therefore Mm,Ψ = S(√mj ψj) is
Let mj < 0 for all j ∈ J, then mj = −pmj2. Therefore
hf,√mj ψji√mjψj = −S(√mj ψj).
Mm,Ψ = −Xj∈J
Theorem 4.7. Let {ψj : j ∈ J} be a sequence of elements in Hilbert C∗-
module H. Let W = {ωj : j ∈ J} be a sequence of positive and semi-
normalized weights. Then the following properties are equivalent:
(1) {ψj : j ∈ J} is a frame;
(2) Mw,Ψ is a positive and invertible operator;
(cid:3)
ON CONTROLLED FRAMES IN HILBERT C ∗-MODULES
13
(3) There are constants C, D > 0 such that for all f ∈ H
Chf, fi ≤Xj∈J
ωjhf, ψjihψj , fi ≤ Dhf, fi.
i.e. the pair {ωj : j ∈ J},{ψj : j ∈ J} forms a weighted frame;
(5) Mw,Ψ is a positive and invertible operator for any positive, semi-
(4) {√ωjψj : j ∈ J} is a frame;
normalized sequence W ′ = {ω
weighted frame.
′
j}j∈J ;
(6) (ωj ψj) is a frame, i.e. the pair {ωj : j ∈ J},{ψj : j ∈ J} forms a
Proof. It is similar to Theorem 4.5 of [5].
(cid:3)
Acknowledgement: The authors would like to thank the refree for use-
ful and helpful comments and suggestions.
References
[1] L. Arambaic, On frames for countably generated Hilbert C ∗-modules, Proc. Amer.
Math. Soc. 135, 469 -- 478, (2007).
[2] M. L. Arias and M. Pacheco, Bessel fusion multipliers. Journal of Mathematical
Analysis and Applications , 348(2), 581 -- 588, (2008).
[3] P. Balazs, Regular and Irregular Gabor Multipliers with Application to Psychoacoustic
Masking, Ph.D Thesis University of Vienna, (2005).
[4] P. Balazs, Basic definition and properties of Bessel multipliers, Journal of Mathemat-
ical Analysis and Applications, 325(1), 571 -- 585, (2007).
[5] P. Balazs, J-P. Antoine and A. Grybos, Wighted and Controlled Frames. Interna-
tional Journal of Wavelets, Multiresolution and Information Processing, 8(1), 109 --
132, (2010).
[6] P. Balazs, D. Bayer and A. Rahimi, Multipliers for continuous frames in Hilbert
spaces, J. Phy. A: Mathematical and Theoretical, 45, (2012).
[7] I. Bogdanova, P. Vandergheynst, J.P. Antoine, L. Jacques, M. Morvidone, Stereo-
graphic wavelet frames on the sphere, Applied Comput. Harmon. Anal. (19), 223 -- 252,
(2005).
[8] O. Christensen, An Introduction to Frames and Riesz Bases, Birkhauser, Boston,
(2003).
[9] J. B. Conway, A Course in Functional Analysis, 2nd edn. (Springer, New York, 1990).
[10] I. Daubechies, A. Grossmann,Y. Meyer, Painless nonorthogonal expansions, J. Math.
Phys. 27 , 1271 -- 1283, (1986).
[11] R. J. Duffin, A.C. Schaeffer, A class of nonharmonic Fourier series, Trans. Amer.
Math. Soc. 72, 341-366, (1952).
[12] M. H. Faroughi, E. Osgooei, A. Rahimi, Some properties of (Xd, X ∗
d ) and
d )-Bessel multipliers, Azarbijan Journal of Mathematics, 3(2), 70 -- 78,
(l∞, Xd, X ∗
(2013).
[13] M. H. Faroughi, E. Osgooei, A. Rahimi, (Xd, X ∗
Banach J. Math. Anal., 7(2), 146 -- 161, (2013).
d )-Bessel Multipliers in Banach spaces,
[14] M. Frank, D.R. Larson, A module frame concept for Hilbert C ∗-modules, in: Func-
tional and Harmonic Analysis of Wavelets, San Antonio, TX, January 1999, Contemp.
Math. 247, Amer. Math. Soc., Providence, RI 207 -- 233, (2000).
[15] M. Frank, D.R. Larson, Frames in Hilbert C ∗-modules and C ∗-algebras, J. Operator
Theory 48, 273 -- 314, (2002).
14
M. RASHIDI-KOUCHI, A. RAHIMI
[16] D. Han, W. Jing, D. Larson, R. Mohapatra, Riesz bases and their dual modular frames
in Hilbert C ∗-modules, J. Math. Anal. Appl. 343, 246 -- 256, (2008).
[17] D. Han, W. Jing, R. Mohapatra, Perturbation of frames and Riesz bases in Hilbert
C ∗-modules, Linear Algebra Appl. 431, 746 -- 759, (2009).
[18] W. Jing, Frames in Hilbert C ∗-modules, Ph. D. thesis, University of Central Florida
Orlando, Florida (2006).
[19] G. G. Kasparov, Hilbert C ∗-modules:
Operator Theory, 4, 133 -- 150, (1980).
theorems of Stinespring and Voiculescu, J.
[20] I. Kaplansky, Algebra of type I, Ann. Math., 56, 460 -- 472, (1952).
[21] A. Khosravi, B. Khosravi, Frames and bases in tensor products of Hilbert spaces and
Hilbert C ∗-modules, Proc. Indian Acad. Sci. Math. Sci. 117, 1 -- 12, (2007).
[22] A. Khosravi, B. Khosravi, g-frames and modular Riesz bases in Hilbert C ∗-modules,
Int. J. Wavelets Multiresolut. Inf. Process., 10(2), 1 -- 12, (2012).
[23] A. Khosravi, M. Mirzaee Azandaryani, Bessel multipliers in Hilbert C ∗-modules, Ba-
nach J. Math. Anal., 9(3), 153 -- 163, (2015).
[24] E. C. Lance, Hilbert C ∗-Modules: A Toolkit for Operator Algebraists, London Math.
Soc. Lecture Note Ser. 210, Cambridge Univ. Press, (1995).
[25] A. Rahimi, Multipliers of Genralized frames in Hilbert spaces, Bulletin of the Iranian
Mathematical Society, Vol., 37(1), 63 -- 80, (2011).
[26] A. Rahimi and P. Balazs, Multipliers of p-Bessel sequences in Banach spaces, Integral
Equations and Operator Theory, 68(2), 193 -- 205, (2010).
[27] A. Rahimi and A. Fereydooni, Controlled G-Frames and Their G-Multipliers in
Hilbert spaces An. St. Univ. Ovidius Constanta, 21(2), 223 -- 236, (2013).
[28] M. Rashidi-Kouchi, A. Nazari, Equivalent continuous g-frames in Hilbert C ∗-modules,
Bull. Math. Anal. Appl., 4 (4), 91 -- 98, (2012).
[29] M. Rashidi-Kouchi, A. Nazari, Continuous g-frame in Hilbert C ∗-modules, Abst.
Appl. Anal., 2011, 1 -- 20, (2011).
[30] M. Rashidi-Kouchi, A. Nazari, M. Amini, On stability of g-frames and g-Riesz bases
in Hilbert C ∗-modules, Int. J.Wavelets Multiresolut. Inf. Process, 12(6), 1 -- 16, (2014).
[31] R. Schatten, Norm Ideals of Completely Continious Operators. Springer Berlin,
(1960).
[32] D. Wang and Guy J. Brown, Computational Auditory Scene Analysis: Principles,
Algorithms, and Applications. Wiley-IEEE Press, (2006).
[33] N. E. Wegge-Olsen, K-theory and C*-algebras a Friendly Approach, Oxford University
Press, Oxford, England, (1993).
Department of Mathematics, Kahnooj Branch, Islamic Azad University,
Kerman, Iran.
E-mail address: m−[email protected]
Department of Mathematics, University of Maragheh, P. O. Box 55181-
83111, Maragheh, Iran.
E-mail address: [email protected]
|
1101.4534 | 2 | 1101 | 2011-02-04T13:12:50 | Growth rates of dimensional invariants of compact quantum groups and a theorem of Hoegh-Krohn, Landstad and Stormer | [
"math.OA",
"math.QA"
] | We give local upper and lower bounds for the eigenvalues of the modular operator associated to an ergodic action of a compact quantum group on a unital C*-algebra. They involve the modular theory of the quantum group and the growth rate of quantum dimensions of its representations and they become sharp if other integral invariants grow subexponentially. For compact groups, this reduces to the finiteness theorem of Hoegh-Krohn, Landstad and Stormer. Consequently, compact quantum groups of Kac type admitting an ergodic action with a non-tracial invariant state must have representations whose dimensions grow exponentially. In particular, S_{-1}U(d) acts ergodically only on tracial C*-algebras. For quantum groups with non-involutive coinverse, we derive a lower bound for the parameters 0<\lambda<1 of factors of type III_\lambda that can possibly arise from the GNS representation of the invariant state of an ergodic action with a factorial centralizer. | math.OA | math | Growth rates of dimensional invariants of compact
quantum groups and a theorem of Høegh-Krohn,
Landstad and Størmer
Claudia Pinzari
Dedicated to the memory of Claudio D'Antoni
.
A
O
h
t
a
m
[
2
v
4
3
5
4
.
1
0
1
1
:
v
i
X
r
a
Abstract
We give local upper and lower bounds for the eigenvalues of the mod-
ular operator associated to an ergodic action of a compact quantum group
on a unital C ∗ -- algebra. They involve the modular theory of the quantum
group and the growth rate of quantum dimensions of its representations
and they become sharp if other integral invariants grow subexponentially.
For compact groups, this reduces to the finiteness theorem of Høegh-Krohn,
Landstad and Størmer. Consequently, compact quantum groups of Kac type
admitting an ergodic action with a non-tracial invariant state must have rep-
resentations whose dimensions grow exponentially. In particular, S−1U (d)
acts ergodically only on tracial C ∗ -- algebras. For quantum groups with non-
involutive coinverse, we derive a lower bound for the parameters 0 < λ < 1
of factors of type IIIλ that can possibly arise from the GNS representation
of the invariant state of an ergodic action with a factorial centralizer.
1 Introduction
In the early 80's Høegh-Krohn, Landstad and Størmer proved that the multiplicity
of an irreducible representation of a compact group acting ergodically on a unital
C∗ -- algebra is bounded above by its dimension and moreover the unique invariant
state is a trace [7]. This result is often used as a finiteness criterion in operator
algebras. Moreover, Wassermann, starting from this, showed the negative result
that SU(2) acts ergodically only on type I von Neumann algebras [14].
If we consider compact quantum groups instead of compact groups, ergodic
theory on operator algebras becomes much richer. For example, as is well known,
finiteness fails, as compact quantum groups may have non-involutive coinverse
and, in this case, the Haar state has a non-trivial modular theory [15]. Boca
generalized some of these results to ergodic actions of compact quantum groups.
He proved that the quantum dimension is an upper bound for the multiplicity and
that the invariant state satisfies the KMS property [3]. Wang has found many
examples of ergodic actions of the free unitary quantum groups on factors of type
1
1 INTRODUCTION
2
II and III [13]. In particular, his examples show that finiteness fails already in
the case of compact quantum groups with involutive coinverse, often called of
Kac type, even though these have trivial modular theory. Indeed, Au(n) can act
ergodically on factors on type III 1
. Bichon, De Rijdt and Vaes have constructed
actions of Ao(F ) with multiplicities larger than the integral dimensions [2].
In
joint papers with Roberts, we have shown that any finite index inclusion of factors
of type II1 gives rise to ergodic actions of Ao(F ) [10]. Moreover, the classification
of ergodic actions of compact quantum groups is related to the classification of
tensor C∗ -- categories with conjugates [11].
n
The aim of this note is to show an analogue of the second part of the finiteness
theorem of [7] for compact quantum groups.
The examples of [2], as well as subsequent developments of the general the-
ory of ergodic actions [9], contributed to a further understanding of the invariant
state. It became clear, for example, that it is always almost periodic in the sense
of Connes [5], meaning that the associated modular operator is diagonal, and that
its eigenvalues exhibit an explicit, separated, dependence on both the modular
theory of the quantum group and the eigenvalues of certain strictly positive matri-
ces canonically associated to the irreducible represenations of the quantum group
belonging in the spectrum of the action.
A consequence of this is that, in the special case where the invariant state of
an action of a non-Kac type compact quantum group is factorial and has factorial
centralizer, then the associated factor is necessarily of type IIIλ with 0 < λ ≤ 1
(cf. Sect. 3 for a more precise statement). Note that the invariant state of the
mentioned examples of Wang satisfy this factoriality condition. Another class of
examples has been recently studied by Vaes and Vergnioux [12]. They considered
the translation action of Ao(F ) over itself and they were able to prove, among
other things, that if F has rank at least 3 and satisfies a suitable condition, then
the Haar state is factorial and moreover the factor generated by the the GNS
construction is full in the sense of Connes [5].
For a general ergodic action, the problem of studying modularity of the in-
variant state becomes that of studying the spectra of the associated matrices. In
retrospect, the theorem of [7] amounts to showing that these matrices are always
trivial in the classical case.
To study these spectra, we introduce two growth rates of dimensional invariants
of the compact quantum group, that of integral dimensions, Dimu, and of quantum
dimensions, Du. These growth rates distinguish between SU (2), SqU (2) for 0 <
q < 1, Ao(F ), for rank(F ) ≥ 3, and Au(F ). More precisely, in the group case,
quantum dimensions are just the integral dimensions, and therefore the growth
rate is always polynomial, as a consequence of Weyl's dimension formula, hence
Du = Dimu = 1 for all representations. This polynomial growth rate played an
important role in the original proof of the finitenss theorem of [7].
Growth rate of quantum dimensions for SqU (2), 0 < q < 1 is instead exponen-
tial. However, integral dimensions of irreducibles are the same as in the classical
case, hence they grow polynomially. This is opposed to Ao(F ), for rank(F ) ≥ 3, for
which both growth rates are exponential. In the case of Au(F ), both dimensional
invariants of the fundamental representation have the largest possible growth rates
2 PRELIMINARIES
3
as all tensor powers of this representation are irreducible.
We shall see that if the growth rate of the integral dimensions is subexponential,
then that of quantum dimensions, Du, is explicitly determined by the eigenvalues
of the modular operator of the quantum group (Prop. 4.3). However, this fact
is not generally true, and Ao(n), for n ≥ 3, or Au(n), for n ≥ 2, are the first
examples.
Given an ergodic action of a compact quantum group on a unital C∗ -- algebra,
our main result consists of showing that Du and D−1
u are upper and lower bounds
for the spectral radii of the mentioned matrices involved in the modular operator
of the ergodic action (see Theorem 4.8). In particular, we reproduce the finiteness
theorem of [7] from the fact that Du = 1 in the group case. Our bounds become
equalities in the case of ergodic actions with high quantum multiplicities in the
sense of [2], but integral multiplicities with subexponential growth rate.
We shall show that the invariant D determines the parameter λ = 1
n of the
Connes invariant S in Wang's example with Au(n). In other words, the rate of
growth of dimensions clarifies which compact quantum groups of Kac type can
act on infinite factors. More precisely, one of the consequences of our result is
that, among compact quantum groups of Kac type, only those having some irre-
ducible representation whose integral dimension grows exponentially can possibly
act on C∗ -- algebras with a non-tracial invariant state. For example, all ergodic
C∗ -- algebras for S−1U (d) are tracial.
For compact quantum groups with non-involutive coinverse, we derive a general
lower bound for the possible parameters 0 < λ < 1 of the type IIIλ factors than can
arise from the GNS representation of the invariant state, provided the centralizer
algebra is a factor. This lower bound involves the modular theory of the quantum
group, the growth rates Du, and the spectrum of the action, see Cor. 4.13. We
thus see that many parameters λ may be excluded if some spectral information is
known. For example, for SqU (2), with 0 < q < 1 then λ ≥ q2r, where r labels
the first spectral irreducible representation. For Ao(F ), with F of rank ≥ 3, then
kF k2 )r where q + 1
q := Trace(F ∗F ) and r is as before. As another example,
λ ≥ (
λ ≥ min{q0,q−1
Trace(F ∗F ) , for Au(F ), with q0 and qn the smallest and largest eigenvalues of
F ∗F if the fundamental representation is spectral.
n }
q
Our methods differ from the original proof of the classical finiteness theorem.
They rely on the duality theorem of [9], which allows a rather simple presentation.
This paper is organized as follows. Sect. 2 is devoted to the preliminaries: we
recall the notion of standard solutions of the conjugate equations and the duality
theorem for ergodic actions.
In Sect. 3 we describe preliminary results on the
modular theory of the invariant state. In Sect. 4 we discuss the main result and
some corollaries.
2 Preliminaries
Standard solutions of the conjugate equations. We shall briefly recall the main
features of tensor C∗ -- categories with conjugates [8]. Let A be a tensor C∗ -- category
(always assumed to be strict, with irreducible tensor unit ι, subobjects and direct
2 PRELIMINARIES
4
sums). Arrows R ∈ (ι, u ⊗ u) and R ∈ (ι, u ⊗ u) are said to define a conjugate u
of the object u if they satisfy the conjugate equations
∗
R
⊗ 1u ◦ 1u ⊗ R = 1u, R∗ ⊗ 1u ◦ 1u ⊗ R = 1u.
If Ru, Ru and Rv, Rv are solutions for u and v then Ru⊗v := 1v ⊗ Ru ⊗ 1v ◦ Rv,
Ru⊗v := 1u ⊗ Rv ⊗ 1u ◦ Ru are solutions for u ⊗ v. Similarly, Ru := Ru, Ru := Ru
are solutions for u. They are called the tensor product and conjugate solutions
respectively. If conjugates exist for every object, then every object is the direct
sum of irreducible objects.
The category Rep(G) of (unitary, finite dimensional) representations of a com-
pact quantum group has conjugates and is embedded in the category of finite di-
mensional Hilbert spaces, hence arrows are linear maps. Solutions of the conjugate
equations take the form R = Pi jψi ⊗ ψi, R = Pk j−1φk ⊗ φk, with j : Hu → Hu
a unique invertible antilinear map between the representation Hilbert spaces and
(ψi), (φk) orthonormal bases. Conjugate and tensor product solutions correspond
to ju := j−1
u , ju⊗v = jv ⊗ juθ, with θ : Hu ⊗ Hv → Hv ⊗ Hu the flip map.
In a tensor C∗ -- category with conjugates it is convenient to select standard
solutions of the conjugate equations, meaning that kRuk = kRuk if u is an ir-
reducible object, while if u ≃ ⊕iui with ui irreducible, Ru := Pi Si ⊗ Si ◦ Rui
Ru := Pi Si ⊗ Si ◦ Rui , where Rui , Rui are standard solutions and {Si ∈ (ui, u)},
{Si ∈ (ui, u)} are two sets of isometries whose ranges are pairwise orthogonal add
up to 1u and 1u, respectively. Clearly, the conjugate of a standard solution is stan-
dard. Standard solutions are unique up to unitary equivalence. Most importantly,
they realize the minimal value of kRukkRuk among all solutions, and they are
characterized, up to scalars, by this property. This minimal value is the quantum
(or intrinsic) dimension of u, denoted by d(u). This implies the following fact,
which will play a role.
2.1. Theorem. [8] The tensor product of standard solutions is standard.
We shall also need the following fact. In the category Rep(G), if ju : Hu → Hu
defines a standard solution of the conjugate equations for u, j∗
uju is a positive
operator on Hu which does not depend on the choice of the standard solution. This
operator correspond to F −1
of Woronowicz [15]. It follows that the spectrum of
j∗
uju as well as its smallest and largest eigenvalues, denoted λu and Λu respectively,
are invariantly associated to u (in fact, to its equivalence class). Note that λu =
Λ−1
u , λu⊗v = λuλv, and similarly for Λu⊗v.
u
Ergodic actions, spectral functor and quasitensor functors. To a certain extent,
we may think of the relationship between an ergodic action of a compact quan-
tum group on a unital C∗-algebra, the associated spectral functor and an abstract
quasitensor functor, as analogous to that between a Lie group, the associated Lie
algebra and an abstract Lie algebra. The analogy is supported by the following
properties of ergodic actions proved in [9], and referred to as a duality theorem for
ergodic actions. The spectral functor of an ergodic action is a quasitensor functor.
For any ergodic action, there always is a maximal ergodic action which has the
same spectral functor as that of the original one. It is the completion in the max-
imal C∗ -- norm of the dense spectral subalgebra of the given ergodic action. The
2 PRELIMINARIES
5
maximal ergodic action with a given spectral functor is unique and canonically
associated with it. Any abstract quasitensor functor from the representation cat-
egory of G to the Hilbert spaces is the spectral functor of an ergodic action. Two
maximal ergodic actions of a given compact quantum group G are conjugate if
and only if the associated spectral functors are related by a unitary natural trans-
formation. Hence quasitensor functors Rep(G) → Hilb classify maximal ergodic
C∗ -- actions of G.
In some more detail, let G be a compact quantum group [16] and
α : C → C ⊗ Q
an ergodic action of G on a unital C∗ -- algebra C (i.e. Cα := {c ∈ C, α(c) = c⊗ I} =
C). Q denotes the Hopf C∗ -- algebra of G. Let u be a representation of G on the
Hilbert space Hu. Consider the space of linear maps T : Hu → C intertwining u
with the action α. The map taking u to this space of intertwiners is a functor from
the representation category of G to the category of vector spaces. However, it is
a contravariant functor, hence for convenience we consider the covariant functor
obtained passing to the category of dual vector spaces. This covariant functor
will be donoted by L (while we used the notation L in previous papers). As a
consequence of ergodicity, the predual of Lu, and hence of Lu itself, is a Hilbert
space. We may explicit Lu = {Pi ψi ⊗ c, α(ci) = Pj cj ⊗ u∗
j,i, ψi o. b.}
and thus think of Lu as the space of fixed points (Hu ⊗ C)u⊗α with inner product
arising from the restriction of the C -- valued inner product of the free Hilbert module
Hu ⊗ C. If A ∈ (u, v) is an intertwiner in Rep(G), LA acts as A ⊗ I from Lu to
Lv.
A ∗ -- functor µ : A → M between two tensor C∗ -- categories is called quasitensor
if there are isometries µu,v ∈ (µu ⊗ µv, µu⊗v), such that
µι = ι,
µu,ι = µι,u = 1µu ,
u,v⊗w ◦ µu⊗v,w = 1µu ⊗ µv,w ◦ µ∗
µ∗
u,v ⊗ 1µw
and natural in u, v,
µ(S ⊗ T ) ◦ µu,v = µu′,v′ ◦ µ(S) ⊗ µ(T ),
(2.1)
(2.2)
(2.3)
(2.4)
for objects u, v, w, u′, v′ of A and arrows S ∈ (u, u′), T ∈ (v, v′). This definition
was given in [9]. Note that the most relevant axiom, (2.3), implies associativity:
µu,v,w := µu⊗v,w ◦ µu,v ⊗ 1µw = µu,v⊗w ◦ 1µu ⊗ µv,w. If all the isometries µu,v
are unitary, (2.3), is equivalent to associativity. In this case (µ, µ) will be called
a relaxed tensor functor. Examples arise from pairs of non-isomorphic compact
quantum groups with tensor equivalent representation categories. A well known
class of examples is SµU (2) and Ao(F ) for suitable conditions on F (cf. Example
4.6). Composition of an equivalence Rep(G) → Rep(G′) with the embedding
functor of Rep(G′) is a relaxed tensor functor Rep(G) → Hilb.
For the spectral functor L of an ergodic action, isometries making it quasitensor
are given by Lu,v(P ψi⊗ci)⊗(φj ⊗dj) = P(ψi⊗φj )⊗dj ci. Note that the spectral
2 PRELIMINARIES
6
space of a non spectral representation is trivial, hence, quasitensor functors, unlike
the relaxed tensor ones, may take a nonzero object to the zero object.
A quasitensor functor (µ, µ) preserves conjugates, in the sense that if a nonzero
object u of A has a conjugate defined by arrows R ∈ (ι, u ⊗ u) and R ∈ (ι, u ⊗ u)
then µu is a conjugate of µu, defined by R = µ∗
u,u ◦ µ(R). This
is a straightforward consequence of the axioms. The property of conservation of
u,u ◦ µ(R), R = µ∗
conjugates of a quasitensor functor µ implies d(µu) ≤ k Rkk Rk ≤ d(u).
In particular, for the spectral functor of an ergodic action we may associate
an antilinear invertible map Ju : Lu → Lu to a solution (Ru, Ru) of the conjugate
equations of a representation u of G by
X
k
JuTk ⊗ Tk = L∗
u,u ◦ L(Ru),
where Tk is an orthonormal basis of Lu. The above relation between intrinsic
dimensions becomes a multiplicity bound of an irreducible representation in the
spectum of the related ergodic action. Indeed, the scalar k Rkk Rk arising from the
solution of the conjugate equations of an irreducible representation u, reduces to
the quantum multiplicity q − mult(u) of [2], while the intrinsic dimension of Lu is
just the integral dimension of Lu, i.e. the ordinary multiplicity mult(u) of u in the
action. The above estimate says mult(u) ≤ q-mult(u) ≤ d(u) a fact shown in [2]
refining the inequality mult(u) ≤ d(u) previously obtained by [3]. In the classical
case, i.e. when G is a compact group, the quantum dimension of a representation
is just its integral dimension, and we thus in turn recover the classical result of
Høegh-Krohn Landstad and Stormer that mult(u) ≤ dim(Hu).
We shall need the following facts proved in [9].
2.2. Lemma. Let G be a compact quantum group acting ergodically on a unital
C∗ -- algebra with spectral functor (L, L). If Ju and Jv are associated to solutions
ju and jv of the conjugate equations for representations u and v of G respectively,
then
a) Ju⊗v Lu,v = Lv,uJv ⊗ JuΘ, where Ju⊗v is associated to the tensor product
solution for u ⊗ v and Θ : Lu ⊗ Lv → Lv ⊗ Lu is the flip map,
b) for any A ∈ (u, v), L(jvAj−1
u )Ju = JvL(A).
Recall that conversely, given a quasitensor functor (µ, µ) : Rep(G) → Hilb, we
may associate a maximal ergodic action of G in the following way. Form the linear
space ◦Cµ := Pu∈Rep(G) µu ⊗Rep(G) Hu, where ⊗Rep(G) indicates a suitable tensor
product treating the arrows of Rep(G) as scalars. The algebraic operations are
defined by, dropping the indices and the natural transformation,
(k ⊗ ψ)(k′ ⊗ ψ′) = (k ⊗ k′) ⊗ (ψ ⊗ ψ′),
(k ⊗ ψ)∗ = J k ⊗ j−1∗ψ.
(2.5)
The action of the quantum group on each subspace µu ⊗ Hu is the tensor product
of the trivial action on the first factor and the representation u on the second.
This action is ergodic.
3 MODULAR THEORY OF THE INVARIANT STATE
7
Most importantly, the linear functional ω on ◦Cµ which annihilates each sub-
space µu ⊗ Hu, u ∈ G, u 6= ι and takes I to 1 is a positive and faithful state, it is
the unique state invariant under the action (i.e. ω ⊗ id ◦ α = ω) . Therefore ◦Cµ
has a C∗ -- norm. It turns out that the maximal C∗ -- norm is finite. We thus have
at our disposal two possible completions of ◦Cµ, which are different in general, the
completion in the maximal C∗ -- norm, denoted Cµ, and the completion in the norm
provided by the GNS representation πω of the invariant state, called the reduced
completion.
The G -- action clearly extends to the maximal completion. In the case of the
reduced completion, note that the action of G on ◦Cµ is in fact only an action
of the dense spectral Hopf ∗ -- subalgebra. This action extends to an action of the
reduced compact quantum group Gred and in turn it lifts to a normal action of the
Hopf -- von Neumann algebra generated by the regular representation of G on the
von Neumann algebra πω(C)′′. In all these cases, the extended action is ergodic,
see Theorem 2.5 of [13].
3 Modular theory of the invariant state
In view of the duality theorem recalled in the previous section, we may and shall
think of the spectral functor of an ergodic C∗ -- action of a compact quantum group
G as an abstract quasitensor functor (µ, µ) : Rep(G) → Hilb. Correspondingly, we
shall represent the dense spectral subalgebra with generators and relations given
in (2.5).
If a = k ⊗ ψ, b = k′ ⊗ ψ′ have support in the irreducible representations u and
v respectively then
ω(a∗b) = δu,vkRuk−2(k′, Ju
ω(ba∗) = δu,vkRuk−2(k′, k)(ψ, (ju
∗Juk)(ψ, ψ′),
∗ju)−1ψ′),
(3.1)
(3.2)
see Sect. 8 in [9] for explicit computations. These formulas may be used to derive
modular properties of the invariant state, in turn generalizing the corresponding
properties of the Haar state [15, 3]. In fact, for every irreducible representation
u choose a standard solution (Ru, Ru). We have a densely defined multiplicative
map such that on µu ⊗ Hu,
σ−i(k ⊗ ψ) = (J ∗
uJu)k ⊗ j∗
ujuψ
with inverse
σi(k ⊗ ψ) = (J ∗
uJu)−1k ⊗ (j∗
uju)−1ψ.
(3.3)
Since a different standard solution is of the form U ju, with U unitary, the asso-
ciated Ju changes into µ(U )Ju, hence σ−i and its inverse do not change. A quick
computation shows that the KMS property holds,
ω(σ−i(b)a∗) = ω(a∗b).
3 MODULAR THEORY OF THE INVARIANT STATE
8
We may choose ju = j−1
It follows that σ−i(a∗) =
σi(a)∗. We collect the conclusions of the above discussion, a slight refinement of
a result of [15, 3] .
u , which implies Ju = J −1
u .
3.1. Theorem. The invariant state of an ergodic action of a compact quantum
group G on a unital C∗ -- algebra satisfies the KMS condition on the dense spectral
subalgebra. It is a trace if and only if for any spectral irreducible representation u
of G,
a) d(u) = dim(u), i.e. the antilinear ju defining a standard solution is antiuni-
tary, and
b) the associated Ju is antiunitary (hence q − mult(u) = mult(u)).
Recall that the condition d(u) = dim(u) for all u means precisely that the quantum
group is of Kac type, i.e.
it has involutive coinverse. For example, quantum
dimensions of the real deformations Gq, 0 < q < 1 of classical compact Lie groups
are known to be strictly larger than the corresponding integral dimensions, hence
the invariant state of any ergodic C∗ -- algebra under the action of any of these
quantum groups is not a trace.
Almost periodicity of the invariant state.
If (C, α, G) is an ergodic action of a
compact quantum group, the dense spectral subalgebra of C is the domain of a
one parameter group of ∗ -- automorphisms, the modular group, given by
σt(k ⊗ ψ) = (Ju
∗Ju)itk ⊗ (j∗
uju)−itψ.
This group extends to a one parameter automorphism group of the maximal com-
pletion Cµ. Moreover, it extends to πω(Cµ) or M := πω(Cµ)′′ as well since it leaves
ω invariant. In this subsection we only consider the extension to the von Neumann
algebra M . Note that the cyclic vector Ω associated to the GNS representation of
ω is separating for M . This may be shown with arguments similar to those of Sect.
4 of [6]. Hence, by the KMS property, σi becomes the restriction of the modular
operator ∆ω associated to ω under the canonical inclusion ◦Cµ → L2(C, ω). Recall
that Connes defined a normal faithful state φ on a von Neumann algebra to be
almost periodic if ∆φ is diagonal. In the case of an ergodic action, the expres-
sions for the inner product (3.1) and for σi, (3.3), show that the cyclic state of
M thus obtained is always almost periodic. Moreover, the point spectrum of ∆ω
is completely determined by the eigenvalues if j∗
uJu for u describing a
complete set of irreducible spectral representations. While the spectrum of j∗
uju is
a structural property of the quantum group, that of J ∗
uJu, depends on the ergodic
action, and this is the most mysterious part.
Indeed, although Ju is explicitly
associated with ju, we can not infer that properties of ju pass to Ju. For example,
we may have ju antiunitary for all spectral u but Ju not antiunitary, or, in other
words, we may have ergodic actions of compact quantum groups of Kac type on a
unital C∗ -- algebra C for which M = πω(C)′′ is a type III factor [13]. Similarly, we
may have ju not antiunitary for all spectral u but Ju always antiunitary, as in the
examples arising from subfactors, described in detail in [10].
uju and J ∗
4 AN ANALOGUE OF THE FINITENESS THEOREM FOR CQG
9
We conclude this section with a few more remarks on the modular theory of
(C, α, G). Let Sp(∆ω) (Spp(∆ω)) denote the spectrum (point spectrum) of ∆ω.
The following fact may be known, a proof is included for convenience.
3.2. Theorem. If an ergodic C∗ -- action of a compact quantum group G on C
admits a spectral irreducible representation u of G such that d(u) > dim(u) then
If in addition both M = πω(C)′′ and the centralizer Mω are
Spp(∆ω) 6= {1}.
factors, then M is of type IIIλ with 0 < λ ≤ 1.
Proof If u is a spectral irreducible representation for which ju is not antiunitary,
then j∗
uju has an eigenvalue < 1 and another > 1 since ju is standard. Hence ∆ω
has an eigenvalue 6= 1 on µu ⊗ Hu. If M and Mω are factors, it is well known
that S(M ) = Sp(∆ω), with S(M ) the Connes invariant (Cor. 3.2.7 in [4]). Hence
S(M ) 6= {1} and S(M ) 6= {0, 1}, so M is not semifinite or of type III0.
In the next section we shall give general lower and upper bounds for the eigen-
values of the modular operator ∆ω, depending only on the quantum group. We
shall derive a general lower bound for the possible parameters 0 < λ < 1 such that
M = πω(C)′′ is a factor of type IIIλ with a factorial centralizer Mω.
4 An analogue of the finiteness theorem for CQG
We have thus seen that the point spectrum of the modular operator ∆ω associ-
ated to an ergodic action is completely determined by the spectra of the positive
uJu) ⊗ (j∗
matrices (J ∗
uju) associated to the spectral irreducible representations u
of the quantum group.
In this section we derive a general estimate for the eigenvalues of the operators
J ∗
uJu. Our estimate involves the growth rate of the quantum dimension of u. It
allows to reproduce the finiteness result of [7], and also to derive other modular
properties of the invariant state.
We define the growth rate of the intrinsic dimension of an object u of a tensor
C∗ -- category,
Du,n := max {d(v), irreducible subobjects v of u⊗n},
Du := lim
n
(Du,n)1/n.
This limit always exists since Du,n is a submultiplicative sequence. Note that
Du ≤ d(u), Du = Dv if u and v are equivalent, and Du = Du. We say that
the intrinsic dimension of u has subexponential (exponential) growth if Du = 1
(Du > 1). Note that if w is a subrepresentation of some tensor power u⊗k of u,
Dw ≤ Dk
u. Hence, if the tensor powers of u contain all the irreducibles and if d(u)
has subexponential growth, so does every irreducible.
4.1. Example. Let G be a compact group. In this case the quantum dimension
of any representation u is just the integral dimension of the corresponding Hilbert
space. It is known that this dimension always has polynomial, and hence subex-
ponential, growth. This fact relies on Weyl's dimension formula and played an
important role in the original proof of the finiteness theorem of [7].
4 AN ANALOGUE OF THE FINITENESS THEOREM FOR CQG
10
Let (C, α) be an ergodic action of a compact quantum group G on a unital C∗ --
algebra. We shall also need to consider the growth rate of integral multiplicities
of irreducible representations of G. Given a representation u, set
Multu,n := max{mult(v), irreducible subrepresentations v of u⊗n},
Multu := lim inf
n
Mult1/n
u,n ,
which enjoys properties similar to those of Du.
In the particular case of the
translation action of G over itself, integral multiplicities mult(u) reduce to integral
dimensions dim(u). We shall accordingly denote the corresponding growth rate by
Dimu and refer to it as the growth rate of integral multiplicities. In both cases, we
have the notion of subexponential or exponential growth. A comparison between
the various growth rates introduced may be easily derived,
Multu ≤ Du, Dimu ≤ Du.
4.2. Example. Let G be a classical compact Lie group and Gq the associated
deformed compact matrix quantum group by a positive parameter 0 < q < 1.
The integral dimensions of irreducibles have subexponential growth as they are
the same as in the classical case. On the other hand, the quantum dimensions of
irreducibles have exponential growth rate. This last assertion is known. However,
it may also be derived from the following proposition. Indeed, Du = 1 would imply
λu = Λu = 1, hence ju antiunitary.
4.3. Proposition. For any representation u of a compact quantum group,
D−1
u ≤ λu ≤ Λu ≤ Du.
If Dimu = 1 then the first and last inequalities are equalities.
Proof The middle inequality being obvious, it suffices to show that Λu ≤ Du,
as λu ≥ D−1
follows passing to the conjugate representation. The n -- th tensor
product solution ju⊗n = (ju ⊗ ··· ⊗ ju)θn for u⊗n, with θn a suitable permutation
operator, is standard if ju is. If v is an irreducible subrepresentation of u⊗n and
jv is a standard solution for v, d(v) = Trace(j∗
u
v jv) ≥ Λv. Hence
Du,n ≥ max{Λv, v irreducible subrepresentation of u⊗n} =
Λu⊗n = Λn
u.
On the other hand d(v) ≤ Λvdim(v) implies
Du,n ≤ Λu⊗n Dimu,n = Λn
uDimu,n,
hence Du ≤ Λu if the integral dimension of u has subexponential growth rate.
Remark This proposition may be used to explain why the spectrum of the as-
sociated matrices j∗
uju has a symmetric shape for the deformed compact matrix
quantum groups Gq.
4 AN ANALOGUE OF THE FINITENESS THEOREM FOR CQG
11
For the Wang-Van Daele quantum groups Ao(F ) and Au(F ), we follow the no-
tation of [1]. We shall always normalize F so that Trace(F ∗F ) = Trace((F ∗F )−1).
Recall that for Ao(F ), the matrix F is required to satisfy F F = ±1.
4.4. Example. If u is the fundamental representation of Au(F ), with rank(F ) ≥
2, all tensor powers u⊗n are irreducible [1], showing that Du,n = d(u)n by multi-
plicativity of quantum dimension. Hence Du = d(u) = Trace(F ∗F ) is the largest
possible value. Similarly, Dimu = dim(u). In particular, the extreme inequalities
in Prop. 4.3 are always strict (the first examples being Au(m), λu = Λu = 1 but
Du = Dimu = m), while the middle inequality is generically strict for rank(F ) ≥ 3,
but it is an equality for rank(F ) = 2 due to Trace(F ∗F ) = Trace((F ∗F )−1).
ur
ur = D−1
ur = qr.
q = Trace(F ∗F ). Hence DAo(F )
4.5. Example. If u1 is the fundamental representation of G = SqU (2), for a
nonzero q ≤ 1, q1/2R = ψ1 ⊗ ψ2 − qψ2 ⊗ ψ1 is a standard solution, hence
u1 ju1 = diag(q,q−1). Let ur be the unique irreducible r + 1 -- dimensional rep-
j∗
resentation. The Clebsch -- Gordan rule u1 ⊗ ur ≃ ur−1 ⊕ ur+1 gives j∗
jur =
diag(qr,qr−2, . . . ,q−r). Hence λur = Λ−1
4.6. Example. Consider the quantum group Ao(F ). It is well known that the
representation categories of Ao(F ) and S∓qU (2) are tensor equivalent if q > 0
is defined by q + 1
= q−r by 4.5. Note that
λur = Λ−1
ur . This follows from the validity for r = 1 and the Clebsch-Gordan rule.
A computation shows that for every F and for the fundamental representation u1,
the extreme inequalities in Prop. 4.3 are strict iff rank(F ) ≥ 3.
4.7. Example. For rank(F ) ≥ 3, the integral dimension of the fundamental rep-
resentation u1 of Ao(F ) has exponential growth. This may be seen in the following
way. Independently of the matrix F , irreducible representations of Ao(F ) satisfy
the same fusion rules as those of SU (2) [1]. Hence, denoting with the same symbol
the corresponding irreducible representations, the integral dimensions are deter-
mined by the Clebsch -- Gordan rule, dim(ur+1) = dim(u1)dim(ur) − dim(ur−1) ≥
2dim(ur), hence Dimu1,r = dim(ur) ≥ 2r−1dim(u1).
ur
The following is our main result.
4.8. Theorem. For any spectral irreducible representation u of an ergodic action
of a compact quantum group on a unital C∗ -- algebra,
a) D−1
u ≤ J ∗
uJu ≤ Du, where Ju is associated to a standard solution ju for u.
b) If the spectral functor is relaxed tensor and if Multu = 1 then 1
Du
respectively the smallest and the largest eigenvalues of J ∗
uJu.
and Du are
Proof a) The first inequality follows from the second applied to u. By Lemma 2.2
a), for any positive integer n, if Ju is associated to any solution ju for u,
kJ ∗
u,...,uJ ∗
uJukn = kJ ∗
uJu ⊗ ··· ⊗ J ∗
u⊗n Ju⊗n µu,...,uk ≤ kJ ∗
u Juk =
u⊗n Ju⊗nk,
kµ∗
4 AN ANALOGUE OF THE FINITENESS THEOREM FOR CQG
12
where Ju⊗n is associated to the n -- th tensor product solution ju⊗n = (ju⊗···⊗ju)θn
for u⊗n. Consider a complete reduction of u⊗n into a direct sum of irreducible
representations v and let Sv ∈ (v, u⊗n) be the isometry associated to v. We
may compute kJ ∗
u⊗nJu⊗nk as the norm of the positive operator -- valued matrix
(µ(S∗
v )Jv = Ju⊗n µ(Sv), with
respect to any solution jv for v, with associated Jv. Now we fix standard solu-
tions for u and all the v. Since ju⊗n is standard, we may find {Sv} such that
ju⊗n Svj−1
u⊗n Ju⊗n µ(Sw))v,w. By lemma 2.2 b), µ(ju⊗n Svj−1
v =: Sv are pairwise orthogonal isometries. Hence
v )J ∗
µ(S∗
v )J ∗
u⊗n Ju⊗n µ(Sw) = J ∗
v µ(S
∗
vSw)Jw = δv,wJ ∗
v Jw.
Combination with the previous estimate gives,
kJ ∗
v {Trace(J ∗
uJuk ≤ kdiagv(J ∗
v Jv)})1/n = (max
v Jv)k1/n = (max
v {kJ ∗
v {k Rvk2})1/n ≤ (max
v Jvk})1/n ≤
v {kRvk2})1/n = D1/n
u,n .
(max
b) If the spectral functor is relaxed tensor, all the µ are unitary, hence the
first and the last inequalities are equalities. On the other hand the estimate
maxv{Trace(J ∗
uJuk = Du if the
integral multiplicity of u has subexponential growth.
v Jvk}Multu,n shows that kJ ∗
v Jv)} ≤ maxv{kJ ∗
4.9. Corollary. [7] The invariant state of an ergodic action of a compact group
on a unital C∗ -- algebra is a trace.
Proof As recalled above, for all u, Du,n has polynomial growth. Hence Ju are all
antiunitary by Theorem 4.8. On the other hand the same holds in the classical
case for standard solutions ju of G -- respresentations, hence we may apply Theorem
3.1.
We next discuss examples satisfying b) with Du > 1 and which are not trans-
lation actions.
4.10. Example. Let Gq be as above a deformed classical compact Lie group,
and let Φ : Rep(G′) → Rep(Gq) be a tensor equivalence, with G′ another compact
matrix quantum group (e.g. G = SU (2) and G′ = Ao(F )). The composition µ :
Rep(G′) → Rep(Gq) → Hilb with the embedding functor of Rep(Gq) is a relaxed
tensor functor. Consider the maximal ergodic action associated to µ. The integral
multiplicity of an irreducible representation u of G′ is the integral dimension of
Φ(u), which is the same as in the classical case, hence MultG′
Φ(u) = 1.
Note that these are just the inverses of the tensor equivalences of [2].
u = DimGq
4.11. Corollary. If a compact quantum group G of Kac type admits an ergodic
action on a unital C∗ -- algebra with non-tracial invariant state then the integral
dimension of some spectral irreducible representation of G has exponential growth.
Proof Since G has involutive coinverse, every standard solution ju is antiunitary. If
all the quantum dimensions of spectral irreducibles u had subexponential growth,
the associated Ju would be antiunitary by Theorem 4.8., hence the invariant state
would be a trace by Theorem 3.1.
4 AN ANALOGUE OF THE FINITENESS THEOREM FOR CQG
13
4.12. Corollary. S−1U (d) acts ergodically only on tracial C∗ -- algebras.
4.13. Corollary. Let G be a compact quantum group acting ergodically on C and
admitting a spectral irreducible representation u s.t. d(u) > dim(u). Assume that
M := πω(C)′′ and Mω are factors.
a) If M is of type IIIλ, 0 < λ < 1, then
λ ≥ sup{
min{λu, Λ−1
u }
Du
, u spectral irr. s.t. d(u) > dim(u)},
b) if the above supremum is 1, M is of type III1.
Proof a) Let u be a spectral irreducible representation such that d(u) > dim(u).
As argued in Theorem 3.2., the modular group of ω does not act trivially on the
spectral subspace µu⊗Hu. On the other hand, the smallest and largest eigenvalues
of the restriction of the modular operator ∆ω to µu ⊗ Hu are bounded below by
. Taking into consideration the fact that the eigenvalues
DuΛu
. b) If the
of ∆ω belong to S(M ), we see that either
supremum is 1, by a), M is not of type IIIλ for any 0 < λ < 1. Hence M must be
of type III1 by Theorem 3.2.
DuΛu ≤ λ or λ−1 ≤ Du
and above by Du
λu
λu
1
1
Remark If the quantum group is of Kac type, under the same assumptions of the
previous corollary, we similarly derive λ ≥ 1
n , with n the minimal dimension of a
spectral irreducible representation for which the restriction of the modular group
on the associated spectral subspace is non-trivial.
n
4.14. Example. The lower bound in the above remark is realized in Wang's
factors [13]. More precisely,
example of ergodic action of Au(n) on the type III 1
in this case the factor is the von Neumann completion of the Cuntz algebra On in
the GNS representation of the canonical state and the modular group is the group
describing the Z -- gradation. This automorphism group acts non-trivially on the
generating Hilbert space of isometries. This Hilbert space carries the fundamental
representation of the quantum group. This example shows that Du, together with
triviality of the modular theory of Au(n), explain completely the factorial type.
−1, e.g. under the condition of Prop. 4.3, the lower
Remark If λu = D−1
bound for λ becomes simply sup{λ2
In particular, for
SqU (2), with 0 < q < 1, under the same assumptions of the previous theorem,
the possible parameters λ satisfy λ ≥ q2r, where r is the smallest strictly positive
integer such that ur is spectral. Hence, for a given q, small values of λ are possible
only if the spectrum of the action has gaps.
u, u : d(u) > dim(u)}.
u = Λu
The following example shows that the assumption of subexponential growth of
integral multiplicities can not be removed in Theorem 4.8 b).
4.15. Example. For Ao(F ), with F of rank ≥ 3, taking into account Example
4.6 and λur = Λ−1
kF k2 )r, where q is defined
by Trace(F ∗F ) = Trace((F ∗F )−1) = q + 1
q and ur is again the first spectral
ur = kFk−2r, we see that λ ≥ (
q
REFERENCES
14
irreducible. In particular, consider the translation action of Ao(F ) over itself. Vaes
and Vergnioux have shown, among other things, that if Trace(F ∗F ) ≥ √5kFk2
then the Haar state h is factorial and the associated von Neumann algebra is a
full factor. Moreover, if the spectrum of (F ∗F )−1 ⊗ F ∗F generates the subgroup
{λn, n ∈ Z}, for some 0 < λ < 1, πh(Ao(F ))′′ is a factor of type IIIλ [12]. The
fundamental representation is spectral. In this example our lower bound λ ≥ q
is not optimal already for F ∗F = diag(λ, 1, λ−1).
kF k2
Acknowledgements. I would like to thank Alessandro Fig`a-Talamanca for a con-
versation related to this note.
References
[1] T. Banica: Le groupe quantique compact libre U(n). Comm. Math. Phys.,
190 (1997), 143 -- 172.
[2] J. Bichon, A. De Rijdt, S. Vaes: Ergodic coactions with large multiplicity
and monoidal equivalence of quantum groups. Comm. Math. Phys., 262
(2006), 703 -- 728.
[3] F.P. Boca: Ergodic actions of compact matrix pseudogroups on C∗-algebras.
In: Recent advances in operator algebras (Orl´eans, 1992). Ast´erisque 232
(1995), 93 -- 109.
[4] A. Connes: Une classification des facteurs de type III. Ann. Sci. ´Ecole Norm.
Sup., 6 (1973), 133 -- 252.
[5] A. Connes: Almost periodic states and factors of type III1. J. Funct. Anal-
ysis, 16 (1974), 415 -- 445.
[6] S. Doplicher, R. Longo, J.E. Roberts, L. Zsido: A remark on quantum group
actions and nuclearity. Rev. Math. Phys., 14 (2002), 787 -- 796.
[7] R. Høegh-Krohn, M.B. Landstad, E. Størmer: Compact ergodic groups of
automorphisms, Ann. of Math., 114 (1981), 75 -- 86.
[8] R. Longo, J.E. Roberts: A theory of dimension. K-Theory, 11, (1997), 103 --
159.
[9] C. Pinzari, J.E. Roberts: A duality theorem for ergodic actions of compact
quantum groups on C∗-algebras. Comm. Math. Phys., 277 (2008), 385 -- 421.
[10] C. Pinzari, J.E. Roberts: Ergodic actions of SµU (2) on C∗-algebras from II1
subfactors. J. Geom. Phys., 60 (2010), 403 -- 416.
[11] C. Pinzari, J.E. Roberts: A theory of induction and classification of tensor
C∗ -- categories, J. Noncomm. Geom., to appear, arXiv:0907.2459.
[12] S. Vaes, R. Vergnioux: The boundary of universal discrete quantum groups,
exactness, and factoriality. Duke Math. J., 140 (2007), 35 -- 84.
REFERENCES
15
[13] S. Wang: Ergodic actions of universal quantum groups on operator algebras.
Comm. Math. Phys., 203 (1999), 481 -- 498.
[14] A. Wassermann: Ergodic actions of compact groups on operator algebras.
III. Classification for SU (2). Invent. Math., 93 (1988), 309 -- 354.
[15] S.L. Woronowicz: Compact matrix pseudogroups. Comm. Math. Phys., 111
(1987), 613 -- 665.
[16] S.L. Woronowicz: Compact quantum groups. In: Sym´etries quantiques (Les
Houches, 1995), 845 -- 884, North-Holland, Amsterdam, 1998.
Claudia Pinzari, Dipartimento di Matematica, Sapienza Universit`a di Roma, 00185 --
Roma, Italy; e-mail: [email protected]
|
1806.08786 | 1 | 1806 | 2018-06-21T22:02:55 | Projective geometry in the Poincar\'e disk of a $C^*$-algebra | [
"math.OA",
"math.DG",
"math.FA"
] | We study the Poincar\'e disk ${\cal D}=\{a\in {\cal A}: \|a\|<1\}$ of a C$^*$-algebra ${\cal A}$ from a projective point of view: ${\cal D}$ is regarded as an open subset of the projective line $\mathbb{P}_1{\cal A}$, the space of complemented rank one submodules of ${\cal A}^2$. We introduce the concept of cross ratio of four points in $\mathbb{P}_1{\cal A}$. Our main result establishes the relation between the exponential map $Exp_{z_0}(z_1)$ of ${\cal D}$ ($z_0,z_1\in {\cal D}$) and the cross ratio of the four-tuple $$ \delta(-\infty), \delta(0)=z_0, \delta(1)=z_1 , \delta(+\infty), $$ where $\delta$ is the unique geodesic of ${\cal D}$ joining $z_0$ and $z_1$ at times $t=0$ and $t=1$, respectively. | math.OA | math |
Projective geometry in the Poincar´e disk of a C∗-algebra
E. Andruchow, G. Corach, L. Recht
June 26, 2018
Abstract
We study the Poincar´e disk D = {a ∈ A : (cid:107)a(cid:107) < 1} of a C∗-algebra A from a projective
point of view: D is regarded as an open subset of the projective line AP1, the space of
complemented rank one submodules of A2. We introduce the concept of cross ratio of
four points in AP1. Our main result establishes the relation between the exponential map
Expz0 (z1) of D (z0, z1 ∈ D) and the cross ratio of the four-tuple
δ(−∞), δ(0) = z0, δ(1) = z1, δ(+∞),
where δ is the unique geodesic of D joining z0 and z1 at times t = 0 and t = 1, respectively.
2010 MSC: 46L05, 58B20, 22E65, 46L08
Keywords: Projective line, Poincar´e disk, C∗-algebra.
Introduction
1
Let A be a unital C∗-algebra and consider the Poincar´e disk
D = {z ∈ A : (cid:107)z(cid:107) < 1}
and the Poincar´e halfspace
H = {h ∈ A :
1
2i
(h − h∗) is positive and invertible}.
This paper is the second part of the study began in [1], where we considered D and H as
homogeneous spaces of the unitary groups U(θD) and U(θH) of certain quadratic forms θD and
θH in A2, respectively. These spaces have reductive connections and invariant Finsler metrics,
which make them non positively curved length spaces (see [10]).
In the present paper we study the D from the projective point of view. More precisely, we
introduce the projective line AP1 of A, and regard D as an open subset of AP1. In this way, the
group U(θ) appears as the group of projective transformations which preserve D. This allows us
to introduce the concept of cross ratio of four points in AP1, in the sense considered by Zelikin
[16], adapted to our situation. Geometrically, the construction of the cross ratio is illustrated
by the following figure:
1
Figure 1.
Here the plane represents A2 and the lines are elements of AP1, i.e. complemented submod-
ules of A generated by a single element. The cross ratio of ∞, 0, (cid:96)1 and (cid:96)2 (in this precise order)
in the projective line AP1 we choose ∞ as the point of
has the following geometric meaning:
infinity, then AP1 \ {∞} turns into an affine line (L in the figure above). In this affine line,
we choose a point 0, and the affine line turns into a vector line. In this vector line we choose
a point (cid:96)1, and the vector line turns into a scalar line (identified with A by means of the basis
(cid:96)1). Finally, if in this scalar line we choose a fourth element (cid:96)2 (cid:54)= 0, this point has a coordinate
in the basis (cid:96)1. Classically, this scalar is the cross ratio of ∞, 0, (cid:96)1, (cid:96)2. These features can be
observed in Figure 1, in the line L. In this paper, the cross ratio of these four lines will be an
endomorphism ϕ of (cid:96)1. The map ϕ associates to each x ∈ (cid:96)1 the point ϕ(x), obtained by two
succesive projections: first to 0, parallel to ∞; next to (cid:96)1, parallel to (cid:96)2. This is clear in the
figure above.
We introduce a fibre bundle Γ whose fibers are C∗-algebras (isomorphic to A), over the base
space D. In the tangent bundle TD, we define an inner product with values in Γ (similarly as a
Riemannian structure). If z ∈ D and X, Y ∈ (TD)z, we denote this inner product by (cid:104)X, Y (cid:105)z.
It is an element in Γz (the fiber of Γ over z). Then, by definition, the cross ratio of four points
in D is an element of Γ.
On the other hand, we compute explicitly the exponential map of the natural connection
over TD, which is bijective at every tangent space, and its inverse Logz : D → (TD)z. We
consider the morphism (cid:104)Logz(w), Logz(w)(cid:105)1/2
between (cid:104)Logz(w), Logz(w)(cid:105)1/2
In Theorem 10.1 and Corollary 10.2 we state our main result, which clarifies the relationship
and the cross ratio of −∞, z, w,∞. Here ±∞ denote the limits
as a distance operator from z to w in D.
z
z
SOT − lim
t→±∞ γ(t),
where γ is the geodesic from z to w in time 1, and the algebra A is considered in its univer-
sal representation. The formulas in Theorem 10.1 and Corollary 10.2 enable one to view the
differential geometry of the Poincar´e disk D as projective geometry of spaces of operators.
of A2, which will be used to define AP1 and APθ
1.
The contents of the paper are the following. In Section 2 we introduce the regular elements
In Section 3 we define the projective AP1 line of A, and introduce the operator cross ratio
it is defined as a
of four submodules (following ideas introduced in [16] for closed subpaces):
(possibly empty) set of module homomorphisms.
2
In Section 4 we introduce our object of study, the θ-hyperbolic part APθ
1 of AP1. We state
1 with the disk D and with the space of projections Qρ (θ-orthogonal
1). We introduce the action of the θ-unitary
the identifications of APθ
projections whose ranges are the elements of APθ
group U(θ), as well as the Borel subgroup B(θ).
In Section 5 we describe the tangent bundle of APθ
1.
In Section 6 we recall the basic facts of the geometry of D done in [1]. Also we compute the
explicit form of the exponential map and its inverse Log.
In Section 7 we compute the limit points of geodesics at t = ±∞. In particular, we show
that the partial isometry of z ∈ D coincides with the SOT− limt→∞ δ(t), where δ is the geodesic
joining 0 and z.
In Section 8 we introduce the U(θ)-invariant Finsler metric in APθ
1.
In Section 9 we consider the cross ratio of the points (cid:96)−∞, (cid:96)0, (cid:96), (cid:96)+∞, where (cid:96)0 is the sub-
module corresponding to the origin in D, (cid:96) is an arbitrary point in (cid:96) (cid:54)= (cid:96)0, and (cid:96)∓∞ are the ∓∞-
limit points of the geodesic starting at (cid:96)0 at t = 0 and reaching (cid:96) at t = 1. We show that this
set is always non empty, and that there is a natural choice of a specific homomorphism cr((cid:96)0, (cid:96)),
or cr(0, z) if (cid:96) (cid:39) z ∈ D (which is the only possible choice for a strongly dense subset of (cid:96) ∈ APθ
1,
in the case when A is a von Neumann algebra). We finish Section 9 by giving a first glimpse
of our main result, which relates cr(0, z) with the metric geometry of D (or equivalently, APθ
1),
namely,
(cid:107)cr(0, z)(cid:107)B((cid:96)z) =
1
2
d(0, z).
In Section 10 we briefly introduce the coefficient bundle over APθ
where (cid:107)cr(0, z)(cid:107)B((cid:96)z) stands for the norm of the endomorphism cr(0, z) : (cid:96)z → (cid:96)z.
1 (using the model Qρ), and
an Hermitian structure on APθ
1, with values in the coefficient bundle. This is done in order to
state our main results of this paper, which are the formulas given in Theorem 10.1 and Corollary
10.2, relating the cross ratio cr((cid:96)1, (cid:96)2) with the logarithm map Log(cid:96)1(cid:96)2.
In Section 11 we consider a relevant example, namely, when there exists a C∗-subalgebra B
of the center of A and a conditional expectation tr : A → B satisfying tr(xy) = tr(yx) for all
x, y ∈ A; an important particular case is when A is commutative. In this case the computations
are much simpler, and the formula relating the cross ratio and the logarithm is
Log(cid:96)1
(cid:96)2
e
= cr((cid:96)1, (cid:96)2).
2 Preliminaries
Let A be a unital C∗-algebra, G the group of invertible elements of A, G+ the subset of G of
positive elements. Consider A2 = A × A as a right A-C∗-module, with A-valued inner product
given by
(cid:19)
(cid:18) x1
x2
(cid:18) y1
(cid:19)
(cid:104)x, y(cid:105) = x∗
1y1 + x∗
2y2,
where x =
, y =
∈ A2. A2 is endowed with the usual C∗-module norm
We shall say that an element x =
y2
(cid:107)x(cid:107) = (cid:107) < x, x > (cid:107)1/2 = (cid:107)x∗
(cid:19)
(cid:18) x1
x2
1x1 + x∗
∈ A2 is regular if
(cid:104)x, x(cid:105) ∈ G+.
2x2(cid:107)1/2.
3
reg the set of regular elements. Clearly A2
reg is an open subset of A2. The
We shall denote by A2
C∗-algebra of adjointable operators of A2 (bounded A-linear operators acting in A2 which admit
an adjoint for the A-valued inner product) identifies with the algebra M2(A) of 2 × 2 matrices
with entries in A, acting by left multiplication. Let Gl2(A) be the group of invertible elements
of M2(A). Let us state the following basic facts:
Proposition 2.1. With the current notations, we have that
reg is an open subset of A.
1. A2
2. Gl2(A) acts on A2
3. G acts on A2
reg by left multiplication.
reg by right multiplication.
4. Both actions commute.
5. For each x ∈ A2
6. For each x ∈ A2
reg the orbit Gl2(A) · x is open (and closed) in A2
reg, the map Gl2(A) → GL2(A) · x ⊂ A2
reg.
reg, g (cid:55)→ g · x defined a principal
fibre bundle with smooth local cross sections.
We shall omit the proof: the proposition holds for An instead of A2. We refer to [5] (Section
1) for complete proofs of these facts, noticing that
An
reg = {x ∈ An : there exists y ∈ An such that y1x1 + ··· + ynxn = 1}.
Indeed, clearly if x ∈ An
k provides an n-tuple
such that y1x1 + ··· + ynxn = 1. Conversely, if y1x1 + ··· + ynxn = 1, and we denote by
y∗ = (y∗
n), by the Schwarz inequality for C∗-modules (see for instance [12], Prop. 1.2.4),
we have
reg, then putting y given by yk =< x, x >−1 x∗
1, . . . , y∗
1 =< y∗, x >= < y∗, x > 2 ≤ (cid:107) < y∗, y∗ > (cid:107) < x, x >,
which implies that < x, x > is invertible.
This set An
reg is known, in the K-theoretic setting, as the set of n-unimodular rows in A .
The paper by M. Rieffel [15] contains a thorough treatment of this subject.
We are interested in the single orbit
(cid:18) 1
0
(cid:19)
.
O := Gl2(A) · e1,
where e1 =
Here and throughout, denote g := (g∗)−1.
We consider the sphere
S = S(A2) = {x ∈ A2 : (cid:104)x, x(cid:105) = 1},
and the intersection
SO = S ∩ O.
The next result will not be needed in the rest of the paper, we state it here to describe the
topology of the sphere S in the case when A = B(H).
4
Theorem 2.2. If A = B(H), for an infinite dimensional Hilbert space H, then:
1. Each x ∈ S defines an isometry x : H → H2 such that R(x)⊥ is infinite dimensional.
2. S is connected.
Proof.
(cid:107)xξ(cid:107)2 = (cid:107)x1ξ(cid:107)2 + (cid:107)x2ξ(cid:107)2 =< x∗
1x1ξ, ξ > + < x∗
Let us check that R(x)⊥ is infinite dimensional. Note that
2x2ξ, ξ >=< ξ, ξ >= (cid:107)ξ(cid:107)2.
R(x)⊥ = {(ξ, η) ∈ H × H : x∗
1ξ = −x∗
2η}.
1) or N (x∗
If either N (x∗
suppose that dim N (x∗
R(x)⊥.
Suppose then that both N (x∗
2) is infinite dimensional, it is clear that R(x)⊥ is infinite dimensional:
1) = +∞; then all pairs of the form (ξ, 0), with ξ ∈ N (x∗
1), belong to
1), N (x∗
{(x1ϕ, x2ψ) : x∗
1x1, the condition x1x∗
2) are finite dimensional. Clearly
1x1ϕ = −x∗
1ϕ = −x∗
2x2ψ} ⊂ R(x)⊥.
2x2ψ is equivalent to ϕ = x∗
2x2 = 1 − x∗
Since x∗
denote ν = ϕ − ψ, then the left-hand set above can be written
1x1ν, x2(x∗
{(x1ϕ, x2(ϕ + ν)) : ϕ = x∗
1x1ν} = {(x1x∗
1x1ν + ν)) : ν ∈ H}.
1x1) is infinite dimensional: its orthogonal complement R(x1x∗
Note that the R(x1x∗
1x1(ϕ − ψ). If we
1x1)⊥ satisfies
R(x1x∗
1x1)⊥ = N (x∗
1x1x∗
1) ⊂ N ((x1x∗
1)2) = N (x1x∗
1) = N (x∗
1),
and thus is finite dimensional. Therefore x is an isometry with infinite co-rank.
Let us check now that the sphere S is connected. Consider a fixed unitary operator U :
H × H → H. Let x, y ∈ S, again, be regarded as isometries from H to H2. Then the operators
U x and U y are isometries in H, with infinite co-rank. In [4] it was shown that the connected
components of the set of isometries in H is parametrized by the co-rank: two isometries belong
to the same component if and only if they have the same co-rank. Moreover, also in [4], it was
shown that these components are the orbits of the left action of the unitary group of H, by left
multiplication. It follows that there exists a selfadjoint operator Z ∈ B(H) such that
U y = eiZU x.
Then x(t) = U∗eitZU x is a continuous curve of isometries from H to H × H such that x(0) = x
and x(1) = y. Then
(cid:18) x1(t)
(cid:19)
∈ S
x2(t)
is a continuous curve in S which connects x and y.
x(t) =
Let us denote by U2(A) the unitary group of M2(A). Since M2(A) is the C∗-algebra of
adjointable operators of the module A2, U2(A) is the group of invertible elements in M2(A)
which preserve the A-valued inner product of A2. In particular, this implies that U2(A) acts on
S, and on SO.
5
Proposition 2.3. SO is the orbit of e1 under the action of U2(A).
Proof. Clearly the orbit U2(A) · e1 is contained in SO. Pick x = ge1 ∈ SO. We must show that
there exists a unitary element u such that ue1 = ge1. Denote r = g∗g and by p0 the selfadjoint
projection
. The fact that ge1 ∈ SO means that
(cid:18) 1 0
(cid:19)
0 0
(cid:104)re1, e1(cid:105) = 1,
i=0 αiti, then q(r)p0 = (cid:80)n
(cid:80)n
or, equivalently, that r p0 = p0. It follows that, for any n ≥ 0, rn p0 = p0, and thus, if q(t) =
i=1 αi p0 = q(1)p0. Let qn(t) be a sequence of polynomials which
t in the spectrum of r (a compact subset of (0, +∞)). Then qn(r) → r1/2.
√
converge uniformly to
On the other hand, it holds
qn(r)p0 = qn(1)p0 → p0.
Then, r1/2 p0 = p0. Consider the polar decomposition of the (invertible) element g in M2(A):
g = ur1/2. Then u is a unitary element such that up0 = ur p0 = g p0. In particular,
ue1 = first column of up = first column of g p = ge1.
Corollary 2.4. The unitary group U of A acts on SO by right multiplication.
Proof. Pick x = u · e1 ∈ SO (u ∈ U2(A)), and v ∈ U. Then
x · v = u · e1 · v = uv · e1,
(cid:19)
(cid:18) v 0
0 v
where uv = u
∈ U2(A).
Remark 2.5. There is a retraction A2
Corollary 2.6. In the case A = B(H), SO = S and O = A2
3 The projective line over A
We shall call rank one submodule a closed submodule (cid:96) ⊂ A2 which is orthocomplemented, i.e.,
reg → S: x (cid:55)→ x(cid:104)x, x(cid:105)−1/2.
reg.
(cid:96) ⊕ (cid:96)⊥ = A2, where (cid:96)⊥ = {y ∈ A2 :< x, y >= 0 for all x ∈ (cid:96)},
and such that there exists x0 ∈ (cid:96) with [x0] = {x0a : a ∈ A} = (cid:96). As the result below shows, not
every element in A2 generates a rank one submodule:
Proposition 3.1. Let x ∈ A2, x (cid:54)= 0. Then (cid:96) = {xa : a ∈ A} is a rank one submodule if and
only if there exists a0 ∈ A such that x0 = xa is also a generator for (cid:96), and < x0, x0 > is a
projection in A.
6
Proof. Suppose that there is a generator x0 of (cid:96) such that < x0, x0 >= p0 is a projection. Note
that x0 = x0p0. Indeed, x0(1 − p0) satisfies < x0(1 − p0), x0(1 − p0) >= (1 − p0)p0(1 − p0) = 0,
i.e., x0(1 − p0) = 0. Then px0 defined by px0(y) = x0 < x0, y > is the orthogonal projection
onto (cid:96). Clearly, it is selfadjoint. It is a projection:
p2
x0(y) = x0 < x0, x0 < x0, y >>= x0p0 < x0, y >= x0 < x0, y >= px0(y).
Clearly, the range of px0 is (cid:96).
Conversely, suppose that (cid:96) = [x] is orthocomplemented. Then there exists a symmetric
adjointable projection p : A2 → (cid:96) ⊂ A2. It is of the form p(y) = xϕ(y), where ϕ : A2 → A is
A-linear and bounded. Then (see [11], p. 13) ϕ is of the form ϕ(y) =< z, y > for some z ∈ A2.
The adjoint of p(y) = x < z, y > is p∗(y) = z < x, y >. Therefore, the fact that p∗ = p, implies
that z = xa for some a ∈ A. Thus, p = xa < x, >= x < xa, >= xa∗ < x, >. Using that
p2 = p, we get that
p = p2 = xa∗ < x, xa < x, >>= xa∗ < x, x > a < x, >= xb < xb, >,
where b = (a∗ < x, x > a)1/2. Denote x0 = xb. The above computation shows that x0 is also a
generator for (cid:96). Then p = px0. Using again the fact that p2 = p, we obtain
x0 < x0, >= p = p2 = x0 < x0, x0 >< x0, > .
Evaluating at x0, we get x0 < x0, x0 >= x0 < x0, x0 >2; applying < x0, > to this equality,
we obtain that c =< x0, x0 > satisfies c2 = c3. Since c ≥ 0, this means that c has two spectral
values 0, 1, and thus c is a (non zero) projection.
In this paper, we shall be interested in rank one modules (cid:96) = [x] with x ∈ O. Note that for
these special generators, the projection p0 in the above proposition is < x, x >= 1.
An equivalence relation is defined in O: x ∼ x(cid:48) iff there exists a ∈ G such that x(cid:48) = xa (or
equivalently, [x] = [x(cid:48)]). Then, it holds that also AP1 = O/ ∼.
AP1 = {(cid:96) = [x] : x ∈ O}.
Define
The map
is onto, and the fibers are the equivalence classes in O.
O → AP1 x (cid:55)→ [x]
Consider the map
O → SO , x (cid:55)→ x(cid:104)x, x(cid:105)−1/2.
Clearly, it is a well defined and continuous retraction.
Remark 3.2. It should be noticed that the projective spaces of C∗-modules over A which
appear in [2] and [3] are strictly bigger than AP1.
We end this section by defining the cross ratio CR((cid:96)1, (cid:96)2, (cid:96)3, (cid:96)4) of four submodules (cid:96)1, (cid:96)2, (cid:96)3, (cid:96)4
in AP1, following ideas by M.I. Zelikin [16].
Definition 3.3. We denote by CR((cid:96)1, (cid:96)2, (cid:96)3, (cid:96)4) the (possibly empty) set of module homomor-
phisms ϕ : (cid:96)3 → (cid:96)3 of the form ϕ = ψη, where the homomorphisms η : (cid:96)3 → (cid:96)2, ψ : (cid:96)2 → (cid:96)3
satisfy x − ψ(x) ∈ (cid:96)4 and y − η(y) ∈ (cid:96)1, for all x ∈ (cid:96)3, y ∈ (cid:96)2.
7
4 The hyperbolic part of the A-projective line: the Poincar´e
disk of A.
In this section we collect from [1] certain facts concerning the hyperbolic geometry of the unit
disk D of A. Recall that, with the appropriate metric, it is a model for the part of AP1 which
carries a nonpositively curved geometry. The main feature is a signed sesquilinear form θ, and
the group which preserves it. More precisely, define
In [1] it was denoted θD. Note that θ(x, y) = (cid:104)
θ : A2 × A2 → A , θ(x, y) = x∗
(cid:18) x1
(cid:18) 1
x2
(cid:19)
(cid:18) y1
(cid:19)
1y1 − x∗
(cid:19)
y2
, ρθ
2y2.
(cid:105), where
ρ = ρθ =
0
0 −1
.
The main tool in the study of the geometry of the Poincar´e disk of A is the subgroup U(θ) of
Gl2(A) which leaves invariant the form θ:
U(θ) = {g ∈ Gl2(A) : θ(gx, gy) = θ(x, y) for all x, y ∈ A2}.
We call U(θ) the unitary group of the form θ.
complemented Banach-Lie subgroup of Gl2(A).
1 of AP1.
Equivalently, g ∈ Gl2(A) belongs to U(θ) iff g∗ρg = ρ. In [1] it was shown that U(θ) is a
Let us define the hyperbolic part APθ
Definition 4.1. The set
APθ
1 = {[x] ∈ AP1 : there exists a generator x0 of [x] such that θ(x0, x0) ∈ G+}
is called the hyperbolic part of AP1.
Remark 4.2. Note that
θ(xa, xa) = a∗x∗
1x1a − a∗x∗
2x2a = a∗(x∗
1x2 − x∗
2x2)a = a∗θ(x, x)a.
Therefore, if θ(x0, x0) ∈ G+ for a generator x0 of [x], then the same is true for any other
generator of [x].
Also, it holds that θ is positive (semidefinite) in [x]. But the converse does not hold: θ might
be positive in [x] without θ(x0, x0) being invertible for any generator x0.
Theorem 4.3. The group U(θ) acts transitively in APθ
given by
1. If x ∈ O and g ∈ U(θ), the action is
g · [x] = [gx].
Before we give the proof of this statement, we introduce certain facts on APθ
1. Note that if
x ∈ O is a generator of [x] ∈ APθ
1, then
θ(gx, gx) = θ(x, x) ∈ G+.
Also it is clear that gx ∈ O. Thus, U(θ) acts in APθ
1. To prove the above theorem, we only need
to show that the action is transitive, i.e., that if x satisfies θ(x, x) ∈ G+ (i.e. x ∈ O), then there
exists h ∈ U(θ) such that x = he1.
8
Definition 4.4. Let us denote by Kθ the hyperboloid defined by the form θ:
Kθ = {x ∈ A2 : θ(x, x) = 1 and x1 ∈ G}.
The space Kθ will be a sort of coordinate space for APθ
1x1 = 1 + x∗
means that x∗
that Kθ ⊂ O. Let us recall the following facts from [1] (Sections 4 and 5):
Proposition 4.5. With the current notations, we have the following:
2x2, so that (cid:104)x, x(cid:105) = 1 + 2x∗
1. Note that Kθ ⊂ A2
reg: θ(x, x) = 1
2x2 ∈ G+. Moreover, we shall see below
1. The group U(θ) acts on Kθ:
if x ∈ Kθ and g ∈ U(θ), then gx ∈ Kθ. The action is
transitive. In particular, Kθ = {ge1 : g ∈ U(θ)} ⊂ O.
2. There is a (complemented Banach-Lie) subgroup Bθ of U(θ) which acts freely and transi-
tively in Kθ. In particular, Kθ has group structure.
3. Kθ is a complemented C∞-submanifold of A2. The map
πθ : Kθ → APθ
1 , πθ(x) = [x]
is a C∞ submersion. In particular, it is onto. If x, y ∈ Kθ satisfy [x] = [y], then there
exists u ∈ U such that y = xu, i.e., the fibers of πθ identify with U.
4. The map πθ is U(θ)-equivariant: if g ∈ U(θ) and x ∈ Kθ,
πθ(gx) = [gx] = g · [x] = g · πθ(x).
Proof. (of Theorem 4.3) Note that e1 ∈ Kθ. Therefore if x ∈ Kθ, there exists g ∈ U(θ) such that
ge1 = x. Since πθ is onto, it follows that any [x] ∈ APθ
1 has a generator (say) x ∈ Kθ. Then
[x] = g · [e1].
Remark 4.6. Let us describe the group Bθ explicitly:
− gx
g − g
− gx
: g ∈ G, x∗ = −x}.
Bθ = {
g + g
g − g
2
2
g + g
2
2
(cid:19)
+ gx
(cid:18) g
+ gx
: g ∈ G, s∗ = s}. The group considered here is
In fact, the group used in [1] is B = {
conjugate of B:
(cid:18) 1
where U = 1√
2
1
i −i
(cid:19)
0
gs g
Bθ = U∗BU,
. Indeed, if g ∈ G, and g =
(cid:19)
(cid:18) g + g − igs g − g − igs
(cid:18) g
0
gs g
g − g + igs g + g + igs
U∗gU =
1
2
∈ B, then
(cid:19)
,
2 s is anti-Hermitian.We call Bθ the Borel subgroup of U(θ). The corresponding Borel
where x = i
subgroup of U(θH ) for the Poincar´e halfspace H was described in [1]. The facts mentioned in
Proposition 4.5 where proved in [1] for the hyperboloid KθH = UKθ, of the form θH (x, y) =
θ(U∗x, U∗y)).
9
Remark 4.7. There are two natural subgroups in Bθ, which are homomorphic images of the
invertible group G of A and the additive group Aah of anti-Hermitian elements of A, which we
denote, respectively, by Gθ and Tθ. Namely,
: g ∈ G} and Tθ = {τx :=
: x∗ = −x}.
(cid:18) 1 − x −x
(cid:19)
x
1 + x
(cid:18) g + g g − g
g − g g + g
(cid:19)
Gθ = {γg :=
1
2
Elementary computations show that
1. Gθ ⊂ Bθ and Tθ ⊂ Bθ.
2. The maps G (cid:51) g (cid:55)→ γg ∈ Gθ and Aah (cid:51) x (cid:55)→ τx ∈ Tθ are group homomorphisms (Aah
considered with its additive structure). In particular, Gθ and Tθ are subgroups of Bθ.
g + g
(cid:19)(cid:18) 1 − x −x
g − g
2
2
x
1 + x
− gx
+ gx
(cid:19)
.
, then
g − g
2
g + g
2
− gx
+ gx
(cid:18) g + g g − g
g − g g + g
g =
1
2
3. Gθ and Tθ generate Bθ. More precisely, if g ∈ Bθ, g =
4. The above factorization is unique. Or, equivalently, Gθ ∩ Tθ contains only the identity
matrix.
1, namely, the unit disk D = {a ∈ A : (cid:107)a(cid:107) < 1}
Let us describe an alternative model for APθ
of A. There is a natural map from Kθ to D:
Lemma 4.8. The map Kθ → D, x (cid:55)→ x2x−1
Proof. Since θ(x, x) = 1, x∗
0 ≤ (x2x−1
2x2 = x∗
1 )∗x2x−1
1x1 − 1. Then
1 = (x−1
1 )∗(x∗
1
is well defined, onto and C∞.
1x1 − 1)x−1
1 = 1 − (x1x∗
1)−1.
Then
1 )∗x2x−1
(cid:18) x1
1 (cid:107) = (cid:107)1 − (x1x∗
1)−1(cid:107) = sup{1 − 1
t
1 (cid:107)2 = (cid:107)(x2x−1
(cid:107)x2x−1
1) is a compact set in (0, +∞) (recall that x1 is invertible). The map is clearly
because σ(x1x∗
C∞. Pick a ∈ D. Then 1 − a∗a ∈ G. Put x1 = (1 − a∗a)−1/2 and x2 = ax1. Then, clearly,
x =
Proposition 4.9. The map Kθ → D induces a C∞ diffeomorphism
belongs to Kθ and is mapped to a.
: t ∈ σ(x1x∗
1)} < 1,
(cid:19)
x2
APθ
1
(cid:39)−→ D ,
[x] (cid:55)→ x2x−1
1 .
Proof. If x, y ∈ O satisfy [x] = [y], then there exists g ∈ G such that y = xg, and thus
y2y−1
1 = x2g(x1g)−1 = x2x−1
1 . Its inverse is
D (cid:51) z (cid:55)→
(cid:20)(cid:18) 1
(cid:19)(cid:21)
∈ APθ
1.
z
10
The map πθ : Kθ → APθ
1 ) is an
It will be useful to describe the action of U(θ) on the model D. By straightforward compu-
1, πθ(x) = [x] (or alternatively, πθ : Kθ → D, πθ(x) = x2x−1
analogue of the classical Hopf fibration.
tations, if g ∈ U(θ) and z ∈ D, then
g · z = (g21 + g22z)(g11 + g12z)−1.
Remark 4.10. In particular, note that if k ∈ U(θ) satisfies k · 0 = 0, then k =
with u1, u2 ∈ UA.
(cid:18) u1
0
(1)
(cid:19)
,
0
u2
We finish this section by addressing the characterization of the local regular structure of
1. Instead of exhibiting an atlas of local coordinates, we refer to an intrinsic model for APθ
1.
APθ
In [1] we studied the space Qρ, which is defined as the space
Qρ = { ∈ M2(A) : 2 = 1 and ρ ∈ G+}.
The elements ∈ M2(A) satisfying 2 = 1, usually called reflections, are in (natural) one to one
2 (1 + ). The condition ρ ∈ G+
correspondence with projections q ∈ M2(A), q2 = q: ↔ q = 1
implies, in particular, that , and thus q, is θ-selfadjoint. More precisely, the projections q in
Qρ, correspond to the submodules (cid:96) ∈ APθ
1 (see [1]), by means of the one to one mapping
Qρ (cid:51) q ←→ (cid:96) = R(q) ∈ APθ
1.
In [14] it was shown that Qρ is a complemented submanifold of M2(A). A benefit we obtain
from this coordinate free identification APθ
1 (cid:39) Qρ, is a description of the tangent spaces of APθ
as subspaces of θ-selfadjoint elements in M2(A). We shall specify this in the next section.
5 The tangent spaces of APθ
1.
In this short section we characterize the tangent spaces of the projective line AP1, and its
hyperbolic part APθ
1. If x0 ∈ O, we identify AP1 with O/x0 · G, because O is open in A2 (and
G is open in A). Then we have
1
(TAP1)[x0] (cid:39) A2 / x0 · A = A2/[x0].
1 is open in AP1.
Note that APθ
Therefore, if [x0] ∈ APθ
1,
(TAPθ
1)[x0] = (TAP1)[x0] (cid:39) A2 / x0 · A = A2/[x0].
If [x] ∈ AP1, let us denote by [x]⊥θ the θ-orthogonal complement of [x].
Lemma 5.1. If x0 ∈ O, then there exists the submodule [x0]⊥θ . It is generated by an element
y0 ∈ A2
Proof. Consider
reg (not necessarily in O).
(cid:19)
.
y0 =
(cid:18) (x∗
1)−1x∗
1
2
Straightforward computations show that θ(x0, y0) = 0. It is easy to see that x0 and y0 generate
A2.
11
Lemma 5.2. If [x0] ∈ APθ
y ∈ [x0]⊥θ , then −θ(y, y) ∈ G+.
Proof. Recall that [x0] ∈ APθ
x1 ∈ G and θ(x0, x0) ∈ G+. Put y0 =
(cid:18) x1
(cid:19)
1, then the form θ is negative and non degenerate in [x0]⊥θ :
if
1 means that for any generator (e.g. x0 =
), it holds that
as above. Then, for any y = y0a ∈ [x0]⊥,
x2
(cid:18) (x∗
1)−1x∗
1
2
(cid:19)
θ(y, y) = a∗θ(y0, y0)a,
1 (cid:107) < 1; thus, 1 −
and
2 − 1 = x2x−1
From Lemma 4.8 and Proposition 4.9, we get that x2x−1
x2x−1
θ(y0, y0) = x2x−1
1 )∗ ∈ G+.
1)−1x∗
1 (x∗
1 )∗ − 1.
1 (x2x−1
1 ∈ D, i.e., (cid:107)x2x−1
1 (x2x−1
For [x0] ∈ APθ
[x] in A2. We choose to identify
1, the tangent space (TAPθ
(TAPθ
1)[x0] = A2/[x0] is isomorphic to any supplement of
1)[x0] (cid:39) [x0]⊥θ .
If (cid:96) = [x0] ∈ APθ
1, the identification of (TAPθ
(2)
Remark 5.3. A remark is in order.
1)(cid:96) with
(cid:96)⊥θ depends on the choice of the generator x0. In order to see how the change of generator
affects this identification, we must refer to an intrinsic model for APθ
1. We choose the model Qρ
described in the first section. Also, to keep matters more simple, consider generators in Kθ, i.e.
x0 satisfies θ(x0, x0) = 1. If (cid:96) = q(cid:96) (the unique θ-orthogonal projection onto (cid:96)) then
(TAPθ
1)(cid:96) = (TQρ)q(cid:96) = {X ∈ M2(A) : X is θ − symmetric and ρ − co-diagonal}.
0 ∈ Kθ is another generator of (cid:96), then there exists a unitary u ∈ UA such that x(cid:48)
0 = x0u.
1)(cid:96) and A2/(cid:96) done by means of x0, the tangent vector X is
If x(cid:48)
In the identification between (TAPθ
identified with Xx0 ∈ (cid:96)⊥θ .
Thus, if we change to x(cid:48)
0, both identifications differ in right multiplication by u.
6 The geometry of the disk
In the previous section we introduced an U(θ)-invariant Finsler metric in the hyperbolic part
of the A-projective line APθ
1 induced by the quadratic form θ. Also, we noted that there is a
natural diffeomorphism [x] (cid:55)→ x2x−1
In [1], we introduced a (nonpositively curved) metric in D, by establishing in turn an iden-
tification between D and a space of positive operators related to the symmetry ρ (related to θ).
Let us briefly describe it:
1 and the unit disk D of A
2 between APθ
Remark 6.1. (several results taken from [1])
1. D is embedded in the space Gl2(A)+ of positive elements in Gl2(A) by means of the map
(cid:18) 2(1 − a∗a)−1 − 1
−2a(1 − a∗a)−1
(cid:19)
−2(1 − a∗a)−1a∗
2a(1 − a∗a)−1a∗ + 1
ΦD : D → GL2(A)+ , ΦD(a) =
12
= −ρ + 2
(cid:18) (1 − a∗a)−1
(cid:18) (1 − a∗a)−1
0
ΦD(a) =
0
(1 − aa∗)−1
0
0
(1 − aa∗)−1
(cid:19)(cid:18) 1 −a∗
(cid:19)
(cid:19)(cid:18) 1 + a∗a −2a∗
−a aa∗
.
−2a
1 + aa∗
(cid:19)
For the last equality, we use that a(1 − a∗a)−1 = (1 − aa∗)−1a. Also note that
where both matrices commute.
2. Therefore, given two points z0, z1 ∈ D there exists a unique geodesic joining them.
In
[1] we computed the velocity of this unique geodesic in the case z0 = 0 and z1 = z is an
arbitrary element of D. The geodesic is given by
Proof. Straightforward computations show that the even and odd powers of
respectively
(cid:18) 0 α∗
(cid:19)2k
α 0
(cid:18) (α∗α)k
0
=
(cid:19)
0
(αα∗)k
and
(cid:19)2k+1
(cid:18)
=
(cid:18) 0 α∗
α 0
13
(cid:18) 0 α∗
α 0
(cid:19)
(cid:19)
are,
.
0
α(α∗α)k
α∗(αα∗)k
0
δ(t) = e
t
0 α∗
∞(cid:88)
α 0
1
· 0,
(z∗z)k,
where
satisfies that δ(0) = 0 and δ(1) = z. Note here that the series (cid:80)∞
2k + 1
α = z
k=0
in the interval (−1, 1), to the function f (t) = 1
explicit form of δ in D.
t2k+1
2k+1 corresponds,
1−t ). We shall compute below the
k=0
2 log( 1+t
(3)
3. The norm of (TD)0 is the usual norm of A.
4. Recall from Proposition 4.5 the hyperboloid Kθ. We have the following conmutative dia-
gram:
Kθ
(cid:39)
πθ
/ D,
πθ
APθ
1
(4)
where the map Kθ → APθ
1 is the restriction of the quotient map (O → AP1). The group
U(θ) acts on the three spaces, and the maps are equivariant with respect to the action.
Let us compute explicitly the form of the geodesic δ joining 0 and z ∈ D at time t = 1:
Lemma 6.2. Given z ∈ D, the unique geodesic δ of D with δ(0) = 0 and δ(1) = z is given by
δ(t) = ω tanh(tα),
(5)
where α is given in (3) above, and ω is the partial isometry in the polar decomposition of α:
α = ωα, performed in A∗∗.
}
}
/
We are interested in the first column of e
(2k)! (α∗α)k
(2k+1)! (α∗α)k
(cid:32) (cid:80)∞
α(cid:80)∞
t2k+1
k=0
k=0
t2k
t
0 α∗
(cid:33)
α 0
=
, which is
(cid:18) cosh(tα)
ω sinh(tα)
(cid:19)
.
Then
δ(t) = ω sinh(tα)(cosh(tα))−1 = ω tanh(tα).
Notice that ω ∈ A∗∗ need not belong to A. However δ(t) ∈ A for all t.
Let us relate the polar decompositions of z and α.
Proposition 6.3. If z ∈ D and α as in (3), then
α =
1
2
ω(log(1 + z) − log(1 − z)),
and
z = ωz,
i.e., the partial isometry ω ∈ A∗∗ is the same for α and z.
∞(cid:88)
k=0
1
2k + 1
(z∗z)kz∗z
∞(cid:88)
k=0
1
2k + 1
(z∗z)k =
(cid:32) ∞(cid:88)
k=0
(cid:33)2
,
1
2k + 1
z2k+1
2 (log(1 + z) − log(1 − z)). Next, put z = µz the polar decomposition of z. Note
Proof. First note that
α2 = α∗α =
i.e., α = 1
that, since α = z(cid:80)∞
k=0
1
2k+1z2k, we have that
∞(cid:88)
α = µz
1
z2k = µα.
2k + 1
k=0
Thus, in order to prove our claim, it suffices to show that both partial isometries µ, ω have the
same initial and final spaces (the result follows, then, by the uniqueness property of the polar
decomposition). The partial isometry µ maps N (z)⊥ = N (z)⊥ onto R(z), whereas ω maps
N (α)⊥ onto R(α). If zξ = 0, then
∞(cid:88)
∞(cid:88)
αξ = z
i.e., N (z) ⊂ N (α). Conversely, in the last expression of α, α =(cid:80)∞
2k + 1
2k + 1
k=0
k=0
1
(z∗z)kξ =
1
(zz∗)kzξ = 0,
2k+1z∗2kz; observe that
1
k=0
∞(cid:88)
1
z∗2k = g(z∗),
2k + 1
k=0
2t (log(1+t)−log(1−t)) (which can be extended continuously as g(0) = 1), is defined
where g(t) = 1
in σ(z∗) ⊂ [0, 1), and is nonvanishing there. Therefore, g(z∗) is invertible. Thus, αξ = 0
implies zξ = 0. Again, using the function g, we get α = zg(z), and thus R(z) = R(α).
14
Corollary 6.4. The exponential map Exp0 of D at 0, and its inverse Log0 can be written
explicitly as follows: if z = ωz ∈ D
ω log(cid:0)(1 + z)(1 − z)−1(cid:1) .
Log0 : D → (TD)0 , Log0(z)
If α = ωα ∈ A (cid:39) (TD)0, then
1
2
Exp0 : (TD)0 → D , Exp0(α) = ω tanh(α).
In particular, if α = Log0(z) (or, equivalently, z = Exp0(α)), then z and α have the same partial
isometry in the polar decomposition. Also,
log(cid:0)(1 + z)(1 − z)−1(cid:1) .
Exp0(α) = tanh(α) and Log0(z) =
1
2
7 Limit points of geodesics
One of our concerns in computing the geodesic δ, and the above results on the polar decompo-
sition, is to establish the following result:
Theorem 7.1. For z ∈ D, let δ be the unique geodesic of D such that δ(0) = 0 and δ(1) = z.
Put z = ωz the polar decomposition (i.e., ω ∈ A∗∗); then
Proof. By formula (5), we only need to compute the limit of tanh(tz) when t → ±∞. The
spectrum σ(z) is contained in [0, 1). Clearly, for any s ∈ [0, 1),
t→−∞ δ(t) = −ω.
SOT − lim
t→∞ δ(t) = ω and SOT − lim
(cid:26) 1 if s ∈ (0, 1)
lim
t→∞ tanh(ts) =
0 if s = 0.
By Lebesgue's bounded convergence theorem, and the Borel functional calculus for bounded
selfadjoint operators, we have that
t→∞ tanh(tα) = χ(0,1)(α) = PN (α)⊥.
lim
Then,
(cid:26) −1
t→∞ δ(t) = ωPN (α)⊥ = ω.
lim
if s ∈ (0, 1)
Similarly, using that limt→−∞ tanh(st) =
0 if s = 0
, we get that
t→−∞ δ(t) = −ωPN (α)⊥ = −ω.
lim
This geometric role of the partial isometry ω in the polar decomposition of z ∈ D (or, more
generally, of every z ∈ A \ {0}) has not been noticed before, to the authors' knowledge.
action of U(θ) to the strong operator border
In order to compute the limit points of arbitrary geodesics, it will be useful to extend the
of D, i.e., to define g · a for a ∈ A∗∗ with (cid:107)a(cid:107) = 1.
∂D := {a ∈ A∗∗ : (cid:107)a(cid:107) = 1}
15
Using the polar decomposition g = ug, u =
(cid:105).
a
(cid:19)
(cid:18) 1
(cid:19)
(cid:18) 1
,
0
u2
a
(cid:19)
(cid:19)
(cid:105),
g11 + g12a = (cid:104)e1, g
(cid:18) u1
0
(cid:104)e1, g
(cid:18) 1
(cid:19)
(cid:18) (1 + b∗b)1/2
a
b
(cid:105) = u∗
1(cid:104)e1,g
Lemma 7.2. If g ∈ U(θ) and a ∈ ∂D, then g11 + g12a is invertible in A∗∗.
Proof. Note that
i.e., we may suppose g ≥ 0, g =
b∗
(1 + bb∗)1/2
, for b ∈ A. Then
g11 + g12a = (1 + b∗b)1/2 + b∗a = (1 + b∗b)1/2(1 + (1 + b∗b)−1/2b∗a).
It suffices to show that (cid:107)(1 + b∗b)−1/2b∗a(cid:107) < 1. Note that, since (cid:107)a(cid:107) = 1,
(cid:107)(1+b∗b)−1/2b∗a(cid:107)2 ≤ (cid:107)(1+b∗b)−1/2b∗(cid:107)2 = (cid:107)(1+b∗b)−1/2b∗b(1+b∗b)−1/2(cid:107) = max{f (t) : t ∈ σ(b∗b)},
(1+t) . Clearly, this number is strictly less than 1.
for f (t) = t
Proposition 7.3. If a ∈ ∂D, and g ∈ U(θ), then
g · a := (g21 + g22a)(g11 + g12a)−1 ∈ ∂D,
defines a left action of U(θ) on ∂D.
Proof. A density argument (or a proof similar as in the previous lemma), shows that if x ∈ A∗∗
with (cid:107)x(cid:107) < 1, then g · x, defined as above, also satifies (cid:107)g · x(cid:107) < 1, and defines an action on the
unit ball of A∗∗. Let a ∈ ∂D. Then, by Kaplansky's density theorem, there exists a sequence
an ∈ D such that an → a in the strong operator topology. We claim that
Lemma 7.4. g · an → g · a strongly.
Proof. Consider the polar decomposition g = ug. We check first that g · an → g · a strongly.
As before, g =
(cid:18) (1 + b∗b)1/2
. Then
(cid:19)
b∗
(1 + bb∗)1/2
b
g · an = (b + (1 + bb∗)1/2an)((1 + b∗b)1/2 + b∗an)−1.
Clearly, b + (1+ bb∗)1/2an → b + (1+ bb∗)1/2a and (1 + b∗b)1/2 + b∗an → (1 + b∗b)1/2 + b∗a strongly.
Moreover,
((1 + b∗b)1/2 + b∗an)−1 = (1 + (1 + b∗b)−1/2b∗an)−1(1 + b∗b)−1/2.
Let us show that the norms of these inverses are uniformly bounded.
It suffices to see that
(cid:107)(1 + (1 + b∗b)−1/2b∗an)−1(cid:107) are uniformly bounded. Denote dn = (1 + b∗b)−1/2b∗an. Then, as
seen above,
(cid:107)dn(cid:107)2 ≤ max{f (t) : t ∈ σ(b∗b)} = r2 < 1.
16
Thus,
(cid:107)(1 + (1 + b∗b)−1/2b∗an)−1(cid:107) ≤ 1
1 − r
.
Therefore the inverses (1 + (1 + b∗b)−1/2b∗an)−1 converge strongly to (1 + (1 + b∗b)−1/2b∗a)−1.
Then, clearly, cn := g · an converge strongly to c := g · a. Since u =
, it is clear
that
0
u2
0
g · an = u · (g · an) = u2cnu∗
1 → u2cu∗
1 = g · a.
(cid:18) u1
(cid:19)
Let us proceed with the proof of Proposition 7.3. Since (cid:107)an(cid:107) < 1, we know that (cid:107)g · an(cid:107) < 1.
From Lemma 7.4, it follows that (cid:107)g · a(cid:107) ≤ 1. Suppose that (cid:107)g · a(cid:107) < 1. Using again the fact that
U(θ) acts on D, this would imply that
a contradiction. Thus, (cid:107)g · a(cid:107) = 1.
g−1 · (g · a) = a ∈ D,
The fact that this rule defines, indeed, a left action, follows from similar density arguments.
Using this result, we can compute the limit points of arbitrary geodesics. Since the action
of U(θ) is transitive, given z1, z2 ∈ D, there exists g ∈ U(θ) such that g · z1 = 0.
Corollary 7.5. Let z0, z1 ∈ D and g ∈ U(θ) such that g · z0 = 0. Let δ be the unique geodesic
of D such that δ(0) = z0 and δ(1) = z1. Denote by δ0 the initial velocity of δ. Then
SOT − lim
t→+∞ δ(t) = g · ω0 and SOT − lim
t→−∞ δ(t) = g · (−ω0),
(cid:19)(cid:18) v1
b∗
where ω0 ∈ A∗∗ is the partial isometry in the polar decomposition of δ0:
Remark 7.6. In order to identify these limit points in D, following the notation of the above
Corollary, note that if g = g∗v =
(the reversed
polar decomposition), then v · ω0 = v2ω0v∗
geodesics are elements in ∂D of the form
1 is a partial isometry. Therefore, the limit points of
(cid:18) (1 + b∗b)1/2
δ0 = ω0 δ0.
(1 + bb∗)1/2
(cid:19)
0
v2
0
b
(b + (1 + bb∗)1/2ω)((1 + b∗b)1/2 + b∗ω)−1 and (b − (1 + bb∗)1/2ω)((1 + b∗b)1/2 − b∗ω)−1,
where b ∈ A is arbitrary and ω ∈ A∗∗ is a partial isometry.
Note that not any partial isometry in A∗∗ occurs in the polar decomposition of an element
in D. For instance, if A = C([0, 1]) (continuous functions in the unit interval), the polar decom-
position of f ∈ A is f = wf, where w ∈ L∞(0, 1) is given by w(t) =
.
An arbitrary partial isometry in L∞(0, 1) is a measurable function whose values are zero or
complex numbers of modulus 1. The set of zeros of such a function is an arbitrary measurable
set, whereas the set of zeros of partial isometries which occur in the polar decomposition of a
continuous function, are closed subsets of [0, 1].
Another way to study the limit points of geodesics, is by using the Borel subgroup Bθ ⊂ U(θ)
instead. Indeed, since the action of this group is transitive in D, any limit point of a geodesic is
either of the form g · v or g · (−v), for g ∈ Bθ. Consider the following example:
(cid:26) f (t)/f (t) if f (t) (cid:54)= 0
0 if f (t) = 0
17
Example 7.7. Suppose that A is a von Neumann algebra, and let p (cid:54)= 0 be a projection in A.
, let us compute g·p. After straightforward computations,
g + g
g − g
2
2
− gx
+ gx
g − g
2
g + g
2
− gx
+ gx
For g =
g · p = (g(1 + p) + g (−1 + p + 2x(1 + p))) (g(1 + p) + g (1 − p − 2x(1 + p)))
−1 .
Note that 1 + p is invertible and that (1 − p)(1 + p)−1 = 1 − p. Then
g · p =(cid:0)1 + g (p − 1 + 2x(1 + p)) (1 + p)−1g−1(cid:1)(cid:0)1 + g (1 − p − 2x(1 + p)) (1 + p)−1g−1(cid:1)−1
=(cid:0)1 + g(p − 1 + 2x)g−1(cid:1)(cid:0)1 + g(1 − p − 2x)g−1(cid:1)−1
.
=
Denote α = g(p− 1)g−1 and β = 2gxg−1. Observe that α is a non-invertible selfadjoint element,
α ≤ 0 and its range is proper and closed; β is an arcitrary anti-selfadjoint element. Then
g · p = (1 + α + β) (1 − (α + β))
−1 .
Note that 1− (α + β) is invertible because Re(1− (α + β)) = 1− α ≥ 1. If one picks p = 1, then
α = 0 and
g · 1 = (1 + β)(1 − β)−1,
which is a unitary operator such that −1 does not belong to its spectrum. In particular, this
shows that the action of Bθ ceases to be transitive in ∂D.
Our next result shows a necessary condition for an element a ∈ A∗∗ with (cid:107)a(cid:107) = 1 to be a
limit point of a geodesic of D
Proposition 7.8. If a ∈ A∗∗ is the limit point at +∞ of a geodesic in D, then 1 − a∗a = hqh,
where h ∈ G+ and q ∈ A∗∗ is a projection. In particular, not every element of norm 1 in A∗∗
is the limit point of a geodesic: such elements satisfy that the defect element 1 − a∗a has closed
range.
Proof. If a is the limit point of a geodesic if and only if there exists g ∈ U(θ) and a partial
isometry ω ∈ A∗∗ such that a = g · a. The definition of the action implies that this equality can
be read as an usual matrix equality
g
(cid:18) 1
(cid:18) b1
ω
(cid:19)
(cid:19)
b2
=
,
(cid:19)
(cid:18) b1
(cid:19)
(cid:18) 1
b2
) = θ(
ω
(cid:18) b1
b2
(cid:19)
,
(cid:18) 1
ω
,
(cid:19)
),
1b1 − b∗
b∗
2b2 = 1 − ω∗ω = q(cid:48),
with b1 ∈ G and b2b−1
1 = a. Using the form θ, and the fact that g ∈ U(θ),
θ(
i.e.
which is a projection in A∗∗. Let b1 = ub1 be the polar decomposition, with u unitary. Then
au = b2b−1
1 u = b2b1−1. Thus,
q(cid:48) = b∗
1b1 − b∗
2b2 = b1(1 − b1−1b∗
2b2b1−1)b1 =,b1(1 − u∗a∗au)b1,
18
i.e.,
for q = uq(cid:48)u∗ a projection in A∗∗ and h = ub1−1u∗ ∈ G+.
1 − a∗a = ub1−1q(cid:48)b1−1u∗ = hqh,
Remark 7.9. We believe that the characterization of the partial isometries which appear in
the polar decompositions of all limit points a ∈ A∗∗ is an interesting open problem.
8 The invariant metric in APθ
1
Given (cid:96) ∈ APθ
1 and V ∈ (TAPθ
The characterization of the tangent spaces done in (2), in Section 5, enables us to define a Finsler
metric, that is, a continuous distribution of norms on the tangent bundle of APθ
1.
1)(cid:96), fix a generator x0 ∈ Kθ for (cid:96), i.e., [x0] = (cid:96) and θ(x0, x0) =
1. Recall, from Remark 5.3, that x0 is determined up to a unitary element of A: if x(cid:48)
0 is another
such generator, then there exists u ∈ UA such that x(cid:48)
0 = x0u. Having fixed a generator for (cid:96), as
we saw in Section 5, (TAPθ
1)(cid:96) identifies with (cid:96)⊥θ , and to the tangent vector V corresponds an
element v ∈ (cid:96)⊥θ . We define:
V (cid:96) := (cid:107)θ(v, v)(cid:107)1/2.
(6)
0 = x0u instead,
Note that
the tangent vector V is represented by v(cid:48) = vu ∈ (cid:96)⊥θ , and therefore
(cid:96) does not depend on the choice of the generator. If we choose x(cid:48)
(cid:107)θ(v(cid:48), v(cid:48))(cid:107) = (cid:107)θ(vu, vu)(cid:107) = (cid:107)u∗θ(v, v)u(cid:107) = (cid:107)θ(v, v)(cid:107).
Next, recall from Lemma 5.2, that θ is negative definite (non degenerate), and therefore the
expression (6) above defines a proper norm in (TAPθ
Remark 8.1. If V ∈ (TAPθ
1)−1x2
some v ∈ [x0]⊥θ ; since y0 =
1
exists a ∈ A such that v = y0a. Then
1)[x0], by the identification in (2), Section 5, V is represented by
, is a generator of [x0]⊥θ , there
(cid:18) (x∗
(cid:18) x1
, for x0 =
(cid:19)
(cid:19)
1)(cid:96).
x2
V [x0] = (cid:107)θ(y0a, y0a)(cid:107)1/2 = (cid:107)a∗(1 − x2x−1
1 )∗a(cid:107)1/2 = (cid:107)(1 − (x2x−1
1 )∗2)1/2a(cid:107).
Note also that the norm of v = y0a in A2 is
(cid:107)y0a(cid:107) = (cid:107)(cid:104)y0a, y0a(cid:105)(cid:107)1/2 = (cid:107)(1 + (x2x−1
1, the norm (cid:96) of (TAPθ
1)[x0],
Proposition 8.2. For any (cid:96) ∈ APθ
Proof. With the current notations, if V = y0a ∈ (TAPθ
V [x0] = (cid:107)(1−(x2x−1
1 )∗2)1/2(1+(x2x−1
1 )∗2)1/2a(cid:107).
1)(cid:96) is complete.
1 )∗2)1/2a(cid:107) = (cid:107)(1−(x2x−1
≤ (cid:107){(1 − (x2x−1
1 )∗2)(1 + (x2x−1
1 )∗2)−1}1/2(cid:107)(cid:107)y0a(cid:107).
1 )∗2)−1/2(1+(x2x−1
1 )∗2)1/2a(cid:107)
Similarly
(cid:107)y0a(cid:107) ≤ (cid:107){(1 + (x2x−1
1 )∗2)(1 − (x2x−1
1 )∗2)−1}1/2(cid:107)V [x0].
It follows that on (TAPθ
is closed in A2, it is complete, and the proof follows.
1)(cid:96)] (cid:39) [y0], the metric (cid:96) and the norm of A2 are equivalent. Since [y0]
19
The distribution APθ
1 (cid:51) (cid:96) (cid:55)→
(cid:96) is clearly continuous. Thus, APθ
1 is endowed with a Finsler
metric.
The following result is tautological, but of the utmost importance for our discussion:
Theorem 8.3. The Finsler metric defined in (6) is invariant under the action of U(θ).
Proof. Pick (cid:96) = [x0] ∈ APθ
1)(cid:96) (as before, V is identified to ∼ y0a) and g ∈ U(θ).
1, V ∈ (TAPθ
The action of U(θ) on APθ
1 induces an action on the tangent spaces. As quotients: g maps
A2/[x0] onto A2/[gx0], because g, being A-linear, maps [x0] onto [gx0]. But as θ-orthogonal
submodules as well: since g preserves θ,
g([y0]⊥θ ) = [gy0]⊥θ .
Then
gV q[x0] = (cid:107)θ(g(y0a), g(y0a))(cid:107)1/2 = (cid:107)θ(y0a, y0a)(cid:107)1/2 = V [x0].
8.1 Invariant metric in D
We need to compute the differential of the map APθ
diagram (4). Recall that (TAPθ
1)[x0] = A2/{x0a : a ∈ A}; in particular
1 → D at [e1]. To do so, we use the above
(TAPθ
1)[e1] = A2/A × {0}.
Thus, any tangent element V ∈ (TAPθ
1)[e1] has a unique representative V =
(cid:20)(cid:18) 0
x
(cid:19)(cid:21)
, for x ∈ A.
Lemma 8.4. The differential of the map APθ
1 , at [e1] is the map
(cid:20)(cid:18) 0
x
(cid:19)(cid:21)
1 → D, [x] (cid:55)→ x2x−1
(cid:55)−→ x , x ∈ A.
Proof. Fix x ∈ A. We use the commutative diagram (4). Let x(t) ∈ Kθ be a smooth curve
such that x(0) = e1, and the derivative of [x(t)] (in APθ
. Then x1(0) =
1, x2(0) = 0, x2(0) = x. If we map x(t) onto D, and differentiate at t = 0 we get:
1) at t = 0 is
x
(cid:20)(cid:18) 0
(cid:19)(cid:21)
x2(t)x−1
1 (t)t=0 = x2(0)x−1
1 (0) − x2(0)x−1
1 (0) x1(0)x−1
1 (0) = x,
d
dt
which proves our claim.
Theorem 8.5. The identification [x] ←→ x2x−1
1 and D. The action is isometric in both
Proof. The group U(θ) acts isometrically on both APθ
spaces. Therefore, it suffices to show that the differential at [e1] is an isometry. By Lemma 8.4,
this map is
1 and D, is isometric.
between APθ
1
(cid:20)(cid:18) 0
(cid:19)(cid:21)
x
20
(cid:55)−→ x.
(cid:19)(cid:21)
(cid:20)(cid:18) 0
(cid:19)(cid:21)
The norm of
(cid:20)(cid:18) 0
x
sends
is computed via the identification (TAPθ
1)[e1] (cid:39) [e1]⊥θ = [e2], which
x
to e2x. The norm (in A2) of this element is
(cid:107)x∗θ(e2, e2)x(cid:107)1/2 = (cid:107)x∗x(cid:107)1/2 = (cid:107)x(cid:107).
On the other hand, the norm of x as a tangent element of D at 0 is the usual norm (cid:107)x(cid:107) (in
A).
8.2 The action of the Borel subgroup Bθ
As we mentioned in Proposition 4.5 and Remark 4.5, the group Bθ, which we call the Borel
group of the form θ, is a subgroup of U(θ) which acts transitively in D. Since it is a subgroup
of U(θ), the action is also isometric (recall also that the action is free in the hyperboloid Kθ).
Therefore it acts transitively and isometrically in APθ
1.
Given z ∈ D, let us find an element g ∈ Bθ such that g · 0 = z. First, we describe the action
of U(θ) on D, which factors through the hyperboloid Kθ:
Remark 8.6. Given g ∈ U(θ) and z ∈ D, we lift z to Kθ by means of the global section
(cid:18) (1 − z∗z)−1/2
z(1 − z∗z)−1/2
(cid:19)
.
D (cid:51) z (cid:55)→
(cid:18) (1 − z∗z)−1/2
z(1 − z∗z)−1/2
(cid:19)
(cid:18) z1
z2
(cid:19)
Kθ (cid:51)
(cid:55)→ z2z−1
1 ∈ D.
Next, we multiply g
, and finally we compose with the fibration
Definition 8.7. If z ∈ D, we define
(cid:18) (1 − z∗z)−1/2
(cid:19)(cid:18) 1 + z∗
z(1 − z∗z)−1/2
0
(1 + z∗)−1
(1 + z∗)z
gz =
(cid:18) (1 + z∗)−1
0
=
(1 + z∗)−1z∗(z + 1)(1 − z∗z)−1/2
(1 + z∗)−1(z + 1)(1 − z∗z)−1/2
(cid:19)(cid:18) (1 − z∗z)−1/2
z∗(z + 1)
z + 1
0
(cid:19)
0
(1 − z∗z)−1/2
(cid:19)
.
Observe that the diagonal matrices on the right and left hand sides do not belong to U(θ),
so that this is not a proper factorization of gz.
Lemma 8.8. If z ∈ D, then gz ∈ Bθ and satisfies gz · 0 = z.
Proof. The origin 0 ∈ D lifts to e1 ∈ Kθ. Recall from Remark 4.6 the form of the elements in
Bθ:
g + g
g − g
2
2
Bθ = {
− gx
+ gx
g − g
2
g + g
2
− gx
+ gx
: g ∈ G, x∗ = −x}.
21
Thus, we are looking for g ∈ G and x ∈ A with x∗ = −x such that
1
2
(g + g) − gx = (1 − z∗z)−1/2 and
(g − g) + gx = z(1 − z∗z)−1/2.
1
2
That is, g = (1 + z)(1 − z∗z)−1/2, thus g = (1 + z∗)−1(1 − z∗z)1/2 and
gx =
1
2
(1 + z∗)−1{z − z∗}(1 − z∗z)−1/2.
Then,
1
2
x =
g∗(1 + z∗)−1{z − z∗}(1 − z∗z)−1/2 =
which is clearly anti-Hermitian, and thus gz ∈ Bθ.
Remark 8.9. Denote by the d the distance defined by the the Finsler metric introduced in D.
In [1] (Section 8) it was shown that
(1 − z∗z)−1/2(z − z∗)(1 − z∗z)−1/2,
1
2
d(0, z) =
1
2
log(
1 + (cid:107)z(cid:107)
1 − (cid:107)z(cid:107) ).
Using the fact that the action of Bθ on D is isometric, and the above construction of gz, for
arbitrary z1, z2 ∈ D, we can compute d(z1, z2) as follows:
1 + (cid:107)g−1
1 − (cid:107)g−1
z1 · z2(cid:107)
z1 · z2(cid:107) ).
1
2
log(
d(z1, z2) = d(0, g−1
(7)
z , recall that elements g in U(θ) are characterized by the relation ρg∗ρ =
(cid:19)
z1 · z2) =
(cid:19)(cid:18)
(cid:19)(cid:18) (1 + z)−1
−z∗(1 + z)
1 + z
0
0
(1 − z∗z)−1/2
−(1 + z∗)z
1 + z∗
0
(1 + z)−1
.
In order to compute g−1
ρ−1. Then
g−1
z =
(cid:18) (1 − z∗z)−1/2
0
Then, after straightforward computations,
1z1)−1/2(1 + z∗
z1 · z2 = (1 − z∗
g−1
1)(1 + z1)−1(z2 − z1)(1 − z∗
1z2)−1(1 − z∗
1z1)1/2.
9 Operator cross ratio in the hyperbolic part of the projective
line
Here we state our main result, relating the metric of APθ
1 introduced in Section 8, with the so
called operator cross ratio, as defined in the Grassmann manifold of a Hilbert space by M.I.
Zelikin [16]. We shall apply these ideas to the rank one submodules in APθ
1. To this effect, the
isometry between APθ
(cid:19)(cid:21)
1 and the disk D will be important.
1, for z ∈ D. Let z = ωz be the polar decomposition. Let δ be
(cid:20)(cid:18) 1
Consider (cid:96) =
∈ APθ
z
(cid:20)(cid:18) 1
(cid:19)(cid:21)
(cid:19)(cid:21)
(cid:20)(cid:18) 1
z
the geodesic of APθ
D: δ(0) = 0 and δ(1) = z. As seen in Section 6,
1 such that δ(0) =
0
and δ(1) =
. Equivalently, regarded in
SOT − lim
t→+∞ δ(t) = ω and SOT − lim
t→−∞ δ(t) = −ω.
22
Four points are determined: −ω, 0, z, ω, or better, four submodules
(cid:19)(cid:21)
(cid:20)(cid:18) 1
−ω
(cid:19)(cid:21)
(cid:20)(cid:18) 1
0
(cid:19)(cid:21)
(cid:20)(cid:18) 1
z
(cid:96)−∞ :=
, (cid:96)0 :=
, (cid:96) =
, (cid:96)+∞ :=
(cid:20)(cid:18) 1
ω
(cid:19)(cid:21)
,
where the limit lines lie in ∂APθ
1.
In Section 3 we defined the operator cross ratio of four elements in AP1, as a (possibly
empty) set of module endomorphisms, following ideas in [16]. Here we compute the operator
cross ratio CR((cid:96)−∞, (cid:96)0, (cid:96), (cid:96)+∞), proving that it is non empty, and that there exists a natural
(cid:96)-endomorphism to choose from this set.
Recall that elements of CR((cid:96)−∞, (cid:96)0, (cid:96), (cid:96)+∞) are (module) endomorphisms of (cid:96), defined as the
composition of the projection from (cid:96) to (cid:96)0 parallel to (cid:96)−∞, followed by the projection from (cid:96)0 to
(cid:96) parallel to (cid:96)+∞.
In coordinates, by choosing generators in the respective submodules
(cid:19)
(cid:18) 1
z
(cid:19)
(cid:18) 1
0
λ,
(cid:55)→
(cid:18) 1 − λ
(cid:19)
z
(cid:19)
(cid:18) 1
−ω
=
µ.
Then 1 − λ = µ and z = −ωµ. Then ωz = −ωµ. If z is invertible (and then ω is unitary)
this implies µ = −z, otherwise this is just one possible solution. Non uniqueness of solutions
of these equations reflect the geometric fact that the modules (cid:96)0 and (cid:96)∞ may not be in direct
sum. Explicitly, all solutions of these equations are of the form
λ = 1 + z − Ω , µ = −z + Ω,
where Ω ∈ A∗∗ is such that ωΩ = 0. In particular, zΩ = zω∗ωΩ = 0. We choose the solution
with Ω = 0. Note that λ = 1 + z, and therefore the first projection in the above composition
is given by
Next
(cid:18) 1 + z
(cid:19)
(cid:55)→
0
(cid:19)
(cid:19)
(cid:18) 1
(cid:18) 1
z
z
(cid:55)→
γ ,
(cid:19)
(cid:18) 1 + z
(cid:18) 1 + z − γ
0
.
−zγ
(cid:19)
=
(cid:18) 1
(cid:19)
.
ω
So that 1 +z− γ = and −zγ = ω, and then (the unique solution if ω is unitary, or a possible
solution that we choose, otherwise) 1 + z − γ = −zω, i.e.,
γ = (1 − z)−1.
Other solutions of the above equation are of the form
γ = (1 − z)−1 + (1 − z)−1Ω(cid:48),
where Ω(cid:48) ∈ A∗∗ is such that ωΩ(cid:48) = 0. In general, the possible endomorphisms (cid:96) → (cid:96) are given
(in these coordinates) by
(1 +z + Ω)(1 −z)−1(1 + Ω(cid:48)) = (1 + z)(1 −z)−1 + Ω(cid:48) + Ω(1−z)−1 + ΩΩ(cid:48),
(cid:19)
(cid:18) 1
z
(cid:19)
(cid:18) 1
z
(cid:55)→
where we use that ωΩ = ωΩ(cid:48) = 0, and thus (1 ± z)±1Ω = Ω (and the same for Ω(cid:48)).
As noted, if z is invertible, there is a unique solution with Ω = Ω(cid:48) = 0. Our choice of cross
ratio, picking Ω = Ω(cid:48) = 0 in any case, is justified below. First, we prove the following fact,
which is certainly well known.
23
Lemma 9.1. Let an ∈ A with (cid:107)an(cid:107) ≤ 1. If an → a strongly, then an → a strongly.
Proof. For any ξ ∈ H,
(cid:107)an2ξ − a2ξ(cid:107)2 = (cid:107)anξ(cid:107)4 + (cid:107)aξ(cid:107)4 − 2Re(cid:104)anξ, anz2ξ(cid:105).
Clearly (cid:107)anξ(cid:107) → (cid:107)aξ(cid:107) and (cid:104)anξ, anz2ξ(cid:105) → (cid:104)aξ, aa2ξ(cid:105) = (cid:107)aξ(cid:107)4, so that an2 → a2 strongly.
It is known that the square root of positive operators is strongly continuous in the unit ball: if
0 ≤ bn ≤ 1 and bn → b strongly, then b1/2
n → b1/2. Let us also sketch a proof of this fact. If
ξ ∈ H,
(cid:107)b1/2
n ξ − b1/2ξ(cid:107)2 = (cid:104)bnξ, ξ(cid:105) + (cid:104)bξ, ξ(cid:105) − 2Re(cid:104)b1/2
n ξ, b1/2ξ(cid:105),
and
(cid:104)ba1/2
n ξ, b1/2ξ(cid:105) =
t1/2dµn(t),
(cid:90) ∞
0
where µn = µbn,ξ,b1/2ξ is the scalar spectral measure of bn, associated to the pair of vectors
ξ, b1/2ξ. If p(t) is a polynomial, than clearly p(bn) → p(b) strongly, because (cid:107)bn(cid:107) ≤ 1. Then
µn converge to µ (the scalar spectral measure of a associated the the vectors ξ, b1/2ξ), when,
regarded as functionals in C(0, 1)∗, they are evaluated at polynomials. Since the norms of these
measures are uniformly bounded,
(cid:107)µn(cid:107) ≤ (cid:107)ξ(cid:107)(cid:107)b1/2
it follows that µn(t1/2) converges weakly to µ(t1/2).
n ξ(cid:107) ≤ (cid:107)ξ(cid:107)2,
Remark 9.2. Dixmier and Marechal [9] proved that the set on invertible elements of a von
Neumann algebra is strong operator dense in the algebra. The argument in [9] proceeds as
follows. Let a = ua be the polar decomposition of a (u ∈ A∗∗). First, the algebra A∗∗ is
factored in its finite and properly infinite parts. In the finite part u can be chosen unitary. In
the properly infinite part, one readily sees that it suffices to consider the cases in which u is an
isometry or a co-isometry. In the case that u is an isometry, Dixmier and Mar´echal prove that
u is the strong limit of unitaries un. If u is a co-isometry, they show that there exist invertible
elements gn which converge strongly to u, with norms (cid:107)gn(cid:107) = 1 (this is clear in the proof, though
it is not stated in their result). Summarizing, if A is a von Neumann algebra, and a ∈ A, there
exist gn ∈ G with (cid:107)gn(cid:107) ≤ (cid:107)a(cid:107) such that
SOT − lim
n→∞ gn = a.
Using this result, it is clear that, if A is a von Neumann algebra, and z ∈ D, then there
exists and zn ∈ G with (cid:107)zn(cid:107) ≤ (cid:107)z(cid:107), such that zn → z strongly.
Proposition 9.3. Let zn, z ∈ D such that zn → z strongly and (cid:107)zn(cid:107) ≤ (cid:107)z(cid:107). Then
(1 + zn)(1 − zn)−1 → (1 + z)(1 − z)−1 strongly.
Proof. First note that zn → z strongly. Then 1 ± zn converges strongly to 1 ± z. Clearly
(cid:107)1 + zn(cid:107) ≤ 2. Let us check that also (cid:107)(1 − zn)−1(cid:107) are uniformly bounded:
1 − (cid:107)z(cid:107) < ∞.
(cid:107)(1 − zn)−1(cid:107) = (cid:107)
znk(cid:107) ≤
∞(cid:88)
(cid:107)zk
n(cid:107) ≤
1
∞(cid:88)
k=0
Therefore, (1 − zn)−1 → (1 − z)−1 strongly, and since the product is strongly continuous on
norm bounded sets, the proof follows.
k=0
24
Definition 9.4. Let z ∈ D. We define cr(0, z) ∈ CR((cid:96)−∞, (cid:96)0, (cid:96)z, (cid:96)∞), for (cid:96)z =
endomorphism
cr(0, z)(cid:0)(cid:18) 1
z
(cid:19)
a(cid:1) =
(cid:18) 1
z
(cid:19)
(1 + z)(1 − z)−1a,
cr(0, z) : (cid:96)z → (cid:96)z ,
for a ∈ A.
(cid:20)(cid:18) 1
z
(cid:19)(cid:21)
, as the
We use the action of U(θ) to extend this definition to any pair z0 (cid:54)= z1 ∈ D.
Definition 9.5. Let z0, z1 ∈ D, z0 (cid:54)= z1. Pick g ∈ U(θ) such that g · 0 = z0 and denote
z = g−1 · z1. We define
cr(z0, z1) = g cr(0, z)g−1.
Before checking that the definition does not depend on the choice of g, we remark the
following. Let δ be the unique geodesic of D such that δ(0) = z0 and δ(1) = z1, and let
z−∞ = SOT − lim
t→−∞ δ(t) and z+∞ = SOT − lim
t→+∞ δ(t).
Then cr(z0, z1) ∈ CR((cid:96)z−∞, (cid:96)z0, (cid:96)z1, (cid:96)z+∞), because g is a module homomorphism which maps (cid:96)z
onto (cid:96)z1. Indeed, if x =
a ∈ (cid:96)z, then clearly
gx =
(g11 + g12z)a =
(g11 + g12z)a ∈ (cid:96)z1.
Let us check that cr(z1, z2) is well defined, i.e., that it does not depend on the choice of g.
To prove this, recall from Remark 4.10 that if k ∈ U(θ) satisfies k · 0 = 0, then k =
(cid:18) u1
0
(cid:19)
0
u2
, with u1, u2 ∈ UA.
Proposition 9.6. With the above notations, the endomorphism
cr(z0, z1) ∈ CR((cid:96)z−∞, (cid:96)z0, (cid:96)z1, (cid:96)z+∞)
does not depend on the choice of g. Namely, if h ∈ U(θ) satisfies h · 0 = z0, and z(cid:48) = h−1 · z1,
then
h cr(0, z(cid:48))h−1 = g cr(0, z)g−1.
(cid:18) u1
0
(cid:19)
0
u2
, for
(cid:19)
(cid:18) 1
(cid:18) 1
z
g · z
(cid:19)
(cid:19)
(cid:18) 1
z1
Proof. Since h · 0 = g · 0, it follows that (g−1h) · 0 = 0, and therefore h = g
u1, u2 ∈ UA. Then
z(cid:48) = h−1 · z1 =
· (g−1 · z1) =
· z = u∗
2zu1,
and
h cr(0, z(cid:48))h−1 = g
0
u2
cr(0, u∗
2zu1)
g−1.
0
u∗
2
(cid:18) u∗
1
0
(cid:19)
0
u∗
2
(cid:18) u∗
1
0
(cid:19)
(cid:18) u∗
1
0
(cid:19)
(cid:18) u1
0
u∗
2
0
(cid:19)
25
Thus, we must show that
left hand side endomorphism transforms the element
(cid:19)
= cr(0, z). Let us see how the
0
u∗
a ∈ (cid:96)z. First, it is sent to
2
(cid:19)
0
u2
cr(0, u∗
2zu1)
(cid:19)(cid:18) 1
(cid:19)
1
0
(cid:18) u∗
(cid:18) 1
(cid:18) 1
z
(cid:19)
(cid:19)
a =
z
u∗
2zu1
u∗
1a.
0
(cid:18) u1
(cid:18) u∗
(cid:18) 1
1
0
0
u∗
2
(cid:19)
(1 + u∗
2zu1)(1 − u∗
2zu1)−1u∗
1a.
u∗
2zu1
(cid:19)
2zu1)1/2 = (u∗
2zu1)∗u∗
(cid:18) u∗
1
u∗
2z
(1 + z)(1 − z)−1a.
1z∗zu1)1/2 = u∗
(cid:19)
(cid:18) u1
0
u2
0
(1 + z)(1 − z)−1a = cr(0, z)
yields
(cid:18) 1
z
(cid:19)
a.
(cid:19)
(cid:18) 1
z
The map cr(0, z(cid:48)) maps this element to
Finally, multiplying on the left by the matrix
Note that u∗
equals
2zu1 = ((u∗
1zu1. Therefore, the above element
Remark 9.7. If A is a von Neumann algebra, by the result of Dixmier and Mar´echal [9], D∩GA
is strongly dense in D. For z ∈ D ∩ GA, the set CR((cid:96)−∞, (cid:96)0, (cid:96)z, (cid:96)∞) consists of a single element
cr(0, z). If z ∈ D is non invertible, there exist zn ∈ D which are invertible such that zn → z
strongly and (cid:107)zn(cid:107) ≤ (cid:107)z(cid:107). Let us see that cr(0, zn) converge in some sense to cr(0, z). First note
that cr(0, zn), cr(0, z) are endomorphisms of different submodules. In order to compare them,
we can regard them as A-module morphisms of A2, embedding each module in A2 using the
θ-orthogonal projections p(cid:96)zn , p(cid:96)z onto the submodules (cid:96)zn, (cid:96)z, respectively. For z(cid:48) ∈ D,
p(cid:96)z(cid:48) (x) = (1 − z(cid:48)2)−1/2θ(
, x)
(1 − z(cid:48)2)−1/2.
(cid:19)
(cid:18) 1
z(cid:48)
(cid:19)
(cid:18) 1
z(cid:48)
We claim that cr(0, zn)p(cid:96)zn (x) → cr(0, z)p(cid:96)z (x) strongly in A2. By Proposition 9.3, we know
that (1 + zn)(1 − zn)−1 → (1 + z)(1 − z)−1 strongly. By a similar argument, it also holds
that (1 − zn2)−1/2 → (1 − z2)−1/2 strongly. Also these operators are uniformly bounded.
Therefore, using that the product is strongly continuous in bounded sets, our claim follows.
Remark 9.8. As a corollary we get that, even if the set CR((cid:96)1, (cid:96)2, (cid:96)3, (cid:96)4) may be empty for
general (cid:96)1, (cid:96)2, (cid:96)3, (cid:96)4, the particular set CR((cid:96)−∞, (cid:96)0, (cid:96)z, (cid:96)∞) is not, and cr(0, z) is a distinguished
element of this set.
APθ
As a first approximation of the deep relationship between the cross ratio and the metric in
1, we can state the following:
26
Theorem 9.9. Let z ∈ D, then
(cid:107)cr(0, z)(cid:107)B((cid:96)z) = d(0, z),
1
2
(cid:19)
where (cid:107) (cid:107)B((cid:96)z) denotes the norm of operators acting in (cid:96)z ⊂ A2.
(cid:18) (1 − z∗z)−1/2
Proof. Choose for (cid:96)z the unital basis ez =
z(1 − z∗z)−1/2
. Then for any x = eza ∈ (cid:96)z,
cr(0, z)x = ez log((1 + z)(1 − z)−1a,
and thus
< cr(0, z)x, cr(0, z)x >= a∗ log((1 + z)(1 − z)−1 < ez, ez > log((1 + z)(1 − z)−1a
= a∗(log((1 + z)(1 − z)−1)2a.
Since (log((1 + z)(1 − z)−1)2 ≤ (cid:107) log((1 + z)(1 − z)−1(cid:107)2, it follows that
a∗(log((1 + z)(1 − z)−1)2a ≤ a∗a(cid:107) log((1 + z)(1 − z)−1(cid:107)2,
and therefore
(cid:107) < cr(0, z)x, cr(0, z)x > (cid:107)1/2 ≤ (cid:107) log((1 + z)(1 − z)−1(cid:107)(cid:107)a∗a(cid:107)1/2
This implies that (cid:107)cr(0, z)(cid:107)B((cid:96)z) ≤ (cid:107) log((1 + z)(1 − z)−1(cid:107). Note that
= (cid:107) log((1 + z)(1 − z)−1(cid:107)(cid:107)x(cid:107).
cr(0, z)ez = ez log((1 + z)(1 − z)−1,
so that
i.e.,
(cid:107)cr(0, z)ez(cid:107)2 = (cid:107) < ez log((1 + z)(1 − z)−1, ez log((1 + z)(1 − z)−1(cid:107)
= (cid:107) log((1 + z)(1 − z)−1(cid:107)2,
(cid:107)cr(0, z)(cid:107)B((cid:96)z) = (cid:107) log((1 + z)(1 − z)−1(cid:107) = 2d(0, z).
10 The coefficient bundle.
Consider the slight variant of the commutative diagram in (4):
πθ
Qρ
πθ
/ D,
Kθ
(cid:39)
27
~
~
/
Recall (from the end of Section 4), that Qρ denotes the space of θ-orthogonal rank one projections
(considered here the coordinate free version of APθ
ξ → Qρ,
1). Let us introduce the canonical bundle
whose fiber over q ∈ Qρ is the module R(q). This is a fiber bundle of right A-modules, which
has a canonical connection. Elements of ξ are pairs (q, x), with q ∈ Qρ and q(x) = x.
Let q ∈ Qρ and ϕ : R(q) → R(q) a right module endomorphism. Pick a normalized generator
x ∈ R(q): θ(x, x) = 1 (i.e., an element of R(q) in Kθ). Then the endomorphism ϕ is determined
by the value ϕ(x) = xa. That is, for any element y = xλ ∈ R(q), ϕ(y) = xaλ. In other words,
if we regard x as a basis for R(q), ϕ can be expressed as λ (cid:55)→ aλ. We call a ∈ A the matrix of
ϕ in the basis x. If x(cid:48) is another basis of R(q) in Kθ, then there exists a unitary u ∈ UA such
that x(cid:48) = xu. If b is the matrix of ϕ in the basis x(cid:48) (i.e., ϕ(x(cid:48)) = x(cid:48)b), then
xub = x(cid:48)b = ϕ(x(cid:48)) = ϕ(xu) = ϕ(x)u = x(cid:48)au.
Then ub = au, which means that the matrix of ϕ in the basis x(cid:48) is b = u∗au .
pairs (x, a), where x ∈ Kθ with q(x) = x, and a ∈ A, with the identification
This shows that we can regard the set End(R(q)) of endomorphisms of R(q), as the set of
(x, a) ∼ (xu, u∗au), u ∈ UA.
Then, the map Γ → Qρ defined by (q, ϕ) (cid:55)→ q for q ∈ Qρ and ϕ ∈ End(R(q)), is a fiber bundle
which we call the coefficient bundle; alternatively (x, a) ∼ (xu, u∗au) (cid:55)→ [x]. Each fiber of Γ is
a C∗-algebra, which is isomorphic to A. The canonical connection of the bundle ξ induces a
connection in Γ , by the rule:
(DX ϕ)(y) = (DX ϕ)y + ϕ(DX y).
Here, ϕ is a local cross section of Γ and y is a local cross section of ξ. We remark that the
connections of ξ and Γ are compatible with the action of U(θ).
We define the basic 1-form. Given x ∈ Kθ and X ∈ (TQρ)q, with q = xθ(x,
), put
κx(X) = Xx.
Note that X is a matrix in M2(A), θ-symmetric and q-codiagonal: Xx ∈ x⊥θ = N (q).
Given X, Y ∈ (TQρ)q, we define the product
(cid:104)X, Y (cid:105)x = −θ(κx(X), κx(Y )) = −θ(Xx, Y x).
If the generator x is changed for x(cid:48) = xu, we have
(cid:104)X, Y (cid:105)x(cid:48) = −θ(X(xu), Y (xu)) = −u∗θ(Xx, Y x)u = u∗(cid:104)X, Y (cid:105)xu.
This means that, given q = [x] = [x(cid:48)], the product (cid:104)X, Y (cid:105)q is well defined as an element of the
fiber Γq.
This product is therefore a Hilbertian product in TQρ, with values in Γ. To this efect, note
that TQρ is a right module over the bundle Γ of coefficients. Indeed, if we fix x ∈ Kθ with
), the map X (cid:55)→ κx(X) = Xx from (TQρ)q to N (q) is one to one. If we change x
q = xθ(x,
with xu, κx(X) changes to κxu(X) = κx(X)u. If X ∈ (TQρ)q and ϕ ∈ Γq, we define Xϕ as
κx(Xϕ) = Xxa, where ϕ is represented by the class of (x, a). With this definition we have
(cid:104)X, Y ϕ(cid:105)q = (cid:104)X, Y (cid:105)qϕ.
28
10.1 The cross ratio, the logarithm and the exponential.
We have just defined a Hilbertian Γ-valued structure in Qρ (cid:39) APθ
particular, the product
1, or, equivalently, in D. In
(cid:19)(cid:21)
(cid:104)Log0(z), Log0(z)(cid:105)0
(cid:20)(cid:18) 1
0
takes values in the set of endomorphisms of (cid:96)0 =
, where Log0 is defined in Corollary
6.4. It is a positive module endomorphism (given by multiplying the generator e1 by a positive
element of a). Thus, it has a unique positive square root (cid:104)Log0(z), Log0(z)(cid:105)1/2
, which we shall
call the θ-modulus mod0(Log0(z)) of Log0(z). Explicitly, in the generator e1, mod0(Log0(z))
On the other hand, we saw that, for z ∈ D, the endomorphism of (cid:96)z denoted by cr(0, z), is
of
consists in multiplying the generator by log(cid:0)(1 + z)(1 − z)−1(cid:1).
given by the same coefficient log(cid:0)(1 + z)(1 − z)−1(cid:1), which multiplies the generator
(cid:18) 1
(cid:19)
0
z
(cid:96)z.
of APθ
the identification D (cid:39) APθ
We shall translate the endomorphism cr(0, z) from (cid:96)z to (cid:96)0 by means of the parallel transport
1, along the geodesic δ, with δ(0) = (cid:96)0 and δ(1) = (cid:96)z (i.e., the same former δ, which under
1 joins δ(0) = 0 and δ(1) = z in D: δ(t) = ω tanh(tα)).
0 α∗
α 0
t
· 0,
The parallel transport of elements of D (or APθ
1) along the geodesic δ(t) = e
where α is, as in Remark 6.1.2
α = z
is given by the left action of the invertible matrix
∞(cid:88)
1
(z∗z)k,
2k + 1
k=0
0 α∗
α 0
t
e
: (cid:96)0 → (cid:96)δ(t). The endomorphism cr(0, z) of (cid:96)z is transported to (cid:96)0 as
0 α∗
α 0
0 α∗
α 0
cr(0, z)e
: (cid:96)0 → (cid:96)0.
−
cr(0, z)0 := e
Our main result (for the origin) is the following:
Theorem 10.1. With the current notation, if z ∈ D (or (cid:96)z ∈ APθ
1),
emod0(Log0(z)) = cr(0, z)0 or, equivalently, mod0(Log0(z)) = log(cr(0, z)0),
(8)
where the exponential in the left hand equality is the usual exponential of A, log in the right
hand equality is the usual logarithm of G, and each endomorphism of (cid:96)0 is identified with its
coefficient in the basis e1 =
(cid:18) 1
0
(cid:19)
.
Proof. Let us prove the first equality. Since we are comparing endomorphisms of (cid:96)0, it suffices
to show that they carry the generator e1 to the same element in A2. Note that
0 α∗
α 0
−
cr(0, z)0(e1) = e
0 α∗
α 0
(cid:18) 1
(cid:19)
0
cr(0, z)e
29
(1 + z)(1 − z)−1 cosh(α).
cosh(α) (see the proof
(cid:19)
;
cr(0, z)
0 α∗
0 α∗
α 0
α 0
ω sinh(α)
(cid:18) cosh(α)
(cid:18) 1
(cid:19)
(cid:18) 1
(cid:19)
(cid:19)
(cid:18) 1
0
z
=
=
0
cosh(α)−1,
(cid:19)
(cid:18) 1
z
z
(cid:18) 1
(cid:19)
(cid:19)
(cid:18) 1
(cid:18) 1
0
0
(cid:19)
0 α∗
α 0
−
= e
using that ω tanh(α) = δ(1) = z (Lemma 6.2), this yields
0 α∗
α 0
−
e
(cid:19)
(cid:18) 1
z
cr(0, z)
−
cosh(α) = e
By the same computation that showed that e
of Lemma 6.2), we have that
0 α∗
α 0
−
e
i.e.,
On the other hand, the endomorphism mod0(Log0(z) sends e1 to e1 log(cid:0)(1 + z)(1 − z)−1(cid:1),
(1 + z)(1 − z)−1.
cr(0, z)0(e1) =
and thus
emod0(Log0(z))(e1) =
(1 + z)(1 − z)−1.
As in Definition 9.5, let z0 (cid:54)= z1 ∈ D ((cid:96)z0 (cid:54)= (cid:96)z1 ∈ APθ
1). Pick g ∈ U(θ) such that g · 0 = z0,
and denote by z = g−1 · z1 as before. Let δ be the geodesic such that δ(0) = 0 and δ(1) = z1.
Then δz0,z1 = g · δ is the geodesic which joins z0 and z1 at t = 0 and t = 1, respectively. Recall
that cr(z0, z1) = gcr(z0, z1)g−1. Likewise, we put
Logz0(z1) := gLog0(z)g−1, and modz0(z1) = (cid:104)Logz0(z1), Logz0(z1)(cid:105)1/2
,
z0
where (cid:104)ϕ, ψ(cid:105)z0
= g(cid:104)gϕg−1, gψg−1(cid:105)0g−1, and Logz0 is the inverse of the exponential expz0 :
(TD)z0 → D. It is not difficult to verify that these definitions do not depend on the choice of g.
Finally, let us denote by cr(z0, z1)z0 the parallel transport of cr(z0, z1) from (cid:96)z1 to (cid:96)z0 along
the geodesic δz0,z1 (obtained by conjugation as in the case of the origin, by the value at t = 1
of the one parameter group in U(θ) which determines δz0,z1). The U(θ)-covariance of the data
involved enables one to prove the following:
Corollary 10.2. With the current notations,
In particular, (cid:107)Logz0(z1)(cid:107)z0 = (cid:107) log(cr(z0, z1)z0)(cid:107).
modz0(Logz0(z1)) = log (cr(z0, z1)z0) .
Figure 2.
30
11 An example.
Suppose that the algebra A has a trace tr onto a central subalgebra, that is, there exists a C∗-
subalgebra B ⊂ Z(A) of the center of A and a conditional expectation tr : A → B satisfying
tr(xy) = tr(yx) for all x, y ∈ A. This happens, for instance, if A is a finite von Neumann
algebra.
A relevant case of this situation is the following. Consider a complex vector bundle E → B
with compact base space B, endowed with a Riemannian metric < e, e(cid:48) >b, b ∈ B, e, e(cid:48) ∈ Eb (the
fiber of E over b). Consider the fiber bundle End(E) → B of endomorphisms of the vector bundle
E, and let A be the algebra Γ(End(E)) of the continuous global cross sections of End(E). Since
each Eb is a (finite dimensional) Hilbert space, End(Eb) is a C∗-algebra. The space Γ(End(E))
of cross sections has therefore the norm (cid:107)ϕ(cid:107) = supb∈B (cid:107)ϕb(cid:107), where ϕb : Eb → Eb and (cid:107)ϕb(cid:107) is
the usual norm of linear operators. With this norm, A is a C∗-algebra. The center Z(A) of
this algebra is the space of scalar sections λ in End(E) (homotetic in each fiber). The central
trace is given by tr : A → Z(A), tr(σ)b = T r(σb), b ∈ B, with T r the usual trace of Eb. More
specifically, B could be a compact manifold, and E the complexification of its tangent bundle,
with an Hermitian metric. This case is interesting due to the following observation:
in our
previous work [1], we noticed the equivalence, as homogeneous spaces, of the disk D and the
Poincar´e halfspace H of the algebra A. This homogeneous space can be thought as the tangent
bundle T G+ of the space G+ of positive and invertible elements of A, as explained in [1]. In
this context, an element of H is a pair (X, a) with a ∈ G+ and X ∈ (T G+)a. The element
a ∈ G+ represents a Riemannian metric in B, and a possible vector X (a selfadjoint element of
A) could be the Ricci curvature of the metric a. In this manner, the geometry of H is linked to
the deformation of the pairs (Riemannian metric, Ricci curvature), viewed as elements of T G+.
Back to the general case (of this example):
trA → B ⊂ Z(A),
we can define a Hilbertian B-valued inner product, by means of
(cid:104)X, Y (cid:105)tr,q = −tr(θ(Xx, Y y)).
31
Indeed, since tr is tracial, the value of −tr(θ(Xx, Y y)) is independent of the choice of x ∈ Kθ
satisfying q = xθ(x, ). On the other hand, cr(z0, z1) is an element of Γz0, which has matrix a in
a unital base x ∈ R(q), as explained before. We put cr(z0, z1)tr, for tr(a). Clearly, cr(z0, z1)tr
does not depend on the basis x. With these notations, the formula in Corollary 10.2, can be
written
(cid:104)Logz0z1, Logz0z1(cid:105)1/2
tr = log cr(z0, z1)tr,
(9)
which is an identity involving elements in B.
More specifically, if A is commutative, we can choose tr the identity A = B, and we have
as elements in A.
Logz0z1 = log cr(z0, z1),
(10)
References
[1] Andruchow, E.; Corach, G.; Recht, L., Poincar´e half-space of a C∗-algebra, Rev. Mat.
Iberoam. (to appear), preprint arXiv:1711.08802.
[2] Andruchow, E.; Corach, G.; Stojanoff, D. Projective spaces of a C∗-algebra, Integral Equa-
tions Operator Theory 37 (2000), 143–168.
[3] Andruchow, E.; Corach, G.; Stojanoff, D., Projective space of a C∗-module, Infin. Dimens.
Anal. Quantum Probab. Relat. Top. 4 (2001), 289–307.
[4] Andruchow, E.; Recht, L.; Varela, A., Metric geodesics of isometries in a Hilbert space and
the extension problem, Proc. Amer. Math. Soc. 135 (2007), 2527–2537
[5] Corach, G.; Larotonda, A. R., Stable range in Banach algebras, J. Pure Appl. Algebra 32
(1984), 289–300.
[6] Corach, G.; Porta, H.; Recht, L., Differential geometry of spaces of relatively regular oper-
ators, Integral Equations Operator Theory 13 (1990), 771–794.
[7] Corach, G.; Porta, H.; Recht, L., The geometry of the space of selfadjoint invertible elements
in a C∗-algebra, Integral Equations Operator Theory 16 (1993), 333–359.
[8] Corach, G.; Porta, H.; Recht, L., The geometry of spaces of projections in C∗-algebras,
Adv. Math. 101 (1993), 59–77.
[9] Dixmier, J.; Mar´echal, O., Vecteurs totalisateurs d'une alg`ebre de von Neumann, Comm.
Math. Phys. 22 (1971), 44–50.
[10] Gromov, M., Metric structures for Riemannian and non-Riemannian spaces, Modern
Birkhuser Classics, Birkhauser Boston, 2007.
[11] Lance, E. C., Hilbert C∗-modules. A toolkit for operator algebraists. London Mathematical
Society Lecture Note Series, 210. Cambridge University Press, Cambridge, 1995.
32
[12] Manuilov, V. M.; Troitsky, E. V., Hilbert C∗-modules. Translated from the 2001 Russian
original by the authors. Translations of Mathematical Monographs, 226. American Mathe-
matical Society, Providence, RI, 2005.
[13] Mata-Lorenzo, L. E.; Recht, L., Infinite-dimensional homogeneous reductive spaces. Acta
Cient. Venezolana 43 (1992), no. 2, 76–90.
[14] Porta, H.; Recht, L., Geometric embeddings of operator spaces, Illinois J. Math. 40 (1996),
151–161.
[15] Rieffel, M., A. Dimension and stable rank in the K-theory of C∗-algebras. Proc. London
Math. Soc. (3) 46 (1983), 301–333.
[16] Zelikin, M. I., Geometry of the cross ratio of operators. (Russian) ; translated from Mat.
Sb. 197 (2006), 39–54 Sb. Math. 197 (2006), 37–51.
Esteban Andruchow
Instituto de Ciencias, Universidad Nacional de Gral. Sarmiento,
J.M. Gutierrez 1150, (1613) Los Polvorines, Argentina
and Instituto Argentino de Matem´atica, 'Alberto P. Calder´on', CONICET,
Saavedra 15 3er. piso, (1083) Buenos Aires, Argentina.
e-mail: [email protected]
Gustavo Corach
Instituto Argentino de Matem´atica, 'Alberto P. Calder´on', CONICET,
Saavedra 15 3er. piso, (1083) Buenos Aires, Argentina,
and Depto. de Matem´atica, Facultad de Ingenier´ıa, Universidad de Buenos Aires, Argentina.
e-mail: [email protected]
L´azaro Recht
Departamento de Matem´atica P y A, Universidad Sim´on Bol´ıvar
Apartado 89000, Caracas 1080A, Venezuela
e-mail: [email protected]
33
|
1510.03328 | 1 | 1510 | 2015-10-12T15:12:47 | Principal functions for bi-free central limit distributions | [
"math.OA"
] | We find the principal function of the completely non-normal operator l(v_1)+l(v_1)^*+i(r(v_2)+r(v_2)^*) on a subspace of the full Fock space F(H) which arises from a bi-free central limit distribution. As an application, we find the essential spectrum of this operator. | math.OA | math |
PRINCIPAL FUNCTIONS FOR BI-FREE CENTRAL LIMIT DISTRIBUTIONS
KENNETH J. DYKEMA AND WONHEE NA
Abstract. We find the principal function of the completely non-normal operator l(v1)+l(v1)∗+i(r(v2)+
r(v2)∗) on a subspace of the full Fock space F(H) which arises from a bi-free central limit distribution.
As an application, we find the essential spectrum of this operator.
1. Introduction and Preliminaries
Bi-free independence was introduced by Voiculescu as a generalization of free independence in a non-
commutative probability space (A, ϕ). He considered a two-faced family of non-commutative random
variables, (X1, X2), in (A, ϕ) and the expectation values for such a combined system of left and right
variables. In [7], Voiculescu proved a bi-free central limit theorem and described the family of distributions
that appear as limits. These are called bi-free central limit distributions.
ii ∈ I(cid:105) → L(X ) and r : C(cid:104)z
1.1. Bi-freeness. Let z = ((zi)i∈I , (zj)j∈J ) be a two-faced family in a non-commutative probability
space (A, ϕ) where I and J are disjoint index sets.
Definition 1.1 ([7]). The two-faced families z(cid:48) and z(cid:48)(cid:48) are said to be bi-freely independent (abbreviated
bi-free) if there exist two vector spaces (X (cid:48),X (cid:48)◦, ξ(cid:48)) and (X (cid:48)(cid:48),X (cid:48)(cid:48)◦, ξ(cid:48)(cid:48)) with specified state vectors and
unital homomorphisms l : C(cid:104)z
jj ∈ J(cid:105) → L(X ) with ∈ {(cid:48),(cid:48)(cid:48) } such
that the two-faced families T = ((λ ◦ l(z
j))j∈J ) have the same joint distribution in
(L(X ), ϕξ) as z(cid:48) and z(cid:48)(cid:48) where (X ,X ◦, ξ) = (X (cid:48),X (cid:48)◦, ξ(cid:48)) ∗ (X (cid:48)(cid:48),X (cid:48)(cid:48)◦, ξ(cid:48)(cid:48)) and λ and ρ are left and right
representations of L(X ) on L(X ).
Definition 1.2 ([7]). For each map α : {1, ..., n} → I (cid:113) J there is a unique universal polynomial Rα in
commuting variables Xα(k1)···α(kr), 1 ≤ k1 < ··· < kr ≤ n such that
(i) Rα is homogeneous of degree n where Xα(k1)···α(kr) is assigned degree r,
(ii) the coefficient of Xα(1)···α(n) is 1, and
j)j∈J ) and z(cid:48)(cid:48) = ((z(cid:48)(cid:48)
(iii) if z(cid:48) = ((z(cid:48)
j )j∈J ) are bi-free pairs of two-faced families of non-
i ))i∈I , (ρ ◦ r(z
i )i∈I , (z(cid:48)(cid:48)
commutative random variables in (A, ϕ), then
i)i∈I , (z(cid:48)
Rα(z(cid:48)) + Rα(z(cid:48)(cid:48)) = Rα(z(cid:48) + z(cid:48)(cid:48))
where Rα(z) = Rα(ϕ(zα(k1) ··· zα(kr))1 ≤ k1 < ··· < kr ≤ n).
These polynomials Rα are called bi-free cumulants.
Theorem 1.3 ([7]). A two-faced family z of non-commutative random variables has a bi-free central
limit distribution if and only if Rα(z) = 0 whenever α : {1, ..., n} → I (cid:113) J with n = 1 or n ≥ 3.
We now recall the notion of a two-faced system with rank ≤ 1 commutation given in [8].
Definition 1.4. An implemented non-commutative probability space is a triple (A, ϕ, P ) where (A, ϕ) is
a non-commutative probability space and P = P 2 ∈ A is an idempotent so that
P aP = ϕ(a)P for all a ∈ A.
Date: October 12, 2015.
2000 Mathematics Subject Classification. 46L54 (47A65).
Key words and phrases. Bi-freeness, bi-free central limit distribution, principal function.
Research supported in part by NSF grant DMS -- 1202660.
1
2
DYKEMA AND NA
An implemented C∗-probability space (A, ϕ, P ) will satisfy additional requirements that (A, ϕ) is a C∗-
probability space and that P = P ∗.
If a two-faced family ((zi)i∈I , (zj)j∈J ) in an implemented non-
commutative probability space (A, ϕ, P ) satisfies that
[zi, zj] = λi,jP for some λi,j ∈ C, i ∈ I, j ∈ J,
then the family ((zi)i∈I , (zj)j∈J ) is called a system with rank ≤ 1 commutation where (λi,j)i∈I,j∈J is the
coefficient matrix of the system.
Definition 1.5. Let H be a complex Hilbert space. Then the full Fock space on H is
F(H) = CΩ ⊕(cid:77)
H⊗n
n≥1
where Ω is called the vacuum vector and has norm one. The vacuum expectation is defined as ϕΩ = (cid:104)·Ω, Ω(cid:105)
on F(H). For ξ ∈ H, the left creation operator l(ξ) ∈ B(F(H)) is given by the formulas l(ξ)Ω = ξ and
l(ξ)(ξ1 ⊗ ··· ⊗ ξn) = ξ ⊗ ξ1 ⊗ ··· ⊗ ξn
for all n ≥ 1 and ξ1,··· , ξn ∈ H. The adjoint l(ξ)∗ of l(ξ) is called the left annihilation operator. The
right creation operator r(ξ) ∈ B(F(H)) is determined by the formulas r(ξ)Ω = ξ and
r(ξ)(ξ1 ⊗ ··· ⊗ ξn) = ξ1 ⊗ ··· ⊗ ξn ⊗ ξ
for all n ≥ 1 and ξ1,··· , ξn ∈ H. The adjoint r(ξ)∗ of r(ξ) is called the right annihilation operator.
Theorem 1.6 (Theorem 7.4 of [7]). For each matrix C = (Ckl)k,l∈I(cid:113)J with complex entries, there is
exactly one bi-free central limit distribution ϕC : C(cid:104)Zkk ∈ I (cid:113) J(cid:105) → C so that
ϕC(ZkZl) = Ckl for each k, l ∈ I (cid:113) J.
If h, h(cid:48) : I (cid:113) J → H are maps into the Hilbert space H and we define
zi = l(h(i)) + l∗(h(cid:48)(i)) if i ∈ I
zj = r(h(j)) + r∗(h(cid:48)(j)) if j ∈ J
then z = ((zi)i∈I , (zj)j∈J ) has a bi-free central limit distribution ϕC where Ckl = (cid:104)h(l), h(cid:48)(k)(cid:105). Every
bi-free central limit distribution when I and J are finite can be obtained in this way.
Remark 1.7. The bi-free two-faced system in Theorem 1.6 is an example of rank ≤ 1 commutation.
Indeed, (B(F(H)), ϕΩ, P ) is an implemented C∗-probability space where ϕΩ is the vacuum expectation
and P is a projection on CΩ. We have [zi, zj] = ((cid:104)h(j), h(cid:48)(i)(cid:105) − (cid:104)h(i), h(cid:48)(j)(cid:105))P .
1.2. Principal function of a completely non-normal operator. Let T be a completely non -- normal
operator on a Hilbert space H with self-commutator T ∗T − T T ∗ = −2C that is trace class. Set U =
2 (T + T ∗) and V = − 1
2 i(T − T ∗). Consider the C*-algebra generated by C and the identity operator on
1
H; it is isometrically isomorphic to C(σ(C)), the complex valued continuous functions on σ(C), by the
Gelfand-Naimark theorem. Consider the function on σ(C),
√−t,
−i
√
0,
t,
t (cid:55)→
t < 0
t = 0
t > 0
and there exists the unique element C in the C*-algebra corresponding to this function by the Gelfand
transform. Note that C 2 = C and C C∗ = C∗ C = C.
The determining function of the operator T is defined to be
E(l, s) = I +
C(V − l)−1(U − s)−1 C
1
i
BIFREE CENTRAL LIMIT DISTRIBUTIONS
3
for l ∈ C \ σ(V ) and s ∈ C \ σ(U ). Then E(l, s), for each fixed l and s, is an invertible element in the
C*-algebra generated by T and I. Since det(I + AB) = det(I + BA) when A is compact with AB and
BA in trace class, we have
(cid:18)
det E(l, s) = det
(cid:19)
1
i
I +
C(V − l)−1(U − s)−1
= det(cid:0)(V − l)(U − s)(V − l)−1(U − s)−1(cid:1) .
(cid:19)
(cid:90)(cid:90)
(cid:18) 1
dδ
δ − l
dγ
γ − s
The principal function g is defined in [4] to be the element of L1(R2) such that
(1)
det E(l, s) = exp
(2)
It is known that supp(g) is contained in {(δ, γ) ∈ R2 γ + iδ ∈ σ(T )}. Moreover, it is a complete unitary
invariant for T if C has one dimensional range; that is, two completely non-normal operators T and T (cid:48)
are unitarily equivalent if and only if their principal functions agree, assuming each of T and T (cid:48) has a
self-commutator with one dimensional range. In Theorem 8.1 of [4], it is proved that
g(δ, γ)
2πi
.
g(δ, γ) = ind(T − (γ + iδ))
if γ + iδ is not in the essential spectrum σe(T ). This result implies that the principal function g of T is
an extension of the Fredholm index of T − z to the whole plane. However, it is not the typical situation
that g assumes only integer values on the plane; indeed the map T (cid:55)→ g is onto, namely (see [5]), any
summable function on R2 with compact support is the principal function of a completely non-normal
operator with a trace class self-commutator.
2. The principal function of certain operators
2.1. Let H be a Hilbert space and v1, v2 ∈ H. We consider the operator T on F(H) given by
T = X1 + iX2,
with X1 = l(v1) + l(v1)∗, X2 = r(v2) + r(v2)∗.
[T, T ∗] = 4(Im(cid:104)v2, v1(cid:105))P.
This arises from the bi-free central limit distribution and was described in Example 3.10 of [8]. As we
discussed in Section 1, we have [X1, X2] = 2i(Im(cid:104)v2, v1(cid:105))P in the implemented C∗-probability space
(B(F(H)), ϕΩ, P ), so that
(3)
Both the spectrum and the essential spectrum of X1 on F(H) equal [−2(cid:107)v1(cid:107), 2(cid:107)v1(cid:107)] and those of X2
equal [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)]. By the following easy lemma, which is well known but whose proof we include
for convenience, the spectrum of the operator T = X1 + iX2 on F(H) is contained in [−2(cid:107)v1(cid:107), 2(cid:107)v1(cid:107)] +
i[−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)]. Throughout this paper, we are interested in non-normal operators T ; so we assume that
Im(cid:104)v2, v1(cid:105) is non-zero.
Lemma 2.1. If A and B are self-adjoint with σ(A) ⊆ [r1, r2] and σ(B) ⊆ [t1, t2], then σ(A + iB) ⊆
[r1, r2] + i[t1, t2].
1, B1 ≥ 0, and B1 is invertible, then A1 + iB1 = B
Proof. If A1 = A∗
1 is invertible
− 1
is self-adjoint. Suppose a + ib /∈ [r1, r2] + i[t1, t2] . Then either a < r1 or a > r2 or
since B
b < t1 or b > t2. If b < t1, then A + iB − (a + ib) = (A − a) + i(B − b) and B − b ≥ 0 is invertible. So
1
a + ib /∈ σ(A + iB). If b > t2, then a + ib− (A + iB) = (a− A) + i(b− B) and b− B ≥ 0 is invertible, so that
a+ib /∈ σ(A+iB). Since (A−a)+i(B−b) = i((B−b)−i(A−a)), we can easily show that A+iB−(a+ib)
is invertible for each case of a < r1 and a > r2. Therefore, σ(A + iB) ⊆ [r1, r2] + i[t1, t2].
(cid:3)
The operator T ∈ B(H) is said to be hyponormal, if its self-commutator T ∗T − T T ∗ is positive.
Furthermore, if there is no reducing subspace of T , the restriction of T to which is normal, then T is said
to be pure hyponormal or completely non-normal hyponormal.
Theorem 2.2 (Theorem 2.1.3 of [6]). Let T ∈ B(H) be a hyponormal operator with [T ∗, T ] = D. Then
there is a unique orthogonal decomposition H = Hp(T ) ⊕ Hn(T ) where Hp(T ) and Hn(T ) are reducing
subspaces for T , such that
− 1
1 A1B
− 1
1 A1B
− 1
1 + i
(cid:16)
(cid:17)
1
2
1
B
B
1
2
2
2
2
2
4
DYKEMA AND NA
(i) Tp = THp(T ) is pure hyponormal,
(ii) Tn = THn(T ) is normal.
Moreover,
(cid:95){T ∗kT lD(H) k, l ∈ N}
and
Hn(T ) = {ζ ∈ H DT ∗lT kζ = 0 for every k, l ∈ N}.
Hp(T ) =
As we can see in (3), if Im(cid:104)v2, v1(cid:105) ≤ 0 (or ≥ 0), then T = X1 + iX2 is a hyponormal operator (or
cohyponormal, respectively) on F(H). By Theorem 2.2, the pure parts Hp(T ) and Hp(T ∗) of T and T ∗
are equal to alg(T, T ∗, 1)Ω.
Assuming that Im(cid:104)v2, v1(cid:105) ≤ 0, if v2 is a scalar mutiple of v1, then alg(T, T ∗, 1)Ω is dense in F(C(cid:104)v1(cid:105))
so that T is pure hyponormal on F(C(cid:104)v1(cid:105)). However, if v2 is not a scalar multiple of v1, then T is not
a pure hyponormal operator on F(C(cid:104)v1, v2(cid:105)), that is, there exists a nontrivial reducing subspace N of T
in F(C(cid:104)v1, v2(cid:105)) such that TN is normal. For, suppose that u is a unit vector which is orthogonal to v1
in C(cid:104)v1, v2(cid:105) and v2 = cv1 + du where c, d ∈ C are non-zero. Since v2 and cc2 v1 − dd2 u are orthogonal to
each other, for each m, n ∈ N,
u ⊗
(l(v1) + l(v1)∗)m
∈ span
(cid:18)
(cid:18)
(cid:18) c
c2 v1 − d
d2 u
(cid:18) c
c2 v1 − d
d2 u
(cid:19)(cid:19)
(cid:19)(cid:19)
∈ span
and
(r(v2) + r(v2)∗)n
u ⊗
u ⊗
1 ⊗ u ⊗
v⊗k
(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) k ∈ N
(cid:27)
(cid:18) c
c2 v1 − d
d2 u
(cid:12)(cid:12)(cid:12)(cid:12) k ∈ N
(cid:27)
(cid:19)
(cid:19)(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) m, n ∈ N
(cid:27)
(cid:26)
(cid:18) c
(cid:26)
c2 v1 − d
d2 u
(cid:18) c
(cid:18)
c2 v1 − d
(cid:12)(cid:12)(cid:12)(cid:12) m, n ∈ N
(cid:27)
d2 u
⊗ v⊗k
u ⊗
2
.
.
Since alg(T, T ∗, 1) = alg(X1, X2, 1) and [X1, X2] = 2i(Im(cid:104)v2, v1(cid:105))P ,
(cid:95)(cid:26)
(cid:95)(cid:26)
N :=
(r(v2) + r(v2)∗)n(l(v1) + l(v1)∗)m
(cid:18) c
c2 v1 − d
d2 u
(cid:19)
⊗ v⊗n
2
=
1 ⊗ u ⊗
v⊗m
is a nontrivial reducing subspace of T in F(C(cid:104)v1, v2(cid:105)) which is orthogonal to CΩ. Clearly, the restrictions
of l(v1) + l(v1)∗ and r(v2) + r(v2)∗ to N commute, so the restriction of T to N is normal.
Now we will characterize the pure part alg(T, T ∗, 1)Ω of T in F(H) when v1 and v2 are linearly
independent.
Proposition 2.3. Let T = l(v1) + l(v1)∗ + i(r(v2) + r(v2)∗) where (cid:107)v1(cid:107) = 1. Suppose v2 = cv1 + du
where c, d ∈ C are non-zero, u ⊥ v1 and (cid:107)u(cid:107) = 1, and let w := 1√
. Let An be the span
of length n tensor products in F(C(cid:104)v1, v2(cid:105)) for each n ∈ N and let A0 = CΩ. Then
(cid:16) cc2 v1 − dd2 u
(cid:17)
2
alg(T, T ∗, 1)Ω =
(An ∩ alg(T, T ∗, 1)Ω)
(cid:77)
n≥0
and for every n ∈ N,
Bn := {v⊗n
1
, v⊗n−1
1
⊗ u, v⊗n−2
1
⊗ u ⊗ v2,··· , v1 ⊗ u ⊗ v⊗n−2
2
, u ⊗ v⊗n−1
2
}
and
n := {v⊗n
B(cid:48)
2
, w ⊗ v⊗n−1
2
, v1 ⊗ w ⊗ v⊗n−2
2
,··· , v⊗n−2
1
⊗ w ⊗ v2, v⊗n−1
1
⊗ w}
are orthogonal bases of An ∩ alg(T, T ∗, 1)Ω. Furthermore, we have the obvious isomorphisms
alg(T, T ∗, 1)Ω ∼= F(C(cid:104)v1(cid:105)) ⊕ (F(C(cid:104)v1(cid:105)) ⊗ u ⊗ F(C(cid:104)v2(cid:105)))
∼= F(C(cid:104)v2(cid:105)) ⊕ (F(C(cid:104)v1(cid:105)) ⊗ w ⊗ F(C(cid:104)v2(cid:105))) .
(4)
(5)
(6)
Proof. We will prove by induction on n that Bn is an orthogonal basis for An ∩ alg(T, T ∗, 1). This is
clear for n = 1. For n = 2, consider the orthogonal basis of A2
Z2 = {v⊗2
1 , v1 ⊗ u, u ⊗ v2, u ⊗ w}
BIFREE CENTRAL LIMIT DISTRIBUTIONS
5
1 , v1⊗u, u⊗v2} ⊆ alg(T, T ∗, 1)Ω and Z2\B2 = {u⊗w} ⊆ (alg(T, T ∗, 1)Ω)⊥
containing B2. Here, B2 = {v⊗2
as we saw in the above argument describing N . Now the assertion is proved for n = 2.
2 , v2⊗ w, w⊗ v2, w⊗2}. Then Z3 := {v1⊗ Z2}∪{u⊗
Consider another orthogonal basis of A2, Z(cid:48)
2} is an orthogonal basis of A3. Since alg(T, T ∗, 1)Ω is a reducing subspace of T , v1⊗B2 ⊆ alg(T, T ∗, 1)Ω
Z(cid:48)
2, only u ⊗ v⊗2
and v1 ⊗ {Z2 \ B2} ⊆ (alg(T, T ∗, 1)Ω)⊥. In u ⊗ Z(cid:48)
is contained in alg(T, T ∗, 1)Ω because
every tensor product in F(C(cid:104)v1, v2(cid:105)) which starts with u and ends with w belongs to N and is therefore
⊥
orthogonal to alg(T, T ∗, 1)Ω; moreover u ⊗ w ⊗ v2 = (r(v2) + r(v2)∗)(u ⊗ w) ∈ (alg(T, T ∗, 1)Ω)
. Hence,
B3 = {v1 ⊗ B2} ∪ {u ⊗ v⊗2
2 } is contained in alg(T, T ∗, 1)Ω and Z3 \ B3 is contained in (alg(T, T ∗, 1)Ω)⊥.
Thus the assertion holds for n = 3.
The induction step for general n proceeds similarly. For each n ∈ N, construct an orthogonal basis Zn
2 = {v⊗2
2
for An as follows.
Zn = {vn
1 } ∪
{vn−j
1 ⊗ u ⊗ Z(cid:48)
j−1}
(cid:91)
1≤j≤n
,
1
1 ⊗ u, v⊗n−1
, v⊗n
where Z(cid:48)
k is the set of all length k tensor products in F(H) whose components consist of v2 and w. The
induction hypothesis is that Bj is an orthogonal basis of Aj ∩ alg(T, T ∗, 1)Ω and Zj \ Bj is orthogonal
to alg(T, T ∗, 1)Ω for each 1 ≤ j ≤ n. Then Zn+1 = {v1 ⊗ Zn} ∪ {u ⊗ Z(cid:48)
n} and it is an orthogonal basis
of An+1. Since alg(T, T ∗, 1)Ω is a reducing subspace of T and is invariant under l(v1) + l(v1)∗, we have
v1 ⊗ Bn = {v⊗n+1
} ⊆ alg(T, T ∗, 1)Ω and v1 ⊗ {Zn \ Bn} ⊆
(alg(T, T ∗, 1)Ω)⊥. In u ⊗ Z(cid:48)
is contained in alg(T, T ∗, 1)Ω and the other elements are
orthogonal to alg(T, T ∗, 1)Ω by the induction hypothesis. Therefore, Bn+1 = {v1 ⊗ Bn} ∪ {u ⊗ v⊗n
2 }
is an orthogonal basis for An+1 ∩ alg(T, T ∗, 1)Ω and Zn+1 \ Bn+1 is an orthogonal basis for An+1 ∩
. Thus, for every n ∈ N, Bn is an orthogonal basis for the set of all length n tensor
(alg(T, T ∗, 1)Ω)
products in alg(T, T ∗, 1)Ω. This finishes the proof by induction.
n is also an orthogonal basis for An ∩ alg(T, T ∗, 1)Ω follows similarly by
induction on n, using the invariance of alg(T, T ∗, 1)Ω under r(v2) + r(v2)∗ rather than l(v1) + l(v1)∗.
(cid:3)
⊗ u ⊗ v2,··· , v1 ⊗ u ⊗ v⊗n−1
The proof that for n ∈ N, B(cid:48)
n, only u ⊗ v⊗n
The equality (4) follows by the above proofs.
⊥
2
1
2
Before we further investigate the operator T = X1 +iX2 having v1 and v2 linearly independent, we will
take a look at the case when the vectors v1 and v2 are linearly dependent. We will refer to the following
result.
Theorem 2.4 ([2]). If T is a hyponormal operator on H, then C∗(T ) is generated by the unilateral shift
if and only if T is unitarily equivalent to S, where S satisfies conditions
(i) S is irreducible;
(ii) self-commutator S∗S − SS∗ is compact;
(iii) σe(S) is a simple closed curve;
(iv) σ(S) is the closure of V , where V is the bounded component of C\σe(S);
(v) for λ ∈ σ(S)\σe(S), ind(S − λ) = −1.
Example 2.5. Let v1 = αv2, α ∈ C, Im α (cid:54)= 0, and (cid:107)v2(cid:107) = 1. Let T be given by
T = l(v1) + l(v1)∗ + i(r(v2) + r(v2)∗)
= (αl(v2) + ir(v2)) + (¯αl(v2)∗ + ir(v2)∗)
on F(C). Then,
and for each n ∈ N,
Therefore,
T (Ω) = (α + i)v2
T (v⊗n
2
) = (α + i)v⊗n+1
2
+ (¯α + i)v⊗n−1
2
.
T = (α + i)U + (¯α + i)U∗,
where U is the unilateral shift on F(C).
and [T ∗, T ] = −4P , so T is cohyponormal.
If α = i, then T = 2iU and [T ∗, T ] = 4P so that T is a hyponormal operator. If α = −i, then T = 2iU∗
6
DYKEMA AND NA
Since the image of the unilateral shift U in the Calkin algebra is a normal operator, by the functional
calculus, we have
σe((α + i)U + (¯α + i)U∗) = {(α + i)t + (¯α + i)¯t t ∈ σe(U )}
= {αt + αt + i(t + ¯t) t ∈ T}.
This curve is the solution set of
x2 + α2y2 − 2(Re α)xy = 4(Im α)2
(7)
in the xy-plane, which is an ellipse centered at the origin. So the essential spectrum of T is a simple
closed curve. Let V0 be the bounded component of C\σe(T ). Then by Theorem 2.4, we have
and for λ ∈ σ(T )\σe(T ),
ind(T − λ) =
Im (α) > 0
Im (α) < 0.
σ(T ) = V0,
(cid:40)−1,
1,
Thus, the principal function is the characteristic function of the interior of the ellipse (7) when Im α < 0,
and is the negative of this when Im α > 0.
In the rest of this paper, we consider the pure part of T = X1 + iX2 acting on alg(T, T ∗, 1)Ω where
2.2.
X1 = l(v1) + l(v1)∗ and X2 = r(v2) + r(v2)∗. So T is a completely non-normal operator.
Now we will find a formula for the principal function of T when v1 and v2 are linearly independent. For
this, we will use equation (2); so we will first establish a formula for det E(l, s) of T . Suppose l ∈ C\σ(X2)
and s ∈ C \ σ(X1). From (1), we have
det E(l, s) = det((X2 − l)(X1 − s)(X2 − l)−1(X1 − s)−1)
= det(((X1 − s)(X2 − l) − 2Im(cid:104)v2, v1(cid:105)iP )(X2 − l)−1(X1 − s)−1)
= det(1 − 2Im(cid:104)v2, v1(cid:105)iP (X2 − l)−1(X1 − s)−1)
= det(1 − 2Im(cid:104)v2, v1(cid:105)iP 2(X2 − l)−1(X1 − s)−1)
= det(1 − 2Im(cid:104)v2, v1(cid:105)iP (X2 − l)−1(X1 − s)−1P )
= det(1 − 2Im(cid:104)v2, v1(cid:105)iϕΩ((X2 − l)−1(X1 − s)−1)P )
= 1 − 2Im(cid:104)v2, v1(cid:105)iϕΩ((X2 − l)−1(X1 − s)−1)
= 1 − 2Im(cid:104)v2, v1(cid:105)iϕΩ((¯s − X1)−1(¯l − X2)−1)
= 1 − 2Im(cid:104)v2, v1(cid:105)iG(X1,X2)(¯s, ¯l)
(8)
where G(X1,X2)(z, w) = ϕ((z− X1)−1(w− X2)−1). Note that G(X1,X2)(z, w) is the germ of a holomorphic
function near (∞,∞) in C∞ × C∞ (see [8]).
2.3. We review the definition and formula of the partial bi-free R-transform, R(a,b)(z, w), defined in [8]
and find det E(l, s) in terms of l and s.
Definition 2.6 ([8]). Let (a, b) be a two-faced pair of non-commutative random variables in (A, ϕ). Set
I = {i} and J = {j} and suppose α : {1,··· , m + n} → I (cid:113) J is given by α(k) = i if 1 ≤ k ≤ m and
α(k) = j if m + 1 ≤ k ≤ m + n. We shall denote the bi-free cummulant Rα as Rm,n. The partial bi-free
R-transform is the generating series
R(a,b)(z, w) =
Rm,n(a, b)zmwn.
(cid:88)
m≥0,n≥0
m+n≥1
BIFREE CENTRAL LIMIT DISTRIBUTIONS
7
Theorem 2.7 (Theorem 2.4 of [8]). We have the equality of germs of holomorphic functions near (0, 0) ∈
C2,
R(a,b)(z, w) = 1 + zRa(z) + wRb(w) −
zw
G(a,b)(Ka(z), Kb(w))
,
where Ra(z) and Rb(w) are one variable R-transforms, and Ka(z) = z−1 + Ra(z) and Kb(w) = w−1 +
Rb(w).
For the given two-faced pair (X1, X2), the definition of the partial bi-free R-transform and Lemma 7.2
of [7] give
R(X1,X2)(z, w) = R2,0(X1, X2) + R0,2(X1, X2) + R1,1(X1, X2)
1 )z2 + ϕ(X 2
= ϕ(X 2
= (cid:107)v1(cid:107)2z2 + (cid:107)v2(cid:107)2w2 + (cid:104)v2, v1(cid:105)zw.
2 )w2 + ϕ(X1X2)zw
From the formula for the partial bi-free R-transform in Theorem 2.7, we also have
R(X1,X2)(z, w) = 1 + (cid:107)v1(cid:107)2z2 + (cid:107)v2(cid:107)2w2 −
Denoting
zw
G(X1,X2)( 1
z + (cid:107)v1(cid:107)2z, 1
w + (cid:107)v2(cid:107)2w)
(9)
.
(10)
1
t1 =
z
for z, w ∈ C \ {0} close to 0, we have
t1 −(cid:112)t2
where the branches of the square roots are (cid:112)t2
z =
large. From the formulas (9) and (10), we get
1 − 4(cid:107)v1(cid:107)2
2(cid:107)v1(cid:107)2
+ (cid:107)v1(cid:107)2z
and t2 =
+ (cid:107)v2(cid:107)2w,
1
w
2 − 4(cid:107)v2(cid:107)2
2(cid:107)v2(cid:107)2
,
and w =
t2 −(cid:112)t2
1 − 4(cid:107)v1(cid:107)2 ≈ t1 and (cid:112)t2
1 − 4(cid:107)v1(cid:107)2)(t2 −(cid:112)t2
zw
1 − (cid:104)v2, v1(cid:105)zw
(t1 −(cid:112)t2
4(cid:107)v1(cid:107)2(cid:107)v2(cid:107)2 − (cid:104)v2, v1(cid:105)(t1 −(cid:112)t2
1 − 4(cid:107)v1(cid:107)2)(t2 −(cid:112)t2
2 − 4(cid:107)v2(cid:107)2)
G(X1,X2)(t1, t2) =
=
when t1 and t2 are large.
Let
2 − 4(cid:107)v2(cid:107)2 ≈ t2 for t1 and t2
2 − 4(cid:107)v2(cid:107)2)
,
(11)
t − √
t2 − 4
2
q(t) =
(t ∈ C \ [−2, 2]).
(12)
The function z (cid:55)→ z + 1
z sends the punctured unit disk {z 0 < z < 1} biholomorphically onto C\ [−2, 2].
The function q is its inverse with respect to composition. We deduce that the identity q(t) = q(¯t) holds
for all t ∈ C \ [−2, 2].
By (8) and (11), for l and s large, we have
(cid:32)
(¯s −(cid:112)¯s2 − 4(cid:107)v1(cid:107)2)(¯l −(cid:112)¯l2 − 4(cid:107)v2(cid:107)2)
4(cid:107)v1(cid:107)2(cid:107)v2(cid:107)2 − (cid:104)v2, v1(cid:105)(¯s −(cid:112)¯s2 − 4(cid:107)v1(cid:107)2)(¯l −(cid:112)¯l2 − 4(cid:107)v2(cid:107)2)
(cid:33)
det E(l, s) = 1 − 2(Im(cid:104)v2, v1(cid:105))i
2(Im(cid:104)v1, v2(cid:105))i(s −(cid:112)s2 − 4(cid:107)v1(cid:107)2)(l −(cid:112)l2 − 4(cid:107)v2(cid:107)2)
4(cid:107)v1(cid:107)2(cid:107)v2(cid:107)2 − (cid:104)v1, v2(cid:105)(s −(cid:112)s2 − 4(cid:107)v1(cid:107)2)(l −(cid:112)l2 − 4(cid:107)v2(cid:107)2)
,
= 1 +
8
and
4(cid:107)v1(cid:107)2(cid:107)v2(cid:107)2 − (cid:104)v1, v2(cid:105)(s −(cid:112)s2 − 4(cid:107)v1(cid:107)2)(l −(cid:112)l2 − 4(cid:107)v2(cid:107)2)
4(cid:107)v1(cid:107)2(cid:107)v2(cid:107)2 − (cid:104)v1, v2(cid:105)(s −(cid:112)s2 − 4(cid:107)v1(cid:107)2)(l −(cid:112)l2 − 4(cid:107)v2(cid:107)2)
DYKEMA AND NA
=
q
=
1 −
1 −
¯α(cid:107)v1(cid:107)(cid:107)v2(cid:107) q
α(cid:107)v1(cid:107)(cid:107)v2(cid:107) q
(cid:16) s(cid:107)v1(cid:107)
(cid:16) s(cid:107)v1(cid:107)
(cid:17)
(cid:16) l(cid:107)v2(cid:107)
(cid:17)
(cid:17) ,
(cid:16) l(cid:107)v2(cid:107)
(cid:17)
(cid:18) ¯s
(¯s −(cid:112)¯s2 − 4(cid:107)v1(cid:107)2) = 2(cid:107)v1(cid:107)q
(cid:18) ¯l
(¯l −(cid:113)
¯l2 − 4(cid:107)v2(cid:107)2) = 2(cid:107)v2(cid:107)q
q
(cid:107)v1(cid:107)
(cid:19)
(cid:19)
where α = (cid:104)v1, v2(cid:105), and for the second equality, we have used
= 2(cid:107)v1(cid:107)q
(cid:107)v2(cid:107)
Since v1 and v2 are linearly independent, α < (cid:107)v1(cid:107)(cid:107)v2(cid:107). Since
(cid:107)v2(cid:107)
= 2(cid:107)v2(cid:107)q
(13)
(cid:18) s
(cid:18) l
(cid:107)v1(cid:107)
(cid:19)
= s −(cid:112)s2 − 4(cid:107)v1(cid:107)2
(cid:19)
= l −(cid:112)l2 − 4(cid:107)v2(cid:107)2.
(cid:12)(cid:12)(cid:12)q
(cid:17)(cid:12)(cid:12)(cid:12) < 1 and
(cid:12)(cid:12)(cid:12)q
(cid:16) s(cid:107)v1(cid:107)
(cid:16) l(cid:107)v2(cid:107)
(cid:17)(cid:12)(cid:12)(cid:12) < 1 for
s ∈ C\ [−2(cid:107)v1(cid:107), 2(cid:107)v1(cid:107)] and l ∈ C\ [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)], the numerator and denominator in (13) do not vanish
for such s and l. So the right-hand side of (13) is a holomorphic function there. Since by definition in
(8), det E(l, s) is holomorphic on (C∞\σ(X2)) × (C∞\σ(X1)), it follows from the analytic continuation
that the formula of det E(l, s) in (13) holds for all s ∈ C \ [−2(cid:107)v1(cid:107), 2(cid:107)v1(cid:107)] and l ∈ C \ [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)].
In this subsection, we find the formula of the principal function g(δ, γ) of T by using the formula
2.4.
(13). The principal function g was defined on R2 by
(cid:19)
det E(l, s) = exp
g(δ, γ)
R
dδ
δ − l
dγ
γ − s
and supp(g) ⊆ {(δ, γ) ∈ R2 γ + iδ ∈ σ(T )}. To find the principal function of T , consider the function f
defined by
f (l, γ) =
g(δ, γ)
dδ
δ − l
.
for l ∈ C\σ(X2) and γ ∈ R. Fixing γ ∈ R, f (l, γ) is a holomorphic function for l ∈ C \ σ(X2). From (13)
and the definition of g(δ, γ), we have
(cid:90)
f (l, γ)
dγ
γ − s
R
= (2πi)Log
(cid:17)
(cid:17)
(cid:16) s(cid:107)v1(cid:107)
(cid:16) s(cid:107)v1(cid:107)
q
q
(cid:16) l(cid:107)v2(cid:107)
(cid:16) l(cid:107)v2(cid:107)
(cid:17)
(cid:17)
,
¯α(cid:107)v1(cid:107)(cid:107)v2(cid:107) q
α(cid:107)v1(cid:107)(cid:107)v2(cid:107) q
(14)
where s ∈ C∞\σ(X1) and l ∈ C∞\σ(X2). Now we will find the function f (l, γ) by using the Stieltjes
inversion formula.
We defined the function q(t) for t ∈ C \ [−2, 2] in (12).
(cid:90)
R
2πi
(cid:90)
(cid:18) 1
(cid:90)
1 −
R
1 −
Lemma 2.8. If t0 ∈ [−2, 2], then
lim
(cid:38)0
Proof. For t0 ∈ (−2, 2),
lim
(cid:38)0
q(t0 + i) = lim
(cid:38)0
q(t0 + i) =
.
0
2
t0 − i(cid:112)4 − t2
t0 + i −(cid:112)(t0 + i)2 − 4
t0 + i −(cid:112)−(4 + 2 − t2
t0 − i(cid:112)4 − t2
2
2
= lim
(cid:38)0
=
0
.
2
0) + 2it0
For, when is large and positive, the branch of a square root is such that(cid:112)−(4 + 2 − t2
BIFREE CENTRAL LIMIT DISTRIBUTIONS
(cid:112)−(4 + 2 − t2
0) + 2it0 = i(cid:112)4 − t2
So lim(cid:38)0
0.
9
0) + 2it0 ≈ t0+i.
(cid:3)
Define a function ζ(t) for t ∈ [−2, 2] by
t − i
ζ(t) =
√
4 − t2
2
.
Then ζ(t) ∈ T for t ∈ [−2, 2], where T is a unit circle in C. By Lemma 2.8, the limit of q(t + i) goes to
ζ(t) as (cid:38) 0, where t ∈ [−2, 2]. Then we have for γ ∈ [−2(cid:107)v1(cid:107), 2(cid:107)v1(cid:107)],
(15)
Fix l ∈ R \ [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)]. Since clearly f (l, γ) = 0 for γ ∈ R \ σ(X1), we suppose γ ∈ σ(X1). Using
(cid:107)v1(cid:107) + i
(cid:107)v1(cid:107)
lim
(cid:38)0
= ζ
=
2
q
(14), the Stieltjes inversion formula, and (15), we have
f (l, γ) =
1
π
lim
(cid:38)0
1 −
(cid:19)
(cid:107)v1(cid:107)
¯α(cid:107)v1(cid:107)(cid:107)v2(cid:107) q
α(cid:107)v1(cid:107)(cid:107)v2(cid:107) q
(cid:18) γ
1 −
(2πi)Log
(cid:16) l(cid:107)v2(cid:107)
(cid:17)
(cid:16) γ(cid:107)v1(cid:107)
(cid:17)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:16) γ(cid:107)v1(cid:107)
(cid:17)
(cid:16) l(cid:107)v2(cid:107)
(cid:17)
(cid:17)(cid:19)(cid:18)
(cid:16) γ(cid:107)v1(cid:107)
(cid:16) l(cid:107)v2(cid:107)
(cid:17)
(cid:17)(cid:19)(cid:18)
(cid:16) l(cid:107)v2(cid:107)
(cid:17)
(cid:16) γ(cid:107)v1(cid:107)
(cid:19)(cid:33)
(cid:18) l
(cid:19)
(cid:18) γ
(cid:19)(cid:33)
(cid:18) l
(cid:18) γ
(cid:19)
¯α(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
α(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:107)v1(cid:107)
(cid:107)v2(cid:107)
α(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
¯α(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
1 −
1 −
α
¯α
q
q
q
q
q
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:107)v1(cid:107)
q
(cid:107)v2(cid:107)
Im
1 −
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1 −
(cid:18)
(cid:18)
(cid:32)
1 −
1 −
1 −
(cid:32)
= 2 log
= log
= Log
− Log
1 −
(cid:114)
γ(cid:107)v1(cid:107) − i
4 −(cid:16) γ(cid:107)v1(cid:107)
(cid:17)2
(cid:16) γ(cid:107)v1(cid:107) + i
(cid:16) γ(cid:107)v1(cid:107) + i
(cid:107)v1(cid:107)
(cid:107)v1(cid:107)
(cid:17)
(cid:17)
q
q
(cid:17)
(cid:17)
(cid:16) l(cid:107)v2(cid:107)
(cid:16) l(cid:107)v2(cid:107)
(cid:19)
∈ T.
(cid:18) γ
α(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
¯α(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:32)
+ Log
(cid:17)
(cid:17)
(cid:16) γ(cid:107)v1(cid:107)
(cid:16) γ(cid:107)v1(cid:107)
q
q
(cid:17)(cid:19)
(cid:16) l(cid:107)v2(cid:107)
(cid:17)(cid:19)
(cid:16) l(cid:107)v2(cid:107)
(cid:19)
(cid:18) γ
(cid:18) γ
(cid:107)v1(cid:107)
¯α
α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
1 −
(cid:32)
− Log
1 −
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:107)v1(cid:107)
(cid:19)(cid:33)
(cid:19)(cid:33)
(cid:18) l
(cid:18) l
(cid:19)
(cid:107)v2(cid:107)
q
q
(cid:107)v2(cid:107)
(16)
, (17)
where Log is the principal branch of the logarithm. This equality holds where γ ∈ σ(X1) and l ∈
R \ [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)].
Fix γ ∈ σ(X1). Since each expression appearing as an argument of Log, above, remains in the
disk of radius 1 centered at 1 for l ∈ C \ [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)]. So the expression (17) is holomorphic on
C \ [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)]. The equality (16) was derived for l ∈ R \ [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)], but as defined, f (l, γ) is
holomorphic in C \ [−2(cid:107)v2(cid:107), 2(cid:107)v2(cid:107)]. By analytic continuation, we have
= Log
1 −
¯α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:107)v1(cid:107)
(cid:32)
− Log
1 −
α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:18) γ
(cid:32)
1 −
(cid:19)(cid:33)
(cid:18) l
(cid:19)
(cid:18) l
(cid:19)
(cid:18) γ
q
(cid:107)v1(cid:107)
(cid:107)v2(cid:107)
q
(cid:107)v2(cid:107)
+ Log
(cid:19)(cid:33)
α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:107)v1(cid:107)
(cid:32)
− Log
1 −
¯α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:18) γ
(cid:19)(cid:33)
(cid:18) l
(cid:19)
(cid:18) l
(cid:19)
(cid:18) γ
q
(cid:107)v1(cid:107)
(cid:107)v2(cid:107)
q
(cid:107)v2(cid:107)
(cid:19)(cid:33)
(cid:90)
g(δ, γ)
R
dδ
δ − l
(cid:32)
for γ ∈ σ(X1) and l ∈ C\σ(X2).
10
DYKEMA AND NA
Now we will apply the Stieltjes inversion formula to f (l, γ) in order to recover the principal function
for δ ∈ σ(X2) as in (15), we get
g(δ, γ) of T . Since we have lim(cid:38)0 q
g(δ, γ) =
=
1
π
1
π
lim
(cid:38)0
(cid:32)
Arg
Im f (δ + i, γ)
¯α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
1 −
(cid:32)
− 1
π
Arg
1 −
α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
= ζ
(cid:107)v2(cid:107)
(cid:16) δ(cid:107)v2(cid:107) + i
(cid:17)
(cid:19)(cid:33)
(cid:18) δ
(cid:19)
(cid:18) γ
(cid:19)(cid:33)
(cid:19)
(cid:18) γ
(cid:18) δ
(cid:17)
(cid:16) γ(cid:107)v1(cid:107)
(cid:107)v1(cid:107)
(cid:107)v2(cid:107)
(cid:107)v1(cid:107)
(cid:107)v2(cid:107)
ζ
ζ
+
(cid:17)
(cid:16) δ(cid:107)v2(cid:107)
(cid:32)
(cid:32)
(cid:17)
− 1
π
Arg
(cid:16) δ(cid:107)v2(cid:107)
Setting
α(cid:107)v1(cid:107)(cid:107)v2(cid:107) = reiφ (0 < r < 1), ζ
= eiθ1 , and ζ
= eiθ2, we get
Arg(1 − rei(−φ+θ1+θ2)) + Arg(1 − rei(φ−θ1+θ2))
1
π
Arg
1 −
α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:19)
(cid:18) γ
(cid:18) γ
(cid:107)v1(cid:107)
(cid:19)(cid:33)
(cid:18) δ
(cid:19)(cid:33)
(cid:18) δ
(cid:19)
(cid:107)v2(cid:107)
ζ
1 −
¯α
(cid:107)v1(cid:107)(cid:107)v2(cid:107) ζ
(cid:107)v1(cid:107)
ζ
(cid:107)v2(cid:107)
.
(cid:18)
1
π
(cid:18)
(cid:18)
g(δ,γ) =
=
1
π
=
1
π
arctan
arctan
(cid:19)
− Arg(1 − rei(φ+θ1+θ2)) − Arg(1 − rei(−φ−θ1+θ2))
+ arctan
(cid:19)
(cid:18) −r sin(−φ + (θ1 + θ2))
(cid:18) −r sin(φ + (θ1 + θ2))
1 − r cos(−φ + (θ1 + θ2)
− arctan
1 − r cos(φ + (θ1 + θ2))
(cid:18) r sin(φ − (θ1 + θ2))
(cid:18) r sin(φ − (θ1 − θ2))
1 − r cos(φ − (θ1 + θ2))
− arctan
1 − r cos(φ − (θ1 − θ2))
(cid:19)
(cid:19)
(cid:19)
+ arctan
(cid:18) −r sin(φ − (θ1 − θ2))
1 − r cos(φ − (θ1 − θ2))
(cid:18) −r sin(−φ − (θ1 − θ2))
(cid:19)
1 − rcos(−φ − (θ1 − θ2))
− arctan
(cid:18) r sin(φ + (θ1 + θ2))
(cid:18) r sin(φ + (θ1 − θ2))
1 − r cos(φ + (θ1 + θ2))
− arctan
1 − r cos(φ + (θ1 − θ2))
(cid:19)
(cid:19)(cid:19)
(18)
(cid:19)(cid:19)
.
3. On the essential spectrum
As an application, we determine the essential spectrum of the operator T whose principal function we
found in Section 2. We will use the following, which follows from Theorem 8.1 of [4]:
Theorem 3.1 ([4]). Suppose T is an operator on a Hilbert space H with self-commutator T ∗T − T T ∗ in
trace class. For γ + iδ not in the essential spectrum of T ,
g(δ, γ) = ind(T − (γ + iδ)),
where g(δ, γ) is the principal function for T .
Lemma 3.2. Let 0 < r < 1 and let
h(r, φ, θ1, θ2)
=
1
π
(cid:18)
arctan
(cid:18) r sin(φ − (θ1 + θ2))
(cid:19)
(cid:18) r sin(φ − (θ1 − θ2))
1 − r cos(φ − (θ1 + θ2))
− arctan
1 − r cos(φ − (θ1 − θ2))
(cid:19)
+ arctan
(cid:18) r sin(φ + (θ1 + θ2))
(cid:19)
(cid:18) r sin(φ + (θ1 − θ2))
1 − r cos(φ + (θ1 + θ2))
− arctan
1 − r cos(φ + (θ1 − θ2))
(cid:19)(cid:19)
.
(a) If φ = 0 or φ = π, then h(r, φ, θ1, θ2) = 0.
(b) If 0 < φ < π, then for all θ1, θ2 ∈ [−π, 0], we have
(cid:18) 2r sin φ
(cid:19)
1 − r2
− 2
π
arctan
≤ h(r, φ, θ1, θ2) ≤ 0,
BIFREE CENTRAL LIMIT DISTRIBUTIONS
11
2 and equality holding on the right only when
with equality holding on the left when θ1 = θ2 = − π
θ1 ∈ {−π, 0} or θ2 ∈ {−π, 0}.
(c) If π < φ < 2π, then for all θ1, θ2 ∈ [−π, 0], we have
0 ≤ h(r, φ, θ1, θ2) ≤ − 2
π
arctan
with equality holding on the right when θ1 = θ2 = − π
θ1 ∈ {−π, 0} or θ2 ∈ {−π, 0}.
(cid:18) 2r sin φ
(cid:19)
1 − r2
,
Proof. Part (a) is clear and we may assume φ ∈ (0, π) ∪ (π, 2π).
Let ν = θ1 + θ2 and µ = θ1 − θ2. Then we are interested in the function
(cid:18)
arctan
(cid:18) r sin(φ − ν)
1 − r cos(φ − ν)
(cid:19)
+ arctan
(cid:19)
(cid:18) r sin(φ + ν)
(cid:19)
(cid:18) r sin(φ − µ)
1 − r cos(φ + ν)
1 − r cos(φ − µ)
− arctan
− arctan
h(r, φ, ν, µ) =
1
π
where
−2π ≤ ν ≤ 0
(cid:18) r sin(φ + µ)
1 − r cos(φ + µ)
(cid:19)(cid:19)
,
(19)
2 and equality holding on the left only when
(20)
In particular, we always have µ ≤ π. Note that the boundaries of the region described by (19) -- (20)
correspond to θ1 ∈ {−π, 0} or θ2 ∈ {−π, 0}, where the function h vanishes.
− min(−ν, 2π + ν) ≤ µ ≤ min(−ν, 2π + ν).
An extreme point of h not on the boundary can occur only where
d
dx
We compute
We also compute
so the function
arctan
(cid:18)
d
dc
∂h
dµ
(cid:19)
∂h
dν
=
(cid:18) r sin(x)
(cid:19)
c − r
1 − r cos(x)
= 0.
r(cos(x) − r)
1 − 2r cos(x) + r2 .
=
1 − 2rc + r2
1 − r2
(1 − 2rc + r2)2 > 0,
=
c (cid:55)→ r(c − r)
1 − 2rc + r2
is strictly increasing on [−1, 1]. Therefore,
(cid:18)
(cid:18) r sin(φ − ν)
1 − r cos(φ − ν)
(cid:19)
∂h
dν
=
d
dν
arctan
(cid:18) r sin(φ + ν)
1 − r cos(φ + ν)
(cid:19)(cid:19)
+ arctan
−r(cos(φ − ν) − r)
1 − 2r cos(φ − ν) + r2 +
r(cos(φ + ν) − r)
1 − 2r cos(φ + ν) + r2
=
vanishes if and only if cos(φ − ν) = cos(φ + ν), which in turn occurs if and only if either ν ∈ πZ or
φ ∈ πZ. We assumed φ /∈ πZ. If ν ∈ {−2π, 0}, then ν is on the boundary of the interval (19), so the only
possibility that is not on the boundary of the region is ν = −π.
dµ = 0 if and only if cos(φ − µ) = cos(φ + µ). Avoiding the boundary, this leaves
only µ = 0. We conclude that the only extreme point of h not on the boundary occurs at (ν, µ) = (−π, 0),
i.e., at (θ1, θ2) = (− π
2 ,− π
Arguing as above, ∂h
(cid:18)
(cid:18) r sin φ
2 ), and the value of h there is
− 2
π
1 − r cos φ
arctan
(cid:19)
(cid:18) r sin φ
(cid:19)(cid:19)
1 + r cos φ
+ arctan
.
(21)
12
We have the identity, for α, β ∈ R,
DYKEMA AND NA
arctan(α) + arctan(β) ∈ arctan
(cid:18) α + β
1 − αβ
(cid:19)
+ πZ.
Letting
since
α =
r sin φ
1 − r cos φ
and β =
r sin φ
1 + r cos φ
,
we find that the quantity (21) equals
− 2
π
arctan
0 < αβ =
r2 sin2 φ
1 − r2 + r2 sin2 φ
(cid:19)
(cid:18) α + β
1 − αβ
= − 2
π
arctan
< 1,
(cid:18) 2r sin φ
1 − r2
(cid:19)
.
(22)
We already observed that on the boundaries of the region described by (19) -- (20), the function h
vanishes and we just showed that the only extreme value not on the boundary is (22), which is attained
when θ1 = θ2 = − π
2 . In particular, h is never vanishing on the interior of the region. This completes the
(cid:3)
proof of (b) and (c).
Theorem 3.3. Let T = l(v1) + l(v1)∗ + i(r(v2) + r(v2)∗) with v1 and v2 linearly independent and
Im(cid:104)v1, v2(cid:105) (cid:54)= 0. Then the essential spectrum σe(T ) of T is the closed rectangle
{γ + iδ ∈ C γ ≤ 2(cid:107)v1(cid:107) and δ ≤ 2(cid:107)v2(cid:107)},
(23)
which equals the spectrum σ(T ) of T .
Proof. By Lemma 2.1, we have that σ(T ) is contained in the rectangle (23). For γ ∈ σ(X1) and δ ∈ σ(X2),
we have the formula of the principal function g(δ, γ) in (18). By Lemma 3.2, −1 < g(δ, γ) ≤ 0 if
Im(cid:104)v1, v2(cid:105) > 0, and 0 ≤ g(δ, γ) < 1 if Im(cid:104)v1, v2(cid:105) < 0. The equality g(δ, γ) = 0 holds only when
γ ∈ {2(cid:107)v1(cid:107),−2(cid:107)v1(cid:107)} or δ ∈ {2(cid:107)v2(cid:107),−2(cid:107)v2(cid:107)}, i.e., when γ and δ are on the boundary of the rectangle (23).
So the function g(δ, γ) does not assume any integer value on the interior of the rectangle. But, by Theorem
3.1, if γ + iδ /∈ σe(T ), then g(δ, γ) = ind(T − (γ + iδ)). So the whole interior of the rectangle is included
in the essential spectrum of T . Since σe(T ) is closed in C and is contained in σ(T ), we have σe(T ) equals
(cid:3)
the rectangle (23).
Proposition 3.4 ([1]). Suppose that T has compact self-commutator T ∗T − T T ∗ on a Hilbert space H
and ind(T − λ) = 0 for all λ ∈ C \ σe(T ). Then T is of the form N + K where N is normal and K is
compact.
Corollary 3.5. The operator T = l(v1) + l(v1)∗ + i(r(v2) + r(v2)∗) with linearly independent v1 and v2
and Im(cid:104)v1, v2(cid:105) (cid:54)= 0 is normal plus compact.
Example 3.6. Let v2 and u be orthogonal vectors in a Hilbert space H with (cid:107)v2(cid:107) = (cid:107)u(cid:107) = 1, and
i ∈ C. Set v1 in H by v1 = αv2 + u. Suppose that T is a bounded operator on the
let α = 1√
2
full Fock space F(H) defined by T = l(v1) + l(v1)∗ + i(r(v2) + r(v2)∗). Then, [T, T ∗] = 2
2P and it
is an one-dimensional projection on F(H). So, by restricting T ∗ to its pure part alg(T, T ∗, 1)Ω, T ∗ is a
completely non-normal hyponormal operator.
δ ≤ 2 and γ ≤ 2
We can find the principal function g(δ, γ) of T by the formula (18). For each pair (δ, γ) such that
2, we have
− 1√
√
√
2
(cid:32)
(cid:18) 1
2
(cid:19)
+
i
ζ
1
2
(cid:32)
(cid:19)
(cid:18) γ√
(cid:18) 1
2
(cid:33)
(cid:18) γ√
+
ζ(δ)
(cid:19)
i
ζ
(cid:18) 1
2
(cid:32)
(cid:33)
1 −
(cid:19)
i
ζ
(cid:19)
(cid:18) γ√
(cid:18) 1
2
− 1
2
(cid:32)
− 1
π
Arg
1 −
1
2
i
+
2
ζ(δ)
(cid:19)
ζ
(cid:33)
(cid:18) γ√
2
Arg
1
π
(cid:19)
ζ(δ)
2
− 1
π
Arg
1 −
− 1
2
2
(cid:19)
(cid:33)
ζ(δ))
,
g(δ, γ) =
1
π
Arg
1 −
BIFREE CENTRAL LIMIT DISTRIBUTIONS
13
4−t2
2
√
for t ∈ [−2, 2]. Since Im(cid:104)v1, v2(cid:105) < 0, we have 0 ≤ g(δ, γ) < 1 for all (δ, γ) ∈ R2. By
where ζ(t) = t−i
Lemma 3.2, g(δ, γ) is vanishing only when (δ, γ) is on the boundary of the rectangle {(δ, γ) ∈ R2 γ ≤
√
2
√
2 and δ ≤ 2}. Therefore, σ(T ) = σe(T ) = {γ + iδ ∈ C γ ≤ 2
2 and δ ≤ 2}. See Figure 1.
Figure 1. The principal function g(δ, γ) of T where v1 = αv2 + u, α = 1√
and (cid:107)v2(cid:107) = (cid:107)u(cid:107) = 1.
2
− 1√
2
i, u ⊥ v1
References
[1] L. G. Brown, R. G. Douglas, and P. A. Fillmore, Unitary equivalence modulo the compact operators and extensions of
C∗-algebras, Proceedings of a Conference on Operator Theory (Dalhousie Univ., Halifax, N.S., 1973), Springer, Berlin,
1973, pp. 58 -- 128. Lecture Notes in Math., Vol. 345.
[2] J. B. Conway and P. McGuire, Operators with C∗-algebra generated by a unilateral shift, Trans. Amer. Math. Soc. 284
(1984), no. 1, 153 -- 161.
[3] R. W. Carey and J. D. Pincus, An invariant for certain operator algebras, Proc. Nat. Acad. Sci. U.S.A. 71 (1974),
1952 -- 1956.
[4]
, Mosaics, principal functions, and mean motion in von Neumann algebras, Acta Math. 138 (1977), no. 3-4,
153 -- 218.
, Construction of seminormal operators with prescribed mosaic, Indiana Univ. Math. J. 23 (1973/74), 1155 -- 1165.
[5]
[6] M. Martin and M. Putinar, Lectures on hyponormal operators, Operator Theory: Advances and Applications, vol. 39,
Birkhauser Verlag, Basel, 1989.
[7] D. Voiculescu, Free probability for pairs of faces I, Comm. Math. Phys. 332 (2014), no. 3, 955 -- 980.
[8]
, Free probability for pairs of faces II: 2-variables bi-free partial R-transform and systems with rank ≤ 1 com-
mutation, preprint, available at arXiv:1308.2035.
K. Dykema, Department of Mathematics, Texas A&M University, College Station, TX 77843-3368, USA
E-mail address: [email protected]
W. Na, Department of Mathematics, Texas A&M University, College Station, TX 77843-3368, USA
E-mail address: [email protected]
|
1502.01530 | 2 | 1502 | 2015-07-14T09:20:39 | Schur idempotents and hyperreflexivity | [
"math.OA",
"math.FA"
] | We show that the set of Schur idempotents with hyperreflexive range is a Boolean lattice which contains all contractions. We establish a preservation result for sums which implies that the weak* closed span of a hyperreflexive and a ternary masa-bimodule is hyperreflexive, and prove that the weak* closed span of finitely many tensor products of a hyperreflexive space and a hyperreflexive range of a Schur idempotent (respectively, a ternary masa-bimodule) is hyperreflexive. | math.OA | math |
SCHUR IDEMPOTENTS AND HYPERREFLEXIVITY
G. K. ELEFTHERAKIS, R. H. LEVENE AND I. G. TODOROV
Abstract. We show that the set of Schur idempotents with hyper-
reflexive range is a Boolean lattice which contains all contractions. We
establish a preservation result for sums which implies that the weak*
closed span of a hyperreflexive and a ternary masa-bimodule is hyper-
reflexive, and prove that the weak* closed span of finitely many tensor
products of a hyperreflexive space and a hyperreflexive range of a Schur
idempotent (respectively, a ternary masa-bimodule) is hyperreflexive.
1. Introduction
Arveson's distance formula [3] has played a fundamental role in operator
algebra theory since its discovery, inspiring a great deal of research in several
distinct settings (see [5] and [6] and the references therein). First established
for nest algebras [2], it is an estimate for the distance of an operator T
to an operator algebra A in terms of the norms of the compressions of T
to suitable "corners" arising from the invariant subspace lattice of A. A
minimax property, the distance formula is not easily verified in practice due
to, firstly, the difficulty of computing specific operator norms, and secondly,
the lack of knowledge of the invariant subspaces of a general A. It however
implies that A is a reflexive operator algebra (see [3] and [14]); the presence
of a distance formula for A is hence known as the hyperreflexivity of A.
Arveson recognised the importance of maximal abelian selfadjoint alge-
bras (masas, for short) in the study of non-selfadjoint (non-abelian) operator
algebras [1] and pioneered the use of masa-bimodules in operator algebra
theory. These are precisely the weak* closed invariant subspaces of weak*
continuous masa-bimodule maps, also known as Schur multipliers -- a class
of transformations that has played a central role in operator space theory
since Haagerup's characterisation [10].
In [8], we studied connections be-
tween Schur idempotents and reflexive masa-bimodules. In [9], this study
was extended by considering tensor products and their relation to operator
synthesis. These papers showed that Schur idempotents are very instru-
mental in questions about reflexivity and related properties, and can be
particularly useful for establishing preservation results.
The present article focuses on the role of Schur idempotents in hyperreflex-
ivity problems. After collecting necessary background and setting notation
in Section 2, in Section 3 we show that the set of all Schur idempotents
with hyperreflexive ranges is a Boolean lattice. While we are not able to
determine whether every Schur idempotent Φ has hyperreflexive range, we
show that, if Φ belongs to the Boolean lattice C generated by the set I1
of contractive Schur idempotents, then it does so. Our results can thus be
Date: 14 July 2015.
1
2
ELEFTHERAKIS, LEVENE AND TODOROV
viewed as a test for the well-known (open) problem of whether the Boolean
lattice C coincides with the set of all Schur idempotents: the existence of a
Schur idempotent with non-hyperreflexive range would imply a negative an-
swer to the latter question. As a corollary, we show that all Schur bounded
patterns [7] give rise to hyperreflexive subspaces.
In Section 4, we examine the behaviour of hyperreflexivity with respect
to linear spans. We show that the sum of a hyperreflexive masa-bimodule
and the hyperreflexive range of a Schur idempotent is hyperreflexive, and
use this to obtain a general result about linear spans (Theorem 4.5) which
implies that the weak* closed linear span of a hyperreflexive masa-bimodule
and a ternary masa-bimodule is hyperreflexive. Ternary masa-bimodules
are subspace versions of type I von Neumann algebras and, together with
the (more general) ternary rings of operators, have been extensively studied
(see, e.g., [4], [8] and [9]).
In Section 5, we obtain results, analogous to the ones from Section 4, but
for intersections as opposed to linear spans. In particular, we prove that
the intersection of an arbitrary hyperreflexive masa-bimodule and a sub-
space belonging to a general class, containing all ternary masa-bimodules,
is hyperreflexive.
In Section 6, we show that (finite, weak* closed) linear spans, each of
whose term is the tensor product of a hyperreflexive space and a ternary
masa-bimodule, is, under some natural condition, necessarily hyperreflexive
(Theorem 6.6). This is achieved by showing first that a similar result holds
in the case where the ternary masa-bimodules are replaced by hyperreflexive
ranges of Schur idempotents.
We wish to note that the results below are formulated for subspaces of
operators acting on a single Hilbert space, but they hold more generally for
subspaces of operators between different spaces; we have chosen to work on
one Hilbert space in order to avoid somewhat cumbersome formulations.
2. Preliminaries
Throughout this paper, we fix a separable Hilbert space H and let B(H)
denote the space of all bounded linear operators on H. The norm on H and
the uniform operator norm on B(H) will both be denoted by k · k. Let X
be a subspace of B(H). If T ∈ B(H), then the distance of T to X is
d(T, X ) = inf
X∈X
kT − Xk
and the Arveson distance of T to X is
α(T, X ) = sup(cid:26) inf
X∈X
kT ξ − Xξk : ξ ∈ H, kξk = 1(cid:27) .
Trivially, α(T, X ) ≤ d(T, X ), and both d and α are order-reversing in the
second variable. We say that X is reflexive [14] if, whenever T ∈ B(H) is such
that T ξ ∈ X ξ for all ξ ∈ H, then T ∈ X . Reflexive spaces are necessarily
closed in the weak operator topology, and a weak* closed subspace X is
reflexive precisely when
α(T, X ) = 0 =⇒ d(T, X ) = 0,
T ∈ B(H).
SCHUR IDEMPOTENTS AND HYPERREFLEXIVITY
3
If X satisfies the stronger condition that there exist k > 0 with
(1)
d(T, X ) ≤ k α(T, X ),
T ∈ B(H),
then X is said to be hyperreflexive; in this case, the least constant k for which
(1) holds is denoted by k(X ) and called the hyperreflexivity constant of X .
The space X is called completely hyperreflexive if X ¯⊗B(H) is hyperreflexive,
where here and in the sequel H is a separable infinite dimensional Hilbert
space and ¯⊗ denotes the spatial weak* closed tensor product. The complete
hyperreflexivity constant kc(X ) of X is by definition the hyperreflexivity con-
stant of X ¯⊗B(H). We remark in passing that whether every hyperreflexive
space is completely hyperreflexive remains an open question [6].
We fix throughout a maximal abelian selfadjoint algebra (for short, masa)
D on H. We denote by P(D) the set of all projections in D. A Schur
multiplier is a weak* continuous D-bimodule map on B(H). The set of all
Schur multipliers is a commutative algebra under pointwise addition and
composition. If Φ is a Schur multiplier, we write kΦk for the norm of Φ as
a linear map on the Banach space B(H).
A Schur idempotent is a Schur multiplier Φ that is also an idempotent.
We denote by I the collection of all Schur idempotents. It is easy to see that
I is a lattice under the operations Φ ∧ Ψ = ΦΨ and Φ ∨ Ψ = Φ + Ψ − ΦΨ,
which is moreover Boolean for the complementation Φ → Φ⊥ def= id −Φ. For
Φ, Ψ ∈ I we write Φ ≤ Ψ if ΦΨ = Φ, and we denote by Ran Φ the range of
Φ. We refer the reader to [8] and [13] for more details on Schur idempotents.
By a D-bimodule (or a masa-bimodule when D is clear from the context)
we mean a subspace X ⊆ B(H) such that DX D ⊆ X . All masa-bimodules
in the sequel are assumed to be weak* closed. If Φ ∈ I then Ran Φ is easily
seen to be a masa-bimodule.
The statements in the next remark are straightforward.
Remark 2.1. We have
α(T, X ) = sup{ hT ξ, ηi : kξk = kηk = 1, hXξ, ηi = 0, for all X ∈ X }.
Furthermore, if X is a D-bimodule then
α(T, X ) = sup(cid:8)kQT P k : P, Q ∈ P(D), QX P = {0}(cid:9).
3. The lattice of hyperreflexive ranges
In this section, we give a characterisation of the Schur idempotents with
hyperreflexive ranges and show that they form a sublattice of the lattice I
of all Schur idempotents. We start by formulating an alternative expression
of the Arveson distance which will prove useful in the sequel.
We write I1 = {Φ ∈ I : kΦk ≤ 1} for the set of contractive Schur idem-
potents. It was shown in [12] that a Schur idempotent Φ belongs to I1 if
and only if there exist families (Pi)i∈N and (Qi)i∈N of mutually orthogonal
projections in D such that
∞
(2)
Φ(T ) =
QiT Pi,
T ∈ B(H),
where the series converges in the weak* topology.
Xi=1
4
ELEFTHERAKIS, LEVENE AND TODOROV
Proposition 3.1. Let X ⊆ B(H) be a weak* closed D-bimodule. Then
α(T, X ) = sup{kΦ(T )k : Φ ∈ I1 and Φ(X ) = {0}}.
Proof. Let M be the supremum on the right hand side. By Remark 2.1,
α(T, X ) = sup{kQT P k : P, Q ∈ P(D) and QX P = {0}}.
Since any map of the form T 7→ QT P (where P, Q ∈ P(D)) is in I1, we
have α(T, X ) ≤ M . Conversely, suppose that Φ ∈ I1 and Φ(X ) = {0}.
Represent Φ as in (2); then QiX Pi = {0} for each i. On the other hand,
kΦ(T )k = supi∈N kQiT Pik ≤ α(T, X ), so M ≤ α(T, X ).
(cid:3)
If Φ ∈ I, write
The following corollary is a direct consequence of Proposition 3.1.
N1(Φ) = {Σ ∈ I1 : ΣΦ = 0}.
Corollary 3.2. If Φ ∈ I and T ∈ B(H) then
α(T, Ran Φ) = sup
kΘ(T )k.
Θ∈N1(Φ)
We next single out a simple condition that formally implies hyperreflex-
ivity. It is based on the fact that, if Φ is a Schur idempotent and T ∈ B(H),
then there is a canonical approximant of T within Ran Φ, namely the oper-
ator Φ(T ).
Definition 3.3. We write H for the set of Schur idempotents Φ ∈ I with
the following property: there exists λ > 0 such that
kΦ⊥(T )k ≤ λ α(T, Ran Φ),
T ∈ B(H).
The least constant λ with this property will be denoted by λ(Φ).
If Φ ∈ I and Ran Φ is hyperreflexive, it will be convenient to denote by
k(Φ) the hyperreflexivity constant k(Ran Φ).
Remark 3.4. Since d(T, Ran Φ) ≤ kT − Φ(T )k = kΦ⊥(T )k, we see that if
Φ ∈ H, then Ran Φ is hyperreflexive and k(Φ) ≤ λ(Φ). We will show shortly
that H is precisely the set of Schur idempotents with hyperreflexive range.
Remark 3.5. In view of Proposition 3.1, if Φ is a Schur idempotent then
Φ⊥ ∈ H precisely when there exists λ > 0 such that
kΦ(T )k ≤ λ sup{kΘ(T )k : Θ ∈ I1, Θ ≤ Φ}.
In particular, if Φ ∈ I1, then Φ⊥ ∈ H and λ(Φ⊥) = 1.
Recall that B(H) is the dual Banach space of the trace class T (H) on
H. Every element ω ∈ T (H) is thus viewed as both an operator on H and
as a (weak* continuous) linear functional on B(H); we denote by hT, ωi the
pairing between T ∈ B(H) and ω ∈ T (H). If f, g ∈ H, we denote by f ⊗ g
the rank one operator on H given by (f ⊗ g)(ξ) = (ξ, g)f , ξ ∈ H. We have
that hT, f ⊗ gi = (T f, g), for a conjugate-linear isometry g → g on H. If
i=1 ωk in the trace norm k · k1, where ωk, k ∈ N, are
ω ∈ T (H) then ω =P∞
operators of rank one such that P∞
If X ⊆ B(H), let
k=1 kωkk1 < ∞.
X⊥ = {ω ∈ T (H) : hT, ωi = 0,
for all T ∈ X }
SCHUR IDEMPOTENTS AND HYPERREFLEXIVITY
5
be the pre-annihilator of X in T (H). The following result was proved by
Arveson [3] in the case the space X is a unital algebra. The proof for the
case where X is a subspace is a straightforward modification of Arveson's
arguments; this is also a special case of [11, Theorem 2.2].
Theorem 3.6. Let X ⊆ B(H) be a reflexive space. The following are equiv-
alent:
(i) X is hyperreflexive and k(X ) ≤ r;
(ii) for every ω ∈ X⊥ and every ε > 0 there exists a sequence (ωi)i∈N ⊂ X⊥
of rank one operators such that
∞
kωik1 < (r + ε)kωk1 and ω =
Xi=1
where the latter series converges in the trace norm.
ωi,
∞
Xi=1
If Φ is a Schur idempotent, we write Φ∗ for the predual of Φ, acting on
the trace class T (H).
Lemma 3.7. If Φ ∈ I and ω ∈ T (H) then Φ⊥
∗ (ω) ∈ (Ran Φ)⊥.
Proof. If T ∈ Ran Φ then
hT, Φ⊥
∗ (ω)i = hΦ(T ), Φ⊥
∗ (ω)i = hΦ⊥Φ(T ), ωi = 0.
(cid:3)
Theorem 3.8. Let Φ be a Schur idempotent. The following are equivalent:
(i) Ran Φ is hyperreflexive;
(ii) Φ ∈ H.
Moreover, if these conditions hold then λ(Φ) ≤ k(Φ)kΦ⊥k.
Proof. (ii)⇒(i) was pointed out in Remark 3.4.
(i)⇒(ii) Let k = k(Φ) and fix T ∈ B(H). For ε > 0 there exist unit
vectors ξ, η ∈ H with
kΦ⊥(T )k − ε < (Φ⊥(T )ξ, η) = hΦ⊥(T ), ξ ⊗ ηi = hT, Φ⊥
∗ (ξ ⊗ η)i.
(3)
By Lemma 3.7, Φ⊥
∗ (ξ ⊗ η) ∈ (Ran Φ)⊥. Clearly,
kΦ⊥
∗ (ξ ⊗ η)k1 ≤ kΦ⊥
∗ k kξ ⊗ ηk1 = kΦ⊥k.
By Theorem 3.6, there exist rank one operators ωk ∈ (Ran Φ)⊥, k ∈ N, such
that
∞
ωk.
Xk=1
∞
Xk=1
By Remark 2.1 and (3),
kωkk1 ≤ (k + ε)kΦ⊥k and Φ⊥
∗ (ξ ⊗ η) =
kΦ⊥(T )k − ε <
hT, ωki ≤
kωkk1α(T, Ran Φ)
∞
∞
Xk=1
Xk=1
≤ (k + ε)kΦ⊥kα(T, Ran Φ).
Since ε is arbitrary, kΦ⊥(T )k ≤ kkΦ⊥kα(T, Ran Φ). Thus, Φ ∈ H and
λ(Φ) ≤ kkΦ⊥k.
(cid:3)
We next prove that the set H is closed under the lattice operations.
Lemma 3.9. Let Φ ∈ H and Σ ∈ I1. Then the Schur idempotent Ψ
(Φ⊥Σ)⊥ belongs to H and λ(Ψ) ≤ λ(Φ).
def
=
6
ELEFTHERAKIS, LEVENE AND TODOROV
Proof. Note that Ψ = Σ⊥ + ΦΣ. Thus, if Θ ∈ N1(Φ), then ΘΣ ∈ N1(Ψ).
Let T ∈ B(H); by Corollary 3.2,
kΨ⊥(T )k = kΦ⊥(Σ(T ))k ≤ λ(Φ) α(Σ(T ), Ran Φ)
sup
kΘΣ(T )k ≤ λ(Φ)
= λ(Φ)
sup
kΛ(T )k
Θ∈N1(Φ)
Λ∈N1(Ψ)
= λ(Φ) α(T, Ran Ψ).
(cid:3)
Theorem 3.10. The set H is a sublattice of I.
Proof. Let Φ1, Φ2 ∈ H and write λi = λ(Φi) and Xi = Ran Φi, i = 1, 2. Set
X def
= X1 ∩ X2 = Ran(Φ1Φ2). For T ∈ B(H), we have
kT − Φ1Φ2(T )k ≤ kT − Φ1(T )k + kΦ1(T ) − Φ1Φ2(T )k
≤ kT − Φ1(T )k + kΦ1kkT − Φ2(T )k
≤ λ1α(T, X1) + λ2kΦ1kα(T, X2).
By the monotonicity of α, we have α(T, Xi) ≤ α(T, X ), i = 1, 2. Thus,
kT − Φ1Φ2(T )k ≤ (λ1 + λ2kΦ1k)α(T, X ).
It follows that Φ1Φ2 ∈ H.
Now let W def= Ran(Φ1 ∨ Φ2) = X1 + X2 and T ∈ B(H). Using Lemma 3.9
and the fact that W ⊆ Ran(Σ⊥ + Φ2Σ) for Σ ∈ N1(Φ1), we have
k(Φ1 ∨ Φ2)⊥(T )k = kΦ⊥
= λ1
Σ∈N1(Φ1)
1 (Φ⊥
2 (T ))k ≤ λ1 α(Φ⊥
sup
2 Σ(T )k
kΦ⊥
2 (T ), Ran Φ1)
≤ λ1λ2
sup
Σ∈N1(Φ1)
≤ λ1λ2 α(T, W).
α(T, Ran(Σ⊥ + Φ2Σ))
It follows that Φ1 ∨ Φ2 ∈ H and λ(Φ1 ∨ Φ2) ≤ λ1λ2.
(cid:3)
Recall that a weak* closed masa-bimodule M is called ternary, if it is a
ternary ring of operators, that is, if ST ∗R ∈ M whenever S, T, R ∈ M (see
e.g. [4]).
Proposition 3.11. If Φ is a contractive Schur idempotent then Φ ∈ H and
λ(Φ) ≤ 2.
Proof. By [12], the space M = Ran Φ is a ternary masa bimodule. Consider
the von Neumann algebra
A =(cid:18) [MM∗]−w∗ M
[M∗M]−w∗ (cid:19) ⊆ B(H ⊕ H)
M∗
and note that D ⊕ D is a masa in B(H ⊕ H), over which A is a bimodule.
By [5, Lemma 8.3] there exists a contractive idempotent D ⊕ D-bimodule
map
Ψ : B(H ⊕ H) → B(H ⊕ H)
such that A = Ran Ψ and
(4)
kT − Ψ(T )k ≤ 2α(T, A)
for all T ∈ B(H ⊕ H).
SCHUR IDEMPOTENTS AND HYPERREFLEXIVITY
7
(Note that Ψ is not necessarily a Schur idempotent on B(H ⊕ H) since it
does not need to be weak*-continuous.) Consider the isometry
and observe that
θ : B(H) → B(H ⊕ H), T 7→(cid:18) 0 T
0 0 (cid:19)
Ψ(θ(T )) = θ(Φ(T )),
for all T ∈ B(H).
Moreover, for T ∈ B(H),
α(θ(T ), A) = sup
kηk=kξk=1
k(θ(T ) − A)(ξ ⊕ η)k
inf
A∈A
≤ sup
kηk=kξk=1
inf
M ∈Ran Φ
k(θ(T ) − θ(M ))(ξ ⊕ η)k
= sup
kξk=1
inf
M ∈Ran Φ
k(T − M )ξk = α(T, M).
By (4),
kΦ⊥(T )k = kT − Φ(T )k = kθ(T ) − Ψ(θ(T ))k ≤ 2α(θ(T ), A) ≤ 2α(T, Ran Φ).
(cid:3)
We write C = C(H) for the Boolean lattice generated by I1 in I.
Corollary 3.12. C ⊆ H.
Proof. Let I⊥
erated by I1∪I⊥
Proposition 3.11 and Remark 3.5, C ⊆ H.
1 = {Φ⊥ : Φ ∈ I1}. It is easy to see that the sublattice of I gen-
1 is Boolean and hence it coincides with C. By Theorem 3.10,
(cid:3)
Question 3.13. Does there exist a Schur idempotent whose range is not
hyperreflexive? In other words, is the second of the inclusions
C ⊆ H ⊆ I
strict? If so, then this would imply that C 6= I, settling in the negative
an open problem of several years' standing which asks whether the Boolean
lattice generated by the contractive Schur idempotents exhausts all Schur
idempotents.
We next show that a class of Schur idempotents, studied by Varopolous [16]
and by Davidson and Donsig [7] (see also [15]), is contained in H. Let H = ℓ2
and D be the masa of diagonal (with respect to the canonical basis) oper-
ators. A Schur bounded pattern [7] is a subset κ ⊆ N × N such that every
bounded function ϕ : N × N → C supported on κ is a Schur multiplier. If κ
is a Schur bounded pattern then the map Φκ of Schur multiplication by the
matrix (ai,j), where ai,j = 1 (resp. ai,j = 0) if (i, j) ∈ κ (resp. (i, j) 6∈ κ) is
a Schur idempotent.
Proposition 3.14. If κ ⊆ N × N is a Schur bounded pattern, then Φκ ∈ H.
Proof. By [7], there exist sets R, C ⊆ N × N and a constant N ∈ N such
that:
(1) {j ∈ N : (i, j) ∈ R} has at most N elements for each i ∈ N;
(2) {i ∈ N : (i, j) ∈ C} has at most N elements for each j ∈ N; and
(3) κ = R ∪ C.
8
ELEFTHERAKIS, LEVENE AND TODOROV
We have Φκ = ΦR ∨ ΦC. By Theorem 3.10 it suffices to show that ΦR ∈ H
and ΦC ∈ H. It is easily seen, however, that Ran ΦR is the sum of at most
N ternary masa-bimodules, so it is hyperreflexive by Corollary 3.12. Hence
ΦR ∈ H, and similarly, ΦC ∈ H.
(cid:3)
4. Hyperreflexivity and spans
In this section, we show that, under certain conditions, hyperreflexivity
is preserved under summation. The main results of the section are Theorem
4.5 and the subsequent Corollary 4.6. The first step is the following lemma.
Lemma 4.1. If U is a hyperreflexive masa-bimodule and Φ ∈ H, then the
algebraic sum U + Ran Φ is hyperreflexive and
k(U + Ran Φ) ≤ k(U )λ(Φ).
Proof. By [8, Corollary 3.4], the algebraic sum W = U + Ran Φ is weak*
closed. Given projections P, Q ∈ D, let ΣQ,P be the (contractive) Schur
idempotent given by ΣQ,P (T ) = QT P , and
Note that
ΨQ,P = Σ⊥
Q,P + ΦΣQ,P = (Φ⊥ΣQ,P )⊥.
Ψ⊥
Q,P (T ) = QΦ⊥(T )P,
T ∈ B(H).
By Lemma 3.9, ΨQ,P ∈ H and λ(ΨQ,P ) ≤ λ(Φ). Using Remark 2.1, and
writing P, Q for projections in D, we have
d(T, U + Ran Φ) = inf{kT − X − Y k : X ∈ U , Y ∈ Ran Φ}
≤ inf{kT − X − Φ(T )k : X ∈ U}
= d(Φ⊥(T ), U )
≤ k(U ) α(Φ⊥(T ), U )
= k(U ) sup{kQΦ⊥(T )P k : QU P = {0}}
= k(U ) sup{kΨ⊥
≤ k(U ) sup{λ(ΨQ,P ) α(T, Ran ΨQ,P ) : QU P = {0}}
≤ k(U ) λ(Φ) α(T, U + Ran Φ),
Q,P (T )k : QU P = {0}}
since, if QU P = {0}, then Ψ⊥
Ran ΨQ,P .
Q,P (U + Ran Φ) = {0}, and hence U + Ran Φ ⊆
(cid:3)
Corollary 4.2. If Φ ∈ I1 and U is a hyperreflexive masa-bimodule, then
the algebraic sum W = U + Ran Φ⊥ is hyperreflexive and k(W) ≤ k(U ).
Proof. Immediate from Remark 3.5 and Lemma 4.1.
(cid:3)
Lemma 4.3. Let Un, n ∈ N, be hyperreflexive spaces, such that Un+1 ⊆
Un for each n ∈ N and supn k(Un) < ∞. Then the space U def
hyperreflexive and k(U ) ≤ lim supn k(Un).
Proof. Let T ∈ B(H). Since
= Tn Un is
there exists Sn ∈ Un such that
d(T, Un) ≤ k(Un) α(T, Un),
kT − Snk < k(Un)α(T, Un) +
1
n
, n ∈ N.
SCHUR IDEMPOTENTS AND HYPERREFLEXIVITY
9
Since α(T, Un) ≤ α(T, U ), n ∈ N, and supn k(Un) < ∞, the sequence
(Sn)n∈N is bounded, and hence after passing to a subsequence we may as-
sume that (Sn) converges in the weak* topology to some operator S. Writing
k = lim supn k(Un), we have
kT − Sk ≤ lim sup
n
kT − Snk ≤ k lim sup
n
α(T, Un).
We thus conclude that
d(T, U ) ≤ kT − Sk ≤ lim sup
n
k(Un) α(T, U ).
(cid:3)
We next introduce a hyperreflexivity analogue of approximately I-injective
masa-bimodules defined in [8]. Let us say that a uniformly bounded se-
quence (Φn)n∈N ⊆ I decreases to a subspace V ⊆ B(H) if Φ1 ≥ Φ2 ≥ . . .
and V = Tn Ran Φn. Recall [8] that in this case, the masa bimodule V is
said to be approximately I-injective.
Definition 4.4. A masa-bimodule V ⊆ B(H) will be called approximately
H-injective if there is a uniformly bounded sequence (Φn)n∈N which de-
creases to V such that
Φn ∈ H for each n ∈ N and sup
n∈N
λ(Φn) < ∞.
The greatest lower bound of the possible values of the latter supremum will
be denoted by λH (V).
Theorem 4.5. If V is an approximately H-injective masa-bimodule and U
is a hyperreflexive masa-bimodule, then the algebraic sum U + V is hyper-
reflexive and
k(U + V) ≤ k(U ) λH (V).
Proof. Let (Φn)n∈N be a uniformly bounded sequence in H decreasing to V
with λ
def= supn λ(Φn) < ∞. By Lemma 4.1,
k(U + Ran Φn) ≤ k(U ) λ(Φn),
so supn k(U + Ran Φn) < ∞. By [8, Corollary 3.4], the space U + V is weak*
closed, and by the proof of [8, Theorem 2.5],
(U + Ran Φn).
k(U + Ran Φn) ≤ λk(U ).
n
By Lemma 4.3, U + V is hyperreflexive and
U + V = U +\n
Ran Φn =\n
(U + Ran Φn)(cid:17) ≤ lim sup
k(U + V) = k(cid:16)\n
Taking an infimum over all possible values of λ, we obtain k(U + V) ≤
k(U ) λH (V).
(cid:3)
Corollary 4.6. If U is a hyperreflexive masa-bimodule and M is a ternary
masa-bimodule, then U + M is hyperreflexive and k(U + M) ≤ 2k(U ).
Proof. It is well-known that every ternary masa-bimodule is the intersec-
tion of a descending sequence of ranges of contractive Schur idempotents
(see, e.g., [8]). By Proposition 3.11, M is approximately H-injective and
λH(M) ≤ 2. The statement now follows from Theorem 4.5.
(cid:3)
10
ELEFTHERAKIS, LEVENE AND TODOROV
5. Hyperreflexivity and intersections
In this section, we show that the intersection of a hyperreflexive masa-
bimodule and an approximately H-injective one is hyperreflexive. We first
establish this statement in a special case.
Lemma 5.1. If U is a hyperreflexive masa-bimodule and Φ ∈ H, then the
intersection U ∩ Ran Φ is hyperreflexive and
k(U ∩ Ran Φ) ≤ λ(Φ) + kΦkk(U ).
Proof. Let W = U ∩ Ran Φ. Since U is invariant under Φ, we have
(5)
W = {Φ(X) : X ∈ U}.
For arbitrary T ∈ B(H) we have
kT − Φ(X)k ≤ kT − Φ(T )k + kΦkkT − Xk
≤ λ(Φ)α(T, Ran Φ) + kΦkkT − Xk.
Thus,
inf
X∈U
kT − Φ(X)k ≤ λ(Φ)α(T, Ran Φ) + kΦk inf
X∈U
kT − Xk
and, by (5),
d(T, W) ≤ λ(Φ)α(T, Ran Φ) + kΦkd(T, U )
≤ λ(Φ)α(T, Ran Φ) + kΦkk(U )α(T, U ).
By the monotonicty of α, we have
d(T, W) ≤ (λ(Φ) + kΦkk(U ))α(T, W).
(cid:3)
Theorem 5.2. If V is an approximately H-injective masa-bimodule and
U is a hyperreflexive masa-bimodule, then the intersection W = U ∩ V is
hyperreflexive and
k(W) ≤ λH (V) + k(U ) + λH (V)k(U ).
Proof. Let (Φn)n∈N be a uniformly bounded sequence in H decreasing to V
with
λ = sup
n∈N
λ(Φn) < ∞.
Since kΦ⊥
n k ≤ λ(Φn) for all n, we have
(6)
kΦnk ≤ 1 + λ.
sup
n∈N
By the proof of [8, Theorem 2.5],
W = ∩∞
n=1(U ∩ Ran Φn).
By (6) and Lemma 5.1,
k(U ∩ Ran Φn) ≤ λ + (1 + λ)k(U ), n ∈ N.
Lemma 4.3 now implies that W is hyperreflexive and
k(W) ≤ λ + (1 + λ)k(U ).
The stated estimate follows after taking the infimum over all possible values
of λ.
(cid:3)
SCHUR IDEMPOTENTS AND HYPERREFLEXIVITY
11
Using Theorem 5.2 and arguing as in the proof of Corollary 4.6, we obtain
the following corollary.
Corollary 5.3. If U is a hyperreflexive masa-bimodule and M is a ternary
masa-bimodule then U ∩ M is hyperreflexive and
k(U ∩ M) ≤ 2 + 3k(U ).
Corollary 5.4. If U is a weak* closed nest algebra bimodule and M is a
ternary masa-bimodule then
k(U ∩ M) ≤ 5.
Proof. The statement is immediate from Corollary 5.3 and the fact that
k(U ) = 1 [6].
(cid:3)
6. Hyperreflexivity and tensor products
In this section, we establish a preservation result for hyperreflexivity under
the formation of tensor products. In addition to the Hilbert space H, we fix a
separable Hilbert space K and a masa in B(K). If U ⊆ B(H) and V ⊆ B(K)
are subspaces, we denote by U ⊗ V their algebraic tensor product, viewed as
a subspace of B(H ⊗ K), so that U ¯⊗V is the weak* closure of U ⊗ V.
Recall that C(H) denotes the Boolean lattice generated by the contractive
Schur idempotents acting on B(H). By (2), it is easy to see that if Φ is a
contractive Schur idempotent on B(H), then Φ ⊗ id is a contractive Schur
idempotent on B(H ⊗ K). Since tensoring with the identity map on K
commutes with the lattice operations, it follows that if Φ ∈ C(H) then
Φ ⊗ id ∈ C(H ⊗ K). By Corollary 3.12, (Ran Φ) ¯⊗B(K) = Ran(Φ ⊗ id)
is hyperreflexive, so Ran Φ is completely hyperreflexive. We let kc(Φ) =
kc(Ran Φ), and λc(Φ) = λ(Φ ⊗ id).
Theorem 6.1. Let Φi ∈ C(H), Xi = Ran Φi, and Ui ⊆ B(K) be a weak*
closed subspace, i = 1, . . . , n. Suppose that, for every non-empty subset
E = {i1, . . . , im} of the set {1, . . . , n}, the space
def= Ui1 + · · · + Uim
UE
w∗
is completely hyperreflexive. Then the space
W = X1 ⊗ U1 + · · · + Xn ⊗ Un
w∗
is completely hyperreflexive.
Proof. It will be convenient to set U∅ = {0}. We first show that W is
hyperreflexive. Let S be the set of all subsets of {1, . . . , n}. For E ∈ S, let
Φi,
ΦE = ^i∈E
where Φ∅ = id and Ψ∅ = 0. Then
and ΨE = _i∈E
Φi,
(7)
ΦEΨ⊥
E c.
id = XE∈S
Note that, for each E ∈ S, we have
(8)
W ⊆ Ran ΨE c ⊗ B(K) + B(H) ⊗ UE
w∗
.
12
ELEFTHERAKIS, LEVENE AND TODOROV
Indeed, for each i, either i ∈ E, in which case Xi ⊗ Ui ⊆ B(H) ⊗ UE, or
i ∈ Ec, in which case Xi ⊗ Ui ⊆ Ran ΨE c ⊗ B(K).
For E ∈ S, set
θE = (ΦEΨ⊥
E c)∗ = (ΦE)∗(Ψ⊥
E c)∗
and let ω ∈ W⊥. Since W is invariant under the map ΦEΨ⊥
E c, we have that
(9)
θE(ω) ∈ W⊥.
By (7), ω =PE∈S θE(ω).
We claim that
(10)
To show (10), suppose first that X ∈ Ran ΨE c and B ∈ B(K). Then
θE(ω) ∈(cid:0)(Ran ΨE c) ⊗ B(K) + B(H) ⊗ UE(cid:1)⊥.
hX ⊗ B, θE(ω)i = hX ⊗ B, (Ψ⊥
E c)∗(θE(ω))i = hΨ⊥
E c(X ⊗ B), θE(ω)i = 0,
and hence
(11)
Now let A ∈ B(H) and Y ∈ UE. Then
θE(ω) ∈(cid:0)(Ran ΨE c) ¯⊗B(K)(cid:1)⊥.
ΦE(A ⊗ Y ) ∈ (∩i∈EXi) ⊗ UE ⊆ W
and, using (9), we see that
hA ⊗ Y, θE(ω)i = hA ⊗ Y, (ΦE)∗(θE(ω))i = hΦE(A ⊗ Y ), θE(ω)i = 0.
Thus,
(12)
θE(ω) ∈ (B(H) ¯⊗UE)⊥.
Now (11) and (12) imply (10).
Let
kE
def
= kc(UE) Yi∈E c
λc(Φi)
and fix ε > 0. By Lemma 4.1, (Ran ΨE c) ¯⊗B(K) + B(H) ¯⊗UE is hyperreflex-
ive and
k((Ran ΨE c) ¯⊗B(K) + B(H) ¯⊗UE) ≤ kc(UE)λc(ΨE c) ≤ kE
since, by Lemma 4.1 and Remark 3.4,
λc(ΨE c) ≤ Yi∈E c
λc(Φi).
It thus follows from (10) and Theorem 3.6 that there exist rank one operators
such that
ωℓ
E ∈(cid:0)(Ran ΨE c) ¯⊗B(K) + B(H) ¯⊗UE(cid:1)⊥,
ℓ ∈ N,
∞
Xℓ=1
and
kωℓ
Ek1 < (kE + ε)kθE(ω)k ≤ (kE + ε)kΦEΨ⊥
E ckkωk1
θE(ω) =
ωℓ
E
∞
Xℓ=1
SCHUR IDEMPOTENTS AND HYPERREFLEXIVITY
13
in the trace norm. It follows that
∞
Xℓ=1
XE∈S
and
in the trace norm.
E ck! kωk1
kωℓ
(kE + ε)kΦEΨ⊥
Ek1 ≤ XE∈S
ω = XE∈S
ωℓ
E
∞
Xℓ=1
Note that, by (8), ωℓ
E ∈ W⊥ for each E ∈ S and each ℓ ∈ N. By
Theorem 3.6, W is hyperreflexive and
(13)
k(W) ≤ XE∈S
kc(UE) Yi∈E c
λc(Φi)kΦEΨ⊥
E ck.
To see that W is completely hyperreflexive, note that, if H is a separable
Hilbert space, then
B(H) ¯⊗W = (B(H) ¯⊗X1) ⊗ U1 + · · · + (B(H) ¯⊗Xn) ⊗ Un
w∗
.
Since B(H) ¯⊗Xi = Ran(id ⊗Φi) and λc(id ⊗Φi) = λc(Φi), i = 1, . . . , n, the
claim now follows from the previous paragraphs.
(cid:3)
Remark 6.2. It should be noted that the (complete) hyperreflexivity of the
spaces UE cannot be omitted from the assumptions of Theorem 6.1. Indeed,
it is implied by its conclusion by taking Φi = id for i ∈ E and Φi = 0 for
i 6∈ E.
Corollary 6.3. Let Φ ∈ C(H), X = Ran Φ and U ⊆ B(K) be a completely
hyperreflexive subspace. Then the space W = X ¯⊗U is hyperreflexive and
k(W) ≤ λc(Φ)kΦ⊥k + kc(U )kΦk.
Proof. The claim is immediate from estimate (13), after taking into account
that λc({0}) = kc({0}) = 1.
(cid:3)
Corollary 6.4. Let {Φ1, . . . , Φn} ⊆ C(H), and {Ψ1, . . . , Ψn} ⊆ C(K). If
Xi = Ran Φi and Yi = Ran Ψi, i = 1, . . . , n, then the space
is hyperreflexive.
X1 ¯⊗Y1 + · · · + Xn ¯⊗Yn
Proof. The statement is immediate from Theorem 6.1, Theorem 3.10 and
the fact that W is weak* closed (see [8, Corollary 3.4]).
(cid:3)
Remark 6.5. The preceding three results hold more generally (with identical
proofs) if we replace C(H) in the hypotheses with the lattice Hc of Schur
idempotents with completely hyperreflexive range:
Hc = {Φ ∈ H : λc(Φ) < ∞}.
On the other hand, C(H) seems a more natural class to work with.
Theorem 6.6. Let Mi ⊆ B(H) be a ternary masa-bimodule and Ui ⊆ B(K)
be a weak* closed subspaces, i = 1, . . . , n. Suppose that for every non-empty
subset E = {i1, . . . , im} of the set {1, . . . , n}, the subspace
UE
def
= Ui1 + · · · + Uim
w∗
14
ELEFTHERAKIS, LEVENE AND TODOROV
is completely hyperreflexive. Then the space
W = M1 ⊗ U1 + · · · + Mn ⊗ Un
w∗
is completely hyperreflexive.
Proof. As in the proof of Corollary 4.6, we may write
Mi =
∞
\j=1
Ran Φi
j ,
i = 1, . . . , n,
where each Φi
all i and j. Fix natural numbers j2, . . . , jn. Letting
j is a contractive Schur idempotent such that Φi
j+1 ≤ Φi
j for
Vj = Ran Φ1
j ⊗ U1 +
w∗
Ran Φi
ji ⊗ Ui
,
j ∈ N,
n
Xi=2
we see that Vj+1 ⊆ Vj for each j and, by Theorem 6.1, supj k(Vj) < ∞. By
[9, Corollary 4.21],
W1
def
= ∩j∈NVj = M1 ⊗ U1 +
w∗
Ran Φi
ji ⊗ Ui
.
n
Xi=2
By Lemma 4.3, the space W1 is hyperreflexive. Continuing inductively, we
see that the space
Wm
def=
m
Xi=1
Mi ⊗ Ui +
w∗
Ran Φi
ji ⊗ Ui
n
Xi=m+1
is hyperreflexive for each m = 1, . . . , n; in particular, the space W = Wn is
hyperreflexive.
Let H be a separable Hilbert space. The space W ¯⊗B(H) is unitarily
equivalent to
(M1 ¯⊗B(H)) ⊗ (U1 ¯⊗B(H)) + · · · + (Mn ¯⊗B(H)) ⊗ (Un ¯⊗B(H))
w∗
.
Since the spaces Mi ¯⊗B(H) are ternary masa bimodules, while the spaces
Ui ¯⊗B(H) are completely hyperreflexive, by the first part of the proof, the
space W ¯⊗B(H) is hyperreflexive.
(cid:3)
Corollary 6.7. If M is a von Neumann algebra of type I and A is a nest
algebra then M ¯⊗A is hyperreflexive and k(M ¯⊗A) ≤ 5.
Proof. Immediate from Theorem 6.6 or, alternatively, from Corollary 5.4.
(cid:3)
Acknowledgements
The authors are grateful to an anonymous referee for helpful comments
and suggestions.
SCHUR IDEMPOTENTS AND HYPERREFLEXIVITY
15
References
[1] W. B. Arveson, Operator algebras and invariant subspaces, Ann. Math. (2) 100
(1974), 433-532.
[2] W. B. Arveson, Interpolation problems in nest algebras, J. Funct. Anal. 20 (1975),
208-233.
[3] W. B. Arveson, Ten lectures on operator algebras, American Mathematical Society,
1984.
[4] D. P. Blecher and C. Le Merdy, Operator algebras and their modules -- an oper-
ator space approach, Oxford University Press, 2004
[5] K. R. Davidson, Nest algebras, Longman Scientific and Technical, 1988.
[6] K. R. Davidson and R. H. Levene, 1-hyperreflexivity and complete hyperreflexivity,
J. Funct. Anal. 235 (2006), no. 2, 666-701.
[7] K. R. Davidson and A. P. Donsig, Norms of Schur multipliers, Illinois J. Math.
51 (2007), no. 3, 743-766.
[8] G. K. Eleftherakis and I. G. Todorov, Ranges of bimodule projections and
reflexivity, J. Funct. Anal. 262 (2012), no. 11, 4891-4915.
[9] G. K. Eleftherakis and I. G. Todorov, Operator synthesis and tensor products,
Trans. Amer. Math. Soc., in print.
[10] U. Haagerup, Decomposition of completely bounded maps on operator algebras, un-
published manuscript.
[11] D. Hadwin, A general view of reflexivity, Trans. Amer. Math. Soc. 344 (1994), no.
1, 325 -- 360.
[12] A. Katavolos and V. Paulsen, On the ranges of bimodule projections, Canad.
Math. Bull. 48 (2005) no. 1, 97-111.
[13] R. H. Levene, Norms of idempotent Schur multipliers, New York J. Math. 20 (2014),
325-352.
[14] A. I. Loginov and V. S. Shulman, Hereditary and intermediate reflexivity of W*-
algebras, Izv. Akad. Nauk SSSR 39 (1975), 1260-1273; Math. USSR-Izv. 9 (1975),
1189-1201.
[15] G. Pisier Multipliers and lacunary sets in non-amenable groups, Amer. J. Math. 117
(1995), no. 2, 337-376.
[16] N. Th. Varopoulos, On an inequality of von Neumann and an application of the
metric theory of tensor products to operator theory, J. Funct. Anal. 16 (1974), 83-100.
Department of Mathematics, Faculty of Sciences, University of Patras,
265 00 Patras, Greece
E-mail address: [email protected]
School of Mathematical Sciences, University College Dublin, Belfield,
Dublin 4, Ireland
E-mail address: [email protected]
Pure Mathematics Research Centre, Queen's University Belfast, Belfast
BT7 1NN, United Kingdom
E-mail address: [email protected]
|
1608.02445 | 1 | 1608 | 2016-08-08T14:11:03 | Polynomial representations of $C^*$-algebras and their applications | [
"math.OA",
"math.RT"
] | This is a sequel to our paper on nonlinear completely positive maps and dilation theory for real involutive algebras, where we have reduced all representation classification problems to the passage from a $C^*$-algebra ${\mathcal A}$ to its symmetric powers $S^n({\mathcal A})$, resp., to holomorphic representations of the multiplicative $*$-semigroup $({\mathcal A},\cdot)$. Here we study the correspondence between representations of ${\mathcal A}$ and of $S^n({\mathcal A})$ in detail. As $S^n({\mathcal A})$ is the fixed point algebra for the natural action of the symmetric group $S_n$ on ${\mathcal A}^{\otimes n}$, this is done by relating representations of $S^n({\mathcal A})$ to those of the crossed product ${\mathcal A}^{\otimes n} \rtimes S_n$ in which it is a hereditary subalgebra. For $C^*$-algebras of type I, we obtain a rather complete description of the equivalence classes of the irreducible representations of $S^n({\mathcal A})$ and we relate this to the Schur--Weyl theory for $C^*$-algebras. Finally we show that if ${\mathcal A}\subseteq B({\mathcal H})$ is a factor of type II or III, then its corresponding multiplicative representation on ${\mathcal H}^{\otimes n}$ is a factor representation of the same type, unlike the classical case ${\mathcal A}=B({\mathcal H})$. | math.OA | math |
Polynomial representations of C ∗-algebras
and their applications
Daniel Beltit¸a∗ and Karl-Hermann Neeb†
Abstract
This is a sequel to our paper on nonlinear completely positive maps and dilation
theory for real involutive algebras, where we have reduced all representation classification
problems to the passage from a C ∗-algebra A to its symmetric powers S n(A), resp., to
holomorphic representations of the multiplicative ∗-semigroup (A, ·). Here we study the
correspondence between representations of A and of S n(A) in detail. As S n(A) is the
fixed point algebra for the natural action of the symmetric group Sn on A⊗n, this is done
by relating representations of S n(A) to those of the crossed product A⊗n ⋊ Sn in which
it is a hereditary subalgebra. For C ∗-algebras of type I, we obtain a rather complete
description of the equivalence classes of the irreducible representations of S n(A) and we
relate this to the Schur -- Weyl theory for C ∗-algebras. Finally we show that if A ⊆ B(H)
is a factor of type II or III, then its corresponding multiplicative representation on H⊗n
is a factor representation of the same type, unlike the classical case A = B(H).
Mathematics Subject Classification 2010: 46L06, 22E66, 46L45
Keywords and phrases: C ∗-algebra, ∗-semigroup, completely positive map
1 Introduction
A real seminormed involutive algebra (or ∗-algebra) is a pair (A, p), consisting of a real
associative algebra A endowed with an involutive antiautomorphism ∗ and a submultiplicative
seminorm p satisfying p(a∗) = p(a) for a ∈ A. Our project, started in [BN12], eventually
aims at a systematic understanding of unitary representations of unitary groups U(A) of real
seminormed involutive algebras (A, p). If A is unital, its unitary group is
U(A) := {a ∈ A : a∗a = aa∗ = 1},
and if A is non-unital, then U(A) := U(A1) ∩ (1 + A), where A1 = A ⊕ R1 is the unitization
of A. Typical examples we have in mind are algebras of the form A = C∞(X, B) for a Banach
∗-algebra B or A = C∞(X, Mn(K)), where X is a smooth manifold and K ∈ {R, C, H} (or
more general algebras of sections of ∗-algebra bundles). Focusing on unitary representations
of U(A) which are boundary values of representations π of the ball semigroups
ball(A) := {a ∈ A : p(a) < 1}
∗Institute of Mathematics "Simion Stoilow" of the Romanian Academy, P.O. Box 1-764, Bucharest, Ro-
mania. Email: [email protected], [email protected]
†Department of Mathematics, Friedrich-Alexander University, Erlangen-Nuremberg, Cauerstrasse 11,
91058 Erlangen, Germany. Email: [email protected]
1
lead us in [BN15] to a complete reduction of this problem to the case where A is a C∗-
algebra and the representation is holomorphic on ball(A). This reduction is achieved by
the fact that π factors through the universal map ηA : A → C∗(A, p), where C∗(A, p) is the
enveloping C∗-algebra of (A, p). 1
For a C∗-algebra A, we write eA for the c0-direct sum of the C∗-algebras Sn(A) ⊆ A⊗n,
where the tensor products are constructed from the maximal C∗-cross norm. We have a
holomorphic ∗-homomorphism of semigroups
Γ : ball(A) → eA, Γ(a) =
∞Xn=0
a⊗n.
One of the main results of [BN15] implies that, for every bounded holomorphic ∗-represen-
tation (π, V ) of the multiplicative ∗-semigroup ball(A), there exists a (linear) representation
Φ : eA → B(V ) with Φ ◦ Γ = π. As eA is the direct sum of the ideals Sn(A), this in turn
reduces all classification issues to representation theory of the C∗-algebras Sn(A). Via the
homogeneous multiplicative map
Γn : A → Sn(A),
a 7→ a⊗n,
(1)
the representations of Sn(A) are in one-to-one correspondence with multiplicative holomor-
phic ∗-representations of the multiplicative ∗-semigroup (A, ·) (or the unit ball ball(A))
which are homogeneous of degree n. We call them polynomial representations of A. As U(A)
is a totally real submanifold of A (resp., of 1+A in the non-unital case), these representations
are uniquely determined by their restrictions to U(A), which are norm-continuous unitary
representations. The multiplicative maps (1) lead to natural maps
Γ∗
Sn(A)b
n−−−−→ U(A)b,
Γ∗
n(π)(a) := π(a⊗n),
which embeds the C∗-algebra dual space Sn(A)b into the set U(A)b of equivalence classes
of unitary irreducible representations of U(A). This method of constructing points in U(A)b
requires a detailed understanding of the representation theory of the C∗-algebra Sn(A) in
terms of the representation theory of A, and this is the main goal of the present paper.
Representations of unitary groups are thus related to the interaction between C∗-algebras
and infinite dimensional analyticity -- a theme that was developed in the comprehensive
monograph [Up85] and has been explored until recently, for instance in the prequel [BN15]
to the present paper. We touch on that theme here again, proving, among other things, that
holomorphy, although much weaker than linearity, is strong enough to imply for instance
complete positivity of multiplicative homomorphisms of C∗-algebras (see Proposition 3.2
below). The present approach to representation theory of infinite dimensional unitary groups
is based on C∗-algebras and their multiplicative representations and is thus complementary
to other recent works on representation theory of infinite dimensional Lie groups, such as
[Wo14] or [DwOl15]. This approach allows us to shed fresh light on, and to partially extend,
some results on Schur -- Weyl duality in infinite dimensions from [BN12], [Nes13], and [EnIz15].
If A is a unital C∗-algebra, its unitary group U(A) is a Banach -- Lie group whose rep-
resentation theory is much more complex than the representation theory of A. That fact
1 As we learned from Jan Stochel, he obtained closely related results in [St92] which has some overlap
with [BN15].
2
is already apparent in the special case of matrix algebras A = Mk(C), when U(A) is the
compact Lie group Uk(C) with a rich representation theory, while the tautological represen-
tation on Ck is the only irreducible representation of Mk(C) up to unitary equivalence. This
classical situation is of course well understood using various well-known tools, including for
instance
• the Peter -- Weyl theorem on compact groups;
• the Cartan -- Weyl Theorem on the highest weights of irreducible representations;
• the Schur -- Weyl theory on tensor realizations of representations of classical groups.
These tools are quite specific to finite dimensions, inasmuch as they rely on Haar measures
or on eigenvectors of the adjoint representation.
We have shown in [BN12] that, for any unital C∗-algebra A, the Schur -- Weyl method leads
for every irreducible representation (π, H) of A to an infinite family (πλ, Sλ(H)) of irreducible
representations of U(A), where λ ∈ Part(n) is a partition of n and H⊗n decomposes under
Sλ(H)⊕dλ, where dλ is the dimension of the irreducible representation of
Sn corresponding to λ. In [BN12] it was left unclear to which extent one thus exhausts the
U(A) asLλ∈Part(n)
unitary dual U(A)b of U(A). As a by-product of our investigation, we prove in Theorem 5.8
that if A is a separable C∗-algebra, then the unitary dual U(A)b is exhausted by the repre-
sentations constructed by the Schur -- Weyl method if and only if A is the C∗-algebra K(H) of
compact operators on a separable complex Hilbert space. Among other things, this explains
why the remarkably complete results of [Ki73] on unitary irreducible representations of U(A)
for A = K(H) were never extended to unitary groups of more general C∗-algebras.
Our approach to the representation theory of Sn(A) is based on the fact that this algebra is
the fixed-point set of the permutation action of the permutation group Sn on the tensor power
A⊗n, hence we can use the old observation of [Ro79] for describing the dual space Sn(A)b
as an open subset of the dual space of the corresponding crossed product A⊗n ⋊ Sn. We
actually extend that approach to unitary equivalence classes of factor representations, because
the unitary group U(A) in general has norm-continuous unitary factor representations of type
II and III. In this connection, some specific applications are given in Theorem 6.4.
The structure of this paper is as follows. In Section 2 we discuss the structure of the
symmetric powers Sn(A) of a few specific types of C∗-algebras. In Section 3 we then take
a closer look at multiplicative homomorphisms, resp., representations of C∗-algebras. Here
our main result is Theorem 3.2, asserting that every multiplicative homomorphism of C∗-
algebras is completely positive, contractive and a sum of homogeneous homomorphisms. To
understand the passage from A to Sn(A), it is natural to study Sn(A) as the fixed point
algebra (A⊗n)Sn for the natural action of the symmetric group Sn on A⊗n. Therefore we
study in Section 4 C∗-dynamical systems (A, G, α), where G is a finite group. Here our main
point is to describe the relation between representations of the fixed point algebra AG and
the crossed product A ⋊α G, in which AG is embedded as a hereditary subalgebra. Here
the main results are the description of the factor representations of the crossed product in
terms of induced representations (Theorem 4.14), its refinement for irreducible representa-
tions (Proposition 4.18) and, eventually, the description of dAG in Theorem 4.23. All this is
applied in Section 5 to the special case (A⊗n, Sn, α). For C∗-algebras of type I, we obtain
in Theorem 5.3 a description of the equivalence classes of the irreducible representations of
3
Sn(A) in terms of A⊗n ⋊α Sn and we also relate this to the Schur -- Weyl theory for C∗-
algebras developed in [BN12]. In the final Section 6 we develop some aspects of Schur -- Weyl
theory for factor representations. This generalizes parts of recent work of Nessonov [Nes13]
and Enomoto/Izumi [EnIz15, Th. 4.1] to more general von Neumann algebras. Our main
results (Proposition 6.2 and Theorem 6.4) imply that the multiplicative n-fold tensor power
representation on H⊗n for a factor M ⊆ B(H) of type II1, II∞ or III is of the same type and
we even obtain some more detailed information on the corresponding projections.
Basic notation. We will use the following notation for any C∗-algebra A:
• U(A) := {u ∈ A u∗u = uu∗ = 1} for the unitary group of a unital C∗-algebra A and
for a non-unital C∗-algebra we put U(A) := U(A1) ∩ (1 + A), where A1 = A ⊕ C1 ⊆
M (A) is the unitization of A, and M (A) is the multiplier algebra of A;
• ball(A) := {a ∈ A kak < 1} for the open ball semigroup of A;
• ball(A) := {a ∈ A kak ≤ 1} for the closed ball semigroup of (A, p).
Both ball(A) and ball(A) are ∗-semigroups, and if A is unital, then 1 ∈ ball(A, p). If H
is a complex Hilbert space and B(H) is the C∗-algebra of all bounded linear operators on H,
then we also write C(H) := ball(B(H)) = {T ∈ B(H) kT k ≤ 1}.
We denote by ≃ the relation of unitary equivalence and by ≈ the (weaker) relation of quasi-
equivalence of group or algebra representations on Hilbert spaces [Dix64]. For any C∗-algebra
every irreducible ∗-representation π : A → B(Hπ), we denote its unitary equivalence class by
A we denote by bA its set of unitary equivalence classes of irreducible ∗-representations. For
[π] ∈ bA. The set bA is considered as a topological space endowed with the Fell topology, which
may be described by the condition that the correspondence J 7→ {[π] ∈ bA J 6⊂ ker π} is a
bijection between the closed two-sided ideals J ⊆ A and the open subsets of bA.
Contents
1 Introduction
2 Some examples of symmetric powers of C∗-algebras
3 Multiplicative homomorphisms of C∗-algebras
4 Representations of crossed products by finite groups
4.1 Commutants of induced representations . . . . . . . . . . . . . . . . . . . . .
4.2 Covariant factor representations as induced representations
. . . . . . . . . .
4.3 Classification of irreducible covariant representations . . . . . . . . . . . . . .
4.4 Representations of fixed-point subalgebras . . . . . . . . . . . . . . . . . . . .
5 Representations of symmetric tensor powers
5.1 The case of C∗-algebras of type I . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Links to Schur -- Weyl theory for irreducible representations . . . . . . . . . . .
6 Schur -- Weyl theory for factor representations
4
1
5
8
10
12
13
17
18
21
22
23
27
2 Some examples of symmetric powers of C ∗-algebras
As mentioned above, in the present paper we develop some mechanisms to get more explicit
access to the C∗-algebras Sn(A) for a given C∗-algebra A. To illustrate that idea, we will
briefly discuss here symmetric tensor powers of some basic examples of C∗-algebras.
Example 2.1. If X is a compact space and A = C(X), then, for every n ∈ N, we have
A⊗n ∼= C(X n) because Hom(A⊗n, C) ∼= Hom(A, C)n. From that we immediately infer that
Sn(A) ∼= C(X n/Sn)
where X n/Sn stands for the quotient space with respect to the natural action of the symmet-
ric group Sn by permutations of the coordinates of points in the nth Cartesian power X n.
Note that X n/Sn need not be a smooth manifold if X is.
If X = {x1, . . . , xN } is a finite set, then C(X) ∼= CN , hence the above remarks imply that
for any n ≥ 1, we have a linear isomorphism
Sn(CN ) ∼= Cr(N,n).
Here r(N, n) := X n/Sn = (cid:0)N +n−1
is the number of combinations of N ele-
ments taken by n at a time, where we allow any element to be selected repeatedly (equiva-
lently, the number of non-decreasing functions {1, . . . , n} → {1, . . . , N }).
n
(cid:1) = (N +n−1)!
n!(N −1)!
Example 2.2. Let A := C(X, B) for a compact space X and a unital C∗-algebra B. It is
well known that C(X) is nuclear and one has a canonical ∗-isomorphism A ∼= C(X) ⊗ B
([Bl06, Th. II.9.4.4], [Mu90, Thm. 6.4.17]). We want to determine the C∗-algebra Sn(A). It
is easy to see that
A⊗n ∼= (C(X) ⊗ B)⊗n ∼= C(X n, B⊗n),
where Sn acts on this algebra by
(σF )(x1, . . . , xn) := σ.F (xσ−1(1), . . . , xσ−1(n)).
Accordingly, Sn(A) can be identified with the set C(X n, B⊗n)Sn of Sn-equivariant maps
F : X n → B⊗n. In this sense it is an equivariant map algebra (cf. [NS13]).
The center of A⊗n contains C(X n), so that we obtain
Z(Sn(A)) ⊇ Sn(C(X)) ∼= C(X n)Sn ∼= C(X n/Sn)
(see [St92, Prop. 12.2] for more details). Since every irreducible representation of the C∗-
algebra Sn(A) determines a central character χ : Sn(C(X)) → C, and all these characters are
given by evaluation in an Sn-orbit x := Sn.(x1, . . . , xn) =: [x1, . . . , xn] ∈ X n/Sn, it factors
through the restriction map given by
evx : Sn(A) ∼= C(X n, B⊗n)Sn → C(x, B⊗n)Sn ,
f 7→ f x.
Let Sn,x be the stabilizer of some representative (x1, . . . , xn) ∈ x, so that x ∼= Sn/Sn,x.
Evaluating in the representative, we see that the image of evx is isomorphic to (B⊗n)Sn,x.
Assuming without loss of generality that {x1, . . . , xn} = k, we see that
Sn,x
∼= Sn1 × · · · × Snk
for
n1 + · · · + nk = n
5
is a product of symmetric groups. This further leads to
C(x, B⊗n)Sn ∼= (B⊗n)Sn,x ∼=
kOj=1
(B⊗nj )Snj ∼=
kOj=1
Snj (B).
Therefore the determination of the irreducible representations of Sn(A) is reduced to the
determination of irreducible representations of finite tensor products of the algebras Sm(B).
If B is of type I, then so is every Sm(B) (see Proposition 4.21(v) below). Further, irreducible
representations of tensor powers of type I algebras are tensor products of irreducible repre-
sentations. This reduces the problem to determine the irreducible representations of Sn(A)
to the corresponding problem for the target algebras Sn(B).
From [BN12] we know that every irreducible representation (π, H) of a C∗-algebra A and
every partition λ of n determine an irreducible representation (πλ, Sλ(H)) of Sn(A). One may
expect to obtain from every faithful irreducible representation (π, H) of A a decomposition
Sn(A) ∼= Mλ∈Part(n)
Sn
λ (A)
(2)
from the corresponding decomposition of Sn(B(H)), acting on H⊗n, where it commutes
with the action of Sn. We shall see in Example 2.5 below that this is far from being true in
general, as the example A = B(H) shows. However, it is true for the ideal K(H) of compact
operators.
Example 2.3. Let A = K(H) be the C∗-algebra of compact operators on the infinite
dimensional complex Hilbert space H. Using the fact that for any finite dimensional complex
vector spaces V1 and V2 one has a canonical isomorphism End(V1) ⊗ End(V2) ∼= End(V1 ⊗ V2)
and approximating the compact operators by finite-rank operators, we obtain, for any integer
n ≥ 1, a canonical isomorphism of C∗-algebras A⊗n ∼= K(H⊗n) (see [RW98, Ex. B.20] for
details). This isomorphism intertwines the permutation action of Sn on A⊗n and the spatial
action of Sn on K(H⊗n) defined via the unitary representation
τ : Sn → U(H⊗n),
τ (σ)(v1 ⊗ · · · ⊗ vn) = vσ−1(1) ⊗ · · · ⊗ vσ−1(n).
defined by permutation of factors in tensor monomials. Therefore, considering the fixed-point
algebras of the above two actions of Sn, we see that the isomorphism A⊗n ∼= K(H⊗n) defines
by restriction an isomorphism
Sn(A) ∼= K(H⊗n) ∩ τ (Sn)′ ∼= Mλ∈Part(n)
K(Sλ(H)),
(3)
where Sλ(H) = B(Vλ, H⊗n)Sn is the multiplicity space for the irreducible representation
(ρλ, Vλ) of Sn in H⊗n. In fact, from the Sn-decomposition
H⊗n ∼= Mλ∈Part(n)
Sλ(H) ⊗ Vλ
it follows that a compact operator on H⊗n commutes with Sn if and only if it is a direct sum
of operators of the form Aλ ⊗ 1, where Aλ ∈ B(Sλ(H)). Note that (3) is the decomposition
of Sn(A) according to the general structure theory of C∗-algebras of compact operators as
developed for instance in [Ar76, Thm. 1.4.5].
6
Example 2.4. If A and B are any C∗-algebras and ϕ : A → B is a ∗-homomorphism, then
for any integer n ≥ 1 one has natural ∗-homomorphisms
ϕ⊗n : A⊗n → B⊗n
and
Sn(ϕ) : Sn(A) → Sn(B),
and •⊗n and Sn(•) are functors from the category of C∗-algebras into itself. The behavior
of these functors with respect to inductive limits of nuclear C∗-algebras sheds some light on
symmetric tensor powers of AF-algebras, that is, C∗-algebras that are inductive limits of
finite dimensional C∗-algebras.
In fact, let A be any nuclear C∗-algebra with a sequence of nuclear closed ∗-subalgebras
A1 ⊆ A2 ⊆ · · · ⊆ A whose union is dense in A. It is easily seen that one has a sequence
of closed ∗-subalgebras A⊗n
2 ⊆ · · · ⊆ A⊗n whose union is dense in A⊗n, and then a
sequence of closed ∗-subalgebras Sn(A1) ⊆ Sn(A2) ⊆ · · · ⊆ Sn(A) whose union is dense in
Sn(A). Thus each of the functors •⊗n and Sn(•) commutes with inductive limits of nuclear
C∗-algebras. Since these functors preserve the category of finite dimensional C∗-algebras, it
then follows that they also preserve the category of AF-algebras.
1 ⊆ A⊗n
We point out however that, as a by-product of Example 2.3, if A is a UHF-algebra
(that is, a unital C∗-algebra that is the inductive limit of a sequence of unital ∗-subalgebras
isomorphic to full matrix algebras) and n ≥ 1, then Sn(A) is not a UHF-algebra. More
precisely, Example 2.3 shows that if A = Mk(C) for some k ≥ 2, then Sn(A) is the direct
sum of at least two full matrix algebras if n ≥ 2.
Now let A ∼= ⊗∞
k=1B(Ek) with Ek ∼= Cnk be an UHF algebra. Since A is a direct limit
of finite dimensional simple matrix algebras AN ∼= B(FN ) with FN ∼= E1 ⊗ · · · ⊗ EN , the
C∗-algebra Sn(A) is the direct limit of the C∗-algebras Sn(AN ) ∼= Lλ∈Part(n) Sn(AN )λ
with Sn(AN )λ ∼= B(Sλ(FN )) (Example 2.3), where the projection onto the λ-summand
corresponds to restriction to the corresponding Sn-isotypic subspace in F ⊗n
N . From the Sn-
equivariance of the inclusions F ⊗n
N +1, it follows that the inclusion AN ֒→ AN +1
induces inclusions Sn(AN )λ ֒→ Sn(AN +1)λ, so that we obtain a direct sum decomposition
N ֒→ F ⊗n
Sn(A) ∼= Mλ∈Part(n)
Sn(A)λ with
Sn(A)λ ∼= lim
−→
Sn(AN )λ,
and the construction implies in particular that the algebras Sn(A)λ are simple and act
irreducibly on Sλ(F ), where F is the Hilbert direct limit of the spaces FN . Therefore we find
a decomposition similar to the one in Example 2.3.
Example 2.5. Let H be an infinite dimensional Hilbert space.
(a) The representation of S2(B(H)) on H⊗2 is not injective. First we recall the non-
uniqueness of C∗-norms on B(H) ⊗ B(H) from [JP95] which implies that the spatial tensor
product (corresponding to the least C∗-norm by [Mu90, Thm. 6.4.18]), which is the image
B(H) ¯⊗2 of B(H)⊗2 under the natural representation on H⊗2 differs from the tensor product
B(H)⊗2 with respect to the maximal C∗-norm. This implies that the kernel J ⊆ B(H)⊗2 of
the representation on H⊗2 is non-zero. Clearly, J is invariant under the flip automorphism
τ (a1⊗a2) = a2⊗a1. If 0 ≤ a ∈ J is non-zero, then the same holds for a+τ (a) ∈ J ∩S2(B(H)).
Therefore the representation of S2(B(H)) on H⊗2 is also not injective.
(b) The representation of S2(B(H)) on H⊗2 is not surjective onto B(H⊗2)S2. Let (ej)j∈J
be an orthonormal basis of H and P ∈ B(H⊗2) be the orthogonal projection onto the subspace
7
spanned by (ej ⊗ ej)j∈J . Then P commutes with the flip action of S2 and it remains to
show that it is not contained in the spatial tensor product B(H)⊗2. To this end, we write
H⊗2 ∼= ⊕j∈J (Hj ⊗ ej) as a direct sum and, accordingly, P as a matrix (Pij )i,j∈J . Then
Pij (v) = δijhv, ejiej is either 0 or the orthogonal projection onto C(ej ⊗ej). In particular, the
subset {Pij : i, j ∈ J} ⊆ B(H) is closed, discrete and infinite, hence not relatively compact.
According to Exercise 11.5.7 in [KR92], this implies that P is not contained in B(H) ¯⊗2.
3 Multiplicative homomorphisms of C ∗-algebras
In this section we take a closer look at multiplicative homomorphisms, resp., representations
of C∗-algebras. Here our main result is Theorem 3.2, asserting that every multiplicative
homomorphism of C∗-algebras is completely positive, contractive and a sum of homogeneous
homomorphisms. We shall use some results and notions from our earlier paper [BN15] on
completely positive nonlinear maps and their dilation theory.
Definition 3.1. If A and B are any unital C∗-algebras, then a map ϕ : A → B is a multi-
plicative homomorphism if it is holomorphic and satisfies
ϕ(1) = 1,
ϕ(ab) = ϕ(a)ϕ(b)
and
ϕ(a∗) = ϕ(a)∗
for
a, b ∈ A.
A linear map ψ : A → B is called contractive if it satisfies ψ(ball(A)) ⊆ ball(B).
Theorem 3.2. Every multiplicative homomorphism of unital C∗-algebras ϕ : A → B is com-
ϕn, where
pletely positive, contractive and has a weakly convergent series expansion ϕ = Pn≥0
ϕn is multiplicative and homogeneous of degree n.
The proof of the above result is based on the following observations. For the case of a
positively homogeneous map ϕ, the complete positivity of ϕ can also be derived from [St92,
Prop. 11.1].
Lemma 3.3. Every dilatable bounded completely positive function ϕ : ball(A) → B(V )
extends to a weakly continuous completely positive function ϕ : ball(A) → B(V ).
Proof. First recall from [BN15] that S◦ := ball(A) is a semigroup ideal in S := ball(A). Let
(π, H, ι) = (πϕ, Hϕ, ιϕ) be the minimal dilation of ϕ given by [BN15, Prop. 3.4]. Then we use
by [BN15, Th. 6.4]. Note that the representation π is completely positive by [BN15, Prop.
[BN15, Prop. 3.21] to extend the representation π to a weakly continuous representationbπ of
S, where the weak continuity follows by the explicit formula ofbπ, since ϕ is norm-continuous
4.1], hence, by continuity, also bπ is completely positive. Then ϕ(s) := ι∗bπ(s)ι is a completely
positive weakly continuous extension of ϕ to S.
Proposition 3.4. Let A be any unital C∗-algebra, and π : ball(A) → B(Hπ) be any holo-
morphic ∗-representation. Then π is completely positive, and extends to a weakly continuous
∗-representation π : ball(A) → C(Hπ). If π is nondegenerate, then lim
π(r1) = 1 in the
r→1−
weak operator topology.
8
Proof. The map π is positive definite because it is a homomorphism of ∗-semigroups. There-
fore, after we will have proved that π is contractive, hence in particular bounded, it will
follow by [BN15, Th. 1.1] that it is completely positive. Then Lemma 3.3 ensures that π
extends to a weakly continuous ∗-representation of ball(A).
The proof of the fact that π is contractive has two stages.
Step 1: We first consider the case A = C, hence ball(A) = D is the unit disk. Then
there exists the norm-convergent power series expansion π(z) = Pn≥0 znPn ∈ B(Hπ) for
all z ∈ D, with Pn ∈ B(Hπ) for all n ≥ 0. For all z, w ∈ D one has π(¯z) = π(z)∗ and
π(zw) = π(z)π(w), and this implies that the coefficients of the above power series satisfy the
conditions Pn = P ∗
n and PnPm = 0 for all n, m ≥ 0 with n 6= m. That is, (Pn)n≥0 is a
sequence of mutually orthogonal projections, and then
n = P 2
(∀z ∈ D)
kπ(z)k ≤ sup
n≥0
zn = 1
which in particular shows that π is bounded. Moreover, one has
the weak operator topology in B(Hπ).
lim
0<r→1−
π(r1) = Pn≥0
Pn in
Step 2: In the general case, for all z ∈ D and a ∈ ball(A) one has π(za) = π(z1)π(a),
hence if π 6≡ 0, then necessarily πD 6≡ 0, which we may assume until the end of this proof.
Then use the above stage to find the sequence of mutually orthogonal projections (Pn)n≥0
in B(Hπ) with π(z1) = Pn≥0
znPn ∈ B(Hπ) for all z ∈ D.
For each a ∈ ball(A) and all z ∈ D one has π(z1)π(a) = π(a)π(z1), and this implies
Pnπ(a) = π(a)Pn, hence πn(·) := Pnπ(·) is a holomorphic ∗-representation of ball(A).
Moreover πn(za) = znπn(a) for all a ∈ ball(A) and z ∈ D, hence one may use [BN15,
Lemma 2.11] to extend πn to a holomorphic ∗-representation πn : (A, ·) → B(Hπ) which
is homogeneous of degree n. Since πn is continuous at 0 ∈ A, there exist r, M > 0 with
kπn(a)k ≤ M if a ∈ A with kak ≤ r. Since πn is homogeneous, it then follows that it is
bounded on ball(A), and then [BN15, Rem. 3.5] implies kπn(a)k ≤ 1 for all a ∈ ball(A).
Pn, we then obtain the weakly convergent series
Denoting P := Pn≥0
Xn≥0
πn(a) = P π(a) = π(a)
and
kπ(a)k ≤ sup
n≥0
kπn(a)k ≤ 1
for every a ∈ ball(A). The condition that π be nondegenerate means P = 1, and the
assertion follows.
Finally, we can now prove Theorem 3.2.
Proof. Using a realization of B as an operator algebra, we may actually assume B = B(H) for
some complex Hilbert space H. Then Proposition 3.4 and its proof show that the restricted
map π := ϕball(A) : ball(A) → B(H) is completely positive and there exists a sequence of
mutually orthogonal projections (Pn)n≥0 such that πn := π(·)Pn = Pnπ(·) is homogeneous
of degree n and completely positive on ball(A), hence in fact πn : A → B(H) is completely
positive.
9
On the other hand, since (Pn)n≥0 are mutually orthogonal projections, we have ϕ =
πn, this series being convergent in the pointwise weak operator topology. Therefore also
Pn≥0
ϕ is completely positive.
Remark 3.5. In connection with Theorem 3.2 we point out that for some simple holomorphic
maps on a C∗-algebra A the complete positivity property is a condition that has strong
structural implications on A. For instance, if A is unital and we define ϕ : A → A, ϕ(a) := a2,
then ϕ is a holomorphic positive contractive map, and the following conditions are equivalent:
(i) The map ϕ is completely positive.
(ii) The map ϕ is a multiplicative homomorphism.
(iii) The C∗-algebra A is commutative.
In fact, one has (i)⇔(iii) for instance by [JiT03, Th. 3], and it is clear that (iii)⇒(ii).
Conversely, if we assume that ϕ is a multiplicative homomorphism, then for arbitrary a, b ∈ A
and t, s ∈ R one has (etaesb)2 = ϕ(etaesb) = ϕ(eta)ϕ(esb) = (eta)2(esb)2, which is equivalent
to esbeta = etaesb. Differentiating this equality with respect to s and t at s = t = 0, we
obtain ba = ab, hence the C∗-algebra A is commutative.
4 Representations of crossed products by finite groups
In this section, unless otherwise mentioned, (A, G, α) is a C∗-dynamical system, where G is
a finite group that acts by ∗-automorphisms of a C∗-algebra A by (g, x) 7→ αg(x). We inves-
tigate here how the covariant representations of these data are related to the representations
of the fixed-point subalgebra
AG := {x ∈ A (∀g ∈ G) αg(x) = x},
which by [Ro79] is ∗-isomorphic to a hereditary subalgebra of the crossed product A ⋊α G.
These results will be later used for maximal C∗-tensor powers A⊗m acted on by the permu-
tation group G = Sm, with the aim of describing irreducible or factor representations of the
symmetric C∗-tensor powers (A⊗m)G = Sm(A), for any C∗-algebra A.
The crossed product corresponding to the above C∗-dynamical system is the C∗-algebra
A ⋊α G := ℓ1(G, A; α) := {f f : G → A} with the multiplication
(f1f2)(g) :=
1
G Xh∈G
f1(h)αh(f2(h−1g))
and the involution f ∗(g) := αg(f (g−1))∗ for all f, f1, f2 ∈ ℓ1(G, A; α) and g ∈ G. A covariant
representation is a pair (π, U ), where π : A → B(H) is a ∗-representation and U : G → B(H)
is a unitary representation satisfying π(αg(x)) = Ugπ(x)U ∗
g for all x ∈ A and g ∈ G. We also
define the ∗-representation
π⋊U : A ⋊α G → B(H),
(π⋊U )(f ) = Xg∈G
π(f (g))Ug.
10
We say that the covariant representation (π, U ) is a factor representation if the von Neumann
algebra generated by (π⋊U )(A ⋊α G), that is (π(A) ∪ π(G))′′, is a factor etc.
The following definition is a specialization of [Tak67, Def. 3.2] to finite groups, except
that we use here the left regular representation of groups.
Definition 4.1. (induced representations) Let (A, G, α) be as above. Assume that G0 is
a subgroup of the finite group G, denote by (A, G0, α0) the corresponding C∗-dynamical
system with α0 := αG0 .
Then for any covariant representation (π0, V ) of (A, G0, α0) on some Hilbert space H0,
its corresponding induced representation is the covariant representation (π, U ) of (A, G, α)
on the Hilbert space H defined as follows:
• H := {f : G → H0 f (gh) = V ∗
h f (g) if g ∈ G, h ∈ G0};
• U : G → B(H), (Ugf )(eg) := f (g−1eg);
• π : A → B(H), (π(x)f )(g) := π0(α−1
g (x))f (g) (multiplication operators).
It is useful to reformulate the above definition as follows, in the spirit of [Se77, Sect. 3.3].
Remark 4.2. Let U : G → B(H) be a unitary representation of a finite group. Assume that
a subgroup G0 ⊆ G and a closed linear subspace H0 ⊆ H satisfy UG0H0 ⊆ H0, and define
V : G0 → B(H0), Vh := UhH0.
The group representation U is unitarily equivalent to the representation induced from
V if for some (and hence for every) system 1 = g0, g1, . . . , gr of representatives of the left
G0-cosets in G we have an orthogonal direct sum decomposition
H =
rMj=0
Ugj H0.
(4)
If, moreover, (A, G, α) is a C∗-dynamical system and one has a covariant representation
(π0, V ) of (A, G0, α0) on H0, where α0 := αG0 and V : G0 → B(H0) is as above, then
the corresponding induced covariant representation is unitarily equivalent to the induced
covariant representation (π, U ) of (A, G, α) on H with π : A → B(H) given by
π(a)Ugj H0 := Ugj π0(a)U −1
gj Ugj H0
for
j = 0, 1, . . . , r, x ∈ A.
Remark 4.3. For later use, we note that in Definition 4.1 there exists a natural unitary
operator between the spaces of fixed points
HG0
0 → HG,
v 7→ fv
where fv : G → H0, fv(g) := G/G0−1/2v for all g ∈ G and v ∈ HG0
0 .
Remark 4.4. In the setting of Remark 4.2, let p0 : H → H denote the orthogonal projection
onto H0. Then
G0 = {g ∈ G Ugp0U −1
g = p0} ⊆ {g ∈ G π0 ◦ αg ≃ π0} ⊆ {g ∈ G π0 ◦ αg ≈ π0}
recalling that ≈ stands for quasi-equivalence of ∗-representations.
11
4.1 Commutants of induced representations
The situation when both inclusions in Remark 4.4 are equalities should be thought of as a
maximality property of the subgroup G0 in the same way as the polarizations at points of
duals of finite dimensional Lie algebras are maximal isotropic subalgebras, which leads to
irreducibility of the corresponding induced representations. That idea will be made precise
in Corollary 4.6 below. To that end we need the following variant of [Tak67, Th. 4.3] for
finite groups acting on C∗-algebras that need not be separable.
Proposition 4.5. Let (π, U ) be the covariant representation of (A, G, α) on H induced by
the covariant representation (π0, V ) of (A, G0, αG0 ) on H0. Pick any complete system 1 =
g0, g1, . . . , gr of representatives of the left cosets of G0 in G and for j = 0, . . . , r let pj ∈ B(H)
be the orthogonal projection on Hj := Ugj H0 and Gj := gjG0g−1
.
j
Then Gj = {g ∈ G UgpjU −1
g = pj}, and if we define (πj, Vj ) as the covariant represen-
tation of (A, Gj , αGj ) on Hj, given by πj (·) := π(·)Hj and Vj(·) := U (·)Hj , then (π, U ) is
induced by (πj , Vj) and the map
Φj : (π⋊U )(A ⋊ G)′ ∩ {p0, p1, . . . , pr}′ → (πj ⋊Vj)(A ⋊ Gj)′,
T 7→ T Hj
is well defined and is a ∗-isomorphism of von Neumann algebras.
Proof. The action of G permutes the subspaces H0, . . . , Hr, and this implies that (π, U ) is
induced by (πj , Vj) for j = 0, . . . , r.
For every T ∈ (π⋊U )(A ⋊ G)′ ∩ {p0, p1, . . . , pr}′ we have T pj = pjT .
In particular,
such an operator T is given by a block diagonal matrix with respect to the decomposition
p0 + · · · + pr = 1, and this also shows that for any j ∈ {0, . . . , r} the map Φj is a ∗-
homomorphism.
To check that Φj is injective, recall that the group G acts transitively on {H0, . . . , Hr}.
G which con-
Therefore the subspace Hj of H is G-cyclic, hence separates the commutant U ′
tains (π ⋊ U )(A ⋊ G)′.
To check that the image of Φj is (πj ⋊Vj)(A ⋊ Gj)′, let Tj ∈ (πj ⋊Vj)(A ⋊ Gj)′. For any
g ∈ G with UgHj = Hk, we then obtain an operator
Tk := UgTjU ∗
g : Hk → Hk
which does not depend on the choice of g in the Gj-coset {x ∈ G : UxHj = Hk}. We thus
obtain a block-diagonal operator T ∈ B(H) with T Hk := Tk for k = 0, . . . , r. It is straight-
forward to verify that T commutes with (π ⋊ U )(A ⋊ G). Clearly also T ∈ {p0, p1, . . . , pr}′
and Φj(T ) = Tj.
Corollary 4.6. Suppose that, in the setting of Proposition 4.5, the following additional
conditions are satisfied:
• G0 = {g ∈ G π0 ◦ αg ≈ π0};
• π0 : A → B(H0) is a factor representation.
Then the map
Φj : (π ⋊ U )(A ⋊ G)′ → (πj ⋊ U ′
j)(A ⋊ Gj )′,
T 7→ pjT Hj
is a well-defined ∗-isomorphism of von Neumann algebras for j = 0, . . . , r.
12
Proof. Using Proposition 4.5, it suffices to show that (π⋊U )(A ⋊ G)′ ⊆ {p0, p1, . . . , pr}′. One
has (π⋊U )(A ⋊ G)′ = (π(A) ∪ π(G))′ = π(A)′ ∩ π(G)′ ⊆ π(A)′ and we will check that for
arbitrary T ∈ π(A)′ one has T pj = pjT for j = 0, . . . , r.
In fact, writing the operators on H as block matrices with respect to the decomposition
p0 + · · · + pr = 1, one has T = (Tjk)0≤j,k≤r and π(x) = diag(π0(x), π1(x), . . . , πr(x)) is a
block diagonal matrix for all x ∈ A. Hence the condition T ∈ π(A)′ is equivalent to the fact
that for all j, k = 0, . . . , r one has
(∀x ∈ A) πj (x)Tjk = Tjkπk(x).
On the other hand the hypotheses imply that if j 6= k then πj A and πkA are factor repre-
sentations that are not quasi-equivalent, hence it follows by [Dix64, Cor. 5.3.6] that there
exists no non-zero intertwiner between them, and then Tjk = 0. Therefore T is given by a
block-diagonal matrix, that is, T pj = pjT for j = 0, . . . , r, and this completes the proof.
Remark 4.7. In the setting of Corollary 4.6, it follows that the induced covariant represen-
tation (π, U ) is irreducible or is a factor representation of some type I, II, or III, if and only
if so is the input representation (π0, V ).
Remark 4.8. In the proof of Proposition 4.5, the idea that T = Φ0(T ) + · · · + Φr(T ) is a
sum of mutually unitarily equivalent operators is related to [Co75, Lemma 3].
On the other hand, since the action of the group G on the set {p0, . . . , pr} is transitive, and
the isotropy groups G0, . . . , Gr are mapped to each other by suitable inner automorphisms
of G, this agrees with the uniqueness assertion on G0 in [Tak67, Th. 5.2].
4.2 Covariant factor representations as induced representations
We now turn to a question that is somehow converse to what we did so far, namely the
question of when a given covariant representation is an induced representation (cf. Propo-
sition 4.12.) We will need the following observation which is essentially a byproduct of the
proof of [AL10, Th. 3.1]. We recall that the action of G on A is called ergodic if AG = C1.
Lemma 4.9. If the action of the finite group G on A is ergodic, then dim A ≤ G < ∞.
Proof. Define
α : ℓ1(G) → B(A), α(ϕ) =
1
G Xg∈G
ϕ(g)αg.
Let π1, . . . , πn be a system of representatives of all equivalence classes of irreducible represen-
tations of G. For j = 1, . . . , n denote by dj the dimension of πj, by χj : G → C the character
of πj, and define Aj := α(χj)A ⊆ A, so that A = A1 ⊕ · · · ⊕ An. On the other hand, since G
is in particular a compact group and its action on A is ergodic by hypothesis, it follows by
[HLS81, Prop. 2.1] that dim Aj ≤ d2
j for j = 1, . . . , n. Therefore dim A ≤
d2
j = G, by
nPj=1
Burnside's Theorem, and this concludes the proof.
Lemma 4.10. Let (π, U ) be a covariant representation of (A, G, α) on H and consider the
action of G on B(H) by (g, T ) 7→ UgT U −1
g .
13
(i) If (π, U ) is a factor representation, then G acts ergodically on the abelian von Neumann
algebra Z := π(A)′′ ∩ π(A)′ and dim Z ≤ G < ∞.
(ii) If (π, U ) is irreducible, then G acts ergodically on π(A)′ and dim π(A)′ ≤ G < ∞.
Proof. If (π, U ) is a factor representation, then one has
Z G = π(A)′′ ∩ π(A)′ ∩ U ′
G = π(A)′′ ∩ (π(A) ∪ UG)′
⊆ (π(A) ∪ UG)′′ ∩ (π(A) ∪ UG)′ = C1
where the latter equality follows by the hypothesis that (π, U ) is a factor representation.
Thus Z G = C1, and then dim Z ≤ G by Lemma 4.9.
If the representation π is irreducible, then π(A)′ ∩ U ′
G = C1, hence the action of G on
π(A)′ is ergodic, and then dim π(A)′ ≤ G by Lemma 4.9 again.
We now give a basic result on ∗-representations with finite dimensional commutant, for
later use along with Lemma 4.10.
Lemma 4.11. Let S be any ∗-semigroup with a ∗-representation π : S → B(H). If the center
of π(S)′ is finite dimensional, then there exist a finite orthogonal direct sum decomposition
H =
Hj with the following properties:
nLj=1
(i) The orthogonal projections of H onto H1, . . . , Hn, respectively, are the minimal central
projections in π(S)′.
(ii) The representations πj := π(·)Hj : S → B(Hj ) for j = 1, . . . , n, are pairwise non-quasi-
equivalent factor representations.
(iii) If π(S)′ is also finite dimensional, then, for j = 1, . . . , n, there exist mutually inequiv-
alent irreducible ∗-representations (ρj, Vj) of S and kj ∈ N0 such that πj ∼= ρ⊕kj
j
.
Proof. Let M := π(S)′ and p1, . . . , pn ∈ M be the minimal central projections. We show
that the assertions hold with Hj := pj(H) for j = 1, . . . , n.
For Assertion (ii), let i, j ∈ {1, 2, . . . , n} with i 6= j. Since πi and πj are factor represen-
tations, it follows just as in [Dix64, Prop. 5.2.9] that it suffices to prove that πi and πj re
disjoint ∗-representations, that is, their only intertwiner is zero. Since the set of intertwiners
corresponds to piMpj, this follows from pipj = 0.
For Assertion (iii), if M is finite dimensional von Neumann algebra, then the ideals pjM
are isomorphic to matrix algebras Mkj (C). Accordingly, the factor representation πj is type I
and a kj -fold multiple of an irreducible representation (ρj, Vj).
The following result is essentially a special case of [Tak67, Th. 5.1 -- 5.2]. We note that
the finite dimensionality of Z follows from Lemma 4.10(i).
Lemma 4.12. Let (π, U ) be a covariant factor representation of (A, G, α) on H and Z :=
π(A)′′ ∩ π(A)′. Then the following assertions hold:
14
(i) Let p ∈ Z be a minimal projection and K := pH, H := {g ∈ G UgpU −1
g = p}, and let
(ρ, V ) be the covariant representation of (A, H, αH ) on K given by ρ(a) := π(a)K and
Vh := UhK. Then H = {g ∈ G ρ ◦ αg ≈ ρ}, ρ is a factor ∗-representation and (π, U )
is unitarily equivalent to the covariant representation induced by (ρ, V ).
(ii) The map Φ : (π⋊U )(A ⋊ G)′ → (ρ⋊V )(A ⋊ H)′, T 7→ pT K, is well defined and is a
∗-isomorphism of von Neumann algebras.
Proof. For (i), we note that Lemma 4.11(ii) implies that ρ is a factor representation since Z
is the center of π(A)′. Moreover, for every g ∈ G, one has
(π ◦ αg)(·) = Ugπ(·)U −1
g
,
so the operators Ug act by conjugation on the von Neumann algebra π(A)′′, hence also on
its center Z. The action of G on Z is ergodic by Lemma 4.10(i), and this implies that the
action of G on the set {p1, . . . , pn} of minimal idempotents in Z is transitive (cf. [Ka13b,
Lemma 3.4]).
Accordingly, G permutes the subspaces Hj := pjH transitively, and this implies by Re-
mark 4.2 that (π, U ) is unitarily equivalent to the covariant representation induced by (ρ, V ).
Since the representations of A on the subspaces Hj are pairwise non-quasi-equivalent by
Lemma 4.11(ii) and the group G acts transitively on {H1, . . . , Hn}, we obtain
H = {g ∈ G ρ ◦ αg ≈ ρ}.
For (ii), note that (π⋊U )(A⋊G)′′ ⊇ π(A)′′ ⊇ Z, hence using the von Neumann Bicommutant
Theorem, (π⋊U )(A ⋊ G)′ = (π⋊U )(A ⋊ G)′′′ ⊆ Z ′ = {p1, . . . , pn}′, and now the conclusion
follows by Corollary 4.6.
Definition 4.13. For any C∗-algebra Y , we denote by Fact(Y ) the class of all unitary
equivalence classes [π] of its factor representations π.
For the C∗-dynamical system (A, G, α), one has a natural action
G × Fact(A) → Fact(A),
(g, [π]) 7→ [π ◦ αg−1 ].
We define the class C(A, G, α) of unitary equivalence classes [(π0, V )] of covariant represen-
tations (π0, V ) of C∗-dynamical systems of the form (A, G0, αG0 ), where [π0] ∈ Fact(A),
[π0 ⋊ V ] ∈ Fact(A ⋊ G0) and G0 is the isotropy group in G of the quasi-equivalence class
of π0. There is a natural action
G × C(A, G, α) → C(A, G, α),
(g, [(π0, V )]) 7→ [(π0 ◦ αg−1 , V ◦ c−1
g gG0g−1 )]
(5)
where cg : G → G, h 7→ ghg−1. This makes sense since gG0g−1 is the isotropy group of
the quasi-equivalence class of the factor representation π0 ◦ αg−1 of A, and the covariant
representation (π0 ◦ αg−1 , V ◦ c−1
g gG0g−1 ) is a factor representation.
It is easily checked that two unitarily equivalent covariant representations give rise by
induction to unitarily equivalent representations, and it then follows by Corollary 4.6 and
Remark 4.7 that the induction of covariant representations defines a map
Θ : C(A, G, α) → Fact(A ⋊α G).
(6)
15
Theorem 4.14. For any C∗-dynamical system (A, G, α) for which the group G is finite,
the map Θ of (6) is constant on the orbits of the group action (5) and defines a bijective
correspondence
Θ : C(A, G, α)/G → Fact(A ⋊α G)
which preserves ∗-isomorphism classes of commutants, irreducibility, and the types I,II,III of
factor representations.
Proof. It follows by Lemma 4.12 that Θ is surjective.
Injectivity of Θ follows by Lemma 4.5, where the subgroup G0 is determined as the
centralizer of a minimal central idempotent in π(A)′, and this determines the corresponding
factor representation π0 of A. By Remark 4.7 the map Θ preserves ∗-isomorphism classes of
commutants, and we are done.
Remark 4.15. Resume the setting of Definition 4.13 and denote by G(A, G, α) the set of
all subgroups G0 ⊆ G for which there exists [π0] ∈ Fact(A) such that G0 is the isotropy
group of the equivalence class of π0. If we consider the natural action of G on G(A, G, α) by
conjugation, then one obtains a G-equivariant correspondence C(A, G, α) → G(A, G, α) that
associates to every [(π0, V )] in C(A, G, α) the domain of definition of V . Using Theorem 4.14,
this leads to a surjective correspondence
Fact(A ⋊α G) ≃ C(A, G, α)/G → G(A, G, α)/G
which in particular defines an equivalence relation on Fact(A ⋊α G) with finitely many equiv-
alence classes.
This remark provides the following necessary criterion for unitary equivalence of two
covariant factor representations (π1, U1) and (π2, U2) of (A, G, α): For j = 1, 2, assume that
πj is induced from some factor covariant representation (πj0, Vj ) of (A, Gj , αGj ) with [πj0] ∈
Fact(A) and Gj the isotropy group of the quasi-equivalence class of πj0. If the subgroups G1
and G2 fail to be conjugated to each other in G, then the covariant factor representations
(π1, U1) and (π2, U2) are not unitarily equivalent, and equivalently, [π1 ⋊ U1] 6= [π2 ⋊ U2] in
Fact(A ⋊ G).
Example 4.16. In the setting of Definition 4.13, consider the subclass C0(A, G, α) that
corresponds to G0 = {1}. Hence the elements of C0(A, G, α) can be identified with the
unitary equivalence classes of factor ∗-representations π0 : A → B(H0) with the property
that for every 1 6= g ∈ G the factor representation π0 ◦ αg is not quasi-equivalent to π0.
Then the corresponding induced representation (π, U ) of (A, G, α) acts on the Hilbert space
H = ℓ2(G, H0) consisting of all functions f : G → H0 and is given by the formulas from
Definition 4.1. The image of the equivalence class [π0] in C(A, G, α)/G is {π0 ◦ αg g ∈ G},
and all the representations from this set induce the same element of Fact(A ⋊α G) by Theo-
rem 4.14.
In particular, this shows that there is no analogue of Theorem 4.14 with unitary equiva-
lence classes of factor representations replaced by quasi-equivalence classes.
Example 4.17. To illustrate Definition 4.13 by a case which is opposite to Example 4.16,
consider the subclass C1(A, G, α) that corresponds to G0 = G. Hence the elements of
C1(A, G, α) are the unitary equivalence classes of covariant factor representations (π0, V ),
16
where π0 : A → B(H0) is also factor ∗-representation. As (π0, V ) is a covariant represen-
tation of (A, G, α), it follows that, for every g ∈ G, the factor representation π0 ◦ αg is
quasi-equivalent (actually unitarily equivalent) to π0. Then the corresponding induced rep-
resentation (π, U ) of (A, G, α) is given by the formulas from Definition 4.1 and is easily
seen to be unitarily equivalent to (π0, V ). The image of the equivalence class [(π0, V )] in
C(A, G, α)/G is the singleton {[(π0, V )]}, and this corresponds to [π0 ⋊ V ] ∈ Fact(A ⋊α G)
via Theorem 4.14.
For later use we also note a simple necessary criterion for unitary equivalence of factor
representations [π0j ⋊ Vj], j=1,2, that belong to the image of C1(A, G, α) in Fact(A ⋊α G): if
[π01 ⋊ V1] = [π02 ⋊ V2], then [π01] = [π02] ∈ Fact(A) and the unitary representations V1 and
V2 of G are unitarily equivalent.
4.3 Classification of irreducible covariant representations
In the setting of Theorem 4.14, the definition of the class C(A, G, α) is somewhat involved, in
the sense that for a covariant representation (π0, V ) of a C∗-dynamical system (A, G0, αG0 ),
the unitary representation V of G0 is rather implicitly described by the set of conditions
π0 ∈ Fact(A), π0 ⋊ V ∈ Fact(A ⋊ G0), G0 is the isotropy group of the quasi-equivalence class
of π0. We will now discuss this issue for the subclass of irreducible covariant representation
(π0, V ), thus recovering the precise description of the dual space of A ⋊ G as in [Tak67],
[AL10], and [Ka13a].
We will basically classify the factor representation ρ from Lemma 4.12 in the case when
the covariant representation (π, U ) is irreducible. See also [AL10, Th. 3.1] and the comments
after [Ka13b, Th. 4.2]. If S is any group, H is any Hilbert space, and V : S → B(H) and
σ : S × S → T are maps satisfying the Vts = σ(t, s)VtVs for all s, t ∈ S, then we say that V
is a σ-projective representation.
For the following proposition, we recall the existence of the minimal central projections
from Lemmas 4.10 and 4.11.
Proposition 4.18. Let (π, U ) be an irreducible covariant representation of (A, G, α) on H.
Let p1, . . . , pn be the minimal central projections in π(A)′. Then the following assertions
hold:
(i) For j ∈ {1, . . . , n} let d2
g = pj}. Then
for an irreducible representation (ρj , Kj) of A and the covariant representa-
j := dim(π(A)′pj) and Gj := {g ∈ G UgpjU −1
πj ∼= ρ⊕dj
tion (πj, Vj ) of (A, Gj , αGj ) on Hj is irreducible.
j
(ii) Gj = {g ∈ G ρj ◦ αg ≃ ρj} and dj = d1 for j = 1, . . . , n.
(iii) The covariant representation (π1, V1) is unitarily equivalent to a covariant representa-
tion (ρ1 ⊗ 1, V (1)
: G1 → U(K1) is a unitary σ-projective represen-
tation for a 2-cocycle σ : G1 × G1 → T whose cohomology class is uniquely determined,
and V (2)
: G1 → Ud1(C) is a unitary irreducible σ-projective representation.
), where V (1)
1 ⊗ V (2)
1
1
1
(iv) The map Θ of (6) defines by restriction a bijection onto (A ⋊ G)b from the G-orbits of
equivalence classes [(ρ1 ⊗ 1, V (1)
)] as above, where V (2)
can be any irreducible
1
1 ⊗ V (2)
1
17
σ-projective unitary representation of the isotropy group G1 of [ρ1] ∈ bA, and unitary
lead to distinct equivalence classes [(ρ1 ⊗ 1, V (1)
inequivalent choices of V (2)
1 ⊗ V (2)
)].
1
1
Proof. (i) The covariant representation (πj, Vj ) of (A, Gj , αGj ) on Hj is irreducible by Propo-
sition 4.12(ii).
(ii) follows by Lemma 4.11(iii).
(iii) follows by [Tak67, Lemmas 5.1-5.2], where the C∗-algebra needs not be separable if
the group is finite. In fact, in this case, [Tak67, Th. 2.6] (needed in the proof of [Tak67,
Lemma 5.1]) is actually the trivial observation that, because
G1 = {g ∈ G ρ1 ◦ αg ≃ ρ1},
there exist a 2-cocycle σ : G1 × G1 → T, and a unitary σ-projective representation
V (1)
1
: G1 → U(K1), with
(ρ1 ◦ αg)(x) = V (1)
1
(g)ρ1(x)V (1)
1
(g)∗
for
g ∈ G1, x ∈ A.
The cohomology class [σ] is uniquely determined because the corresponding central T-extension
of G1 is isomorphic to the pullback of the T-extension of PU(K1) := U(K1)/T1 under the
corresponding projective representation. Then one has
(V (1)
1
(g) ⊗ 1)π1(x)(V (1)
1
(g) ⊗ 1)∗ = (π1 ◦ αg)(x) = Vgπ1(x)Vg,
1
(g) ⊗ 1)∗Vg belongs to the commutant of π1 ∼= ρ⊕d1
hence (V (1)
1 ∈
Ud1(C) with (V (1)
(g). Since V is a unitary representation, V ≃
V (1)
1 ⊗ V (2)
is a σ-projective
unitary representation of G1, which is also irreducible, exactly as in the proof of [Tak67,
Lemma 5.2].
is a σ-projective representation, it follows that V (2)
(g) ⊗ 1)∗Vg = 1 ⊗ (V (2)
, hence there exists a V (2)
, and V (1)
1
1
1
1
1
1
Conversely, given [ρ1] ∈ bA, one determines its isotropy group G1 with respect to the action
of G on bA, then a cocycle σ ∈ Z 2(G1, T) and a σ-projective representation V (1)
: G1 → B(K1),
and then every irreducible σ-projective unitary representation V (2)
gives rise to a covariant
representation (ρ1 ⊗ 1, V (1)
) of (A, G1, αG1) which is an irreducible representation
and induces an irreducible covariant representation of (A, G, α). Thus (iv) follows by Theo-
rem 4.14, and we are done.
1 ⊗ V (2)
1
1
1
4.4 Representations of fixed-point subalgebras
We will need the following slight generalization of the well-known fact that factor ∗-repre-
sentations are nondegenerate.
Lemma 4.19. Let B0 be a ∗-subalgebra of some ∗-algebra B. Then, for any factor represen-
tation π : B → B(H), one has B0 6⊂ ker π if and only if the subspace span π(BB0)H is dense
in H.
Proof. The hypothesis that π is a factor representation is equivalent to the fact that there
exists no nontrivial closed linear subspace of H that it is invariant both under π(B) and its
18
commutant π(B)′, since the orthogonal projection on such a subspace would belong to the
center π(B)′ ∩ π(B)′′ of π(B)′′.
We may assume H 6= {0} and the assertion will follow by the above observation as soon
as we will have proved that if B0 6⊂ ker π then π(BB0)H 6= {0}, since this subset of H is
invariant under π(B) and π(B)′. In fact, since B0 6⊂ ker π, there exists b0 ∈ B0 with π(b0) 6= 0
and then π(b∗
0)π(b0)H ⊆ π(B)π(B0)H is non-zero.
Proposition 4.20. Let B0 ⊆ B be a ∗-subalgebra of some Banach ∗-algebra with approximate
unit, with B0BB0 ⊆ B0. Denote by FactB0(B) the class of all [π] ∈ Fact(B) with B0 6⊂ ker π
and denote by Ψ([π]) the unitary equivalence class of the B0-nondegenerate part of πB for
[π] ∈ FactB0(B). Then the correspondence
Ψ : FactB0(B) → Fact(B0)
is a bijection and a homeomorphism with respect to the regional topologies, which preserves
∗-isomorphism classes of commutants, irreducibility, and the types I,II,III of factor represen-
tations.
Proof. This follows by [FeDo88, vol. II, XI.7.6, Prop. and Rem. 1], using the above Lemma 4.19.
For the following statement we recall that for any C∗-algebra B we denote by M (B) its
multiplier algebra, which is a unital C∗-algebra with a canonical embedding B ֒→ M (B), and
one has B = M (B) if and only if B is unital.
Proposition 4.21. Denote AG := {x ∈ A (∀g ∈ G) αg(x) = x} and define eA := A + C1 ⊆
M (A). Let p ∈ ℓ1(G, eA; α) ⊆ M (A ⋊α G) be the constant function that is equal to 1 ∈ eA at
every point of G. Then the following assertions hold:
(i) One has p = p∗ = p2, and by restriction and co-restriction the injective linear map
A → A ⋊α G, x 7→ α(·, x), yields a ∗-isomorphism ι : AG → p(A ⋊α G)p.
the canonical extension of π⋊U from A ⋊α G to M (A ⋊α G), and by (π ⋊ U )G the
(ii) For every factor covariant representation (π, U ) of (A, G, α), we will write (π ⋊ U )efor
representation of AG obtained as the nondegenerate part of (π ⋊ U )e◦ ι. We also define
the following open subset of Fact(A ⋊ G),
Then the map
Factp(A ⋊α G) := {[π ⋊ U ] ∈ Fact(A ⋊α G) (π ⋊ U )e(p) 6= 0}.
Factp(A ⋊α G) → Fact(AG),
[π⋊U ] 7→ [(π⋊U )G]
is a well-defined homeomorphism which preserves ∗-isomorphism classes of commutants.
(iii) One has [π⋊U ] ∈ Factp(A ⋊α G) if and only if
[π⋊U ] ∈ Fact(A ⋊α G)
and Xs∈G
Us 6= 0,
and in this case PU := 1
Us is the orthogonal projection onto the essential space
of the ∗-representation (π⋊U ) ◦ ι of AG.
G Ps∈G
19
(iv) Let G0 ⊆ G be any subgroup and (π0, V ) be any covariant representation of the C∗-
dynamical system (A, G0, αG0 ) for which the corresponding induced covariant repre-
sentation (π, U ) of (G, A, α) is irreducible. Then [π⋊U ] ∈ Factp(A ⋊α G) if and only
Vs 6= 0. In particular, this is always the case if G0 = {1}.
if Ps∈G0
(v) The C∗-algebra A is of type I if and only if AG is, and if and only if A ⋊α G is.
Proof. Since G is in particular a compact group, Assertion (i) follows by [Ro79] (see also
[Bl06, II.10.4.18]).
For Assertion (ii), denote B := A ⋊α G = ℓ1(G, A; α), B0 := pBp, and note that, since
p = p∗ = p2 ∈ M (B), pBp is a hereditary subalgebra of B, that is, B0BB0 ⊆ B0. Then the
assertion follows by Proposition 4.20 along with Assertion (i).
For Assertion (iii), recall that if (π, U ) is a covariant representation of (A, G, α), then the
corresponding representation of B = ℓ1(G, A; α) is
π ⋊ U : B → B(H),
(π ⋊ U )(f ) =
1
G Xs∈G
π(f (s))Us.
If (π, U ) is a factor representation, then also π ⋊ U : B → B(H) is a factor representation
hence it is nondegenerate, and then it extends to M (B) ⊇ ℓ1(G, eA; α). By the above formula
one then obtains for the constant function p : G → eA,
G Xs∈G
(π ⋊ U )(p) =
1
Us.
Therefore the condition (π ⋊ U )(p) 6= 0 is equivalent to Ps∈G
Us 6= 0.
For Assertion (iv), let {1 = g0, g1, . . . , gn} be a complete system of representatives of the
G0-left cosets sets in G. Note that
(cid:16)Xg∈G
Ug(cid:17)H = {v ∈ H (∀g ∈ G) Ugv = v} =: HG
and similarly
(cid:16)Xs∈G0
Vg(cid:17)H0 = {v0 ∈ H0 (∀s ∈ G0) Vsv0 = v0} =: HG0
0 ,
hence we must show that HG 6= {0} if and only if HG0
0
6= {0}, which follows by Remark 4.3.
Finally, Assertion (iv) follows by [Ri80, Th. 4.1], and we are done.
Our next aim is to describe the dual spacedAG using Proposition 4.21. In order to simplify
the corresponding statement (Theorem 4.23) we make the following definition.
Definition 4.22. For the C∗-dynamical system (A, G, α), let π0 : A → B(Hπ0 ) be an arbi-
trary irreducible ∗-representation and
Gπ0 := {g ∈ G π0 ◦ αg ≃ π0}.
20
Let ω0 : Gπ0 × Gπ0 → T be a 2-cocycle and V0 : Gπ0 → U(Hπ0) be any ω0-projective unitary
representation satisfying
(∀g ∈ Gπ0 )(∀a ∈ A) V0(g)π0(a)V0(g)∗ = π0(αg(a)).
Let Γ(A, G, α) be the class of all covariant representations (π, V ) of (A, Gπ0 , αGπ0
as
) obtained
π := π0 ⊗ idHU0
and V := V0 ⊗ U0
where U0 : Gπ0 → B(HU0 ) is an arbitrary irreducible ω0-projective unitary representation,
and let Γ0(A, G, α) ⊆ Γ(A, G, α) be the subclass defined by imposing the additional condition
Xg∈Gπ0
V0(g) ⊗ U0(g) 6= 0.
(7)
on by G via α and via the conjugation action of G on itself as in (5) (see also [Tak67, Th. 7.2],
Then the set eΓ0(A, G, α) of all unitary equivalence classes in Γ0(A, G, α), is naturally acted
[Ka13a, Sect. 2]), and we denote by eΓ0/G the corresponding quotient set.
Finally, we define
(8)
I : eΓ0(A, G, α)/G →dAG
to be the correspondence that takes the unitary equivalence class [(π, V )] of a covariant repre-
V and then composing
sentation of (A, Gπ0 , αGπ0
that correspondence with the map [Π] 7→ [ΠG] given by Proposition 4.21(ii).
) to the unitary equivalence class of IndG
Gπ0
The following result should be compared with [Ka13a, Th. 3.3] (or [Tak67, Th. 7.2] if
the C∗-algebra A is of type I).
Theorem 4.23. For every C∗-dynamical system (A, G, α), whose group G is finite, the map
Proof. Using the notation of Definition 4.22, one can see that Proposition 4.18 establishes
I : eΓ0(A, G, α)/G →dAG is a bijection.
the bijection (8) from the set of G-orbits in eΓ(A, G, α) onto (A ⋊ G)b.
In some more detail, to check surjectivity of I, we use our Lemma 4.12 to show that every
irreducible covariant representation of (A, G, α) is induced from a factor representation of
type I. Now the conclusion follows by Proposition 4.21(ii) and (iv).
5 Representations of symmetric tensor powers
It follows by [BN15] that the representations of exponential C∗-algebras eA = ⊕c0
Sn(A)
play an important role in the representation theory of ball semigroups. Since any exponential
C∗-algebra is the c0-direct sum of the symmetric tensor powers Sn(A) = (A⊗n)Sn of a C∗-
algebra, one needs to understand the representations of the algebras Sn(A) in terms of the
representations of A.
n∈N0
Definition 5.1. If A is a C∗-algebra, then for every permutation σ ∈ Sn we denote by
ασ : A⊗n → A⊗n its corresponding permutation action, and we denote by (A⊗n, Sn, α) the
C∗-dynamical system obtained in this way.
21
We record the following fact for later use.
Lemma 5.2. If n ≥ 2, A1, . . . , An are C∗-algebras and A := A1 ⊗ · · · ⊗ An, then the map
cA1 × · · · × cAn → bA,
([π1], . . . , [πn]) 7→ [π1 ⊗ · · · ⊗ πn]
(9)
is injective, and it is also surjective if at least n − 1 of the C∗-algebras A1, . . . , An are of
type I.
Proof. If n = 2, the assertion follows by [OT66, p. 333], or [Bl06, IV.3.4.22/26], then use
induction.
5.1 The case of C ∗-algebras of type I
We are now in a position to describe the representations of the algebras Sn(A) in terms of the
representations of A for any C∗-algebra A of type I, by basically specializing Theorem 4.23
to the C∗-dynamical system (A⊗n, Sn, α). For the following statement we recall that a full
corner of a C∗-algebra A is any subalgebra of the form pAp, where p = p2 = p∗ ∈ M (A),
that is not contained in any closed two-sided ideal of A, (see [Br77]).
Theorem 5.3. Let A be a C∗-algebra of type I and n ≥ 1. For the C∗-dynamical system
(A⊗n, Sn, α), define the projection p ∈ M (A⊗n ⋊α Sn) and the ∗-isomorphism
ι : Sn(A) → p(A⊗n ⋊α Sn)p
as in Proposition 4.21(i). Then the following assertions hold:
(i) The hereditary subalgebra p(A⊗n ⋊α Sn)p is a full corner of A⊗n ⋊α Sn.
(ii) For every irreducible representation Π of A⊗n ⋊α Sn one has eΠ(p) 6= 0.
(iii) The map [Π] 7→ [ΠG] from Proposition 4.21(ii) is a homeomorphism from (A⊗n ⋊α Sn)b
(iv) For a partition q1 + · · · + qm = n, let q := (q1, . . . , qm) and Sq := Sq1 × · · · × Sqm ⊆ Sn.
Let Γ0 be the family of covariant representations (Π1, W1) of (A⊗n, Sq, αSq) obtained
as
onto Sn(A)b.
Π1 := π⊗q1
1 ⊗ · · · ⊗ π⊗qm
m ⊗ idHU01
⊗ · · · ⊗ idHU0m
and W1 := V0 ⊗ U01 ⊗ · · · ⊗ U0m
where πk : A → B(Hπk ) for k = 1, . . . , m are mutually inequivalent irreducible ∗-
representations, V0 : Sq → U(H⊗q1
πm ) is the (external) tensor product of
permutation representations, and U0k : Sqk → U(HU0k ) is an arbitrary irreducible uni-
π1 ⊗ · · · ⊗ H⊗qm
tary representation for k = 1, . . . , m. Then the set eΓ0 of all unitary equivalence classes
in Γ0, is naturally acted upon by Sn via α, and the bijection eΓ0/Sn → (A⊗n ⋊α Sn)b
takes every covariant representation (Π1, W1) to its induced covariant representation of
(A⊗n, Sn, α).
22
Proof. (i) By [Br77, Lemma 1] this is a consequence of (ii).
(ii) Since A is a C∗-algebra of type I, it follows by Lemma 5.2 that the map (9) with
A1 = · · · = An = A is bijective for every n ≥ 1. This implies that, for an irreducible repre-
sentation Π0 : A⊗n → B(HΠ0 ), there exist uniquely determined irreducible representations
π1, . . . , πn of A with Π0 ∼= π1 ⊗ · · · ⊗ πn. After conjugating with a suitable permutation, we
may assume
Π0 ∼= ρ⊗q1
1 ⊗ · · · ⊗ ρ⊗qm
m ,
so that the isotropy group of [Π0] is conjugate to Sq.
Then, in the notation of Theorem 4.23, one has ω0 ≡ 1 and V0 : (Sn)Π0 → U(HΠ0 )
is the (external) tensor product of the permutation representations V0k : Sqk → U(H⊗qk
ρk ).
Since ω0 ≡ 1, it follows that U0 : (Sn)Π0 → U(HU0 ) is an arbitrary irreducible unitary
representation satisfying (7). By [OT66, Cor. to Th. 2], there exist uniquely determined
irreducible unitary representations U0k : Sqk → U(HU0k ) for k = 1, . . . , m whose (external)
tensor product is U0. Then, using the fact that, if Tk ∈ B(H0k) for k = 1, . . . , m, then
T1 ⊗ · · · ⊗ Tm 6= 0 if and only if Tk 6= 0 for k = 1, . . . , m, it is easily seen that condition (7)
is equivalent to
Tk := Xσ∈Sqk
V0k(σ) ⊗ U0k(σ) 6= 0 for k = 1, . . . , m.
To see that the above condition is satisfied, note that the irreducibility of the representa-
tion U0k implies
1
qk! Xσ∈Sqk
U0k(σ) = 1 for k = 1, . . . , m.
Hence for any nonzero vector v ∈ HU0k and for every nonzero symmetric tensor w ∈
Sqk (Hπj0
one has
) ⊆ H⊗qk
πj0
k
k
0 6= w ⊗ (qk!v) = w ⊗ Xσ∈Sqk
=(cid:16) Xσ∈Sqk
V0k(σ) ⊗ U0k(σ)(cid:17)(w ⊗ v).
U0k(σ)v = Xσ∈Sqk
V0k(σ)w ⊗ U0k(σ)v
This completes the proof of Assertion (ii).
(iii) follows by Theorem 4.23, and we are done.
Example 5.4. Consider the C∗-algebra A = S∞(H) of compact operators on some infinite
ical representation of A. Then, for any n ≥ 2, in Theorem 5.3 one has m = 1, and it follows
dimensional complex Hilbert space. Then it is known that bA = {[π]}, where π is the tautolog-
that Sn(A)b is parametrized by cSn, just as the holomorphic Schur -- Weyl representations. See
Subsection 5.2 below for more details.
5.2 Links to Schur -- Weyl theory for irreducible representations
Let A be any C∗-algebra and recall from the introduction that its unitary group is
U(A) := (1 + A) ∩ U(M (A)),
23
where M (A) is the multiplier C∗-algebra of A, and M (A) = A if A is unital.
It follows
from [BN12] that to every irreducible ∗-representation π : A → B(ℓ2(J)) there corresponds
an infinite family of unitary irreducible representations πA
λ of U(A) obtained by Schur -- Weyl
theory (see [BN12, Def. 4.1]), where λ belongs to the additive group PJ ∼= Z(J) of all finitely
supported Z-valued functions on J. Here a set J = J[π] is fixed for very equivalence class
[π] ∈ bA. We then define the Schur -- Weyl map of A as
ΨA : G[π]∈ bA
PJ[π]/S(J[π]) → U(A)b,
[λ] 7→ [πA
λ ]
for [λ] ∈ PJ[π]/S(J[π]) and [π] ∈ bA,
(10)
where the notation is as in [BN12, Th. 3.2].
In particular, for λ ∈ PJ , we denote by [λ]
its equivalence class modulo the natural action of the group S(J) of finitely supported per-
mutations of J. As an application of Theorem 4.14 (via Theorem 4.23) we will show in
Theorem 5.8 that the Schur -- Weyl map ΨA is always injective but, if A is separable and
is not surjective.
non-isomorphic to K(ℓ2(N)), then bA contains at least two elements and this implies that ΨA
n = Pj λj , then the representation πA
On the other hand, using [BN15, Th. 6.8], one easily sees that if λj ≥ 0 for every j and
λ uniquely extends to an irreducible ∗-representation
λ and termed a Schur -- Weyl representation in the following,
of Sn(A), denoted again by πA
where by extension we actually mean a factorization through the ∗-morphism
U(A) ֒→ Sn(A),
a 7→ a⊗n.
To begin with, we will show in Proposition 5.7 how the equivalence classes [πA
can be recovered in the setting of the above Theorem 4.14. To this end we need the following
general lemma.
λ ] ∈ Sn(A)b
Lemma 5.5. Let (π, U ) be a covariant representation of a C∗-dynamical system (B, G, α),
with the finite group G, on a Hilbert space H for which π is irreducible. For any unitary
irreducible representation λ : G → B(V) define the Hilbert space HV := B2(V, H) ≃ H ⊗ V ∗
and the ∗-representation πV : B → B(HV ), πV (x)T := π(x)T . Then the following assertions
hold:
(i) The pair (πV , U ⊗ λ) is an irreducible covariant representation of (B, G, α).
(ii) Let ι : BG → B ⊆ B ⋊ G be the canonical embedding as in Proposition 4.21(i).
If
HomG(V, H) 6= {0}, then this is the essential space of (πV ⋊ (U ⊗ λ)) ◦ ι : BG → B(HV )
and the corresponding nondegenerate representation of BG on HomG(V, H) is irreducible
and a subrepresentation of πV BG.
Proof. For every g ∈ G and x ∈ B one has
(U ⊗ λ)(g)πV (x)(U ⊗ λ)(g−1) = (Ugπ(x)Ug−1 ) ⊗ idV ∗ = π(αg(x)) ⊗ idV ∗ = πV (αg(x)),
so that (πV , U ⊗ λ) is a covariant representation of (B, G, α).
To prove that (πV , U ⊗λ) is irreducible, we must check that πV (B)′∩(U ⊗λ)′
G = C idH⊗V ∗.
One has πV (B)′ = (π(B)⊗C idV ∗ )′ = π(B)′ ⊗B(V ∗) = C idH ⊗B(V ∗) because π is irreducible.
Hence we must prove that if T ∈ B(V ∗) satisfies idH ⊗T ∈ (U ⊗ λ)′
G, then T ∈ C idV ∗ . The
24
′
condition on T is equivalent to T ∈ λ
G hence, using the hypothesis that λ is an irreducible
representation hence so is λ, one obtains T ∈ C idV ∗, and this completes the proof of the first
assertion.
For the second assertion, it follows by Proposition 4.21(ii) and (iii) that HomG(V, H),
that is, (H ⊗ V ∗)G, is the essential space of (πV ⋊ (U ⊗ λ)) ◦ ι, and the corresponding
nondegenerate representation of BG is irreducible because πV ⋊ (U ⊗ λ) is irreducible. To see
that this irreducible representation of BG is a subrepresentation of πV BG , that is, it is equal
to x 7→ πV (x)HomG(V,H), note that, for x ∈ BG, one has
((πV ⋊ (U ⊗ λ)) ◦ ι)(x) =
1
G Xg∈G
(πV(cid:0)ι(x)(g)(cid:1)(U ⊗ λ)g =
1
G Xg∈G
πV (x)(U ⊗ λ)g
= πV (x)P = P πV (x),
where P := 1
(U ⊗ λ)g is the orthogonal projection of H ⊗ V ∗ onto (H ⊗ V ∗)G. This
G Pg∈G
completes the proof.
Remark 5.6. In connection with Lemma 5.5, we note that the factor covariant represen-
tations (π, U ) of (B, G, α) for which the irreducible decomposition of U contains a fixed
irreducible representation λ of G were classified in [La80, Th. 1] even for compact groups G.
Proposition 5.7. Let A be a unital C∗-algebra and fix some integer n ≥ 1 and a unitary
irreducible representation λ : Sn → B(V). For any irreducible ∗-representation π : A → B(H)
denote by W (π) : Sn → B(H⊗n) the corresponding permutation representation and define
(π⊗n)V : A⊗n → B(H⊗n ⊗ V ∗),
(π⊗n)V (x) := π⊗n(x) ⊗ idV ∗ .
Then ((π⊗n)V , W (π) ⊗ λ) is an irreducible covariant representation of (A⊗n, Sn, α) and the
image of its Sn-orbit through the map I of Theorem 4.23 is [πA
Proof. The covariant representation ((π⊗n)V , W (π)⊗λ) is irreducible by Lemma 5.5, because
π⊗n : A⊗n → B(H⊗n) is irreducible. Using the notation of Definition 4.22, one has π0 = π⊗n
and G = Gπ0 = Sn, hence IndG
(π0) = π0. Moreover, ω0 ≡ 1, V0 = W (π), and U0 = λ.
Condition (7) is satisfied, since this can be proved just as the relation Tk 6= 0 in the last
part of the proof of Theorem 5.3(ii). The image of the Sn-orbit of the equivalence class of
(πV , W (π)⊗λ) under I is the equivalence class of the nondegenerate part of the representation
(π0 ⋊ (V0 ⊗ U0)) ◦ ι = (π⊗n ⋊ (W (π) ⊗ λ)) ◦ ι. This nondegenerate part is unitarily equivalent
to the representation
Gπ0
λ ] ∈ Sn(A)b ֒→ U(A)b.
Sn(A) = (A⊗n)Sn → B(HomSn (V, H⊗n)),
x 7→ π⊗n(x) ⊗ idV ∗
(11)
by Lemma 5.5(ii), since
(π⊗n(x) ⊗ idV ∗)(T ) = π⊗n(x)T for every T ∈ HomSn(V, H⊗n) ≃ (H⊗n ⊗ V ∗)Sn .
But then, by composing (11) with the power map U(A) → Sn(A), a 7→ a⊗n one obtains a
unitary irreducible representation of U(A) (compare [BN15, Cor. 6.10]) on the Hilbert space
HomSn(V, H⊗n) =: Sλ(H⊗n) which is unitarily equivalent to the Schur -- Weyl representation
πA
λ (see [BN12, Rem. A.6]).
25
In Assertion (ii) of the following statement we use the notation C1(A⊗n, Sn, α) as intro-
duced in Example 4.17.
Theorem 5.8. For any C∗-algebra A, the following assertions hold:
(i) The Schur -- Weyl map ΨA from (10) is injective.
(ii) One has Im ΨA ⊆ Sn≥0
representations which belong to C1(A⊗n, Sn, α), for all n ≥ 1.
Sn(A)b ֒→ U(A)band Im ΨA ∩ Sn(A)b consists of restrictions of
(iii) The map ΨA is bijective if A is isomorphic to the C∗-algebra of compact operators on
some complex Hilbert space.
(iv) If πj : A → B(Hj), j = 1, 2, are inequivalent irreducible ∗-representations, then the map
a 7→ π1(a) ⊗ π2(a) is a unitary irreducible representation of U(A) on H1 ⊗ H2 whose
equivalence class does not belong to the image of the Schur -- Weyl map.
Proof. (i), (ii) Recall from (10) that the Schur -- Weyl map is
ΨA : G[π]∈ bA
[(π2)A
λ2
[λ] 7→ [πA
λ ]
PJ[π] /S(J[π]) → U(A)b,
for [λ] ∈ PJ[π]/S(J[π]) and [π] ∈ bA.
To prove that ΨA is injective, let [πj] ∈ bA and λj ∈ PJ[πj ] for j = 1, 2 with [(π1)A
] ∈ U(A)b. We must check that [π1] = [π2] := [π] ∈ bA and [λ1] = [λ2] ∈ PJ[π] /S(J[π]).
Sn(A)b is the image through the bijection I (Theorem 4.23) of the irreducible covariant
λ2 ] ∈ Sn(A)b ֒→
bA, implies
Denote also by λj : Sn → B(Vj ) the unitary irreducible representation associated with
λj ∈ PJ[π] , where n ≥ max{ supp λ1, supp λ2}. It follows by Proposition 5.7 that [πλ] ∈
)V , W (πj) ⊗ λj) of (A⊗n, Sn, α). Then [(π1)A
representation ((π⊗n
λ1 ] = [(π2)A
] =
λ1
[(π⊗n
1
)V1 ⋊ (W (π1) ⊗ λ1)] = [(π⊗n
2
j
)V2 ⋊ (W (π2) ⊗ λ2)] ∈ (A⊗n ⋊ Sn)b.
Now we can use Example 4.17 and the injectivity of the map (9) in Lemma 5.2 to obtain
[π1] = [π2], that is, [π1] = [π2] := [π] ∈ bA. It then follows by [BN12, Th. 4.4] that also
[λ1] = [λ2] ∈ PJ[π] /S(J[π]), and this completes the proof of the fact that ΨA is injective.
(iii) If A is isomorphic to the C∗-algebra of compact operators on some complex Hilbert
space, then ΨA is surjective by [Ne14, Th. 3.21].
(iv) Now let πj : A → B(Hj ) for j = 1, 2 be inequivalent irreducible ∗-representations
and define the irreducible ∗-representation π0 := π1 ⊗ π2 : A⊗2 → B(H1 ⊗ H2). Using again
the injectivity of the map (9) in Lemma 5.2 it follows that the isotropy group of [π0] for
the natural action of S2 on dA⊗2 is {1}, that is, [π] belongs to the class C0(A⊗2, S2, α) from
Example 4.16. This shows that the image of [π0] under I does not belong to the image of
the Schur -- Weyl map ΨA. It only remains to compute the image of [π0] under the map I.
To this end we first compute the corresponding induced representation (π, U ) of (A⊗2, S2, α)
given by Definition 4.1. Since G0 = {1}, we have H = H0 ⊕ H0 thought of as column vectors
and
U : S2 ≃ Z2 → B(H) with U0 = 1 and U1 =(cid:18)0 1
1 0(cid:19)
26
and also
π : A⊗2 → B(H),
π(a1 ⊗ a2) =(cid:18)π1(a1) ⊗ π2(a2)
0
π1(a2) ⊗ π2(a1)(cid:19)
0
for all a1, a2 ∈ A.
elements of A⊗2 with the multiplication given by
If we view the elements of the crossed product A⊗2 ⋊ S2 as pairs of
(a1 ⊗ a2,b1 ⊗ b2) · (c1 ⊗ c2, d1 ⊗ d2)
= ((a1c1) ⊗ (a2c2) + (b1d1) ⊗ (b2d2), (a1d1) ⊗ (a2d2) + (b1c2) ⊗ (b2c1))
then the irreducible representation π ⋊ U : A⊗2 ⋊ S2 → B(H) is given by
(π ⋊ U )(a1 ⊗ a2, b1 ⊗ b2) = π(a1 ⊗ a2)U0 + π(b1 ⊗ b2)U1
=(cid:18)π1(a1) ⊗ π2(a2) π1(b2) ⊗ π2(b1)
π1(b1) ⊗ π2(b2) π1(a2) ⊗ π2(a1)(cid:19)
degenerate representation on H whose essential subspace is the image of (cid:18)1 1
and by composing π ⋊ U with ι : S2(A) → A⊗2 ⋊ S2, ι(a ⊗ a) = (a ⊗ a, a ⊗ a), we obtain a
1 1(cid:19), i.e., the
"diagonal subspace" of H = H0 ⊕ H0. The nondegenerate part of (π ⋊ U ) ◦ ι is unitarily
equivalent to the representation
S2(A) → B(H0),
a ⊗ a 7→ π1(a) ⊗ π2(a)
(and is irreducible by Proposition 4.21(iii)). Composing it further with the embedding
U(A) → S2(A), a 7→ a ⊗ a, we obtain the irreducible representation of U(A) from the
statement. This concludes the proof.
6 Schur -- Weyl theory for factor representations
In Propositions 6.1 -- 6.2 below we give a generalization of [Nes13, Th. 1(1)] dealing with
factors of type II1 to more general von Neumann algebras, and this also provides a partial
generalization of [EnIz15, Th. 4.1] where a Schur -- Weyl duality property was established for
the standard representation of the hyperfinite factor of type II1. In the case M = B(H) the
following proposition also recovers the Schur -- Weyl duality studied in [BN12].
Proposition 6.1. Let M ⊆ B(H) be a von Neumann algebra and for any n ≥ 1 consider
the unitary permutation representation V : Sn → B(H⊗n) and its corresponding action by
∗-automorphisms α : Sn → Aut(M⊗n). Then for the homomorphism of multiplicative ∗-
semigroups Γ : M → (M⊗n)Sn , Γ(a) := a⊗n, one has
Γ(U(M))′′ = (M⊗n)Sn
and Γ(U(M))′ = (VSn ∪ (M′)⊗n)′′.
Proof. The inclusion Γ(U(M))′′ ⊆ (M⊗n)Sn is clear. For the opposite inclusion we use the
differential of Γ at 0 ∈ M,
dΓ : M → (M⊗n)Sn ,
dΓ(a) =
nXk=1
1⊗(k−1) ⊗ a ⊗ 1⊗(n−k).
27
By the formulas dΓ(a) = 1
i
easily obtains the equality Γ(U(M))′′ = dΓ(M)′′, hence it suffices to prove that (M⊗n)Sn ⊆
dΓ(M)′′. As one has the surjective map
dt(cid:12)(cid:12)t=0Γ(eita) and Γ(eia) = eidΓ(a) for all a = a∗ ∈ M, one
d
E : M⊗n → (M⊗n)Sn, E(b) =
1
n! Xσ∈Sn
ασ(b)
we will have to prove
(∀b1, . . . , bn ∈ M) E(b1 ⊗ · · · ⊗ bn) ∈ dΓ(M)′′.
(12)
We will prove this by recurrence after r := {j ∈ {1, . . . , n} bj 6= 1}. If r = 1, this is clear
since E(b1 ⊗ · · · ⊗ bn) = 1
dΓ(bj1), where j1 is the unique j ∈ {1, . . . , n} with bj 6= 1. Now
n!
assume r + 1 ≤ n and define ℓ0 := max{j ∈ {1, . . . , n} bj 6= 1}, hence bℓ0 6= 1. Also define
cj := bj if j ∈ {1, . . . , n} \ {ℓ0} and cℓ0 := 1. Then one has
dΓ(bℓ0) · n!E(c1 ⊗ · · · ⊗ cn) =(cid:16) nXk=1
nXk=1 Xσ∈Sn
=
cσ(1) ⊗ · · · ⊗ (bℓ0cσ(k)) ⊗ · · · ⊗ cσ(n)
1⊗(k−1) ⊗ bℓ0 ⊗ 1⊗(n−k)(cid:17)(cid:16)Xσ∈Sn
cσ(1) ⊗ · · · ⊗ cσ(n)(cid:17)
If we split the above sum according to the pairs (σ, k) ∈ Sn×{1, . . . , n} which satisfy σ(k) = ℓ0
and σ(k) 6= ℓ0, respectively, then we obtain
dΓ(bℓ0) · n!E(c1 ⊗ · · · ⊗ cn) =n!E(b1 ⊗ · · · ⊗ bn)
+ n! Xk∈{1,...,n−1}\{ℓ0}
E(b1 ⊗ · · · ⊗ (bℓ0bk) ⊗ · · · ⊗ bn)
where in every summand of the second sum, the ℓ0-th factor in the tensor product is equal
to 1. Solving the above equation for E(b1 ⊗ · · ·⊗ bn) and using the recurrence hypothesis, one
completes the proof of (12), and we have already seen above that this implies the assertion.
It remains to prove the second equality from the statement. Using the commutation
2 and the Bicommutant Theorem,
theorem for von Neumann algebras (M1 ⊗M2)′ = M′
one obtains
1 ⊗M′
(VSn ∪ (M′)⊗n)′ = (VSn ∪ (M⊗n)′)′ = V ′
Sn ∩ M⊗n = (M⊗n)Sn = Γ(U(M))′′
where we used also the first of the asserted equalities from the statement, which was already
proved. Now the second equality from the statement follows by taking commutants in the
above equalities and using once again the Bicommutant Theorem.
The following proposition shows that, beyond the classical Schur -- Weyl theory when M =
B(H), the picture is completely different.
Proposition 6.2. Assume the setting of Proposition 6.1. If M is a factor of type II1, II∞,
or III, then Γ : U(M) → B(H⊗n), Γ(u) = u⊗n, is a factor representation of the same type
as M.
28
Proof. By Proposition 6.1 it suffices to prove that (M⊗n)Sn is a factor of the same type
as M.
Since M is not a factor of type I, it follows by [Sa75, Th. 5] that for every σ ∈ Sn \ {1}
the automorphism α(σ) of M⊗n is not inner, that is, it is not of the form x 7→ uxu∗ for any
unitary element u ∈ M. Then, since M is a factor, it follows by [Sa71, Prop. 2.6.7] that
M⊗n is a factor, hence it follows by [Au76, Cor. to Prop. II.3] that (M⊗n)Sn is in turn a
factor.
To check that (M⊗n)Sn has the same type as M, we discuss below separately the cases
that can occur, using the faithful normal conditional expectation
E : M⊗n → (M⊗n)Sn , E(a) =
1
n! Xσ∈Sn
ασ(a).
If 0 ≤ a ∈ M⊗n then, using the fact that 0 ≤ ασ(a) if σ ∈ Sn \ {1} and a = ασ(a) if σ = 1,
one obtains a ≤ n!E(a), hence the conditional expectation E has finite index in the sense of
[FrKi98]. It then follows by [FrKi98, Prop. 2.2] that the factors M⊗n and (M⊗n)Sn have
the same type II1, II∞, or III.
If M is of type III, then M⊗n is of type III by [Sa71, Th. 2.6.4]. If M is of type II1
or II∞, then M⊗n is of the same type by [Sa71, Prop. 2.6.1/3]. This completes the proof.
Remark 6.3. With the notation of the proof of Proposition 6.2, the index of the conditional
expectation E is Ind E = Sn = n! by [Lo92, Ex. 1.2]. Moreover, if M is of type IIIλ for
some λ ∈ [0, 1], then more refined information on the type of (M⊗n)Sn can be obtained by
[Lo92, Th. 2.7].
The following theorem extends [Nes13, Th. 1(2)] to infinite factors.
Theorem 6.4. In the setting of Proposition 6.2 we define for every non-zero projection
p = p2 = p∗ ∈ Γ(U(M))′ with range Hp = pH the representation
Γp : U(M) → U(Hp), Γp(u) := Γ(u)Hp .
(i) Γp is a factor representation of the same type II or III as M and Γp ≈ Γ.
(ii) If M is a factor of type II, then for any faithful normal trace τ on Γ(U(M))′ and finite
j ∈ Γ(U(M))′ for j = 1, 2, one has Γp1 ≤ Γp2 if and only if
projections pj = p2
τ (p1) ≤ τ (p2), and Γp1 ≃ Γp2 if and only if τ (p1) = τ (p2).
j = p∗
(iii) If M is a countably decomposable factor of type III, then Γp ≃ Γ.
Proof. (i) It is easily seen that Γp(U(M))′′ = Γ(U(M))p (reduced von Neumann algebra),
hence Γp is a factor representation, by [SZ79, Th. 3.13]. Moreover, the reduction map
Γ(U(M)) → Γ(U(M))p is a ∗-isomorphism by [SZ79, Prop. 3.14], hence Γ and Γp are
quasi-equivalent factor representations.
(ii) If M is a factor of type II, then Γ(U(M))′′ is a factor of type II by Proposition 6.2,
hence by [Dix69, Ch. I, §6, no. 8, Cor. 1 of Prop. 13] also Γ(U(M))′ is a factor of type II,
and then it has a faithful normal trace τ . If pj = p2
j ∈ Γ(U(M))′ are finite projections
for j = 1, 2, then one has τ (p1) ≤ τ (p2) (respectively τ (p1) = τ (p2)) if and only if p1 ≺ p2
j = p∗
29
(respectively p1 ∼ p2) in Γ(U(M))′ by [Dix69, Ch.
III, §2, Prop. 13], which is further
equivalent to Γp1 ≤ Γp2 by [Dix64, Cor. 5.1.4] (respectively Γp1 = Γp2 by [Dix64, Cor.
5.1.3]).
(iii) If M is a countably decomposable factor of type III and 0 6= pj = p2
j ∈ Γ(U(M))′
for j = 1, 2, then p1 ∼ p2 in Γ(U(M))′ by [Sa71, Prop. 2.2.14], hence Γp1 and Γp2 are unitarily
equivalent again by [Dix64, Cor. 5.1.3].
j = p∗
Remark 6.5. It follows by Proposition 6.1 that to every central projection χ ∈ C[Sn] there
corresponds an orthogonal projection V (χ) ∈ Γ(U(M))′. In the case when M is a factor of
type II, the traces of these projections were computed in the proof of [Nes13, Th. 1(3)] and
in [DaKa06, Cor. 5].
When M is a factor of type II1 with its faithful normal tracial state τ , the character of
the representation Γ is (τ U(M))n (which belongs to the list in [EnIz15, Th. 1.5]) and this is
equal to the character of the representation Γp if 0 6= p = p2 = p∗ ∈ Γ(U(M))′, because, by
Theorem 6.4, the representations Γ and Γp are quasi-equivalent.
Acknowledgment
We wish to thank Jan Stochel for informing us about his paper [St92].
References
[AL10]
[Ar76]
[Au76]
[BN12]
Arias, A., and F. Latr´emoli`ere, Irreducible representations of C∗-crossed products
by finite groups. J. Ramanujan Math. Soc. 25 (2010), no. 2, 193 -- 231.
Arveson, W., "An Invitation to C∗-algebras", Graduate Texts in Mathematics 39,
Springer, 1976.
Aubert, P.-L., Th´eorie de Galois pour une W ∗-alg`ebre, Comment. Math. Helv. 51
(1976), no. 3, 411 -- 433.
Beltit¸a, D., and K.-H. Neeb, Schur -- Weyl theory for C∗-algebras, Math. Nachr.
285 (2012), no. 10, 1170 -- 1198.
[BN15] -- , Nonlinear completely positive maps and dilation theory for real involutive al-
gebras, Integral Equations Operator Theory 83 (2015), no. 4, 517 -- 562.
[Bl06]
[Br77]
[Co75]
Blackadar, B., "Operator Algebras," Encyclopedia of Mathematical Sciences Vol.
122, Springer-Verlag, Berlin, 2006.
Brown, L. G., Stable isomorphism of hereditary subalgebras of C∗-algebras, Pacific
J. Math. 71 (1977), no. 2, 335 -- 348.
Corwin, L., Induced representations of discrete groups. Proc. Amer. Math. Soc. 47
(1975), 279 -- 287.
[DaKa06] Daletskii, A., and A. Kalyuzhnyi, Permutations in tensor products of factors, and
L2 cohomology of configuration spaces, Methods Funct. Anal. Topology 12 (2006),
no. 4, 341 -- 352.
30
[DwOl15] Dawson, M., and G. ´Olafsson, A survey of amenability theory for direct-limit
groups, in J.G. Christensen, S. Dann, A. Mayeli, and G. ´Olafsson (eds.), "Trends
in harmonic analysis and its applications"', Contemp. Math., 650, Amer. Math.
Soc., Providence, RI, 2015, pp. 89 -- 109.
[Dix64] Dixmier, J., "Les C∗-alg`ebres et leurs repr´esentations," Gauthier-Villars, Paris,
1964.
[Dix69] -- , "Les alg`ebres d'op´erateurs dans l'espace hilbertien (alg`ebres de von Neu-
mann)". Deuxi`eme ´edition, revue et augment´ee. Cahiers Scientifiques, Fasc. XXV.
Gauthier-Villars Editeur, Paris, 1969.
[EnIz15] Enomoto, T., and M. Izumi, Indecomposable characters of infinite dimensional
groups associated with operator algebras, J. Math. Soc. Japan (to appear).
[FeDo88] Fell, J. M. G., and R. S. Doran, "Representations of ∗-Algebras, Locally Compact
Groups, and Banach-∗-Algebraic Bundles I, II," Academic Press, 1988.
[FrKi98] Frank, M., and E. Kirchberg, On conditional expectations of finite index, J. Oper-
ator Theory 40 (1998), no. 1, 87 -- 111.
[HLS81] Høegh-Krohn, R., and M. B. Landstad, and E. Størmer, Compact ergodic groups
of automorphisms, Ann. of Math. (2) 114 (1981), no. 1, 75 -- 86.
[JiT03]
[JP95]
Ji, G., and J. Tomiyama, On characterizations of commutativity of C∗-algebras,
Proc. Amer. Math. Soc. 131 (2003), no. 12, 3845 -- 3849.
Junge, M., and G. Pisier, Bilinear forms on exact operator spaces and
B(H) ⊗ B(H), Geom. Funct. Anal. 5 (1995), no. 2, 329 -- 363.
[KR92] Kadison, R.V., and J.R. Ringrose, "Fundamentals of the theory of operator al-
gebras. Vol. IV. Special topics. Advanced theoryan exercise approach." Birkhuser
Boston, Inc., Boston, MA, 1992.
[Ka13a] Kamalov, F., The dual structure of crossed product C∗-algebras with finite groups,
Bull. Aust. Math. Soc. 88 (2013), no. 2, 243 -- 249.
[Ka13b] -- , Covariant representations of C∗-dynamical systems with compact groups, J.
Operator Theory 70 (2013), no. 1, 259 -- 272.
[Ki73]
[La80]
[Lo92]
Kirillov, A. A., Representation of the infinite dimensional unitary group, Dokl.
Akad. Nauk. SSSR 212 (1973), 288 -- 290.
Landstad, M. B., Algebras of spherical functions associated with covariant systems
over a compact group, Math. Scand. 47 (1980), no. 1, 137 -- 149.
Loi, Ph. H., On the theory of index for type III factors, J. Operator Theory 28
(1992), no. 2, 251 -- 265.
[Mu90] Murphy, G. J., "C∗-algebras and Operator Theory," Academic Press, 1990
31
[Ne14]
[NS13]
Neeb, K.-H., Unitary representations of unitary groups, in: G. Mason, I. Penkov
and J.A. Wolf (eds), "Developments and retrospectives in Lie theory", Geometric
and analytic methods. Developments in Mathematics, 37. Springer, Cham, 2014,
pp. 197 -- 243.
Neher, E., and A. Savage, A survey of equivariant map algebras with open problems,
In: V. Chari, J. Greenstein, K.C. Misra, K.N. Raghavan, S. Viswanath (eds.),
"Recent developments in algebraic and combinatorial aspects of representation
theory", Contemp. Math., 602, Amer. Math. Soc., Providence, 2013, pp. 165 -- 182.
[Nes13] Nessonov, N. I., Schur -- Weyl duality for the unitary groups of II1-factors, Preprint,
arXiv.RT:1312.0824
[OT66] Okayasu, T., and M. Takesaki, Dual spaces of tensor products of C∗-algebras,
Tohoku Math. J. (2) 18 (1966), 332 -- 337.
[RW98] Raeburn, I., and D. P. Williams, "Morita equivalence and continuous-trace C∗-
algebras". Mathematical Surveys and Monographs, 60. American Mathematical
Society, Providence, RI, 1998.
[Ri80]
[Ro79]
[Sa71]
Rieffel, M. A., Actions of finite groups on C∗-algebras, Math. Scand. 47 (1980),
no. 1, 157 -- 176.
Rosenberg, J., Appendix to O. Bratteli's paper on "Crossed products of UHF alge-
bras by product type actions", Duke Math. J. 46 (1979), no. 1, 25 -- 26.
Sakai, Sh., C∗-algebras and W ∗-algebras, Ergebnisse der Mathematik und ihrer
Grenzgebiete, Band 60. Springer-Verlag, New York-Heidelberg, 1971.
[Sa75] -- , Automorphisms and tensor products of operator algebras, Amer. J. Math. 97
(1975), no. 4, 889 -- 896.
[Se77]
[St92]
[SZ79]
Serre, J.-P., "Linear representations of finite groups." Graduate Texts in Mathe-
matics, Vol. 42. Springer-Verlag, New York-Heidelberg, 1977.
Stochel, J., Decomposition and disintegration of positive definite kernels on convex
∗-semigroups, Ann. Polon. Math. 56 (1992), no. 3, 243 -- 294.
Stratila, S¸., L. Zsid´o, Lectures on von Neumann algebras. Editura Academiei,
Bucharest; Abacus Press, Tunbridge Wells, 1979.
[Tak67] Takesaki, M., Covariant representations of C∗-algebras and their locally compact
automorphism groups, Acta Math. 119 (1967), 273 -- 303.
[Up85]
Upmeier, H., "Symmetric Banach Manifolds and Jordan C∗-algebras", North-
Holland Mathematics Studies, 104. Notas de Matem´atica, 96. North-Holland Pub-
lishing Co., Amsterdam, 1985.
[Wo14] Wolf, J.A., Principal series representations of infinite dimensional Lie groups,
I: Minimal parabolic subgroups, in R. Howe, M. Hunziker, and J.F. Willenbring
(eds.), "Symmetry: representation theory and its applications", Progr. Math.,
257, Birkhauser/Springer, New York, 2014, pp. 519 -- 538.
32
|
1502.01873 | 1 | 1502 | 2015-02-06T12:42:14 | Limit distributions of Gaussian block ensembles | [
"math.OA",
"math.PR"
] | It has been shown by Voiculescu that important classes of square independent random matrices are asymptotically free, where freeness is a noncommutative analog of classical independence. Recently, we introduced the concept of matricial freeness, which is similar to freeness in free probability, but it also has some matricial features. Using this new concept of noncommutative independence, we described the asymptotics of blocks and symmetric blocks of certain classes of independent random matrices. In this paper, we present the main results obtained in this framework, concentrating on the ensembles of blocks of Gaussian random matrices. | math.OA | math | LIMIT DISTRIBUTIONS OF GAUSSIAN BLOCK ENSEMBLES
ROMUALD LENCZEWSKI
Abstract. It has been shown by Voiculescu that important classes of square inde-
pendent random matrices are asymptotically free, where freeness is a noncommutative
analog of classical independence. Recently, we introduced the concept of matricial
freeness, which is similar to freeness in free probability, but it also has some matri-
cial features. Using this new concept of noncommutative independence, we described
the asymptotics of blocks and symmetric blocks of certain classes of independent ran-
dom matrices. In this paper, we present the main results obtained in this framework,
concentrating on the ensembles of blocks of Gaussian random matrices.
5
1
0
2
b
e
F
6
]
.
A
O
h
t
a
m
[
1
v
3
7
8
1
0
.
2
0
5
1
:
v
i
X
r
a
1. Introduction
Our main objective is to describe asymptotic joint distributions of rectangular blocks
of independent random matrices, called random blocks, under the expectation of nor-
malized partial traces. For that purpose, we use a new concept of noncommutative
independence called matricial freeness and associated arrays of operators which give
Hilbert space realizations of these distributions. The well-known connection between
free probability and the asymptotics of independent square random matrices under the
expectation of normalized trace can also be reproduced in this framework.
The most fundamental results of this nature were obtained by Voiculescu [16], who
showed that certain ensembles of independent n n random matrices tY pu, nq : u P
Uu were asymptotically free under the expectation of the normalized complete trace,
τpnq " E b Trpnq, where Trpnq stands for the trace divided by n.
In particular, if
the entries of Y pu, nq are i.i.d. complex p0, 1{nq-Gaussian random variables, we can
symbolically write
lim
nÑ8
Y pu, nq " ηpuq
where tηpuq : u P Uu is the standard free circular system of operators and convergence
is understood in the sense of mixed moments under τpnq. The operators ηpuq live in
the free Fock space and have the standard circular distribution (uniform distribution
on the unit disc in the complex plane) in the vacuum state. A similar result holds for
Hermitian Gaussian random matrices whose limit joint distributions are described by
mixed moments of free Gaussian operators with semicircle distributions.
Gaussian random matrices studied by Voiculescu had i.i.d. entries, except that in
the case of Hermitian ensembles it holds that Yi,jpu, nq " Yj,ipu, nq. If the entries are
independent but not identically distributed, standard free probability may not suffice
to describe asymptotic distributions of these matrices. One of the approaches is to
employ the much more general scheme of freeness with amalgamation as in the papers
of Shlyakhtenko [14] and Benaych-Georges [3]. Recently, we have developed another
2010 Mathematics Subject Classification: 46L54, 15B52, 60F99
Key words and phrases: free probability, random matrix, freeness, matricial freeness
1
2
R. LENCZEWSKI
approach [7,8] in the case when the entries are independent and block-identically dis-
tributed (i.b.i.d.). This approach is based on the concept of matricial freeness [6],
which can be viewed as a matricial generalization of freeness, lying somewhere between
freeness and freeness with amalgamation.
In this context, we decompose matrices Y pu, nq with i.b.i.d. entries into rectangular
blocks tSp,qpu, nq : 1 ď p, q ď ru, or symmetric blocks tTp,qpu, nq : 1 ď p ď q ď ru. The
block analog of the above Voiculescu's result for the ensemble of independent Gaussian
matrices can then be written in the form
lim
nÑ8
Sp,qpu, nq " ζp,qpuq
where ζp,qpuq are certain operators living in the matricially free Fock space, a matricial
analog of the free Fock space, into which ηpuq decompose. The convergence is in the
sense of mixed *-moments under the expectation of normalized partial traces over the
subsets of basis vectors defined by the block decomposition, denoted τqpnq " E Trqpnq,
where 1 ď q ď r. In turn, the corresponding moments of operators are computed with
respect to the family of vacuum states. A similar theorem holds for symmetric blocks
of Hermitian and non-Hermitian Gaussian random matrices.
This block model encodes more degrees of freedom than the usual framework of
square random matrices with i.i.d. Gaussian entries. Apart from block variances, the
most important parameters are the asymptotic dimensions of blocks, namely
dq " lim
nÑ8
nq
n
,
where n " n1 ` . . . ` nr is the n-dependent partition of n defined by the block decom-
position of Y pu, nq. We assume that these limits exist, but it is possible that some of
them vanish. This leads to three types of blocks or symmetric blocks: balanced (if both
their asymptotic dimensions are positive), unbalanced (if one asymptotic dimension is
positive and the other one vanishes) and evanescent (if both asymptotic dimensions
vanish). For details, see [8].
Using all types of blocks, we can construct random matrix models for other notions
of independence, such as monotone independence, boolean independence, c-freeness
(conditional freeness) and s-freeness (freeness with subordination). Moreover, the limit
moments can be written in a quite explicit form as polynomials in d1, . . . , dr, with
block variances as additional parameters. In some cases, this enables us to construct
simple random matrix models for certain families of probability measures,
like the
model for free Meixner laws [1,4] constructed in [9]. At the same time, studying such
polynomials can lead to new results in free probability. For instance, a study of the
limit distributions of products of independent rectangular Gaussian random matrices
produced polynomials which can be viewed as multivariate Fuss-Narayana polynomials
and, moreover, turned out to be the moments of the free multiplicative convolution of
Marchenko-Pastur distributions with arbitrary shape parameters [10].
In order to make this paper easy to follow, we avoid technical details and only sketch
some proofs. For complete proofs, we refer the reader to [7,8,9,10].
2. Matricial operator systems
In the framework of free probability, an important role is played by semicircular
and circular systems of operators. Such systems were introduced by Voiculescu, who
LIMIT DISTRIBUTIONS OF GAUSSIAN BLOCK ENSEMBLES
3
used them to construct the generators of free group factors in order to prove certain
isomorphisms between them [16].
These generators play also an important role in the framework of matricial freeness.
For the definition of that concept, which can be viewed as a matricial generalization of
freeness involving a family of states instead of one state, we refer the reader to [6]. In this
paper, we prefer to restrict our attention to important examples of arrays of operators
[7,8] which are matricially free with respect to a family of states associated with vacuum
vectors in the 'matricial' analog of the free Fock space, called the matricially free Fock
space.
Definition 2.1. Let pHp,qpuqq, u P U, be a family of r r arrays of Hilbert spaces.
The matricially free Fock space is the Hilbert space direct sum of the form
M "
ràq"1
Mq,
where
Mq " CΩq '
8àm"1 àp1 ,...,pmPrrs
u1 ,...,unPU
Hp1,p2pu1q b Hp2,p3pu2q b . . . b Hpm,qpumq,
with rrs :" t1, 2, . . . , ru and tΩ1, . . . , Ωru being the set of unit (vacuum) vectors,
equipped with the canonical inner product. Let tΨ1, . . . , Ψru be the corresponding
states replacing the single vacuum state in the free Fock space.
In fact, for most purposes, it suffices to take each Hilbert space to be one-dimensional,
namely Hp,qpuq " Cep,qpuq, where the vectors ep,qpuq form an orthonormal basis. Arrays
of certain partial isometries living in this Fock space which remind free creation oper-
ators ℓpuq living in the free Fock space serve as generators of certain Toeplitz-Cuntz-
Krieger algebras [5]. These partial isometries, when multiplied by positive scalars,
become matricial analogs of free creation operators.
Definition 2.2. Let Bpuq " pbp,qpuqq be an array of positive real numbers for any u P U.
We associate with each such array the array of matricially free creation operators whose
non-trivial action onto the basis vectors is
℘p,qpuqΩq " bbp,qpuqep,qpuq
℘p,qpuqpeq,tpsqq " bbp,qpuqpep,qpuq b eq,tpsqq
℘p,qpuqpeq,tpsq b wq " bbp,qpuqpep,qpuq b eq,tpsq b wq
for any p, q, t P rrs and u, s P U, where eq,tpsq b w is a basis vector. Their actions onto
the remaining basis vectors give zero. The corresponding matricially free annihilation
operators are their adjoints ℘
p,qpuq. In some cases, it will be convenient to use the same
notation even if bp,qpuq " 0, when we obtain trivial operators.
Certain linear combinations of the matricially free creation and annihilation oper-
ators are of special interest. We have studied several matricial systems of operators
constructed from the matricially free creation and annihilation operators or their sym-
metrized counterparts when describing the asymptotic distributions of blocks or sym-
metric blocks of Hermitian and non-Hermitian Gaussian random matrices [7,8]. We
4
R. LENCZEWSKI
define them below, mentioning also the corresponding ensembles of random blocks. We
also give their realizations as operator-valued matrices, which identifies some of them
with the generators of free group factors used by Voiculescu in [16].
(1) Non-trivial matricially free creation and annihilation operators can be realized
as operator-valued matrices from the C -algebra A b MnpCq, where A is the
C -algebra generated by the family tℓpp, q, uq : 1 ď p, q ď r, u P Uu of *-free
creation operators living in the free Fock space. Namely,
℘p,qpuq " ℓpp, q, uq b epp, qq
p,qpuq " ℓpp, q, uq b epq, pq,
℘
where tepp, qq : 1 ď p, q ď ru is the array of matrix units. If ϕ is the vacuum
state in this free Fock space and ψq is the vector state on MnpCq defined by the
basis vector eq of Cn, then the state Ψq can be identified with ϕ b ψq, as we
showed in [11].
(2) Matricially free Gaussian operators are self-adjoint operators of the form
ωp,qpuq " ℘p,qpuq ` ℘p,qpuq,
and they are the canonical Gaussian operators in our framework. It turns out
that they describe the limit distributions of unbalanced symmetric blocks of
Hermitian Gaussian random matrices.
(3) It is convenient to introduce the symmetrized creation operators as
p℘p,qpuq "" ℘p,qpuq ` ℘q,ppuq
℘q,qpuq
if p ă q
if p " q
and the symmetrized annihilation operators as their adjoints p℘
(4) In order to describe the limit distributions of balanced symmetric blocks of
Hermitian Gaussian random matrices, we need to use symmetrized Gaussian
operators
p,qpuq.
pωp,qpuq " p℘p,qpuq ` p℘p,qpuq,
where all operators are non-trivial. It is easy to see that if all creation and anni-
hilation operators involved are non-trivial, then the above ones can be identified
with the operator-valued matrices of the form
pωp,qpuq "" gpp, q, uq b epp, qq ` gpp, q, uq b epq, pq
spq, uq b epq, qq
if p ă q
if p " q
where tgpp, q, uq : 1 ď p ď q ď r, u P Uu is a family of circular operators and
tspq, uq : 1 ď q ď r, u P Uu is a family of semicircular operators. These matrices
are generators of free group factors introduced by Voiculescu [16].
(5) In order to describe the limit distributions of the usual (non-symmetric) blocks
of non-Hermitian Gaussian random matrices, we need to use matricial R-circular
operators [11]
ζp,qpuq " ℘p,qpu1q ` ℘q,ppu2q,
where the notation means that in order to construct one such operator labelled
by u one needs two operators, which are labelled by u1 and u2, thus the index
LIMIT DISTRIBUTIONS OF GAUSSIAN BLOCK ENSEMBLES
5
set U has to be doubled. If the added operators are non-trivial, then we can
identify the above ones with operator-valued matrices of the form
ζp,qpuq " gpp, q, uq b epp, qq
for any 1 ď p, q ď r and any u P U, where tgpp, q, uq : 1 ď p, q ď r, u P Uu
is a family of circular operators (here, the covariances of ℘p,qpuq are symmetric
and identical for u1 and u2). Sums ζpuq " řp,q ζp,qpuq are R-cyclic matrices
introduced in [13].
(6) In order to describe the limit joint distributions of the symmetric blocks of
non-Hermitian Gaussian random matrices, we need to use matricial circular
operators [8] which are natural symmetrizations of ζp,qpuq, namely
where u P U and 1 ď p ď q ď r. If the added operators are non-trivial, then
the above ones can be identified with the operator-valued matrices of the form
ηp,qpuq " p℘p,qpu1q ` p℘p,qpu2q,
ηp,qpuq "" gpp, q, uq b epp, qq ` gpq, p, uq b epq, pq
spq, uq b epq, qq
if p ă q
if p " q
where tgpp, q, uq : 1 ď p, q ď r, u P Uu is a family of circular operators and
tspq, uq : 1 ď q ď r, u P Uu is a family of semicircular operators.
3. Hermitian Gaussian Block Ensemble
In free probability, we study the mixed moments of independent n n random ma-
trices tY pu, nq : u P Uu under the expectation of the normalized trace
where
τpnq " E Trpnq
TrpnqpAq "
TrpAq
1
n
for any n n matrix A as n Ñ 8. Let us recall the asymptotic freeness result of
Voiculescu for independent Hermitian Gaussian random matrices [15].
Theorem 3.1. If we are given an ensemble of independent Hermitian n n random
matrices tY pu, nq : u P Uu, whose entries Yi,jpu, nq satisfy Yi,jpu, nq " Yj,ipu, nq, are
complex p0, 1{nq-Gaussian if i ‰ j and real p0, 1{nq-Gaussian if i " j and the family
tYi,jpu, nq : 1 ď i ď j ď nu is independent for any u, then
lim
nÑ8
Y pu, nq " ωpuq,
which should be understood as the convergence of mixed moments of matrices under
τpnq to the mixed moments of the corresponding free standard semicircular operators
ωpuq under the vacuum state Φ in the free Fock space.
Remark 3.1. Let us make some remarks which will enable us to formulate our main
results.
(1) We stated the above theorem in a simplified form, which will also be used when
we state our results on the asymptotic distributions of random blocks. A more
explicit formulation says that
lim
nÑ8
τpnqpY pu1, nq . . . Y pum, nqq " Φpωpu1q . . . ωpumqq,
6
R. LENCZEWSKI
for any u1, . . . , um P U, where tωpuq : u P Uu is a standard free semicircular
family, which means in particular that Φpωpuq2q " 1 for any u. Alternatively,
one could say that the family tY pu, nq : u P Uu is asymptotically free under
τpnq and that the limit distribution of each Y pu, nq is the standard semicircular
Wigner distribution Wp0, 1q with density p2πq´1?4 ´ x2 on r´2, 2s.
(2) In order to define a family of partial traces, take the decomposition of the set
rns :" t1, 2, . . . , nu into r disjoint intervals
and denote by nq the cardinality of Nq. Next, let
rns :" N1 Y . . . Y Nr
Ipnq " D1 ` . . . ` Dq
be the corresponding decomposition of the n n identity matrix Ipnq, that is
pDkqi,j " 1 whenever i " j P Nk, with the remaining entries equal to zero, where
1 ď k ď r. Of course, the objects Nq, nq, Dq depend on n, but this is supressed
in the notation.
(3) By partial traces we then understand states of the form
where
τqpnq " E Trqpnq
TrqpnqpAq "
1
nq
TrpDqADqq
and 1 ď q ď r.
of the form
(4) By random blocks of a random matrix Y pu, nq we shall understand nn matrices
for any 1 ď p, q ď r and u P U, where r, n P N.
Hermitian, then Sq,ppu, nq " Sp,qpu, nq. Clearly, we have the decomposition
In particular, if Y pu, nq is
Sp,qpu, nq " DpY pu, nqDq,
Y pu, nq " ÿ1ďp,qďr
Sp,qpu, nq
for any u, n.
(5) We assume that all variables Yi,jpu, nq which belong to the same block Sp,qpu, nq
have the same covariance EpYi,jpu, nqYi,jpu, nqq " vp,qpuq{n whenever pi, jq P
Np Nq. We denote by V puq " pvp,qpuqq the corresponding matrices of block
covariances. Apart from the dimension matrix
D " diagpd1, . . . , drq,
these matrices are additional parameters of the ensemble.
distributions will be expressed in terms of matrices Bpuq " DV puq.
n n matrices of the form
(6) By random symmetric blocks of a random matrix Y pu, nq we shall understand
In fact, the limit
for any 1 ď p ď q ď r and u P U, where r, n P N. We have the decomposition
Tp,qpu, nq "" Sp,qpu, nq ` Sq,ppu, nq
if p ă q
if p " q
Sq,qpu, nq
Y pu, nq " ÿ1ďpďqďr
Tp,qpu, nq
LIMIT DISTRIBUTIONS OF GAUSSIAN BLOCK ENSEMBLES
7
(7) When speaking of limit distributions of mixed moments of symmetric blocks, we
for any u, n. Clearly, if Y pu, nq is Hermitian, then Tp,qpu, nq " Tp,qpu, nq.
will use a simplified formulation, similar to that in the free case. For instance
in the Hermitian case will mean that
lim
nÑ8
Tp,qpu, nq " pωp,qpuq
lim
nÑ8
τqpnqpTp1,q1pu1, nq . . . Tpm,qmpum, nqq
" Ψqppωp1,q1pu1q . . .pωpm,qmpumqq
for any 1 ď p1 ď q1 ď r, . . . , 1 ď pm ď qm ď r, 1 ď q ď r and u1, . . . , um P U.
The operators will always belong to one of the families defined in Section 2. A
similar formulation will be used for blocks Sp,qpu, nq.
We can describe limit joint distributions of blocks and symmetric blocks of Hermitian
random matrices under rather general assumptions. It suffices to assume that the family
tY pu, nq : u P Uu is asymptotically free and asymptotically free from tD1, . . . , Dru under
τpnq and that their norms are uniformly boubded almost surely. This class includes
unitarily invariant random matrices whose limit moments are compactly supported
probability measures on the real line. This general version was proved in [8, Theorem
6.1]. For the sake of simplicity, we restrict our attention here to the case of Hermitian
Gaussian random matrices.
Theorem 3.2. If we are given an ensemble tY pu, nq : u P Uu of independent Hermitian
nn random matrices whose entries Yi,jpu, nq satisfy Yi,jpu, nq " Yj,ipu, nq, are complex
p0, vp,qpuq{nq-Gaussian if i ‰ j and pi, jq P Np Nq, and real p0, vq,qpuq{nq-Gaussian if
i " j P Nq and the family tYi,jpu, nq : 1 ď i ď j ď nu is independent for any u, then
lim
nÑ8
Tp,qpu, nq " pωp,qpuq
for any p ď q and u P U, where convergence is in the sense of mixed moments under
partial traces and the arrays ppωp,qpuqq are associated with the symmetric covariance
matrices Bpuq " DV puq, respectively.
Sketch of the proof. First, let us assume that vp,qpuq " 1 for any p, q, u. In that case,
we can use asymptotic freeness of the family tY pu, nq : u P Uu and its asymptotic free-
ness with respect to the family tD1, . . . , Dru (these are deterministic diagonal matrices)
to describe the limit joint distributions of the blocks Sp,qpu, nq under τpnq since
Sp,qpu, nq " DpY pu, nqDq
for any p, q, u. Moreover, we know that the limit distribution of each Y pu, nq is the
standard semicircle Wigner law Wp0, 1q and a direct computation gives τpnqpDqq "
nq{n Ñ dq for any q. Now, it is not hard to show that the family of operators
ωpuq :"ÿp,q
ωp,qpuq,
where ωp,qpuq " ℘p,qpuq ` ℘p,qpuq for any u and ℘p,qpuq has covariance dp, is free with
respect to Ψ " řq dqΨq. We abuse the notation a little since by ωpuq we denoted a
standard semicircular operator on the free Fock space, but this is justified by the fact
that each ωpuq has the standard semicircle distribution under Ψ. Therefore, we have in
fact a realization of the standard free semicircular family tωpuq : u P Uu as operators
8
R. LENCZEWSKI
living in the matricially free Fock space M. Now, it suffices to find an appropiate
limit realization for tD1, . . . , Dru. We have shown in [8] that it is given by the family
tP1, . . . , Pru, where
using the tensor product realization described in Section 2, since the family tωpuq : u P
Uu is free from tP1, . . . , Pru and ΨpPqq " dq. Therefore, any limit mixed *-moment
of the random blocks Sp,qpuq under τpnq can be expressed as a mixed *-moment of
the corresponding operators Pp ωpuqPq under Ψ. Symbolically, since limnÑ8 Y pu, nq "
ωpuq, we have
Pq " 1 b epq, qq,
lim
nÑ8
Sp,qpu, nq " Pp ωpuqPq " ℘p,qpuq ` ℘q,ppuq :" ςp,qpuq
except that the moments of blocks are computed under τpnq and those of the operators
Pp ωpuqPq are computed under Ψ. In order to pass from τpnq to τqpnq, notice that
τqpnqpAq "
τpnqpDqADqq
n
nq
so the limit mixed *-moments of random blocks under τqpnq are given by the mixed
*-moments of the above operators under Ψq, where 1 ď q ď r. All this holds provided
dq ‰ 0. The case when dq " 0 is slightly more complicated and is omitted here (we
refer the reader to [8]). It easily follows that
lim
nÑ8
Tp,qpu, nq " pωp,qpuq
for any 1 ď p ď q ď r and u P U. It can also be seen that one can rescale blocks
Sp,qpu, nq, which means that one can rescale the block covariances and take arbitrary
non-negative vp,qpuq (except that we must have the symmetry vp,qpuq " vq,ppuq since
Y pu, nq is Hermitian). This proves that in the case when the covariances are equal to
vp,qpuq within block Sp,qpuq, the limit operator ςp,qpuq gets rescaled by vp,qpuq and thus
the covariance of ℘p,qpuq becomes bp,qpuq " dpvp,qpuq. This completes the proof.
Corollary 3.1. In particular,
(cid:4)
(1) if dp " 1 and dq " 0, then limnÑ8 Tp,qpu, nq " ωp,qpuq,
(2) if dp " 0 and dq " 1, then limnÑ8 Tp,qpu, nq " ωq,ppuq,
(3) if dp " 0 and dq " 0, then limnÑ8 Tp,qpu, nq " 0.
Proof. If dp " 1 and dq " 0, then ωq,ppuq " 0 since bq,ppuq " dqvq,ppuq " 0 and thus
pωp,qpuq reduces to ωp,qpuq. In turn, if Tp,qpu, nq is evanescent, then ωp,qpuq " ωq,ppuq " 0
and thus pωp,qpuq " 0, which completes the proof.
(cid:4).
Remark 3.2. It should be remarked that the matricially free Gaussian operators are
not operatorial realizations of the usual (non-symmetric) blocks Sp,qpuq of Gaussian
Hermitian matrices. In fact, it follows from the proof of Theorem 3.2 that
Sp,qpuq "" gpp, q, uq b epp, qq
spq, uq b epq, qq
lim
nÑ8
if p ‰ q
if p " q
for any 1 ď p, q ď r and u P U, where convergence is understood in the usual sense
(mixed moments of blocks under partial traces τqpnq converge to mixed moments of
the corresponding operators under Ψq). Here, tgpp, q, uq : 1 ď p ď q ď r, u P Uu is a
family of circular operators and tspq, uq : 1 ď q ď r, u P Uu is a family of semicircular
operators and we assume that gpq, p, uq " gpp, q, uq for p ă q.
LIMIT DISTRIBUTIONS OF GAUSSIAN BLOCK ENSEMBLES
9
Example 3.1. Fix u P U (omitted in our notations) and p ‰ q. Let gpp, qq " ℓ1 ` ℓ
and thus gpq, pq " ℓ
1 ` ℓ2 for some *-free creation operators ℓ1, ℓ2. In view of Remark
3.2, we get
2
lim
nÑ8
τqpnqpSq,ppnqSp,qpnqSq,ppnqSp,qpnqq " Ψqpςq,pςp,qςq,pςp,qq
1 ` ℓ2qpℓ1 ` ℓ
p ` dpdqqv2
2ℓ2q " Ψpp℘
" ϕppℓ
" pd2
p,q℘p,qq " dpvp,q and ϕpℓ
p,q
where ϕpℓ
1ℓ1q " Ψqp℘
the symmetry vp,q " vq,p.
2qpℓ
1 ` ℓ2qpℓ1 ` ℓ
2qq
q,p℘q,pq " dpvq,p and we use
4. Ginibre Block Ensemble
Blocks of non-Hermitian Gaussian random matrices can be treated in a similar way.
The associated ensmble of blocks cna be called the Ginibre Block Ensemble. The main
difference is that the limit joint distributions of blocks are described by matricial R-
circular operators ζp,qpuq and those of the symmetric blocks by matricial circular oper-
ators ηp,qpuq.
Let us recall Voiculescu's theorem on the asymptotic freeness of the ensemble of
independent Gaussian random matrices with i.i.d. entries (Ginibre Ensemble) [15].
Theorem 4.1. If we are given an ensemble of independent n n random matrices
tY pu, nq : u P Uu, whose entries Yi,jpu, nq are complex p0, 1{nq-Gaussian for any i, j
and the family tYi,jpu, nq : 1 ď i, j ď nu is independent, then
lim
nÑ8
Y pu, nq " ηpuq,
where convergence is in the sense of mixed moments of matrices under τpnq to the mixed
moments of the corresponding free standard circular operators ηpuq under the vacuum
state Φ in the free Fock space.
Let us now formulate an analogous theorem for blocks and symmetric blocks of
Gaussian random matrices with i.b.i.d. entries.
Theorem 4.2. If we are given an ensemble of independent n n random matrices
tY pu, nq : u P Uu, whose entries Yi,jpu, nq are complex p0, vp,qpuq{nq-Gaussian for any
pi, jq P Np Nq and the family tYi,jpu, nq : 1 ď i, j ď nu is independent, then
lim
nÑ8
Sp,qpu, nq " ζp,qpuq,
where convergence is in the sense of mixed moments of blocks under partial traces τqpnq
to the mixed moments of the corresponding matricial systems of operators under the
vacuum states Ψq, respectively, in the matricially free Fock space.
Sketch of the proof. The proof is similar to that of Theorem 3.1. In the case when all
variables are i.i.d., we can use the asymptotic *-freeness of the family tY pu, nq : u P Uu
under τpnq as n Ñ 8 as well as their asymptotic *-freeness from the family of diagonal
matrices. One can check that the family tηpuq : u P Uu, where
gpp, q, uq b epp, qq,
ηpuq "ÿp,q
ζp,qpuq "ÿp,q
10
R. LENCZEWSKI
is *-free under Ψ as well as *-free with respect to tP1, . . . , Pru under Ψ, where again
Pq " 1 b epq, qq. Moreover, each ηpuq has the standard circular distribution under
Ψ "řq Ψq. Since the asymptotic joint distribution of tD1, . . . , Dru under τpnq agrees
with that of tP1, . . . , Pru under Ψ, we must have
lim
nÑ8
Sp,qpu, nq " lim
nÑ8
DpY pu, nqDq " Pp ηpuqPq " ζp,qpuq
where convergence is understood as described in Remark 3.1, which completes the proof.
(cid:4)
Corollary 4.1. Under the assumptions of Theorem 4.2, it holds that
for any p, q, u
lim
nÑ8
Tp,qpu, nq " ηp,qpuq
Proof. This is an easy consequence of Theorem 4.2.
(cid:4)
Example 4.1. Computations of limit mixed (*-) moments of blocks reduce to the com-
putation of moments involving matricial R-circular systems of operators. For instance,
lim
nÑ8
p,qζ
τqpnqpSp,qpnqSq,ppnqSq,ppnqSp,qpnqq " Ψqpζ
" ϕpℓ
1ℓ1qϕpℓ
" dpdqvp,qvq,p
3ℓ3q
q,pζq,pζp,qq
where p ‰ q, since
ζp,q " pℓ1 ` ℓ
2q b epp, qq,
ζq,p " pℓ3 ` ℓ
4q b epq, pq
where tℓ1, ℓ2, ℓ3, ℓ4u is a *-free system of free creation operators with covariances ϕpℓ
dpvp,q and ϕpℓ
ators can be written in the form ℓ1 ` ℓ
1ℓ1q "
3ℓ3q " dqvq,p. In fact, it is well known that any pair of free circular oper-
2 and ℓ3 ` ℓ
4, respectively.
5. Products of independent Gaussian random matrices
The first concrete application of our method concerns products of independent rect-
angular Gaussian random matrices [10]. For any given p P N and any n P N, consider
the product of independent rectangular Gaussian random matrices
Bpnq " X1pnqX2pnq . . . Xppnq,
where n P N and all entries of these matrices are assumed to be independent p0, 1{nq-
If Xjpnq is an nj´1 nj matrix for any 1 ď j ď p, we assume
Gaussian variables.
that
lim
nÑ8
nj
n " dj
for any j P t0, . . . , pu (it is convenient to start with n0 rather than with n1). Let τ0pnq
be the trace over the set of first n0 basis vectors composed with classical expectation.
Theorem 5.1. Under the above assumptions, it holds that
lim
nÑ8
where
Pkpd0, d1, . . . , dpq "
τ0pnq´pBpnqBpnqqk¯ " Pkpd0, d1, . . . , dpq,
jp dj0´1
j1 . . . k
j0 k
k k
ÿj0`...`jp"pk`1
1
0
dj1
1 . . . djp
p ,
LIMIT DISTRIBUTIONS OF GAUSSIAN BLOCK ENSEMBLES
11
for any natural k and j0, j1, . . . , jp. These polynomials are called multivariate Fuss-
Narayana polynomials.
Sketch of the proof. The proof is based on embedding the matrices X1pnq, . . . , Xppnq
in symmetric blocks T1,2pnq, . . . , Tp,p`1pnq, respectively, of a large square Gaussian ran-
dom matrix Y pnq of dimension NN, where N " n1`. . .`np`1, with p0, 1{nq-Gaussian
entries for any n (we use only one matrix and thus we omit u in our notations). Com-
puting the moments of BpnqBpnq under τ0pnq becomes now the normalized partial
trace over the first subset of basis vectors. We then use Corollary 4.1 to realize the
limit moments in terms of η1,2, . . . , ηp,p`1 and their adjoints. These limit moments can
be computed explicitly, which was done in [10], which completes the proof.
(cid:4)
Let us make some additional remarks on the above result:
(1) The special case of p " 1 corresponds to Wishart matrices. If we set d0 " 1 and
d1 " t, we obtain
Pkp1, tq " ÿi`j"k`1
1
ik
kk
jtj,
the moments of the Marchenko-Pastur distribution with shape parameter t ą 0,
namely
t " maxt1 ´ t, 0uδ0 ` apx ´ aqpb ´ xq
2πx
11ra,bspxqdx
where a " p1 ´ ?tq2 and b " p1 ` ?tq2.
(2) The moments of the Marchenko-Pastur distributions are known to have the form
of Narayana polynomials
Nkptq "
kÿj"1
1
j ´ 1 k
jk ´ 1
j ´ 1tj
for any k P N. It is easy to show that Pkp1, tq " Nkptq, but our formula is more
suitable for multivariate generalizations.
(3) If d0 " 1, the multivariate Fuss-Narayana polynomials become the moments of
the p-fold free multiplicative convolution of the Marchenko-Pastur distributions
with arbitrary shape parameters. Namely,
Pkp1, t1, . . . , tpq " mkpt1 b t2 b . . . b tpq
where mkpµq stands for the kth moment of µ and b stands for the free multi-
plicative convolution.
(4) A less general class of polynomials was studied in the context of free Bessel laws
πp, t " bpp´1q
1
b t,
where p P N, defined by Banica, Belinschi, Capitaine and Collins [2]. The
moments of free Bessel laws are polynomials in one variable t,
Qkptq "
kÿj"1
1
j ´ 1 pk
jk ´ 1
j ´ 1tj
12
R. LENCZEWSKI
called Fuss-Narayana polynomials. Our polynomials are natural multivariate
generalizations of these polynomials. Clearly, Qkptq " Pkp1, . . . , 1, tq, where 1
appears p times.
6. New random matrix models
The results on the asymptotics of random blocks can be applied to the construction
In this paper, we choose
of some new random matrix models, as we showed in [9].
to present a simple version of such a model for monotone independence [12].
In a
similar way, one can construct random matrix models for boolean independence an
s-free independence (freeness with subordination). In fact, more general versions, not
restricted to Gaussian random matrices have been constructed in [8].
Monotone independence will appear in the study of the asymptotic joint distributions
of two independent Hermitian Gaussian random matrices of the same block form. Our
assumptions are the following:
(A1) We have a family of independent Gaussian random matrices
Cpu, nq Dpu, nq
Y pu, nq " Apu, nq Bpu, nq
where u P U, and blocks pApu, nqq are evanescent, symmetric blocks built from
pBpu, nqq and pCpu, nqq are unbalanced, and blocks pDpu, nqq are balanced,
(A2) the matrices are Hermitian, thus the off-diagonal blocks are Hermitian conjugate
and the diagonal blocks are Hermitian
(A3) the complex Gaussian variables Yi,jpu, nq have zero mean and have identical
covariances within blocks, namely
EpYi,jpu, nqYi,jpu, nqq "
vp,qpuq
n
(A4) the decomposition of the identity matrix corresponding to the block decompo-
whenever the pair pi, jq belongs to the block indexed by pp, qq,
sition is given by
for any n P N.
Ipnq " D1 ` D2
Theorem 6.1. Under assumptions (A1)-(A4), if U " t1, 2u and vp,qpuq " 1 for any
p, q, u, the pair tBp1, nq`Cp1, nq, Y p2, nqu is asymptotically monotone independent with
respect to the partial trace τ1pnq.
Proof. By Theorem 3.2, the proof reduces to showing that the pair
tω2,1p1q, ω2,1p2q ` ω2,2p2qu
is monotone independent with respect to Ψ1 since, by assumption, the asymptotic
dimensions are d1 " 0 and d2 " 1, which means that the remaining operators can be
neglected. Denote a " ω2,1p1q and b " ω2,1p2q ` ω2,2p2q. We need to show that
Ψ1pw1a1b1a2w2q " Ψ1pb1qΨ1pw1a1a2w2q
for any a1, a2 P Cra, 11s, b1 P Crb, 12s, where 11 " 12,1 and 12 " 1 and w1, w2 are
arbitrary elements of Cxa, b, 11, 12y. It suffices to consider the action of a and b onto
LIMIT DISTRIBUTIONS OF GAUSSIAN BLOCK ENSEMBLES
13
their invariant subspace in M of the form
M1 " CΩ1 ' pFp2q b Hp1qq ' pFp2q b Hp2qq
where Fp2q " FpCe2,2p2qq with the vacuum vector Ω and Hpuq " Ce2,1puq for u P t1, 2u,
where we identify Ω b e2,1puq with e2,1puq. Now, the range of any polynomial in a is
CΩ1 ' Hp1q since
akΩ1 "" Ω1
e2,1p1q
if k is even
if k is odd
and 11Ω1 " Ω1, 11e2,1p1q " e2,1p1q. Therefore, it suffices to compute the action of any
polynomial in b onto Ω1 and e2,1p1q. Now, the action of powers of b onto Ω1 and onto
e2,1p1q is the same as the action of the free Gaussian operator onto the vacuum vector
in the free Fock space. Namely, we have
b2kΩ1 " CkΩ1 mod pM1 a CΩ1q
and
where Ck is the kth Catalan number and
b2ke2,1p1q " Cke2,1p1q mod pM1 a pCΩ1 ' Hp1qq
b2k´1Ω1 " b2k´1e2,1p1q " 0 mod pM1 a pCΩ1 ' Hp1qq
for any k P N. Thus Ψ1pb2kq " Ck and, moreover, since M1 a pCΩ1 ' Hp1qq Ă Kera,
the required condition for monotone independence holds if b1 is a positive power of b.
It is easy to see that it also holds if b1 " 12, which completes the proof.
(cid:4)
Interestingly enough, unbalanced symmetric blocks were also used in the construction
of a simple random matrix model for free Meixner laws [1,4]. These are probability
measures on the real line associated with the sequences of Jacobi parameters of the
form
pα1, α2, α2, . . .q and pβ1, β2, β2, . . .q
and thus we can asy that they are associated with quadruples pα1, α2, β1, β2q. Let us
formulate the theorem in the most interesting case when β1 and β2 are positive.
Theorem 6.2. Under assumptions (A1)-(A4), let β1puq " v2,1puq ą 0 and β2puq "
v2,2 ą 0, for any u P U. Then
(1) the asymptotic distributions of the matrices
Mpu, nq " Y pu, nq ` α1puqD1pnq ` α2puqD2pnq
under the partial trace τ1pnq are the free Meixner distributions associated with
the parameters pα1puq, α2puq, β1puq, β2puqq, respectively,
(2) the family tMpu, nq : u P Uu is asymptotically conditionally free with respect to
the pair of partial traces pτ1pnq, τ2pnqq as n Ñ 8.
Sketch of the proof. It follows from Theorem 3.2 that
lim
nÑ8
Y pu, nq " ω2,1puq ` ω2,2puq
where the variances of ω2,1puq and ω2,2puq are b2,1puq " d2v2,1puq " β1puq and b2,2puq "
d2v2,2puq " β2puq since d1 " 0 and d2 " 1. The fact that d1 " 0 is the reason why
14
R. LENCZEWSKI
ω1,1puq and ω1,2puq become trivial operators and that is why they do not show up in
the limit realization. Therefore,
lim
nÑ8
Mpu, nq " γpuq :" ω2,1puq ` ω2,2puq ` α1puqP1 ` α2puqP2,
where P1 and P2 are as in the proof of Theorem 3.2. This implies, in particular, that the
asymptotic distribution of Mpu, nq under τ1pnq agrees with that of γpuq under Ψ1. The
proof that the moments of γpuq are the moments of the free Meixner law associated with
the parameters pα1puq, α2puq, β1puq, β2puqq is purely combinatorial and can be found in
[9]. We also refer the reader to [9] for the proof of asymptotic conditional independence
of the family of such matrices.
(cid:4).
References
[1] M. Anshelevich, Free martingale polynomials, J. Funct. Anal. 201(2003), 228-261.
[2] T. Banica, S.T. Belinschi, M. Capitaine, B. Collins, Free Bessel laws, Canad. J. Math. 63 (2011),
3-37.
[3] F. Benaych-Georges, Rectangular random matrices, related convolution, Probab. Theory Relat.
Fields 144 (2009), 471-515.
[4] M. Bozejko, W. Bryc, On a class of free L´evy laws related to a regression problem, J. Funct. Anal.
236 (2006), 59-77.
[5] J. Cuntz, W. Krieger, A class of C -algebras and topological Markov chains, Invent. Math. 56
(1980), 251-268.
[6] R. Lenczewski, Matricially free random variables, J. Funct. Anal. 258 (2010), 4075-4121.
[7] R. Lenczewski, Asymptotic properties of random matrices and pseudomatrices, Adv. Math. 228
(2011), 2403-2440.
[8] R. Lenczewski, Limit distributions of random matrices, Adv. Math. 263 (2014), 253-320.
[9] R. Lenczewski, Random matrix model for free Meixner laws, Int. Math. Res. Notices (2014),
rnu041, 26 pages.
[10] R. Lenczewski, R. Sa lapata, Multivariate Fuss-Narayana polynomials with appplication to random
matrices, Electron. J. Combin. 20 (2013), Issue 2 (electronic).
[11] R. Lenczewski, Matricial R-circular systems and random matrices, arXiv:1311.6420 (2014).
[12] N. Muraki, Monotonic independence, monotonic central limit theorem and monotonic law of small
numbers, Infin. Dimens. Anal. Quantum Probab. Relat. Top. 4 (2001), 39-58.
[13] A. Nica, D. Shlyakhtenko, R. Speicher, R-cyclic families of matrices in free probability, J. Funct.
Anal. 188 (2002), 227-271.
[14] D. Shlyakhtenko, Random Gaussian band matrices and freeness with amalgamation, Int. Math.
Res. Notices 20 (1996), 1013-1025.
[15] D. Voiculescu, Limit laws for random matrices and free products, Invent. Math. 104 (1991),
201-220.
[16] D. Voiculescu, Circular and semicircular systems and free product factors, Progress in Math. 92,
Birkhauser, 1990.
Romuald Lenczewski,
Instytut Matematyki i Informatyki, Politechnika Wroc lawska,
Wybrze ze Wyspia´nskiego 27, 50-370 Wroc law, Poland
E-mail address: [email protected]
|
1305.1920 | 1 | 1305 | 2013-05-08T19:30:19 | Freely Independent Random Variables with Non-Atomic Distributions | [
"math.OA"
] | We examine the distributions of non-commutative polynomials of non-atomic, freely independent random variables. In particular, we obtain an analogue of the Strong Atiyah Conjecture for free groups thus proving that the measure of each atom of any $n \times n$ matricial polynomial of non-atomic, freely independent random variables is an integer multiple of $n^{-1}$. In addition, we show that the Cauchy transform of the distribution of any matricial polynomial of freely independent semicircular variables is algebraic and thus the polynomial has a distribution that is real-analytic except at a finite number of points. | math.OA | math |
FREELY INDEPENDENT RANDOM VARIABLES WITH
NON-ATOMIC DISTRIBUTIONS
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
Abstract. We examine the distributions of non-commutative polynomials of
non-atomic, freely independent random variables.
In particular, we obtain
an analogue of the Strong Atiyah Conjecture for free groups thus proving
that the measure of each atom of any n × n matricial polynomial of non-
atomic, freely independent random variables is an integer multiple of n−1.
In addition, we show that the Cauchy transform of the distribution of any
matricial polynomial of freely independent semicircular variables is algebraic
and thus the polynomial has a distribution that is real-analytic except at a
finite number of points.
1. Introduction
One of the essential themes in the study of free probability [23] and its applica-
tions to random matrix theory is to determine specific properties of the spectral dis-
tribution of a fixed (matricial) polynomial in freely independent random variables.
For example, some of the earliest work in free probability theory concerns free con-
volution, which is the study of the distribution of the polynomial P (X, Y ) = X + Y
in two freely independent random variables.
In particular, the recent paper [3]
of Belinschi, Mai, and Speicher uses an analytic theory for operator-valued addi-
tive free convolution and Anderson's self-adjoint linearization trick to provide an
algorithm for determining distributions of arbitrary polynomials. Combining the
previously known results from [12], [7], [1], and [15] along with the results con-
tained in this paper, we obtain the following summary of the known properties of
distributions of matrices whose entries are polynomials in several free variables (or,
equivalently, polynomials in free variables having matricial coefficients).
Theorem 1.1. Let X1, . . . , Xn be normal, freely independent random variables
and let [pi,j] be an ℓ × ℓ matrix whose entries are non-commuting polynomials in n
variables and their adjoints such that [pi,j(X1, . . . , Xn)] is normal. Then
(1) if there exists {dj}n
j=1 ⊆ N such that the measure of each atom in the prob-
ability distribution of Xj is an integer multiple of 1
, then the measure of
dj
each atom in the probability distribution of [pi,j(X1, . . . , Xn)] is an integer
multiple of 1
In particular,
dℓ where d :=Qn
j=1 dj.
Date: June 19, 2018.
2010 Mathematics Subject Classification. 46L54.
Key words and phrases. Freely independent random variables, non-atomic distributions,
Atiyah Property for tracial ∗-algebras, free entropy, semicircular variables.
This research was supported in part by NSF grants DMS-090076, DMS-1161411 and by NSERC
PGS.
1
2
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
(2) if the probability distribution of each Xj is non-atomic, then the measure of
each atom in the probability distribution of [pi,j(X1, . . . , Xn)] is an integer
multiple of 1
ℓ .
If, in addition, X1, . . . , Xn are freely independent semicircular variables or freely
independent Haar unitaries and [pi,j(X1, . . . , Xn)] is self-adjoint, then
(3) the spectrum of [pi,j(X1, . . . , Xn)] is a union of at most ℓ disjoint sets each
of which is either a closed interval or a point, and
(4) the measure of each connected subset of the spectrum of [pi,j(X1, . . . , Xn)]
is a multiple of 1
ℓ .
Furthermore, if µ is the spectral distribution of [pi,j(X1, . . . , Xn)], if K is the sup-
port of µ, and if Gµ is the Cauchy transform of µ, then
(5) Gµ is an algebraic formal power series and thus
(6) there exists a finite subset A of R such that if I is a connected component of
R \ A and µI is the restriction of µ to I, then µI = 0 whenever I \ K 6= ∅
and if I ⊆ K, then µI has probability density function Im(g)I where g is
an analytic function defined on
W := {z ∈ C Im(z) < δ} \ [a∈A
{a − it t ∈ [0, ∞)}
for some δ > 0 such that g agrees with Gµ on {z ∈ C 0 < Im(z) < δ} and
for each a ∈ A there exists an N ∈ N and an ǫ > 0 such that (z − a)N g(z)
admits an expansion on W ∩ {z ∈ C z − a < ǫ} as a convergent power
series in rN (z − a) where rN (z) is the analytic N th-root of z defined with
branch C \ {−it t ∈ [0, ∞)}.
Finally, if the support of µ is contained in [0, ∞), then
(7) limǫ→0R 1
ǫ ln(t) dµ(t) > −∞.
In this theorem, by a polynomial in X1, . . . , Xn we mean a fixed element of the
∗-algebra generated by X1, . . . , Xn.
Parts (3) and (4) of Theorem 1.1 follow directly from [12, Corollary 3.2] which
computes the K-groups of C∗
red(Fn), the reduced group C∗-algebras of the free
groups. The characterization of the K0-group immediate implies that the normal-
ized trace of any projection in Mℓ(C∗
red(Fn)) is an integer multiple of ℓ−1. Notice
that part (4) of Theorem 1.1 does not imply part (2) of Theorem 1.1 in the setting of
part (4) as atoms may occur inside a closed interval of the spectrum. Alternatively,
these results were obtained using random matrix techniques in [7].
Notice that part (2) of Theorem 1.1 applies when X1, . . . , Xn are freely indepen-
dent semicircular variables. Since freely independent semicircular variables describe
the non-commutative law of certain independent large random matrices (see [23])
we obtain the following application to random matrix theory.
For each N ∈ N let X1(N ), . . . , Xn(N ) be self-adjoint Gaussian random matri-
ces (or, more generally, matrices with independent, identically distributed entries
satisfying certain moment conditions; see [23] or [8] for details) and let p be an
arbitrary non-constant non-commutative polynomial in n variables which is self-
adjoint in the sense that Y (N ) = p(X1(N ), . . . , Xn(N )) is always a self-adjoint
matrix. Let µN be the empirical spectral measure of Y (N ) (that is, µN [a, b] is the
average proportion of eigenvalues of Y (N ) which lie in [a, b]).
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 3
Corollary 1.2. With the notation as above, the measures µN converge to a non-
atomic limiting measure µ.
Indeed, by a result of Voiculescu (see [23] or [8]), it is known that µN converges
weakly to a measure µ that is the law of p(X1, . . . , Xn) where X1, . . . , Xn are freely
independent semicircular variables. Thus part (2) of Theorem 1.1 implies that µ
has no atoms provided p is non-constant.
The motivation for the proof of Theorem 1.1 part (2) stems from the knowledge
that the statement of the theorem holds by the Strong Atiyah Conjecture for the
free groups in the case when X1, . . . , Xn are freely independent Haar unitaries.
The Strong Atiyah Conjecture (motivated by the work in [2] and proved for a class
of groups that includes free groups by Linnell in [9]; also see [10] and references
therein) states that the kernel projection of an arbitrary matrix with entries taken
from the group ring CFn of a free group on n generators must have integer von Neu-
mann trace. To prove our theorem, we consider the analogue of the Strong Atiyah
Conjecture for ∗-subalgebras of a tracial von Neumann algebra. We call this notion
the Strong Atiyah Property (since it is known that the Strong Atiyah Conjecture
does not hold even for arbitrary group algebras; see [6] or [10] for example). It is
not hard to see that the Strong Atiyah Property holds for ∗-algebras generated by
a single normal element with non-atomic spectral measure. Our main result states
that the Strong Atiyah Property for ∗-algebras is stable under taking free products
(in the sense of free probability theory [23]) with the group algebra of a free group.
Our proof closely follows [16] with the main difference of being adapted for free
products of algebras and not groups. Using this result, we are able to conclude
that the Strong Atiyah Property holds for any ∗-algebra generated by X1, . . . , Xn
provided that Xj are free and each has a non-atomic distribution.
The proof that part (5) of Theorem 1.1 is true in the case X1, . . . , Xn are freely
independent Haar unitaries is contained in the proof of [15, Theorem 3.6].
In
Section 5 we will adapt the proof of [15, Theorem 3.6] to the semicircular case
(see Theorem 5.4). The main idea of the proof is to use the fact that if a certain
tracial map on formal power series in a single variable with coefficients in a tracial
∗-algebra A maps rational formal power series to algebraic formal power series,
then the Cauchy transform of a measure associated to a self-adjoint element of A is
algebraic (see Lemma 5.7). The proof that the tracial map is as desired in the case
A is generated by semicircular variables follows from demonstrating that a specific
formal power series in non-commuting variables is algebraic via a specific property
of the semicircular variables (see Lemma 5.12).
It is an interesting question whether the Cauchy transform of any polynomial in
freely independent random variable X1, . . . , Xn is algebraic provided the Cauchy
transform of each Xj is algebraic.
The question of whether the Cauchy transform of a measure is an algebraic
power series as in part (5) of Theorem 1.1 has previously been studied in particular
cases. For example [13, Example 3.8] demonstrates that the Cauchy transform of
the quarter-circular distribution is not algebraic. Furthermore [13, Corollary 9.5]
demonstrates that if µ and ν are compactly supported probability measures on R
which have algebraic Cauchy transforms and are the weak limits of the empirical
spectral measures of N × N random matrices, then the free additive convolution
µ ⊞ ν (see [19]) is algebraic. Moreover, [13, Corollary 9.6] demonstrates that if,
in addition, µ and ν have support contained in the positive real axis, then the
4
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
free multiplicative convolution µ ⊠ ν (see [20]) is algebraic. This question was also
considered in [1] for limit laws of certain random matrices. In fact a result much
like ours was hinted at in that paper. Using [1, Theorem 2.9] we see that part (6)
of Theorem 1.1 is implied by part (5) of Theorem 1.1. In particular, part (6) of
Theorem 1.1 directly provides information about the probability density function
of µ by the Stietjes inversion formula.
Finally, in Section 5, we will prove part (7) of Theorem 1.1 by following the
proof of [15, Theorem 3.6] which demonstrates that if the Cauchy transform of a
measure is algebraic, then the Novikov-Shubin invariants of the measure are non-
zero. Our interest in part (7) of Theorem 1.1 comes from the following question:
if p is an arbitrary, non-constant, self-adjoint polynomial in n free semicircular
variables, must it be the case that the free entropy (as defined in [21]) of p is finite?
Indeed elementary arguments may be used to show that if S is a semicircular
variable and p is a non-constant polynomial such that p(S) is self-adjoint, then
the spectral measure of p(S) has finite free entropy. Further evidence that this
must be true comes from a strengthened version of part (2) of Theorem 1.1 for
matrices of the form [pi,j] where each pi,j ∈ Alg(S1, . . . , Sn) ⊗ Alg(S1, . . . , Sn),
which we prove below. In particular, it follows that the vector of non-commutative
difference quotients JP := [∂1P, . . . , ∂nP ] (see [22]) has maximal rank whenever P
is a non-constant, non-commutative polynomial in n free semicircular variables.
Given the success of [3] in providing an algorithm for determining the distri-
butions of (matricial) polynomials in semicircular variables, it would also be of
interest if an alternate proof of Theorem 1.1 could be constructed using the ideas
and techniques from [3].
2. The Atiyah Property for Tracial ∗-Algebras
In this section we will introduce the notion of the Atiyah Property for tracial
∗-algebra. In addition, several examples of tracial ∗-algebras that satisfy the Atiyah
Property, which will be of use in Section 3, will be provided.
If ℓ ∈ N and τ is a linear functional on an algebra A, then τℓ will denote the
linear functional on Mℓ(A) given by
τℓ([Ai,j]) =
τ (Ai,i)
ℓ
Xi=1
for all [Ai,j] ∈ Mℓ(A). Notice that if τ is tracial (that is, τ (AB) = τ (BA) for all
A, B ∈ A), then τℓ is tracial.
Definition 2.1. Let A be a ∗-subalgebra of B(H), let τ be a vector state that is
tracial on A, and let Γ be an additive subgroup of R containing Z. We say that
(A, τ ) has the Atiyah Property with group Γ if for any n, m ∈ N and A ∈ Mm,n(A)
the kernel of the induced operator LA : H⊕n → H⊕m given by LA(ξ) = Aξ satisfies
τm(ker(LA)) ∈ Γ. We say that (A, τ ) has the Strong Atiyah Property if (A, τ ) has
the Atiyah Property with group Z.
Of course the case that Γ = R is of no interest in the above definition. By the fact
that ker(LA) = ker(LA∗A), it suffices to consider n = m in the above definition.
In this case it is easy to see that ker(LA) = Im(LA∗ ) so we may replace kernels
with images in the above definition. Furthermore, if A is equipped with a C∗-norm
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 5
and τ is faithful on the C∗-algebra generated by A, the tracial representation of
A ⊆ B(H) clearly does not matter.
It is clear that if G is a group that satisfies the Strong Atiyah Conjecture (e.g. any
free group) and τG is the canonical tracial state on L(G) (the group von Neumann
algebra), then (CG, τG) has the Strong Atiyah Property. The following provides
examples of a tracial ∗-algebras that have the Atiyah Property. In particular, the
following result implies that the tracial ∗-algebra generated by a single semicircular
variable has the Strong Atiyah Property with respect to the canonical tracial state
(see [23] or [8]).
Lemma 2.2. Let µ be a compactly supported probability measure on C. Let Γ be
the topological closure of the additive subgroup of R generated by 1 and the measures
of the atoms of µ and let (A, τ ) be the tracial ∗-subalgebra of L∞(µ) ⊆ B(L2(µ))
generated by multiplication by polynomials with trace
Then (A, τ ) has the Atiyah Property with group Γ.
τ (Mp) =ZC
p dµ.
Proof. Let δt denote the point-mass measure at t ∈ C. Then we can write
where ν is a non-atomic, compactly supported measure on C and αt ∈ Γ for all t.
Therefore ν(C) ∈ Γ by the construction of Γ.
µ = ν +Xt
αtδt
To see that (A, τ ) has the Atiyah Property with group Γ, fix ℓ ∈ N and let [Ai,j ] ∈
Mℓ(A). Viewing each Ai,j as a polynomial, we can view [Ai,j ] as a measureable
function from C to Mℓ(C). Moreover, if P is the projection onto the image of
[Ai,j] (which is in the von Neumann algebra generated by Mℓ(A) and thus is in
L∞(µ)⊗Mℓ(C)) and Pt ∈ Mℓ(C) is the projection onto the image of [Ai,j(t)], it is
elementary to see that P (t) = Pt µ-almost everywhere. Hence
τℓ(P ) =ZC
tr(P (t)) dµ(t) =ZC
rank([Ai,j (t)]) dµ(t).
Recall the rank of a matrix M ∈ Mℓ(C) may be obtained by computing the
maximum size of a submatrix with non-zero determinant. However, the pointwise
determinant of submatrices of [Ai,j(t)] is a polynomial in t and thus is either zero
everywhere or non-zero except at a finite number of points. Hence we obtain that
t 7→ rank([Ai,j (t)]) is an integer-valued function that is constant except at a finite
number of points which may or may not be atoms of µ. It is then easy to deduce
that τℓ(P ) is an integer-valued combination of elements of Γ and thus lies in Γ. (cid:3)
Extending these integration techniques, we obtain the following result for the
product of measures on C. Notice that the tracial ∗-algebra constructed is the
tensor product of tracial ∗-algebras from Lemma 2.2.
Lemma 2.3. Let n ∈ N and let {µj}n
ability measures on C. Let µ be the product measure of {µj}n
the tracial ∗-algebra generated by multiplication by the coordinate functions {xj}n
with trace
j=1 be non-atomic, compactly supported prob-
j=1 and let (A, τ ) be
j=1
τ (Mf ) =ZCn
f dµ.
6
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
Then (A, τ ) has the Strong Atiyah Property .
Proof. We claim that if p(x1, . . . , xn) is a polynomial and V is the zero set of
p(x1, . . . , xn), then µ(V ) ∈ {0, 1} and µ(V ) = 1 only occurs when p(x1, . . . , xn) is
the zero polynomial. To prove this claim, we proceed by induction on n with
the case n = 1 following from Lemma 2.2. Suppose the claim holds for n −
1. Let p(x1, . . . , xn) be any polynomial and let ν be the product measure of
{µj}n−1
j=1 . Clearly the claim is trivial if p(x1, . . . , xn) is the zero polynomial so
suppose p(x1, . . . , xn) is not the zero polynomial. For each t ∈ C let
Vt := {(x1, . . . , xn−1) ∈ Cn p(x1, . . . , xn−1, t) = 0}.
Therefore the zero set of p(x1, . . . , xn) is St∈C Vt and ν(Vt) ∈ {0, 1} for each t ∈ C
by the induction hypothesis. If ν(Vt) = 1, then p(x1, . . . , xn−1, t) must be the zero
polynomial which implies xn − t divides p(x1, . . . , xn) since we can write
p(x1, . . . , xn) =
n−1
Xk=1 Xik≥0
pi1,...,in−1(xn)xi1
1 · · · xin−1
n−1
where pi1,...,in−1 are polynomials and if pi1,...,in−1(t) 6= 0 for at least one i1, . . . , in−1,
then clearly p(x1, . . . , xn−1, t) would not be the zero polynomial. By degree ar-
guments there are at most a finite number of t ∈ C such that xn − t divides
p(x1, . . . , xn) so ν(Vt) = 0 except for a finite number of t ∈ C. Since µn contains
no atoms, by integrating using Fubini's Theorem we easily obtain that the zero set
of p(x1, . . . , xn) has zero µ-measure as desired.
To see that (A, τ ) has the Strong Atiyah Property, fix ℓ ∈ N and let [Ai,j] ∈
Mℓ(A). Thus each Ai,j is a multivariable polynomial. If P is the projection onto
the image of [Ai,j], then, as in the proof of Lemma 2.2, we obtain that
τℓ(P ) =ZCn
rank([Ai,j (t1, . . . , tn)]) dµ(t1, . . . , tn).
Since the rank of a matrix can be determined by computing the largest non-
zero determinant of a submatrix and since the determinant of any submatrix of
[Ai,j(x1, . . . , xn)] is a polynomial in x1, . . . , xn whose zero set either has zero or full
µ-measure, the result is complete.
(cid:3)
Next we endeavour to extend the above result to include compactly supported
probability measures on R that have atoms. We will only focus on measures with
atoms that lie in certain subgroups of Q since the main result of Section 3 will also
only apply to these groups.
To discuss such measures, for an additive subgroup Γ of Q and a d ∈ N we define
1
d
Γ :=(cid:26) 1
d
g g ∈ Γ(cid:27) ,
which is clearly an additive subgroup of Q that contains Z if Γ contains Z. As such,
the following result is trivial.
Lemma 2.4. Let (A, τ ) be a tracial ∗-algebra that has the Atiyah Property with
group Γ and let ℓ ∈ N. Then (Mℓ(A), 1
ℓ Γ.
ℓ τℓ) has the Atiyah Property with group 1
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 7
Theorem 2.5. Let n ∈ N and let {µj}n
sures on C. Let µ be the product measure of {µj}n
∗-algebra generated by multiplication by the coordinate functions {xj}n
j=1 be compactly supported probability mea-
j=1 and let (A, τ ) be the tracial
j=1 with trace
τ (Mf ) =ZCn
f dµ.
Suppose for each j ∈ {1, . . . , n} there exists a dj ∈ N such that the atoms of µj have
measures contained in 1
j=1 dj, then (A, τ ) has the Atiyah Property
dj
with group 1
Z. If d := Qn
d Z.
Proof. By assumptions, for each j ∈ {1, . . . , n} we can write
µj = µ′′
j +Xk
αk
dj
δtk
j := 1
µ′′(C) µ′′
j
if µ′′
j
j is an non-atomic measure. Notice µ′′
where δt represents the point-mass probability measure at t, the sum is finite,
αk ∈ N, tk1 6= tk2 if k1 6= k2, and µ′′
Z.
Let µ′
j be the Lebesgue measure on [0, 1] if
µ′′
j = 0. Therefore the tracial ∗-algebra generated by polynomials acting on L2(µj)
can represented a tracial ∗-algebra of diagonal matrices in Mdj (B(L2(µ′
j)) (with
respect to the canonical normalized matrix trace) where the polynomial x maps to
the matrix with x appearing on the diagonal dj µ′′
j (C) times and each tk appearing
on the diagonal αk times.
6= 0 and let µ′
j (C) ∈ 1
dj
Let µ′ be the product measure of {µ′
j}n
j=1 and let (Aµ′ , τµ′ ) be the tracial ∗-
j=1 with trace
algebra generated by multiplication by the coordinate functions {xj}n
τµ(Mf ) =RCn f dµ′. By taking tensor products of the tracial ∗-algebras generated
by polynomials acting on L2(µj), it is easily seen using the above representations
that (A, τ ) can be represented in the tracial ∗-algebra (Md(Aµ′ ), 1
d (τµ′ )d). Since
Lemma 2.3 implies (Aµ′ , τµ′ ) has the Strong Atiyah Property, Lemma 2.4 implies
(Md(Aµ′ ), 1
d Z completing the proof.
(cid:3)
d (τµ′ )d) has the Atiyah Property with group 1
3. Atiyah Property for Freely Independent Random Variables
The goal of this section is to use the Atiyah Property for tracial ∗-algebras to gain
information about the distributions of matricial polynomials of freely independent
random variables.
In particular, Theorem 3.1 will enable the extensions of the
results from Section 2 to the non-commutative setting as seen in Theorem 3.4 thus
completing the proof of part (1) of Theorem 1.1. The proof of Theorem 3.1, which
is based on the proof of [16, Proposition 3] (or the updated version [17, Proposition
6.1]), will be postponed until the next section in order to focus on the applications
of Theorem 3.1.
Recall that given unital ∗-algebras Ai ⊆ B(Hi) with vector states τi that are
tracial on Ai, we can consider the ∗-subalgebra A1 ∗ A2 inside the reduced free
product C∗-algebra (B(H1), τ1)∗(B(H2), τ2) generated by A1 and A2. The canonical
vector state τ1 ∗ τ2 is then a tracial state on A1 ∗ A2 (see [23] or [8]). Similarly we
can consider the ∗-subalgebra A1 ⊙ A2 inside the C∗-algebra B(H1 ⊗ H2) generated
by T ⊗ IH2 and IH1 ⊗ S for all T ∈ A1 and S ∈ A2. With this notation, it is easy
to state the following technical result.
8
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
Theorem 3.1. Let n ∈ N, let Fn be the free group on n generators, let CFn be
the group ∗-algebra equipped with the C∗-norm defined by the left regular represen-
tation, and let τFn be the canonical trace on L(Fn). Let A and B be ∗-subalgebras
of the tracial von Neumann algebras with separable preduals (M, τM) and (N, τN)
respectively. Suppose that (A ⊙ B, τM⊗τN) has the Atiyah Property with group 1
d Z
for some d ∈ N. Then ((A ∗ CFn) ⊙ B, (τM ∗ τFn)⊗τN) has the Atiyah Property with
group 1
d Z.
Clearly Theorem 3.1 implies the following two results.
Corollary 3.2. If A and B are as in Theorem 3.1 and n, m ∈ N, then ((A∗ CFn)⊙
(B ∗ CFm), (τM ∗ τFn)⊗(τN ∗ τFm)) has the Atiyah Property with group 1
d Z.
Proof. This is a simple application of Theorem 3.1 twice using A = B and B =
A ∗ CFn the second time.
(cid:3)
Corollary 3.3. Let A be a ∗-subalgebra of a tracial von Neumann algebra with
separable predual (M, τM). Suppose (A, τM) has the Atiyah Property with group
1
d Z for some d ∈ N. Then (A ∗ CFn, τM ∗ τFn ) has the Atiyah Property with group
1
d Z.
Proof. Take B = C in Theorem 3.1.
(cid:3)
Using Theorem 3.1 along with the examples of Section 2, we obtain the following
result which provides important information about the spectral distributions of
matricial polynomials of normal, freely independent random variables.
Theorem 3.4. Let n ∈ N and let X1, . . . , Xn be normal, freely independent random
variables with probability measures µj as distribution respectively. Suppose for each
j ∈ {1, . . . , n} there exists a dj ∈ N such that the atoms of µj have measures
contained in 1
Z. If A is the unital ∗-algebra generated by X1, . . . , Xn (obtained by
dj
taking a reduced free product of tracial ∗-algebras), τ is the canonical trace on A,
j=1 dj, then (A, τ ) has the Atiyah Property with group 1
d Z.
and d :=Qn
Furthermore, if [pi,j] is an ℓ × ℓ matrix whose entries are non-commutative poly-
nomials in n variables and their adjoints such that [pi,j(X1, . . . , Xn)] is normal,
then the measure of any atom of the spectral distribution of [pi,j(X1, . . . , Xn)] with
respect to the normalized trace 1
ℓ τℓ is in 1
dℓ Z.
Proof. Let µ be the product measure of {µj}n
algebra generated by multiplication by the coordinate functions {xj}n
j=1 and let (A0, τ0) be the tracial ∗-
j=1 on L2(µ)
with trace τ0(Mf ) = RCn f dµ. Clearly each Xj has a representation in A0 as
multiplication by the coordinate function xj so we will view Xj ∈ A0 for all j ∈
{1, . . . , n}. Let U := λ(1) be the canonical generating unitary operator for L(Z).
Then it is easy to see that X1, U X2U ∗, . . ., U nXn(U n)∗ are freely independent in
A0 ∗ CZ with respect to the trace τ0 ∗ τZ. However, since (A0, τ0) has the Atiyah
Property with group 1
d Z by Theorem 2.5, (A0 ∗ CZ, τ0 ∗ τZ) has the Atiyah Property
with group 1
d Z by Theorem 3.1. Hence (A, τ ) has the Atiyah Property with group
1
d Z by taking the canonical isomorphism of tracial ∗-algebras.
Next suppose that [pi,j ] is an ℓ × ℓ matrix whose entries are non-commutative
polynomials in n variables and their adjoints such that [pi,j(X1, . . . , Xn)] is normal
and the spectral distribution of [pi,j(X1, . . . , Xn)] has an atom. By translation
we may assume that this atom occurs at zero and thus corresponds to the kernel
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 9
projection of [pi,j(X1, . . . , Xn)]. Since (A, τ ) has the Atiyah Property with group
d Z we obtain that the measure of the atom is in 1
1
(cid:3)
dℓ Z.
As an application of the above result, we recall that Voiculescu developed in
[19] the notion of the additive free product of measures in which if {Xj}n
j=1 are
self-adjoint, freely independent random variables with probability measures µj as
distribution respectively, then the additive free product measure µ := µ1 ⊞ · · · ⊞ µn
is the distribution of X1 + · · · + Xn in the reduced free product C∗-algebra. Hence
Theorem 3.4 implies the following specific case of [4, Theorem 7.4].
Corollary 3.5 (see [4, Theorem 7.4]). If n ∈ N and {µj}n
j=1 are non-atomic,
compactly supported probability measures on R, then µ1 ⊞ · · · ⊞ µn has no atoms.
Proof. Since each µj contains no atoms, we can apply Theorem 3.4 to conclude
that µ := µ1 ⊞ · · · ⊞ µn may only have atoms in Z. Since µ is a probability measure,
if µ has an atom, then µ must be a point-mass measure which would imply that
X1 + · · · + Xn = αI for some α ∈ R contradicting the fact that X1, . . ., Xn are
freely independent.
(cid:3)
To complete this section, we can extend Theorem 3.4 to tensor products of tracial
∗-algebras generated by self-adjoint, freely independent random variables.
Corollary 3.6. Let n, m ∈ N and let X1, . . . , Xn and Y1, . . . , Ym be collections of
normal, freely independent random variables with probability measures µj and νk as
distribution respectively. Let (A, τA) and (B, τB) be the tracial ∗-algebras generated
by the reduced free products of {X1, . . . , Xn} and {Y1, . . . , Ym} respectively. Suppose
for each j ∈ {1, . . . , n} and k ∈ {1, . . . , m} there exists a dj, d′
k ∈ N such that the
atoms of µj and νk have measures contained in 1
dj
Z respectively. If
Z and 1
d′
k
n
m
d :=
dj ·
Yj=1
d′
k,
Yk=1
d Z.
j=1 and let ν be the product mea-
k=1. Let (A0, τA,0) be the tracial ∗-algebra generated by multiplication
then (A ⊗ B, τA⊗τB) has the Atiyah Property with group 1
Proof. Let µ be the product measure of {µj}n
sure of {νk}m
by the coordinate functions {xj}n
and let (B0, τB,0) be the tracial ∗-algebra generated by multiplication by the co-
ordinate functions {yk}m
(A0 ⊙ B0, τA,0⊗τB,0) has the Atiyah Property with group 1
d Z by Theorem 2.5.
The remainder of the proof follows the proof of Theorem 3.4 by an application of
Corollary 3.2.
(cid:3)
j=1 on L2(µ) with trace τA,0(Mf ) = RCn f dµ
k=1 on L2(ν) with trace τB,0(Mf ) = RCm f dν. Therefore
Notice that Corollary 3.6 has the following interesting application. For any
n, m ∈ N let P1, . . . , Pm ∈ A := Alg(S1, . . . , Sn) be polynomials in n free semicir-
cular variables S1, . . . , Sn and let ∂j be the non-commutative difference quotient
derivations (see [22]). Let JP := [∂iPj]ij which is an n × m matrix with entries
in A ⊗ A. The matrix JP is the non-commutative Jacobian of P := (P1, . . . , Pm).
We define the rank of JP to be the (non-normalized) trace of its image projection
in Mn(W ∗(A ⊗ A)).
Corollary 3.7. With the above notation, rank(JP ) ∈ {0, 1, . . . , min(m, n)}. In
particular, if {Pj}m
j=1 are not all constant, then rank(JP ) ≥ 1.
10
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
4. Proof of Theorem 3.1
This section is devoted to the proof of Theorem 3.1, which underlies all results
of Section 3. Our proof is essentially the same as the argument of Schick in [16]
adapted for the case of algebras. This proof has themes similar to those used in
[14, Lemma 10.43], which makes use of the notion of a Fredholm module to show
that the free groups satisfy the Strong Atiyah Conjecture. The idea of applying
Fredholm modules has its roots in a proof of the Kadison Conjecture for free groups
on two generators from [5].
Proof of Theorem 3.1. Let H := L2(M, τM). Thus M has left and right actions on
H. Similarly, let K := L2(N, τN). For a right-(M⊗N)⊕ℓ invariant subspace L of
(H ⊗ K)⊕ℓ, we define
dimM⊗N(L) := trM⊗N(Q) = (τM⊗τN)ℓ(Q)
where Q is the orthogonal projection onto L (which is an element of Mℓ(M⊗N)
acting on the left).
For later convenience we desire to construct a certain isomorphism of Hilbert
spaces that commonly appears in the proof that Fn satisfies the Strong Atiyah
Conjecture. We desire a bijection
ψ : {δh h ∈ Fn \ {e}} → {δh ⊗ ei h ∈ Fn, i ∈ {1, . . . , n}}.
i=1 are the canonical orthonormal basis for Cn) as this will clearly
(where {ei}n
produce a unitary operator
Ψ : ℓ2(Fn) \ (Cδe) → ℓ2(Fn) ⊗ Cn.
Let {ui}n
i=1 be generators for Fn. Consider the Cayley graph of Fn with edges
{g, gui}. For each h ∈ Fn \ {e} let e(h) be the first edge of the geodesic from h to
e. Thus we may write e(h) = {ψ0(h), ψ0(h)ur(h)} for some r(h) ∈ {1, . . . , n}. Thus
if we define
ψ(δh) := δψ0(h) ⊗ er(h),
we clearly obtain a bijection.
Let λ denote the left regular representation of Fn on ℓ2(Fn). We claim that Ψ
has the property that for each T ∈ CFn the set of {δh}h∈Fn\{e} such that Ψ(λ(T )δh)
does not make sense (i.e. hλ(T )δh, δei 6= 0) or
Ψ(λ(T )δh) 6= (λ(T ) ⊗ ICn )Ψ(δh)
is finite. To see this notice for fixed g, h ∈ Fn the only way that λ(g)(δh) /∈
ℓ2(Fn) ⊖ (Cδe) is if gh = e and the only way that Ψ(λ(g)δh) 6= (λ(g) ⊗ ICn )Ψ(δh)
can occur is if when reducing gh a term from g cancels the second-last letter in h
(which occurs for a finite number of h for a given g). Thus the claim follows by the
linearity of Ψ.
Let {ζj}j∈Z be any orthonormal basis for K with ζ0 a trace vector. We claim we
may assume that there exists an orthonormal basis {ξj}j∈Z of H such that ξ0 is a
trace vector and
{k ∈ Z hT ξj, ξki 6= 0}
is finite for each j ∈ Z and T ∈ A. To see this, we first may assume that A is
finitely generated by self-adjoint operators {Ak}m
k=1 since we need only check the
Atiyah Property for one matrix with entries in (A ∗ CFn) ⊙ B at a time and a finite
j}j∈Z is any orthonormal basis of H
number of elements of A will appear.
If {ξ′
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 11
with ξ′
the Gram-Schmidt Orthogonalization Process to
0 = ξ0 a trace vector, then the desired basis will be produced by applying
{Ai1 · · · Aim ξ′
j j ∈ Z, m ∈ N ∪ {0}, {ik}m
k=1 ⊆ {1, . . . , n}}
starting with ξ′
0.
Recall (A ∗ CFn) ⊙ B acts on ((H, ξ0) ∗ (ℓ2(Fn), δe)) ⊗ K and
(H, ξ0) ∗ (ℓ2(Fn), δe) = Cξ0 ⊕(cid:16)M C (ξj1 ⊗ δg1 ⊗ · · · )(cid:17) ⊕(cid:16)M C (δg1 ⊗ ξj1 ⊗ · · · )(cid:17)
(where ξ0 = δe) where all the tensors in the direct sums have finite length (ending
at any point), alternate between basis elements of H and ℓ2(Fn), jk ∈ N, ik ∈
Z \ {0}, and gk ∈ Fn \ {e}. Notice that the union of the vectors used in the above
definition of (H, ξ0) ∗ (ℓ2(Fn), δe) is an orthonormal basis for (H, ξ0) ∗ (ℓ2(Fn), δe).
For convenience of notation, ξ0 ⊗ δg1 ⊗ · · · := δg1 ⊗ · · · , · · · ⊗ δgm ⊗ ξ0 := · · · ⊗ δgm,
and · · · ⊗ ξjm ⊗ δe = · · · ⊗ ξjm .
Define the Hilbert spaces
L+ := ((H, ξ0) ∗ (ℓ2(Fn), δe)) ⊗ K and L− := (L+ ⊗ Cn ⊗ H) ⊕ (H ⊗ K).
Notice that (A ∗ CFn) ⊙ B has a canonical left action on L+ and thus induces
a canonical left action on L− by letting an operator T ∈ (A ∗ CFn) ⊙ B act via
(T ⊗ ICn ⊗ IH) ⊕ 0. Thus we may view L+ and L− as left (A ∗ CFn) ⊙ B-modules.
Similarly, M⊗N has a canonical right action on H ⊗ K and thus on L+ by
(· · · ⊗ δgm ⊗ ξjm ⊗ ζ)T = · · · ⊗ δgm ⊗ ((ξjm ⊗ ζ)T ))
for all ζ ∈ K. Hence L+ is also a right M⊗N-module. It is clear that the right
action of M⊗N and the left action of (A ∗ CFn) ⊙ B on L+ commute.
We desire to construct a bijection φ between the canonical basis elements of L+
It is clear that if
and L− which will induce a unitary operator Φ : L+ → L−.
Λ := Λ0 ∪ Λ′ where Λ0 = {ξj ⊗ ζj′ }j,j′∈Z and
m ≥ 1, {gk}m
j0, jm, j′ ∈ Z, {jk}m−1
k=1 ∈ Fn \ {e},
k=1 ⊆ Z \ {0} (cid:27) ,
Λ′ :=(cid:26)(ξj0 ⊗ δg1 ⊗ · · · ⊗ δgm ⊗ ξjm ) ⊗ ζj′ (cid:12)(cid:12)(cid:12)(cid:12)
then Λ is an orthonormal basis of L+. Furthermore
Θ := {0 ⊕ (ξj ⊗ ζj′ )}j,j′∈Z ∪ {(η ⊗ ei ⊗ ξj) ⊕ 0 η ∈ Λ, j ∈ Z, i ∈ {1, . . . , n}}
is an orthonormal basis of L−. Define φ : Λ → Θ by defining φΛ0 via
φ(ξj ⊗ ζj′ ) = 0 ⊕ (ξj ⊗ ζj′ )
for all j, j′ ∈ Z and by defining φΛ′ via the following rule: for
η = (ξj0 ⊗ δg1 ⊗ · · · ⊗ δgm ⊗ ξjm ) ⊗ ζj′ ∈ Λ
define
φ(η) = (((ξj0 ⊗ δg1 ⊗ · · · ⊗ δgm−1 ⊗ ξjm−1 ⊗ δψ0(g)) ⊗ ζj′ ) ⊗ er(g) ⊗ ξjm ) ⊕ 0
(where if ψ0(g) = e, we reduce the length of the first tensor by removing δe). Since
Ψ is a bijection on the given basis elements, it is elementary to verify that φ is a
bijection and thus induces a Hilbert space isomorphism Φ : L+ → L−.
Define a right M⊗N-module structure on L− by defining ηT := Φ((Φ−1(η))T )
for all T ∈ M⊗N and η ∈ L−. It is easy to see that
((η ⊗ ei ⊗ ξk) ⊕ (ξj ⊗ ζj′ ))(T ⊗ S) = (η(IH ⊗ S) ⊗ ei ⊗ ξkT ) ⊕ (ξj T ⊗ ζj′ S)
12
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
for all T ∈ M and S ∈ N. Hence 0 ⊕ (H ⊗ K) and (L+ ⊗ Cn ⊗ H) ⊕ 0 are a right
M⊗N-invariant subspace of L−. It is clear that the right action of M⊗N on L−
commutes with the left action of (A ∗ CFn) ⊙ B on L−.
Define Ξ to be the union of {ξ0 ⊗ ζ0} with
(cid:26)(ξj0 ⊗ δg1 ⊗ · · · ⊗ ξjm−1 ⊗ δgm) ⊗ ζ0 (cid:12)(cid:12)(cid:12)(cid:12)
m ≥ 1, {gk}m
j0 ∈ Z, {jk}m−1
k=1 ∈ Fn \ {e},
k=1 ⊆ Z \ {0} (cid:27) .
It is clear that Ξ is a set of orthonormal vectors in L+ each of which generates a
one-M⊗N-dimensional right M⊗N-submodule of L+ that are pairwise orthogonal
and whose union is dense in L+ (as ξ0 and ζ0 are cyclic vectors for the right actions).
By the definition of Φ it is clear that Φ(Ξ) is a set of orthonormal vectors in L− each
of which generates a one-M⊗N-dimensional right M⊗N-submodule of L− that are
pairwise orthogonal and whose union is dense in L−.
We claim if T ∈ (A ∗ CFn) ⊙ B, then
{ξ ∈ Ξ hT ξ, ξj ⊗ ζj′ iL+ 6= 0 for some j, j′ ∈ Z or Φ(T (ξ)) 6= T (Φ(ξ))}
is a finite subset (containing ξ0). By linearity it suffices to prove the claim when
T is a product of elements from A ∪ B ∪ {λ(h)}h∈Fn. First we will prove the claim
when T ∈ A ∪ B. However, it clearly follows that hT ξ, ξj ⊗ ζj′ iL+ 6= 0 for some
j, j′ ∈ Z or Φ(T (ξ)) 6= T Φ(ξ) only if ξ = ξ0 ⊗ ζ0.
Next we will prove the claim for T ∈ {λ(h)}h∈Fn\{e}. Fix h ∈ Fn, fix T = λ(h),
and fix
ξ = ξj0 ⊗ δg1 ⊗ · · · ξjm−1 ⊗ δgm ⊗ ζ0 ∈ Ξ \ {ξ0 ⊗ ζ0}.
If m > 1 or j0 6= 0, then hT ξ, ξj ⊗ ζj′ i = 0 for all j, j′ ∈ Z and Φ(T (ξ)) = T (Φ(ξ))
are clear. Otherwise ξ = δg1 ⊗ ζ0 and it clear that hT ξ, ξj ⊗ ζj′ i 6= 0 for some
j, j′ ∈ Z only if hg1 = e and Φ(T (ξ)) = T (Φ(ξ)) unless Ψ(T δg1) 6= (T ⊗ ICn )Ψ(δg1 ).
Since the number of such g1 is finite, the claim holds in this case.
Next notice for any element ξ ∈ Ξ and any element T of A ∪ {λ(h)}h∈Fn that
T ξ is a finite linear combination of elements of Ξ ∪ {ξj ⊗ ζ0}j∈Z by the choice of
the orthonormal basis {ξj}j∈Z. Furthermore, for any element ξ ∈ Ξ ∪ {ξj ⊗ ζ0}j∈Z
and any element T of A ∪ {λ(h)}h∈Fn there are only a finite number of elements
η of Ξ such that hT η, ξiL+ 6= 0. Therefore if T1, . . . , Tn ∈ A ∪ {λ(h)}h∈Fn, then
the set of all ξ ∈ Ξ such that hT1 · · · Tnξ, ξj ⊗ ζj′ iL+ 6= 0 for some j, j′ ∈ Z,
hT2 · · · Tnξ, ξj ⊗ ζj′ iL+ 6= 0 for some j, j′ ∈ Z, or Φ(T1 · · · Tnξ) 6= T1Φ(T2 · · · Tnξ) is
finite. Thus the claim then follows by recursion and the fact that the B-operator
commute with elements of A ∗ CFn and with Φ.
The above construction show that we have two representations of (A ∗ CFn) ⊙ B
that differ by a A ⊙ B-finite rank operator. In order to complete the proof, we need
a way to analyze the trace of such operators. Fix ℓ ∈ N and fix
A := [Ai,j] ∈ Mℓ((A ∗ CFn) ⊙ B).
± and let P± ∈ B(L⊕ℓ
The left actions of (A ∗ CFn) ⊙ B on L± allows A to act on L⊕ℓ
± . Let A± be the left
action of A on L⊕ℓ
± ) be the projection onto the image of A±.
Thus we desire to show that ((τM ∗ τFn )⊗τN)ℓ(P+) ∈ 1
d Z. Since the right action of
M⊗N on L± commutes with the left action of (A ∗ CFn) ⊙ B, we easily obtain that
all operators under consideration commute with the diagonal right action of M⊗N
on these spaces.
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 13
Notice that there are only finitely many elements of (A ∗ CFn) ⊙ B that appear
in A. For each of these elements T , we recall that
{ξ ∈ Ξ hT ξ, ξj ⊗ ζj′ iL+ 6= 0 for some j, j′ ∈ Z or Φ(T (ξ)) 6= T (Φ(ξ))}
is finite. Let L+,0 be the finite M⊗N-dimensional right M⊗N-submodule of L+
spanned by the vectors that appear in the above set for at least one T ∈ (A∗CFn)⊙B
appearing in A. Thus L+,c := L+ ⊖ L+,0 is a right M⊗N-submodule of L+.
Let L−,c := Φ(L+,c), which is a right M⊗N-submodule of L−. Therefore, since
L+,0 contained all ξ ∈ Ξ where Φ(T (ξ)) 6= T (Φ(ξ)) for some T ∈ (A ∗ CFn) ⊙ B
appearing in A and since the right M⊗N-actions commutes with the left action of
T and with Φ, we clearly obtain that
A+L+,c = Φ−1 ◦ A− ◦ ΦL+,c.
By progressively adding the right M⊗N-submodule of L+ generated by a single
element of Ξ we can choose an increasing sequence
of finite M⊗N-dimensional right M⊗N-submodules of L+ such that
L+,0 ⊂ L+,1 ⊂ L+,2 ⊂ · · · ⊂ L+
L+,j.
L+ = [j≥0
Let L−,j := Φ(L+,j) for all j ∈ N∪{0}. Hence each L−,j is a right M⊗N-submodule
of L− generated by a finite number of elements of Φ(Ξ). Notice that Λ0 ⊆ L+,0 so
0 ⊕ (H ⊗ K) ⊆ L−,0. By construction, it is clear that
A±(L⊕ℓ
± ) = [j≥0
A±(L⊕ℓ
±,j).
For each j ∈ N ∪ {0} let P±,j be the orthogonal projections onto A±(L⊕ℓ
±,j ).
Since only finitely many elements of (A ∗ CFn) ⊙ B appear in A, by our selection
right M⊗N-modules generated by elements of Ξ we see that A+ has finite propa-
gation; that is, for every j ∈ N there exists an nj ∈ N such that A+(L⊕ℓ
+,nj .
Indeed an element of B does not modify the submodule, {λ(h)}h∈Fn permutes the
elements of Ξ, and an element of A maps an element of Ξ to at most a finite-M⊗N-
dimensional M⊗N-module by the choice of the basis {ξj}j∈Z. Similarly, as the left
action of (A ∗ CFn) ⊙ B on L− has the same form and the right M⊗N-modules L−,j
are generated by elements of Φ(Ξ), A− also has propagation so we may assume that
A−(L⊕ℓ
−,nj by choosing nj sufficiently large.
+,j) ⊆ L⊕ℓ
−,j) ⊆ L⊕ℓ
The above allows us to view A±(L⊕ℓ
±,j) as images of rectangular matrices with
entries in A ⊙ B acting on the left from (H ⊗ K)⊕qj to (H ⊗ K)⊕pj
for some
appropriate choice of qj and pj. Indeed an element from CFn acting on an element
of Ξ or Φ(Ξ) acts as a scalar matrix since {λ(h)}h∈Fn sends the right M⊗N-basis
vectors Ξ and Φ(Ξ) to scalar multiples of other elements of Ξ and Φ(Ξ) respectively.
Furthermore, each element T ∈ A acts by the usual left action of A on H ⊆ L+
(which corresponds to the action of A ⊗ IK on the right M⊗N-module generated by
ξ0 ⊗ζ0 ∈ Ξ) and otherwise act by sending the other elements of Ξ and every element
of Φ(Ξ) to a finite linear combination of elements of Ξ and Φ(Ξ) respectively and
thus can be viewed as scalar matrices on these right M-modules. Furthermore, it
is clear that an element of B acts via IH ⊗ B on each of the one-M⊗N-dimensional
14
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
right M⊗N-modules spanned by an element of Ξ or Φ(Ξ). Thus the claim follows.
Therefore, since A ⊙ B has the Atiyah Property with group 1
d Z, we obtain that
trM⊗N(P±,j ) = dimM(A±(L⊕ℓ
±,j )) ∈
1
d
Z.
Notice that
A±(L⊕ℓ
±,0), A±(L⊕ℓ
±,c), and each A±((L±,j ∩ L+,c)⊕ℓ)
are all closed right M⊗N-modules (note L±,j ∩ L±,c = L±,j ⊖ L±,0). We claim that
±,0) ∩ A±(L⊕ℓ
dimM⊗N(cid:18)A±(L⊕ℓ
= limj→∞ dimM⊗N(cid:16)A±(L⊕ℓ
±,c)(cid:19)
±,0) ∩ A±((L±,j ∩ L±,c)⊕ℓ)(cid:17) .
To see this, it suffices by the continuity of von Neumann dimension (see [10, proof
of Theorem 1.12]) to show that
A±(L⊕ℓ
±,0) ∩ A±(L⊕ℓ
±,c) = [j≥0
A±(L⊕ℓ
±,0) ∩ A±((L±,j ∩ L±,c)⊕ℓ).
To see this, notice one inclusion is trivial. For the other inclusion, recall that A±
has finite propagation so there exists an n0 ∈ N such that A±(L⊕ℓ
±,0) ⊆ L⊕ℓ
±,n0 so
A±(L⊕ℓ
±,0) ∩ A±(L⊕ℓ
±,c) = A±(L⊕ℓ
= A±(L⊕ℓ
±,0) ∩ L⊕ℓ
±,0) ∩ L⊕ℓ
±,n0 ∩ A±(L⊕ℓ
±,c)
for some sufficiently large m ∈ N. Specifically, to choose m, we notice, by the
same arguments that Φ almost commutes with the left actions, that there exists
an m ∈ N such that if η ∈ L±,m+k ⊖ L±,m for any k ≥ 1, then every entry of A
applied to η is orthogonal to L±,n0 (that is, there are a finite number of elements
η of Ξ for which there is an entry T in A such that T η has non-zero inner product
with an element of L±,n0 ∩ Ξ). To see the above equality for this m ∈ N, we notice
that one inclusion is trivial. For the other inclusion, fix
ξ ∈ L⊕ℓ
±,n0 ∩
[j≥1
A±((L±,j ∩ L±,c)⊕ℓ)
.
Thus there exists ηj ∈ (L±,c ∩ L±,j)⊕ℓ such that ξ = limj→∞ A±ηj . Therefore, if
P is the projection of L⊕ℓ
±,c onto (L±,m ∩ L±,c)⊕ℓ, then
where ωj ∈ (L⊕ℓ
±,n0 )⊥. Therefore, since
Aηj = A(P ηj) + ωj
lim
j→∞
A±ηj = ξ ∈ L⊕ℓ
±,n0 ,
We claim that
L⊕ℓ
±,n0 ∩
[j≥1
A±((L±,j ∩ L±,c)⊕ℓ)
±,n0 ∩(cid:16)Sj≥1 A±((L±,j ∩ L±,c)⊕ℓ)(cid:17) .
= L⊕ℓ
±,n0 ∩ A±((L±,m ∩ L±,c)⊕ℓ)
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 15
we obtain that limj→∞ ωj = 0 and ξ = limj→∞ A(P ζj ) where P ζj ∈ (L±,m ∩
L±,c)⊕ℓ as desired. Hence the claim is complete. Thus
A±(L⊕ℓ
±,0) ∩ A±(L⊕ℓ
±,c) = A±(L⊕ℓ
= A±(L⊕ℓ
±,n0 ∩ A±((L±,m ∩ L±,c)⊕ℓ)
±,0) ∩ L⊕ℓ
±,0) ∩ A±((L±,m ∩ L±,c)⊕ℓ)
which completes the claim.
⊆ Sj≥0 A±(L⊕ℓ
±,0) ∩ A±((L±,j ∩ L±,c)⊕ℓ)
Let P±,c to be the orthogonal projections onto A±(L⊕ℓ
±,c) and for each j ∈ N∪{0}
let P±,j,c be the orthogonal projection onto A±(L±,j ∩ L±,c)⊕ℓ. Notice that P±,c
and each P±,j,c need not be in the von Neumann algebra generated by Mℓ((A ∗
CFn) ⊙ B) but do commute with the right M⊗N-action on their respective spaces.
Since
A+L+,c = Φ−1 ◦ A− ◦ ΦL+,c,
we obtain that P+,j,c = Φ−1 ◦ P−,j,c ◦ Φ for all j ∈ N∪{0} and P+,c = Φ−1 ◦ P−,c ◦ Φ.
Hence
hP+,cη, ηiL⊕ℓ
+
= hP−,cΦ(η), Φ(η)iL⊕ℓ
−
and hP+,j,cη, ηiL⊕ℓ
+
= hP−,j,cΦ(η), Φ(η)iL⊕ℓ
−
for all j ∈ N ∪ {0} and η ∈ L⊕ℓ
+ .
Let Q± := P± − P±,c and for each j ∈ N ∪ {0} define Q±,j := P±,j − P±,j,c.
Clearly these are projections onto the complements of smaller projections in larger
projections. We claim that
trM⊗N(Q±) = lim
j→∞
trM⊗N(Q±,j).
To begin, let A0 denote the restriction of A± to L⊕ℓ
that
±,0. We claim for each fixed j ∈ N
0 −→ ker(Q±,jA0) −→ L⊕ℓ
±,0
Q±,j A0−→ Im(Q±,j) −→ 0
is a weakly exact sequence (that is, the images are dense in the kernels). To see this,
it suffices to check weak exactness at Im(Q±,j). It is clear that Q±,j(A±(L⊕ℓ
±,j)) is
dense in Im(Q±,j). However
A±(L⊕ℓ
±,j) = A±(L⊕ℓ
±,0) + A±((L±,j ∩ L±,c)⊕ℓ)
and it is clear that Q±,j(A((L±,j ∩ L±,c)⊕ℓ)) = 0. Thus Q±,j(A±(L⊕ℓ
±,0)) =
Q±,j(A±(L⊕ℓ
±,j )) is dense in Im(Q±,j). Since each term in the weak exact sequence
is a right M⊗N-module and weak exact sequence preserve M⊗N-dimension (see
[10, proof of Theorem 1.12]), we obtain that
dimM⊗N(L⊕ℓ
±,0) = dimM⊗N(Im(Q±,j)) + dimM⊗N(ker(Q±,jA0))
(which are all finite as dimM⊗N(L⊕ℓ
clear that
±,0) is finite by construction). Furthermore, it is
ker(Q±,jA0) = {η ∈ L⊕ℓ
±,0 Q±,jA0η = 0}.
Hence the sequence
0 −→ ker(A0) −→ ker(Q±,jA0)
A0−→ A±(L⊕ℓ
±,0) ∩ ker(Q±,j) −→ 0
is weakly exact. This implies the sequence
0 −→ ker(A0) −→ ker(Q±,jA0)
A0−→ A±(L⊕ℓ
±,0) ∩ A±((L±,j ∩ L±,c)⊕ℓ) −→ 0
16
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
is a weakly exact sequence since it is elementary to verify that
A±(L⊕ℓ
±,0) ∩ ker(Q±,j) = A±(L⊕ℓ
±,0) ∩ A±((L±,j ∩ L±,c)⊕ℓ).
Hence we obtain that
dimM⊗N(ker(Q±,jA0))
= dimM⊗N(ker(A0)) + dimM⊗N(cid:16)A±(L⊕ℓ
±,0) ∩ A±((L±,j ∩ L±,c)⊕ℓ)(cid:17) .
By combining the two above dimension equations we obtain that
dimM⊗N(Im(Q±,j)) = dimM⊗N(L⊕ℓ
±,0) − dimM⊗N(ker(A0))
for each j ∈ N. Similarly, by repeating the same arguments we obtain that
−dimM⊗N(cid:16)A±(L⊕ℓ
±,0) ∩ A±((L±,j ∩ L±,c)⊕ℓ)(cid:17)
dimM⊗N(Im(Q±)) = dimM⊗N(L⊕ℓ
±,0) − dimM⊗N(ker(A0))
−dimM⊗N(cid:16)A±(L⊕ℓ
±,0) ∩ A±((L±,c)⊕ℓ)(cid:17) .
Therefore, as all the terms in the above dimension equations are finite (in fact
bounded by dimM⊗N(L⊕ℓ
±,0)),
trM⊗N(Q±) = dimM⊗N(Im(Q±))
= limj→∞ dimM⊗N(Im(Q±,j)) = limj→∞ trM⊗N(Q±,j).
We will now use Ξ and Φ(Ξ) to compute traces. For each η ∈ Ξ and i ∈ {1, . . . , ℓ}
let
where η is in the ith spot and similarly let
ηi = (0, 0, . . . , 0, η, 0, . . . , 0) ∈ L⊕ℓ
+
φ(ηi) = (0, . . . , 0, φ(η), 0, . . . , 0) ∈ L⊕ℓ
− .
Since Ξ and Φ(Ξ) are orthonormal M⊗N-bases for L+ and L− respectively, we
easily obtain that
and
ℓ
trM⊗N(Q+) = Xη∈Ξ
Xi=1
hQ+ηi, ηiiL⊕ℓ
+
ℓ
trM⊗N(Q−) = Xη∈Ξ
Xi=1
hQ−φ(ηi), φ(ηi)iL⊕ℓ
−
.
Furthermore, we notice if η = ξ0 ⊗ ζ0 ∈ Ξ, then
whereas
ℓ
Xi=1
hP+ηi, ηiiL⊕ℓ
+
= ((τM ∗ τFn )⊗τN)ℓ(P+)
ℓ
ℓ
by the definition of A− and P−. Finally, we claim that
Xi=1
hP−φ(ηi), φ(ηi)iL⊕ℓ
−
=
0 = 0
Xi=1
ℓ
Xi=1
hP+ηi, ηiiL⊕ℓ
+
− hP−φ(ηi), φ(ηi)iL⊕ℓ
−
= 0
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 17
for all η ∈ Ξ \ {ξ0 ⊗ ζ0}. To see this, suppose
η = (ξj0 ⊗ δg1 ⊗ · · · ⊗ δgm ) ⊗ ζ0 ∈ Ξ \ {ξ0 ⊗ ζ0}.
Then, by considering the above expression of φ(η) and the right action of L(Fn)
on (H, ξ0) ∗ (ℓ2(Fn), δe), there exists a unitary operator Uη ∈ L(Fn) such that Uη
commutes with the left actions of M, L(Fn), and N on L+ ⊗ Cn ⊗ H such that
Uηφ(η) = η⊗ei0 ⊗ξ0 for some i0 ∈ {1, . . . , n}. Since every element T ∈ (A∗CFn)⊙B
acts on L− via (T ⊗ ICn ⊗ IH) ⊕ 0H⊗K, P− is (P+ ⊗ ICn ⊗ IH) ⊕ 0(H⊗K)⊕ℓ so
i=1hP−φ(ηi), φ(ηi)iL⊕ℓ
−
Pℓ
as claimed. Hence
η (η ⊗ ei0 ⊗ ξ0), U ∗
i=1hP−U ∗
i=1hP−(η ⊗ ei0 ⊗ ξ0), η ⊗ ei0 ⊗ ξ0iL⊕ℓ
i=1hP+ηi, ηiiL⊕ℓ
−
+
η (η ⊗ ei0 ⊗ ξ0)iL⊕ℓ
−
= Pℓ
= Pℓ
= Pℓ
ℓ
Xη∈Ξ
Xi=1(cid:16)hP+ηi, ηiiL⊕ℓ
+
− hP−φ(ηi), φ(ηi)iL⊕ℓ
− (cid:17) = (τ ∗ τFn)ℓ(P+).
Thus the proof will be complete if the left-hand side of the above equation is in 1
d Z.
To begin we notice for all η ∈ Ξ and i ∈ {1, . . . , ℓ} that
− hP−φ(ηi), φ(ηi)iL⊕ℓ
− hP−,cφ(ηi), φ(ηi)iL⊕ℓ
− hQ−φ(ηi), φ(ηi)iL⊕ℓ
hP+ηi, ηiiL⊕ℓ
= hP+,cηi, ηiiL⊕ℓ
+hQ+ηi, ηiiL⊕ℓ
−
−
+
+
−
+
= 0 + hQ+ηi, ηiiL⊕ℓ
+
− hQ−φ(ηi), φ(ηi)iL⊕ℓ
−
.
Similarly, we obtain for all η ∈ Ξ, i ∈ {1, . . . , ℓ}, and j ∈ N that
hP+,j ηi, ηiiL⊕ℓ
= hP+,c,jηi, ηiiL⊕ℓ
+hQ+,jηi, ηiiL⊕ℓ
− hP−,jφ(ηi), φ(ηi)iL⊕ℓ
− hP−,c,jφ(ηi), φ(ηi)iL⊕ℓ
− hQ−,jφ(ηi), φ(ηi)iL⊕ℓ
−
−
+
+
−
+
− hQ−,jφ(ηi), φ(ηi)iL⊕ℓ
−
±,j)) ∈ 1
d Z for all j ∈ N, and since Q±,j have
+,0)), the following computation is valid:
+
= 0 + hQ+,jηi, ηiiL⊕ℓ
Since trM⊗N(P±,j) = dimM⊗N(A±(L⊕ℓ
finite M⊗N-rank (bounded by dimM⊗N(L⊕ℓ
trM⊗N(Q+,j) − trM⊗N(Q−,j)
i=1hQ+,jηi, ηiiL⊕ℓ
i=1hP+,jηi, ηiiL⊕ℓ
+
+
=Pη∈ΞPℓ
=Pη∈ΞPℓ
= trM⊗N(P+,j) − trM⊗N(P−,j) ∈ 1
d Z.
− hQ−,jφ(ηi), φ(ηi)iL⊕ℓ
− hP−,jφ(ηi), φ(ηi)iL⊕ℓ
−
−
Therefore, since Q+ and Q− have finite M⊗N-rank (specifically bounded above by
dimM⊗N(L⊕ℓ
+,0)), we obtain that
trM⊗N(Q+) − trM⊗N(Q−) = lim
j→∞
trM⊗N(Q+,j) − trM⊗N(Q−,j) ∈
1
d
Z.
Hence
((τM ∗ τFn )⊗τN)ℓ(P+) = Pη∈ΞPℓ
= Pη∈ΞPℓ
i=1(cid:16)hP+ηi, ηiiL⊕ℓ
i=1(cid:16)hQ+ηi, ηiiL⊕ℓ
+
+
= trM⊗N(Q+) − trM⊗N(Q−) ∈ 1
d Z
− hP−φ(ηi), φ(ηi)iL⊕ℓ
− hQ−φ(ηi), φ(ηi)iL⊕ℓ
− (cid:17)
− (cid:17)
18
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
which completes the proof.
(cid:3)
5. Algebraic Cauchy Transforms of Polynomials in Semicircular
Variables
In this section we will demonstrate that the Cauchy transform of any self-adjoint
matricial polynomial of semicircular variables is algebraic (see Theorem 5.4). Know-
ing that the Cauchy transform of a measure is algebraic provides information about
the spectral distribution of operators as seen in Theorem 1.1. To begin, we recall
the notion of a formal power series in commuting variables.
Definition 5.1. Let n ∈ N and let X = {z1, . . . , zn}. For a ring R, a formal power
series in commuting variables X with coefficients in R is a map P : (N ∪ {0})n → R
which we will write as
P =
n
Xj=0 Xkj ≥0
P (k1, . . . , kn)zk1
1 · · · zkn
n .
A formal power series P is called a polynomial if P (k1, . . . , kn) = 0 except for a finite
number of n-tuples (k1, . . . , kn). The set of all formal power series with coefficients
in R will be denoted R[[X]] and the set of all polynomials with coefficients in R
will be denoted R[X].
The set of formal power series over a ring R can be given a ring structure. Indeed,
if addition on R[[X]] is defined coordinate-wise and the product of P, Q ∈ R[[X]] is
defined via the rule
(P + Q)(k1, . . . , kn) =
n
kj
Xj=0
Xℓj =0
P (ℓ1, . . . , ℓn)Q(k1 − ℓ1, . . . , kn − ℓn),
it is elementary to verify that R[[X]] is a ring. Clearly R[X] is a subring of R[[X]]
which enables us to construct the quotient field of R[X]. The quotient field of R[X]
will be denoted R(X).
With the above definitions, we have the following definition essential to this
section.
Definition 5.2. Let n ∈ N, let X = {z1, . . . , zn}, and let R be an integral domain.
A formal power P ∈ R[[X]] is said to be algebraic if there exists an m ∈ N and
{qj}m
j=0 ⊆ R(X) not all zero such that
m
qjP j = 0.
Xj=0
Equivalently, by clearing denominators, we can require {qj}m
all algebraic elements of R[[X]] is denoted Ralg[[X]].
j=0 ⊆ R[X]. The set of
Our main interest lies in demonstrating that certain formal power series relating
to measures are algebraic. Thus we recall the following definition.
Definition 5.3. Let µ be a compactly supported probability measure on R. The
Cauchy transform of µ, denoted Gµ, is the function defined on {z ∈ C Im(z) > 0}
by
Gµ(z) =ZR
1
z − t
dµ(t).
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 19
Notice for large enough z it is clear that Gµ has a Laurent expansion that
z }]]. Thus it makes sense to ask whether Gµ
defines a formal power series in C[[{ 1
is algebraic.
In order to state the main result of this section, we will need some additional
notation. Let M be a finite von Neumann algebra with a faithful normal tracial
state τ . Let A ∈ M be a fixed self-adjoint operator. Since A is a self-adjoint
element in a von Neumann algebra, for each t ∈ R let EA(t) ∈ M be the spectral
projection of A onto (−∞, t]. The spectral density function of A, denoted FA, is
the function on [− kAk , kAk] defined by FA(t) = τ (EA(t)). Clearly FA is a right
continuous function that is bounded above by 1. In turn, FA defines the spectral
measure of A, denoted µA, by the equation
µA((t1, t2]) = FA(t2) − FA(t1).
Notice that µA is a Borel probability measure supported on [− kAk , kAk]. Recall
the spectral measure has the unique property that if f is a continuous function on
the spectrum of A, then
τ (f (A)) =Z kAk
0
f (t) dµA(t).
With the above notation, we have the following important result which provides
information about spectral distributions as indicated in Section 1.
Theorem 5.4. Let n, ℓ ∈ N, let S1, . . . , Sn be freely independent semicircular vari-
ables, let A be the ∗-algebra generated by S1, . . . , Sn, and let A ∈ Mℓ(A) be a
fixed self-adjoint operator. The Cauchy transform of the spectral measure of A is
algebraic.
In order to prove Theorem 5.4 we will mimic the proof of [15, Theorem 3.6]
which proves said result when S1, . . . , Sn are replaced with freely independent Haar
unitaries. In order to mimic the proof in [15], we recall another type of formal power
series in commuting variables.
Definition 5.5. Let S be a ring and let R be a subring of S. It is said that R is
rationally closed in S if for every matrix with entries in R which is invertible when
viewed as a matrix with entries in S, the entries of the inverse lies in R.
The rational closure of R in S, denoted R(R ⊆ S), is the smallest subring of S
containing R that is rationally closed.
For an arbitrary ring R and finite set X, the rational closure R(R[X] ⊆ R[[X]])
is called the ring of rational power series over R and is denoted Rrat[[X]].
It turns out that the key to showing the Cauchy transform GµA is algebraic for
all positive matrices A with entries in a tracial ∗-algebra is intrinsically related to
the following map.
Definition 5.6. Let M be a finite von Neumann algebra with faithful, normal,
tracial state τ . The tracial map on formal power series in one variable is the map
T rM : M[[{z}]] → C[[{z}]] defined by
T rM
Xn≥0
Tnzn
= Xn≥0
τ (Tn)zn.
20
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
In particular, the beginning of the proof of [15, Theorem 3.6] demonstrates the
following.
Lemma 5.7. Let M be a finite von Neumann algebra with faithful, normal, tracial
state τ and let A be a subalgebra of M. If
T rM(Arat[[{z}]]) ⊆ Calg[[{z}]],
then the Cauchy transform GµA is algebraic for every positive matrix A ∈ Mℓ(A)
and any ℓ ∈ N.
Proof. As in the proof of [15, Theorem 3.6], for an arbitrary ℓ ∈ N and positive
matrix A ∈ Mℓ(A), the entries of z(IMℓ(A) − Az)−1 (which can be viewed as
an element of Mℓ(A)[[{z}]] by expanding the result when kAk z < 1) lie in the
rational closure Arat[[{z}]]. By assumption, the formal power series
ℓ
q(z) := T rMℓ(M)(cid:0)z(IMn(A) − Az)−1(cid:1) =
Xj=1
T rM((z(IMℓ(A) − Az)−1)jj )
is an element of Calg[[{z}]]. Thus q(z−1) is an element of Calg[[{ 1
the canonical trace on Mℓ(M), it is well-known that
z }]]. If τMℓ(M) is
GµA (z) = τMℓ(M)((zIMn(A) − A)−1) = q(z−1)
in the domain {z ∈ C Im(z) > 0, z > kAk}. Hence GµA ∈ Calg[[{ 1
desired.
z }]] as
(cid:3)
Thus the proof of Theorem 5.4 will be complete provided the assumptions of
Lemma 5.7 can be verified. Following [15], it is necessary to examine formal power
series in non-commuting variables.
Definition 5.8. Let X be a finite set (which will be called an alphabet) and let
W (X) denote the set of all words with letters in X. The empty word will be
denoted by e. For a ring R, a formal power series with non-commuting variables X
with coefficients in R is a map P : W (X) → R which we will write as
P = Xw∈W (X)
P (w)w.
A formal power series P is called a polynomial P (w) = 0 except for a finite number
of words w ∈ W (X). The set of all formal power series with coefficients in R will
be denoted RhhXii and the set of all polynomials with coefficients in R will be
denoted RhXi.
The set of formal power series over a ring R can be given a ring structure. Indeed,
if addition on RhhXii is defined coordinate-wise, and multiplication is defined via
the rule
Xw∈W (X)
P (w)w
·
Xw∈W (X)
Q(w)w
= Xw∈W (X)
Xu,v∈W (X),uv=w
P (u)Q(v)
w
(notice that for each w ∈ W (X) there are a finite number of pairs u, v ∈ W (X)
such that w = uv), it is elementary to verify that RhhXii is a ring. Thus it makes
sense to consider the rational closure of RhXi inside RhhXii which will be denoted
RrathhXii.
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 21
As with formal power series in commuting variables, there is a notion of an
algebraic formal power series in non-commuting variables. The definition of such
a formal power series is more technical than in the commutative case and is based
on the following definition.
Definition 5.9 (Schutzenberger). Let X := {x1, . . . , xn} be an alphabet and let
Z := {z1, . . . , zm} be an alphabet disjoint from X. A proper algebraic system over
a ring R is a set of equations zi = pi(x1, . . . , xn, z1, . . . , zm) for i ∈ {1, . . . , m} where
each pi is an element of RhX ∪ Zi that has no constant term nor term of the form
αzj where α ∈ R and j ∈ {1, . . . , n}.
A solution to a proper algebraic system is an m-tuple (P1, . . . , Pm) ∈ RhhXiim
such that Pj(e) = 0 and pj(x1, . . . , xn, P1, . . . , Pm) = Pj for all j ∈ {1, . . . , m}.
Definition 5.10. A formal power series P ∈ RhhXii is said to be algebraic if
P − P (e)e is a component of the solution of a proper algebraic system. The set all
algebraic formal power series in RhhXii will be denoted by RalghhXii.
In order to prove the assumptions of Lemma 5.7 hold in the context of Theorem
5.4, the proof of [15, Theorem 2.19(ii)] will be mimicked. To do so, it is necessary
to show that a certain formal power series in non-commuting variables is algebraic.
The following formula involving traces of words of semicircular variables plays a
crucial role.
Lemma 5.11 (See [22, Section 3]). Let n ∈ N, let S1, . . . , Sn be freely independent
semicircular variables (with second moments 1), let A be the ∗-algebra generated
by S1, . . . , Sn, let τ be the canonical trace on A, and let X := {x1, . . . , xn} be an
alphabet. For each j ∈ {1, . . . , n} and w ∈ W (X),
τ (Sj w(S1, . . . , Sn)) =
Xu,v∈W (X),w=uxj v
τ (u(S1, . . . , Sn))τ (v(S1, . . . , Sn))
where, for a word w0 ∈ W (X), w0(S1, . . . , Sn) is the element of A obtained by
substituting Si for xi.
Lemma 5.12. With the notation as in Lemma 5.11, the formal power series
Psemi ∈ ChhXii defined by
Psemi := Xw∈W (X)
τ (w(S1, . . . , Sn))w
is algebraic.
Proof. By Lemma 5.11 we easily obtain that
Psemi − e
= Pn
= Pn
= Pn
= Pn
j=1Pw∈W (X) τ (Sj w(S1, . . . , Sn))xj w
j=1Pw,u,v∈W (X),w=uxjv τ (u(S1, . . . , Sn))τ (v(S1, . . . , Sn))xj uxjv
j=1Pu,v∈W (X) τ (u(S1, . . . , Sn))τ (v(S1, . . . , Sn))xj uxjv
j=1 xjPsemixj Psemi.
Hence it is elementary to verify that Psemi − e is a solution to the proper algebraic
system
z =
xjzxjz + x2
j z + xjzxj + x2
j .
n
Xj=1
Thus Psemi is algebraic by definition.
(cid:3)
22
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
Using Lemma 5.12 it is easy to verify the proof of [15, Theorem 2.19(ii)] general-
izes enough to complete the proof of Theorem 5.4. We will only sketch the changes
to the proof of [15, Theorem 2.19(ii)] as it nearly follows verbatim.
Proof of Theorem 5.4. Let M be the von Neumann algebra generated by S1, . . . , Sn.
By Lemma 5.7 it suffices to show that the tracial map on formal power series
T rM : M[[{z}]] → C[[{z}]] has the property that
T rM(Arat[[{z}]]) ⊆ Calg[[{z}]].
Let S := {x1, . . . , xn} be an alphabet. As in the proof of [15, Theorem 2.19(ii)],
there is a canonical way to view
(ChSi)rat[[{z}]] ⊆ (C(z))rathhSii.
Consider the injective homomorphisms π : W (S) → A uniquely defined by
π(xj ) = Sj for all j ∈ {1, . . . , n}. Clearly π extends to a homomorphism π :
ChSi → A and thus also extends to a homomorphism π : (ChSi)[[{z}]] → A[[{z}]]
by applying π coordinate-wise.
Let P ∈ Arat[[{z}]] be arbitrary. Using algebraic properties, the proof of [15,
Theorem 2.19(ii)] implies that
P ∈ π ((ChSi)rat[[{z}]]) .
Choose P ∈ (ChSi)rat[[{z}]] ⊆ (C(z))rat[[S]] such that π(P ) = P . Recall that
Psemi := Xw∈W (S)
τ (w(S1, . . . , Sn))w ∈ CalghhSii ⊆ (C(z))alghhSii
by Lemma 5.12. Hence the Haadamard Product
P ⊙ Psemi := Xw∈W (S)
P (w)Psemi(w)w = Xw∈W (S)
τ (w(S1, . . . , Sn))P (w)w
is an element of (C(z))alghhSii by a theorem of Schutzenberger from [18].
Since P ⊙ Psemi ∈ (C(z))alghhSii, if we substitute 1 ∈ C for every element of S
we obtain a well-defined power series in C[[{z}]]. Indeed if
pm(S1, . . . , Sn)zm
P = Xm≥0
for some non-commutative polynomials pm in n variables, then
(pm(x1, . . . , xn) + qm(x1, . . . , xn))zm
P = Xm≥0
for some non-commutative polynomials qm in n variables such that qm(S1, . . . , Sn) =
0. Hence
P ⊙ Psemi = Xw∈W (S)
τ (w(S1, . . . , Sn))
Xm≥0
(coef (pm, w) + coef (qm, w))zm
w
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 23
where coef (p, w) is the element of C that is the coefficient of w in p. Therefore, by
replacing each w with the scalar 1, we obtain
Pw∈W (S) τ (w(S1, . . . , Sn))(cid:16)Pm≥0(coef (pm, w) + coef (qm, w))zm(cid:17)
= Pm≥0 τ(cid:16)Pw∈W (S)(coef (pm, w) + coef (qm, w))w(S1, . . . , Sn)(cid:17) zm
= Pm≥0 τ (pm(S1, . . . , Sn) + qm(S1, . . . , Sn)) zm
= Pm≥0 τ (pm(S1, . . . , Sn))zm = T rM(P )
as desired. Thus the proof of [15, Theorem 2.19(ii)] implies that T rM(P ) is an
element of Calg[[{z}]] as desired.
(cid:3)
With the proof of Theorem 5.4 complete, we turn our attention to further infor-
mation that Sauer's results from [15] imply. The main purpose of [15] was to show
the rationality and positivity of the Novikov-Shubin invariant for matrices with
entries in the group algebra of a virtually free group. In particular, the Novikov-
Shubin invariants are well-defined for any finite, tracial von Neumann algebra.
Definition 5.13. Let M be a finite von Neumann algebra with faithful, normal,
tracial state τ . For a positive operator A ∈ M with spectral distribution FA, the
Novikov-Shubin invariant α(A) ∈ [0, ∞] ∪ {∞+} of A is defined as
α(A) :=( lim inf t→0+
∞+
ln(FA(t)−FA(0))
ln(t)
if FA(t) > FA(0) for all t > 0
otherwise
.
For a positive operator A in a finite von Neumann algebra M, it is easy to see
that α(A) = ∞+ implies that zero is isolated in the spectrum of A. Furthermore,
if α(A) = λ ∈ [0, ∞), then FA(t) − FA(0) behaves like tλ as t tends to zero.
The Novikov-Shubin invariants are of interest in the context of Theorem 5.4 due
to the following result which is directly implied by the proof of [15, Theorem 3.6].
Lemma 5.14 (See [15, Theorem 3.6] for a proof). Let M be a finite von Neumann
algebra with faithful, normal, tracial state τ . Let A ∈ M be a positive operator and
let µA is the spectral measure of A. If the Cauchy transform GµA is algebraic, then
the Novikov-Shubin invariant α(A) is a non-zero rational number or ∞+.
The Novikov-Shubin invariants are of interest in terms of determining the decay
of the spectral density function at zero due to the following result.
Lemma 5.15 (See [10, Theorem 3.14(4)]). Let M be a finite von Neumann algebra
with faithful, normal, tracial state τ . If A ∈ M is a positive operator and FA is the
spectral density function of A, then
lim
ǫ→0Z kAk
ǫ
1
t
(FA(t) − FA(0)) dt < ∞
provided α(A) 6= 0.
Proof. If α(A) = ∞+, then FA(t) − FA(0) is a right continuous function bounded
that is zero on a neighbourhood of zero. Hence the result follows. If α(A) ∈ (0, ∞],
then it is trivial to verify from Definition 5.13 that there exists a δ > 0 and an
λ ∈ (0, α(A)) such that F (t) − F (0) ≤ tλ for all 0 ≤ t ≤ δ. Hence
Thus the result follows as FA(t)−FA(0) is a right continuous function bounded. (cid:3)
tλ−1 dt < ∞.
0 ≤Z δ
0
1
t
(FA(t) − FA(0)) dt ≤Z δ
0
24
DIMITRI SHLYAKHTENKO AND PAUL SKOUFRANIS
Furthermore, the following result provides information on how to extract infor-
mation from the conclusion of Lemma 5.15 to obtain information about integrating
logarithms against the spectral measure.
Lemma 5.16 (See [10, Lemma 3.15(1)]). Let M be a finite von Neumann algebra
with faithful, normal, tracial state τ . If A ∈ M is a positive operator, FA is the
spectral density function of A, and µA is the spectral measure of A, then
if and only if
1
t
lim
ǫ→0Z kAk
ǫ→0Z kAk
lim
ǫ
ǫ
(FA(t) − FA(0)) dt < ∞
ln(t) dµA(t) > −∞.
Combining the above results, we obtain the following.
Theorem 5.17. Let n, ℓ ∈ N, let X1, . . . , Xn be freely independent semicircular
variables or freely independent Haar unitaries, and let A be the ∗-algebra generated
by X1, . . . , Xn. Then
lim
ǫ→0Z kAk
ǫ
ln(t) dµA(t) > −∞
for all positive A ∈ Mm(A) \ {0}. Furthermore, if µA does not have an atom at
zero (e.g. when ℓ = 1 by Theorem 3.4), then
Z kAk
0
ln(t) dµA(t) > −∞.
References
[1] G. Anderson and O. Zeitouni, A law of large numbers for finite-range dependent random
matrices, Communications on Pure and Applied Mathematics 61 (2008), no. 8, 1118-1154.
[2] M. F. Atiyah, Elliptic operators and compact groups, Lecture Notes in Math., vol. 401,
Springer, 1974.
[3] S. T. Belinschi, T. Mai, and R. Speicher, Analytic subordination theory of operator-valued free
additive convolution and the solution of a general random matrix problem, 24 pp., available
at arXiv:1303.3196.
[4] H. Bercovici and D. Voiculescu, Regularity questions for free convolution, Oper. Theory Adv.
Appl. 104 (1998), 37-47.
[5] A. Connes, Noncommutative geometry, Academic Press Inc., San Diego, CA, 1994.
[6] R. I. Grigorchuk and A. Zuk, The lamplighter group as a group generated by a 2-state au-
tomaton, and its spectrum, Geom. Dedicata 87 (2001), no. 1, 209-244.
[7] U. Haagerup, H. Schultz, and S. Thorbjørnsen, A random matrix approach to the lack of
projections in C ∗
red(F2), Advances in Mathematics 204 (2006), no. 1, 1-83.
[8] F. Hiai and D. Petz, The Semicircle Law, Free Random Variables and Entropy, Mathematical
Surveys and Monographs, vol. 77, American Mathematical Society, 2000.
[9] P. A. Linnell, Division rings and group von Neumann algebras, Forum Math. 5 (1993), no. 5,
561-576.
[10] W. Luck, L2-Invariants: Theory and Applications to Geometry and K-Theory, A Series of
Modern Surveys in Mathematics, vol. 44, Springer, 2002.
[11] M. Pimsner and D. Voiculescu, Exact sequences for K-groups and Ext-groups of certain
cross-product C∗-algebras, J. Operator Theory. 4 (1980), no. 1, 93-118.
[12]
, K-groups of reduced crossed products by free groups, J. Operator Theory. 8 (1982),
no. 1, 131-156.
[13] N. R. Rao and A. Edelman, The polynomial method for random matrices, Foundations of
Computational Mathematics 8 (2008), no. 6, 649-702.
FREELY INDEPENDENT RANDOM VARIABLES WITH NON-ATOMIC DISTRIBUTIONS 25
[14] H. Reich, Group von Neumann algebras and related algebras, Ph.D. Thesis, Universitat
Gottingen, http://www.mi.fu-berlin.de/math/groups/top/members/publ/diss.pdf.
[15] R. Sauer, Power series over the group ring of a free group and applications to Novikov-Shubin
invariants, High-dimensional manifold topology (2003), 449-468.
[16] T. Schick, Integrality of L2-Betti numbers, Math. Ann. 317 (2000), no. 4, 727-750.
[17]
[18] M. P. Schutzenberger, On a theorem of R. Jungen, Proc. Amer. Math. Soc. 13 (1962), no. 6,
, Integrality of L2-Betti numbers, 24 pp., available at arXiv:math/0001101v4.
885-890.
[19] D. Voiculescu, Addition of certain non-commuting random variables, J. Funct. Anal. 66
(1986), no. 3, 323-346.
[20]
[21]
[22]
, Multiplication of certain non-commuting random variables, J. Operator Theory 18
(1987), no. 2, 223-235.
, The analogues of entropy and of Fisher's information measure in free probability
theory, I, Communications in mathematical physics 155 (1993), no. 1, 71-92.
, The analogues of entropy and of Fisher's information measure in free probability
theory V. Noncommutative Hilbert Transforms, Inventiones mathematicae 132 (1998), no. 1,
189-227.
[23] D. V. Voiculescu, K. Dykema, and A. Nica, Free Random Variables, CRM Monograph Series,
vol. 1, American Mathematical Society, 1992.
Department of Mathematics, UCLA, Los Angeles, California, 90095, USA
E-mail address: [email protected]
Department of Mathematics, UCLA, Los Angeles, California, 90095, USA
E-mail address: [email protected]
|
1511.07987 | 2 | 1511 | 2016-02-22T20:25:24 | On a generalized Semrl's theorem for weak-2-local derivations on B(H) | [
"math.OA"
] | We prove that, for every complex Hilbert space $H$, every weak-2-local derivation on $B(H)$ or on $K(H)$ is a linear derivation. We also establish that every weak-2-local derivation on an atomic von Neumann algebra or on a compact C$^*$-algebra is a linear derivation. | math.OA | math |
ON A GENERALIZED SEMRL'S THEOREM FOR
WEAK-2-LOCAL DERIVATIONS ON B(H)
JUAN CARLOS CABELLO AND ANTONIO M. PERALTA
Abstract. We prove that, for every complex Hilbert space H, every weak-2-
local derivation on B(H) or on K(H) is a linear derivation. We also establish
that every weak-2-local derivation on an atomic von Neumann algebra or on a
compact C∗-algebra is a linear derivation.
1. Introduction
Let S be a subset of the space L(X, Y ) of all linear maps between Banach
spaces X and Y . Following [2, 3] and [4], we shall say that a (non-necessarily
linear nor continuous) mapping ∆ : X → Y is a weak-2-local S map (respectively,
a 2-local S-map) if for each x, y ∈ X and φ ∈ Y ∗ (respectively, for each x, y ∈ X),
there exists Tx,y,φ ∈ S, depending on x, y and φ (respectively, Tx,y ∈ S, depending
on x and y), satisfying
φ∆(x) = φTx,y,φ(x), and φ∆(y) = φTx,y,φ(y)
(respectively, ∆(x) = Tx,y(x), and ∆(y) = Tx,y(y)).
When A is a Banach algebra and S is the set of derivations (respectively,
homomorphisms or automorphisms) on A, weak-2-local S maps on A are called
weak-2-local derivations (respectively, weak-2-local homomorphisms or weak-2-
local automorphisms). 2-local ∗-derivations and 2-local ∗-homomorphisms on C∗-
algebras are similarly defined. We recall that a ∗-derivation on a C∗-algebra A is
a derivation D : A → A satisfying D(a∗) = D(a)∗ (a ∈ A).
The notion of 2-local derivations goes back, formally, to 1997 when P. Semrl
introduces the formal definition and proves that, for every infinite dimensional
separable Hilbert space H, every 2-local automorphism (respectively, every 2-
local derivation) on B(H) is an automorphism (respectively, a derivation). Sh.
Ayupov and K. Kudaybergenov proved that Semrl's theorem also holds for arbi-
trary Hilbert spaces [2]. In 2014, Ayupov and Kudaybergenov prove that every
2-local derivation on a von Neumann algebra is a derivation (see [3]).
Date: October 11, 2018.
2010 Mathematics Subject Classification. 47B49, 46L05, 46L40, 46T20, 47L99.
Key words and phrases. derivation; 2-local linear map; 2-local symmetric maps; 2-local ∗-
derivation; 2-local derivation; weak-2-local derivation.
Authors partially supported by the Spanish Ministry of Economy and Competitiveness and
European Regional Development Fund project no. MTM2014-58984-P and Junta de Andaluc´ıa
grants FQM375 and FQM290.
1
2
J.C. CABELLO AND A.M. PERALTA
Results on weak-2-local maps are even more recent. In a very recent contribu-
tion, M. Niazi and the second author of this note prove the following generaliza-
tion of the previously mentioned results.
Theorem 1.1. [7, Theorem 3.10] Let H be a separable complex Hilbert space.
Then every (non-necessarily linear nor continuous) weak-2-local ∗-derivation on
B(H) is linear and a ∗-derivation.
(cid:3)
The same authors prove that for finite dimensional C∗-algebras the conclusions
are stronger:
Theorem 1.2. [7, Corollary 2.13] Every weak-2-local derivation on a finite di-
mensional C∗-algebra is a linear derivation.
(cid:3)
Let ∆ : A → B be a mapping between C∗-algebras. We consider a new
mapping ∆♯ : A → B given by ∆♯(x) := ∆(x∗)∗ (x ∈ A). Obviously, ∆♯♯ = ∆,
∆(Asa) ⊆ Bsa for every ∆ satisfying ∆ = ∆♯, where Asa and Bsa denote the
self-adjoint parts of A and B, respectively. The mapping ∆ is linear if and
only if ∆♯ enjoys the same property. The mapping ∆ is called symmetric if
∆♯ = ∆ (equivalently, ∆(x∗) = ∆(x)∗, for all x ∈ X). Henceforth, the set of all
symmetric maps from A into B will be denoted by S(A, B). Weak-2-local S(A, B)
maps between A and B will be called weak-2-local symmetric maps, while 2-local
L(A, B) maps between A and B will be called weak-2-local linear maps.
The study on weak-2-local maps has been also pursued in [4], where we obtained
that every weak-2-local symmetric map between C∗-algebras is linear (see [4,
Theorem 2.5]). Among the consequences of this result, we also establish that
every weak-2-local ∗-derivation on a general C∗-algebra is a (linear) ∗-derivation
(cf. [4, Corollary 2.10]).
One of the main problems that remains unsolved in this line reads as follows:
Problem 1.3. Is every weak-2-local derivation on a general C∗-algebra A a
derivation?
We shall justify later that every weak-2-local derivation ∆ on A writes as a
linear combination ∆ = ∆1 + ∆2, where ∆1 = ∆+∆♯
are weak-
2-local derivations and symmetric maps. Thus, we shall deduce that the above
Problem 1.3 is equivalent to the following question.
and ∆2 = ∆−∆♯
2
2i
Problem 1.4. Let ∆ : A → A be a weak-2-local derivation on a C∗-algebra which
is also a symmetric map (i.e. ∆♯ = ∆). Is ∆ a weak-2-local symmetric map? --
or, equivalently, Is ∆ a linear derivation?
The above problems are natural questions arisen in an attempt to generalize
the above mentioned results by Semrl's [10] and Ayupov and Kudaybergenov
[2, 3]. Both remain open even in the intriguing case of A = B(H).
In this paper we provide a complete positive answer to both problems in several
cases. In Theorem 3.1 we prove that every weak-2-local derivation on A = B(H)
is a linear derivation. This generalizes the results in [10], [2, 3] and [7]. We
also establish that this weak-2-local stability of derivations is also true when A
A GENERALIZED SEMRL'S THEOREM FOR WEAK-2-LOCAL DERIVATIONS
3
coincides with K(H) (see Theorem 3.2), when A is an atomic von Neuman algebra
(cf. Corollary 3.5), and when A is a compact C∗-algebra (cf. Corollary 3.6). The
techniques and arguments provided in this note are completely new compared
with those in previous references. The note is divided in two main sections. In
Section 2 we establish a certain boundedness principle showing that for each weak-
2-local derivation ∆ on B(H), or on K(H), the mappings a 7→ pF ∆(pF apF )pF
are uniformly bounded when pF runs in the set of all finite-rank projections on H
(compare Theorems 2.15 and 2.17). In Section 3 we derive the main results of the
paper from an identity principle, which assures that a weak-2-local derivation ∆
on B(H) with ∆♯ = ∆, coincide with a ∗-derivation D if and only if they coincide
on every finite-rank projection in B(H) (see Theorem 2.9).
2. Boundedness of weak-2-local derivations on the lattice of
projections in B(H)
We recall some basic properties on weak-2-local maps which have been bor-
rowed from [4] and [6].
Lemma 2.1. ([4, Lemma 2.1], [6, Lemma 2.1]) Let X and Y be Banach spaces
and let S be a subset of the space L(X, Y ). Then the following properties hold:
(a) Every weak-2-local S map ∆ : X → Y is 1-homogeneous, that is, ∆(λx) =
λ∆(x), for every x ∈ X, λ ∈ C;
(b) Suppose there exists C > 0 such that every linear map T ∈ S is continuous
with kT k ≤ C. Then every weak-2-local S map ∆ : X → Y is C-Lipschitzian,
that is, k∆(x) − ∆(y)k ≤ Ckx − yk, for every x, y ∈ X;
(c) If S is a (real) linear subspace of L(X, Y ), then every (real) linear combination
of weak-2-local S maps is a weak-2-local S map;
(d) Suppose A and B are C∗-algebras and S is a real linear subspace of L(A, B).
If a mapping ∆ : A → B is a weak-2-local S map then for each ϕ ∈ B∗
sa
and every x, y ∈ A, there exists Tx,y,ϕ ∈ S satisfying ϕ∆(x) = ϕTx,y,ϕ(x) and
ϕ∆(y) = ϕTx,y,ϕ(y).
(e) Suppose A and B are C∗-algebras and S is a real linear subspace of L(A, B)
with S ♯ = S (in particular when S = S(A, B) is the set of all symmetric
linear maps from A into B). Then a mapping ∆ : A → B is a weak-2-local
S map if and only if ∆♯ is a weak-2-local S map.
(cid:3)
Henceforth, H will denote an arbitrary complex Hilbert space. The symbols
B(H) and K(H) will denote the C∗-algebras of all bounded and compact linear
operators on H, respectively.
If H is finite dimensional, then every weak-2-
local derivation on B(H) is a linear derivation (compare Theorem 1.2). We may
therefore assume that H is infinite dimensional.
Following standard notation, an element x in a C∗-algebra A is said to be
finite (respectively, compact) in A, if the wedge operator x ∧ x : A → A, given
by x ∧ x(a) = xax, is a finite-rank (respectively, compact) operator on A.
It
is known that the ideal F (A) of finite elements in A coincides with Soc(A), the
socle of A, that is, the sum of all minimal right (equivalently left) ideals of A,
and that K(A) = Soc(A) is the ideal of compact elements in A. Moreover, if H
4
J.C. CABELLO AND A.M. PERALTA
is a Hilbert space, then F (L(H)) = F (H) and K(L(H)) = K(H) are the ideals
of finite-rank and compact elements in B(H), respectively.
Suppose ∆ : B(H) → B(H) is a weak-2-local derivation. By [7, Lemma 3.4]
we know that ∆(K(H)) ⊆ K(H) and ∆K(H) : K(H) → K(H) is a weak-2-local
derivation. Proposition 3.1 in [7] proves that ∆(a + b) = ∆(a) + ∆(b), for every
a, b ∈ F (H).
Lemma 2.2. [7, Lemma 3.4 and Proposition 3.1] Let ∆ : B(H) → B(H) be a
weak-2-local derivation. Then ∆F (H) : F (H) → B(H) is linear.
(cid:3)
Let us revisit some basic facts on commutators. We recall that every derivation
on a C∗-algebra is continuous (cf.
[9, Lemma 4.1.3]). A celebrated result of S.
Sakai establishes that every derivation on a von Neumann algebra M is inner,
that is, if D : M → M is a derivation then there exists z ∈ M such that
D(x) = [z, x] = zx − xz for every x ∈ M (see [9, Theorem 4.1.6]). The element
z given by Sakai's theorem is not unique; however, we can choose z satisfying
kzk ≤ kDk.
Let us consider two elements z, w in a C∗-algebra A such that the derivations
[z, .] and [w, .] coincide as linear maps on A. Since [z, x] = [w, x] for every x ∈ A,
we deduce that z − w lies in the center of A. The reciprocal statement is also
true, therefore [z, .] = [w, .] on A if and only if z − w lies in the center, Z(A), of A.
It is known that a derivation of the form [z, .] is symmetric (i.e. a ∗-derivation)
if and only if z = w + c, where w = −w∗ and c lies in the center of A.
From now on, the set of all finite dimensional subspaces of H will be denoted
by F(H). We consider in F(H) the natural order given by inclusion. For each
F ∈ F(H), pF will denote the orthogonal projection of H onto F .
Lemma 2.3. Let ∆ : B(H) → B(H) be a weak-2-local derivation. For each
F ∈ F(H) there exists zF ∈ pF B(H)pF satisfying
pF ∆(pF apF )pF = [zF , pF apF ],
for every a ∈ B(H).
zF ∈ pF B(H)pF satisfying zF = −z∗
.
F
If ∆ is symmetric (i.e. ∆♯ = ∆), then we can choose
Proof. Let F be a finite dimensional subspace of H. By [6, Proposition 2.7] the
mapping pF ∆pF pF B(H)pF
: pF B(H)pF → pF B(H)pF , a 7→ pF ∆(pF apF )pF is a
weak-2-local derivation. Having in mind that pF B(H)pF is a finite dimensional
C∗-algebra, we deduce from Theorem 1.2 that pF ∆pF pF B(H)pF
is a linear deriva-
tion. By Sakai's theorem there exists zF ∈ pF B(H)pF satisfying the desired
conclusion.
If ∆ is symmetric we can easily check that pF ∆pF pF B(H)pF
also is symmetric,
and hence a ∗-derivation on pF B(H)pF . In this case, we can obviously replace zF
with
(cid:3)
to get the final statement in the lemma.
zF −z∗
F
2
Remark 2.4. Let ∆ : B(H) → B(H) be a weak-2-local derivation with ∆♯ = ∆,
and let F be a subspace in F(H). It is clear that the element zF given by the
A GENERALIZED SEMRL'S THEOREM FOR WEAK-2-LOCAL DERIVATIONS
5
above Lemma 2.3 is not unique. We can consider the set
Given z1, z2 ∈ [zF ] it follows that z1 − z2 ∈ Z(pF B(H)pF ) = CpF , and since
(z1 − z2)∗ = −(z1 − z2), it follows that there exists λ ∈ R such that z2 = z1 + iλpF .
[zF ] :=nz ∈ pF B(H)pF : z∗ = −z and pF ∆pF pF B(H)pF
It is easy to check that there exists a uniqueezF ∈ [zF ] satisfying
From now on, given an element a in a C∗-algebra A, the spectrum of a will be
denoted by σ(a). Our next remark gathers some information about the norm of
an inner ∗-derivation on B(H).
kezF k = min{kzk : z ∈ [zF ]}.
= [z, .]o.
Remark 2.5. Let z be an element in B(H). J.G. Stampfli proves in [11, Theorem
4] that
(cid:13)(cid:13)(cid:13)[z, .](cid:13)(cid:13)(cid:13) = inf
λ∈C(cid:13)(cid:13)(cid:13)z − λIdH(cid:13)(cid:13)(cid:13),
where k[z, .]k denotes the norm of the inner derivation [z, .] in B(B(H)).
For a compact subset K ⊂ C, the radius, ρ(K), of K is the radius of the
smallest disk containing K. In general, two times the radius of a compact set
K does not coincide with its diameter. In general, 2ρ(K) ≥ diam(K). However,
when K ⊂ R or K ⊂ iR, we can easily see that 2ρ(K) = diam(K).
When z is a normal operator in B(H) we further know that k[z, .]k = 2ρ(σ(z))
(compare [11, Corollary 1]). In particular, for each z in B(H) with z = z∗ or
z = −z∗, we have
k[z, .]k = 2ρ(σ(z)) = diam(σ(z)) ≤ 2kzk.
(1)
Let us observe that if 0 ∈ σ(z), then kzk ≤diam(σ(z)), for every z = ±z∗.
Given a projection p in a unital C∗-algebra A we shall denote by p⊥ the pro-
jection 1 − p.
Lemma 2.6. Let ∆ : B(H) → B(H) be a weak-2-local derivation with ∆♯ = ∆.
Suppose F1 and F2 are finite dimensional subspaces of H with F1 ⊆ F2. We
employ the notation given in Remark 2.4. Then for each z1 ∈ (cid:2)zF1(cid:3) and each
z2 ∈(cid:2)zF2(cid:3) we have
[z1, .] = [pF1
z2pF1
, .]
as operators on pF1
B(H)pF1
and z2) such that z1 + iλpF1
and diam(σ(z1)) =diam(σ(z′
. Consequently, there exists a real λ (depending on z1
= pF1
. In particular, diam(σ(z1)) ≤diam(σ(z2))
1)) for every z1, z′
z2pF1
].
1 ∈ [zF1
Proof. By Lemma 2.3 and Remark 2.4 we have
pF1
∆(pF1
apF1
)pF1
= [z1, pF1
apF1
],
and
pF2
∆(pF2
for every a ∈ B(H). Since pF1
)pF2
= [z2, pF2
, it follows that
apF2
],
apF2
≤ pF2
[z2, pF1
[pF1
z2pF1
, pF1
apF1
] = pF1
apF1
]pF1
= pF1
pF2
∆(pF1
apF1
)pF2
pF1
6
J.C. CABELLO AND A.M. PERALTA
for every a ∈ B(H), which proves the first statement in the lemma.
= pF1
∆(pF1
apF1
)pF1
= [z1, pF1
apF1
]
B(H)pF1
Since Z(pF1
that z1 + iλpF1
z2pF1
By Remark 2.5 we have
= pF1
) = CpF1
, z∗
1 = −z1 and z∗
(compare Remark 2.4).
2 = −z2, there exists λ ∈ R such
diam(σ(z1)) = k[z1, .]k = k[pF1
z2pF1
, .](pF1
B(H)pF1
k ≤ k[z2, .]k = diam(σ(z2))
)k
= kpF1
[z2, pF1
.pF1
]pF1
(cid:3)
Proposition 2.7. Let ∆ : B(H) → B(H) be a weak-2-local derivation with
∆♯ = ∆. Suppose that the set Diam(∆) = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H)}
is unbounded. Then for each G ∈ F(H), the set
Diam+
G = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H), F ⊇ G}
is unbounded.
Proof. Let us fix an arbitrary G ∈ F(H). For each F ∈ F(H) we can find
K ∈ F(H) with G, F ⊆ K. Applying Lemma 2.6 we have
diam(σ(wF )), diam(σ(wG)) ≤ diam(σ(wK )),
for every wF ∈ [zF ], wG ∈ [zG] and wK ∈ [zK ]. The unboundedness of Diam
implies the same property for Diam+
G.
(cid:3)
2.1. An identity principle for weak-2-local derivations. Let ∆ : B(H) →
B(H) be a weak-2-local derivation with ∆♯ = ∆. Suppose there exists G ∈ F(H)
such that the set
Diam−
G⊥ =(cid:8)diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H), pF ≤ p⊥
G(cid:9)
is bounded. For each F ∈ F(H) with pF ≤ p⊥
to satisfy
kezF k ≤ diam(σ(ezF )) ≤ 2kezF k.
is bounded in p⊥
G with z0 = −z∗
≤p⊥
G
F
B(H)p⊥
G .
G , the element ezF has been chosen
0 and a subnet (ezF )F ∈Λ converging
G . By Alaoglu's theo-
B(H)p⊥
G
F ∈F(H),p
Therefore, the net (ezF )
rem, we can find z0 ∈ p⊥
B(H)p⊥
to z0 in the weak∗-topology of p⊥
If the set Diam(∆) = Diam−
G
G
bounded, we can similarly define, via Alaoglu's theorem, an element z0 = −z∗
{0}⊥ = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H)} is
0 ∈
Proposition 2.8. Let ∆ : B(H) → B(H) be a weak-2-local derivation with
∆♯ = ∆. Suppose there exists G ∈ F(H) such that the set
B(H) which is the weak∗-limit of a convenient subnet of (ezF )F ∈F(H).
G⊥ =(cid:8)diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H), pF ≤ p⊥
G(cid:9)
G (z0 = −z∗
0) be the element determined in the
G = [z0, p], for every projection p ∈ F (H) with
B(H)p⊥
G ∆(p)p⊥
Diam−
is bounded, and let z0 ∈ p⊥
previous paragraph. Then p⊥
p ≤ p⊥
G .
G
A GENERALIZED SEMRL'S THEOREM FOR WEAK-2-LOCAL DERIVATIONS
7
If the set Diam(∆) = Diam−
{0}⊥ = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H)} is
bounded, then ∆(p) = [z0, p], for every projection p ∈ F (H).
Proof. Let us fix a finite-rank projection p ∈ F (H) with p ≤ p⊥
converges to z0 in the weak∗ topology of p⊥
fixed before Proposition 2.8, there exists F0 ∈ Λ such that p ≤ pF0
(we observe
that, under these hypothesis, there exits a monotone final function h : F(H) → Λ
G . Since (ezF )F ∈Λ
G , where (ezF )F ∈Λ is the subnet
which defines the subnet). The subnet (ezF )F0⊆F ∈Λ converges to z0 in the weak∗
Clearly, the net (pF )F ∈F(H) converges to the projection p⊥
topology of B(H).
B(H)p⊥
G
ogy of B(H). Therefore the subnet (pF )F0⊆F ∈Λ → p⊥
B(H). Since for each F ∈ Λ with F0 ⊆ F we have p ≤ pF0
[7, Lemma 3.2], Lemma 2.3 and Remark 2.4, that
G in the strong∗ topol-
G in the strong∗ topology of
≤ pF , we deduce, via
pF ∆(p)pF = pF ∆(pF ppF )pF = pF [ezF , pF ppF ]pF = [ezF , p].
(2)
It is known that the product of every von Neumann algebra is jointly strong∗-
continuous on bounded sets (see [9, Proposition 1.8.12]), we thus deduce that
G in the strong∗ topology of B(H), and
the net (pF ∆(p)pF )F0⊆F ∈Λ → p⊥
G also in the weak∗ topology (compare [9,
hence (pF ∆(p)pF )F0⊆F ∈Λ → p⊥
G ∆(p)p⊥
Theorem 1.8.9]). This shows that the left-hand side in (2) converges to p⊥
in the weak∗-topology of B(H).
G ∆(p)p⊥
G ∆(p)p⊥
G
Finally, the separate weak∗-continuity of the product of B(H) (cf. [9, Theorem
G = [z0, p] in
G = [z0, p] as we desired. The second
(cid:3)
1.7.8]) shows that the right-hand side in (2) converges to p⊥
the weak∗-topology. Therefore, p⊥
statement follows from the same arguments.
G [z0, p]p⊥
G ∆(p)p⊥
We can state now an identity principle for weak-2-local derivations on B(H).
Theorem 2.9. Let ∆ : B(H) → B(H) be a weak-2-local derivation with ∆♯ = ∆.
Let p0 ∈ F (H) be a finite rank projection. Suppose z0 is a skew symmetric element
in (1−p0)B(H)(1−p0) such that (1−p0)∆(p)(1−p0) = [z0, p], for every finite-rank
projection p ∈ (1 − p0)B(H)(1 − p0). Then
(1 − p0)∆((1 − p0)a(1 − p0))(1 − p0) = [z0, (1 − p0)a(1 − p0)],
for every a ∈ B(H). If in addition p0 = 0, then ∆ = [z0, .] is a linear derivation
on B(H).
Proof. Let D : B(H) → B(H) denote the ∗-derivation defined by D(a) = [z0, a]
(a ∈ B(H)). Lemma 2.2 (see also [7, Lemma 3.4 and Proposition 3.1]) assures
that ∆F (H) : F (H) → F (H) is a linear mapping. Since every element in F (H)
can be written as a finite linear combination of finite-rank projections in B(H),
it follows from our hypothesis that
(1 − p0)∆(1 − p0)(1−p0)F (H)(1−p0) = D(1−p0)F (H)(1−p0) = [z0, .](1−p0)F (H)(1−p0). (3)
Fix a in (1 − p0)B(H)(1 − p0) and a finite-rank projection p1 ≤ 1 − p0. Having
1 ap1 ∈ (1 − p0)F (H)(1 − p0), Lemma 3.2 in [7],
in mind that p1ap1 + p1ap⊥
1 + p⊥
8
J.C. CABELLO AND A.M. PERALTA
and (3), we conclude that
p1∆(a)p1 = p1∆(p1ap1+p1ap⊥
The net (pF ) F ∈F(H)
≤1−p0
deduce from (4) that
F
p
1 +p⊥
1 ap1)p1 = p1[z0, (p1ap1+p1ap⊥
1 ap1)]p1. (4)
converges to 1 − p0 in the strong∗ topology of B(H). We
1 +p⊥
pF ∆(a)pF = pF [z0, pF a + p⊥
F
apF ]pF ,
for every F ∈ F(H) with pF ≤ 1 − p0. Taking strong∗-limits in the above identity,
it follows from the joint strong∗-continuity of the product in B(H) that
(1 − p0)∆(a)(1 − p0) = [z0, a],
which finishes the proof.
(cid:3)
Our next result is a consequence of Proposition 2.8 and Theorem 2.9.
Corollary 2.10. Let ∆ : B(H) → B(H) be a weak-2-local derivation with ∆♯ =
∆. Suppose that one of the following statements holds:
(a) The set Diam(∆) = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H)} is bounded;
(b) The set(cid:8)(cid:13)(cid:13)pF ∆pF pF B(H)pF(cid:13)(cid:13) : F ∈ F(H)(cid:9) is bounded.
Then ∆ is a linear derivation.
Proof. If ∆ satisfies (a), the the conclusion follows straightforwardly from Propo-
sition 2.8 and Theorem 2.9. If we assume (b) we simply observe that for each
F ∈ F(H) we have
kezF k ≤ diam(σ(ezF )) =(cid:13)(cid:13)[ezF , .]pF B(H)pF(cid:13)(cid:13) =(cid:13)(cid:13)pF ∆pF pF B(H)pF(cid:13)(cid:13) ≤ 2kezF k
(compare Remarks 2.4 and 2.5).
The following lemma states a simple property of derivations on Mn. The proof
(cid:3)
is left to the reader.
Lemma 2.11. Let D : Mn → Mn be a ∗-derivation. Suppose p1 is a rank one
projections in Mn. If D(a) = 0 for every a = p⊥
1 in Mn, then there exists
α ∈ iR such that D(x) = [αp1, x] for all x ∈ Mn.
(cid:3)
1 ap⊥
We state now an infinite dimensional analog of the previous lemma.
Proposition 2.12. Let ∆ : B(H) → B(H) be a weak-2-local derivation with
∆♯ = ∆. Suppose p0 is rank one projection in Mn such that ∆(a) = 0 for every
a = p⊥
0 ap⊥
0 in B(H), then there exists α ∈ iR such that ∆(x) = [αp0, x] for all
x ∈ B(H).
Proof. Take a finite rank projection p ≤ p⊥
0 . Since
∆p = (p+p0)∆(p0+p)(p0+p)B(H)(p0+p) : (p0+p)B(H)(p0+p) → (p0+p)B(H)(p0+p)
is a weak-2-local derivation with ∆♯
p = ∆p (compare [6, Proposition 2.7]) and
(p0 + p)B(H)(p0 + p) ∼= Mm for a suitable m, we deduce from [7, Theorem
2.12] that ∆p is a ∗-derivation. We also know that ∆p(a) = 0 for every a ∈
(p0+p)B(H)(p0+p) with a = pap. Lemma 2.11 implies the existence of α(p) ∈ iR,
depending on p, such that ∆p(x) = [α(p)p0, x] for all x ∈ (p0 + p)B(H)(p0 + p).
A GENERALIZED SEMRL'S THEOREM FOR WEAK-2-LOCAL DERIVATIONS
9
We claim that α(p) doesn't depend on p.
Indeed, let p1, p2 be finite rank
projections with pj ≤ p⊥
0 . We can find a third finite rank projection p3 ≤ p⊥
0
such that p1, p2 ≤ p3. We know that ∆pj (x) = [α(pj)p0, x] for all x ∈ (p0 +
pj)B(H)(p0 + pj) for all j = 1, 2, 3. Since for each j = 1, 2,
(p0 + pj)∆p3(p0 + pj)(p0+pj)B(H)(p0+pj) = ∆pj ,
we can easily see that α(pj) = α(p3) for every j = 1, 2, which proves the claim.
Therefore, there exists α ∈ iR such that
for all x ∈ (p0 + p)B(H)(p0 + p) and every finite rank projection p ≤ p⊥
0 .
(p + p0)∆(x)(p0 + p) = [αp0, x]
(5)
Let us fix F ∈ F(H). We can find another finite rank projection p1 ≤
p⊥
0 such that pF ≤ p0 + p1. We have shown that ∆p1 = (p0 + p1)∆(p0 +
p1)(p0+p1)B(H)(p0+p1) = [αp0, .](p0+p1)B(H)(p0+p1), and hence k∆p1k ≤ 2α. Since
pF ∆p1pF pF B(H)pF
, we can also conclude that
= pF ∆pF pF B(H)pF
for every F ∈ F(H). Corollary 2.10 implies that ∆ is a linear ∗-derivation. The
continuity and linearity of ∆ combined with (5) give the desired statement. (cid:3)
(cid:13)(cid:13)pF ∆pF pF B(H)pF(cid:13)(cid:13) ≤ 2α,
Theorem 2.13. Let ∆ : B(H) → B(H) be a weak-2-local derivation with ∆♯ =
∆. Suppose there exists G ∈ F(H) such that the set
is bounded. Then ∆ is a linear ∗-derivation.
G⊥ =(cid:8)diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H), pF ≤ p⊥
G(cid:9)
Diam−
Proof. Combining Proposition 2.8 and Theorem 2.9 we deduce the existence of
z0 = −z∗
0 in (1 − pG)B(H)(1 − pG) such that
(1 − pG)∆(a)(1 − pG) = [z0, (1 − pG)a(1 − pG)],
for every a ∈ B(H). The mapping ∆1 = ∆ − [z0, .] is a weak-2-local derivation
on B(H) with ∆1 = ∆♯
1, and satisfies
(1 − pG)∆1(a)(1 − pG) = 0,
(6)
for all a ∈ (1 − pG)B(H)(1 − pG).
Let q1, . . . , qm be mutually orthogonal rank one projections such that pG =
q1 + . . . + qm.
Let {ξj : j ∈ J} be an orthonormal basis of p⊥
G (H). For each j ∈ J we denote
by pj the rank-projection corresponding to the orthogonal projection of H onto
Cξj. By Proposition 2.7 in [6] the mapping (qm + pj)∆1(qm + pj)(qm+pj )B(H)(qm+pj )
is a linear ∗-derivation on (qm + pj)B(H)(qm + pj). Therefore, there exists zj =
αj
00
−αj
αj
0j
0j αj
jj ! = −z∗
j ∈ M2(C) such that (qm + pj)∆1(qm + pj)(a) = [zj, a], for
every a ∈ (qm + pj)B(H)(qm + pj). We deduce from (6) that αj
j). We have thus defined a family (αj
0j) ⊂ C.
jj = 0 (for every
10
J.C. CABELLO AND A.M. PERALTA
The same arguments give above show, via [6, Proposition 2.7] and (6), that for
each finite subset J0 ⊂ J, with k0 = ♯J0, and pJ0
)∆1(a)(qm + pJ0
(qm + pJ0
)B(H)(qm+pJ0
for all a ∈ (qm+pJ0
skew symmetric matrix given by zJ0
the unique minimal partial isometry satisfying e0je∗
α00 is a suitable complex number.
= α00qk0 +Pj∈J0
), where zJ0
=Pj∈J0
pj that
(7)
, a],
) = [zJ0
identifies with the (k0+1)×(k0+1)
0j, where e0j is
0je0j = pj, and
0j e∗
0j = qm and e∗
αj
0je0j − αj
0j2 is summable.
Indeed, for each finite
subset J0 ⊂ J, we can show from (7) and [7, Lemma 3.2] that
We claim that the family Pj∈J αj
Xj∈J0
0j = (qm + pJ0
αj
0je0j + αj
)∆1(pJ0
0je∗
and hence
)(qm + pJ0
) = (qm + pJ0
)∆1(p⊥
G )(qm + pJ0
),
αj
0j2 = k(qm + pJ0
)∆1(pJ0
)(qm + pJ0
)k2 ≤ k∆1(p⊥
G )k2,
the claim.
which assures the boundedness of the set {Pj∈J αj
Thanks to the claim, the element z1 =Pj∈J αj
skew symmetric element in B(H). We further know, from (7), that
0j2 : J0 ⊂ J finite } and proves
0je0j − αj
0je∗
0j is a well-defined
(qm + pJ0
)∆1(a)(qm + pJ0
) = (qm + pJ0
)[z1, a](qm + pJ0
),
(8)
pj, and every element a in pJ0
B(H)pJ0
.
Xj∈J0
=Pj∈J0
for every finite subset J0 ⊂ J, pJ0
In the case a = pJ0
(qm + pJ0
)∆1(pJ0
) we get
Lemma 3.2 in [7] implies that
(qm + pJ0
)∆1(p⊥
G )(qm + pJ0
)(qm + pJ0
) = (qm + pJ0
)[z1, pJ0
](qm + pJ0
).
= (qm + pJ0
)∆1(pJ0
)(qm + pJ0
](qm + pJ0
)
) = (qm + pJ0
)∆1(pJ0
) = (qm + pJ0
G ](qm + pJ0
)[z1, p⊥
)[z1, pJ0
).
= (qm + pJ0
+ (p⊥
G − pJ0
))(qm + pJ0
)
Letting pJ0
ր p⊥
G in the strong∗-topology, we get
(qm + p⊥
G )∆1(p⊥
G )(qm + p⊥
G ) = (qm + p⊥
G )[z1, p⊥
G ](qm + p⊥
0je0j + αj
αj
0je∗
0j.
G ) =bz1 =Xj∈J
G be a finite rank projection. We
where the last equality follows from [7, Lemma 3.2]. We similarly prove p[z1, p]qm =
G )p = qm∆1(p + (p⊥
G − p))p = qm∆1(p)p,
deduce from the last identity that
Clearly, z1 = qmbz1p⊥
G − p⊥
Gbz1qm. Let p ≤ p⊥
qm[z1, p]p = qmbz1p = qm∆1(p⊥
−pbz1qm = p∆1(p)qm, and hence, by (6),
G )∆1(p)(qm + p⊥
(qm + p⊥
G ) = (qm + p⊥
G )[z1, p](qm + p⊥
G ).
A GENERALIZED SEMRL'S THEOREM FOR WEAK-2-LOCAL DERIVATIONS
11
Now, Proposition 3.1 in [7] shows that ∆1 is linear on F (H). We thus deduce
from the above that
(qm + p⊥
G )∆1(a)(qm + p⊥
G ) = (qm + p⊥
G )[z1, a](qm + p⊥
G ),
(9)
for every a ∈ p⊥
G F (H)p⊥
G
.
We claim now that
(qm + p⊥
G
for every element a in p⊥
a projection pJ0
(qm + pJ0
that
(qm+p⊥
G )[z1, a](qm+p⊥
)⊥a(qm + pJ0
G )∆1(a)(qm + p⊥
G ) = (qm + p⊥
G )[z1, a](qm + p⊥
G ),
. For this purpose, let us fix a ∈ p⊥
B(H)p⊥
B(H)p⊥
, with J0 a finite subset of J. Having in mind that (qm + pJ0
, and
)a +
G , a new application of [7, Lemma 3.2] proves
G F (H)p⊥
) ∈ p⊥
G
G
G
G ) = (qm+p⊥
G )[z1, (qm+pJ0
)⊥a(qm+pJ0
)](qm+p⊥
G )
)a+(qm+pJ0
)⊥a(qm + pJ0
= (qm + pJ0
)∆1((qm + pJ0
)a + (qm + pJ0
))(qm + pJ0
)
= (qm + pJ0
If in the previous identity we let pJ0
equality stated in the claim.
The mapping (qm + p⊥
G )∆1(qm + p⊥
G )B(H)(qm + p⊥
tion on (qm + p⊥
that (qm + p⊥
p⊥
B(H)p⊥
G )∆1(a)(qm + p⊥
. We set
G
G
)∆1(a)(qm + pJ0
ր p⊥
).
G in the strong∗-topology we obtain the
)B(H)(qm+p⊥
G
G )(qm+p⊥
) is a weak-2-local deriva-
G ) (see [6, Proposition 2.7]). We know from (9)
G ), for every element a in
G )[z1, a](qm + p⊥
G ) = (qm + p⊥
G
∆2 = (qm + p⊥
G )∆1(qm + p⊥
G )(qm+p⊥
G
)B(H)(qm+p⊥
G
) − (qm + p⊥
G )[z1, ](qm + p⊥
G ).
Then ∆2 is a weak-2-local derivation on (qm + p⊥
for every a ∈ p⊥
on (qm + p⊥
(qm + p⊥
G ) and ∆2(a) = 0
. Proposition 2.12 proves that ∆2 is a linear ∗-derivation
G ), which implies the same conclusion for the mapping
G )B(H)(qm + p⊥
G )(qm+p⊥
G )B(H)(qm + p⊥
G )∆(qm + p⊥
B(H)p⊥
).
G
G
)B(H)(qm+p⊥
G
G
If we set G1 = m−1Xj=1
qj! (H) ( G, we conclude that the set
Diam−
G⊥
1
=ndiam(σ(wF )) : wF ∈ [zF ], F ∈ F(H), pF ≤ p⊥
G1o
is bounded (just apply that p⊥
G1
If we apply the above reasoning to G1, pm−1, and ∆, we deduce that
B(H)p⊥
G1
∆p⊥
G1
is a bounded linear ∗-derivation).
p⊥
G1
(qm−1 + p⊥
G1
)∆(qm−1 + p⊥
G1
)(qm−1+p⊥
G1
)B(H)(qm−1+p⊥
G1
)
is a bounded linear ∗-derivation. Repeating these arguments a finite number of
steps we prove that ∆ is a bounded linear ∗-derivation.
(cid:3)
The key technical result needed in our arguments follows now as a direct con-
sequence of the preceding proposition.
12
J.C. CABELLO AND A.M. PERALTA
Corollary 2.14. Let ∆ : B(H) → B(H) be a weak-2-local derivation with ∆♯ =
∆. Suppose that the set Diam(∆) = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H)} is
unbounded. Then there exists a sequence (Fn) ⊂ F(H) such that pFn ⊥ pFm for
every n 6= m, and diam(σ(ezFn )) ≥ 4n for every natural n.
Proof. If there exists G ∈ F(H) such that
Diam−
G⊥ =(cid:8)diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H), pF ≤ p⊥
G(cid:9)
is bounded, then Theorem 2.13 implies that ∆ is a linear ∗-derivation, which
contradicts the unboundedness of the set
We can therefore assume that Diam−
Diam(∆) =(cid:8)diam(σ(wF )) =(cid:13)(cid:13)pF ∆pF pF B(H)pF(cid:13)(cid:13) : wF ∈ [zF ], F ∈ F(H)(cid:9) .
We shall argue by induction. Let us fix F1 ∈ F(H) with diam(σ(ezF1
In the notation employed before, the set Diam−
F1
ping p⊥
F1
tion and a symmetric mapping (compare [6, Proposition 2.7]). Therefore the
set Diam(p⊥
) must be unbounded. We can find F2 ∈ F(H) with
F1
)) ≥ 4.
⊥ is unbounded. The map-
is a weak-2-local deriva-
G⊥ is unbounded for every G ∈ F(H).
B(H)p⊥
F1
B(H)p⊥
F1
→ p⊥
F1
B(H)p⊥
F1
∆p⊥
F1
∆p⊥
F1
: p⊥
F1
p⊥
F1
pF2
⊥ pF1
Suppose we have defined F1, . . . , Fn satisfying the desired conditions. Set Kn :=
F1 ⊕ℓ2 . . . ⊕ℓ2 Fn ∈ F(H). According to the arguments at the beginning of the
proof, Diam−
⊥ is unbounded. Therefore, we can find Fn+1 ∈ F(H) such that
Kn
pFn+1
(cid:3)
and diam(σ(ezF2
for every j = 1, . . . , n and diam(σ(ezFn+1
We shall show next that every weak-2-local derivation on B(H) is bounded on
)) ≥ 4n+1.
⊥ pFj
the lattice of projections of B(H).
p⊥
F1
B(H)p⊥
F1
)) ≥ 42.
Theorem 2.15. Let ∆ : B(H) → B(H) be a weak-2-local derivation with ∆♯ =
∆. Then the following statements hold:
(a) The set Diam(∆) = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H)} is bounded;
0 and a
of mutually orthogonal rank one projections (pk) ⊆ B(H) satisfying p2n−1, p2n ≤
Consequently, by Alaoglu's theorem, we can find z0 ∈ B(H) with z0 = −z∗
Proof. (a) Arguing by contradiction, we suppose that Diam(∆) is unbounded.
By Corollary 2.14, there exists a sequence (Fn) ⊂ F(H) such that pFn ⊥ pFm for
(b) The set {kezF k : F ∈ F(H)} is bounded;
subnet (ezF )F ∈Λ of (ezF )F ∈F(H) converging to z0 in the weak∗-topology of B(H).
every n 6= m, and diam(σ(ezFn )) ≥ 4n for every natural n. We can pick a sequence
pFn , ezFn = iλ2n−1p2n−1 + iλ2np2n + (pFn − p2n−1 − p2n)ezFn (pFn − p2n−1 − p2n)
(λ2n−1, λ2n ∈ R), and λ2n−1 − λ2n = λ2n−1 − λ2n = diam(σ(ezFn )) ≥ 4n.
∞Xn=1
Let en be the unique rank-2 partial isometry in B(H) defined by en = ξ2n ⊗
ξ2n−1 + ξ2n−1 ⊗ ξ2n, where ξ2n and ξ2n−1 are norm one vectors in p2n(H) and
p2n−1(H), respectively. Since en ⊥ em, for every n 6= m, the series
en converges
A GENERALIZED SEMRL'S THEOREM FOR WEAK-2-LOCAL DERIVATIONS
13
2nXk=1
nMk=1
to an element a0 ∈ B(H). Set s2n :=
pk ≤ pKn , where Kn =
Fk. Clearly,
2na0s⊥
a0 = s2na0s2n + s⊥
Remark 2.4 and Lemma 2.6) and [7, Lemma 3.2] we have
s2n∆(a0)s2n = s2n∆(s2na0s2n)s2n = s2n[ezKn
2n. Applying the properties of ezFn (compare Lemma 2.3,
="i
nXk=1
λ2k−1p2k−1 + λ2kp2k, s2na0s2n# .
, s2na0s2n]s2n
1
2k ωξ2k−1,ξ2k, where, following
the standard notation, ωξ2k−1,ξ2k(a) = hξ2k−1, a(ξ2k)i (a ∈ B(H)). We deduce from
the above that kφ0k ≤ 1 and
k=1
Let us consider consider the functional φ0 =Pn
nXk=1
nXk=1
1
2k λ2k−1 − λ2k >
k∆(a0)k ≥ φ0(s2n∆(a0)s2n) =
nXk=1
=
which is impossible.
1
2k (λ2k−1 − λ2k)
1
2k 4k =
2k,
nXk=1
(b) Take F ∈ F(H) and any z ∈ [zF ]. If we choose iλ ∈ σ(zF ), the inequalities
kezF k ≤ kz − iλpF k ≤ diam(σ(z − iλpF )) = diam(σ(z)) = diam(σ(ezF )),
hold because 0 ∈ σ(z − iλpF ) and (z − iλpF )∗ = −(z − iλpF ). Finally, the desired
conclusion follows from statement (a).
(cid:3)
We can provide now a positive answer to Problem 1.4 in the case A = B(H).
Theorem 2.16. Let ∆ : B(H) → B(H) be a weak-2-local derivation with ∆♯ =
∆. Then ∆ is a linear ∗-derivation.
Proof. By Theorem 2.15, the set
Diam(∆) = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H)}
is bounded. The desired conclusion follows from Corollary 2.10.
(cid:3)
All the results from Lemma 2.3 to Proposition 2.14 remain valid when ∆ :
K(H) → K(H) is a weak-2-local derivation with ∆♯ = ∆. Actually, the conclu-
sion of Theorem 2.15 also holds for every such a mapping ∆ with practically the
same proof, but replacing a0 =
en ∈ B(H) with a0 =
en ∈ K(H),
because in that case we would have
k∆(a0)k ≥ φ0(s2n∆(a0)s2n) =
λ2k−1 − λ2k >
∞Xn=1
1
3(cid:19)k
2k(cid:18) 2
nXk=1
3(cid:19)n
∞Xn=1(cid:18)2
nXk=1(cid:18)4
3(cid:19)k
,
obtaining the desired contradiction. We have thus obtained an appropriate ver-
sion of Theorem 2.15 for weak-2-local derivations on K(H).
14
J.C. CABELLO AND A.M. PERALTA
Theorem 2.17. Let ∆ : K(H) → K(H) be a weak-2-local derivation with ∆♯ =
∆. Then the following statements hold:
(a) The set Diam(∆) = {diam(σ(wF )) : wF ∈ [zF ], F ∈ F(H)} is bounded;
Consequently, by Alaoglu's theorem, we can find z0 ∈ B(H) with z0 = −z∗
(b) The set {kezF k : F ∈ F(H)} is bounded;
subnet (ezF )F ∈Λ of (ezF )F ∈F(H) converging to z0 in the weak∗-topology of B(H). (cid:3)
Applying a subtle adaptation of the previous arguments we get the following.
Theorem 2.18. Let ∆ : K(H) → K(H) be a weak-2-local derivation with ∆♯ =
∆. Then ∆ is a linear ∗-derivation.
0 and a
3. weak-2-local derivations on B(H)
We can culminate now the study of weak-2-local derivations on B(H) with the
promised solution to Problem 1.3 in the case A = B(H).
Theorem 3.1. Let H be an arbitrary complex Hilbert space, and let ∆ be a
weak-2-local derivation on B(H). Then ∆ is a linear derivation.
Proof. We have already commented that H can be assumed to infinite dimen-
sional. Suppose ∆ : B(H) → B(H) is a weak-2-local derivation. Since the set
S = Der(A), of all derivations on B(H), is a linear subspace of B(B(H)), we
deduce from Lemma 2.1(c) and (e) that ∆1 = ∆+∆♯
are weak-
2-local derivations on B(H). Since ∆1 = ∆♯
2, Theorem 2.9 proves
that ∆1 and ∆2 are linear ∗-derivations on B(H), and thus, ∆ = ∆1 + i∆2 is a
linear derivation on B(H).
(cid:3)
and ∆2 = ∆−∆♯
1 and ∆2 = ∆♯
2
2i
According to Theorem 2.18, the arguments developed to prove Theorem 3.1
are also valid to obtain the following:
Theorem 3.2. Let H be an arbitrary complex Hilbert space, and let ∆ be a
weak-2-local derivation on K(H). Then ∆ is a linear derivation.
(cid:3)
We begin with a suitable generalization of [7, Lemma 3.2].
Lemma 3.3. Let A1 and A2 be C∗-algebras, and let ∆ : A1 ⊕∞ A2 → A1 ⊕∞ A2
be a weak-2-local derivation. Then ∆(Aj) ⊆ Aj for every j = 1, 2. Moreover, if
πj denotes the projection of A1 ⊕∞ A2 onto Aj, we have πj∆(a1 + a2) = πj∆(aj),
for every a1 ∈ A1, a2 ∈ A2 and j = 1, 2.
Proof. Let us fix a1 ∈ A1. Every C∗-algebra admits a bounded approximate unit
(cf. [8, Theorem 1.4.2]), thus, by Cohen's factorisation theorem (cf. [5, Theorem
VIII.32.22 and Corollary VIII.32.26]), there exist b1, c1 ∈ A1 satisfying a1 = b1c1.
We recall that A∗ = A∗
2, there exists a
derivation Da1,φ : A1 ⊕∞ A2 → A1 ⊕∞ A2 satisfying
φ∆a1,φ(a1) = φDa1,φ(a1) = φDa1,φ(b1c1) = φ(Da1,φ(b1)c1) + φ(b1Da1,φ(c1)) = 0,
2. By hypothesis, for each φ ∈ A∗
1 ⊕ℓ1 A∗
where in the last equalities we applied that Da1,φ(b1)c1 and b1Da1,φ(c1) both lie
in A1 and φ ∈ A∗
2. We deduce, via Hahn-Banach theorem, that ∆(a1) ∈ A1.
A GENERALIZED SEMRL'S THEOREM FOR WEAK-2-LOCAL DERIVATIONS
15
The above arguments also show that, for each derivation D : A1 ⊕∞ A2 →
A1 ⊕∞ A2 we have D(Aj) ⊆ Aj for every j = 1, 2. It follows from the hypothesis
that, for each φ ∈ A∗
1, a1 ∈ A1 and a2 ∈ A2, there exists a derivation Dφ,a1+a2,a1 :
A1 ⊕∞ A2 → A1 ⊕∞ A2 satisfying
φ∆(a1) = φDφ,a1+a2,a1(a1), and φ∆(a1 + a2) = φDφ,a1+a2,a1(a1 + a2).
In particular, φ∆(a1) = φ∆(a1 + a2), for every φ ∈ A∗
π1∆(a1 + a2).
1. It follows that π1∆(a1) =
(cid:3)
For further purposes, we shall also explore the stability of the above results
under ℓ∞- and c0-sums.
Proposition 3.4. Let (Aj) be an arbitrary family of C∗-algebras. Suppose that
for each j, every weak-2-local derivation on Aj is a linear derivation. Then the
following statements hold:
j∈J Aj be a weak-2-local derivation. Let πj
denote the natural projection of A onto Aj. If we fix an index j0 ∈ J, it follows
(a) Every weak-2-local derivation on A =Lℓ∞ Aj is a linear derivation;
(b) Every weak-2-local derivation on A =Lc0 Aj is a linear derivation.
Proof. (a) Let ∆ : Lℓ∞
from Lemma 3.3 that ∆(Aj0) ⊆ Aj0 and ∆ ℓ∞Mj06=j∈J
j∈J Aj → Lℓ∞
from the assumptions that ∆Aj : Aj → Aj is a linear derivation for every j.
Aj! ⊆
ℓ∞Mj06=j∈J
Aj. We deduce
We shall finish the proof by showing that {k∆Aj k : j ∈ J} is a bounded set.
Otherwise, there exist infinite sequences (jn) ⊆ J, (ajn) ⊂ A, with ajn ∈ Ajn,
kajnk ≤ 1, and k∆(ajn)k > 4n, for every natural n. Let a0 =
ajn ∈ A. For
∞Xn=1
each natural n, a0 = ajn + (a0 − ajn) with ajn ⊥ (a0 − ajn) in A. It follows from
the above properties and the second statement in Lemma 3.3 that
k∆(a0)k ≥ kπjn∆(a0)k = k∆(ajn)k > 4n,
for every n ∈ N, which is impossible.
(b) The proof of (a) but replacing a0 =
valid in this case.
ajn with a0 =
∞Xn=1
1
2n
∞Xn=1
ajn ∈ A remains
(cid:3)
Following standard notation, we shall say that a von Neumann algebra M is
that a Banach algebra is called dual or compact if, for every a ∈ A, the operator
A → A, b 7→ aba is compact. By [1], compact C∗-algebras are precisely the
atomic if M =Lℓ∞ B(Hα), where each Hα is a complex Hilbert space. We recall
algebras of the form (Li∈I K(Hi))c0, where each Hi is a complex Hilbert space.
We finish this note with a couple of corollaries which follow straightforwardly
from Theorems 3.1, 3.2 and Proposition 3.4.
Corollary 3.5. Every weak-2-local derivation on an atomic von Neumann algebra
is a linear derivation.
(cid:3)
16
J.C. CABELLO AND A.M. PERALTA
Corollary 3.6. Every weak-2-local derivation on a compact C∗-algebra is a linear
derivation.
(cid:3)
References
[1] J. Alexander, Compact Banach algebras, Proc. London Math. Soc. 18, 1-18, (1968).
[2] Sh. Ayupov, K. Kudaybergenov, 2-local derivations and automorphisms on B(H), J. Math.
Anal. Appl. 395, no. 1, 15-18 (2012).
[3] Sh. Ayupov, K.K. Kudaybergenov, 2-local derivations on von Neumann algebras, Positivity
19, No. 3, 445-455 (2015). DOI 10.1007/s11117-014-0307-3
[4] J.C. Cabello, A.M. Peralta, Weak-2-local symmetric maps on C∗-algebras, Linear Algebra
Appl. 494, 32-43 (2016).
[5] E. Hewitt, K.A. Ross, Abstract harmonic analysis. Vol. II: Structure and analysis for
compact groups. Analysis on locally compact Abelian groups, Springer-Verlag, New York-
Berlin, 1970.
[6] M. Niazi and A.M. Peralta, Weak-2-local derivations on Mn, to appear in FILOMAT.
[7] M. Niazi and A.M. Peralta, Weak-2-local ∗-derivations on B(H) are linear ∗-derivations,
Linear Algebra Appl. 487, 276-300 (2015).
[8] G.K. Pedersen, C∗-algebras and their automorphism groups, Academic Press, London 1979.
[9] S. Sakai: C∗-algebras and W ∗-algebras. Springer Verlag. Berlin (1971).
[10] P. Semrl, Local automorphisms and derivations on B(H), Proc. Amer. Math. Soc. 125,
2677-2680 (1997).
[11] J.G. Stampfli, The norm of a derivation, Pacific J. Math. 33, No. 3, 737-747 (1970).
Departamento de An´alisis Matem´atico, Universidad de Granada, Facultad de
Ciencias 18071, Granada, Spain
E-mail address: [email protected], [email protected]
|
1606.03728 | 2 | 1606 | 2017-11-08T18:16:55 | Landstad-Vaes theory for locally compact quantum groups | [
"math.OA"
] | Landstad-Vaes theory deals with the structure of the crossed product of a C$^*$-algebra by an action of locally compact (quantum) group. In particular it describes the position of original algebra inside crossed product. The problem was solved in 1979 by Landstad for locally compact groups and in 2005 by Vaes for regular locally compact quantum groups. To extend the result to non-regular groups we modify the notion of $G$-dynamical system introducing the concept of weak action of quantum groups on C$^*$-algebras. It is still possible to define crossed product (by weak action) and characterise the position of original algebra inside the crossed product. The crossed product is unique up to an isomorphism. At the end we discuss a few applications. | math.OA | math |
LANDSTAD-VAES THEORY FOR
LOCALLY COMPACT QUANTUM GROUPS
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
Abstract. Landstad-Vaes theory deals with the structure of the crossed product of a C∗-algebra
by an action of locally compact (quantum) group. In particular it describes the position of original
algebra inside crossed product. The problem was solved in 1979 by Landstad for locally compact
groups and in 2005 by Vaes for regular locally compact quantum groups. To extend the result to non-
regular groups we modify the notion of G-dynamical system introducing the concept of weak action
of quantum groups on C∗-algebras.
It is still possible to define crossed product (by weak action)
and characterise the position of original algebra inside the crossed product. The crossed product is
unique up to an isomorphism. At the end we discuss a few applications.
0. Introduction
The concept of crossed product of a C∗-algebra by an action of a locally compact (quantum) group
comes from the desire to unite in a single object the C∗-algebra and the unitaries implementing the
action. For the theory of crossed product of C∗-algebras by actions of locally compact groups see [13].
One of the most used formula in the operator algebra theory is the implementing of an automorphism
by a unitary operator:
(0.1)
In this formula U is a unitary operator acting on a Hilbert space K and d runs over a C∗-algebra D
of operators acting on K. It is assumed that α(d) ∈ D for any d ∈ D and that α(D) = D. Then α is
an automorphism of C∗-algebra D. We say that automorphism α is implemented by U .
α(d) = U dU ∗.
Any automorphism of a C∗-algebra D can be implemented. For any α there exist a Hilbert space
K and a pair (j, U ), where j is an embedding of D in B(K) and U is a unitary operator acting on K
such that (identifying d with j(d)) we have (0.1). We say that (j, U ) is a covariant representation of
D.
It is interesting to extend D by including U . Let
(0.2)
B =nU nd : n ∈ Z, d ∈ DoCLS
,
where CLS stays for norm closed linear span. Then B is a C∗-algebra, U ∈ M(B), D ⊂ B and DB = B.
The latter means that the embedding D ⊂ B is a morphism from D into B. In general, for given D
and α, the algebra B may depend on the used covariant representation. For instance, if α is inner then
we may take U ∈ M(D) and then B = D.
To obtain a more interesting algebra B we have to assume that U is in a sense independent of
elements of D. One of the symptom of this independence is the existence of dual action. We say that
B admits a dual action if for any z ∈ S1 there exists an automorphism βz of B such that βz(U ) = zU
and βz(d) = d for any d ∈ D. It turns out that in any case one can find a covariant representation such
that the algebra (0.2) admits a dual action. Moreover the algebra B with the dual action is unique (up
to isomorphism): It does not depend on the choice of covariant representation. This unique C∗-algebra
is denoted by D ⋉α Z and called the crossed product of D by the automorphism α. It is equipped with
the dual action β of S1 and distinguished element U ∈ M(D ⋉α Z) such that
(0.3)
for any z ∈ S1.
βz(U ) = zU
2000 Mathematics Subject Classification. 46L55 (46L08 81R50).
Key words and phrases. Dynamical system, Crossed product, Landstad conditions, Landstad algebra, Non-regular
quantum groups.
Supported by the Alexander von Humboldt-Stiftung. Sutanu Roy was partially supported by Inspire faculty award
given by D.S.T., Government of India. S.L. Woronowicz was partially supported by the National Science Center (NCN)
grant no. 2015/17/B/ST1/00085.
1
2
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
Conversely assume that B is a C∗-algebra equipped with an action β of S1 and a distinguished
element U ∈ M(B) such that formula (0.3) holds. Then
D =(cid:8)d ∈ B : βz(d) = d for all z ∈ S1(cid:9)
is a C∗-subalgebra of B, DB = B, formula (0.1) defines an automorphism α of D and relation (0.2)
holds.
In the above the single automorphism may be replaced by a locally compact quantum group of
In this case the
automorphisms. We shall consider locally compact quantum group G = (A, ∆).
formula (0.1) takes the form
(0.4)
α(d) = U (d ⊗ IA)U ∗,
where U ∈ M(B0(K) ⊗ A) is a unitary representation of G acting on K. It is assumed that α(d) ∈
M(D ⊗ A) for any d ∈ D and that (using the notation (1.2))
(0.5)
α(D)(ID ⊗A) = D ⊗ A.
Then α ∈ Mor(D, D ⊗ A). We say that α is an action of G on D implemented by a representation U .
In general, a right action of G on a C∗-algebra D is an injective morphism α ∈ Mor(D, D ⊗ A) such
that (α ⊗ idA) ◦ α = (idD ⊗ ∆) ◦ α. The action is said to be continous if the Podleś condition (0.5)
holds. In the similar way one defines left actions.
Any continuous action α of a locally compact group G on a C∗-algebra D can be implemented. For
any α there exist a Hilbert space K and a pair (j, U ), where j is an embedding of D in B(K) and U
is a unitary representation of G acting on K such that (identifying d with j(d)) we have (0.4). We
say that (j, U ) is a covariant representation of (D, α). Moreover one may assume that U is weakly
contained in the regular representation. It means that U is of the form
U = (ψ ⊗ idA)V,
where ψ ∈ Rep(bA, K) and V ∈ M(bA ⊗ A) is the canonical bicharacter establishing duality between
G = (A, ∆) and bG = (bA,b∆). In what follows we shall use shorthand1 notation Vψ2 = (ψ ⊗ idA)V .
Again it is interesting to extend D by including ψ-copy of bA. We shall use notation (1.2) (see next
section). Let
independent of elements of D. One of the symptom of this independence is the existence of dual
the next section for the concept of morphism in the category of C∗-algebras. In general, for given D
and α, the algebra B may depend on the used covariant representation.
Then B is a C∗-algebra, D, ψ(bA) ⊂ M(B) and DB = ψ(bA)B = B. The latter means that the
embedding D ⊂ M(B) is a morphism from D into B and ψ is a morphism from bA into B. We refer to
To obtain a more interesting algebra B we have to assume that elements of bA are in a sense
action. We say that B admits a dual action if there exists an injective morphism β ∈ Mor(B, bA ⊗ B)
such that β(ψ(ba)) = (id bA ⊗ ψ)b∆(ba) and β(d) = I bA ⊗ d for any d ∈ D and ba ∈ bA. It turns out that
crossed product of D by the action α. It is equipped with the dual action β of bG and the morphism
ψ ∈ Mor(bA, B) such that
for any ba ∈ bA. In general, a triple (B, β, ψ), where B is a C∗-algebra, β ∈ Mor(B, bA ⊗ B) is a left
action of bG on B and ψ is an injective morphism from bA into B is called G-product if formula (0.7)
in any case one can find a covariant representation such that the algebra (0.6) admits a dual action.
Moreover the algebra B with the dual action is unique (up to isomorphism): It does not depend on
the choice of covariant representation. This unique C∗-algebra is denoted by D ⋉α G and called the
β(ψ(ba)) = (id bA ⊗ ψ)b∆(ba)
One of the aims of the Landstad theory was to describe position of D within M(D ⋉α G). It could
(0.7)
holds.
be easily shown that elements d ∈ D satisfy the following three conditions:
1This is extended leg numbering notation
(0.6)
B =nψ(ba)d :ba ∈ bA, d ∈ DoCLS
= ψ(bA)D.
L1. β(d) = I bA ⊗ d,
LANDSTAD-VAES THEORY
3
L2. ψ(ba)d ∈ B for anyba ∈ bA,
L3. (cid:16)Vψ2(d ⊗ IA)V ∗
ψ2(cid:17) (IB ⊗a) ∈ M(B) ⊗ A for any a ∈ A.
The above conditions were (in a more classical language) formulated by Magnus Landstad2 (see also
section 7.8 of [13]). Assuming that G is a locally compact group Landstad was able to show that his
conditions completely characterise those elements of M(B) that belong to D. More than that Landstad
proved (for classical G) that any G-product (B, β, ψ) comes from crossed product construction:
where D is the C∗-algebra consisting of all elements of d ∈ M(B) satisfying Conditions L1, L2 and L3,
and
B = D ⋉α G,
(0.8)
for any d ∈ D.
α(d) = Vψ2(d ⊗ IA)V ∗
ψ2
Instead of looking for conditions characterising elements of D within M(D ⋉α G) one may formulate
properties concerning the C∗-algebra D itself. It could be easily shown that
V1. β(d) = I bA ⊗ d for any d ∈ D,
V2. B = ψ(bA)D,
V3. (cid:16)Vψ2(D ⊗ IA)V ∗
ψ2(cid:17) (IB ⊗A) = D ⊗ A.
These conditions were formulated by Stefaan Vaes in [18]. Assuming that G is a locally compact
regular quantum group Vaes was able to show that for any G-product (B, β, ψ) there exists unique
C∗-subalgebra D of M(B) satisfying Conditions V1, V2 and V3. This subalgebra is equipped with
the left action α ∈ Mor(D, A ⊗ D) of G introduced by (0.8) and (B, β, ψ) comes from crossed product
construction: B = D ⋉α G. This way Vaes extended Landstad theory to regular locally compact
quantum groups.
It is interesting to compare Landstad and Vaes conditions. Let (B, β, ψ) be a G-product, D ⊂ M(B)
be a C∗-subalgebra and d ∈ D. Then L1 coincides with V1, L2 follows from V2 and assuming V3 we
see that for any a ∈ A we have (cid:16)Vψ2(d ⊗ IA)V ∗
follows from V3.
ψ2(cid:17) (IB ⊗a) ∈ D ⊗ A ⊂ M(B) ⊗ A. It shows that L3
Let (B, β, ψ) be a G-product. In general (for non-regular G) a subalgebra D ⊂ M(B) satisfying Vaes
conditions may not exist. To regain the existence of D we have to replace V3 by a weaker Condition
C3. Roughly speaking, C3 means that the slices of Vψ2(D ⊗ IA)V ∗
ψ2 generate C∗-algebra D.
In what follows the subalgebra G ⊂ M(B) satisfying Conditions V1, V2 and C3 will be called
Landstad algebra of (B, β, ψ). We shall prove that any G-product admits unique Landstad algebra.
Unfortunately now (when Condition V3 is not satisfied) formula (0.8) does not define a action of G
on D. In general α(d) /∈ M(D ⊗ A). To deal with the problem we invent the notion of weak action
adapted to this situation. In brief instead of (0.5) we assume that slices of α(d) belongs to D and that
the set of all slices generate C∗-algebra D. In the following a pair (D, α), where D is a C∗-algebra
and α is a weak action of G on D, will be called a weak G-dynamical system. For regular groups the
concepts of weak and continuous actions coincides.
Working with weak actions we have to reconsider the concept of crossed product. Given a G-
dynamical system (D.α), we shall construct a G-product (B, β, ψ) such that D plays the role of
Landstad algebra of (B, β, ψ), α is implemented by Vψ2 and ψ ∈ Mor(bA, B) is the canonical embedding.
In other words B is in a sense crossed product of D by the action α and β is the dual action. We are
still able to show that (B, β, ψ) is unique. It means that the correspondence between G-products and
G-dynamical systems is one to one.
Let us shortly discuss the content of the paper. In section 1 we explain the notation used in the
paper. In particular we recall the category of C∗-algebras (concepts of morphisms and composition of
morphisms). We also collect all informations concerning quantum groups used in the paper. Section 2
2see formulae (3.6) - (3.8) in [10]. In fact in the second condition Landstad demanded additionally that dψ(ba) ∈ B.
However this requirement is redundant, it follows from other Landstad conditions (see last section)
4
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
contains main definitions and results. We introduce (recall) the concepts of G-dynamical system and
G-product and describe the duality between them. In Section 3 we analyse the concept of weak action.
The most important result is Proposition 3.2 that enables crossed product construction. At the end we
show that for regular groups any weak action is continuous (satisfies Podleś condition). The existence
and uniqueness of Landstad algebra for any G-product is discussed in section 4. At the end of the
section we show the uniqueness of G-product corresponding to any weak G-dynamical system. Section
5 is devoted to the crossed product construction. Next, in Section 6 we discuss in detail weak actions
implemented by a unitary representation of G. An application to the Kasprzak version of Rieffel
deformation is recalled in Section 7. Finally in Section 8 we show (for coameanable G) that one of the
original Landstad condition is a consequence of the others.
Throughout the paper we shall use the following notation: For any separable Hilbert space K we
set
1. Notation
B(K) = the von Neumann algebra of all bounded operators acting on K,
B0(K) = the C∗-algebra of all compact operators acting on K,
B(K)∗ = the set of all normal functionals on B(K)
= the set of all continuous functionals on B0(K),
C∗(K) =(cid:26)A ⊂ B(K) :
A is separable C∗-algebra
such that AK = K
(cid:27) .
Then B(K) and B0(K) are C∗-algebras, C∗(K) is a set of C∗-algebras, B0(K) ∈ C∗(K). In this paper
phrase 'C∗-algebra generated by a set' means 'the smallest C∗-algebra containing the set'.
We recall that B(K)∗ is a bimodule over B(K). For any µ ∈ B(K)∗ and a ∈ B(K), µa and aµ are
normal functionals on B(K) such that
(1.1)
for any m ∈ B(K).
(µa)(m) = µ(am),
(aµ)(m) = µ(ma)
Let X and Y be a norm closed subsets of a C∗-algebra. We set
(1.2)
where CLS stays for norm closed linear span.
XY =(cid:26)xy :
x ∈ X
y ∈ Y (cid:27)CLS
,
For any C∗-algebra A, M(A) will denote the multiplier algebra (cf [13]) of A. Then A is an essential
ideal in M(A). We shall use the category of C∗-algebras introduced in [20, 19]. It will be denoted by
C∗. Objects are C∗-algebras. For any C∗-algebras A and B, Mor(A, B) is the set of all ∗-algebra homo-
morphisms ϕ from A into M(B) such that ϕ(A)B = B. Any ϕ ∈ Mor(A, B) admits unique extension
are C∗-algebras) then the composition of morphisms ψ ◦ ϕ ∈ Mor(A, C) is defined as composition of
to a unital ∗-homomorphism eϕ : M(A) −→ M(B). If ϕ ∈ Mor(A, B) and ψ ∈ Mor(B, C) (A, B, C
∗-algebra homomorphisms: ψ ◦ ϕ = ψ ◦eϕ.
Let A be a C∗-algebra. Depending on the context we shall use two symbols: idA and IA to denote
the identity map acting on A.
idA will denote the identity morphism acting on A, whereas IA will
denote the unit element of the multiplier algebra M(A)3. To simplify notation we write IK and idK
instead of IB0(K) and idB0(K). If A ∈ C∗(K) then IA = IK. We shall often omit the index 'A' (or 'K')
when the algebra A (or the Hilbert space K) is obviously implied by the context.
By definition, representations acting on a Hilbert space K are morphisms into B0(K):
for any
C∗-algebra X, Rep(X, K) = Mor(X, B0(K)). We know that M(B0(K)) = B(K). Therefore represen-
tations are non-degenerate ∗-algebra homomorphisms into B(K). ∗-algebra homomorphism π : X −→
B(K) is non-degenerate if π(X) B0(K) = B0(K). Equivalently π is non-degenerate if 0 ∈ K is the only
vector killed by π(x) for all x ∈ X.
3Identifying elements of M(A) with left multipliers acting on A we have IA = idA.
LANDSTAD-VAES THEORY
5
The category C∗ is equipped with a monoidal structure. For any C∗-algebras X and Y , the tensor
product X ⊗ Y is a C∗-algebra. One can easily define tensor product of morphisms. Then ⊗ becomes
an associative functor from C∗ × C∗ to C∗. In this paper we use exclusively minimal (spacial) tensor
product of C∗-algebras.
Proposition 1.1. Let X, Y ∈ C∗(K), where K is a Hilbert space. Assume that
XY = Y X.
Then Z = XY ∈ C∗(K), X, Y ⊂ M(Z) and the embeddings are morphisms from X and Y into Z.
We shall refer to this situation by saying that Z is a crossed product of X and Y . The obvious
proof is left to the reader.
We shall use the following shorthand notation borrowed from [3]: If (Xω)ω∈Ω is a family of subsets
of a Banach space X, then the smallest Banach subspace of X containing all Xω (ω ∈ Ω) will be
denoted by
So we have:
Typically we shall deal with expressions of the form
hXω : ω ∈ Ωi .
hXω : ω ∈ Ωi = [ω∈Ω
Xω!CLS
h(ω ⊗ idK)Z : ω ∈ B(H)∗i ,
.
where Z is a linear subset of B(H ⊗ K) and H and K are Hilbert spaces. Let A ∈ C∗(H). By the
factorisation theorem [4] any ω ∈ B(H)∗ is of the form ω′a and of the form aω′ where ω′ ∈ B(H)∗ and
a ∈ A. Therefore
(1.3)
h(ω ⊗ idK)Z : ω ∈ B(H)∗i =h(ω ⊗ idK)(A ⊗ IK)Z : ω ∈ B(H)∗i
=h(ω ⊗ idK)Z(A ⊗ IK) : ω ∈ B(H)∗i .
We shall refer to these formulas by saying that ω emits A to the right (upper formula) or to the left
(lower formula). Reading from the right we say that ω absorbs A.
In the present paper we consider actions of a locally compact quantum group on C∗-algebras. The
group will be denoted by G. We shall not use the full power of the theory developed by Kustermans
and Vaes in [9] (see also [11]).
Instead we shall assume that G is constructed from a manageable
(modular) multiplicative unitary V in the way described in [24] (see also [15, 16]). In particular we do
not assume the existence of Haar measures.
The Haar measure plays an essential role in the Landstad considerations. To construct β invariant
elements in M(B) he applies βg (g ∈ G) to specially chosen elements of B and then integrates over
G using the Haar measure. On the other hand in Vaes approach [18] (at least in the part that
intersects with our interest) the Haar measure plays purely decorative role and can be removed from
considerations. Then the proofs become more transparent.
In this paper a locally compact quantum group G is a pair G = (A, ∆), where A is a C∗-algebra
and ∆ ∈ Mor(A, A ⊗ A) having a number of properties. One of the property is coassociativity of ∆:
another one is the cancelation property:
(∆ ⊗ idA) ◦ ∆ = (idA ⊗ ∆) ◦ ∆,
(A ⊗ IA)∆(A) = A ⊗ A = ∆(A)(IA ⊗A).
Instead of listing the properties we assume that (A, ∆) comes from a manageable multiplicative
unitary by the construction described in [24]. Elements of A should be considered as continuous
vanishing at infinity functions on G, whereas ∆ encodes the group multiplication on G. Locally compact
quantum groups appear in pairs. For any G = (A.∆) we have dual group bG = (bA,b∆). The duality
between G and bG is described by a bicharacter V ∈ M(bA ⊗ A). It satisfies bicharacter equations:
(id bA ⊗ ∆)V = V12V13,
(1.4)
(b∆ ⊗ idA)V = V23V13,
6
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
An important role in the theory of quantum groups is played by Heisenberg and anti-Heisenberg
The role of G and bG is symmetric. Replacing V by bV = flip(V ∗) we obtain bicharacter describing
duality between bG and G.
pairs. These are pairs of representations of A and bA acting on the same Hillbert space. Let H be a
Hilbert space, σ ∈ Rep(A, H) andbσ ∈ Rep(bA, H). We say that (σ,bσ) is a Heisenberg pair if
Similarly let H be a Hilbert space, ρ ∈ Rep(A, H) and bρ ∈ Rep(bA, H). We say that (ρ,bρ) is an
Vbσ3V1σ = V1σV13Vbσ3.
anti-Heisenberg pair if
(1.5)
V1ρVbρ3 = Vbρ3V13V1ρ.
(1.6)
It is known that representations appearing in Heisenberg and in anti-Heisenberg pairs are faithful.
In the above formulas extended leg numbering notation is used. Both sides of (1.5) belongs to
ρ(a) = σ(aR)⊤,
bR)⊤.
The existence of Heisenberg and anti-Heisenberg pairs is one of the basic features of the theory of
quantum groups. In [24] we start with a multiplicative unitary operator acting on H ⊗ H. Then A
conjugate to H. Then we have canonical anti-unitary mapping H ∋ x 7→ x ∈ H and transposition
map B(H) ∋ m 7→ m⊤ ∈ B(H) introduced by the formula m⊤x = m∗x for any x ∈ H. For any a ∈ A
M(bA ⊗ B0(H) ⊗ A). By definition Vbσ3 = I bA ⊗(bσ ⊗ idA)V and V1σ = (id bA ⊗σ)V ⊗ IA. Similarly both
sides of (1.6) belongs to M(bA ⊗ B0(H) ⊗ A), V1ρ = (id bA ⊗ρ)V ⊗ IA and Vbρ3 = I bA ⊗(bρ ⊗ idA)V .
Inserting on (1.5) V = flip(bV ∗) one can easily show that bVσ3bV1bσ = bV1bσbV13bVσ3. It shows that (bσ, σ)
is a Heisenberg pair for bG.
and bA appear as C∗-algebras acting on H and the embeddings A ֒→ B(H) and bA ֒→ B(H) form a
Heisenberg pair. Let (σ,bσ) be a Heisenberg pair acting on a Hilbert space H. Using unitary antipodes
R and bR (see [24]) we can construct anti-Heisenberg pair acting on H. H is the Hilbert space complex-
andba ∈ bA we set
bρ(ba) =bσ(ba
Then (ρ,bρ) is an anti-Heisenberg pair acting on H.
Let (σ,bσ) be a Heisenberg pair acting on H and M (cM resp.) be the weak closure of σ(A) (bσ(bA)
the set of all linear functionals on A (bA resp.)
resp.). We denote by A∗ (bA∗ resp.)
extensions to normal functionals on M (cM resp):
A∗ =(cid:8)µ ◦ σ : µ ∈ B(H)∗(cid:9) ,
bA∗ =(cid:8)µ ◦bσ : µ ∈ B(H)∗(cid:9) .
We know that σ andbσ are faithful. Therefore A∗ and bA∗ are weakly dense in the set of all continuous
functionals on A and bA resp.. It turns out that A∗ and bA∗ are independent of the choice of Heisenberg
pair. We may even replace (σ,bσ) by an anti-Heisenberg pair.
functionals on bA and normal representations of bA. Representations in Heisenberg and anti-Heisenberg
Algebras A and bA coincide with the closures of the sets of slices of V :
A =n(bω ⊗ idA)V :bω ∈ bA∗onorm closure
bA =n(id bA ⊗ ω)V : ω ∈ A∗onorm closure
Elements of A∗ are called normal functionals on A. A representation φ ∈ Rep(A, K) is called
normal if its matrix elements are normal: µ ◦ φ ∈ A∗ for any µ ∈ B(K)∗. Similarly one defines normal
Let (σ,bσ) be a Heisenberg pair and (ρ,bρ) be an anti-Heisenberg pair. Combining (1.4) with (1.5)
pairs are normal.
and (1.6) we get
that admit
(1.7)
,
.
(id bA ⊗ (σ ⊗ idA)∆)V = V1σV13 = Vbσ3V1σV ∗
bσ3,
1ρ,
((id bA ⊗bρ)b∆ ⊗ idA)V = Vbρ3V13 = V1ρVbρ3V ∗
(cid:0)id bA ⊗ (ρ ⊗ idA) ◦ flip ◦∆(cid:1) V = V13V1ρ = V
(cid:16)(id bA ⊗bσ) ◦ flip ◦b∆ ⊗ idA(cid:17) V = V13Vbσ3 = V
bρ3V1ρVbρ3,
1σVbσ3V1σ
∗
∗
LANDSTAD-VAES THEORY
7
Computing appropriate slices and taking into account (1.7) we obtain
(σ ⊗ idA)∆(a) = Vbσ2(σ(a) ⊗ IA)V ∗
bσ2,
1ρ,
(1.8)
(ρ ⊗ idA) ◦ flip ◦∆(a) = V
(id bA ⊗bρ)b∆(ba) = V1ρ(I bA ⊗bρ(ba))V ∗
1σ(cid:0)I bA ⊗bσ(ba)(cid:1) V1σ
(id bA ⊗bσ) ◦ flip ◦b∆(ba) = V
bρ2 (ρ(a) ⊗ IA) Vbρ2,
∗
∗
ω′ ∗ ω = (ω′ ⊗ ω) ◦ ∆.
for any a ∈ A andba ∈ bA.
(1.9)
We shall use convolution product of normal functionals. For any ω′, ω ∈ A∗ we set
Using first formula of (1.8) one can easily show that ω′ ∗ ω ∈ A∗.
shall use a canonical Heisenberg pair in section 5 to construct crossed product. To remove possible
doubts about this construction we have to show that there exists a canonical Heisenberg pair.
Let (σ,bσ) be a Heisenberg pair acting on a Hilbert space H. We say that (σ,bσ) is canonical if there
exists ρ ∈ Rep(A, H) such that (ρ, σ) is a commuting pair4 and (ρ,bσ) is an anti-Heisenberg pair. We
Proposition 1.2. Let (σ,bσ) be a Heisenberg pair acting on a Hilbert space H and (ρ,bρ) be an anti-
Heisenberg pair acting on a Hilbert space H . For any a ∈ A andba ∈ bA we set
Then (σ′,bσ′) is a canonical Heisenberg pair acting on H ′ = H ⊗ H.
Proof. We shall prove that (σ′,bσ′) is a Heisenberg pair. In the following considerations we deal with
elements of M(bA⊗B0(H ⊗H)⊗A) = M(bA⊗B0(H)⊗B0(H)⊗A). Clearly V1σ′ = V1σ and Vbσ′4 = Vbσ4Vbρ4
(the latter follows from the second formula of (1.4)). We assumed that (σ,bσ) is a Heisenberg pair:
Vbσ4V1σ = V1σV14Vbσ4. The reader should notice that nontrivial legs of V1σ and Vbρ4 belong to disjoint
sets. Therefore the two unitaries commute. Now we have:
bσ′(ba) = (bρ ⊗ bσ)b∆(ba).
σ′(a) = IH ⊗ σ(a),
Vbσ′4V1σ′ = Vbσ4Vbρ4V1σ = Vbσ4V1σVbρ4 = V1σV14Vbσ4Vbρ4 = V1σ′ V14Vbσ′4
It shows that (σ′,bσ′) is a Heisenberg pair.
For any a ∈ A we set ρ′(a) = ρ(a) ⊗ IH . Then ρ′ ∈ Rep(A, H ⊗ H). We shall prove that (ρ′,bσ′)
is an anti-Heisenberg pair. Clearly V1ρ′ = V1ρ and Vbσ′4 = Vbσ4Vbρ4. We assumed that (ρ,bρ) is an
anti-Heisenberg pair: V1ρVbρ4 = Vbρ4V14V1ρ. The reader should notice that nontrivial legs of V1ρ and
Vbσ4 belong to disjoint sets. Therefore the two unitaries commute. Now we have:
V1ρ′ Vbσ′4 = V1ρVbσ4Vbρ4 = Vbσ4V1ρVbρ4 = Vbσ4Vbρ4V14V1ρ = Vbσ′4V14V1ρ′ .
semi-regularity.
An important concept of the theory of locally compact quantum groups is that of regularity and
It shows that (ρ′,bσ′) is an anti-Heisenberg pair. Obviously (ρ′.σ′) is a commuting pair. Therefore the
Heisenberg pair (σ′,bσ′) is canonical.
It was introduced by Baaj and Skandalis in [2] and Baaj in [1]. Let (σ,bσ) be a
Heisenberg pair acting on H, W = Vbσσ = (bσ ⊗ σ)V and
C =n(idH ⊗ ω)(ΣW ) : ω ∈ B(H)∗oCLS
where Σ be the flip operator acting on H ⊗ H: Σ(x ⊗ y) = y ⊗ x for all x, y ∈ H. We say that W
is semi-regular if B0(H) ⊂ C and regular if B0(H) = C. It was shown in [2] that W is regular if and
only if
(cid:3)
,
(1.10)
(bA ⊗ IA)V (I bA ⊗A) = bA ⊗ A.
This formula no longer depends on the choice of Heisenberg pair. This is an equality of subsets of
unitary W related to the group). It is known that V
M(bA ⊗ A). Due to this fact regularity is the property of the group (not of a particular multiplicative
bR ⊗ R to the both sides (1.10) we obtain equivalent condition:
(I bA ⊗A)V (bA ⊗ IA) = bA ⊗ A.
bR⊗R = V . Applying anti-multiplicative involution
4it means that [ρ(a), σ(a′)] = 0 for any a, a′ ∈ A
(1.11)
8
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
2. Main definitions and results
Let G = (A, ∆) be a locally compact quantum group and D be a C∗-algebra. A weak action of G
on D is a bilinear mapping α : A∗ × D −→ D subject to a number of conditions. One of the conditions
is faithfulness: α is called faithful if for any non-zero d ∈ D there exists ω ∈ A∗ such that α(ω, d) 6= 0.
To formulate the definition of weak action in a compact form we need the concept of faithful
presentation.
(2.1)
for any d ∈ D and ω ∈ A∗.
Definition 2.1. Let K and H be Hilbert spaces, π ∈ Rep(D, K) and eπ ∈ Mor(D, B0(K) ⊗ A). We
say that (π,eπ) is a faithful presentation of α on K if π is faithful and
π(cid:16)α(ω, d)(cid:17) = (idK ⊗ ω)eπ(d)
Clearlyeπ is injective for faithful α. Now we are ready to formulate our main definition.
WA1. D is generated byhα(ω, D) : ω ∈ A∗i,
Definition 2.2. Let D be a C∗-algebra and α be a faithful bilinear mapping from A∗ × D into D. We
say that α is a weak right action of G on D if the following three conditions are fulfilled:
WA0. α admits a faithful presentation.
WA2. α(ω, α(ω′, d)) = α(ω ∗ ω′, d) for any d ∈ D and ω, ω′ ∈ A∗.
Alternatively we say that the pair (D, α) is a weak G-dynamical system. Let (D, α) and (D′, α′)
be weak G-dynamical systems and ı : D → D′ be an isomorphism of C∗-algebras. We say that ı is an
isomorphism of dynamical systems if ı intertwins α and α′:
for any d ∈ D and ω ∈ A∗.
ı(α(ω, d)) = α′(ω, ı(d))
It would be nice to replace Condition WA0 by a number of algebraic and topological conditions
imposed directly on the bilinear map α. This issue goes beyond the present paper. Eventually we shall
return to it later.
Let (D, α) be a weak G-dynamical system and
(2.2)
We say that D1 is a range of α. The range D1 is a norm closed subspace of C∗-algebra D, so it
carries a structure of an operator space in the sense of [5]. Moreover D1 is invariant under Hermitian
conjugation. In general (when G is not semiregular) D1 is not a C∗-algebra (see [3, Proposition 5.7]).
According to Condition WA1, D1 generates the C∗-algebra D. Therefore
D1 =hα(ω, D) : ω ∈ A∗i .
D =hDk : k = 1, 2, 3, . . .i ,
π(D1) =h(idK ⊗ ω)eπ(D) : ω ∈ A∗i .
shows that
where D2 = D1D1, D3 = D1D1D1 and so on. Let (π,eπ) be a faithful presentation of α. Then (2.1)
(2.3)
The weakness is a weak condition imposed on actions of quantum groups on C∗-algebras. Stronger
are weak continuity and continuity [3]. We say that α is a weakly continuous action of G if there exixts
eα ∈ Mor(D, D ⊗ A) such that
for any ω ∈ A∗ and d ∈ D. If moreover the Podleś condition eα(D)(ID ⊗A) = D ⊗ A is satisfied then
α(ω, d) = (idD ⊗ ω)eα(d)
α is called continuous. Clearly continuous actions are weakly continuous.
Theorem 2.3. Let α be a weak action of a regular locally compact quantum group G on a C∗-algebra
D. Then α is a continuous action.
This theorem belongs essentially to Baaj, Skandalis and Vaes [3]. We shall present a proof because
our setting is slightly more general than the one used in [3] and because we deal with right actions.
Left actions considered in [3] did not require the use of the anti-Heisenberg pair.
Definition 2.4. G-product is a triple (B, β, ψ), consisting of a C∗-algebra B, a left continuous action
LANDSTAD-VAES THEORY
9
β ∈ Mor(B, bA ⊗ B) of bG on B and an injective morphism ψ ∈ Mor(bA, B) such that the diagram
B
ψ
(2.4)
is commutative.
b∆
bA
bA ⊗ bA
id bA ⊗ ψ
β
/ bA ⊗ B
We recall that β is a continuous left action of bG on B if β is an injective morphism β ∈ Mor(B, bA⊗B)
such that (id bA ⊗β)β = (b∆ ⊗ idB)β and
Let (B1, β1, ψ1) and (B2, β2, ψ2) be G-products and : B1 −→ B2 be an isomorphism of C∗-
(2.5)
(bA ⊗ IB)β(B) = bA ⊗ B.
algebras. We say that is an isomorphism of the G-products if the diagram
(2.6)
bA
ψ1
8qqqqqqqqqqqqq
&▼▼▼▼▼▼▼▼▼▼▼▼▼
ψ2
B1
B2
β 1
β 2
bA ⊗ B1
id bA ⊗
/ bA ⊗ B2
is commutative.
Definition 2.5. Let (B, β, ψ) be a G-product and D ⊂ M(B) be a C∗-subalgebra. We say that D is a
Landstad algebra of (B, β, ψ) if the following three conditions are satisfied:
C1. β(d) = I bA ⊗ d for any d ∈ D,
C2. B = ψ(bA)D,
C3. The C∗-algebra generated by
coincides with D.
(cid:2)(idB ⊗ ω)(cid:0)Vψ2(D ⊗ IH )V ∗
ψ2(cid:1) : ω ∈ A∗(cid:3)
We know that Bψ(bA) = B (this is because ψ ∈ Mor(bA, B)). Therefore BD = Bψ(bA)D = BB = B.
It shows that Landstad algebra is a non-degenerate subalgebra od M(B). In other words the embedding
D ֒→ M(B) is a morphism from D into B (it belongs to Mor(D, B)). Let (B, β, ψ) be a G-product
and D be its Landstad algebra. Then using Condition C1 and commutativity of (2.6) we get
(2.7)
(2.8)
β(cid:0)ψ(ba)d(cid:1) =(cid:16)(id bA ⊗ ψ)b∆(ba)(cid:17) (I bA ⊗ d).
for anyba ∈ bA and d ∈ D. According to Condition C2, this formula determines β completely.
Theorem 2.6. Any G-product admits one and only one Landstad algebra. Let (B, β, ψ) be a G-product
and D be its Landstad algebra. Then the formula
for any ω ∈ A∗ and d ∈ D defines a weak right action α of G on D.
ψ2(cid:1)
α(ω, d) = (idB ⊗ ω)(cid:0)Vψ2(d ⊗ IA)V ∗
The resulting weak G-dynamical system (D, α) will be called Landstad G-dynamical system of
(B, β, ψ). It turns out that Landstad G-dynamical system determines the related G-product.
Theorem 2.7. Let (B1, β1, ψ1) and (B2, β2, ψ2) be G-products, (D1, α1) and (D2, α2) be corresponding
Landstad G-dynamical systems and ı : D1 −→ D2 be an isomorphism of (D1, α1) and (D2, α2). Then
G-products (B1, β1, ψ1) and (B2, β2, ψ2) are isomorphic: there exists unique isomorphism : B1 −→ B2
of the G-products extending ı.
To establish the one to one correspondence between weak G-dynamical systems and G-products we
have to construct a G-product corresponding to a given weak G-dynamical system. The construction
is described in the following theorem.
/
/
/
/
/
8
&
/
10
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
and d ∈ D we have
Theorem 2.8. Let (D, α) be a weak G-dynamical system, (π,eπ) be a faithful presentation of α on a
Hilbert space K and (σ,bσ) be a canonical Heisenberg pair acting on a Hilbert space H. For anyba ∈ bA
we set ψ(ba) = IK ⊗bσ(ba). Let B = ψ(bA)(idK ⊗ σ)eπ(D). Then B ⊂ B(K ⊗ H) and
1. B is a C∗-algebra, ψ(bA), (id bA ⊗ σ)eπ(D) ⊂ M(B) and the embeddings are morphisms from ψ(bA)
and (idK ⊗ σ)eπ(D) into B. Consequently ψ ∈ Mor(bA, B) and (idK ⊗ σ)eπ ∈ Mor(D, B).
2. B is equipped with unique continuous left action β ∈ Mor(B, bA ⊗ B) of bG such that for anyba ∈ bA
3. (B, β, ψ) is a G-product, Landstad algebra of (B, β, ψ) coincides with (idK ⊗ σ)eπ(D). Original
precisely the map (idK ⊗ σ)eπ : D −→ (idK ⊗ σ)eπ(D) is an isomorphism of the weak G-dynamical
α of bG and denoted by D ⋊α bG. In the same case β is called the dual action and often denoted by bα.
β(cid:0)ψ(ba)(cid:1) = (id bA ⊗ ψ)b∆(ba),
β(cid:0)(idK ⊗ σ)eπ(d)(cid:1) = I bA ⊗ (idK ⊗ σ)eπ(d).
The C∗-algebra B appearing in the above Theorem is called a crossed product of D by the action
weak G-dynamical system (D, α) is isomorphic to the Landstad G-dynamical system of (B, β, ψ). More
3. Weak actions
systems.
In this section we investigate properties of G-dynamical systems.
In particular we shall prove
Theorem 2.3. We start with the following Proposition describing analytical properties that follow from
Definition 2.2.
Proposition 3.1. Let (D, α) be a G-dynamical system, (π,eπ) be a faithful presentation of α on a
Hilbert space K, (σ,bσ) be a Heisenberg pair acting on a Hilbert space H and (ρ,bρ) be an anti-Heisenberg
pair acting on a Hilbert space H. Then for any ω ∈ A∗, µ ∈ B(H)∗ and d ∈ D we have
∗
eπ(α(ω, d))1σ = (idK⊗H ⊗ ω)(cid:16)Vbσ3eπ(d)1σV
eπ(α(µ ◦ ρ, d)) = (idK ⊗ µ ⊗ idA)(cid:16)V
bσ3(cid:17) ,
bρ3eπ(d)1ρVbρ3(cid:17) .
∗
Proof. For any ω′ ∈ A∗ we have
(3.1)
(3.2)
Therefore
(3.3)
(idK ⊗ ω′ ⊗ ω)(idK ⊗ ∆)eπ(d) = (idK ⊗ ω′ ∗ ω)eπ(d) = π(α(ω′ ∗ ω, d))
= π(α(ω′, α(ω, d))) = (idK ⊗ ω′)eπ(α(ω, d)).
First formula of (1.8) shows that
Comparing this formula with (3.3) we get (3.1).
Third formula of (1.8) shows that
eπ(α(ω, d)) = (idK ⊗ idA ⊗ ω)(idK ⊗ ∆)eπ(d).
(idK ⊗ σ ⊗ idA)(idK ⊗ ∆)eπ(d) = Vbσ3eπ(d)1σV
(idA ⊗ µ ◦ ρ)∆(a) = (µ ⊗ idA)(cid:16)V
bρ2 (ρ(a) ⊗ IA) Vbρ2(cid:17)
bσ3.
∗
∗
for any a ∈ A. Therefore
On the other hand, inserting ω = µ ◦ ρ in (3.3) we obtain
(idK ⊗ (idA ⊗ µ ◦ ρ)∆)eπ(d) = (idK ⊗ µ ⊗ idA)(cid:16)V
eπ(α(µ ◦ ρ, d)) = (idK ⊗ (idA ⊗ µ ◦ ρ)∆)eπ(d).
∗
bρ3eπ(d)1ρVbρ3(cid:17) .
and (3.2) follows.
(cid:3)
Proposition 3.2. Let (D, α) be a weak G-dynamical system and D1 ⊂ D be the range of α (cf (2.2)).
acting on a Hilbert space H. Then
Morover let (π,eπ) be a faithful presentation of α on a Hilbert space K and (σ,bσ) be a Heisenberg pair
1.(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D)1σ is a C∗-algebra of operators acting on K ⊗ H in a non-degenerate way,
LANDSTAD-VAES THEORY
11
2. We have
(3.4)
(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D1)1σ =(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D)1σ.
bσ3(cid:17) : ω ∈ A∗i .
eπ(D1)1σ =h(idK⊗H ⊗ ω)(cid:16)Vbσ3eπ(D)1σV
∗
Proof. We shall use formula (3.1). Taking the closed linear span over all ω ∈ A∗ and d ∈ D we get
We compute left hand side of (3.4). We know that unitary V belongs to M (bA ⊗ A). Therefore
(bA ⊗ A)V = bA ⊗ A. In the following computation we use emit-absorb rules introduced by (1.3). At
first (in the second equality) ω emits A to the right, later (in the forth equality) ω absorbs A back.
Finally in the last equality follows from second formula of (1.7).
bσ3(cid:17) : ω ∈ A∗i
∗
(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D1)1σ =h(idK⊗H ⊗ ω)(cid:16)(cid:16)IK ⊗bσ(bA) ⊗ IA(cid:17) Vbσ3eπ(D)1σV
bσ3(cid:17) : ω ∈ A∗i
bσ3(cid:17) : ω ∈ A∗i
bσ3(cid:17) : ω ∈ A∗i
=h(idK⊗H ⊗ ω)(cid:16)(cid:16)(bA ⊗ A)V(cid:17)bσ3eπ(D)1σV
=h(idK⊗H ⊗ ω)(cid:16)(bA ⊗ A)bσ3eπ(D)1σV
=(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D)1σh(idK⊗H ⊗ ω)(cid:16)V
=(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D)1σ(cid:16)IK ⊗bσ(bA)(cid:17) .
∗
∗
∗
We showed that
(3.5)
The expression on the right hand side is invariant under hermitian conjugation. Therefore
(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D1)1σ =(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D)1σ(cid:16)IK ⊗bσ(bA)(cid:17) .
where n = 1, 2, 3 . . . and Dn is the product of n copies of D1. Taking closed linear span over all n and
remembering that D is generated by D1 we see that
Proposition 3.3. Let (D, α) be a weak G-dynamical system, D1 ⊂ D be the range of α (cf (2.2)) and
(cid:3)
(3.6)
Iterating this formula we obtain
(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D1)1σ =eπ(D1)1σ(cid:16)IK ⊗bσ(bA)(cid:17) .
(cid:16)IK ⊗bσ(bA)(cid:17)eπ(Dn)1σ =eπ(Dn)1σ(cid:16)IK ⊗bσ(bA)(cid:17) ,
(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D)1σ =eπ(D)1σ(cid:16)IK ⊗bσ(bA)(cid:17) .
(cid:16)IK ⊗bσ(bA)(cid:17)eπ(D)1σ ∈ C∗(K ⊗ H).
(π,eπ) be a faithful presentation of α on a Hilbert space K. Then
Combining (3.6) with (3.5) we obtain (3.4).
Proposition 1.1 shows that
(3.7)
Moreover, if G is a regular group (cf. (1.10)) then
(IK ⊗A)eπ(D1)(IK ⊗A) = π(D1) ⊗ A.
eπ(D1)(IK ⊗A) = π(D1) ⊗ A.
(3.8)
(3.9)
Taking the closed linear span over all µ ∈ B(H)∗ and d ∈ D we obtain
Proof. Let (ρ,bρ) be an anti-Heisenberg pair acting on a Hilbert space H. We shall use formula (3.2).
eπ(D1) =h(idK ⊗ µ ⊗ idA)(cid:16)V
∗
bρ3eπ(D)1ρVbρ3(cid:17) : µ ∈ B(H)∗i .
We computeeπ(D1)(IK ⊗A). We know that unitary V belongs to M (bA ⊗ A). Therefore V (bA ⊗ A) =
bA ⊗ A. In the following computation we use emit-absorb rules introduced by (1.3). At first (in the
Denote by Φ[X] the right hand side of (3.10).
It makes sense for any closed linear subspace X of
(3.10)
(3.11)
M(bA ⊗ A). In particular
eπ(D1)(IK ⊗A) =h(idK ⊗ µ ⊗ idA)(cid:16)X bρ3eπ(D)1ρ(cid:17) : µ ∈ B(H)∗i .
Φ[bA ⊗ A] =h(idK ⊗ µ ⊗ idA)(cid:16)(bA ⊗ A)bρ3eπ(D)1ρ(cid:17) : µ ∈ B(H)∗i
=h(idK ⊗ µ ⊗ idA)(cid:16)(IK⊗H ⊗A)eπ(D)1ρ(cid:17) : µ ∈ B(H)∗i
=h(idK ⊗ µ ◦ ρ)eπ(D) : µ ∈ B(H)∗i ⊗ A = π(D1) ⊗ A.
12
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
∗
second equality) µ emits bρ(bA) to the left, later (in the forth equality) µ absorbsbρ(bA) back. Finally in
the last equality µ emits bρ(bA) to the right.
eπ(D1)(IK ⊗A) =h(idK ⊗ µ ⊗ idA)(cid:16)V
=(cid:20)(idK ⊗ µ ⊗ idA)(cid:18)V
=h(idK ⊗ µ ⊗ idA)(cid:16)V
=(cid:20)(idK ⊗ µ ⊗ idA)(cid:18)(cid:16)V
=(cid:20)(idK ⊗ µ ⊗ idA)(cid:18)(cid:16)(bA ⊗ IA)V
Let X be a subset of M(bA ⊗ A) introduced by formula
X = (bA ⊗ IA)V
bρ3eπ(D)1ρVbρ3(IK⊗H ⊗A)(cid:17) : µ ∈ B(H)∗i
bρ3eπ(D)1ρ(cid:16)V (bA ⊗ A)(cid:17) bρ3(cid:19) : µ ∈ B(H)∗(cid:21)
bρ3eπ(D)1ρ(bA ⊗ A)bρ3(cid:17) : µ ∈ B(H)∗i
(I bA ⊗A)(cid:17) bρ3eπ(D)1ρ(cid:19) : µ ∈ B(H)∗(cid:21)
(I bA ⊗A)(cid:17)bρ3eπ(D)1ρ(cid:19) : µ ∈ B(H)∗(cid:21) .
(I bA ⊗A).
Then
∗
∗
∗
∗
∗
In the second equality µ absorbs bρ(bA) from the right, at the end we use (2.3). If G is regular then
X = bA ⊗ A (cf (1.11)) and (3.8) follows immediately from (3.11).
We go back to general case. To compute left hand side of (3.7) we use (3.10):
(IK ⊗A)eπ(D1)(IK ⊗A) =h(idK ⊗ µ ⊗ idA)(cid:16)(cid:0)(I bA ⊗A)X(cid:1)bρ3eπ(D)1ρ(cid:17) : µ ∈ B(H)∗i
On the other hand (I bA ⊗A)X = (bA ⊗ A)V ∗(I bA ⊗A) = bA ⊗ A. Using again (3.11) we get (3.7).
Proof of Theorem 2.3. Let (π,eπ) be a faithful presentation of α on a Hilbert space K. Identifying D
we write eα instead ofeπ. Then (3.8) takes the form
with π(D) we may assume that D ∈ C∗(K) and that π is the embedding D ֒→ B(K). In this context
Now we are able to present the prove of Baaj-Skandalis-Vaes theorem.
= Φ[(I bA ⊗A)X].
(cid:3)
Let n ∈ N and Dn be the product of n copies of D1. We have:
eα(D1)(IK ⊗A) = D1 ⊗ A.
obtain Podleś condition:
eα(D2)(IK ⊗A) =eα(D1)eα(D1)(IK ⊗A) =eα(D1)(IK ⊗A)(D1 ⊗ IK) = D2 ⊗ A
and similarly eα(Dn)(IK ⊗A) = Dn ⊗ A for any n. Taking the closed linear span over all natural n we
It shows that eα(D) ⊂ M(D ⊗ A) and eα ∈ Mor(D, D ⊗ A). To end the proof it is sufficient to notice
that in present notation (2.1) takes the form
eα(D)(IK ⊗A) = D ⊗ A.
α(ω, d) = (idD ⊗ ω)eα(d).
It means that action α is continuous.
(cid:3)
LANDSTAD-VAES THEORY
13
4. G-products
We shall analyse a structure of a G-product (B, β, ψ). The section contains the proofs of Theorems
2.6 and 2.7. At the beginning we follow the path elaborated by S. Vaes in [18, Proof of Thm 6.7]. Let
presented in this section. It is important to realise, which of them are independent of the choice of
(σ,bσ) be a Heisenberg pair acting on a Hilbert space H. The pair will be used in many constructions
(σ,bσ). We shall discuss this issue later.
Proposition 4.1. Let (B, β, ψ) be a G-product, β′ = flip ◦β and ϕ ∈ Mor(B, B⊗B0(H)) be an injective
morphism such that
(4.1)
for any b ∈ B. Then
(4.2)
(4.3)
ϕ(b) = Vψσβ′(b)1bσV ∗
ψσ
(β ⊗ idH )ϕ(b) = I bA ⊗ ϕ(b)
ϕ ◦ ψ(ba) = IB ⊗bσ(ba),
for anyba ∈ bA and b ∈ B.
Proof. Let ba ∈ bA. According to (2.4), β(ψ(ba)) = b∆(ba)1ψ. Therefore β′(ψ(ba)) = (flip ◦b∆(ba))ψ2 and
using the last equality in (1.8) we get
∗
Now (4.2) follows immediately from (4.1).
β′(ψ(ba))1bσ = (flip ◦b∆(ba))ψ bσ = V
ψσ (IB ⊗bσ(ba)) V ψσ.
Let b ∈ B. Then
(4.4)
(β ⊗ idH )ϕ(b) = (β ⊗ idH ) (Vψσ) (β ⊗ idH )(cid:16)β′(b)1bσ(cid:17) (β ⊗ idH )(cid:0)V ∗
ψσ(cid:1) .
Taking into account (2.4) and second formula of (1.4) we obtain:
(β ⊗ idH )(cid:16)Vψσ(cid:17) = (β ◦ ψ ⊗ σ)V =(cid:16)(id bA ⊗ ψ)b∆ ⊗ σ(cid:17) V
=(cid:0)id bA ⊗ ψ ⊗ σ(cid:1)(cid:16)b∆ ⊗ idA(cid:17) V =(cid:0)id bA ⊗ ψ ⊗ σ(cid:1) (V23V13) = VψσV1σ.
We know that β is a left action of bG: (id bA ⊗β)β = (b∆ ⊗ idB)β. A moment of reflection shows that
(β ⊗ id bA)β′ = flip12(idB ⊗ flipb∆)β′. Taking into account the last formula of (1.8) we obtain
(β ⊗ idH )(cid:16)β′(b)1bσ(cid:17) = (β ⊗ bσ)β′(b) = (idB0(H)⊗B ⊗bσ) flip12(idB ⊗ flipb∆)β′(b)
= flip12(idB ⊗ (id bA ⊗bσ) flipb∆)β′(b)
2σβ′(b)1bσV2σ(cid:17) = V ∗
= flip12(cid:16)V
∗
1σβ′(b)2bσV1σ.
Inserting the results of above computations to (4.4) we see that
(β ⊗ idH )ϕ(b) = VψσV1σV ∗
1σβ′(b)2bσV1σ V ∗
1σV ∗
ψσ
= Vψσβ′(b)2bσV ∗
ψσ = ϕ(b)23 = IB ⊗ ϕ(b).
Formula (4.2) is proven
(cid:3)
Proposition 4.2. Let (B, β, ψ) be a G-product, ϕ ∈ Mor(B, B ⊗ B0(H)) be the injective morphism
introduced by (4.1) and
(4.5)
D1 =h(idB ⊗ µ)ϕ(B) : µ ∈ B(H)∗i .
Then D1 is a norm closed linear subset of M(B). Elements of D1 are β-invariant:
(4.6)
for any d ∈ D1 and
(4.7)
β(d) = I bA ⊗ d
B = ψ(bA)D1.
14
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
Proof. Relation (4.6) follows immediately from (4.3).
B ⊗ B0(H) and
We know that V1σ is a unitary multiplier of bA ⊗ B0(H). Therefore Vψσ is a unitary multiplier of
(B ⊗ B0(H))V ∗
Moreover Podleś condition (2.5) easily shows that
(bA ⊗ B0(H))V1σ = bA ⊗ B0(H)
ψσ = B ⊗ B0(H)
Using the above formulae we obtain
(IB ⊗ B0(H))β′(B)1bσ = B ⊗ B0(H).
Taking into account the equality ψ(bA)B = B (this is because ψ ∈ Mor(bA, B)) we get
ψσ
ψσ
ψσ
ψσ
(ψ(bA) ⊗ B0(H))ϕ(B) = (ψ(bA) ⊗ B0(H))Vψσβ′(B)1bσV ∗
= (ψ ⊗ idH )(cid:16)(bA ⊗ B0(H))V1σ(cid:17) β′(B)1bσV ∗
= (ψ ⊗ idH )(cid:16)bA ⊗ B0(H)(cid:17) β′(B)1bσV ∗
= (ψ(bA) ⊗ IH ) (IB ⊗ B0(H)) β′(B)1bσV ∗
= (ψ(bA) ⊗ IH ) (B ⊗ B0(H)) V ∗
= ψ(bA)B ⊗ B0(H).
(ψ(bA) ⊗ B0(H))ϕ(B) = B ⊗ B0(H).
ψ(bA)D1 =hψ(bA)(idB ⊗ ω)ϕ(B) : ω ∈ B(H)∗i
=h(idB ⊗ ω)n(ψ(bA) ⊗ B0(H))ϕ(B)o : ω ∈ B(H)∗i
=h(idB ⊗ ω)nB ⊗ B0(H)o : ω ∈ B(H)∗i = B.
Now, using (1.3) we obtain
ψσ = (ψ(bA) ⊗ IH ) (B ⊗ B0(H))
Formula (4.7) is proven.
(cid:3)
With the notation introduced in the previous Propositions we have
Proposition 4.3. C∗-algebra generated by D1 is the only Landstad algebra of (B, β, ψ).
Therefore
Proof. Let D be a C∗-subalgebra of M(B) satisfying Conditions C1 and C2 of Definition 2.5. Condition
C1 shows that β(D) = I bA ⊗D. Therefore ϕ(D) = (idB ⊗ σ)(cid:16)Vψ2(D ⊗ IA)V ∗
ψ2(cid:17). I bA was replaced by
IA because bσ(I bA) = IH = σ(IA). We know that ω ∈ A∗ if and only if ω = µ ◦ σ, where µ ∈ B(H)∗.
h(idB ⊗ ω)(cid:16)Vψ2(D ⊗ I bA)V ∗
ψ2(cid:17) : ω ∈ A∗i =h(idB ⊗ µ)ϕ(D) : µ ∈ B(H)∗i
h(idB ⊗ µ)ϕ(D) : µ ∈ B(H)∗i =h(idB ⊗ µ)(cid:16)(cid:16)IB ⊗bσ(bA)(cid:17) ϕ(D)(cid:17) : µ ∈ B(H)∗i
We shall prove that the latter set coincides with D1. Indeed:
=h(idB ⊗ µ)ϕ(cid:16)ψ(bA)D(cid:17) : µ ∈ B(H)∗i
=h(idB ⊗ µ)ϕ(B) : µ ∈ B(H)∗i = D1.
The first equality follows from (1.3), the second one from (4.2) and the third one from C2. We showed
that
(4.8)
h(idB ⊗ ω)(cid:16)Vψ2(D ⊗ I bA)V ∗
ψ2(cid:17) : ω ∈ A∗i = D1.
Assume now that D is a Landstad algebra for (B, β, ψ). Then Condition C3 shows that D is
generated by D1.
Conversely assume that D is the C∗-algebra generated by D1. Then D ⊂ M(B). A moment of
reflection shows that
D =(cid:16)D1 ∪ D1D(cid:17)CLS
.
LANDSTAD-VAES THEORY
15
We know that ψ(bA)D1 = B (see (4.7)). The simple observation: ψ(bA)D1D = BD ⊂ B combined with
the above formula shows that ψ(bA)D = B. Condition C2 is verified.
According to (4.6) elements of D1 are β-invariant. The same is true for elements of D and Condition
C1 is verified.
Now we may use (4.8). Remembering that D is generated by D1 we see that Condition C3 is
satisfied. It means that D is a Landstad algebra for (B, β, ψ).
The existence and uniqueness of Landstad algebra is shown.
(cid:3)
Now we are able to prove Theorem 2.6.
Proof. Let D ⊂ M(B) be the Landstad algebra of G-product (B, β, ψ) (see Definition 2.5). It follows
immediately from C3, that formula (2.8) defines a mapping α acting from A∗ × D into D satisfying
Condition WA1 of Definition 2.2. We shall check Condition WA0.
One may assume that B ∈ C∗(K), where K is a Hilbert space. Then D ⊂ M(B) ⊂ B(K). Denote
by π′ the embedding of B into B(K), by π the restriction of π′ to D (so π is the embedding of D into
B(K)) and byeπ ∈ Mor(D, B0(K) ⊗ A) the morphism introduced by the formula
Let ω ∈ A∗. Taking into account (2.8) we obtain
ψ2(cid:1)
eπ(d) = (π′ ⊗ idH )(cid:0)Vψ2(d ⊗ IH )V ∗
ψ2(cid:17)
π(α(ω, d)) = (π′ ⊗ ω)(cid:16)Vψ2(d ⊗ IH )V ∗
is verified.
This formula coincides with (2.1). Therefore (π,eπ) is a faithful presentation of α and Condition WA0
We shall show that α introduced by (2.8) obeys Condition WA2. This is a matter of easy compu-
tation. Let ω, ω′ ∈ A∗ and d ∈ D. Then
= (idK ⊗ ω)eπ(d).
(4.9)
ψ2(cid:17)
α(ω ∗ ω′, d) = (idB ⊗ (ω ∗ ω′))(cid:16)Vψ2(d ⊗ IA)V ∗
ψ2(cid:17)
= (idB ⊗ ω ⊗ ω′)(idB ⊗ ∆)(cid:16)Vψ2(d ⊗ IA)V ∗
ψ2(cid:17)
= (idB ⊗ ω ⊗ ω′)(cid:16)Vψ2Vψ3(d ⊗ IA ⊗ IA)V ∗
= (idB ⊗ ω)(cid:16)Vψ2(α(ω′, d) ⊗ IA)V ∗
ψ2(cid:17) = α(ω, α(ω′, d)).
ψ3V ∗
We have shown that α is a weak right action of G on D. The proof of Theorem 2.6 is complete.
(cid:3)
For any d ∈ D and µ ∈ B0(H) we have
(4.10)
Indeed, β(d) = I bA ⊗ d, β′(d) = d ⊗ I bA, β′(d)1bσ = d ⊗ IH and ϕ(d) = Vψσ(d ⊗ IH )V
(idB ⊗ µ)ϕ(d) = α(µ ◦ σ, d).
∗
ψσ. Therefore
and using (2.8) we get (4.10).
(idB ⊗ µ)ϕ(d) = (idB ⊗ µ ◦ σ)(cid:16)Vψ2(d ⊗ IA)V
∗
ψ2(cid:17)
To prove the uniqueness Theorem 2.7 we start with the following proposition:
Proposition 4.4. Let (B1, β1, ψ1) and (B2, β2, ψ2) be G-products and (D1, α1), (D2, α2) be corre-
sponding Landstad weak G-dynamical systems and ı : D1 −→ D2 be an isomorphism of the weak
G-dynamical systems (D1, α1) and (D2, α2). It means that
(4.11)
for any ω ∈ A∗ and d ∈ D1. Then B1 and B2 are isomorphic as Banach spaces. More precisely: there
exists a linear, isometric map : B1 −→ B2 such that (B1) = B2 and
ı(α1(ω, d)) = α2(ω, ı(d))
(4.12)
for anyba,ba′ ∈ bA and d ∈ D1.
(cid:0)ψ1(ba)d(cid:1) = ψ2(ba)ı(d),
(cid:0)dψ1(ba′)(cid:1) = ı(d)ψ2(ba′),
(cid:0)ψ1(ba)dψ1(ba′)(cid:1) = ψ2(ba)ı(d)ψ2(ba′)
16
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
three formulas (4.12) defines on a dense subset of B1 with values running over a dense subset of B2.
Proof. Let D be the Landstad algebra of a G-product (B, β, ψ). We know that ψ(bA)D = B. Applying
to the both sides hermitian conjugation we get Dψ(bA) = B. Moreover ψ(bA)Dψ(bA) = B.
ψ ∈ Mor(bA, B). Therefore ψ(bA)B = B and ψ(bA)Dψ(bA) = ψ(bA)B = B. It shows that each of the
Let (σ,bσ) be a Heisenberg pair acting on a Hillbert space H, ϕ ∈ Mor(B, B ⊗ B0(H)) be the
use the B(H)-bimodul structure of B(H)∗. According to (1.1), for any µ ∈ B(H)∗ and ba,ba′ ∈ bA,
µbσ(ba), bσ(ba′)µ andbσ(ba′)µbσ(ba) are normal linear functionals on B(H) such that
morphism introduced by (4.1) and α be the weak action of G on D introduced by (2.8). We shall
Indeed
for any m ∈ B(H). Let d ∈ D. We shall prove that
(4.13)
(µbσ(ba)) (m) = µ(bσ(ba)m),
(bσ(ba′)µ) (m) = µ(mbσ(ba′)),
(bσ(ba′)µbσ(ba)) (m) = µ(bσ(ba)mbσ(ba′))
α((µbσ(ba)) ◦ σ, d) = (idB ⊗ µ)ϕ(ψ(ba)d),
α((bσ(ba′)µ) ◦ σ, d) = (idB ⊗ µ)ϕ(dψ(ba′))
α((bσ(ba′)µbσ(ba)) ◦ σ, d) = (idB ⊗ µ)ϕ(ψ(ba)dψ(ba′)).
Indeed taking into account (4.10) and (4.2) we have
In the same way one can prove two remaining formulas of (4.13).
Let b1 ∈ B1 and b2 ∈ B2 be elements of the form
α((µbσ(ba)) ◦ σ, d) = (idB ⊗ µbσ(ba))ϕ(d) = (idB ⊗ µ)(cid:0)(IB ⊗bσ(ba))ϕ(d)(cid:1)
= (idB ⊗ µ)(cid:0)ϕ(ψ(ba))ϕ(d)(cid:1) = (idB ⊗ µ)ϕ(ψ(ba)d).
r−1Xk=1
t−1Xk=s
ψ1(bak)dk +
ψ1(bak)dkψ1(ba′
t−1Xk=s
r−1Xk=1
ψ2(bak)ı(dk) +
ψ2(bak)ı(dk)ψ2(ba′
s−1Xk=r
dkψ1(ba′
s−1Xk=r
ı(dk)ψ2(ba′
k) +
k) +
k),
b1 =
b2 =
k),
where 1 ≤ r ≤ s ≤ t are integers, dk ∈ D1 for 1 ≤ k < t,bak ∈ bA for 1 ≤ k ≤ r and s ≤ k < t and
k ∈ bA for r ≤ k < t. We look for an isometry : B1 −→ B2 such that (b1) = b2. To prove the
ba′
existence of it is sufficient to show that
(4.14)
Let ϕ1 ∈ Mor(B1, B1 ⊗ B0(H)) and ϕ2 ∈ Mor(B2, B2 ⊗ B0(H)) be morhisms related to G-products
(B1, β1, ψ1) and (B2, β2, ψ2) by the formula (4.1). Combining (4.13) with (4.11) we see that
(cid:13)(cid:13)b1(cid:13)(cid:13) =(cid:13)(cid:13)b2(cid:13)(cid:13) .
(4.15)
ı(cid:0)(idB1 ⊗ µ)ϕ1(ψ1(ba)d)(cid:1) = (idB2 ⊗ µ)ϕ2(ψ2(ba)ı(d))
ı(cid:0)(idB1 ⊗ µ)ϕ1(dψ1(ba′))(cid:1) = (idB2 ⊗ µ)ϕ2(ı(d)ψ2(ba′))
ı(cid:0)(idB1 ⊗ µ)ϕ1(ψ1(ba)dψ1(ba′))(cid:1) = (idB2 ⊗ µ)ϕ2(ψ2(ba)ı(d)ψ2(ba′))
for anyba,ba′ ∈ bA, d ∈ D1 and µ ∈ B(H)∗. It shows that
ı(cid:0)(idB1 ⊗ µ)ϕ1(b1)(cid:1) = (idB2 ⊗ µ)ϕ2(b2)
for any µ ∈ B(H)∗.
(4.16)
Let m, n be one-dimensional operators acting on H: m = x) (y and n = z) (u , where x, y, z, u
are elements of H and let µ be a normal linear functional on B(H) such that µ(a) = (y a z) for any
a ∈ B(H).. Then
(IB1 ⊗ m)ϕ1(b1)(IB1 ⊗ n) = (idB1 ⊗ µ)ϕ1(b1) ⊗ x) (u ∈ D1 ⊗ B0(H),
(IB2 ⊗ m)ϕ2(b2)(IB2 ⊗ n) = (idB2 ⊗ µ)ϕ2(b2) ⊗ x) (u ∈ D2 ⊗ B0(H).
Formula (4.16) shows now that
(ı ⊗ idH )(cid:0)(IB1 ⊗ m)ϕ1(b1)(IB1 ⊗ n)(cid:1) = (IB2 ⊗ m)ϕ2(b2)(IB2 ⊗ n).
LANDSTAD-VAES THEORY
17
Any compact operator is a norm limit of sums of one-dimensional operators. Therefore the above
formula holds for all m, n ∈ B0(H). ı is an isomorphism of C∗-algebras. Therefore
Replacing m and n by approximate units of B0(H) and passing to the limit we get
(cid:13)(cid:13)(IB1 ⊗ m)ϕ1(b1)(IB1 ⊗ n)(cid:13)(cid:13) =(cid:13)(cid:13)(IB2 ⊗ m)ϕ2(b2)(IB2 ⊗ n)(cid:13)(cid:13) .
(cid:13)(cid:13)ϕ1(b1)(cid:13)(cid:13) =(cid:13)(cid:13)ϕ2(b2)(cid:13)(cid:13) .
Remembering that ϕ1 and ϕ2 are injective morphisms we obtain (4.14).
(cid:3)
Now the proof of the uniqueness theorem is easy.
Proof of Theorem 2.7. At first we show that the isometry : B1 −→ B2 introduced by (4.12) is a
C∗-algebra isomorphism i.e.
(4.17)
for any b, b′ ∈ B1.
(b∗) = (b)∗,
(bb′) = (b)(b′)
∗
∗
∗
) = (d
∗
∗
)) = ı(d
ψ1(ba
((ψ1(ba)d)
Letba ∈ bA and d ∈ D1. Then using the second and the first formula of (4.12) we obtain
) =(cid:0)ψ2(ba)ı(d)(cid:1)∗
Remembering that ψ1(bA)D1 = B1 we get the first formula of (4.17).
Letba,ba′ ∈ bA and d, d′ ∈ D1. Then using (4.12) we obtain
Remembering that ψ1(bA)D1 = D1ψ1(bA) = B1 we get the second formula of (4.17).
((ψ1(ba)dd′ψ1(ba′)) = ψ2(ba)ı(dd′)ψ2(ba′) = ψ2(ba)ı(d)ı(d′)ψ2(ba′) = (ψ1(ba)d)(d′ψ1(ba′)).
= (ψ1(ba)d)
)ψ2(ba
Now we know that ∈ Mor(B1, B2). Extending to multiplier algebras we may rewrite first formula
∗
.
of (4.12) in the following way:
It shows that
(4.18)
(cid:0)ψ1(ba)(cid:1) (d) = ψ2(ba)ı(d).
(cid:0)ψ1(ba)(cid:1) = ψ2(ba),
(d) = ı(d)
triangle in diagram (2.6) is commutative. We shall prove that the square in (2.6) is also commutative.
for anyba ∈ bA and d ∈ D1. Second equality says that is an extension of ı, the first one shows that the
Letba ∈ bA and d ∈ D!. Then using (2.4) and the triangle in (2.6) we get
(id bA ⊗)β1(ψ1(ba)) = (id bA ⊗)(id bA ⊗ ψ1)b∆(ba) = (id bA ⊗ ψ2)b∆(ba) = β2(ψ2(ba)).
Similarly using Condition C1 of Definition 2.5 we get
(id bA ⊗)β1(d) = (id bA ⊗)(I bA ⊗ d) = I bA ⊗ı(d) = β2(ı(d)).
Combining the two formulas and using the first relation of (4.12) we obtain
(id bA ⊗)β1(ψ1(ba)d) = β2(ψ2(ba)ı(d)) = β2((ψ1(ba)d))
Remembering that ψ1(bA)D1 = B1 we conclude that
(id bA ⊗)β1(b) = β2((b))
for any b ∈ B1. It shows that (2.6) is a commutative diagram, so is an isomorphism of G-products
(B1, β1, ψ1) and (B2, β2, ψ2).
Conversely if is an isomorphism of G-products (B1, β1, ψ1) and (B2, β2, ψ2) extending ı then
coincides with B1 we see that is uniquely determined.
our considerations. Definitions of ϕ ∈ Mor(B, B0(H) ⊗ B) and of D1, D ⊂ M(B) contain explicitly σ
satisfies relations (4.18). The latter implies the first relation of (4.12). Remembering that ψ1(bA)D1
The Heisenberg pair (σ,bσ) chosen at the beginning of present section plays an important role in
andbσ. However D as unique Landstad algebra does not depend on the choice of (σ,bσ). Formula (4.8)
shows that also D1 remains the same when we change (σ,bσ).
(cid:3)
18
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
This section is devoted to the proof of Theorem 2.8. Let (D, α) be a weak G-dynamical system,
5. Crossed product construction
(5.1)
(5.2)
for any ω ∈ A∗ and d ∈ D.
Then B is a closed linear subset of B(K ⊗ H). We have to show that the above assumptions imply all
three statements of Theorem 2.8. Statement 1 is already established (see Proposition 3.2.1).
K, H be Hilbert spaces, (σ,bσ) be a canonical Heisenberg pair acting on H and (π,eπ) be a faithful
presentation of α on K. Theneπ ∈ Mor(D, B0(K) ⊗ A) is injective and
For anyba ∈ bA we set ψ(ba) = IK ⊗bσ(ba). Clearly ψ ∈ Rep(bA, K ⊗ H). Let
π(α(ω, d)) = (idK ⊗ ω)eπ(d)
B = ψ(bA)(idK ⊗ σ)eπ(D).
In this section (σ,bσ) is a canonical Heisenberg pair acting on a
(ρ,bσ) is an anti-Heisenberg pair. We shall deal with elements of M(bA ⊗ B0(K ⊗ H)). Among them we
Then β is a ∗-algebra homomorphism from B to M(bA ⊗ B0(K ⊗ H)). We shall prove that
have unitary V1ρ having the second leg (the one in K) trivial. We recall that B ⊂ B(K ⊗ H). For any
b ∈ B we set
Proof of statement 2 of Theorem 2.8.
Hilbert space H. Let ρ be a representation of A acting on H such that (ρ, σ) is a commuting pair and
β(b) = V1ρ(I bA ⊗b)V ∗
1ρ.
(5.3)
β(ψ(ba)) = (id bA ⊗ ψ)b∆(ba)
β((idK ⊗ σ)eπ(d)) = I bA ⊗ (idK ⊗ σ)eπ(d)
for anyba ∈ bA and d ∈ D.
Indeed, we know that (ρ,bσ) is an anti-Heisenberg pair. In this case second formula of (1.8) takes
the form (id bA ⊗bσ)b∆(ba) = V1ρ(I bA ⊗bσ(ba))V ∗
of M(bA ⊗ B0(K) ⊗ B0(H)) having trivial second leg we obtain first formula of (5.3). We know that
(ρ, σ) is a commuting pair. Therefore V1ρ commutes with I bA ⊗ (idK ⊗ σ)eπ(d) for all d ∈ D and second
1ρ. Reinterpreting this equation as the equality of elements
formula of (5.3) follows.
Combining the two formulae of (5.3) we get
(5.4)
β(ψ(ba)eπ(d)) = (id bA ⊗ ψ)b∆(ba)(I bA ⊗eπ(d))
for anyba ∈ bA and d ∈ D. We shall prove that β ∈ Mor(B, bA ⊗ B). Indeed
(bA ⊗ IB)β(B) = (bA ⊗ IB)(id bA ⊗ ψ)b∆(bA)(I bA ⊗eπ(D))
= (id bA ⊗ ψ)(cid:16)(bA ⊗ I bA)b∆(bA)(cid:17) (I bA ⊗eπ(D))
= (id bA ⊗ ψ)(bA ⊗ bA)(I bA ⊗eπ(D)) = bA ⊗ B.
(bA ⊗ IB)β(B) = bA ⊗ B.
So we have
(5.5)
It means that
First formula of (5.3) shows that (2.4) is a commutative diagram. Using commutativity of (2.4) and
β ∈ Mor(B, A ⊗ B).
Multiplying both sides (from the left) by I bA ⊗B we obtain (bA ⊗ B)β(B) = bA ⊗ B.
coassociativity of b∆ one can easily show that (id bA ⊗β)β ◦ ψ = (b∆ ⊗ idB)β ◦ ψ. Consequently
(id bA ⊗β)β(ψ(ba)) = (b∆ ⊗ idB)β(ψ(ba))
for anyba ∈ bA. Moreover for any d ∈ D we have:
(id⊗ β)β((idK ⊗ σ)bπ(d)) = (id⊗ β)(I bA ⊗ (idK ⊗ σ)bπ(d)) = I bA ⊗ I bA ⊗ (idK ⊗ σ)bπ(d)
= (b∆ ⊗ idB)(I bA ⊗ (idK ⊗ σ)bπ(d)) = (b∆ ⊗ idB)β((idK ⊗ σ)bπ(d)).
Combining two last formulae and remembering that B = ψ(bA)(idK ⊗ σ)bπ(D) we obtain that
(id bA ⊗β)β(b) = (b∆ ⊗ idB)β(b)
LANDSTAD-VAES THEORY
19
for any b ∈ B. It means that β is a left action of G on B. The action is continuous: (5.5) is the Podleś
condition for this action. Statement 2 is shown.
(cid:3)
Proof of statement 3 of Theorem 2.8. We already know that the diagram (2.4) is commutative. Hence
(B, β, ψ) is a G-product.
C3'. The C∗-algebra generated by
(5.6)
characterising Landstad algebra take the form
We have to show that (idK ⊗ σ)eπ(D) is a Landstad algebra for (B, β, ψ). Now the conditions
C1'. β((idK ⊗ σ)eπ(d)) = I bA ⊗ (idK ⊗ σ)eπ(d) for any d ∈ D,
C2'. B = ψ(bA)(idK ⊗ σ)eπ(D),
h(idB ⊗ ω)(cid:0)Vψ3(idK ⊗ σ)eπ(D)12V ∗
coincides with (idK ⊗ σ)eπ(D).
for two legs: ψ(ba) = IK ⊗bσ(ba) ∈ B(K ⊗ H). Clearly Vψ3 = Vbσ3.
shows that the set (5.6) coincides with (idK ⊗ σ)eπ(D1), where D1 is given by (2.2). Now Condition
C3' follows directly from Condition WA1 of Definition 2.2. Hence eπ(D) is the Landstad algebra of
C∗-algebra D and the Landstad algebra of (B, β, ψ) are obviously isomorphic: (idK ⊗ σ)eπ is the
isomorphism. Let ((idK ⊗ σ)eπ(D), α′) be the Landstad G-dynamical system of (B, β, ψ). Now formula
Adapting Condition C3 to the present context we had to replace Vψ2 by Vψ3, because ψ itself stays
Conditions C2' and C1' are already verified (cf. (5.2) and the second formula of (5.3)). Formula (3.1)
ψ3(cid:1) : ω ∈ A∗i
(B, β, ψ).
(2.8) takes the form
for any ω ∈ A∗ and d ∈ D. Remembering that Vψ3 = Vbσ3 and taking into account (3.1) we get
ψ3(cid:1)
α′(ω, (idK ⊗ σ)eπ(d)) = (idB ⊗ ω)(cid:0)Vψ3eπ(d)1σV ∗
α′(ω, (idK ⊗ σ)eπ(d)) = (idK ⊗ σ)eπ(α(ω, d)).
In other words (idK ⊗ σ)eπ is an isomorphism of weak G-dynamical systems. We see that the Landstad
G-dynamical system of (B, β, ψ) and the original weak G-dynamical system (D, α) are isomorphic.
The proofs of statement 3 and of Theorem 2.8 are complete.
(cid:3)
6. Weak action implemented by a unitary representation.
Let C be a C∗ algebra and U be a unitary element of M(C ⊗ A). We say that U is a unitary
representation of G if
(6.1)
(idC ⊗ ∆)U = U12U13.
Interesting examples of weak G-dynamical systems are provided by the following Theorem:
Theorem 6.1. Let C is a C∗-algebra, U ∈ M(C ⊗ A) be a unitary representation of G and
D1 =h(idC ⊗ ω)(cid:16)U (C ⊗ IA)U
∗(cid:17) : ω ∈ A∗i
Then D1 is a norm-closed subspace of M(C). Let D ⊂ M(C) be a C∗-subalgebra generated by D1. For
any ω ∈ A∗ and d ∈ D we set
(6.2)
α(ω, d) = (idC ⊗ ω) (U (d ⊗ IA)U ∗) .
Then α is a right weak action of G on D.
We say that α is a weak action implemented by U . The reader should notice that D need not
coincide with C. Corresponding G-product is described by the following Theorem:
Theorem 6.2. Let C is a C∗-algebra and U ∈ M(C ⊗ A) be a unitary representation of G. Then there
exists unique morphism ψ ∈ Mor(bA, bA ⊗ C) such that
(6.3)
(ψ ⊗ idA)V = U23V13,
20
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
It makes the diagram
(6.4)
b∆
bA
bA ⊗ bA
ψ
id bA ⊗ ψ
b∆⊗idC
bA ⊗ C
/ bA ⊗ bA ⊗ C
With the notation introduced in the above Theorems we have
commutative. Comparing this diagram with (2.4) we see that (bA ⊗ C,b∆ ⊗ idC , ψ) is a G-product.
Proposition 6.3. Let (D′, α′) be the Landstad G-dynamical system related to (bA ⊗ C,b∆ ⊗ idC, ψ).
Then
D′ = I bA ⊗ D
and
(6.5)
for any ω ∈ A∗ and d ∈ D.
α′(ω, I bA ⊗ d) = I bA ⊗ α(ω, d)
At first we shall prove Theorem 6.2 and Proposition 6.3. Then Theorem 6.1 will follow from an
obvious isomorphism connecting (D, α) and (D′, α′).
Proof of Theorem 6.2. We chose a Heisenberg pair acting on a Hilbert space H. Applying idC ⊗ σ⊗idA
to the both sides of (6.1) and using first formula of (1.8) we get U1σU13 =(cid:0) idC ⊗ (σ ⊗ idA)∆(cid:1) U =
bσ3. Therefore
Vbσ3U1σV
∗
∗
U
1σVbσ3U1σ = U13Vbσ3
∗
(6.6)
eψ(ba) = flip(cid:16)U
(eψ ⊗ idA)V = flip12(cid:16)U
For anyba ∈ bA we set
Then U1σ, IC ⊗bσ(ba) ∈ M(C ⊗ B0(H)), eψ(ba) ∈ M(B0(H) ⊗ C) and eψ ∈ Mor(bA, B0(H) ⊗ C). We have
The reader should notice that in the above computation U13Vbσ3 = (idC ⊗bσ ⊗ idA)U13V23, whereas
U23Vbσ3 = (bσ ⊗ idC ⊗ idA)U23V13. Clearly U23Vbσ3 ∈ M(B0(H) ⊗ C ⊗ A). More precisely U23Vbσ3 belongs
to M(bσ(bA) ⊗ C ⊗ A). Now Theorem 1.6.6 of [24] shows that eψ ∈ Mor(bA,bσ(bA) ⊗ C). We know thatbσ
is faithful. Therefore eψ is of the form eψ = (bσ ⊗ idC)ψ, where ψ ∈ Mor(bA, bA ⊗ C). With this notation
1σ(IC ⊗bσ(ba))U1σ(cid:17) .
1σVbσ3U1σ(cid:17) = flip12(cid:0)U13Vbσ3(cid:1) = U23Vbσ3.
(6.6) implies (6.3).
Taking into account (6.3) we obtain
∗
(cid:16)(b∆ ⊗ idC )ψ ⊗ idA(cid:17) V =(cid:16)b∆ ⊗ idC ⊗ idA(cid:17) U23V13 = U34V24V14
=(cid:0)id bA ⊗ ψ ⊗ idA(cid:1) V23V13 =(cid:16)(id bA ⊗ ψ)b∆ ⊗ idA(cid:17) V.
Now second formula of (1.7) shows that (6.4) is a commutative diagram. Using coassociativity of b∆
and the cancelation property one can easily show that b∆ ⊗ idC ∈ Mor(bA ⊗ C, bA⊗bA⊗ C) is a continuous
left action of bG on bA ⊗ C. Hence (bA ⊗ C,b∆ ⊗ idC , ψ) is a G-product.
Proof of Proposition 6.3. Let B = bA ⊗ C, β = b∆ ⊗ idC ∈ Mor(B, bA ⊗ B) and ψ ∈ Mor(bA, B) be the
morphism introduced in Theorem 6.2. We shall prove that
(cid:3)
(6.7)
Indeed
B = (I bA ⊗ C)ψ(bA).
ψ(bA) = ψ(cid:16)h(id bA ⊗ ω)V : ω ∈ A∗i(cid:17) =h(id bA⊗C ⊗ ω)U23V13 : ω ∈ A∗i
/
/
/
LANDSTAD-VAES THEORY
21
Therefore
(I bA ⊗C)ψ(bA) =h(id bA ⊗ idC ⊗ ω)(cid:16)(I bA ⊗C ⊗ IA)U23V13(cid:17) : ω ∈ A∗i
=h(id bA ⊗ idC ⊗ ω)(cid:16)(I bA ⊗C ⊗ A)U23V13(cid:17) : ω ∈ A∗i
=h(id bA ⊗ idC ⊗ ω)(cid:16)(I bA ⊗C ⊗ A)V13(cid:17) : ω ∈ A∗i
=h(id bA ⊗ ω)V : ω ∈ A∗i ⊗ C = bA ⊗ C
and (6.7) follows. In the above computation ω emits A to the right, next U is absorbed by C ⊗ A (this
is because U is a unitary element of M(C ⊗ A)) and finally ω absorbs A.
To determine the Landstad algebra of (bA ⊗ C,b∆ ⊗ idC, ψ) we shall use the procedure described in
Section 4. Let (σ,bσ) be a Heisenberg pair acting one Hilbert space H and ϕ ∈ Mor(B, B ⊗ B0(H)) be
the morphism introduced by (4.1). We have
β(I bA ⊗ C) = I bA ⊗ I bA ⊗ C,
β′(I bA ⊗ C) = I bA ⊗ C ⊗ I bA,
ϕ(I bA ⊗ C) = Vψσ(I bA ⊗ C ⊗ IH )V
∗
ψσ.
Formula (6.3) says that Vψ3 = U23V13. Hence Vψσ = U2σV1σ. Second leg of V1σ is trivial. Therefore
V1σ commutes with I bA ⊗ C ⊗ IH and
ϕ(I bA ⊗ C) = U2σ(I bA ⊗ C ⊗ IH )U
∗
2σ
= I bA ⊗ (idC ⊗ σ)(cid:16)U (C ⊗ IA)U
∗(cid:17) .
∗(cid:17)(cid:16)IC ⊗bσ(bA)(cid:17) .
Formula (4.2) shows that ϕ(ψ(bA)) = IB ⊗bσ(bA). Taking into account (6.7) we obtain
ϕ(B) = I bA ⊗ (idC ⊗ σ)(cid:16)U (C ⊗ IA)U
In this section D1 and D denote operator spaces introduced in Theorem 6.1. To avoid conflict of
notation with the one used earlier we decorate D1 in (4.5) with prime:
D′
1 =h(idB ⊗ µ)ϕ(B) : µ ∈ B(H)∗i
= I bA ⊗h(idC ⊗ µ)(cid:16)(idC ⊗ σ)(cid:16)U (C ⊗ IA)U
Absorbingbσ(bA) by µ and replacing µ ◦ σ by ω we obtain
1 = I bA ⊗h(idC ⊗ ω)(cid:16)U (C ⊗ IA)U
By Proposition 4.3, the Landstad algebra of (bA ⊗ bA,b∆ ⊗ idC , ψ) coincides with I bA ⊗D. According to
∗(cid:17)(cid:16)IC ⊗bσ(bA)(cid:17)(cid:17) : µ ∈ B(H)∗i .
∗(cid:17) : ω ∈ A∗i = I bA ⊗D1.
(6.3) Vψ3 = U23V13. Now, formula (2.8) takes the form
D′
α′(ω, I bA ⊗ d) = (id bA ⊗ idC ⊗ ω)(cid:0)U23V13(I bA ⊗ d ⊗ IA)V ∗
= I bA ⊗ (idC ⊗ ω) (U (d ⊗ IA)U ∗) = I bA ⊗ α(ω, d)
23(cid:1)
13U ∗
for any ω ∈ A∗ and d ∈ D.
(cid:3)
Proof of Theorem 6.1.
isomorphism of C∗-algebras. Formula (6.5) takes the form
Let ı(d) = I bA ⊗ d any for d ∈ D. Then ı : D −→ I bA ⊗ D = D′ is an
α′(ω, ı(d)) = ı(α(ω, d)).
It shows that (D′, α′) and (D, α) are isomorphic. By Proposition (6.3), (D′, α′) is a weak G-dynamical
system. So is (D, α).
(cid:3)
We say that U is regular if (IC ⊗A)U (C ⊗ IA) = C ⊗ A. If U is regular then D = D1 = C. In
general D 6= C. To obtain the most obvious example of the above construction we set: C = bA and
22
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
U = V . Then ψ = b∆ and diagram (6.4) takes the form
b∆
b∆
bA
bA ⊗ bA
b∆⊗id bA
bA ⊗ bA
/ bA ⊗ bA ⊗ bA
id bA ⊗ b∆
(6.8)
we have
D1 =h(id bA ⊗ ω)(cid:16)V (bA ⊗ IH )V ∗(cid:17) : ω ∈ B(H)∗i
This is the commutative diagram stating the coassociativity of b∆. This way for any locally compact
quantum group G we have canonically associated G-product (bA ⊗ bA,b∆ ⊗ id bA,b∆). Now
and D is the C∗-algebra generated by D1. D1 and Dare subsets of M(bA). For any ω ∈ A∗ and d ∈ D
(bA ⊗ bA,b∆ ⊗ id bA,b∆). If G is regular then D = bA. Otherwise D′ 6= bA.
The first known example of non-regular locally compact quantum group was quantum deformation
Eq(2) of the group of motions of Euclidean plane (with real deformation parameter 0 < q < 1, see
[22, 21] and [1]). Let5 A = C0(Eq(2)) and ∆ ∈ Mor(A, A ⊗ A) be the corresponding comultiplication:
Eq(2) = (A, ∆). Then (A ⊗ A, ∆ ⊗ idA, ∆) is a \Eq(2)-product. Let D′ = IA ⊗D be its Landstad
algebra. Condition C2 says that
α is a right weak action of G on D and (D, α) is a G-dynamical system corresponding to the G-product
α(ω, d) = (id bA ⊗ ω) (V (d ⊗ IH )V ∗) .
A boring calculations entering deeply into anatomy of Eq(2) shows that D is unital. Therefore ∆(A) ⊂
A ⊗ A. For the first time this unexpected result appeared in [23].
∆(A)(IA ⊗D) = A ⊗ A.
7. Kasprzak approach to Rieffel deformation.
The following example comes from the Kasprzak theory [6, 7, 8]. He was able to find an elegant
realisation of Rieffel deformation [14] of C∗-algebras endowed with an action of a group. Kasprzak
(and Rieffel) worked with locally compact abelian group, but due to the further developement we may
consider any locally compact quantum group G. We shall use the notation introduced in previous
sections.
(7.1)
Assume that we have a unitary two-cocycle. This is a unitary element Ω ∈ M(bA ⊗ bA) such that
(Ω ⊗ I bA)(b∆ ⊗ id bA)(Ω) = (I bA ⊗Ω)(id bA ⊗b∆)(Ω).
We shall also assume that the Drinfeld twist induced by Ω is trivial:
(7.2)
for anyba ∈ bA.
Ω∗b∆(ba)Ω = b∆(ba)
Let D be a C∗-algebra equipped with a right weak action α of G. Then (D, α) be a G-dynamical sys-
tem. Using Theorem 2.8 we may find G-product (B, β, ψ) with Landstad dynamical system isomorphic
to (D, α).
Theorem 7.1. For any b ∈ B we set
βΩ(b) = Ω∗
1ψβ(b)Ω1ψ.
Then βΩ(b) ∈ M(bA ⊗ B), βΩ ∈ Mor(B, bA ⊗ B) is a continuous left action of bG on B and (B, βΩ, ψ)
is a G-product.
Proof. We have to show that
(7.3)
(7.4)
(7.5)
(id bA ⊗βΩ)βΩ(b) = (b∆ ⊗ idB)βΩ(b),
(bA ⊗ IB)βΩ(B) = bA ⊗ B,
βΩ(ψ(ba)) = (id bA ⊗ ψ)b∆(ba).
5notice change of notation: the role of G and bG are interchanged.
/
/
/
Relation (7.5) is easy to verify:
LANDSTAD-VAES THEORY
23
βΩ(ψ(ba)) = Ω∗
1ψ(id bA ⊗ ψ)b∆(ba))Ω1ψ
1ψβ(ψ(ba))Ω1ψ = Ω∗
= (id bA ⊗ ψ)(cid:16)Ω∗b∆(ba))Ω(cid:17) = (id bA ⊗ ψ)b∆(ba).
We already know that Ω1,βΩψ = (id bA ⊗βΩψ)Ω = (id bA ⊗ (id bA ⊗ ψ)b∆)Ω = Ω1,(id bA ⊗ ψ) b∆. We compute
1,βΩψ(id bA ⊗βΩ)β(b)Ω1,βΩψ = Ω∗
2ψ(id bA ⊗β)β(b)Ω2ψΩ1,βΩψ.
(id bA ⊗βΩ)βΩ(b) = Ω∗
1,βΩψΩ∗
On the other hand
(b∆ ⊗ idB)βΩ(b) = Ω∗
b∆,ψ
(b∆ ⊗ idB)β(b)Ω b∆,ψ.
b∆,ψ
To prove (7.3) it is enough to show that Ω2ψΩ1,βΩψΩ∗
last moment (7.1) we obtain
Ω2ψΩ1,βΩψΩ∗
= Ω2ψΩ1,(id bA ⊗ ψ) b∆Ω∗
b∆,ψ
b∆,ψ
commutes with (b∆ ⊗ idB)β(b). Using at the
= (id bA ⊗ id bA ⊗ ψ)(cid:16)(I bA ⊗Ω)(id bA ⊗b∆)Ω(b∆ ⊗ id bA)Ω∗(cid:17)
= Ω ⊗ IB .
Now the commutativity follows immediately form (7.2).
Applying to the both sides id bA ⊗ ψ and taking into account commutativity of (2.4) we get
We shall show (7.4). Ω is a unitary multiplier of bA ⊗ bA. Therefore (bA ⊗ bA)Ω∗ = bA ⊗ bA. Using the
cancelation formula bA ⊗ bA = (bA ⊗ I bA)b∆(bA) and relation (7.2) we obtain
(bA ⊗ I bA)Ω∗b∆(bA) = (bA ⊗ I bA)b∆(bA).
1ψβ(ψ(bA)) = (bA ⊗ IB)β(ψ(bA)).
(bA ⊗ IB)Ω∗
(bA ⊗ IB)Ω∗
1ψβ(B) = (bA ⊗ IB)β(B) = bA ⊗ B.
(bA ⊗ IB)βΩ(B) = (bA ⊗ B)Ω1ψ = bA ⊗ B.
We know that ψ ∈ Mor(bA, B). Therefore ψ(bA)B = B. Multiplying both sides of the above formula by
β(B) and using Podleś condition for the action β we have
Let D′ be the Landstad algebra related to G-product (B, βΩ, ψ). According to Kasprzak, D′ may
be considered as Rieffel deformation of D. Recently Kasprzak theory was extended by Neshveyev and
Tuset [12] for non-trivial Drinfeld twist i.e. when (7.2) does not hold. Then they had to consider
deformations of G and ψ.
Finally
(cid:3)
8. A remark on Landstad conditions
This section is not in the main stream of the paper.
Kasprzak that was used to simplify second Landstad condition (cf.
this section we assume that A admits a continuous counit. (i.e: G is coameanable).
It contains a generalisation of a result of
formulae (3) and (4) of [6]). In
Theorem 8.1. Let (B, β, ψ) be a G-product and d ∈ M(B). Assume that
(8.1)
Vψ2(d ⊗ IA)V ∗
ψ2(IB ⊗a) ∈ M(B) ⊗ A
for any a ∈ A. Then the following conditions are equivalent:
(1) ψ(ba)d ∈ B for anyba ∈ bA,
(2) dψ(ba′) ∈ B for anyba′ ∈ bA,
(3) ψ(ba)dψ(ba′) ∈ B for anyba,ba′ ∈ bA.
Proposition 8.2. There exists a bounded net (beλ)λ∈Λ of elements of bA converging strictly to I bA such
We shall use the following result:
that
(8.2)
for any d ∈ M(B) satisfying relation (8.1). Square bracket in the above formula denotes commutator.
lim
λ∈Λ(cid:13)(cid:13)(cid:2)d, ψ(beλ)(cid:3)(cid:13)(cid:13) = 0
24
SUTANU ROY AND STANISŁAW LECH WORONOWICZ
Proof. Let e be a counit of A and (ωλ)λ∈Λ be a net of normal states on A weakly converging to e:
for any x ∈ A. Then for any r ∈ M(B) ⊗ A we have
(8.3)
norm-lim
λ∈Λ
(idB ⊗ ωλ)r = (idB ⊗ e)r.
lim
λ∈Λ
ωλ(x) = e(x)
Indeed (8.3) is obvious for r ∈ M(B) ⊗alg A. Moreover the set of all r ∈ M(B) ⊗ A satisfying (8.3) is
closed in norm topology. Remembering that M(B) ⊗ A is a norm closure of M(B) ⊗alg A we obtain
(8.3) in full generality.
We fix an element a ∈ A such that e(a) = 1 and set
(8.4)
beλ = (idB ⊗ ωλ)(cid:0)V ∗(I bA ⊗ a)(cid:1) .
Thenbeλ ∈ bA for any λ ∈ Λ. With an easy calculation we obtain
(cid:2)d, ψ(beλ)(cid:3) = (idB ⊗ ωλ)X,
ψ2hVψ2(d ⊗ IA)V ∗
ψ2(IB ⊗ a) − V ∗
X = (d ⊗ IA)V ∗
where
= V ∗
ψ2(IB ⊗a) − d ⊗ ai .
ψ2(d ⊗ a)
Assume that d ∈ M(B) satisfies (8.1). Then the expression in square bracket belongs to M(B) ⊗
A. Therefore X ∗X ∈ M(B) ⊗ A.
It is known that (idB ⊗ e)V = I bA. Therefore (e is a character)
(idB ⊗ e)X = d − d = 0 and (idB ⊗ e)(X ∗X) = 0. Formula (8.3) shows now that
Clearly idB ⊗ ωλ is a completely positive unital mapping. Using Kadison inequality (cf [17, Corollary
1.3.2, page 9]) we obtain
k(idB ⊗ ωλ)(X ∗X)k = 0.
lim
λ∈Λ
(cid:13)(cid:13)(cid:2)d, ψ(beλ)(cid:3)(cid:13)(cid:13)2
= k(idB ⊗ ωλ)(X ∗)(idB ⊗ ωλ)(X)k
≤ k(idB ⊗ ωλ)(X ∗X)k .
Formula (8.2) is shown. To end the proof we have to show that the net (beλ)λ∈Λ converges strictly to
I bA ∈ M(bA). To this end we have to choose the net (ωλ)λ∈Λ in a more specific way.
λ)λ∈Λ be a net of normal states on A weakly converging to e and c be an element of A such
Let (ω′
that e(c) = 1. For any a ∈ A and λ ∈ Λ we set
ωλ(a) =
λ(c∗ac)
ω′
ω′
λ(c∗c)
.
Then (ωλ)λ∈Λ is a net of normal states on A weakly converging to e. Now (8.4) takes the form
and
(idB ⊗ ω′
(idB ⊗ ω′
(idB ⊗ ω′
ω′
λ(c∗c)
λ)(cid:0)(I bA ⊗c∗)V ∗(I bA ⊗ca)(cid:1)
λ)(cid:0)(ba ⊗ c∗)V ∗(I bA ⊗ca)(cid:1)
λ)(cid:0)(I bA ⊗c∗)V ∗(ba ⊗ ca)(cid:1)
λ(c∗c)
λ(c∗c)
ω′
ω′
,
beλ =
babeλ =
beλba =
for anyba ∈ bA. The reader should notice that (ba ⊗ c∗)V ∗(I bA ⊗ca) and (I bA ⊗c∗)V ∗(ba ⊗ ca) belong to
B ⊗ A. Formula (8.3) shows now thatbabeλ andbeλba converge in norm toba. It means thatbeλ converges
strictly to I bA ∈ M(bA).
Proof of Theorem 8.1. We know that ψ(ba), ψ(ba′) ∈ M(B). Therefore (3) follows from (1) and from
(2). We shall prove that (1) follows from (3). One can easily verify that
(cid:3)
ψ(ba)d − ψ(ba)dψ(beλ) = ψ(ba −babeλ)d − ψ(ba)(cid:2)d, ψ(beλ)(cid:3) .
Proposition 8.2 shows now that
Remembering that B is norm-closed in M(B) we see that (3) implies (1). In the similar way one shows
that (3) implies (2)
(cid:3)
ψ(ba)d = norm-lim
λ∈Λ
ψ(ba)dψ(beλ).
LANDSTAD-VAES THEORY
25
Acknowlegdement
The research started in Oberwolfach within the program "research in pairs" in August 2014 and
ended in Warsaw in last week of June 2017. The authors are very grateful to the Oberwolfach Research
Institute for Mathematics and to Banach Center in Warsaw for creating the perfect conditions for
fruitful work.
References
[1] S. Baaj. Represéntation regulière du groupe quantique Eµ(2) de Woronowicz. Comptes rendus de l'Académie des
sciences. Série 1, Mathématique, 314(13): 1021 -- 1026, 1992.
[2] S. Baaj and G. Skandalis. Unitaries multiplicatifs et dualité pour les produits croisé de C ∗-algèbres. Annales Sci-
entifiques de l'Ecole Normale Supérieure, 26(4): 425 -- 488, 1993.
[3] S. Baaj, G. Skandalis, and S. Vaes. Non-semi-regular quantum groups coming from number theory. Communication
in Mathematical Physics, 235(1): 139 -- 167, 2003.
[4] Robert S. Doran and Josef Wichmann. Approximate identities and factorization in Banach modules, volume 768 of
Lecture Notes in Mathematics. Springer-Verlag, Berlin-New York, 1979.
[5] Edward G. Effros and Zhong-Jin Ruan. Operator spaces, volume 23 of London Mathematical Society Monographs.
New Series. The Clarendon Press, Oxford University Press, New York, 2000.
[6] Paweł Kasprzak. Rieffel deformation via crossed products. Journal of Functional Analysis, 257(5): 1288 -- 1332, 2009.
[7] Paweł Kasprzak. Rieffel deformation of group coactions. Communications in Mathematical Physics, 300(3):741 -- 763,
2010.
[8] Paweł Kasprzak. Rieffel deformation of homogeneous spaces. Journal of Functional Analysis, 260(1): 146 -- 163, 2011.
[9] J. Kustermans and S. Vaes. Locally compact quantum groups. Annales Scientifiques de l'Ecole Normale Supérieure,
33(6): 837 -- 934, 2000.
[10] Magnus B. Landstad. Daulity theory for covariant systems. Trans. AMS, 248(2): 223 -- 267, 1979.
[11] Tetsuya Masuda, Yoshiomi Nakagami, and S.L. Woronowicz. A C∗-algebraic framework for quantum groups. Inter-
nat. J. Math., 14(9):903 -- 1001, 2003.
[12] Sergey Neshveyev and Lars Tuset. Deformation of C∗-algebras by cocycles on locally compact quantum groups.
Advances in Mathematics, 254:454 -- 496, March 2014.
[13] Gert K. Pedersen. C∗-algebras and their Automorphism Groups. Academic Press, London, New York, San Francisco,
1979.
[14] Rieffel, Marc A. Deformation quantization for actions of Rd. Memoirs of the American Mathematical Society,
106(506):x+93, 1993.
[15] Piotr M. Sołtan and S.L. Woronowicz. A remark on manageable multiplicative unitaries. Lett. Math. Phys.,
57(3):239 -- 252, 2001.
[16] Piotr M. Sołtan and S.L. Woronowicz. From multiplicative unitaries to quantum groups. II. J. Funct. Anal.,
252(1):42 -- 67, 2007.
[17] Erling Størmer. Positive linear maps of operator algebras. Springer Monographs in Mathematics. Springer, Heidel-
berg, 2013.
[18] S. Vaes. A new approach to induction and imprimitivity results. Journal of Functional Analysis, 229: 317 -- 374, 2005.
[19] J. M. Vallin. C∗-algèbre de Hopf et C∗-algèbre de Kac. Proceedings of London Mathematical Society, 50(3): 131 -- 174,
1985.
[20] S.L. Woronowicz. Pseudospaces, pseudogroups and Pontriagin duality. In Mathematical problems in theoretical
physics (Proc. Internat. Conf. Math. Phys., Lausanne, 1979), volume 116 of Lecture Notes in Phys., pages 407 --
412. Springer, Berlin-New York, 1980.
[21] S.L. Woronowicz. Quantum E(2) group and its Pontryagin dual. Lett. Math. Phys., 23(4):251 -- 263, 1991.
[22] S.L. Woronowicz. Unbounded elements affiliated with C∗-algebras and noncompact quantum groups. Comm. Math.
Phys., 136(2):399 -- 432, 1991.
[23] S.L. Woronowicz. Quantum SU(2) and E(2) groups. Contraction procedure. Comm. Math. Phys., 149(3):637 -- 652,
1992.
[24] S.L. Woronowicz. From multiplicative unitaries to quantum groups. Internat. J. Math., 7(1):127 -- 149, 1996.
E-mail address: [email protected]
School of Mathematical Sciences,, National Institute of Science Education and Research, Bhubanes-
war, HBNI, Jatni, India
E-mail address: [email protected]
Institute of Mathematics of the Polish Academy of Sciences, ul. Śniadeckich 8, 00-656 Warszawa,
Poland and Department of Mathematical Methods in Physics, Faculty of Physics, University of Warsaw,
ul. Pasteura 5, 02-093 Warszawa Poland.
|
1908.00937 | 2 | 1908 | 2019-11-04T17:12:11 | Decomposing nuclear maps | [
"math.OA"
] | We show that the strengthened version of the completely positive approximation property of Brown, Carri\'on, and White---where the downward maps are asymptotically order zero and the upward maps are convex combinations of order zero maps---is enjoyed by every nuclear order zero map. | math.OA | math |
DECOMPOSING NUCLEAR MAPS
JOS´E R. CARRI ´ON AND CHRISTOPHER SCHAFHAUSER
Abstract. We show that the strengthened version of the completely
positive approximation property of Brown, Carri´on, and White -- where
the downward maps are asymptotically order zero and the upward maps
are convex combinations of order zero maps -- is enjoyed by every nuclear
order zero map.
Choi and Effros [9] and Kirchberg [16] proved that nuclearity for C ∗-
algebras is equivalent to the completely positive approximation property: a
C ∗-algebra A is nuclear if and only if there exist nets of finite dimensional
C ∗-algebras Fi and of contractive completely positive (c.c.p.) maps
(1)
A
ψi−→ Fi
φi−→ A
the composition of which tends to the identity map on A in the point-norm
topology, in the sense that
(2)
k(φiψi)(a) − ak → 0
for all a ∈ A.
See [6] for the general theory of nuclearity for C ∗-algebras.
This approximation property has been strengthened by several authors
in various cases [2, 24, 15, 5]. Notably, in [5], building on techniques from
[15], it was shown that all nuclear C ∗-algebras enjoy completely positive
approximations as in (2), where in addition all of the maps φi are convex
combinations of order zero maps and the maps ψi are asymptotically order
zero, in the sense that kψi(a)ψi(b)k → 0 whenever a and b are positive
elements of A with ab = 0.
In this short note, we show that this improved approximation property
holds for any c.c.p. nuclear order zero map between two C ∗-algebras. Recall
that a c.c.p. map θ : A → B has order zero if θ(a)θ(b) = 0 whenever a and
b are positive elements in A with ab = 0; see [23].
Theorem 1. Let A and B be C ∗-algebras. If θ : A → B is a c.c.p. nuclear
order zero map, then there are nets of finite dimensional C ∗-algebras (Fi)
and of c.c.p. maps A
φi−→ B satisfying
(i) k(φiψi)(a) − θ(a)k → 0 for all a ∈ A,
(ii) kψi(a)ψi(b)k → 0 for all a, b ∈ A such that ab = 0, and
(iii) every φi is a convex combination of c.c.p. order zero maps.
ψi−→ Fi
Date: November 5, 2019.
2010 Mathematics Subject Classification. Primary: 46L05.
1
2
J. CARRI ´ON AND C. SCHAFHAUSER
The refined versions of the completely positive approximation property
originate in a search for a theory of non-commutative covering dimension
of C ∗-algebras in the work of Kirchberg and Winter on decomposition rank
[18] and of Winter and Zacharias on nuclear dimension [23]. In both [18]
and [23] one finds conditions asking for an approximate factorization of the
identity map similar to that in Theorem 1, but with more control on the
upward maps φi.
The refined approximation property given in [5] -- which is Theorem 1 in
the special case when θ is the identity map on a nuclear C ∗-algebra -- is a cru-
cial part of [8], where it is shown that unital, separable, simple, nuclear, Z-
stable C ∗-algebras have nuclear dimension at most 1. This latter result plays
an important role in the classification theory for nuclear C ∗-algebras through
the results of [22]. The improved approximation properties mentioned above
have also seen application to near-inclusions of C ∗-algebras [15]. Given the
growing interest in the structure and classification of ∗-homomorphisms (as
opposed to C ∗-algebras) in papers such as [10, 13, 20, 8, 7, 3], we hope that
the results presented here will prove just as useful as their counterparts for
C ∗-algebras have.
The outline of the proof of Theorem 1 is not altogether new: the same
broad steps used in [15] and [5] apply. We point out, however, that the
proofs here, even once restricted to the case of the identity map on a nu-
clear C ∗-algebra, are new. For instance, we avoid the use of amenable and
quasidiagonal traces, which are critical in [5].
In order to prove Theorem 1, we will prove the following von Neumann
algebraic version first. Theorem 1 will then follow from a Hahn-Banach
argument similar to that used in [5] and [15].
Theorem 2. Let A be a C ∗-algebra and let N be a von Neumann algebra. If
θ : A → N is a weakly nuclear ∗-homomorphism, then there are nets of finite
φi−→ N satisfying
dimensional C ∗-algebras (Fi) and of c.c.p. maps A
(i) (φiψi)(a) → θ(a) in the σ-strong∗ topology (and therefore also in
ψi−→ Fi
the σ-weak topology) for all a ∈ A,
(ii) kψi(a)ψi(b)k → 0 for all a, b ∈ A such that ab = 0, and
(iii) every φi is a ∗-homomorphism.
In fact, the approximate factorizations satisfying conditions (i) and (ii) of
Theorem 2 can be arranged for all c.c.p. weakly nuclear maps.
Proposition 3. Let A be a C ∗-algebra and let N be a von Neumann algebra.
If θ : A → N is a c.c.p. weakly nuclear map, then there are nets of finite
φi−→ N satisfying
dimensional C ∗-algebras (Fi) and of c.c.p. maps A
ψi−→ Fi
(i) (φiψi)(a) → θ(a) in the σ-strong∗ topology for all a ∈ A, and
(ii) kψi(a)ψi(b)k → 0 for all a, b ∈ A such that ab = 0.
If A is quasidiagonal, we may further arrange for
DECOMPOSING NUCLEAR MAPS
3
(ii′) kψi(a)ψi(b) − ψi(ab)k → 0 for all a, b ∈ A.
Remark 4. If N is a von Neumann algebra and ρ is a normal state on N ,
then the seminorm k · k♯
(3)
kxk♯
2,ρ on N is defined by
2,ρ = ρ(cid:16) xx∗ + x∗x
(cid:17)1/2
2
,
x ∈ N.
If {ρi} is a separating family of normal states on N , then the topology
generated by {k · k♯
2,ρi} agrees with the σ-strong∗ topology on any bounded
subset of N ; see [1, III.2.2.19].
Proof of Proposition 3. First we assume A is quasidiagonal and prove the
stronger approximation condition. We may assume A is unital. Let F ⊆
A \ {0} and S ⊆ S(N ) ∩ N∗ be finite sets, and let ε > 0. Because θ is weakly
nuclear, there exist an integer ℓ ≥ 1 and u.c.p. maps A
satisfying
(4)
kφ′(ψ′(x)) − θ(x)k♯
2,ρ < kxkε
ψ′
−→ Mℓ(C)
φ′
−→ N
for all x ∈ F and ρ ∈ S. We will prove the existence of an integer k and of
u.c.p. maps A
ψ
−→ Mk(C)
(a) kφ(ψ(x)) − θ(x)k♯
(b) kψ(x)ψ(y) − ψ(xy)k < kxkkykε for all x, y ∈ F.
φ
−→ N satisfying
2,ρ < 5kxkε for all x ∈ F and ρ ∈ S, and
Replacing A with C ∗(F ∪ {1A}) ⊆ A, we may assume A is separable. Fix
a separable Hilbert space H and a faithful representation A ⊆ B(H) with
A ∩ K(H) = 0. After identifying Mℓ(C) with B(Cℓ), Voiculescu's Theorem1
provides an isometry v : Cℓ → H such that kv∗xv − ψ′(x)k < kxkε for all
x ∈ F.
Since A is quasidiagonal, there is a finite rank projection p ∈ B(H) such
that kpx − xpk < εkxk for all x ∈ F and kpv − vk < ε. Identify B(pH) with
Mk(C) for some integer k ≥ 1 and define ψ : A → Mk(C) by ψ(a) = pap.
Then
(5)
and
(6)
kv∗ψ(x)v − ψ′(x)k < 3kxkε
kψ(x)ψ(y) − ψ(xy)k < kxkkykε
for all x, y ∈ F. Note that (4) and (5) imply
kφ′(v∗ψ(x)v) − θ(x)k♯
(7)
2,ρ < 5kxkε
for all x ∈ F and ρ ∈ S. To complete the proof (in the quasidiagonal case),
define φ = φ′(v∗(·)v).
Now we handle the case of a general C ∗-algebra A. Let π : C0(0, 1]⊗A → A
be the ∗-homomorphism given by π(f ⊗ a) = f (1)a for all f ∈ C0(0, 1],
a ∈ A. By the homotopy invariance of quasidiagonality [21], C0(0, 1] ⊗ A
1See [11, Theorem II.5.3] for the version we are using here.
4
J. CARRI ´ON AND C. SCHAFHAUSER
is quasidiagonal. Applying the first part of the proof to the c.c.p. weakly
φi−→ N satisfying (i) and
i(id(0,1] ⊗a). (cid:3)
nuclear map θπ, there is a net C0(0, 1] ⊗ A
(ii′). The result follows by defining ψi : A → Fi by ψi(a) = ψ′
ψ′
i−→ Fi
Remark 5. If A and B are C ∗-algebras and θ : A → B is a c.c.p. nuclear
map, then a statement analogous to Proposition 3 holds with the approx-
imations in (i) taken in the operator norm. The proof is the same except
with operator norm estimates in (4), (7), and (a).
Proof of Theorem 2. Through a standard argument, we may assume A is
separable and N has separable predual. Note that N decomposes as a direct
sum N = N1 ⊕ N∞ with N1 finite and N∞ properly infinite. Then the map θ
decomposes accordingly as θ = θ1 ⊕θ∞ for weakly nuclear ∗-homomorphisms
θi : A → Ni, i ∈ {1, ∞}. By handling each summand separately, we may
assume that either N is finite or N is properly infinite.
Properly Infinite Case: We may assume A is unital. Fix a faithful
normal state ρ on N , a finite set F ⊆ A of unitaries, and a finite set G ⊆ A×A
such that xy = 0 for all (x, y) ∈ G, and let ε > 0. By Proposition 3, there are
a finite dimensional C ∗-algebra F and c.c.p. maps A
ψ
−→ F
φ′
−→ N satisfying
(a) kφ′(ψ(x)) − θ(x)k♯
(b) kψ(x)ψ(y)k < ε for all (x, y) ∈ G.
2,ρ < ε for all x ∈ F, and
At this point we can follow the proof of Lemma 2.4 of [5], which borrows
heavily from Theorem 2.2 of [14]. Indeed the last two paragraphs of the
proof of Lemma 2.4 of [5], with θ and φ′ in place of π∞ and θ, produce a
∗-homomorphism φ : F → M such that
(8)
for all x ∈ F.
k(φψ)(x) − θ(x)k♯
2,ρ < 2ε1/2
Finite Case: Let τN be a faithful normal trace on N so that on bounded
subsets of N , the σ-strong∗ topology is induced by the norm
(9)
kxk2,τN = τN (x∗x)1/2,
x ∈ N.
Let τA = τN θ, and note that θ induces a faithful normal ∗-homomorphism
¯θ : πτA(A)′′ → N . As N is finite, there is normal expectation N → ¯θ(πτA(A)′′)
(see Lemma 1.5.11 in [6]). Therefore, the corestriction of πτA to πτA(A)′′ is
weakly nuclear. It follows that πτA(A)′′ is hyperfinite by the equivalence of
(5) and (6) in Theorem 3.2.2 of [4]. After replacing N with πτA(A)′′ and θ
with πτA, we may assume that N is hyperfinite, which we do for the rest of
the proof.
Let Fi be an increasing sequence of finite dimensional C ∗-subalgebras of
N with σ-strong∗ dense union and let φi : Fi → N denote the inclusion
maps. Let τi = τN Fi be the induced trace on Fi. Fix a trace-preserving
expectation Ei : N → Fi and set ψ′
i) induces a
i = Eiθ. The sequence (ψ′
DECOMPOSING NUCLEAR MAPS
5
∗-homomorphism to the tracial von Neumann algebra ultraproduct
ω
(10)
Y(Fi, τi)
and a c.c.p. map to the C ∗-algebra ultraproduct
ψω : A → F ω =
(11)
ψω : A → Fω = Y
ω
Fi.
Let q : Fω → F ω be the quotient map and note that qψω = ψω. Let J = ker q.
Proposition 4.6 of [17] proves that the uniform 2-norm null sequences in
a norm ultrapower form a σ-ideal (Definition 4.4 of [17]), by a standard
application of Kirchberg's ǫ-test. A very slight modification of the proof
shows that J is a σ-ideal of Fω. Then there is a positive contraction e ∈
J ∩ ψω(A)′ such that ex = x for all x ∈ J ∩ C ∗(ψω(A)).
Consider the c.c.p. map Ψ : A → Fω defined by Ψ(x) = (1 − e)ψω(x),
x ∈ A. We claim Ψ is order zero. Indeed, if x, y ∈ A+ satisfy xy = 0,
then q(ψω(x)ψω(y)) = ψω(xy) = 0, so ψω(x)ψω(y) ∈ J. Then Ψ(x)Ψ(y) =
(1 − e)2ψω(x)ψω(y) = (1 − e)(cid:0)ψω(x)ψω(y) − ψω(x)ψω(y)(cid:1) = 0, as required.
Represent e by a sequence of positive contractions (ei) with ei ∈ Fi and
define
(12)
Then the maps A
ψi = (1 − ei)1/2ψ′
ψi−→ Fi
i(·)(1 − ei)1/2 : A → Fi.
φi−→ N have the desired properties.
(cid:3)
Finally, we prove Theorem 1.
Proof of Theorem 1. The structure theorem for order zero maps (see Corol-
lary 4.1 of [23]) implies there is a ∗-homomorphism ´θ : C0(0, 1]⊗ A → B such
that θ(a) = ´θ(id(0,1] ⊗ a) for a ∈ A. By Theorem 2.10 in [12], ´θ is nuclear
(since θ is nuclear). Hence, after replacing A with C0(0, 1] ⊗ A and θ with
´θ, we may assume θ is a ∗-homomorphism.
Let F be a finite subset of A and let ε > 0. Let K be the subset of
B(A, B) consisting of all c.c.p. maps η : A → B for which there exist a finite
dimensional C ∗-algebra F and c.c.p. maps A
ψ
−→ F
φ
−→ B satisfying:
• η = φψ,
• kψ(a)ψ(b)k < ε for all a, b ∈ F with ab = 0, and
• φ is a convex combination of c.c.p. order zero maps.
To prove the theorem, it suffices to show that θ is in the point-norm closure
of K.
ψi−→ Fi
By Theorem 2, there are nets of finite dimensional C ∗-algebras (Fi) and
φi−→ B ∗∗ satisfying conditions (i) -- (iii) of Theorem 2.
c.c.p. maps A
Lemma 1.1 of [15] implies that for each i there is a net (φλ
i : Fi → B)λ∈Λ of
c.c.p. order zero maps such that φλ
i (x) → φi(x) in the σ-strong∗ topology
(and therefore in the σ-weak topology) for all x ∈ Fi. Therefore, we may
6
J. CARRI ´ON AND C. SCHAFHAUSER
assume that the image of φi is contained in B for all i. Then, for all suffi-
ciently large i, we have that φi ◦ ψi ∈ K. Thus θ is in the point-weak closure
of K. As K is a convex set in B(A, B), its point-norm and point-weak clo-
sures coincide, by the Hahn-Banach Theorem2. Next, recall that the weak
topology on B is the same as the restriction to B of the weak-∗ topology on
B ∗∗ (with B regarded as a subspace of B ∗∗ via the standard map). But the
weak∗-topology on B ∗∗ is precisely the σ-weak topology (see e.g. [1, I.8.6]),
so the theorem follows.
(cid:3)
For a nuclear ∗-homomorphism θ : A → B as in Theorem 1, it is natural to
consider a stronger approximation property where the maps ψi are required
to be approximately multiplicative in the sense that
(ii′) kψi(a)ψi(b) − ψi(ab)k → 0 for all a, b ∈ A.
Question 6. For which nuclear ∗-homomorphisms θ : A → B between C ∗-
φi−→ B of θ
algebras A and B are there approximate factorizations A
as in Theorem 1 that also satisfy (ii′)?
ψi−→ Fi
A characterization of such θ is likely related to some appropriate form of
quasidiagonality. In particular, note that if θ is faithful and (φi), (ψi), and
(Fi) are as in Question 6, then for all a ∈ A, we have
(13)
kψi(a)k ≥ kφi(ψi(a))k → kθ(a)k = kak.
Now, the net (ψi) is an approximate embedding of A into finite dimensional
C ∗-algebras, and this shows that A is quasidiagonal. We end with some
further comments and partial results related to the Question 6.
First, Theorem 2.2 of [5] states that, for a separable nuclear C ∗-algebra
A, idA enjoys approximate factorizations as in Question 6 if and only if A is
quasidiagonal and all traces on A are quasidiagonal. It is not hard to show
that the separability assumption is superfluous.
Second, if A is quasidiagonal and B has no traces, then every nuclear
∗-homomorphism θ : A → B enjoys approximate factorizations as in Ques-
tion 6. This follows as in the proof of Theorem 1 with only slight modifica-
tions. The key observations are that
(a) because A is quasidiagonal, the approximate factorizations of the
θ−→ B ֒→ B ∗∗ given in Proposition 3 can be taken to
composition A
satisfy (ii′), and
(b) because B has no traces, B ∗∗ is properly infinite, and so the finite
case in the proof of Theorem 2 is not needed.
Third, one might also compare Question 6 with the recent results of [3],
which address the nuclear dimension and decomposition rank of O∞-stable
and O2-stable ∗-homomorphisms. In particular, Theorem D of [3] states
that a full O2-stable ∗-homomorphism with a separable and exact domain
2In fact, this is consequence of the Hahn-Banach Separation Theorem. See Lemma 2.3.4
of [6] for a proof.
DECOMPOSING NUCLEAR MAPS
7
has nuclear dimension zero if and only if it is nuclear and its domain is
quasidiagonal. Noting that decomposition rank zero is equivalent to nuclear
dimension zero, Proposition 1.7 of [3] implies such a map enjoys approxi-
mate factorizations as in Question 6. Again, via standard reductions, the
separability of A is not needed in this result.
Acknowledgments. The authors are happy to record their gratitude to
Stuart White, who asked the question leading to the main result of this note.
References
1. Bruce Blackadar, Operator algebras, Encyclopaedia of Mathematical Sciences, vol.
122, Springer-Verlag, Berlin, 2006, Theory of C ∗-algebras and von Neumann algebras,
Operator Algebras and Non-commutative Geometry, III. MR 2188261
2. Bruce Blackadar and Eberhard Kirchberg, Generalized inductive limits of finite-
dimensional C ∗-algebras, Math. Ann. 307 (1997), no. 3, 343 -- 380. MR 1437044
3. Joan Bosa, James Gabe, Aidan Sims, and Stuart White, The nuclear dimension of
O∞-stable C ∗-algebras, arXiv:1906:02066, 2019.
4. Nathanial P. Brown, Invariant means and finite representation theory of C ∗-algebras,
Mem. Amer. Math. Soc. 184 (2006), no. 865, viii+105. MR 2263412
5. Nathanial P. Brown, Jos´e R. Carri´on, and Stuart White, Decomposable approximations
revisited, Operator algebras and applications -- the Abel Symposium 2015, Abel Symp.,
vol. 12, Springer, [Cham], 2017, pp. 45 -- 65. MR 3837591
6. Nathanial P. Brown and Narutaka Ozawa, C ∗-algebras and finite-dimensional approx-
imations, Graduate Studies in Mathematics, vol. 88, American Mathematical Society,
Providence, RI, 2008. MR 2391387
7. Jos´e Carri´on, James Gabe, Christopher Schafhauser, Aaron Tikuisis, and Stuart
White, Classification of ∗-homomorphims I: simple nuclear C ∗-algebras, in prepara-
tion, 2019.
8. Jorge Castillejos, Samuel Evington, Aaron Tikuisis, Stuart White, and Wilhelm Win-
ter, Nuclear dimension of simple C ∗-algebras, arXiv:1901.05853, 2018.
9. Man Duen Choi and Edward G. Effros, Nuclear C ∗-algebras and the approximation
property, Amer. J. Math. 100 (1978), no. 1, 61 -- 79. MR 0482238
10. Marius Dadarlat, Morphisms of simple tracially AF algebras, Internat. J. Math. 15
(2004), no. 9, 919 -- 957. MR 2106154
11. Kenneth R. Davidson, C ∗-algebras by example, Fields Institute Monographs, vol. 6,
American Mathematical Society, Providence, RI, 1996. MR 1402012
12. James Gabe, Quasidiagonal traces on exact C ∗-algebras, J. Funct. Anal. 272 (2017),
no. 3, 1104 -- 1120. MR 3579134
13.
, A new proof of Kirchberg's O2-stable classification, arXiv:1706.03690, to ap-
pear in J. Reine Angew. Math., 2018.
14. Uffe Haagerup, A new proof of the equivalence of injectivity and hyperfiniteness for fac-
tors on a separable Hilbert space, J. Funct. Anal. 62 (1985), no. 2, 160 -- 201. MR 791846
15. Ilan Hirshberg, Eberhard Kirchberg, and Stuart White, Decomposable approximations
of nuclear C ∗-algebras, Adv. Math. 230 (2012), no. 3, 1029 -- 1039. MR 2921170
16. Eberhard Kirchberg, C ∗-nuclearity implies CPAP, Math. Nachr. 76 (1977), 203 -- 212.
MR 0512362
17. Eberhard Kirchberg and Mikael Rørdam, Central sequence C ∗-algebras and tenso-
rial absorption of the Jiang-Su algebra, J. Reine Angew. Math. 695 (2014), 175 -- 214.
MR 3276157
18. Eberhard Kirchberg and Wilhelm Winter, Covering dimension and quasidiagonality,
Internat. J. Math. 15 (2004), no. 1, 63 -- 85. MR 2039212
8
J. CARRI ´ON AND C. SCHAFHAUSER
19. Yasuhiko Sato, Stuart White, and Wilhelm Winter, Nuclear dimension and Z-stability,
Invent. Math. 202 (2015), no. 2, 893 -- 921. MR 3418247
20. Christopher Schafhauser, Subalgebras of simple AF-algebras, arXiv:1807.07381, 2018.
21. Dan Voiculescu, A note on quasi-diagonal C ∗-algebras and homotopy, Duke Math. J.
62 (1991), no. 2, 267 -- 271. MR 1104525
22. Wilhelm Winter, Classifying crossed product C ∗-algebras, Amer. J. Math. 138 (2016),
no. 3, 793 -- 820. MR 3506386
23. Wilhelm Winter and Joachim Zacharias, Completely positive maps of order zero,
Munster J. Math. 2 (2009), 311 -- 324. MR 2545617
24.
, The nuclear dimension of C ∗-algebras, Adv. Math. 224 (2010), no. 2, 461 --
498. MR 2609012
Department of Mathematics, Texas Christian University, Fort Worth, Texas
76129, USA.
E-mail address: [email protected]
Department of Mathematics, University of Nebraska -- Lincoln, Lincoln, Ne-
braska 68588, USA
E-mail address: [email protected]
|
1205.3553 | 1 | 1205 | 2012-05-16T05:16:37 | Orbit Representations from Linear mod 1 Transformations | [
"math.OA"
] | We show that every point $x_0\in [0,1]$ carries a representation of a $C^*$-algebra that encodes the orbit structure of the linear mod 1 interval map $f_{\beta,\alpha}(x)=\beta x +\alpha$. Such $C^*$-algebra is generated by partial isometries arising from the subintervals of monotonicity of the underlying map $f_{\beta,\alpha}$. Then we prove that such representation is irreducible. Moreover two such of representations are unitarily equivalent if and only if the points belong to the same generalized orbit, for every $\alpha\in [0,1[$ and $\beta\geq 1$. | math.OA | math |
Symmetry, Integrability and Geometry: Methods and Applications
SIGMA 8 (2012), 029, 9 pages
Orbit Representations
from Linear mod 1 Transformations
Carlos CORREIA RAMOS †, Nuno MARTINS ‡ and Paulo R. PINTO ‡
† Centro de Investiga¸cao em Matem´atica e Aplica¸coes,
R. Romao Ramalho, 59, 7000-671 ´Evora, Portugal
E-mail: [email protected]
‡ Department of Mathematics, CAMGSD, Instituto Superior T´ecnico,
Technical University of Lisbon, Av. Rovisco Pais, 1049-001 Lisboa, Portugal
E-mail: [email protected], [email protected]
Received March 14, 2012, in final form May 09, 2012; Published online May 16, 2012
http://dx.doi.org/10.3842/SIGMA.2012.029
Abstract. We show that every point x0 ∈ [0, 1] carries a representation of a C∗-algebra
that encodes the orbit structure of the linear mod 1 interval map fβ,α(x) = βx + α. Such
C∗-algebra is generated by partial isometries arising from the subintervals of monotonicity
of the underlying map fβ,α. Then we prove that such representation is irreducible. Moreover
two such of representations are unitarily equivalent if and only if the points belong to the
same generalized orbit, for every α ∈ [0, 1[ and β ≥ 1.
Key words: interval maps; symbolic dynamics; C∗-algebras; representations of algebras
2010 Mathematics Subject Classification: 46L55; 37B10; 46L05
Introduction
1
A famous class of representations of the Cuntz algebra On called permutative representations
were studied and classified by Bratteli and Jorgensen in [2, 3]. From the applications viewpoint,
and besides its own right, applications of representation theory of Cuntz and Cuntz -- Krieger
algebras to wavelets, fractals, dynamical systems, see e.g. [2, 3, 13], and quantum field theory
in [1] are particularly remarkable. For example, it is known that these representations of the
Cuntz algebra serve as a computational tool for wavelets analysts, see [12]. This is clear because
such a representation on a Hilbert space H induces a subdivision of H into orthogonal subspaces.
Then the problem in wavelet theory is to build orthonormal bases in L2(R) from these data.
Indeed this can be done [9] and these wavelet bases have the advantages over the earlier known
basis constructions (one advantage is the efficiency of computation). This method has also been
applied to the context of fractals that arise from affine iterated function systems [10]. Some of
these results have been extended to the more general class of Cuntz -- Krieger algebras, see [7, 13]
and subshift C∗-algebras [4, 11, 15] (whose underlying subshift is not necessarily of finite type)
in [5].
Symbolic dynamics is one of the main tools that we have used in [5, 7] to construct representa-
tions of Cuntz, Cuntz -- Krieger and subshift C∗-algebras. The C∗-algebra is naturally associated
to the given interval map and the Hilbert spaces naturally arise from the generalized orbits of
the interval map. For a particular family of interval maps, we were able to recover Bratteli and
Jorgensen permutative representations in [7] among the class of Markov maps (which underline
the Cuntz -- Krieger algebras of the transition matrix). We remark that while these Cuntz and
Cuntz -- Krieger algebras are naturally associated to the so-called Markov or periodic dynamical
systems, the subshift C∗-algebras are ready to incorporate bigger classes of interval maps.
2
C. Correia Ramos, N. Martins and P.R. Pinto
The interval maps that we treat in [5] are unimodal maps (that have precisely two subintervals
of monotonicity). Then the representations of the subshift C∗-algebra constructed in [5] are
shown to coincide with the ones constructed in [7] from the Cuntz -- Krieger algebra, provided the
underlying dynamical system is periodic and therefore has a finite transition Markov Matrix.
However, the proof of the irreducibility of the subshift C∗-algebras representations (for unimodal
maps without a finite transition Markov matrix) rely on the structure of these unimodal interval
maps where the C∗-algebra is generated by two partial isometries.
In this paper we construct representations (of a subshift C∗-algebra generated by n partial
isometries) from a family of interval maps and prove the irreducibility of these representations
(avoiding the unimodal maps techniques used in [5]).
Namely, we yield and study representations of a certain C∗-algebra on the generalized orbit
i∈Z f j
β,α(x0) of every point x0 ∈ [0, 1] from the interval map fβ,α : [0, 1] → [0, 1] defined by
fβ,α(x) = βx + α (mod 1) with β ≥ 1
(cid:83)
and α ∈ [0, 1[,
(1)
by fixing the parameters α and β. The underlying C∗-algebra OΛfβ,α
is generated by n partial
isometries where n is the number of monotonicity subintervals of fβ,α. See Fig. 1 for a graph of
one such map.
Figure 1. Graph of fβ,α with α =
√
2 − 1 and β = 2 (n = 3).
We show that the representation is irreducible. Moreover the representations of the same
algebra on the orbits of the two points x0 and y0 are unitarily equivalent if and only if the orbits
coincide. If the parameters α and β are so that the dynamical system ([0, 1], fβ,α) is periodic, the
above results were obtained by [6, 7], where the relevant C∗-algebra is the Cuntz -- Krieger OAfβ,α
and Afβ,α is the underlying Markov transition 0-1 (finite) matrix of fβ,α. A further particular
case is obtained when β = n is an integer and α = 0, in which case the n×n matrix Afn,0 = (ai,j)
is full, aij = 1 for all i, j, and thus recovering the Cuntz algebra On representations yielded
in [2, 3] using wavelet theory framework.
Of course we may fairly easily prove that we do get representations of such C∗-algebras in
the context of piecewise monotone maps, but the main concern is how to show irreducibility
(and unitarily equivalence) of such the representations, thus giving a rich family of representa-
tions attached to interval maps. The periodic or non-periodic cases for which we have n = 2
subintervals of monotonicity were carried out in [5]. We generalize here the construction of
the representations for generic piecewise monotone interval maps and prove irreducibility and
unitarily equivalence for the dynamical systems arising from equation (1) above -- and obviously
we look for values of the parameters α and β for which we do not have a finite Markov transition
matrix for fβ,α and thus we really get representations that cannot be recovered from [5, 6, 7].
A more detailed description of the paper is as follows. In Section 2 we provide some back-
ground material first on the operator algebras setup and then in symbolic dynamics [16]. The
0.00.20.40.60.81.00.00.20.40.60.81.0 Orbit Representations from Linear mod 1 Transformations
3
main results are in Section 3. We consider a partition I of the interval I = [0, 1] into subin-
tervals so that the restriction of fβ,α to each of these subintervals is monotone. Then for every
x0 ∈ I we explicitly define in equation (6) a linear operator on the Hilbert space Hx0 that arises
from the generalized orbit of x0, for every such subinterval. The Hilbert space Hx0 encodes the
generalized orbit orbit(x0) of x0 and in fact every ξ ∈ orbit(x0) is regarded as a vector ξ(cid:105) in Hx0
using Dirac's notation. Then we prove that these linear operators do satisfy the relation they
→ B(Hx0) of the C∗-algebra OAfβ,α
ought to satisfy, leading to a representation ρx0 : OAfβ,α
,
as in Proposition 2.
Then the main result of this paper is Theorem 1 where we show that representation ρx0 of the
underlying C∗-algebra OAfβ,α
is irreducible and that two such representations ρx0 and ρy0 are
unitarily equivalent if and only if y0 belongs to the generalized orbit of x0. The new ingredient
involved in the proof of Theorem 1 is the computation of the commutant A(cid:48)
β,α of C∗-algebra Aβ,α
(generated by the operators defined in equations (9) and (10)) in B(Hx0). Indeed, as soon as
we prove that the commutant is trivial A(cid:48)
β,α contains
ρx0(OAfβ,α
β,α = C1, we only have to show that A(cid:48)
)(cid:48) as in Proposition 3.
2 Preliminaries
In this section we provide some necessary background, starting with the operator algebras we
obtain from dynamical systems. A representation of a ∗-algebra A on a complex Hilbert space H
is a ∗-homomorphism π : A → B(H) into the ∗-algebra B(H) of bounded linear operators
on H. Usually representations are studied up to unitary equivalence. Two representations
π : A → B(H) and (cid:101)π : A → B((cid:101)H) are (unitarily) equivalent if there is a unitary operator
U : H → (cid:101)H (i.e., U is a surjective isometry) such that
U π(a) =(cid:101)π(a)U for every a ∈ A,
and in this case we write π ∼ π. A representation π : A → B(H) of some ∗-algebra is said to be
irreducible if there is no non-trivial subspace of H invariant with respect to all operators π(a)
with a ∈ A. A well known result, see e.g. [17, Proposition 3.13], says that π is irreducible if and
only if
x ∈ B(H) : xπ(a) = π(a)x for all a ∈ A =⇒ x = λ1,
(2)
for some complex number λ, where 1 denotes the identity of B(H). By the very definition of
commutant, (2) can be restated as follows: π(A)(cid:48) = C1. We will be interested in some classes
of C∗-algebras (= Banach ∗-algebras such that aa∗ = a2 holds for all a, see e.g. [17]).
Besides, if we have a representation π : A → B(H) of a C∗-algebra A, then π being a ∗-
homomorphism implies that π(a) ≤ a for all a ∈ A, thus π is automatically continuous, see
also e.g. [17, Secttion 1.5.7]).
N
Subshift C∗-algebras
Let Λ ⊆ Σ
be a subshift with a finite alphabet Σ = {1, . . . , n}. Exel [11] and Matsumoto [15]
constructed C∗-algebras associated to Λ. Carlsen and Silvestrov [4] unified the two constructions
that led them to a C∗-algebra OΛ, which is unital and generated partial isometries {ti}i∈Σ. Then
the partial isometries that generate OΛ obey the following relations:
(cid:88)
tit∗
i = 1,
t∗
αtαtβ = tβt∗
αβtαβ,
t∗
αtαt∗
βtβ = t∗
βtβt∗
αtα,
where tα = tα1 ··· tαα and tβ = tβ1 ··· tββ with α, β admissible words (if α = (α1, . . . , αk) with
αi ∈ Σ we denote by α the length k of α). The algebra OΛ is called the C∗-algebra associated
n(cid:88)
n(cid:88)
m(cid:91)
4
C. Correia Ramos, N. Martins and P.R. Pinto
to the subshift Λ or subshift C∗-algebra. Important properties of the subshift C∗-algebra OΛ
If Λ is a subshift
(e.g. simplicity) are naturally inherited from properties of the subshift Λ.
of finite type, then OΛ is nothing but the well known Cuntz -- Krieger algebra [15], where the
Cuntz -- Krieger algebra OA associated to a 0-1 matrix A = (aij) is the C∗-algebra [8] generated
by (non-zero) partial isometries s1, . . . , sn satisfying:
s∗
i si =
aijsjs∗
j
(i = 1, . . . , n),
sis∗
i = 1,
(3)
j=1
i=1
and the Cuntz algebra On is the Cuntz -- Krieger algebra OA with A full aij = 1 for all 1 ≤ i, j ≤ n.
partition of open sub-intervals of I, I = {I1, . . . , In} such that(cid:83)n
2.1 Symbolic dynamics on piecewise monotone interval maps
Let f : I → I be a piecewise monotone map of the interval I into itself, that is, there is a minimal
j=1 Ij = I and fIj is continuous
monotone, for every j = 1, . . . , n. We define fj := fIj . The inverse branches are denoted by
f−1
: f (Ij) → Ij. Let χIi be the characteristic function on the interval Ii. The following are
j
naturally satisfied
f ◦ f−1
(x) = χf (Ij )(x)x,
f−1
j
◦ fIj (x) = χIj (x)x.
j
intervals of I so that(cid:83)m
Let {1, 2, . . . , m} be the alphabet associated to some partition {J1, . . . , Jm} of open sub-
j=1 Jj = I, not necessarily I. The address map, is defined by
ad :
Jj → {1, 2, . . . , m},
ad(x) = i
if x ∈ Ji.
j=1
We define
Ωf :=
(cid:110)
x ∈ I : f k(x) ∈ m(cid:91)
j=1
(cid:111)
.
Jj for all k = 0, 1, . . .
Note that Ωf = I. The itinerary map it : Ωf → {1, 2, . . . , m}N
it(x) = ad(x)ad(f (x))ad(cid:0)f 2(x)(cid:1)···
is defined by
and let
Λf = it(Ωf ).
(4)
The space Λf is invariant under the shift map σ : {1, 2, . . . , m}N → {1, 2, . . . , m}N
defined by
σ(i1i2 ··· ) = (i2i3 ··· ),
and we have it ◦ f = σ ◦ it. We will use σ meaning in fact σΛf . A sequence in {1, 2, . . . , m}N
is
called admissible, with respect to f , if it occurs as an itinerary for some point x in I, that is, if
it belongs to Λf . An admissible word is a finite sub-sequence of some admissible sequence. The
set of admissible words of size k is denoted by Wk = Wk(f ). Given i1 ··· ik ∈ Wk, we define
Ii1···ik as the set of points x in Ωf which satisfy
ad(x) = i1, . . . , ad(cid:0)f k(x)(cid:1) = ik.
Orbit Representations from Linear mod 1 Transformations
5
2.2 Linear mod 1 interval maps
Now, let us consider the family of linear mod 1 transformations as in equation (1). In the sequel
we will denote fβ,α by f . The behavior of the dynamical system (I, f ) is characterized by the
sequences it(f (c+
j )), for each discontinuity point cj, see [14]. Let us consider the
partition of monotonicity I = {I1, . . . , In} of f , with
j )) and it(f (c−
I1 = ]0, (1 − α)/β[ , . . . , Ij = ](j − α)/β, (j + 1 − α)/β[ , . . . ,
. . . , In = ](n − 1 − α)/β, 1[ ,
which is the minimal partition of monotonicity for f (n = [β] + 1 with [β] being the integral
part of β and α > 0. For α = 0: n = [β] if β is an integer, and n = [β] + 1 if β /∈ N).
A characterization of the values of α, for which there is a Markov partition, is partially given
by the following:
Proposition 1. If itf (0) = (ξ1, ξ2, . . . , ξl, . . . ) and itf (1) are periodic (with ξl = ξ1) then
ξl + ξl−1β + ξl−2β2 + ··· + ξ1βl−1
1 + β + β2 + ··· + βl−1
.
α =
In particular α ∈ Q(β).
Proof . See [14, Proposition 2.6] for full details.
(5)
(cid:4)
(6)
3 Subshift algebras from linear mod 1 transformations
As in [7], we consider the equivalence relation
Rf = {(x, y) : f n(x) = f m(y) for some n, m ∈ N0}.
We write x ∼ y whenever (x, y) ∈ Rf . Consider the equivalence class Rf (x) (= (cid:83)
j∈Z f j(x)
also called the generalized orbit of x) and set Hx the Hilbert space l2(Rf (x)) with canonical
orthonormal basis {y(cid:105) : y ∈ Rf (x)}, in Dirac notation. Note that Hx = Hy (are the same
Hilbert spaces) whenever x ∼ y. The inner product (·,·) is given by
(cid:104)yz(cid:105) = (y(cid:105) ,z(cid:105)) = δy,z.
Let now f be the linear mod 1 transformation defined in equation (1) and I = {I1, . . . , In}
be the partition of monotonicity as written down in equation (5). For every i = 1, . . . , n, let
fi := fIi be the restriction of f to the subinterval Ii of the partition I. For every i ∈ {1, 2, . . . , n}
let us define an operator Ti on Hx defined first on the orthonormal basis as follows:
Ti y(cid:105) = χf (Ii)(y)(cid:12)(cid:12)f−1
i
(y)(cid:11)
and then extend it by linearity and continuity to Hx. Note that χf (Ii)(x) = 1 if and only if there
is a pre-image of x in Ii. The we have
Indeed, on one hand
i y(cid:105) = χIi(y)f (y)(cid:105) .
T ∗
(y(cid:105), Tiz(cid:105)) =(cid:0)y(cid:105), χf (Ii)f−1
i
(z)(cid:105)(cid:1)
(7)
and on the other hand we have
(T ∗
i y(cid:105),z(cid:105)) = (χIi(y)f (y)(cid:105),z(cid:105)) = χIi(y)δf (y),z.
So since χf (Ii)f−1
(z) = χIi(y)δf (y),z we have shown that the adjoint of Ti is given by equa-
tion (7). We further remark that Ti is a partial isometry: namely, Ti is an isometry on its
restriction to span{y(cid:105) : y ∈ f (Ii)} ∩ H x and vanishes in the remaining part of Hx.
i
6
C. Correia Ramos, N. Martins and P.R. Pinto
Lemma 1. The operators Ti satisfy the relations
TiT ∗
i = 1,
T ∗
µ TµTν = TνT ∗
µνTµν
and
T ∗
µ TµT ∗
ν Tν = T ∗
ν TνT ∗
µ Tµ,
n(cid:88)
i=1
for µ, ν given admissible words.
Proof . Consider TiT ∗
i acting on a vector y(cid:105) of the canonical basis of Hx,
i y(cid:105) = χIi(y)Ti f (y)(cid:105) = χIi(y)χf (Ii)(f (y))(cid:12)(cid:12)f−1
◦ f (y)(cid:11) = χIi(y)y(cid:105) ,
i
TiT ∗
since χIi(y)χf (Ii)(f (y)) = χIi(y). Then
(T1T ∗
1 + ··· + TnT ∗
n )y(cid:105) = (χI1(y) + ··· + χIn(y))y(cid:105) = y(cid:105) .
µ TµTν acting on a vector y(cid:105) of the canonical basis for some µ = µ1 ··· µk,
Now, consider T ∗
ν = ν1 ··· νr admissible words,
µ TµTν y(cid:105) = T ∗
T ∗
= T ∗
µ Tµχf r(Iν )(y)(cid:12)(cid:12)f−1
µ χf k+r(Iµν )(y)(cid:12)(cid:12)f−1
= χf k+r(Iµν )(y)(cid:12)(cid:12)f−1
νr (y)(cid:11)
νr (y)(cid:11) .
µk
ν1 ◦ ··· ◦ f−1
µ1 ◦ ··· ◦ f−1
ν1 ◦ ··· ◦ f−1
νr (y)(cid:11)
◦ f−1
ν1 ◦ ··· ◦ f−1
On the other hand
TνT ∗
µνTµν y(cid:105) = TνT ∗
µνχf k+r(Iµν )(y)(cid:12)(cid:12)f−1
= Tνχf k+r(Iµν )(y)y(cid:105) = χf k+r(Iµν )(y)(cid:12)(cid:12)f−1
µ1 ◦ ··· ◦ f−1
µk
νr (y)(cid:11)
νr (y)(cid:11) .
◦ f−1
ν1 ◦ ··· ◦ f−1
ν1 ◦ ··· ◦ f−1
Finally since T ∗
easily conclude that T ∗
µ Tµy(cid:105) = χf k(Iµ)(y)y(cid:105) and T ∗
µ Tµ.
ν Tν = T ∗
µ TµT ∗
ν TνT ∗
(8)
ν Tνy(cid:105) = χf r(Iν )(y)y(cid:105) for admissible words µ, ν, we
(cid:4)
As an immediate consequence of Lemma 1 (and above equation (8)) we obtain the following.
Proposition 2. Let OΛf be the subshift algebra associated to the subshift Λf as defined in (4)
above, then ρx : OΛf → B(Hx) defined by ti → Ti is a representation of OΛf .
We remark here that T ∗
i Ti = 1 for all i = 2, . . . , n − 1 and T ∗
1 T1 = 1 if and only if α = 0.
n Tn = 1 if and only if β = n ∈ N. Therefore ρx is a representation of a Cuntz algebra
Besides T ∗
if and only if α = 0 and β is a positive integer. In this case, the interval map f is a Markov map
and moreover the partition with the monotonicity intervals, as in (5), reduces to
I1 = ]0, 1/n[,
I2 = ]1/n, 2/n[,
. . . ,
Ij = ](j − 1)/n, j/n[,
. . . ,
In = ](n − 1)/n, 1[
and coincides with the (minimal) Markov partition [7]. From the viewpoint of interval maps,
these Cuntz algebra On representations were treated in [7, Remark 2.9].
We remark that if fβ,α is a linear mod 1 map with α /∈ Q and β = 1 then f is not a Markov
map by Proposition 1 and thus the representation ρx of Proposition 2 is never a representation
of a Cuntz -- Krieger algebra.
Orbit Representations from Linear mod 1 Transformations
7
3.1 Irreducibility of the representations
For the linear mod 1 transformation map f and the linear operators T1, . . . , Tn ∈ B(Hx) defined
in equation (6), we may consider the following operator
V = T ∗
1 + ··· + T ∗
n ,
(9)
which satisfies V y(cid:105) = f (y)(cid:105) on every vector basis y(cid:105). In general V is not unitary (unless β = 1
so that f becomes an invertible function). Let U be the diagonal operator
U y(cid:105) = e2πiy y(cid:105) ,
(10)
which is an unitary operator, with U∗ y(cid:105) = e−2πiy y(cid:105) . In order to emphasize that U, V ∈ B(Hx),
we write Ux and Vx for the above operators U and V , respectively.
For a self-adjoint set of operators S ⊆ B(H) on some Hilbert space H, containing the
identity 1, the von Neumann algebra generated by S equals the double commutant S(cid:48)(cid:48), which
in turn is also equal to the closure of S under the strong operator topology (this is the famous
bicommutant von Neumann theorem e.g. the textbook [17]). We note that S(cid:48) = {t ∈ B(H) :
ts = st, for all s ∈ S} and S(cid:48)(cid:48) = (S(cid:48))(cid:48) and S(cid:48)(cid:48)(cid:48) = S(cid:48). Also si converges to s in the strong operator
topology if (si − s)ξ → 0 for every vector ξ ∈ H.
Let Aβ,α = C∗(U, V ) be the C∗-subalgebra of B(Hx) generated by U and V , and consider
the representation ρx from Proposition 2.
Proposition 3. We have Aβ,α ⊆ ρx(OΛf )(cid:48)(cid:48).
Proof . By definition of the C∗-algebra Aβ,α, we only need to prove that U , V belong
to ρx(OΛf )(cid:48)(cid:48).
It is clear that V ∈ ρx(OΛf ) ⊆ ρx(OΛf )(cid:48)(cid:48). We now show that U ∈ ρx(OΛf )(cid:48)(cid:48).
For each µ ∈ Wk, let m(µ) be some point in Iµ ∩ Rf (x). Note that if we have it(y) = (αj)∞
j=1,
for some point y ∈ Rf (x) then lim
j→∞ m(α1 ··· αj) = y, and the limit is independent on the
Let Mk = (cid:80)
choice of m(α1 ··· αj) ∈ Iα1···αj , since for each j ∈ N there is r > j so that Iα1···αj ⊃ Iα1···αr .
k→∞ Mk = U in the strong topology, since
k→∞(cid:107)Mkv − U v(cid:107) = 0, for every v ∈ Hx. Therefore, U is in the von Neumann algebra generated
(cid:4)
by the operators T1, . . . , Tn.
Lemma 2. Let x ∈ I and f be the linear mod 1 transformation (1) with fixed α and β. Let
Q ∈ B(Hx) be an operator commuting with both U and V , then Q = λI for some λ ∈ C.
µ . We can see that lim
e2πim(µ)TµT ∗
µ∈Wk
lim
Proof . First of all we remark that the
e2πiz
are all distinct,
(11)
with z ∈ Rf (x), since z ∈ [0, 1]. Let Q ∈ B(Hx) commuting with both U and V . For each
z ∈ Rf (x) let µz := e2πiz. By definition Uz(cid:105) = µzz(cid:105), i.e., every µz is an eigenvalue of U . We
easily get
U Qz(cid:105) = µzQz(cid:105)
(12)
by applying the definitions and the fact that U and Q commute. For every z ∈ Rf (x), set
ξz = Qz(cid:105). Then we can write equation (12) as follows: U ξz = µzξz. Since {z(cid:105), z ∈ Rf (x)} is
an o.n. basis of Hx, there are constants cw with w ∈ Rf (x) such that ξz =(cid:80) cww(cid:105). Since
(cid:88)
µwcww(cid:105)
U ξz =
w∈Rf (x)
8
and
(cid:88)
w∈Rf (x)
µzξz =
µzcww(cid:105),
C. Correia Ramos, N. Martins and P.R. Pinto
we conclude that cw = 0 for all w (cid:54)= z, because the µw's are all distinct by (11) and {w(cid:105)} is
an o.n. basis of Hx. It follows that ξz = czz(cid:105) or equivalently Qz(cid:105) = czz(cid:105) for some cz ∈ C.
But V z(cid:105) = f (z)(cid:105), so V Q = QV gives cz = cf (z). Therefore Q is a multiple of the identity
(cid:4)
operator.
Theorem 1. The representation ρx of the subshift C∗-algebra OΛf as in Proposition 2 is irre-
ducible. Moreover ρx ∼ ρy if and only if x ∼ y.
Proof . To prove that ρx is irreducible we prove that ρx(OΛf )(cid:48) = CI. First note that from
Proposition 3 we have C∗(U, V ) ⊆ ρx(OΛf )(cid:48)(cid:48) and so taking commutant and using von Neumann
bicommutant theorem, we conclude that ρx(OΛf )(cid:48) ⊆ C∗(U, V )(cid:48).
Now let Q ∈ ρx(OΛf )(cid:48). So we can conclude that Q ∈ C∗(U, V )(cid:48) and thus Q commutes with
both U and V . By Lemma 2 we conclude that Q = λI for some λ ∈ C. Therefore ρx(OΛf )(cid:48) = CI
and so ρx is an irreducible representation of OΛf .
It is clear that if Rf (x) = Rf (y), then ρx and ρy are unitarily equivalent. Notice that Ux
and Uy have the same eigenvalues if and only if x ∼ y. Hence ρx and ρy can be unitarily
equivalent only when x ∼ y.
(cid:4)
Remark 1. If β = n is an integer and α = 0 then the partition of I into monotonicity subinter-
vals of fn,0 is a Markov partition, the subshift Λfn,0 is the full shift and the underlying C∗-algebra
is the Cuntz algebra On. Furthermore we recover in Theorem 1 our previous result obtained
in [7].
Acknowledgment
First author acknowledges CIMA-UE for financial support. The other authors were partially
supported by the Funda¸cao para a Ciencia e a Tecnologia through the Program POCI 2010/FE-
DER.
References
[1] Abe M., Kawamura K., Recursive fermion system in Cuntz algebra. I. Embeddings of fermion algebra into
Cuntz algebra, Comm. Math. Phys. 228 (2002), 85 -- 101, math-ph/0110003.
[2] Bratteli O., Jorgensen P.E.T., Iterated function systems and permutation representations of the Cuntz
algebra, Mem. Amer. Math. Soc. 139 (1999), no. 663, 89 pages, funct-an/9612002.
[3] Bratteli O., Jorgensen P.E.T., Ostrovs'kyı V., Representation theory and numerical AF-invariants. The
representations and centralizers of certain states on Od, Mem. Amer. Math. Soc. 168 (2004), no. 797,
178 pages, math.OA/9907036.
[4] Carlsen T.M., Silvestrov S., C∗-crossed products and shift spaces, Expo. Math. 25 (2007), 275 -- 307,
math.OA/0512488.
[5] Correia Ramos C., Martins N., Pinto P.R., On C∗-algebras from interval maps, Complex Anal. Oper. Theory,
to appear.
[6] Correia Ramos C., Martins N., Pinto P.R., Orbit representations and circle maps, in Operator Algebras,
Operator Theory and Applications, Oper. Theory Adv. Appl., Vol. 181, Birkhauser Verlag, Basel, 2008,
417 -- 427.
[7] Correia Ramos C., Martins N., Pinto P.R., Sousa Ramos J., Cuntz -- Krieger algebras representations from
orbits of interval maps, J. Math. Anal. Appl. 341 (2008), 825 -- 833.
Orbit Representations from Linear mod 1 Transformations
9
[8] Cuntz J., Krieger W., A class of C∗-algebras and topological Markov chains, Invent. Math. 56 (1980),
251 -- 268.
[9] Daubechies I., Ten lectures on wavelets, CBMS-NSF Regional Conference Series in Applied Mathematics,
Vol. 61, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 1992.
[10] Dutkay D.E., Jorgensen P.E.T., Wavelet constructions in non-linear dynamics, Electron. Res. Announc.
Amer. Math. Soc. 11 (2005), 21 -- 33, math.DS/0501145.
[11] Exel R., A new look at the crossed-product of a C∗-algebra by an endomorphism, Ergodic Theory Dynam.
Systems 23 (2003), 1733 -- 1750, math.OA/0012084.
[12] Jorgensen P.E.T., Certain representations of the Cuntz relations, and a question on wavelets decompositions,
in Operator Theory, Operator Algebras, and Applications, Contemp. Math., Vol. 414, Amer. Math. Soc.,
Providence, RI, 2006, 165 -- 188, math.CA/0405372.
[13] Marcolli M., Paolucci A.M., Cuntz -- Krieger algebras and wavelets on fractals, Complex Anal. Oper. Theory
5 (2011), 41 -- 81, arXiv:0908.0596.
[14] Martins N., Sousa Ramos J., Cuntz -- Krieger algebras arising from linear mod one transformations,
in
Differential Equations and Dynamical Systems (Lisbon, 2000), Fields Inst. Commun., Vol. 31, Amer. Math.
Soc., Providence, RI, 2002, 265 -- 273.
[15] Matsumoto K., On C∗-algebras associated with subshifts, Internat. J. Math. 8 (1997), 357 -- 374.
[16] Milnor J., Thurston W., On iterated maps of the interval, in Dynamical Systems (College Park, MD, 1986 --
1987), Lecture Notes in Math., Vol. 1342, Springer, Berlin, 1988, 465 -- 563.
[17] Pedersen G.K., C∗-algebras and their automorphism groups, London Mathematical Society Monographs,
Vol. 14, Academic Press Inc., London, 1979.
|
1711.09526 | 1 | 1711 | 2017-11-27T04:24:10 | An infinite quantum Ramsey theorem | [
"math.OA",
"math-ph",
"math.CO",
"math.FA",
"math-ph"
] | We prove an infinite Ramsey theorem for noncommutative graphs realized as unital self-adjoint subspaces of linear operators acting on an infinite dimensional Hilbert space. Specifically, we prove that if V is such a subspace, then provided there is no obvious obstruction, there is an infinite rank projection P with the property that the compression PVP is either maximal or minimal in a certain natural sense. | math.OA | math | AN INFINITE QUANTUM RAMSEY THEOREM
MATTHEW KENNEDY, TARAS KOLOMATSKI, AND DANIEL SPIVAK
Abstract. We prove an infinite Ramsey theorem for noncommu-
tative graphs realized as unital self-adjoint subspaces of linear oper-
ators acting on an infinite dimensional Hilbert space. Specifically,
we prove that if V is such a subspace, then provided there is no
obvious obstruction, there is an infinite rank projection P with the
property that the compression P V P is either maximal or minimal
in a certain natural sense.
.
A
O
h
t
a
m
[
1
v
6
2
5
9
0
.
1
1
7
1
:
v
i
X
r
a
1. Introduction
The notion of a noncommutative graph first appeared in the work
of Duan, Severini and Winter [2] in quantum information theory as a
generalization, for a quantum channel, of the confusability graph of a
channel from classical information theory. In this setting, a noncom-
mutative graph takes the form of an operator system, which is a unital
self-adjoint subspace of operators acting on a Hilbert space.
The idea of viewing an operator system as a noncommutative gener-
alization of a graph appears to be much more broadly applicable. For
example, the work of Stahlke [10] and Weaver [11], as well as many
others (see e.g. [3, 6, 8]) shows that there are useful generalizations for
operator systems of many graph-theoretic constructions and results.
Recently, Weaver [12] proved an analogue of Ramsey's theorem for
operator systems acting on finite dimensional spaces. This result has
concrete applications, for example, to the theory of quantum error cor-
rection. But perhaps more interesting, Weaver's work suggests that
operator systems provide a new and potentially very interesting set-
ting for Ramsey theory.
In this paper, we adopt this perspective and prove an analogue of
the infinite Ramsey theorem for operator systems acting on infinite
dimensional spaces.
2010 Mathematics Subject Classification. Primary 05C55, 05D10, 13C99, 15A60,
46L07 .
Key words and phrases. Ramsey theory, graph theory, operator systems, opera-
tor theory, quantum information theory, quantum error correction.
First author supported by NSERC Grant Number 418585.
1
2
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
For a complex Hilbert space H, let B(H) denote the space of bounded
(i.e. continuous) linear operators on H. If H is finite dimensional, say
dim H = n, then B(H) can be identified with the n×n matrices Mn(C).
A subspace V ⊆ B(H) is said to be an operator system if it is unital,
meaning that it contains the identity operator IH, and self-adjoint,
meaning that A∗ ∈ V whenever A ∈ V, where A∗ denotes the adjoint
of A. For a general reference on operator systems, we direct the reader
to Paulsen's book [7].
An easy way to see an operator system as a generalization as a graph
is to consider the operator system associated to a graph G = (V, E)
with vertex set V and edge set E. Let HG be a Hilbert space with
dim HG = V and let (hv)v∈V be an orthonormal basis for HG indexed
by V . For vectors x, y ∈ HG, define a rank one operator xy∗ ∈ B(HG)
by xy∗(z) = hz, yix for z ∈ HG. The operator system VG ⊆ B(HG)
corresponding to G is defined by
VG = span{IHG, hvh∗
w : (v, w) ∈ E}.
It is clear that VG completely encodes G. A result of Paulsen and
Ortiz [8] implies that if G′ = (V ′, E′) is another graph with V = V ′,
then G and G′ are isomorphic if and only if the corresponding operator
systems VG and VG′ are completely order isomorphic, i.e.
isomorphic
in the category of operator systems.
Recall that a subset of vertices C ⊆ V is said to be an anticlique
if there are no edges between any pair of vertices in C. On the other
hand, C is said to be a clique if every pair of distinct vertices in C is
adjacent.
Let PC ∈ B(HG) denote the projection onto {hv : v ∈ C}. It is easy
to see that C is an anticlique if and only if PCVGPC = CPC, where
PCVGPC denotes the compression PCVGPC = {PCAPC : A ∈ VG}. If
C has finite cardinality, say C = k, then it is also easy to see that C
is a clique if and only if the compression PCVGPC is maximal, in the
sense that dim PCVGPC = k2 or equivalently, PCVGPC = PCB(HG)PC.
Let V ⊆ B(H) be an arbitrary operator system. Motivated by the
above considerations, for k ≥ 1 Weaver says that a rank k projection
P ∈ B(H) is a quantum k-anticlique for V if P VP = CP . On the
other hand, Weaver says that P is a quantum k-clique for V if P VP =
P B(H)P .
Ramsey's theorem [9] asserts that for k ≥ 1 there is N ≥ 1 such
that every graph with at least N vertices has either a k-clique or a
k-anticlique. Weaver's noncommutative Ramsey theorem [12] is an
analogue of this result for operator systems.
AN INFINITE QUANTUM RAMSEY THEOREM
3
Theorem 1.1 (Weaver). Let V ⊆ B(H) be an operator system on a
finite dimensional Hilbert space H. For k ≥ 1 there is N ≥ 1 such
that if dim H ≥ N, then V has either a quantum k-clique or a quantum
k-anticlique.
In order to motivate the definition of an infinite quantum clique,
consider an infinite graph G with corresponding operator system VG
as above. Let C ⊆ V be a subset of vertices and let PC denote the
projection onto {hv : v ∈ C} as above. We equip B(HG) with the
weak (or weak operator) topology, which is the topology of coordinate-
wise convergence, meaning that a net of operators (Ti) in B(HG) weakly
converges to an operator T ∈ B(HG) if and only if limihTix, yi = hT x, yi
for all x, y ∈ HG.
It is easy to check that C is an infinite clique
if and only if PC has infinite rank and PCVGPC is weakly dense in
PCB(HG)PC.
Definition 1.2. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. An infinite quantum clique of V is an
infinite rank orthogonal projection P ∈ B(H) with the property that
P VP = {P V P : V ∈ V} is weakly dense in P B(H)P , i.e. the weak
closure of the compression P VP is maximal. An infinite quantum anti-
clique of V is an infinite rank orthogonal projection P ∈ B(H) with the
property that P VP = CP , i.e. the compression P VP is minimal. An
infinite quantum obstruction is an infinite rank projection P ∈ B(H)
with the property that P VP is finite dimensional, every operator in
P VP can be written as the sum of a scalar multiple of P and a com-
pact operator and P VP contains a self-adjoint compact operator with
range dense in P H.
The infinite Ramsey theorem asserts that a graph on infinitely many
vertices has either an infinite anticlique or an infinite clique. The main
result in our paper is an analogue of this result for operator systems.
Theorem 1.3. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. Then it has either an infinite quantum
clique, an infinite quantum anticlique or an infinite quantum obstruc-
tion.
As a corollary to Theorem 1.3, we obtain a dichotomy that is slightly
simpler to state but provides less information. Before stating it, we
require the following definition.
Definition 1.4. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. An infinite quantum almost anticlique of
V is an infinite rank projection P ∈ B(H) such that the compression
4
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
P VP is finite dimensional and every operator in P VP can be written
as the sum of a scalar multiple of P and a compact operator.
Corollary 1.5. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. Then it has either an infinite quantum
clique or an infinite quantum almost anticlique.
The next two examples show that there are operator systems on
infinite dimensional Hilbert spaces with no infinite quantum anticlique
and no infinite quantum clique. By Theorem 1.3 an operator system
with this property necessarily has an infinite quantum obstruction.
Example 1.6. Let V ⊆ B(ℓ2(N)) be any finite dimensional operator
system containing a self-adjoint compact operator K with dense range.
Since V is finite dimensional, it clearly has no infinite quantum clique.
On the other hand, for any infinite rank orthogonal projection P ∈
B(ℓ2(N)), the compression P KP is nonzero and compact, so P VP 6=
CP . Therefore, V also has no infinite quantum anticlique. The identity
operator is an infinite quantum obstruction for V.
The next example is an infinite dimensional variant of the example
from [12, Proposition 2.3].
Example 1.7. Let H = ℓ2(N) and let (en) denote the standard or-
thonormal basis of H. Let K ∈ B(H) denote the compact operator
defined by
K = Xn≥1
1
n
ene∗
n,
where the sum is taken in the weak topology. Note that K has dense
range. Define an operator system V ⊆ B(H) by
V = span{IH, K, e1e∗
n, ene∗
1 : n ≥ 1}.
Let P ∈ B(H) be any infinite rank projection. Then P VP contains
the nonzero compact operator P KP , so P is not an infinite quantum
anticlique for V. We will show that P is not an infinite quantum clique
for V.
First suppose P e1 = 0. Then for n ≥ 1, P e1e∗
1P =
0, so P VP = span{P, P KP }. Hence P is an infinite quantum obstruc-
tion for V.
nP = 0 and P ene∗
Next suppose P e1 6= 0. Let (fn) be an orthonormal basis for P H with
nP = kP e1kf1v∗
n
f1 = P e1/kP e1k. For n ≥ 1, let vn = P en. Then P e1e∗
and P ene∗
1P = kP e1kvnf ∗
1 , so
P VP = span{P, P KP, f1v∗
n, vnf ∗
1 : n ≥ 1}.
AN INFINITE QUANTUM RAMSEY THEOREM
5
To see that P is not an infinite quantum clique for V, note that if P VP
is weakly dense in P B(H)P , then for any infinite rank projection Q ∈
B(H) with Q ≤ P , QVQ is weakly dense in QB(H)Q. In other words, if
P is an infinite quantum anticlique for V, then so is Q. However, let Q
denote the infinite rank projection onto the closure of span{fn : n ≥ 2}.
Then Q ≤ P , but for n ≥ 1, Qf1v∗
1 Q = 0, so
QVQ = span{Q, QKQ}. Hence Q is an infinite quantum obstruction
for V.
nQ = 0 and Qvnf ∗
In the definition of an infinite quantum clique, it is tempting to
replace the condition that P VP is weakly dense in P B(H)P with the
condition that P VP = P B(H)P , where V denotes the weak closure of
V. The next example shows that our main result would not hold with
this definition.
Example 1.8. Let H = ℓ2(N ∪ {0}) and let (en) denote the standard
orthonormal basis of H. For n ≥ 1, let
n + ne0e∗
n + nene∗
0,
An = ene∗
and define an operator system V0 ⊆ B(H) by V0 = span{I, An : n ≥ 1}.
Then every operator A ∈ V0 satisfies hAei, eji = 0 for i, j ≥ 1 with
i 6= j.
Let V denote the weak closure of V0. Then V is a weakly closed
operator system and every operator B ∈ V satisfies hBei, eji = 0 for
i, j ≥ 1 with i 6= j. In particular, every operator in V can be written as
the sum of a diagonal operator and an operator of rank at most two.
For m ≥ 1, let
Tm = 2me0e∗
0 + e0e∗
m + eme∗
0 − 2meme∗
m.
Then tr(TmI) = 0 and tr(TmAn) = 0 for all n ≥ 1, so tr(TmA) = 0 for
all A ∈ V0. Hence tr(TmB) = 0 for all B ∈ V, or equivalently,
2mhBem, emi = hBe0, emi + hBem, e0i + 2mhBe0, e0i.
For B ∈ V with kBk ≤ 1, let B′ = B − hBe0, e0iI. Then kB′k ≤ 2
and hB′e0, e0i = 0, so it follows from above that for m ≥ 1,
hB′em, emi =
1
2m
hB′e0, eni + hB′en, e0i ≤
2
m
.
In particular, limmhB′em, emi = 0. Hence B′ can be written as the sum
of a compact diagonal operator and an operator of rank at most two.
In particular, B′ is compact. It follows that V ⊆ CIH + K(H), where
K(H) denotes the space of compact operators on H. Since K(H) is
an ideal in B(H), this implies that if P ∈ B(H) is a projection, then
P VP ⊆ CP +K(H). However, if P has infinite rank, then CP +K(H) is
6
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
a proper subset of P B(H)P . Hence there is no infinite rank projection
P ∈ B(H) such that P VP = P B(H)P .
Next, we claim that V does not have an infinite quantum anticlique.
To see this, suppose for the sake of contradiction that P ∈ B(H) is
an infinite rank projection such that P VP = CP . For n ≥ 1, the fact
that P AnP is compact implies that P AnP = 0. Hence for nonzero
f ∈ P H ∩ {e0}⊥,
0 = hAnf, f i
= hf, enihen, f i + nhf, enihe0, f i + nhe0, f ihen, f i
= hf, eni2.
Since {en : n ≥ 1} is an orthonormal basis for {e0}⊥, this implies
f = 0. Hence P H ∩ {e0}⊥ = 0. But it is not difficult to show that
P H ∩ {e0}⊥ 6= 0 (see Lemma 3.2), giving a contradiction.
Finally, we claim that V does not have an infinite quantum obstruc-
tion. To see this, suppose for the sake of contradiction that P ∈ B(H)
is an infinite rank projection with the property that P VP is finite di-
mensional and contains a self-adjoint compact operator K with range
dense in P H. Since P VP is finite dimensional, so is P V0P , say
P V0P = span{P, P An1P, . . . , P AnkP }.
But P VP is contained in the weak closure of P V0P and, being finite
dimensional, P V0P is already weakly closed. So P VP = P V0P . In par-
ticular, P VP is spanned by P and finitely many finite rank operators,
so it does not contain any compact operators of infinite rank, giving a
contradiction.
The notion of a quantum anticlique arises in quantum information
theory, specifically in the theory of quantum error correction. Let H
and K be Hilbert spaces and let E : Bt(H) → Bt(K) be a quantum
channel, i.e. a completely positive trace preserving map between the
spaces Bt(H) and Bt(K) of trace class operators on H and K respec-
tively (see e.g.
[1] for details). Then there are operators (Ei)i∈I in
B(H, K) such that
E(T ) = Xi∈I
EiT E∗
i
for T ∈ Bt(H), where the sum is taken in the weak topology. The
operator system associated to E is
VE = span{IH, E∗
i Ej : i, j ∈ I}.
AN INFINITE QUANTUM RAMSEY THEOREM
7
In the finite dimensional setting, a quantum anticlique for VE corre-
sponds to an error correcting code for E [4, 5]. In the infinite dimen-
sional setting, an infinite quantum anticlique also gives rise to an error
correcting code for E [1].
In Section 2, we establish a key technical result about dilations of
operator systems. To prove our main result, we first reduce to the
case of a diagonalizable operator system in Section 3. The case of a
diagonalizable operator system is then handled in Section 4. The main
result is proved in Section 5.
Acknowledgements. The authors are grateful to Nik Weaver for sug-
gesting the inclusion of Corollary 1.5 and Example 1.7.
2. Dilation lemma
Lemma 2.1. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. Let K be another infinite dimensional
Hilbert space and let T ∈ B(H, K) be an operator such that T VT ∗ is
weakly dense in B(K). Then the projection P ∈ B(H) onto ker(T )⊥ is
an infinite quantum clique for V.
Proof. Let T ∗ = V ∗S be the polar decomposition for T ∗, so that V ∗ ∈
B(K, H) is an isometry and S = T ∗ = (T T ∗)1/2 ∈ B(K). Then
P = V ∗V . Let V ′ denote the operator system V ′ = V VV ∗ ⊆ B(K).
Then we can write
T VT ∗ = SV VV ∗S = SV ′S.
By the spectral theorem, there is a projection-valued measure E on
R such that
S = Z kSk
0
s dEs
For t > 0, define operators St, Rt ∈ B(K) by
Rt = Z kSk
t
s−1 dEs,
St = Z kSk
t
s dEs.
Also, let Et = E((t, kSk)). Then RtS = RtSt = Et.
For t > 0,
EtV ′Et = RtSV ′SRt = RtT VT ∗Rt.
Since T VT ∗ is weakly dense in B(K), EtT VT ∗Et is weakly dense in
EtB(K)Et. Since Rt is invertible in EtB(K)Et, this implies that RtT VT ∗Rt
(and hence EtV ′Et) is weakly dense in EtB(K)Et.
Since (Et) weakly converges to IK, it follows that V ′ is weakly dense
in B(K). Hence the compression P VP = V ∗V ′V is dense in P B(H)P .
(cid:3)
8
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
Lemma 2.2. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. Suppose there is an isometry V : ℓ2(N) →
H ⊕ H such that V ∗(V ⊕ 0H)V is weakly dense in B(ℓ2(N)). Then V
has an infinite quantum clique.
Proof. Define T ∈ B(H, ℓ2(N)) by T h = V ∗(h, 0) for h ∈ H. Then
T VT ∗ = V ∗(V ⊕ 0H)V is weakly dense in B(ℓ2(N)) by assumption. We
conclude from Lemma 2.1 that V has an infinite quantum clique. (cid:3)
Lemma 2.3. Let (An) be a sequence of operators on a countably infinite
dimensional Hilbert space H with orthonormal basis (hn). For n ≥ 1
and i, j ≥ 1, let αn
ij = hAnhj, hii, so that
An = Xi,j≥1
ijhih∗
αn
j ,
where the sum is taken in the weak topology. Let (xn) be a sequence of
vectors in ℓ2(N) such that kxnk ≤ 2−(n+1) for n ≥ 1. For n ≥ 1, let
Tn = Xi,j≥1
ijxix∗
αn
j ,
where the sum is taken in the weak topology. Then there is an isometry
V : ℓ2(N) → H ⊕ H such that for n ≥ 1,
V ∗(An ⊕ 0H)V = Tn.
Proof. Let (en) denote the standard orthonormal basis for ℓ2(N). Define
a linear operator R : ℓ2(N) → ℓ2(N) by Ren = xn for n ≥ 1. Since
kxnk ≤ 2−(n+1), it is not difficult to check that kRk ≤ 1. Define a
linear operator S : ℓ2(N) → ℓ2(N) ⊕ ℓ2(N) by S = R ⊕ (1 − R∗R)1/2.
Then S is an isometry, so the sequence (yn) in ℓ2(N) ⊕ ℓ2(N) defined
by yn = Sen for n ≥ 1 is orthonormal. Let Y denote the closure of
span{yn : n ≥ 1}.
Define an isometry W : Y → H ⊕ H by W yn = hn ⊕ 0H for n ≥ 1.
Then we can extend W to an isometry from ℓ2(N) to H ⊕ H that we
continue to denote by W .
Let Q1 : ℓ2(N) ⊕ ℓ2(N) → ℓ2(N) denote the projection onto the first
summand and define V : ℓ2(N) → H ⊕ H by V = W Q∗
1. Then V is
an isometry and V ∗(hn ⊕ 0H) = Q1W ∗W yn = Q1yn = xn for n ≥ 1.
Hence
V ∗(An ⊕ 0H)V = Xi,j≥1
ijV ∗(hi ⊕ 0H)(hj ⊕ 0H)∗V = Xi,j≥1
αk
αn
ijxix∗
j = Tn.
(cid:3)
AN INFINITE QUANTUM RAMSEY THEOREM
9
3. Reduction to the diagonalizable case
For a Hilbert space H and a (not necessarily closed) subspace X ⊆ H,
we will write H ⊖ X for H ∩ X ⊥.
Lemma 3.1. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. Then there is an infinite rank projection
P ∈ B(H) satisfying one of the following conditions:
(1) For all nonzero vectors x ∈ P H, dim P H ⊖ P VP x < ∞.
(2) The compression P VP is diagonalizable.
In particular, if dim V < ∞, then there is an infinite rank projection
P ∈ B(H) such that the compression P VP is diagonalizable.
Proof. Suppose there is no infinite rank projection P ∈ B(H) satisfying
the first condition. Let P1 = IH . By assumption there is x1 ∈ H such
that dim H ⊖ VP1x1 = ∞. We inductively define a decreasing sequence
of infinite rank projections (Pn) in B(H) and unit vectors (xn) in H
such that for n ≥ 2, PnH = Pn−1H ⊖ Pn−1VPn−1xn−1 and xn ∈ PnH.
For k ≥ 2, suppose P1, . . . , Pk−1 and x1, . . . , xk−1 have been chosen.
Let Pk ∈ B(H) denote the projection onto Pk−1H ⊖ Pk−1VPk−1xk−1.
Then by assumption dim PkH = ∞ and there is xk ∈ PkH such that
dim PkH ⊖ PkVPkxk = ∞.
Proceeding inductively, we obtain the sequences (Pn) and (xn). We
claim hAxk, xli = 0 for A ∈ V and k, l ≥ 1 with k 6= l. To see this,
first note that since V is self-adjoint, we can assume k < l. Then by
construction, Pk ≥ Pl so xl = Plxl = Pkxl and xl ∈ PkH ⊖ PkVPkxk.
Hence
hAxk, xli = hAPkxk, Pkxli = hPkAPkxk, xli = 0.
It follows that if P is the projection onto the closure of span{xn : n ≥
1}, then the compression P VP is diagonalizable.
(cid:3)
Lemma 3.2. Let H be an infinite dimensional Hilbert space. Let X ⊆
H be a (potentially non-closed) subspace and let Y ⊆ H be a closed
subspace. Suppose dim X > dim Y ⊥ and dim Y ⊥ < ∞. Then X ∩ Y 6=
0.
Proof. Let n = dim Y ⊥ and let x1, . . . , xn+1 ∈ X be linearly inde-
pendent. For each k, write xi = yi + zi for yk ∈ Y and zk ∈ Y ⊥.
Since dim Y ⊥ = n, the vectors z1, . . . , zn+1 are linearly dependent,
and hence there are scalars α1, . . . , αn+1 ∈ C, not all zero, such that
α1z1 + · · · + αn+1zn+1 = 0. Since x1, . . . , xn+1 are linearly independent,
0 6= α1x1 + · · · + αn+1xn+1 ∈ Y .
(cid:3)
Lemma 3.3. Let V ⊆ B(H) be an operator system on an infinite di-
mensional Hilbert space H such that for all nonzero vectors h ∈ H,
10
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
dim Vh = ∞. Then there is a sequence (An) in V and an orthonor-
mal sequence (xn) in H such that for each n, hAnxn, xn+1i = 1 and
hAkxi, xji = 0 if max{i, j} > n + 1 and i 6= j.
Proof. We inductively construct the sequences (An) in V and (xn) in
H such that for n ≥ 1, Anxn = xn+1 and
xn+1 ∈ Vxn ∩ {xi, Ajxi, A∗
j xi : 1 ≤ i ≤ n and 1 ≤ j ≤ n − 1}⊥.
Let x1 ∈ H be any unit vector. Then by assumption dim Vx1 = ∞,
so by Lemma 3.2 there is a unit vector x2 ∈ Vx1 ∩ {x1}⊥. Choose
A1 ∈ V such that A1x1 = x2.
For k ≥ 2, suppose A1, . . . , Ak−1 and x1, . . . , xk have been chosen.
By assumption dim Vxk = ∞, so by Lemma 3.2 there is a unit vector
xk+1 ∈ Vxk ∩ {xi, Ajxi, A∗
j xi : 1 ≤ i ≤ k and 1 ≤ j ≤ k − 1}⊥.
Choose Ak ∈ V such that Akxk = xk+1.
Proceeding inductively, we obtain the sequences (An) and (xn). For
n ≥ 1, hAnxn, xn+1i = hxn+1, xn+1i = 1. Finally, suppose k, l ≥ 1 with
max{k, l} > n + 1 and k 6= l. If k < l then xl ∈ {Anxk}⊥. Otherwise,
if k > l, then xk ∈ {A∗
(cid:3)
nxl}⊥. Either way, hAnxk, xli = 0.
Lemma 3.4. Let V ⊆ B(H) be an operator system on a countably
infinite dimensional Hilbert space. Let (An) be a sequence in V and let
(hn) be an orthonormal basis for H such that for n ≥ 1, hAnhn, hn+1i =
1 and hAnhi, hji = 0 if max{i, j} > n + 1 and i 6= j. Suppose that for
all N ≥ 1, there is a nonzero finite rank operator in span{An : n ≥ N}.
Then there is a sequence (Bn) of self-adjoint operators in V and an
orthonormal sequence (fn) in H such that for n ≥ 1, hBnfn, fni = 1
and hBnfi, fji = 0 if max{i, j} > n.
Proof. We inductively construct the sequences (Bn) in V and (fn) in
H along with a strictly increasing sequence (Nn) in N such that Bn ∈
span{Aj, A∗
j : j > Nn}, fn ∈ span{hi : Nn−1 < i < Nn}, hBnfn, fni = 1
and hBnfi, fji = 0 if max{i, j} > n.
For N ≥ 1, let PN denote the projection onto span{hn : 1 ≤ n ≤ N}.
By assumption there is a nonzero finite rank operator F1 ∈ span{Aj :
j ≥ 1}. Choose N1 ≥ 1 such that F1 = PN1F1PN1.
If Re(F1) 6= 0,
then let B1 = Re(F1). Otherwise, let B1 = Im(F1). Then 0 6= B1 =
PN1B1PN1. Hence there is unit vector f1 ∈ H such that hB1f1, f1i 6= 0.
We can multiply B1 by a scalar if necessary to ensure that hB1f1, f1i =
1.
For k ≥ 2, suppose B1, . . . , Bk−1, f1, . . . , fk−1 and N1, . . . , Nk−1 have
been chosen. By assumption there is a nonzero finite rank operator
AN INFINITE QUANTUM RAMSEY THEOREM
11
Fk ∈ span{Aj : j > Nk−1}. If Re(Fk) 6= 0 then let Bk = Re(Fk). Oth-
erwise, let Bk = Im(Fk). Choose Nk ∈ N such that Bk = PNkBkPNk.
By construction, (PNk − PNk−1)Bk(PNk − PNk−1) 6= 0. Hence there is
a unit vector fk ∈ span{hi : Nk−1 < i ≤ Nk} such that hBkfk, fki 6= 0.
We can multiply Bk by a scalar if necessary to ensure that hBkfk, fki =
1.
Proceeding inductively, we obtain the sequences (Bn) and (fn). For
n ≥ 1, hBnfn, fni = 1. Furthermore, for k, l ≥ 1 with max{k, l} ≥ n,
fk, fl ⊥ PNnH. Since Bn = PNnBnPNn, it follows that hBnfk, fli =
0.
(cid:3)
The proof of the next lemma is inspired by the proof of Lemma
[12, Lemma 4.3].
Lemma 3.5. Let V ⊆ B(H) be an operator system on a countably
infinite dimensional Hilbert space. Let (An) be a self-adjoint sequence
of operators in V and let (hn) be an orthonormal basis for H such that
for n ≥ 1, hAnhn, hni = 1 and hAnhi, hji = 0 if max{i, j} > n. Then
V has an infinite quantum clique.
Proof. Let (en) be the standard orthonormal basis for ℓ2(N). For m ≥ 1
we can identify the m × m matrices Mm(C) with the set of operators
T ∈ B(ℓ2(N)) satisfying hT ei, eji = 0 if max{i, j} > m. For n ≥ 1 and
i, j ≥ 1, let αn
ij = hAnhj, hii, so that
An = X1≤i,j≤n
ijhih∗
αn
j ,
where the sum is taken in the weak topology. Note that since An is
self-adjoint, αn
ij = αn
ji.
We will inductively construct sequences (xn) in ℓ2(N) and (Tn) in
B(ℓ2(N)). For n ≥ 1, the operator Tn will be defined by
Tn = X1≤i,j≤n
ijxix∗
αn
j .
We will choose xn such that kxnk ≤ 2−(n+1) and xn ∈ span{ei : 1 ≤
i ≤ m} if (m − 1)2 < n ≤ m2. In addition, we will require that the
operators T1, . . . , Tn are linearly independent. This will imply that for
m ≥ 1, span{T1, . . . , Tm2} = Mm(C).
Let x1 = 2−2e1. For k ≥ 2, suppose x1, . . . , xk−1 have been chosen
and let m ≥ 1 satisfy (m − 1)2 < k ≤ m2. Suppose for the moment
that xk ∈ ℓ2(N) has been chosen. Let
y = X1≤i≤k−1
ikxi
αk
12
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
and let
T = X1≤i,j≤k−1
αk
ijxix∗
j − yy∗.
Then since αk
kk = 1,
Tk = X1≤i,j≤k−1
αk
ijxix∗
j − yy∗ + (y + xk)(y + xk)∗
= T + yy∗ + yx∗
k + xky∗ + xkx∗
k.
Since Mm(C) is spanned by positive rank one matrices and k ≤ m2,
there is z ∈ span{ei : 1 ≤ i ≤ m} such that zz∗ /∈ span{T1, . . . , Tk−1}.
It is easy to check there are at most finitely many α > 0 such that
T + yy∗ + αyz∗ + αzy∗ + α2zz∗ ∈ span{T1, . . . , Tk−1}.
Hence there is a choice of α > 0 such that for xk = αz, the operators
T1, . . . , Tk are linearly independent and kxkk ≤ 2−(k+1). Proceeding
inductively, we obtain the sequences (Tn) and (xn) in ℓ2(N) as required.
We now apply Lemma 2.3 to obtain an isometry V : ℓ2(N) → H ⊕ H
such that V ∗(An ⊕ 0H)V = Tn for n ≥ 1. By construction, span{Tn :
n ≥ 1} is weakly dense in B(ℓ2(N)). Hence by Lemma 2.2, we conclude
that V has an infinite quantum clique.
(cid:3)
Lemma 3.6. Let V ⊆ B(H) be an operator system on a countably
infinite dimensional Hilbert space. Let (An) be a linearly independent
sequence of operators in V and let (hn) be an orthonormal basis for
H such that for n ≥ 1 there is Nn ≥ 1 such that hAnhi, hji = 0
if max{i, j} > Nn and i 6= j. Suppose there are no nonzero finite
rank operators in span{An : n ≥ 1}. Then there is an infinite rank
projection P ∈ B(H) and an orthonormal basis (fn) for P H such that
the sequence of compressions (P AnP ) is linearly independent and for
n ≥ 1, hAnfi, fji = 0 if max{i3, j3} > n and i 6= j.
Proof. We can assume that the sequence (Nn) is strictly increasing.
We inductively construct a sequence (fn) in H and a sequence (Mn) of
indices such that for n ≥ 1, fn = hMn, Mn > Nn3 and the compressions
PnA1Pn, . . . , PnAnPn are linearly independent, where Pn denotes the
projection onto span{f1, . . . , fn}.
Since A1 has infinite rank there is M1 > N1 such that hA1hM1, hM1i 6=
0. Let f1 = hM1. For k ≥ 2 suppose f1, . . . , fk−1 and M1, . . . , Mk−1
have been chosen.
Suppose first there is an operator A ∈ span{A1, . . . , Ak−1} such that
Pk−1APk−1 = Pk−1AkPk−1. Since Pk−1A1Pk−1, . . . , Pk−1Ak−1Pk−1 are
linearly independent by the induction hypothesis, A is unique. Also,
since A1, . . . , Ak are linearly independent, A − Ak has infinite rank,
AN INFINITE QUANTUM RAMSEY THEOREM
13
so there is Mk > Nk3 such that h(A − Ak)hMk, hMki 6= 0. Then
hAhMk, hMki 6= hAkhMk, hMki. Otherwise, if there is no such operator
A ∈ span{A1, . . . , Ak−1}, then choose any Mk > Nk3. Either way, let-
ting fk = hMk and letting Pk denote the projection onto span{f1, . . . , fk},
it follows that PkA1Pk, . . . , PkAkPk are linearly independent. Proceed-
ing inductively, we obtain the sequence (hn).
For n ≥ 1 and i, j ≥ 1 with max{i3, j3} > n and i 6= j, it follows
from above that either Mi > Ni3 > Nn or Mj > Nj 3 > Nn. Hence
hAnfi, fji = hAnhMi, hMj i = 0.
We can take P to be the projection onto the closure of span{fn : n ≥
1}.
(cid:3)
Lemma 3.7. Let V ⊆ B(H) be an operator system on a countably
infinite dimensional Hilbert space. Let (An) be a linearly independent
sequence of operators in V and let (hn) be an orthonormal basis for
H such that for n ≥ 1, hAnhi, hji = 0 if max{i3, j3} > n and i 6= j.
Then there is a sequence (Bn) in V that is linearly independent and
simultaneously diagonalizable.
Proof. For n ≥ 1 we can identify the n × n matrices Mn(C) with the
set of operators T ∈ B(H) satisfying hT hi, hji = 0 if max{i, j} > n.
For n ≥ 2, the 3n2 − 3n + 1 operators A(n−1)3+1, . . . , An3 can each
be written as the sum of an operator in Mn(C) and an operator in
B(H) that is diagonal with respect to the orthonormal basis (hn).
Since 3n2 − 3n + 1 ≥ n2 = dim Mn(C), there is a nonzero operator
Bn−1 ∈ span{A(n−1)3+1, . . . , An3} that is diagonal with respect to (hn).
The linear independence of the sequence (An) implies the linear inde-
pendence of the sequence (Bn).
(cid:3)
Proposition 3.8. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space. Suppose there is no infinite rank projection
P ∈ B(H) such that the compression P VP is diagonalizable. Then
either V has an infinite quantum clique or there is an infinite rank
projection P ∈ B(H) and an operator system V ′ ⊆ V such that the
compression P V ′P is infinite dimensional and diagonalizable.
Proof. Suppose there is no infinite rank projection P ∈ B(H) such that
the compression P VP is diagonalizable. Then by Lemma 3.1 there is
an infinite rank projection P ∈ B(H) such that for every nonzero vector
x ∈ P H, dim P H ⊖ P VP x < ∞. By replacing V with P VP , we can
assume that for every nonzero vector x ∈ H, dim H ⊖ Vx < ∞. In
particular, dim Vx = ∞.
14
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
By Lemma 3.3, there is a sequence (An) in V and an orthonor-
mal sequence (xn) in H such that for n ≥ 1, hAnxn, xn+1i = 1 and
hAnxi, xji = 0 if max{i, j} > n + 1 and i 6= j.
Suppose first that for all N ≥ 1 there is a nonzero finite rank operator
in span{An : n ≥ N}. Then applying Lemma 3.4 followed by Lemma
3.5, we conclude that V has an infinite quantum clique.
Otherwise, by truncating (An), we can assume that span{An : n ≥ 1}
does not contain any nonzero finite rank operators. Then we can apply
Lemma 3.6, followed by Lemma 3.7 to obtain an infinite rank projection
P ∈ B(H) and a sequence (Cn) in V such that the sequence (P CnP )
In this case we can take
is linearly independent and diagonalizable.
V ′ = span{Cn : n ≥ 1}.
(cid:3)
4. Diagonalizable case
Lemma 4.1. Let H be a countably infinite dimensional Hilbert space
and let A1, . . . , An ∈ B(H) be self-adjoint and simultaneously diagonal-
izable. Then there is an infinite rank projection P ∈ B(H) and scalars
α1, . . . , αn ∈ R such that for k ≥ 1 the compression P (Ak − αkIH)P is
self-adjoint and compact.
Proof. First consider the case of a single self-adjoint diagonalizable op-
erator A ∈ B(H). Let (hn) be an orthonormal basis for H that diag-
onalizes A. For n ≥ 1, let αn = hAhn, hni. Since A is bounded, the
sequence (αn) is bounded. Hence there is a subsequence (αnk) con-
verging to α ∈ R. Let P denote the projection onto the closure of
span{hnk : k ≥ 1}. Then the compression P (A − αIH)P is a diagonal-
izable operator with diagonal entries converging to zero. In particular
it is compact. The case of multiple operators follows from an iteration
of the above argument.
(cid:3)
Lemma 4.2. Let V ⊆ B(H) be a finite dimensional diagonalizable
operator system on an infinite dimensional Hilbert space. Then either
V has an infinite quantum anticlique or it has an infinite quantum
obstruction.
Proof. Choose self-adjoint operators A1, . . . , An ∈ V such that V =
span{IH, A1, . . . , An}. By Lemma 4.1, there is an infinite rank pro-
jection P ∈ B(H) and scalars α1, . . . , αn ∈ R such that for k ≥ 1,
P (Ak − αkIH)P is self-adjoint and compact. Hence by replacing each
Ak with P (Ak − αkIh)P and replacing V with the compression P VP ,
we can assume that each Ak is compact.
Let (hn) be an orthonormal basis for H that simultaneously diago-
nalizes A1, . . . , An. If each Ak has finite rank, then for sufficiently large
AN INFINITE QUANTUM RAMSEY THEOREM
15
N the projection onto the closure of span{en : n ≥ N} is an infinite
quantum anticlique for V. Otherwise, if some Ak has infinite rank, then
the projection onto the closure of the range of Ak is an infinite quantum
obstruction for V.
(cid:3)
Lemma 4.3. Let V ⊆ B(H) be a simultaneously diagonalizable oper-
ator system on an infinite dimensional Hilbert space and let (An) be
a linearly independent sequence of operators in V. Then there is a se-
quence (Bn) of operators in V, an infinite rank projection P ∈ B(H)
and an orthonormal basis (fn) for P H such that the sequence of com-
pressions (P BnP ) is linearly independent and for n ≥ 1, hBnfi, fii = 0
for 1 ≤ i < n and hBnfn, fni = 1.
Proof. Let (hn) be an orthonormal basis for H that simultaneously
diagonalizes (An). We inductively construct sequences (Bk) and (fk)
and a sequence (nk) in N such that for k ≥ 1, Bk ∈ span{A1, . . . , Ak},
fk = hnk, hBkfi, fii = 0 for 1 ≤ i < k and hBkfk, fki = 1.
For k = 1 choose n1 ≥ 1 such that hA1hn1, hn1i 6= 0. Then we
can take B1 = A1/hA1hn1, hn1i and f1 = hn1. Suppose f1, . . . , fk−1
and B1, . . . , Bk−1 have been chosen. By the induction hypothesis and
the linear independence of the sequence (An) there is nonzero A ∈
span{A1, . . . , Ak} such that hAhni, hnii = hAfi, fii = 0 for 1 ≤ i <
k. Choose nk ∈ N such that hAhnk, hnki 6= 0. Then we can take
Bk = A/hAenk, enki and fk = hnk. Proceeding inductively, we obtain
the sequences (Bn) and (fn) as desired. We can take P to be the
projection onto the closure of span{fn : n ≥ 1}.
(cid:3)
Lemma 4.4. Let V ⊆ B(H) be an operator system on a countably
infinite dimensional Hilbert space. Let (An) be a linearly independent
sequence of simultaneously diagonalizable operators in V and let (hn) be
an orthonormal basis that simultaneously diagonalizes (An). Suppose
that for each n, hAnhi, hii = 0 for 1 ≤ i < n and hAnhn, hni = 1. Then
V has an infinite quantum clique.
Proof. Let (en) denote the standard orthonormal basis for the Hilbert
space ℓ2(N). We identify the m × m matrices Mm(C) with the set of
operators T ∈ B(ℓ2(N)) satisfying hT ei, eji = 0 for max{i, j} > m. For
n ≥ 1 and i ≥ 1, let αn
ii = hAnhi, hii, so that
αn
iihih∗
i ,
An = Xi≥1
where the sum is taken in the weak topology.
We will inductively construct a sequence (xn) of nonzero vectors
in ℓ2(N) satisfying very restrictive norm conditions. Also, for n ≥ 1
16
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
and m ≥ 1 satisfying (m − 1)2 < n ≤ m2, we will require that
xn ∈ span{e1, . . . , em}. Finally, we will require that the rank one op-
erators x1x∗
n are linearly independent and that the operators
An
1 , . . . , An
j is defined
by
1, . . . , xnx∗
n are linearly independent, where for 1 ≤ j ≤ n, An
j = X1≤i≤n
An
αj
iixix∗
i ∈ Mm(C).
For k = 1, we can take x1 = 2−2e1. For k ≥ 2, suppose x1, . . . , xk−1
have been chosen and let m ≥ 1 satisfy (m − 1)2 < k ≤ m2. Since
Mm(C) is spanned by positive rank one matrices, there is a vector xk ∈
span{e1, . . . , em} such that xkx∗
k−1}. Observe
that for 1 ≤ j ≤ k, Ak
By assumption, Ak−1
k−1 are linearly independent. Since small
perturbations of finitely many linearly independent matrices are lin-
early independent, we can scale xk if necessary to ensure that Ak
1, . . . , Ak
are linearly independent.
k. In particular, Ak
k /∈ span{x1x∗
1, . . . , xk−1x∗
k = xkx∗
k.
j = Ak−1
, . . . , Ak−1
j + αj
kkxkx∗
1
k−1
We claim that Ak
1, . . . , Ak
k are linearly independent. To see this, sup-
pose there are scalars β1, . . . , βk ∈ C such that Pk
kkxkx∗
k.
βjAk−1
k = −
j −
βjαj
xkx∗
k−1
k−1
j=1 βjAk
j = 0. Then
The linear independence of x1x∗
j = 0
kk = −1, which contradicts the linear independence of
j=1 βjAk−1
Xj=1
Xj=1
k implies Pk−1
1, . . . , xkx∗
1
By construction, for m ≥ 1, Am2
1 , . . . , Am2
s. Then there are scalars γrs,m2
1
m2 span Mm(C). For 1 ≤
m2 ∈ C
, . . . , γrs,m2
r, s ≤ m, let Ers = ere∗
such that
and Pk−1
j=1 βjαj
, . . . , Ak−1
k−1.
Ak−1
Ers =
m2
Xj=1
γrs,m2
j
Am2
j
.
For k > m2, we may further scale xk if necessary, while preserving the
linear independence of Ak
k, to ensure that
1, . . . , Ak
kxkk2 ≤
1
2k+1mPm2
j=1 γrs,m2
j
.
kAjk
Proceeding inductively, we obtain the sequence (xn) as desired.
By Lemma 2.2 there is an isometry V : ℓ2(N) → H ⊕ H such that
V ∗(Ak ⊕ 0H)V = Xi≥1
iixix∗
αk
i ,
AN INFINITE QUANTUM RAMSEY THEOREM
17
where the sum is taken in the weak topology. For m ≥ 1 and 1 ≤ r, s ≤
m, we estimate
m2
Xj=1
γrs,m2
j
Ers−
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Ers −
V ∗AjV(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
m2
γrs,m2
j
Xj=1
Am2
j −
m2
Xj=1
γrs,m2
j Xk>m2
= (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
= (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
m2
Xj=1
1
m
.
≤
≤
m2
Xj=1
γrs,m2
j Xk>m2
αj
kkxkx∗
k(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
γrs,m2
j
kAjk Xk>m2
kxkk2
αj
kkxkx∗
k(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
It follows that Ers belongs to the norm closure of V ∗VV . Hence
the weak closure of V ∗VV contains all of the finite rank operators
in B(ℓ2(N)).
In particular, it is weakly dense in B(ℓ2(N)). We now
conclude from Lemma 2.3 that V has an infinite quantum clique. (cid:3)
Proposition 4.5. Let H be a Hilbert space of countably infinite di-
mension and let V ⊆ B(H) be a diagonalizable operator system.
If
the dimension of V is finite, then either it has an infinite quantum an-
ticlique or it has an infinite quantum obstruction. Otherwise, if the
dimension of V is infinite, then it has an infinite quantum clique.
Proof. If the dimension of V is infinite, then applying Lemma 4.3 fol-
lowed by Lemma 4.4, we conclude that V has an infinite quantum
clique. Otherwise, if the dimension of V is finite, then Lemma 4.2
implies that either V has an infinite quantum anticlique or it has an
infinite quantum obstruction.
(cid:3)
5. Main results
Theorem 5.1. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. Then it has either an infinite quantum
clique, an infinite quantum anticlique or an infinite quantum obstruc-
tion.
Proof. By compressing to a subspace, we can assume that H is count-
ably infinite dimensional. If there is an infinite rank projection P such
that the compression P VP is diagonalizable, then we can replace V
18
M. KENNEDY, T. KOLOMATSKI, AND D. SPIVAK
with P VP and apply Proposition 4.5. Otherwise, by Proposition 3.8,
either V has an infinite quantum clique, or there is an infinite rank
projection P ∈ B(H) and an operator subsystem V ′ ⊆ V such that the
compression P V ′P is infinite dimensional and diagonalizable. In the
latter case, applying Proposition 3.8 gives the desired result.
(cid:3)
The following corollary is an immediate consequence of Definition 1.2,
Definition 1.4 and Theorem 5.1.
Corollary 5.2. Let V ⊆ B(H) be an operator system on an infinite
dimensional Hilbert space H. Then it has either an infinite quantum
clique or an infinite quantum almost anticlique.
References
[1] C. B´eny, A. Kempf, and D.W. Kribs, Quantum error correction on infinite-
dimensional Hilbert spaces, Journal of Mathematical Physics 50 (2009), 062108.
[2] R. Duan, S. Severini, and A. Winter, Zero-error communication via quantum
channels, noncommutative graphs, and a quantum Lov´asz number, IEEE Trans-
actions on Information Theory 59 (2013), 1164-1174.
[3] S. Kim and A. Mehta, Chromatic numbers and a Lov´asz type inequality for
non-commutative graphs, arXiv preprint arXiv:1709.05595 (2017).
[4] E. Knill and R. Laflamme, Theory of quantum error-correcting codes, Physical
Review A 55 (1997), 900-911.
[5] E. Knill, R. Laflamme, and L. Viola, Theory of quantum error correction for
general noise, Physical Review Letters 84 (2000), 2525-2528.
[6] R.H. Levene, V.I. Paulsen, and I.G. Todorov, Complexity and capacity bounds
for quantum channels, arXiv preprint arXiv:1710.06456 (2017).
[7] V. Paulsen, Completely bounded maps and operator algebras, Vol. 78, Cam-
bridge University Press, 2002.
[8] V.I. Paulsen and C.M. Ortiz, Lov´asz theta type norms and operator systems,
Linear Algebra and its Applications 477 (2015), 128-147.
[9] F.P. Ramsey, On a problem of formal logic, Proceedings of the London Math-
ematical Society 30 (1930), 264-286.
[10] Stahlke, Quantum zero-error source-channel coding and non-commutative
graph theory, IEEE Transactions on Information Theory 62 (2016), 554-577.
[11] N. Weaver, Quantum graphs as
quantum relations,
arXiv preprint
arXiv:1506.03892 (2015).
[12]
, A "quantum" Ramsey theorem for operator systems, Proceedings of
the American Mathematical Society 145 (2017), 4595-4605.
AN INFINITE QUANTUM RAMSEY THEOREM
19
Department of Pure Mathematics, University of Waterloo, Water-
loo, Ontario, N2L 3G1, Canada
E-mail address: [email protected]
Department of Pure Mathematics and Mathematical Statistics, Uni-
versity of Cambridge, Cambridge, CB3 0WA, United Kingdom
E-mail address: [email protected]
Department of Mathematics, University of Toronto, Toronto, On-
tario, M5S 2E4, Canada
E-mail address: [email protected]
|
1807.08665 | 1 | 1807 | 2018-07-23T15:13:36 | Generalized gauge actions on $k$-graph $C^*$-algebras: KMS states and Hausdorff structure | [
"math.OA",
"math.MG"
] | For a finite, strongly connected $k$-graph $\Lambda$, an Huef, Laca, Raeburn and Sims studied the KMS states associated to the preferred dynamics of the $k$-graph $C^*$-algebra $C^*(\Lambda)$. They found that these KMS states are determined by the periodicity of $\Lambda$ and a certain Borel probability measure $M$ on the infinite path space $\Lambda^\infty$ of $\Lambda$. Here we consider different dynamics on $C^*(\Lambda)$, which arise from a functor $y: \Lambda \to \mathbb{R}_+$ and were first proposed by McNamara in his thesis. We show that the KMS states associated to McNamara's dynamics are again parametrized by the periodicity group of $\Lambda$ and a family of Borel probability measures on the infinite path space. Indeed, these measures also arise as Hausdorff measures on $\Lambda^\infty$, and the associated Hausdorff dimension is intimately linked to the inverse temperatures at which KMS states exist. Our construction of the metrics underlying the Hausdorff structure uses the functors $y: \Lambda \to \mathbb{R}_+$; the stationary $k$-Bratteli diagram associated to $\Lambda$; and the concept of exponentially self-similar weights on Bratteli diagrams. | math.OA | math |
GENERALIZED GAUGE ACTIONS ON k-GRAPH
C ∗-ALGEBRAS: KMS STATES AND HAUSDORFF
STRUCTURE
CARLA FARSI, ELIZABETH GILLASPY, NADIA S. LARSEN AND JUDITH A.
PACKER
Abstract. For a finite, strongly connected k-graph Λ, an Huef, Laca,
Raeburn and Sims studied the KMS states associated to the preferred
dynamics of the k-graph C ∗-algebra C ∗(Λ). They found that these KMS
states are determined by the periodicity of Λ and a certain Borel proba-
bility measure M on the infinite path space Λ∞ of Λ. Here we consider
different dynamics on C ∗(Λ), which arise from a functor y : Λ → R+
and were first proposed by McNamara in his thesis. We show that the
KMS states associated to McNamara's dynamics are again parametrized
by the periodicity group of Λ and a family of Borel probability mea-
sures on the infinite path space. Indeed, these measures also arise as
Hausdorff measures on Λ∞, and the associated Hausdorff dimension is
intimately linked to the inverse temperatures at which KMS states exist.
Our construction of the metrics underlying the Hausdorff structure uses
the functors y : Λ → R+; the stationary k-Bratteli diagram associated
to Λ; and the concept of exponentially self-similar weights on Bratteli
diagrams.
1. Introduction
KMS states have their origin in equilibrium statistical mechanics and
have long been a very fruitful tool in the study of operator algebras. In this
paper, we identify links between KMS states on C ∗-algebras of higher rank
graphs, and the Hausdorff measure and Hausdorff dimension associated to
ultrametrics on Bratteli diagrams that exhibit a certain self-similarity.
Given a C ∗-algebra A with a one-parameter group of automorphisms
(γt)t∈R, a state φ on A satisfies the Kubo-Martin-Schwinger (KMS) condition
at inverse temperature β ∈ R if φ(ab) = φ(bγiβ(a)) for all analytic elements
a, b ∈ A, where an element x ∈ A is analytic if the function t 7→ γt(x)
extends to an entire function. Some fundamental examples of dynamical
systems (A, γ) with a unique KMS state at a distinguished inverse tem-
perature β are the Cuntz algebras On with the gauge action, for n ≥ 1,
where β = log n, see [OP78], and (when A is an irreducible matrix) the
Cuntz-Krieger algebras OA with the gauge action, where β = log ρ(A) with
ρ(A) the spectral radius of A, see [EFW84]. This last result was generalized
by Exel in [Exe04], and Exel and Laca in [EL03], where they considered
generalized gauge actions γ on Cuntz-Krieger algebras. Again, if A is an ir-
reducible matrix, the dynamical system (OA, γ) admits a unique KMS state.
Date: 18 July 2018.
1
2
FARSI, GILLASPY, LARSEN AND PACKER
Links between these unique KMS states and harmonic measures were estab-
lished by Okayasu in [Oka02, Oka03]. From a quite different perspective, it
is natural to ask what are the possible inverse temperatures of dynamical
systems, and in this respect Bratteli-Elliott-Herman [EBH80] constructed
simple C ∗-algebras A and associated dynamics γ which attain any closed
subset of R as a possible range of inverse temperatures for KMSβ states.
Recently there has been great interest in KMS states for gauge actions
on higher-rank graph C ∗-algebras, which are a generalization of Cuntz-
Krieger algebras: see for example [Yan10, Yan12, Yan17, HLRS14, HLRS15,
LLN+15, Chr17]. The first main aim of the present paper is the analysis, for
the C ∗-algebras of finite, strongly connected higher-rank graphs, of the KMS
states associated to the generalized gauge actions on higher-rank graph C ∗-
algebras which were first introduced by McNamara in his thesis [McN15].
Proposition 4.5 identifies the inverse temperatures β such that these actions
admit KMSβ states, while Theorem 4.8 describes the KMS states.
Towards explaining this aim in more detail, let us first recall that Kumjian
and Pask introduced higher-rank graphs (or k-graphs) and their C ∗-algebras
in [KP00] as a simultaneous generalization of the higher-rank Cuntz-Krieger
algebras of Robertson and Steger [RS99] and the C ∗-algebras of directed
graphs. In addition to their graph-theoretical description, the C ∗-algebras
associated to higher-rank graphs also admit a groupoid description as well as
a universal presentation in terms of generators and relations. This flexibility
has led to applications of k-graph C ∗-algebras in a variety of contexts (such
as the question of nuclear dimension for Kirchberg algebras [RSS15] and K-
theory computations for quantum spheres [HNP+]) and has also facilitated
the analysis of structural properties of k-graph C ∗-algebras. For example,
the ideal structure of k-graph C ∗-algebras C ∗(Λ) [RSY03, RS07, CKSS14] is
completely determined by the underlying higher-rank graph Λ, whereas the
groupoid perspective enabled the characterization of Cartan subalgebras of
C ∗(Λ) [BNR+16].
In the groupoid perspective, as explained by Renault already in [Ren80],
time evolutions (dynamics) on the C ∗-algebra of a groupoid G are imple-
mented by continuous cocycles on G, and the task of understanding the
KMS states on C ∗(G) requires, at a minimum, identifying the quasi-invariant
measures on the unit space of G. There are now refinements of Renault's
result, see for example [Nes13, Tho14, Chr17]. In particular, Christensen's
recent preprint [Chr17] combines quasi-invariant measures with a certain
group of symmetries to describe KMS states on groupoid C ∗-algebras. This
perspective is particularly well suited to our case of interest, namely, the
KMS states associated to generalized gauge actions on k-graph C ∗-algebras.
Much of the structural analysis of k-graph C ∗-algebras C ∗(Λ) is facilitated
by the gauge action, a natural action of Tk on C ∗(Λ); the existing literature
on KMS states for k-graph C ∗-algebras is no exception. Restricting the
gauge action to a subgroup R ∼= R of Tk gives rise to a dynamics, that
is, a one-parameter action α of the real line on C ∗(Λ). The KMS states
and the possible range of inverse temperatures for the dynamical system
(C ∗(Λ), α) carry interesting information about the underlying k-graph Λ,
cf. [Yan12, HLRS14, HLRS15, LLN+15, FGKP16, Yan17]. In particular, the
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
3
analysis of [HLRS15] links the KMS states associated to the gauge action
with the simplicity of C ∗(Λ):
for finite, strongly connected k-graphs Λ,
simplicity of C ∗(Λ) is equivalent to the existence of a unique KMSβ state at
the inverse temperature β = 1, and moreover to triviality of the periodicity
group Per Λ (which is defined in terms of shift invariant infinite paths in the
k-graph; see Definition 3.8).
For the method developed in [HLRS15] to classify KMS states associ-
ated to dynamics of the form α described in the previous paragraph, a
key ingredient is the construction of a certain Borel probability measure M
on the infinite path space Λ∞ of Λ. This measure M is also intrinsically
linked to both the fractal geometry and the noncommutative geometry of
Λ∞ [PB09, FGJ+18b]. In a different but related development, Ionescu and
Kumjian established connections between KMS states and Hausdorff struc-
ture for certain Renault-Deaconu groupoid C ∗-algebras in [IK13]. Their
results apply in particular to the C ∗-algebras associated to directed graphs
(which are k-graphs for k = 1) equipped with a generalized gauge dynamics.
In this paper, we extend and sharpen these results for finite, strongly
connected k-graphs, that is, k-graphs Λ such that there are finitely many
vertices in Λ, and the set vΛw of paths with source w and range v is finite and
nonempty for each pair (v, w) of vertices. We first analyze the KMS states
on C ∗(Λ) for the generalized gauge dynamics αy,θ introduced by McNamara
in [McN15], where θ is a positive real number and y is an R+-functor, or
weight functor as defined by McNamara in [McN15]. Very simple examples
involving a 2-graph with a single vertex (cf. Section 5.1 below) show that
there are multitudes of choices of R+-functors, and it is this flexibility we
want to explore more closely.
McNamara's thesis [McN15] characterizes the KMS states of (C ∗(Λ), αy,θ)
under the additional hypothesis that each of the coordinate matrices {Ai}k
i=1
of Λ (defined in Equation (2.1)) is irreducible. This hypothesis implies,
but is strictly stronger than, our standing hypothesis that Λ be finite and
strongly connected; see [HLRS15, Lemma 4.1 and Example 4.3]. The ac-
tions αy,θ are constructed using a new family of matrices {Bi(y, θ)}k
i=1 that
takes into account the coordinate matrices A1, . . . , Ak as well as y and θ.
Lemma 3.3 follows [HLRS15, Proposition 3.1] to establish that when Λ is fi-
nite and strongly connected, the matrices Bi(y, θ) admit a common positive
eigenvector which is unique up to scaling; we call it the Perron-Frobenius
eigenvector of the family {Bi(y, θ)}k
i=1. We prove in Theorem 3.6 that for a
fixed y, the spectral radii and the Perron-Frobenius eigenvector of the fam-
ily B1(y, θ), . . . , Bk(y, θ) vary smoothly with θ. This is a new sort of insight
that would not have been available via considering only the gauge dynamics
and its variations, and we use it to prove that in some cases the unique
inverse temperature β > 0 at which (C ∗(Λ), αy,θ) admits KMSβ states is
precisely θ, see Propositions 4.6 and 4.7.
As hinted above, our analysis of the KMS states of (C ∗(Λ), αy,θ) comes in
two steps: first we describe the unique quasi-invariant measure µy,θ associ-
ated to αy,θ (Proposition 3.5 and Remark 4.3), and then Theorem 4.8 shows
that Christensen's group of symmetries for this dynamics is precisely Per Λ.
4
FARSI, GILLASPY, LARSEN AND PACKER
We also show (see Proposition 4.5) that the dynamical system (C ∗(Λ), αy,θ)
admits KMSβ states iff αy,β = αy,θ.
One immediate consequence of Theorem 4.8 is a new proof of the struc-
tural result from [HLRS15] that identifies existence of a unique KMSβ state
with simplicity of C ∗(Λ) and aperiodicity of Λ. Moreover, if Λ has irreducible
coordinate matrices, we show that a certain technical condition (which arose
already in [McN15]) gives rise to a criterion for the aperiodicity of Λ that
may prove versatile in applications; see Corollary 4.11.
Given the structural similarity between the KMS states associated to αy,θ,
and those which are tied to the actions α already studied by [HLRS15], we
pause to reassure the reader that our added generality does indeed give
rise to new examples of actions and measures. To that end, in Section 5
we study two higher-rank graphs which, while simple to draw, admit R+-
functors y leading to a diverse family of measures µy,θ and actions αy,θ.
Their periodicity groups are also described, leading to a complete picture of
the associated KMS states.
Inspired by Ionescu and Kumjian [IK13], we then proceed to our second
main aim of the paper; namely, we relate these KMS states to Hausdorff
structures on Λ∞. The same data of an R+-functor y and a positive real
number θ which gives rise to the generalized gauge actions αy,θ also leads to
an ultrametric dy,θ on Λ∞ (Proposition 6.9). The construction of the ultra-
metric dy,θ uses the stationary k-Bratteli diagram associated to Λ, which was
introduced in [FGJ+18b], as well as a new concept, that of an exponentially
self-similar weight on a Bratteli diagram (Definition 6.12).
The final main result of the paper is Corollary 6.17, which proves that
the Hausdorff dimension of (Λ∞, dy,θ) is θ and the associated Hausdorff
measure is µy,θ. In fact, we establish a result about Hausdorff dimension in
a greater generality involving weights on Bratteli diagrams with a certain
self-similarity property; see Theorem 6.16.
Acknowledgments. C.F. and J.P. were partially supported by two indi-
vidual grants from the Simons Foundation (C.F. #523991; J.P. #316981).
E.G. was partially supported by the Deutsches Forschungsgemeinschaft via
the SFB 878 "Groups, Geometry, and Actions" of the Universitat Munster.
The authors heartily thank Sooran Kang for a number of insightful con-
versations throughout the course of this work, and in particular for bringing
McNamara's thesis [McN15] to our attention. We also thank A. Sims for en-
couraging us, after the second author's talk at the workshop "Applications
of operator algebras: order, disorder and symmetry" at ICMS Edinburgh
during June 2017, to consider general finite, strongly connected k-graphs in
our investigation. Finally, thanks to J. Christensen for alerting us to his
recent preprint [Chr17], which was instrumental in proving Theorem 4.8.
The research presented here was completed in large part during visits of
E.G. and C.F. to Oslo and of N.L. to Boulder; we thank the host universities
and their Operator Algebras research groups for their hospitality.
2. Preliminaries on higher-rank graphs
We begin by fixing the notation which will be used throughout the paper.
For a matrix A, we write ρ(A) for its spectral radius. By N we denote the
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
5
monoid of natural numbers {0, 1, 2 . . . } under addition. For k ∈ N, k ≥ 1,
we let Nk be the monoid of k-tuples of natural numbers under addition, with
standard basis vectors ei for i = 1, . . . , k. If γ ∈ Rk, we will write γi ∈ R to
denote the ith component of γ. In other words, we have γ =Pk
i=1 γiei. If
γ is a vector in Rk and m ∈ Nk, we write γm for the productQk
i=1 γmi
As a category, Nk has one object (namely 0), and composition of mor-
phisms n, m ∈ Nk is given by addition. In keeping with the use of n ∈ Nk
to denote a morphism in the category Nk, for any category Λ, we write
.
i
λ ∈ Λ ⇔ λ ∈ Mor(Λ).
We recall the basic facts of the construction, due to Kumjian and Pask,
of k-graphs and their C ∗-algebras [KP00]. Let k ∈ N, k ≥ 1. A k-graph
Λ consists of a countable small category and a degree functor d : Λ → Nk,
i.e. a map such that d(ην) = d(η) + d(ν) for all η, ν ∈ Λ, satisfying the
following factorization property: whenever d(λ) = m + n for λ ∈ Λ and
some m, n ∈ Nk, there are unique η, ν ∈ Λ such that λ = ην, d(η) = m and
d(ν) = n.
We denote by Λn the set of morphisms (or paths) of degree n ∈ Nk. Thus
Λ0 is the set of identity morphisms; these are referred to as the vertices of Λ.
We also identify Λ0 with the set of objects of Λ, and so the codomain and
domain maps become functions r, s : Λ → Λ0. In this paper, we consider
only finite k-graphs, meaning that Λn < ∞ for all n ∈ Nk. For v, w ∈ Λ0
and n ∈ Nk, denote by vΛnw the set of λ ∈ Λn such that s(λ) = w and
r(λ) = v.
It is often helpful to think of Λei as the "edges of color i" in Λ. With
this perspective, the factorization property implies that (for i 6= j) any
morphism λ ∈ Λei+ej can be represented either as a color-i edge followed by
a color-j edge, or a color-j edge followed by a color-i edge. In other words,
(cf. [HRSW13]) a k-graph can be described by a directed graph with k colors
of edges, together with a pairing which identifies each color-i -- color-j path
with a unique color-j -- color-i path.
The infinite path space Λ∞ of Λ formally consists of degree-preserving
functors from the k-graph Ωk into Λ. The category Ωk has object set Nk
and morphism set
Mor(Ωk) = {(m, n) ∈ Nk × Nk m ≤ n};
its structure maps are given by r(m, n) = m, s(m, n) = n, and the composi-
tion rule is (m, n)(n, p) = (m, p). To view Ωk as a k-graph, we equip it with
the degree functor d : Ωk → Nk given by d(m, n) = n− m. The infinite path
space comes equipped with a family of shift maps {σj j ∈ Nk}, given by
σj(x)(m, n) := x(m + j, n + j).
Observe that, since Ωk has a terminal object (namely 0 ∈ Nk) but no
initial object, our infinite paths will have a range but no source. Moreover,
the fact that Ωk contains infinitely many morphisms of each degree implies
that the same is true about every infinite path.
In particular, using the
"edge-colored directed graph" perspective on k-graphs, every infinite path
x ∈ Λ∞ contains infinitely many edges of each color.
6
FARSI, GILLASPY, LARSEN AND PACKER
We view Λ∞ as being topologized by the cylinder sets {Z(λ)}λ∈Λ, where
Z(λ) = {x ∈ Λ∞ : x(0, d(λ)) = λ}
is the collection of infinite paths with initial segment λ. The topology on
Λ∞ whose basic open sets are {Z(λ)}λ∈Λ is locally compact and Hausdorff;
indeed, each cylinder set Z(λ) is compact (as well as open) in the cylinder
set topology. If Λ is finite then Λ∞ is compact.
The coordinate matrices A1, . . . , Ak ∈ MatΛ0(N) of Λ are given by
Ai(v, w) = vΛeiw
(2.1)
for v, w in Λ0. By the factorization property, the matrices Ai pairwise com-
mute. For n ∈ Nk, we define
(2.2)
Note that An(v, w) = vΛnw for all v, w. The family {A1, . . . , Ak} is an
irreducible family of matrices, cf. [HLRS15, Section 3], if each Ai is nonzero
and there is a finite set F ⊂ Nk such that the matrix
An :=Qk
i=1 Ani
i .
AF :=Xn∈F
An
is positive; explicitly, for every pair (v, w) ∈ Λ0 × Λ0 we have AF (v, w) > 0.
A k-graph Λ is said to be strongly connected if vΛw 6= ∅ for all v, w ∈ Λ0.
The following characterization then holds.
Lemma 2.1. ([HLRS15, Lemma 4.1]) For a finite k-graph the following are
equivalent:
(i) Λ is strongly connected.
(ii) {A1, . . . , Ak} is an irreducible family of matrices.
(iii) There exists a finite set F ⊂ Nk such that for all v, w ∈ Λ0 there is
λ ∈ Λ with d(λ) ∈ F such that s(λ) = w and r(λ) = v.
The k-graph algebra C ∗(Λ) of a (finite) k-graph Λ is the C ∗-algebra that
λsλ = ss(λ) for all λ ∈ Λ;
is universal for families {sλ λ ∈ Λ} satisfying
(CK1) {sv v ∈ Λ0} is a family of mutually orthogonal projections;
(CK2) sλsν = sλν whenever s(λ) = r(ν);
(CK3) s∗
(CK4) For any v ∈ Λ0 and any n ∈ Nk, we have sv =Pλ∈vΛn sλs∗
Λmin(α, β) := {(ξ, ζ) ∈ Λ × Λ αξ = βζ and d(αξ) = d(α) ∨ d(β)}.
Relations (CK4) and (CK3) enable us to rewrite any element s∗
as a finite sum of terms of the form sλs∗
η; to be precise, let
λ.
αsβ ∈ C ∗(Λ)
Here d(α) ∨ d(β) denotes the coordinatewise maximum of d(α), d(β) ∈ Nk.
Moreover, relation (CK4) combines with the fact that each sα is a partial
isometry to tell us thatPρ:α6=ρ∈r(α)Λd(α) sρs∗
ρsα = 0. Consequently,
sξs∗
ξs∗
αsβsζs∗
ζ
s∗
αsβ =
(2.3)
ξ∈s(α)Λd(α)∨d(β)−d(α),ζ∈s(β)Λd(α)∨d(β)−d(β)
= X(ξ,ζ)∈Λmin(α,β)
sξs∗
αξsβζs∗
ζ = X(ξ,ζ)∈Λmin(α,β)
sξs∗
ζ.
X
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
7
η λ, η ∈ Λ} densely spans C ∗(Λ).
We can also view C ∗(Λ) as a groupoid C ∗-algebra [KP00, Corollary 3.5(i)].
In other words, {sλs∗
Namely, if we set
GΛ := {(x, n, y) ∈ Λ∞×Zk×Λ∞ ∃ j, ℓ ∈ Nk s.t. σj(x) = σℓ(y) and n = j−ℓ},
then the isomorphism C ∗(Λ) ∼= C ∗(GΛ) identifies sλs∗
µ ∈ C ∗(Λ) with the
characteristic function χZ(λ,µ) ∈ Cc(GΛ) ⊆ C ∗(GΛ), where
Z(λ, µ) = {(x, d(λ) − d(µ), y) ∈ GΛ x ∈ Z(λ), y ∈ Z(µ)}.
The sets {Z(λ, µ) s(λ) = s(µ)} constitute a compact open basis for the
topology on GΛ, which is an ample groupoid.
Observe that Λ∞ ∼= {(x, 0, x) x ∈ Λ∞} is the unit space of GΛ; therefore,
C(Λ∞) can be viewed as a subalgebra of C ∗(GΛ) ∼= C ∗(Λ). The identification
χZ(λ,µ) ↔ sλs∗
µ, and the fact that the cylinder sets Z(λ) are compact and
open in Λ∞, allows one to see that C(Λ∞) ∼= span{sλs∗
λ λ ∈ Λ}. Indeed,
we have a canonical conditional expectation Ψ : C ∗(Λ) → C(Λ∞) which is
given on the generators by Ψ(sλs∗
µ) = δλ,µsλs∗
λ.
3. R+-functors and measures on the infinite path space
The following definition is based on thinking of the non-negative real num-
bers as a category, with one object (namely 0) and composition of morphisms
given by addition. Throughout the paper, we will write
and R>0 := (0,∞).
R+ := [0,∞)
Definition 3.1 ([McN15]). Let Λ be a higher-rank graph. A R+-functor
(called a weight functor in [McN15, Section 5.3]) on Λ is a function y : Λ →
[0,∞) such that
• y(v) = 0 for every v ∈ Λ0, and
• y(λν) = y(λ) + y(ν) whenever s(λ) = r(ν).
We next recall another definition due to McNamara which constructs
generalized coordinate matrices from a R+-functor and a nonnegative pa-
rameter.
Definition 3.2. ([McN15, Definition 5.10]) Let Λ be a finite k-graph. For
y : Λ → [0,∞) a R+-functor and θ ∈ [0,∞), define matrices Bi(y, θ) for
each i = 1, . . . , k by
(3.1)
Bi(y, θ)v,w := Xλ∈vΛei w
e−θy(λ).
It is established in [McN15, Lemma 5.11] that for any choice of y and θ, the
matrices {Bi(y, θ)}i=1,...,k pairwise commute. Further, in [McN15, Lemma
5.13] McNamara proves that whenever all of the matrices A1, . . . , Ak are
irreducible, so are all the matrices Bi(y, θ).
We next notice that if Λ is a finite strongly connected graph, then for
every choice of R+-functor y and θ in [0,∞), the matrices {Bi(y, θ)}i=1,...,k
form an irreducible family.
8
FARSI, GILLASPY, LARSEN AND PACKER
Lemma 3.3. Let Λ be a strongly connected k-graph, y : Λ → [0,∞) a R+-
functor and θ ∈ [0,∞). Then {Bi(y, θ)}i=1,...,k is an irreducible family of
matrices.
Proof. For each n ∈ Nk we have (cf. [McN15, Definition 5.12]) a new Λ0× Λ0
matrix:
(3.2)
B(y, θ)n :=
Bi(y, θ)ni.
kYi=1
The fact that the matrices Bi(y, θ) pairwise commute implies that this def-
inition is independent of our choice of ordering on the generators of Nk.
Since y is additive, the (v, w) entry in B(y, θ)n is given by
(3.3)
B(y, θ)n
e−θy(λ).
v,w = Xλ∈vΛnw
B(y, θ)F :=Xn∈F
B(y, θ)n.
Let F ⊂ Nk be the finite set of degrees given by Lemma 2.1. Similar to the
definition of AF , but using an upper subscript to ease the notation, let
(cid:3)
It suffices to prove that B(y, θ)F is positive, that is, for each v, w in Λ0 we
have B(y, θ)F
v,w > 0. By construction,
B(y, θ)F
v,w =Xn∈F Xλ∈vΛnw
e−θy(λ)
and by our assumption,Pn∈F vΛnw = AF (v, w) 6= 0.
vector ξy,θ ∈ (R>0)Λ0
v = 1 and
withPv∈Λ0 ξy,θ
Bi(y, θ)ξy,θ = ρ(Bi(y, θ))ξy,θ
for all 1 ≤ i ≤ k,
Consequently, [HLRS15, Proposition 3.1] implies that there is a unique
where ρ(Bi(y, θ)) > 0 is the spectral radius of the matrix Bi(y, θ).
Definition 3.4 (Notation). Throughout this paper, we will write
ρ(B(y, θ)) := (ρ(B1(y, θ)), . . . , ρ(Bk(y, θ))).
i=1 ρ(Bi(y, θ))mi.
For m ∈ Zk, we let ρ(B(y, θ))m denote the productQk
Measures on the infinite path space of a finite, strongly connected k-graph
have recently gained attention starting with the construction in [HLRS15,
Proposition 8.1] that was motivated by the study of KMS states, and contin-
ued by constructions relating to Hausdorff and Markov measures in [FGJ+17,
FGJ+18a]. Here we construct a Borel probability measure on Λ∞ for any
R+-functor and parameter θ ∈ [0,∞).
Proposition 3.5. Let Λ be a finite, strongly connected k-graph, y a R+-
functor on Λ, and θ ∈ [0,∞). The measure µy,θ on Λ∞ given by
(3.4)
is the unique measure µ on Λ∞ such that µ(Z(v)) > 0 for all v ∈ Λ0 and
(3.5)
µy,θ(Z(λ)) = e−θy(λ)ρ(B(y, θ))−d(λ)ξy,θ
s(λ)
µ(Z(λ)) = e−θy(λ)ρ(B(y, θ))−d(λ)µ(Z(s(λ)).
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
9
Moreover, the measures µy,θ are all probability measures.
Proof. The fact that y(v) = 0 whenever v ∈ Λ0 implies that µy,θ(Z(v)) =
ξy,θ
v > 0 for all v ∈ Λ0, and that µy,θ(Z(v)) satisfies Equation (3.5). Since
µy,θ(Λ∞) = Xv∈Λ0
µy,θ(Z(v)) = Xv∈Λ0
ξy,θ
v = 1,
showing that µy,θ is indeed a measure will imply that µy,θ is a probability
measure.
To establish the claim that µy,θ is a measure, we first observe that it is
finitely additive on cylinder sets Z(λ) with d(λ) = (n, . . . , n) ∈ Nk for some
n ∈ N. To see this, it suffices to show that
µy,θ(Z(v)) = Xλ∈vΛ(n,...,n)
µy,θ(Z(λ))
for any n ∈ N. Observe that
Xλ∈vΛ(n,...,n)
µy,θ(Z(λ)) = ρ(B(y, θ))−(n,...,n) Xw∈Λ0
= ξy,θ
v = µ(Z(v))
B(y, θ)(n,...,n)(v, w)ξy,θ
w
by the definition of the matrices Bi(y, θ) and their common eigenvector ξy,θ.
In other words, µy,θ is indeed finitely additive on square cylinder sets.
The fact that µy,θ defines a measure on Λ∞ now follows from the Kol-
mogorov Extension Theorem (cf. Lemma 2.12 of [FGJ+18a]).
Suppose now that a measure µ satisfies Equation (3.5). Without loss of
+ by
generality, we will further assume that µ(Λ∞) = 1. Define m ∈ RΛ0
m(v) = µ(Z(v)). For any 1 ≤ i ≤ k, Equation (3.5) implies that
m(v) = Xλ∈vΛei
µ(Z(λ)) = Xλ∈vΛei
= ρ(Bi(y, θ))−1 Xw∈Λ0
Bi(y, θ)v,wm(w).
e−θy(λ)ρ(Bi(y, θ))−1m(s(λ))
that m = ξy,θ, so
In other words, m ∈ RΛ0
+ is a positive eigenvector for each matrix Bi(y, θ),
with eigenvalue ρ(Bi(y, θ)). Moreover, our hypothesis that µ(Λ∞) = 1 im-
plies that Pv∈Λ0 m(v) = 1. Proposition 3.1 of [HLRS15] therefore implies
for all v ∈ Λ0. Equation (3.5) now implies that µ(Z(λ)) = µy,θ(Z(λ)) for all
λ ∈ Λ. Since the topology (and the associated Borel structure) of Λ∞ are
generated by the cylinder sets Z(λ), it follows that µ = µy,θ.
µ(Z(v)) = µy,θ(Z(v))
(cid:3)
Theorem 3.6. Let Λ be a finite, strongly connected k-graph, y a R+-functor
on Λ, and θ ∈ [0,∞). The common eigenvector ξy,θ for the matrices Bi(y, θ),
as well as the spectral radii {ρ(Bi(y, θ))}k
i=1, depend smoothly on the entries
of the matrix Bi(y, θ). In particular, the spectral radii and the eigenvector
ξy,θ vary smoothly with θ.
10
FARSI, GILLASPY, LARSEN AND PACKER
Proof. Inspired by [Yeo15], we use the Implicit Function Theorem. Suppose
that Λ has n vertices. Since Λ is strongly connected, we know from Lemma
3.3 that there is a finite subset F ⊆ Nk (which does not depend on y or
θ) such that B(y, θ)F is a positive matrix. Moreover, ξy,θ is the unique
eigenvector for B(y, θ)F with eigenvalue ρ(B(y, θ)F ) and ℓ1-norm 1.
Observe first that there must exist at least one i, 1 ≤ i ≤ k, such that for
some f ∈ F we have fi 6= 0. (If this assertion were false then B(y, θ)F = I
would be the identity matrix, which is not positive.) By re-ordering the
indices i if necessary, we will assume that
(3.6)
To a vector ~x in Rkn2
,
∃ f ∈ F : f1 6= 0.
~x = (~x1, . . . , ~xk),
~xℓ = (xℓ
we associate the matrices Xℓ := (xℓ
a family X1, . . . , Xk of matrices, we define
1n, xℓ
21, . . . , xℓ
11, . . . , xℓ
n1, . . . , xℓ
ij)i,j ∈ Mn(R), for 1 ≤ ℓ ≤ k. Then, for
2n, . . . , xℓ
where
nn),
X F =Xf ∈F
kYi=1
X fi
i .
Similarly, if ~p = (p1, . . . , pk), we define pF :=Pf ∈FQk
Let ~v = (v1, . . . , vn). We now define a smooth function f : Rkn2+k+n →
i=1 pfi
i .
Rk+n by
f (~x, ~p, ~v)
= −1 +
nXi=1
vi, [X2(~v)]1 − p2v1, . . . , [Xk(~v)]1 − pkv1, (X F − pF · I)~v! .
11, . . . , xk
If ~v is an eigenvector for each matrix Xℓ = (xℓ
ij)i,j with eigenvalue pℓ,
nn, ~p, ~v) = ~0. Moreover, the (k + n) ×
and k~vk1 = 1, then f (x1
(k + n) matrix of partial derivatives of f with respect to the variables
p1, . . . , pk, v1, . . . , vn is given by F = (Fi,j)i,j, where if 2 ≤ i, j ≤ k and
1 ≤ h ≤ n, we have
Fi,j =
Fi,k+h =
([Xi(~v)]1−piv1) = −δi,jv1;
([Xi(~v)]1−piv1) = xi
1h.
∂
∂pj
∂
∂vh
Note that F1,j = ∂
∂pj
Moreover, F1,k+h = ∂
∂vh
Fk+h,i =
∂
Fi,1 =
([Xi(~v)]1 − piv1) = 0 ∀ 1 ≤ i ≤ k.
m=1 vm − 1) = 1. If 1 ≤ i ≤ k and 1 ≤ h ≤ n,
h=1 vh) = 0 for all 1 ≤ j ≤ k. Similarly,
∂
∂p1
(Pn
(Pn
∂pi−pF vh +
(X F )h,jvj = −vhXf ∈F
nXj=1
∂vj−pF vi +
(X F )i,jvj = X F
nXj=1
i Yj6=i
i,j − δi,jpF .
fipfi−1
p
fj
j .
∂
Finally, if 1 ≤ i, j ≤ n we have
Fk+i,k+j =
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
11
Applied to the situation when Xi = Bi(y, θ), ~v = ξy,θ, and pi = ρ(Bi(y, θ)) =:
fi
0
~0
0
0
1
···
1
ρi, the Implicit Function Theorem tells us that whenever the (k+n)×(k+n)
matrix
(Bi(y, θ)1,h)2≤i≤k, 1≤h≤n
(B(y, θ)F − ρF · I)
j=1 ρjfj(cid:17)h,i
i=1 depend smoothly on the entries of the matrices Bi(y, θ).
···
1 , . . . ,−ξy,θ
diag (−ξy,θ
1 )
(cid:16)−ξy,θ
ρiQk
h Pf ∈F
F =
is invertible, the eigenvalues ρ(Bi(y, θ)) and the common eigenvector ξy,θ of
{Bi(y, θ)}k
We now proceed to show that F is invertible. Suppose that F(cid:18)q
and −ξy,θ
Then kζk1 = 0. Moreover, for all 2 ≤ i ≤ k,
(3.7)
ζ(cid:19) = 0.
[Bi(y, θ)(ζ)]1 = qiξy,θ
1
(3.8)
h
fiqi
ρi
ρjfj =
nXm=1
kYj=1
kXi=1Xf ∈F
(BF − ρF · I)(ζ) =
kXi=1Xf ∈F
h,mζm − ρF ζh, equivalently,
BF
kYj=1
pjfj ξy,θ.
fiqi
pi
In particular, Equation (3.8) implies that (BF − ρF · I)(ζ) is a multiple of
ξy,θ.
Let J be the Jordan form of BF , and let S := {ξy,θ, x1, . . . , xn−1} be
a corresponding basis of Rn. Observe that [BF − ρF · I]S is an upper
If ρF = α0, α1, . . . , αn−1 are the eigenvalues of BF ,
triangular matrix.
counted with multiplicity, then the diagonal of [BF − ρF · I]S is given by
0, α1 − ρF , . . . , αn−1 − ρF .
Writing ζ = z0ξy,θ +Pn−1
i=1 zixi, the fact that [BF − ρF · I]S is upper
triangular and has first row zero implies that (BF − ρF · I)(ζ) is a linear
combination of the vectors (xi)n−1
i=1 :
[BF − ρF · I]S (ζ) =
n−1Xi=1
[(αi − ρF )zi + Ji,i+1zi+1]xi,
where we add zi+1 only for those indices i for which Ji,i+1 = 1.
Since (BF − ρF · I)(ζ) is a multiple of ξy,θ whenever F(cid:18)q
ζ(cid:19) = 0, the fact
that αn−1 6= ρF implies that zn−1 = 0 in this case. Proceeding "backwards"
by induction (from n − 1 towards 1) reveals that, in fact, zi = 0 for all
1 ≤ i ≤ n − 1. In other words,
F(cid:18)q
ζ(cid:19) = 0 ⇒ ζ = z0ξy,θ.
In particular, Bi(y, θ)(ζ) = z0ρiξy,θ and (computed in the original basis)
kζk1 = z0. Thus, the fact that kζk1 = 0 implies z0 = 0, and hence ζ = 0.
Moreover, using these formulas in Equation (3.7) implies that
qi = z0ρi ∀ 2 ≤ i ≤ k and hence
qi = 0 if i > 1.
12
FARSI, GILLASPY, LARSEN AND PACKER
Recall from Equation (3.8) that z0 =Pk
follows from the previous equation that
i=1Pf ∈F
j=1 ρjfj. It then
fiqi
ρi Qk
q1Xf ∈F
f1
ρ1
kYj=1
ρjfj = 0.
Every term in the sum is non-negative, since ρj > 0 for all j and fj ≥ 0.
Moreover, Equation (3.6) guarantees that there is at least one nonzero term
in the sum; consequently, q1 = 0. We conclude that if F(cid:18)q
ζ(cid:19) = 0 then q = 0
and ζ = 0, as claimed.
(cid:3)
Corollary 3.7. Let Λ be a finite, strongly connected k-graph equipped with
a R+-functor y. In the weak*-topology on the space of measures on Λ∞, the
function θ 7→ µy,θ is continuous on R+.
Proof. Theorem 3.6 and the definition (3.4) of µy,θ combine to tell us that
for each fixed λ ∈ Λ, the function
θ 7→ µy,θ(Z(λ))
is continuous (in fact smooth). Since {χZ(λ) : λ ∈ Λ} densely spans C(Λ∞),
whose dual is the space of measures on Λ∞, standard measure-theoretic
arguments enable us to complete the proof.
(cid:3)
We conclude this section with some remarks about the relationship be-
tween the measures µy,θ and the periodicity of Λ.
Definition 3.8. [RS07, Lemma 3.2] A k-graph Λ has periodicity at v ∈
Λ0 or is not aperiodic if there exists m 6= n ∈ Nk such that for all x ∈
Z(v), σm(x) = σn(x). We define
Per(v) := Z{m − n σm(x) = σn(x), ∀ x ∈ Z(v)}.
If Λ is strongly connected, then [HLRS15, Lemma 5.1] establishes that
Per(v) = Per(w) =: PerΛ
PerΛ = {d(λ) − d(ν) λx = νx, ∀ x ∈ Λ∞}.
for any v, w ∈ Λ0. In fact,
(3.9)
We define PΛ := {(λ, ν) ∈ Λ × Λ λx = νx, ∀ x ∈ Λ∞}.
Remark 3.9. We observe that Lemma 8.4 of [HLRS15], and its proof, are
still valid if we replace the measure M by µy,θ, and replace ρ(Λ)−d(λ) by
e−θy(λ)ρ(B(y, θ))−d(λ) wherever the former appears in the proof. The key
idea of this proof is that {M (Z(v)) : v ∈ Λ0} is an eigenvector for each
matrix Ai, with eigenvalue ρ(Λ)ei. In our case, Proposition 3.5 guarantees
that {µy,θ(Z(v)) : v ∈ Λ0} is an eigenvector for each matrix Bi(y, θ) with
eigenvalue ρ(Bi(y, θ)). Consequently, Proposition 8.2 of [HLRS15] also holds
for the measures µy,θ, so we conclude that
(3.10)
µy,θ ({x ∈ Λ∞ σm(x) = σn(x)}) =(1, m − n ∈ PerΛ
else.
0,
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
13
4. KMS states associated to (y, θ)
Throughout this section Λ will be a finite, strongly connected k-graph.
Each R+-functor y gives us a family of generalized gauge actions {αy,θ}θ>0
on C ∗(Λ) (denoted αy in Remark 5.25 of [McN15]): For t ∈ R and any
generator sλ of C ∗(Λ),
(4.1)
αy,θ
t
(sλ) := eity(λ)(cid:16)ρ(B(y, θ))d(λ)(cid:17)it/θ
sλ = eit(y(λ)+ 1
θ ln(ρ(B(y,θ))d(λ)))sλ.
In general, these actions are not the same as the actions αr considered in
[HLRS14] and [HLRS15]. As was remarked already in [McN15], αy,θ differs
from αr as soon as y is not of the form y(λ) = r · d(λ) for some fixed
r ∈ (0,∞)k and all λ ∈ Λ. Section 5 describes a variety of examples of
R+-functors y which are not of the form y(λ) = r · d(λ).
In this section, we will compute the KMSβ states associated to the actions
αy,θ. Proposition 4.5 establishes that KMSβ states for αy,θ exist precisely
when αy,θ = αy,β. The KMSβ states for αy,θ are therefore described in
Theorem 4.8 below. A similar reduction in computational complexity occurs
for the actions αr mentioned above. Indeed, as explained at the beginning
of Section 7 of [HLRS15], any KMSβ state associated to an action of the
form αr must also be a KMS1 state for αln(ρ(Λ)).
In order to compute the KMS states for a C ∗-dynamical system (A, γ),
one often first identifies a good dense subset of A on which γ is analytic (it
is well-known that the γ-analytic elements are dense in A, see for example
[BR], but for Cuntz-Krieger type algebras there usually is a good spanning
set for A inside the analytic elements). In our setting, this dense subset is
span{sλs∗
ν : λ, ν ∈ Λ}. Indeed, as observed in [HLRS14, page 269] for the
case of the gauge action, any element of C ∗(Λ) of the form sλs∗
ν is αy,θ-
analytic for every R+-functor y and all θ > 0. To see this, note that the
function t 7→ αy,θ
ν) on R admits the extension
(sλs∗
ζ 7→ eiζ(y(λ)−y(ν))+ln(ρ(B(y,θ))(d(λ)−d(ν))/θ)sλs∗
ν,
which is an entire function. The fact that span{sλs∗
ν : λ, ν ∈ Λ} is dense in
C ∗(Λ) thus implies that the KMSβ states for αy,θ are precisely the norm-one
positive linear functionals φ : C ∗(Λ) → C such that for any (λ, ν), (ρ, η) with
s(λ) = s(ν) and s(ρ) = s(η),
t
(4.2)
φ(sλs∗
νsρs∗
η) = φ(sρs∗
ηαy,θ
iβ (sλs∗
ν)).
We next observe that in the groupoid picture of k-graph algebras, these
actions αy,θ arise from a cocycle cy,θ on the k-graph groupoid GΛ as in [Nes13]
or [Tho14];
(4.3) cy,θ(λz, d(λ)− d(ν), νz) = y(λ)− y(ν) + log(cid:16)ρ (B(y, θ))(d(λ)−d(ν))/θ(cid:17) .
Proposition 4.1. The function cy,θ : GΛ → R is well-defined and satisfies
cy,θ(gh) = cy,θ(g) + cy,θ(h).
Proof. Suppose that g = (λz, d(λ) − d(ν), νz) = (λ′z′, d(λ′) − d(ν′), ν′z′).
To show that cy,θ is well defined, it suffices to show that y(λ) − y(ν) =
14
FARSI, GILLASPY, LARSEN AND PACKER
y(λ′) − y(ν′). To that end, write d(λ) − d(λ′) = (n1, . . . , nk) ∈ Zk and, for
each 1 ≤ i ≤ k, define
(mλ)i = max{0,−ni};
(mλ′)i = max{0, ni}.
i=1] we have d(λ) + mλ = d(λ′) +
Then [defining mλ ∈ Nk by mλ = ((mλ)i)k
mλ′ and, since d(ν) − d(ν′) = d(λ) − d(λ′) = (n1, . . . , nk), we also have
d(ν) + mλ = d(ν′) + mλ′.
Write λz = z(0, mλ) and λ′
z′ = z′(0, mλ′ ). By construction,
λλz = λ′λ′
z′
and
νλz = ν′λ′
z′.
The additivity of y now implies that, as desired,
y(λ) − y(ν) = y(λλz) − y(νλz) = y(λ′λ′
Showing that cy,θ is multiplicative uses a similar argument. Given g, h ∈
z′) = y(λ′) − y(ν′).
z′) − y(ν′λ′
GΛ with s(g) = r(h), we can choose λg, νg = λh, νh ∈ Λ such that
g = (λgz, d(λg) − d(νg), νgz),
h = (λhz, d(λh) − d(νh), νhz),
gh = (λgz, d(λg) − d(νh), νhz);
cf. [KPS15, Lemma 6.3]. To check that cy,θ is multiplicative, it now suffices
to observe that
y(λg) − y(νg) + y(λh) − y(νh) = y(λg) − y(νh).
(cid:3)
The following definition is an application of the definition of quasi-invariance
from [Nes13] to our setting, taking for our sets U the basic open sets
Z(λ, ν) = {(λz, d(λ) − d(ν), νz) : z ∈ Λ∞} of GΛ. We have also invoked
the fact that cy,θ is constant on the sets Z(λ, ν).
Definition 4.2. Let y be a R+-functor and θ, β ∈ (0,∞). We say that a
measure µ on Λ∞ is quasi-invariant with Radon-Nikodym cocycle e−βcy,θ if,
for any (λ, ν) ∈ Λ × Λ with s(λ) = s(ν) and all z ∈ Λ∞,
e−βcy,θ(λz,d(λ)−d(ν),νz)µ(Z(ν)) = µ(Z(λ)).
Equivalently, µ is quasi-invariant with Radon-Nikodym cocycle e−βcy,θ iff
eβy(ν)ρ(B(y, θ))
whenever s(ν) = s(λ).
β
θ d(ν)µ(Z(ν)) = eβy(λ)ρ(B(y, θ))
β
θ d(λ)µ(Z(λ))
Remark 4.3.
(1) If β = θ, the measure µy,β is quasi-invariant with
Radon-Nikodym cocycle e−θcy,θ. In fact, Proposition 3.5 establishes
that µy,θ is the unique such measure.
(2) It is relatively straightforward to check that µy,θ is (GΛ, cy,θ) con-
formal in the sense of [Tho14]. Consequently, Proposition 4.4 below
could also be derived from [Tho14, Proposition 2.1].
Proposition 4.4. Let Λ be a finite, strongly connected k-graph, equipped
with a R+-functor y, and choose β, θ ∈ (0,∞). Let Ψ denote the canoni-
cal conditional expectation Ψ : C ∗(Λ) → C(Λ∞). For any quasi-invariant
probability measure µ with Radon-Nikodym cocycle e−βcy,θ , the function
ψβ(a) :=ZΛ∞
Ψ(a) dµ,
for a ∈ C ∗(Λ), defines a KMSβ state for (C ∗(Λ), αy,θ).
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
15
Proof. Since µ is a probability measure, ψβ is easily verified to be a positive
linear functional of norm 1. Since the elements sλs∗
ν in a dense spanning
family for C ∗(Λ) are αy,θ-analytic, it suffices to verify condition (4.2).
One easily sees that
αy,θ
iβ (sλs∗
ν) = e−β(y(λ)−y(ν))ρ(B(y, θ))
β
θ (d(ν)−d(λ))sλs∗
ν,
so, using Equation (2.3), ψβ(sγs∗
ζαy,θ
iβ (sλs∗
ν)) becomes
e−β(y(λ)−y(ν))ρ(B(y, θ))
ψβ(sγρ1s∗
νρ2)
= e−β(y(λ)−y(ν))ρ(B(y, θ))
µ(Z(νρ2)),
β
θ d(ν)−d(λ) X(ρ1,ρ2)∈Λmin(ζ,λ)
θ d(ν)−d(λ) X(ρ1,ρ2)∈S1
β
Similarly, setting S2 = {(η1, η2) ∈ Λmin(γ, ν) : λη2 = ζη1},
ζη1)
where S1 = {(ρ1, ρ2) ∈ Λmin(ζ, λ) : γρ1 = νρ2}.
ζ) = X(η1,η2)∈Λmin(γ,ν)
= X(η1,η2)∈S2
ψβ(sλη2s∗
µ(Z(λη2)).
ψβ(sλs∗
νsγs∗
Note that, since (η1, η2) is an extension of (ζ, λ) for any (η1, η2) ∈ S2, we
must have
d(ηi) ≥ d(ρi) for i = 1, 2
for any (ρ1, ρ2) ∈ Λmin(ζ, λ). Similarly, since (ρ1, ρ2) is an extension of
(γ, ν) for any (ρ1, ρ2) ∈ S1, we must have d(ρi) ≥ d(ηi) for each i. Hence
d(ρi) = d(ηi) for all (ρ1, ρ2) ∈ S1, (η1, η2) ∈ S2. In other words, (ρ1, ρ2) is a
minimal extension of (γ, ν) and (η1, η2) is a minimal extension of (ζ, λ), for
all (ρ1, ρ2) ∈ S1, (η1, η2) ∈ S2. It follows that
Now, we see that ψβ(sγs∗
S1 = Λmin(ζ, λ) ∩ Λmin(γ, ν) = S2.
ν)) transforms as
iβ (sλs∗
ζαy,θ
e−β(y(λ)−y(ν))ρ(B(y, θ))
e−βy(ν)ρ(B(y, θ))− β
θ d(ν)µ(Z(ρ2))
β
θ (d(ν)−d(λ)) X(ρ1,ρ2)∈S1
θ d(λ) X(η1,η2)∈S2=S1
µ(Z(η2))
= e−βy(λ)ρ(B(y, θ))− β
= X(η1,η2)∈S2
µ(Z(λη2)) = ψβ(sλs∗
νsγs∗
ζ),
which is (4.2). Hence ψβ is a KMSβ state as claimed.
(cid:3)
Proposition 4.5. Fix a finite, strongly connected k-graph Λ, a R+-functor
y on Λ, and β, θ ∈ (0,∞). There exist KMSβ states for (C ∗(Λ), αy,θ) iff
αy,θ = αy,β.
16
FARSI, GILLASPY, LARSEN AND PACKER
Proof. Choose a KMSβ state φ for (C ∗(Λ), αy,θ). By the Cuntz-Krieger
relations and the KMS condition, for any 1 ≤ i ≤ k and v ∈ Λ0,
φ(s∗
λsλ)e−βy(λ)(cid:16)ρ(B(y, θ))−d(λ)(cid:17)β/θ
φ(ps(λ))e−βy(λ)ρ(Bi(y, θ))−β/θ
φ(sλs∗
λ) =Xλ
φ(pv) = Xλ∈vΛei
= Xλ∈vΛei
= ρ(Bi(y, θ))−β/θ Xw∈Λ0
φ(pw) Xλ∈vΛei w
e−βy(λ).
Since Pλ∈vΛei w e−βy(λ) = Bi(y, β)v,w, we see that (φ(pv))v∈Λ0 is an eigen-
vector for Bi(y, β) with eigenvalue ρ(Bi(y, θ))β/θ. Moreover,
φ(pv) = φ(1) = 1,
Xv∈Λ0
so [HLRS15, Proposition 3.1(a)] implies that (φ(pv))v∈Λ0 = ξy,β and
It follows that αy,θ = αy,β.
ρ(Bi(y, θ))1/θ = ρ(Bi(y, β))1/β ∀ i.
Conversely, if αy,β = αy,θ, then the KMSβ states for the two actions are
(cid:3)
the same (and constitute a nontrivial set by Proposition 4.4).
In the case when Λ has only one vertex, we obtain a slightly sharper
result. Such higher-rank graphs have been extensively studied by Davidson,
Power, and Yang (cf. [DY09a, DY09b, DPY10]).
Proposition 4.6. Let Λ be a finite k-graph with one vertex. Choose a R+-
functor y on Λ and β, θ ∈ (0,∞). If there exist KMSβ states for αy,θ, then
β = θ.
Proof. In the one-vertex case, each adjacency matrix Bi(y, θ) has only one
(positive) entry, which is also its spectral radius:
ρ(Bi(y, θ)) = Xh∈Λei
e−θy(h).
Since the function θ → ρ(Bi(y, θ)) is differentiable by Theorem 3.6, the
function
ψi(θ) := ρ(Bi(y, θ))1/θ =(cid:16) Xh∈Λei
e−θy(h)(cid:17)1/θ
is also differentiable, and
dψi
dθ
(ρ(Bi(y, θ)))
1
θ
=
1−θ
θ
< 0.
d
dθ
ρ(Bi(y, θ)) =
1
θ
(ρ(Bi(y, θ)))
1−θ
θ Xh∈Λei
−y(h)e−θy(h)
Consulting the formula (4.1) for the action αy,θ reveals then that αy,β 6= αy,θ
if β 6= θ ∈ R+. The result now follows from Proposition 4.5.
(cid:3)
It would be interesting to know if the conclusion of Proposition 4.6 is valid
for k-graphs with more than one vertex. Proposition 4.7 offers a partial
result in this direction. In his thesis, McNamara identified a different set of
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
17
hypotheses guaranteeing the uniqueness result of Proposition 4.6; we discuss
this result in Remark 4.10 below.
Proposition 4.7. Fix a finite, strongly connected k-graph Λ, a R+-functor
y on Λ, and suppose that for some interval (a, b) ⊂ R+
ρ(Bi(y, θ)) > 1, ∀ θ ∈ (a, b), ∀ 1 ≤ i ≤ k.
For β, θ ∈ (a, b), there exist KMSβ states for (C ∗(Λ), αy,θ) iff β = θ.
Proof. This follows from the explicit computation of the derivative of
ψi(θ) = ρ(Bi(y, θ))1/θ;
we will show that dψi
dθ < 0 on the entire interval (a, b) under the given
hypotheses. Once this is established, the same argument used in the proof
of Proposition 4.6 ends the proof.
To show dψi
dθ < 0 on (a, b), we explicitly compute
d
ln ρ(Bi(y,θ))
dψi
dθ
dψi
dθ
θ
=
dθ(cid:16)e
(cid:17)
= ρ(Bi(y, θ))1/θ"−
< 0 ⇐⇒ (cid:20)−
1
θ
ln(ρ(Bi(y, θ))) +
⇐⇒
d
dθ (ln ρ(Bi(y, θ)))
ln(ρ(Bi(y, θ)))
<
1
θ
.
Now ρ(Bi(y, θ)) > 1 implies ln(ρ(Bi(y, θ))) > 0. Therefore
1
θ2 ln(ρ(Bi(y, θ))) +
θ
d
dθ (ln ρ(Bi(y, θ)))
# .
(ln ρ(Bi(y, θ)))(cid:21) < 0
d
dθ
Gelfand's formula implies that ρ(Bi(y, θ)) is a non-increasing function of
θ. Taking derivatives reveals that ln(ρ(Bi(y, θ))) is also a non-increasing
function of θ. It follows that
(ln ρ(Bi(y, θ)))′
ln(ρ(Bi(y, θ))) ≤ 0 <
1
θ
,
thus proving dψi
dθ < 0.
(cid:3)
For the remainder of the section, we assume a R+-functor y and β ∈ (0,∞)
are given. We will characterize the KMSβ states of (C ∗(Λ), αy,β ). Recall
from [BR] that the KMSβ-states for β ∈ R form a Choquet simplex, and in
particular are determined by the extremal KMSβ-states.
In the terminology of [Chr17] or [Nes13], Proposition 3.5 above implies
that µy,β is the unique measure on Λ∞ which is e−βcy,β -quasi-invariant.
Therefore, Lemma 4.1 and Theorem 5.2 of [Chr17] imply that the extremal
KMSβ states for αy,β are in bijection with a certain subgroup bB of Tk.
Theorem 4.8 below establishes that B is equal to PerΛ ⊆ Zk, the periodicity
group of Λ.
We observe that Christensen's Theorem 5.2 is a refinement of Neshveyev's
description [Nes13] of KMS states on groupoid C ∗-algebras. While both
Christensen and Neshveyev use quasi-invariant measures to parametrize
KMS states, Christensen replaces Neshveyev's measurable fields of states
18
FARSI, GILLASPY, LARSEN AND PACKER
with the group B, which consists of symmetries of the simplex of KMSβ
states.
In the case that y ≡ 0 and β = 1, so that αy,β is the preferred dynamics
on C ∗(Λ), Theorem 4.8 reduces to [HLRS15, Theorem 7.1]. Our proof of
Theorem 4.8 combines [HLRS15, Theorem 7.1] with [Chr17, Theorem 5.2]
to show that C ∗(Per Λ) parametrizes KMSβ states for any of the dynamics
in the family {αy,β}β>0.
Theorem 4.8. Let Λ be a finite, strongly connected k-graph, with a R+-
functor y, and fix β ∈ (0,∞). The simplex of KMSβ states for (C ∗(Λ), αy,β)
is affinely isomorphic to the state space of C ∗(Per Λ).
Proof. Recall the isomorphism C ∗(GΛ) ∼= C ∗(Λ) sending χZ(ρ,η) from the
dense subalgebra Cc(GΛ) ⊆ C ∗(GΛ) to sρs∗
η ∈ C ∗(Λ), where
Z(ρ, η) := {(x, d(ρ) − d(η), y) x ∈ Z(ρ), y ∈ Z(η)} ⊆ GΛ
for ρ, η ∈ Λ. Recall also that αy,β is determined by the cocycle cy,β. We
aim to apply [Chr17, Theorem 5.2] to GΛ and cy,β. For this we must first
identify the extremal KMSβ states of (C ∗(GΛ), αy,β); these are given as
in Equation (5.1) of [Chr17, Theorem 5.1] by integrating functions f ∈
Cc(GΛ) with respect to measures which are extremal points in the set of
e−βcy,β -quasi-invariant measures on Λ∞. In our case, there is a unique such
measure, namely µy,β, see Proposition 3.5. Since any KMS state must be
linear and continuous, we may therefore assume that Christensen's functions
f are of the form χZ(ρ,η) for ρ, η ∈ Λ. Equivalently, we apply Equation
(5.1) of [Chr17] to monomials sρs∗
η in C ∗(Λ). Note also that in our setting,
Christensen's group A is Zk and Φ : GΛ → A is given by Φ(x, n, y) = n.
ν corresponds to a periodic pair (λ, ν) ∈
PΛ as in Definition 3.8. Then Z(λ) = Z(ν), and Equation (5.1) of [Chr17]
implies that any extremal KMSβ state ω for αy,β must satisfy
Assume first that the monomial sλs∗
ω(sλs∗
ν) = zd(λ)−d(ν)µy,β(Z(λ)) = zd(λ)−d(ν)µy,β(Z(ν))
for some z ∈ Tk. Write ω1 for the extremal KMSβ state associated to
z = (1, 1, . . . , 1), cf. [Chr17, Proposition 4.2].
For an arbitrary pair (ρ, η), Theorem 5.1 of [Chr17] reveals that every
extremal KMSβ state for αy,β will be of the form
ω(sρs∗
η) = zd(ρ)−d(η)µy,β(Aρ,η),
where Aρ,η = {x ∈ Z(ρ) ∩ Z(η) σd(ρ)+m(x) = σd(η)+m(x) for some m ∈
Nk}. However, by Remark 3.9 we see that µy,β(Aρ,η) will be zero unless
d(ρ) − d(η) ∈ PerΛ.
We claim that if d(ρ)−d(η) ∈ PerΛ and Z(ρ)∩Z(η) 6= ∅, then (ρ, η) ∈ PΛ.
Indeed, with v = r(ρ) = r(η), the fact that σd(ρ)(y) = σd(η)(y) for all
y ∈ Z(v) implies, by Lemma 5.1(b) of [HLRS15], that there is a unique
η ∈ Λd(η) such that ρx′ = ηx′ for all x′ ∈ Z(s(ρ)). This in particular means
that (ρ, η) ∈ PΛ. Therefore, if there exists x = ρx1 = ηx2 ∈ Z(ρ) ∩ Z(η) for
some x1, x2 ∈ Λ∞, then
ηx1 = ρx1 = ηx2 ⇒ ηx2(0, m) = ηx1(0, m)
for any m ∈ Nk. Since d(η) = d(η), it follows that η = η and x1 = x2.
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
19
In other words, any extremal KMSβ state for αy,β must be of the form
(4.4)
ωz(sρs∗
η) =(zd(ρ)−d(η)µy,β(Z(ρ)) = zd(ρ)−d(η)µy,β(Z(η)),
Theorem 5.2(2) of [Chr17] describes a homeomorphism between the dual
0,
for some z ∈ Tk.
bB of a certain subgroup B ⊆ A = Zk and the set of extremal KMSβ states
for αy,β More precisely, B is defined in terms of a subgroup N ⊆ Tk. In our
setting,
(ρ, η) ∈ PΛ
else
N = {z ∈ Tk : ω1(sρs∗
η) = zd(ρ)−d(η)ω1(sρs∗
η) ∀ (ρ, η) ∈ Λ ∗s Λ}.
Since ω1(sρs∗
η) = 0 unless (ρ, η) ∈ PΛ, it follows that
(4.5)
Consequently,
N = {z ∈ Tk : zd(ρ)−d(η) = 1 ∀ (ρ, η) ∈ PΛ}.
B = N ⊥ = {m ∈ Zk : zm = 1 ∀ z ∈ N}
clearly contains PerΛ.
Our goal is to show that B = PerΛ. To this end, note that N is in-
dependent of the choice of y and β. Furthermore, when y = 0 we obtain
Bi(0, β) = Ai, for any β > 0. Thus, µ0,β agrees with the measure M from
Proposition 8.1 of [HLRS15], and α0,1 agrees with the preferred dynamics α
from [HLRS15].
Theorem 7.1 of [HLRS15] establishes that the extremal KMS1 states for
α are in bijection with the pure states of C0(\Per Λ), that is, with the points
of \Per Λ. Moreover, Remark 10.4 of [HLRS15] shows that this bijection (just
as in Theorem 5.2 of [Chr17]) assigns the state
φz(sλs∗
ν) =(M (Z(λ))zd(λ)−d(ν),
0,
(λ, ν) ∈ PΛ
else
to z ∈ \Per Λ. Therefore, we must have bB = \Per Λ and hence B = Per Λ.
Consequently, for any R+-functor y and β ∈ R+, the simplex of KMSβ states
of (C ∗(Λ), αy,β) is affinely isomorphic to the state space of C ∗(Per Λ).
(cid:3)
Corollary 4.9. Let Λ be a finite, strongly connected higher-rank graph,
and fix β ∈ (0,∞) and an R+-functor y on Λ. The C ∗-dynamical system
(C ∗(Λ), αy,β ) admits a unique KMSβ state iff Λ is aperiodic, iff C ∗(Λ) is
simple.
Proof. The last equivalence was established in [HLRS15, Theorem 11.1]. For
the first equivalence, observe that by Theorem 4.8, uniqueness of the KMSβ
state is equivalent to the triviality of C ∗(Per Λ) -- in other words, to the
aperiodicity of Λ.
(cid:3)
Remark 4.10. In his thesis [McN15], McNamara considers finite k-graphs
Λ which are coordinatewise irreducible, in the sense that each coordinate
matrix Ai, for i = 1, . . . , k, is irreducible. In particular, [McN15, Theorem
20
FARSI, GILLASPY, LARSEN AND PACKER
5.30] establishes that given a R+-functor y on such a k-graph Λ and β ∈
(0,∞), if the statement
(4.6)
1
β
1
β
y(λ) +
ln(ρ(B(y, β))d(λ)) = y(ν) +
ln(ρ(B(y, β))d(ν)) ⇒ d(λ) = d(ν)
holds, then there is a unique KMS state φ of (C ∗(Λ), αy,β) occurring at
(inverse) temperature β. Moreover, this KMS state satisfies φ(sλs∗
ν) =
ξy,β
s(λ). Observe that φ is the state we ob-
δλ,νe−βy(λ)(cid:0)ρ(B(y, β))−d(λ)(cid:1)1/β
tained in Proposition 4.4 above.
Combining McNamara's result with our Theorem 4.8 above implies that
if Λ is coordinatewise irreducible and Equation (4.6) holds for at least one
pair (y, β), then Λ must be aperiodic. Given the potential importance of
this result for applications, we also offer a direct proof which does not rely
on Theorem 4.8.
Corollary 4.11. Let Λ be a finite, coordinatewise irreducible k-graph. If
there exists an R+-functor y on Λ and β ∈ (0,∞) such that Equation (4.6)
holds, then Λ is aperiodic.
Proof. We argue by contrapositive. If Per Λ 6= 0, choose (λ, ν) ∈ PΛ with
d(λ) 6= d(ν). By construction, Z(λ) = Z(ν), so
µy,β(Z(λ)) = µy,β(Z(ν))
for any R+-functor y and β ∈ (0,∞). The fact that s(λ) = s(ν) (and hence
s(λ) = ξy,β
ξy,β
s(ν) > 0) whenever (λ, ν) ∈ PΛ then implies that
eβy(λ)ρ(B(y, β))d(λ) = eβy(ν)ρ(B(y, β))d(ν).
Taking logarithms of both sides and dividing by β yields the left-hand side
of Equation (4.6), yet d(λ) 6= d(ν).
(cid:3)
The preceding Corollary generalizes [HLRS15, Corollary 7.2], which es-
tablishes that for periodic k-graphs which are coordinatewise irreducible,
the set {ln(ρ(Ai))}k
i=1 is rationally dependent. Indeed, rational dependence
of the set {ln(ρ(Ai))}k
i=1 implies that Equation (4.6) fails for y = 0 and
β = 1. Corollary 4.11 implies that for periodic k-graphs which are coordi-
natewise irreducible, Equation (4.6) must fail for all choices of y and β. In
other words, Corollary 4.11 offers an expanded set of strategies for detecting
aperiodicity of the k-graph, and hence the simplicity of C ∗(Λ).
5. Examples of R+-functors
Before addressing the relationship between R+-functors and Hausdorff
measures in Section 6, we pause to discuss the range of possibilities for R+-
functors on two examples of finite, strongly connected 2-graphs. We also
describe the associated actions and quasi-invariant measures. Finally, we
identify the periodicity groups of these 2-graphs; Theorem 4.8 then enables
us to reconstruct the KMS states associated to these 2-graphs and R+-
functors.
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
21
5.1. An example from [McN15]. We begin with an example which was
studied in Section 5.11 of McNamara's thesis [McN15]. Namely, Λ is a 2-
graph with one vertex v and two blue edges (called e1, e2) and two red edges
(called f1, f2) and factorization relations
e1f1 = f1e1, e1f2 = f1e2, e2f1 = f2e1, e2f2 = f2e2.
Consequently, Z(ei) = Z(fi) for i = 1, 2. It follows that Λ is periodic and
Per (Λ) ⊇ Z{(1,−1)}.
if Per (Λ) )
Z{(1,−1)}, then there would exist integers {ni, mi}ℓ
i=1 such that
1 f mℓ
).
Indeed, this inclusion is an equality:
Z(v) = Z(f n1
1 f m1
2
2
··· f nℓ
Since Z(f1) ⊔ Z(f2) = Z(v), however, this is impossible.
To describe all the R+-functors y on this 2-graph, set ei = y(ei), and
fi = y(fi). The factorization relations then tell us that all the R+-functors
must satisfy
e1 + f2 = f1 + e2, e2 + f1 = f2 + e1.
In other words, we have 3 free variables e1, e2, f1; and f2 = f1 + e2 − e1.
Since Λ0 = 1, we have
B1(y, θ) =(cid:0)e−θe1 + e−θe2(cid:1) ,
B2(y, θ) =(cid:0)e−θf1 + e−θf2(cid:1) =(cid:0)e−θf1 + e−θ(f1+e2−e1),(cid:1)
and ξy,θ = (1) for any y, θ.
The fact that Λ has only one vertex implies that each matrix Bi(y, β) will
be irreducible, for any choice of y and β. By Corollary 4.11, the periodicity
of Λ means that Equation (4.6) will never hold, regardless of our choice of
y and β. Indeed, although d(f1) 6= d(e1), we always have
y(e1) +
Consequently, since
1
β
ln(ρ(B1(y, β)) = y(f1) +
ln(ρ(B2(y, β)).
1
β
αy,β
t
(sλ) = eit(y(λ)+ 1
β ln(ρ(B(y,β))d(λ)))sλ,
the action αy,β scales both se1 and sf1 by the same complex number.
Recall that every infinite path in Λ∞ can be written uniquely as a one-
sided infinite sequence of edges which alternate blue-red-blue-red.
In Λ,
all edges are composable, so Λ∞ is naturally homeomorphic to the infinite
productQN{0, 1}. Moreover, µy,θ(Z(e2)) = 1 − µy,θ(Z(e1)) because Λ∞ =
Z(e1)⊔Z(e2). By our identification of Λ∞ withQN{0, 1}, we can view µy,θ as
a Markov measure µx onQN{0, 1}, where x = µy,θ(Z(e1)) = µy,θ(Z(f1)). In
the notation of Section 3.1 of [DJ14], the Markov measure µx onQn∈N{0, 1}
corresponds to the matrix
Tx =(cid:18) x
1 − x
1 − x
x (cid:19) ,
and assigns measure µx(Z(a1 ··· an)) = x#{i:ai=0}(1 − x)#{j:aj =1} to the
cylinder set Z(a1 ··· an). Indeed, [DJ14, Theorem 3.9] implies that if x 6= x′
then µx and µx′ are mutually singular. It follows that for x = µy,θ(Z(e1)) 6=
1/2, the measure µy,θ is mutually singular with respect to the measure M
from [HLRS15, Proposition 8.1].
22
FARSI, GILLASPY, LARSEN AND PACKER
In fact, the correspondence taking (y, θ) to x such that µy,θ = µx is
surjective. That is, given x ∈ (0, 1/2), we will describe a way to choose a
pair (y, θ) such that µy,θ(Z(e1)) = µx. Having chosen x ∈ (0, 1/2) and y(e1),
choose θ > ln((1−x)/x)
> 0 and define
y(e1)
y(e2) = y(e1) −
1
θ
ln(cid:18) 1 − x
x (cid:19) .
Note that y(e2) will be positive whenever θ > ln((1−x)/x)
. Setting y(fi) :=
y(ei) completes the definition of the R+-functor. However, other construc-
tions of weight functors are also possible; cf. [McN15, Example 5.31].
y(e1)
Remark 5.1. This example can be extended to the setting of 2-graphs with
one vertex and N edges of each color, using the Markov measures associated
to N × N matrices from [DJ14].
Remark 5.2. For a fixed R+-functor y, Corollary 3.7 tells us that varying
θ produces a continuous family of measures µy,θ. However, if x 6= x′, the
Markov measures µx and µx′ are mutually singular. Thus, equivalence and
continuity of a family of measures are different concepts.
5.2. An example from [LLN+15]. Another motivating example for us was
the 2-graph of [LLN+15, Example 7.7], which is described by the edge-
colored directed graph
u
d0
a0
c0
b0
v
d1
a1
c1
b1
w
with factorization rules
a0b0 = d0c0
a0b1 = d0c1
a1b1 = d1c1
c1d1 = b0a0
a1b0 = d1c0
c0d0 = b1a1.
Again, for any edge f , write f for the value y(f ).
The linear system arising from the factorization relations that a R+-
functor on Λ must satisfy consists of 6 equations, which we write in com-
pressed form as
ai + bi = di + ci, i = 0, 1
ai + bi = di + c1−i, i = 0, 1
ai + bi = d1−i + c1−i, i = 0, 1,
This system has 4 free variables (b1, c1, d0, d1).
Now, suppose we have chosen a R+-functor y on Λ and θ ∈ (0,∞). Define,
for i = 0, 1,
Ai := e−θ ai Bi := e−θ bi Ci := e−θ ci Di := e−θ di;
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
23
0
0 B1
0.
Straightforward computations reveal that
and the unimodular positive eigenvector for B1(y, θ) is
then B1(y, θ) =
0
0 C1
0 D0
B0
0 D1
0 A0
C0
0 A1
0, B2(y, θ) =
B1(cid:17)1/2
ρ(B1(y, θ)) =pA0C0 + A1C1 = √2 C1(cid:16) D1
A0 + A1 + ρ(B1(y, θ))
.
D0 + D1 + ρ(B2(y, θ))
.
ρ(B1(y, θ))
ρ(B2(y, θ))
D0
D1
A0
A1
1
1
Similarly, ρ(B2(y, θ)) = √B0D0 + B1D1 = √2 (B1 D1)1/2, and the unimod-
ular positive eigenvector for B2(y, θ) is
Lemma 2.1 implies that B1(y, θ) and B2(y, θ) have a unique common positive
unimodular eigenvector ξy,θ, so the eigenvectors for B1(y, θ) and B2(y, θ)
must be equal. Moreover, we have
ρ(B1(y, θ))ρ(B2(y, θ)) = 2 C1D1 = 2 A0B0.
With the above information, we can now compute the probability measure
µy,θ on some cylinder sets Z(λ). First, observe that
µy,θ(Z(a0b0)) = A0 B0(ρ(B1(y, θ)))−1(ρ(B2(y, θ)))−1ξy,θ
s(a0b0)
=
ξy,θ
u .
1
2
Proposition 4.6 of [FGJ+18a] explains how a matrix Tx =(cid:18) x
x (cid:19),
1 − x
for x ∈ (0, 1), can be used to construct Markov measures on Λ∞. For dif-
ferent values of x, the associated Markov measures are inequivalent. We
observe also that the Markov measures µx from [FGJ+18a, Proposition 4.6]
are not probability measures; rather, µx(Λ∞) = 2 .
1 − x
However, the measure µy,θ can only be a (rescaled) Markov measure for
x = 1/2. To see this, we recall from [FGJ+18a] that
µx(Z(a0b0)) = Tx(1, 1)Tx(1, 1) = x2,
Therefore, if µy,θ = 1/2µx for some x, then
while
µx(Z(u)) = Tx(1, 1) = x.
1
2
=
µy,θ(Z(a0b0)
µy,θ(Z(u))
=
µx(Z(a0b0))
µx(Z(u))
= x.
Note that the Markov measure µ1/2 also assigns µ1/2(Z(w)) = 1/2, and
µ1/2(Z(v)) = 1.
Now, recall from [FGJ+18a] that µ1/2 = 2M where M denotes the mea-
sure from [HLRS15, Proposition 8.1]. This measure M also arises as µy,θ
when y = 0.
In other words, the only way that µy,θ can be a (rescaled)
Markov measure is if µy,θ = M .
24
FARSI, GILLASPY, LARSEN AND PACKER
We can completely characterize the KMSβ states of (C ∗(Λ), αy,β ). Indeed,
by [LLN+15, Example 7.7], we know that Per(Λ) = 2Z(1,−1). Theorem 4.8
therefore implies that the simplex of KMSβ states is isomorphic to the tracial
state space of C(T).
6. Weights, ultrametrics, and Hausdorff structure
In this section, we use the same data (an R+-functor and a positive num-
ber θ) that we employed in Section 4 to define the generalized gauge action
αy,θ for a different purpose: namely, we construct an ultrametric dy,θ on the
infinite path space Λ∞, which we view as a Cantor set. We then compute
the Hausdorff dimension and Hausdorff measure of (Λ∞, dy,θ): Corollary
6.17 establishes that the Hausdorff dimension of (Λ∞, dy,θ) is θ -- the same
as the inverse temperature for which we characterized the KMS states for
the associated dynamics αy,θ of C ∗(Λ) in Theorem 4.8 -- and the associated
Hausdorff measure is precisely our unique quasi-invariant measure µy,θ. In
fact, we prove a result about Hausdorff dimension in a greater generality in-
volving weights on Bratteli diagrams with a certain self-similarity property,
see Theorem 6.16.
The examples of k-graphs and R+-functors which we discussed in Section
5 satisfy the hypotheses of Corollary 6.17.
In particular, the 2-graph of
Section 5.1 admits R+-functors giving rise to a large family of inequivalent
Hausdorff measures on Λ∞.
Following [PB09, JS11b], our ultrametrics dy,θ are constructed using weights
on Bratteli diagrams. Thus, we begin by reviewing the construction of a
Bratteli diagram associated to a higher-rank graph from [FGJ+18b] (Defi-
nition 6.1) and discussing how to use a R+-functor to construct weights on
the Bratteli diagram (Propositions 6.6 and 6.7). We then show, in Propo-
sition 6.9, that the ultrametric dy,θ arising from such a weight metrizes the
cylinder set topology on Λ∞.
We note that weights on Bratteli diagrams and the associated ultrametrics
have been studied by many authors [PB09, JS11b, FGJ+18b]. In particular,
Pearson and Bellissard [PB09] were motivated by work of Michon [Mic93],
who introduced the notion of a weighted tree in his study of Gibbs measures
on Cantor sets; see also [JS11a].
6.1. Defining weights and metrics on Bratteli diagrams.
Definition 6.1 ([FGJ+18b] Definition 2.5). Let Λ be a finite k-graph with
coordinate matrices A1, . . . , An. The stationary k-Bratteli diagram asso-
ciated to Λ, which we will call BΛ, is given by a filtered set of vertices
go from Vn to Vn−1, such that:
V =Fn∈N Vn and a filtered set of edges E =Fn≥1 En, where the edges in En
(a) For each n ∈ N, Vn = Λ0 consists of the vertices of Λ.
(b) When n ≡ i (mod k), there are Ai(p, q) edges whose range is the
vertex p of Vn−1 and whose source is the vertex q of Vn.
A path (finite or infinite) in the Bratteli diagram BΛ is a path with range in
V0. We write η for the length (number of edges) of a finite path η in the
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
25
Bratteli diagram, and F nBΛ for the finite paths of length n. We also write
FBΛ :=[n
F nBΛ.
Proposition 2.10 and Remark 2.11 of [FGJ+18b] discuss the relationship
between paths (finite and infinite) in Λ and BΛ. In particular, every finite
path in BΛ is represented by a string of composable edges in Λ. Conse-
quently, if η ∈ FBΛ we will also write η ∈ Λ to denote the unique morphism
in Λ represented by the string η of composable edges. However, not every
finite path in Λ corresponds to a finite path in BΛ. For example, a path in
Λ consisting of two red edges will not occur in BΛ.
The space of infinite paths in BΛ is also denoted the boundary of the
It is canonically equipped with the
Bratteli diagram in some references.
cylinder set topology, whose basic open sets are {Z(λ)}λ∈F BΛ , where Z(λ)
is the set of infinite paths whose initial segment is λ. Proposition 2.10 of
[FGJ+18b] shows that when we equip both spaces with the cylinder set
topology, Λ∞ is homeomorphic to the space of infinite paths in BΛ.
The following is a slight modification of the definition of a weight on a
Bratteli diagram as introduced in, for example, [PB09, JS11b, FGJ+18b].
Although we state Definition 6.2 for arbitrary Bratteli diagrams, in this
paper we will apply it mainly to stationary k-Bratteli diagrams.
Definition 6.2 (compare with [PB09] Definition 8, [JS11b] Definition 2.9,
[FGJ+18b] Definition 2.14). A weight on a Bratteli diagram B is a function
w : FB → R+ such that
(i) w(v) ≤ 1 for all v ∈ V0.
(ii) limn→∞ sup{w(γ) γ ∈ F nB} = 0.
(iii) If η is a sub-path of γ, then w(γ) ≤ w(η).
In this paper, we work primarily with Bratteli diagrams associated to
finite k-graphs. Under these hypotheses, the supremum in condition (ii)
above is actually a maximum.
Definition 6.3. A metric d on a space X is an ultrametric if
(6.1)
d(x, y) ≤ max{d(x, z), d(y, z) z ∈ X}.
The following Proposition shows that the first part of the conclusion of
Proposition 2.15 of [FGJ+18b] still holds with our revised definition of a
weight. The second part of that proposition, asserting that the ultrametric
topology agrees with the cylinder set topology, need not hold in general but
it does hold in our case of interest; see Proposition 6.9 below.
Proposition 6.4. Let w be a weight on a Bratteli diagram B. The formula
(6.2)
dw(x, z) =
1,
0,
w(x ∧ z),
r(x) 6= r(z)
x = z
else.
defines an ultrametric on the space XB of infinite paths in B. Here x ∧ z ∈
FBΛ denotes the longest common initial segment of x and z.
Proof. This follows verbatim from the first part of the proof of [FGJ+18b,
Proposition 2.15].
(cid:3)
26
FARSI, GILLASPY, LARSEN AND PACKER
The following Lemma establishes conditions under which the hypotheses
of Proposition 6.6 are satisfied. These conditions are not necessary; for
example, the coordinate matrices for the 2-graph studied in Section 5.2
satisfy the conclusion of Lemma 6.5 (and thus the hypotheses of Proposition
6.6) but not the hypotheses of Lemma 6.5.
Lemma 6.5. Let B be a nonnegative matrix with at least two non-zero
entries per row. Then the spectral radius of B is strictly greater than any of
the entries of B.
Proof. Write R for the positive square root of BBt, and notice that R is
Hermitian. By [Sch86, Theorem 2] we have that
ρ(R) ≤ ρ(B).
By the spectral theorem, it also follows that
ρ(R)2 = ρ(R2) = ρ(BBt).
Now assume that m = Bq,r is the largest entry of B; this implies that
the (q, q)-entry of BBt is strictly greater than m2. Now by using Rayleigh
quotients to bound the spectral radius for BBt, we get, if we denote by ej
the standard basis for Rn, that
ρ(R2) ≥ (BBteq, eq) > m2 ⇒ pρ(R2) > m.
It follows that m < ρ(R) ≤ ρ(B), as desired.
(cid:3)
In preparation for the next two propositions, we first note that due to
the fact that every path in FBΛ is given by a string of composable edges
in Λ, and hence represents a unique morphism in Λ, cf. [FGJ+18b, Remark
2.11], we can (and will) interpret a R+-functor y also as an additive functor
y : FBΛ → R+. Second, we identify a necessary condition on the matrices
Bi(y, θ) which ensures that we obtain a weight on FBΛ. Since the condition
will differ in the one-vertex case and the general finite k-graph case, we list
it for easy reference in the following two formulations:
(w-I) ∀i = 1, . . . , k, ρ(Bi(y, θ)) > 1.
(w-II) ∃i = 1, . . . , k such that ρ(Bi(y, θ)) > max{Bi(y, θ)v,w v, w ∈ Λ0}.
Observe that condition (w-I) is satisfied for all θ near 0 if the k-graph has
at least two edges of every color; the example of Section 5.1 satisfies this
condition.
Proposition 6.6. Let Λ be a finite, strongly connected k-graph with at least
two vertices. Let y be a R+-functor on Λ and θ ∈ R+ such that condition
(w-II) holds. Then the function wy,θ : FBΛ → R>0 given by
(6.3)
wy,θ(λ) = e−y(λ)(cid:16)ρ(B(y, θ))−d(λ)ξy,θ
s(λ)(cid:17)1/θ
is a weight on BΛ.
Proof. Since ξy,θ ∈ (R+)Λ0
has ℓ1-norm 1, wy,θ satisfies the first condition of
Definition 6.2. We next check the third condition. Let η ∈ FBΛ be a finite
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
27
path; suppose that η = qk + (i − 1), so that every edge extending η will
have degree ei. Writing w := s(η), we compute:
v
e−y(λ)θξy,θ
s(λ).
Bi(y, θ)w,vξy,θ
ρ(B(y, θ))−d(η)ξy,θ
Moreover, each summand is strictly positive, and therefore
w = ρ(B(y, θ))−d(η)−ei Xv∈Λ0
= ρ(B(y, θ))−d(η)−ei Xλ∈wΛei
w ≥ ρ(B(y, θ))−d(η)−ei e−y(λ)θξy,θ
ρ(B(y, θ))−d(η)ξy,θ
for any λ ∈ Eqk+i = Λei. Hence,
s(λ)(cid:17)1/θ
w (cid:17)1/θ
≥ e−y(λ)(cid:16)ρ(B(y, θ))−d(η)−ei ξy,θ
(cid:16)ρ(B(y, θ))−d(η)ξy,θ
≥ e−y(η)e−y(λ)(cid:16)ρ(B(y, θ))−d(ηλ)ξy,θ
s(η)(cid:17)1/θ
e−y(η)(cid:16)ρ(B(y, θ))−d(η)ξy,θ
for any such λ.
extension ηλ of η,
The additivity of y thus implies that
s(λ)
It follows that, given any finite path η ∈ FBΛ and any
s(λ)(cid:17)1/θ
.
(6.4)
wy,θ(ηλ) ≤ wy,θ(η),
so the third condition of Definition 6.2 is satisfied.
For the second condition, first note that our calculations above imply that
sup{wy,θ(γ) γ ∈ F n+1BΛ} ≤ sup{wy,θ(γ) γ ∈ F nBΛ}.
Moreover, for any non-negative matrix B, [Sch86] implies that
ρ(B)2 ≥ ρ(BBt).
The fact that Λ is strongly connected, and hence source-free by [HLRS15,
Lemma 2.1], implies that the matrix Bi(y, θ) has a nonzero entry in each row.
Therefore, every diagonal of Bi(y, θ)Bi(y, θ)t is nonzero, and ρ(Bi(y, θ)Bi(y, θ)t) ≥
maxv∈Λ0{Bi(y, θ)2
vv}. It follows that
ρi := ρ(Bi(y, θ)) ≥ max
v∈Λ0
Bi(y, θ)vv.
≤ ρ1/θ
j
.
We furthermore recall that, for each 1 ≤ j ≤ k; v, w ∈ Λ0; and f ∈ vΛej w,
we have e−y(f )θ ≤ Bj(y, θ)vw. It follows that
(6.5)
Writing n = qk + t for 0 ≤ t ≤ k − 1, the sequence
(6.6)
sup{e−y(f ) d(f ) = ej} ≤(cid:0)sup{Bj(y, θ)vw v, w ∈ Λ0}(cid:1)1/θ
(cid:16)sup{e−y(λ)ρ(B(y, θ))−d(λ)/θ λ ∈ F nBΛ}(cid:17)n∈N
tends to zero iff wy,θ satisfies the second condition of a weight, because the
fact that Λ is finite implies that the set {(ξy,θ
v )1/θ : v ∈ Λ0} is bounded (and
bounded away from zero).
Since each term in this sequence is bounded by 1, our assumption that
sup{Bi(y, θ)vw v, w ∈ Λ0} < ρi for at least one i, combined with Equation
(6.5), forces the sequence (6.6) to tend to zero. Consequently, the sequence
(sup{wy,θ(γ) γ ∈ F nBΛ})n∈N -- being bounded above by the product of the
28
FARSI, GILLASPY, LARSEN AND PACKER
sequence (6.6) and the maximum of {(ξy,θ
n → ∞.
v )1/θ}v∈Λ0 -- also tends to zero as
(cid:3)
Before stating the following Proposition, we remind the reader that if
Λ0 = 1, then the unimodular Perron-Frobenius eigenvector ξy,θ ∈ (R>0)Λ0
must be the constant vector (1). With this in mind, Equations (6.3) and
(6.7) give the same formula in the case of one-vertex k-graphs.
Proposition 6.7. Let Λ be a finite, strongly connected one-vertex k-graph.
Suppose that an R+-functor y and θ ∈ R>0 have been chosen such that
condition (w-I) holds. Then the function wy,θ : FBΛ → R>0 given by
(6.7)
wy,θ(λ) = e−y(λ)(cid:16)ρ(B(y, θ))−d(λ)(cid:17)1/θ
is a weight on BΛ.
Proof. First notice that condition (i) of Definition 6.2 holds immediately;
wy,θ(v) = 1 for the unique v ∈ Λ0. Condition (iii) follows immediately
from condition (w-I). To check condition (ii) of Definition 6.2, we simply
observe that, again thanks to condition (w-I), that(cid:0)ρ(B(y, θ))−d(λ)(cid:1)1/θ
→ 0
as d(λ) → ∞.
(cid:3)
The next Lemma establishes the crucial condition of our weights wy,θ,
which guarantees that the associated ultrametric dy,θ metrizes the cylinder
set topology on XBΛ ∼= Λ∞, the infinite path space of Λ. (However, Lemma
6.8 does not actually require that the function wy,θ be a weight.) We will
also rely on Lemma 6.8 to prove Corollary 6.17.
Lemma 6.8. Let Λ be a finite and strongly connected k-graph, y a R+-
functor on Λ and θ ∈ (0,∞). Let wy,θ denote the function from Equation
(6.3). For any finite path λ ∈ BΛ, and any m ∈ N,
(6.8)
wy,θ(λη)θ.
wy,θ(λ)θ = Xλη∈F λ+mBΛ
Proof. The fact that ξy,θ is an eigenvector for each matrix Bi(y, θ) with
eigenvalue ρi implies that, for any path λη ∈ FBΛ,
wy,θ(λ)θ = e−θy(λ)ρ(B(y, θ))−d(λ)ξy,θ
s(λ)
B(y, θ)d(η)
s(λ),vξy,θ
v
e−θy(η)ξy,θ
s(η)
= e−θy(λ)ρ(B(y, θ))−d(λ)−d(η) Xv∈Λ0
= e−θy(λ)ρ(B(y, θ))−d(λ)−d(η) Xλη∈F ληBΛ
= Xλη∈F ληBΛ
= Xη:λη∈F ληBΛ
wy,θ(λη)θ.
e−θy(λη)ρ(B(y, θ))−d(λ)−d(η)ξy,θ
s(η)
Furthermore, since Λ is strongly connected, there is a path λη ∈ F λ+mBΛ
for any m ∈ N. Since η was arbitrary, this finishes the proof.
(cid:3)
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
29
Proposition 6.9. Let Λ be a finite, strongly connected k-graph. Suppose
that θ ∈ R+ and a R+-functor y on Λ exist such that condition (w-I) holds
if Λ0 = 1 and otherwise condition (w-II) holds. Then the formula dy,θ
associated to wy,θ as in Proposition 6.4,
(6.9)
dy,θ(x, z) = inf(cid:26)e−y(λ)(cid:16)ρ(B(y, θ))−d(λ)ξy,θ
s(λ)(cid:17)1/θ
= inf{wy,θ(λ) x, z ∈ Z(λ), λ ∈ FBΛ}
x, z ∈ Z(λ), λ ∈ FBΛ(cid:27)
is an ultrametric on Λ∞. Furthermore, this ultrametric induces the cylinder
set topology on Λ∞.
Proof. For the first statement, combine Proposition 6.4 and Proposition 6.6
or Proposition 6.7.
For the second statement, we will prove that if wy,θ(γ) = r then for any
x ∈ Z(γ), we have B(x; r) = Z(γ). Observe first that
B(x; r) = {z ∈ Λ∞ dy,θ(x, z) ≤ r}
= {z ∈ Λ∞ ∃ λ ∈ FBΛ s.t. x, z ∈ Z(λ) and wy,θ(λ) ≤ r}
⊇ Z(γ).
To see that B(x; r) = Z(γ), choose z ∈ B(x; r); we will show that z ∈ Z(γ).
Write λ for the longest path in FBΛ such that x, z ∈ Z(λ); then dy,θ(x, z) =
wy,θ(λ). By hypothesis,
wy,θ(λ) ≤ r = wy,θ(γ).
Moreover, since x ∈ Z(λ)∩ Z(γ) we must have that one of λ, γ is a sub-path
of the other. If γ is a sub-path of λ then the fact that z ∈ Z(λ) implies that
z ∈ Z(γ) and the proof is finished.
On the other hand, if λ is a sub-path of γ, condition (iii) of Definition 6.2
forces wy,θ(λ) ≥ wy,θ(γ). Thus,
(6.10)
wy,θ(λ) ≤ wy,θ(γ) ≤ wy,θ(λ) ⇒ wy,θ(λ) = wy,θ(γ) = r.
By Lemma 6.8, since λ is a sub-path of γ, we know that
wy,θ(λη)θ.
wy,θ(λ)θ = Xλη∈F γBΛ
wy,θ(λη)θ = wy,θ(γ)θ + Xλη6=γ
Equation (6.10) now implies that Pλη6=γ wy,θ(λη)θ = 0; since wy,θ(ν) > 0
for any path ν ∈ FBΛ, it follows that
{λη ∈ F γBΛ λη 6= γ} = ∅.
In other words, Z(λ) = Z(γ), so z ∈ Z(γ) as desired.
Corollary 6.10. Suppose that Λ is a finite, strongly connected k-graph.
Choose θ ∈ R+ and a R+-functor y on Λ such that condition (w-I) holds if
Λ0 = 1 and otherwise condition (w-II) holds. Then for any path λ ∈ FBΛ,
(cid:3)
diam Z(λ) = wy,θ(λ).
30
FARSI, GILLASPY, LARSEN AND PACKER
Proof. The proof of Proposition 6.9 establishes that for any x ∈ Z(λ),
Z(λ) = B(x; r)
(cid:3)
where r = wy,θ(λ). Therefore diam Z(λ) = r = wy,θ(λ).
6.2. The Hausdorff structure of (Λ∞, dy,θ). We conclude the paper by
computing the Hausdorff measure and dimension of the ultrametric Cantor
sets (Λ∞, dy,θ), using the detailed understanding of the weights dy,θ obtained
in Section 6.1, and showing that the Hausdorff measure is precisely the quasi-
invariant measure µy,θ. For this computation, the condition (6.8) satisfied
by wy,θ is crucial; indeed, we are able to compute the Hausdorff measure and
dimension of (XB, dǫ) whenever ǫ is a weight on the Bratteli diagram B which
satisfies Equation (6.8). To formalize this, we introduce (see Definition
6.12) the notion of an exponentially self-similar weight with exponent θ that
is modeled on Equation (6.8). We mention that self-similar conditions on
weighted Cantor sets or Bratteli diagrams have been introduced before, see
for example [JS11a, Definition 2.6]. The existing definitions do not apply to
our case of interest, however (see Remark 6.13 below).
Definition 6.11. [Rog70, Definition 16] Let (X, d) be a metric space and
fix s ∈ R≥0. The Hausdorff measure of dimension s of a compact subset Z
of X is
H s(Z) = lim
δ→0
infXUi∈F
(diam Ui)s F < ∞, ∪iUi = Z, diam Ui < δ for all i
It is standard to show that H s(Z) is a decreasing function of s, and that
there is a unique s ∈ R such that H t(X) = ∞ for all t < s and that
H t(X) = 0 for all t > s. This value of s is called the Hausdorff dimension
of X.
Definition 6.12. Let B be a Bratteli diagram and let XB be its infinite
path space equipped with the cylinder set topology.
.
(i) We say B is a Cantorian Bratteli diagram if XB is a Cantor set.
(ii) An exponentially self-similar weight with exponent θ ∈ [0,∞) is a
weight ǫ on a Cantorian Bratteli diagram B satisfying the equation
for all finite paths λ ∈ FB and all m ∈ N.
Remark 6.13. There does not appear to be any relation between our ex-
ponentially self-similar weights and the self-similar metrics on Bratteli di-
agrams discussed in [JS11a]. Lemma 6.8 implies that for any R+-functor
y and any θ ∈ R+, the weights wy,θ are exponentially self-similar weights
with exponent θ. However, the associated metric dy,θ does not satisfy the
self-similarity condition in [JS11a, Definition 2.6], because inf{e−y(λ) : λ ∈
FBΛ} = 0. Moreover, the conditions placed on the constants {aγ : γ ∈ FB}
in [JS11a, Definition 2.6] are too weak to guarantee Equation (6.11).
The proofs of the following Proposition and Corollary are identical to the
proofs of Proposition 6.9 and Corollary 6.10, so we omit the details.
(6.11)
ǫ(λ)θ = Xλη∈F λ+mB
ǫ(λη)θ
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
31
Proposition 6.14. Suppose that B is a Cantorian Bratteli diagram with
weight ǫ. The formula
dǫ(x, y) := ǫ(x ∧ y)
(6.12)
defines an ultrametric on the infinite path space XB of B. Furthermore, if ǫ
is exponentially self-similar with exponent θ, then the ultrametric dǫ induces
the cylinder set topology on XB.
Corollary 6.15. Suppose that B is a Cantorian Bratteli diagram with ex-
ponentially self-similar weight ǫ. Then in (XB, dǫ) we have
diam Z(λ) = ǫ(λ).
We now prove that for Cantorian Bratteli diagrams with exponentially
self-similar weights, the exponent and weight are intimately tied to the Haus-
dorff dimension and measure of the ultrametric space (XB, dǫ); compare the
results below with [Edg08, pages 203-206].
Theorem 6.16. Suppose that B is a weighted Cantorian Bratteli diagram
for a self-similar weight ǫ with exponent θ. Then the Hausdorff dimension
of the ultrametric Cantor set (XB, dǫ) is θ.
Proof. We first compute the Hausdorff measure of dimension θ of (XB, dǫ).
Consider a finite cover {Ui}i=1,...,n of XB, and the associated sum
(diam Ui)θ.
nXi=1
Observe first that we can assume each Ui to be a cylinder set Z(µi). Indeed,
given Ui, we can pick x = x1x2 ··· ∈ Ui, and define Bx,i to be the closed ball
of center x and radius diam Ui, namely,
Bx,i = {y ∈ XB dǫ(x, y) ≤ diam Ui}.
We will show that Bx,i = Z(x1 ··· xℓi) for some ℓi ∈ N; that diam Bx,i ≤
diam Ui; and that Bx,i ⊇ Ui. Thus, in order to minimize the sum used
to compute the Hausdorff measure, we may assume without loss of gener-
ality that each open set Ui is a cylinder set, by replacing Ui with Bx,i =
Z(x1 ··· xℓi).
The fact that Bx,i ⊇ Ui is immediate from the observation that dǫ(x, z) ≤
diam Ui for all z ∈ Ui. To estimate the diameter of Bx,i, choose y, z ∈ Bx,i
and observe that
dǫ(y, z) ≤ max{dǫ(x, y), dǫ(x, z)} ≤ diam Ui.
Taking supremums reveals that diam Bx,i ≤ diam Ui.
ǫ, there is a smallest ℓi ∈ N such that
We now check that Bx,i = Z(x1 ··· xℓi). By the definition of the weight
ǫ(x1 ··· xℓi) ≤ diam Bx,i.
If y ∈ Bx,i, by Equation (6.12) we have that
diam Bx,i ≥ dǫ(x, y) = wǫ(x ∧ y) = ǫ(x1 ··· xmi)
Z(µi) = XB, diam Z(µi) < δ for all i)
(6.13)
inf( nXi=1
(diam Z(µi))θ [i
Given one cover U = {Z(µi)}n
and take the limit as δ → 0 of these infima.
i=1 of XB in the set (6.13), let M be the
maximum of the lengths of the paths µ1, . . . , µn. Moreover, by equation
(6.11), replacing each Z(µi) by the collection
{Z(µiηij) ηij = M − µi}j
(which makes all of the cylinder sets in the open cover U of the same length)
(Note that
does not change the sum arising in our computation (6.13).
Z(µi) = Fj Z(µiηij), so the new collection of sets does indeed cover XB
whenever {Z(µi)}n
Therefore we can assume that all of the cylinder sets Z(µ1), . . . , Z(µn)
are associated to finite paths in B which all have the same length M , and
that all of the cylinder sets are also pairwise disjoint, since
µi = µj ⇒ Z(µi) ∩ Z(µj) = δi,jZ(µj).
i=1 does.)
Since {Z(µi)}i covers XB, the collection U = {Z(µ1), . . . , Z(µn)} must there-
fore contain precisely all of the cylinder sets of length M in XB.
Now by applying Corollary 6.15, we see that
H θ(XB) = lim
δ→0
= lim
δ→0
= lim
δ→0
inf XZ(µi)∈U
(diam Z(µi))θ diam Z(µi) < δ ∀ i
inf Xµ=M
(ǫ(µ))θ max{ǫ(µ) µ = M} < δ
infXv∈Λ0
ǫ(v)θ
,
32
FARSI, GILLASPY, LARSEN AND PACKER
for some N ∋ mi ≥ ℓi (thanks to Definition 6.2 and the minimality of ℓi). It
follows that y ∈ Z(x1 ··· xℓi). On the other hand, if z ∈ Z(x1 ··· xℓi), then
dǫ(z, x) = ǫ(z ∧ x) ≤ ǫ(x1 ··· xℓi) ≤ diam Ui,
so z ∈ Bx,i by construction. In other words, Bx,i = Z(x1 ··· xℓi) as claimed;
set µi := x1 ··· xℓi, i = 1, . . . , n.
Thus, in estimating H θ(XB), we need to compute
which is strictly between 0 and infinity. Definition 6.11 now implies that θ is
the Hausdorff dimension of (XB, dǫ). Moreover, the arguments above imply
that the Hausdorff measure H θ is given on cylinder sets by
H θ(Z(λ)) = ǫ(λ)θ.
(cid:3)
Corollary 6.17. Let Λ be a finite, strongly connected k-graph, y a R+-
functor on Λ and θ ∈ (0,∞) such that condition (w-I) holds if Λ0 = 1
and otherwise condition (w-II) holds. Then the Hausdorff dimension of the
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
33
ultrametric Cantor set (Λ∞, dy,θ) is θ. Moreover, the Hausdorff measure H θ
agrees with µy,θ.
Proof. Since, by Lemma 6.8, wy,θ is a self-similar weight on Λ∞ ∼= XBΛ with
exponent θ, the assertion about Hausdorff dimension follows directly from
Theorem 6.16. To see that H θ = µy,θ, observe that
(6.14) H θ(Z(λ)) = wy,θ(λ)θ = e−θy(λ)ρ(B(y, θ))−d(λ)ξy,θ
s(λ) = µy,θ(Z(λ)).
Since the cylinder sets generate the ultrametric topology on Λ∞ by Propo-
sition 6.9, µy,θ = H θ as claimed.
(cid:3)
References
[BNR+16] J.H. Brown, G. Nagy, S. Reznikoff, A. Sims, and D.P. Williams, Cartan sub-
algebras in C ∗-algebras of Hausdorff ´etale groupoids, Integral Equations Op-
erator Theory 85 (2016), no. 1, 109 -- 126.
O. Bratteli and D.W. Robinson, Operator algebras and quantum statistical
mechanics. 2, equilibrium states. Models in quantum statistical mechanics,
Springer-Verlag, Berlin.
J. Christensen, Symmetries of the KMS simplex, arXiv:1710.04412, 2017.
[Chr17]
[CKSS14] T.M. Carlsen, S. Kang, J. Shotwell, and A. Sims, The primitive ideals of the
Cuntz-Krieger algebra of a row-finite higher-rank graph with no sources, J.
Funct. Anal. 266 (2014), no. 4, 2570 -- 2589.
D. Dutkay and P.E.T. Jorgensen, Monic representations of the Cuntz algebra
and Markov measures, J. Funct. Anal. 267 (2014), no. 4, 1011 -- 1034.
[BR]
[DJ14]
[DPY10] K.R. Davidson, S.C. Power, and D. Yang, Dilation theory for rank 2 graph
[DY09a]
algebras, J. Operator Theory 63 (2010), no. 2, 245 -- 270.
K.R. Davidson and D. Yang, Periodicity in rank 2 graph algebras, Canad. J.
Math. 61 (2009), no. 6, 1239 -- 1261.
[DY09b]
, Representations of higher rank graph algebras, New York J. Math. 15
(2009), 169 -- 198.
[EBH80] G.A. Elliott, O. Bratteli, and R.H. Herman, On the possible temperatures of
[Edg08]
a dynamical system, Comm. Math. Phys. 74 (1980), 281 -- 295.
G. Edgar, Measure, topology, and fractal geometry, Springer, London-New
York, 2008.
[EFW84] M. Enomoto, M. Fujii, and Y. Watatani, KMS states for gauge action on OA,
[EL03]
[Exe04]
Math. Japon. 29 (1984), no. 4, 607 -- 619.
Ruy Exel and Marcelo Laca, Partial dynamical systems and the KMS condi-
tion, Comm. Math. Phys. 232 (2003), no. 2, 223 -- 277.
R. Exel, KMS states for generalized gauge actions on Cuntz-Krieger algebras,
Bull. Brazilian Math. Soc. (New Series) 35 (2004), no. 1, 1 -- 12.
[FGJ+17] C. Farsi, E. Gillaspy, A. Julien, S. Kang, and J. Packer, Wavelets and spec-
tral triples for fractal representations of Cuntz algebras, Problems and recent
methods in operator theory, Contemp. Math., vol. 687, Amer. Math. Soc.,
Providence, RI, 2017, pp. 103 -- 133.
[FGJ+18a] C. Farsi, E. Gillaspy, P.E.T. Jorgensen, S. Kang, and J. Packer, Representa-
tions of higher-rank graph C ∗-algebras associated to λ-semibranching function
systems, arXiv:1803:08779, 2018.
[FGJ+18b] C. Farsi, E. Gillaspy, A. Julien, S. Kang, and J. Packer, Spectral triples and
wavelets for higher-rank graphs, arXiv:1803:09304, 2018.
[FGKP16] C. Farsi, E. Gillaspy, S. Kang, and J. Packer, Separable representations, KMS
states, and wavelets for higher-rank graphs, J. Math. Anal. Appl. 434 (2016),
no. 1, 241 -- 270.
[HLRS14] A. an Huef, M. Laca, I. Raeburn, and A. Sims, KMS states on C ∗-algebras
associated to higher-rank graphs, J. Funct. Anal. 266 (2014), no. 1, 265 -- 283.
34
FARSI, GILLASPY, LARSEN AND PACKER
[HLRS15] A. an Huef, M. Laca, I. Raeburn, and A. Sims, KMS states on the C ∗-algebra
of a higher-rank graph and periodicity in the path space, J. Funct. Anal. 268
(2015), 1840 -- 1875.
P.M. Hajac, R. Nest, D. Pask, A. Sims, and B. Zieli´nski, The K-theory
of twisted multipullback quantum odd spheres and complex projective spaces,
arXiv:1512.08816.
[HNP+]
[HRSW13] R. Hazlewood, I. Raeburn, A. Sims, and S.B.G. Webster, Remarks on some
fundamental results about higher-rank graphs and their C ∗-algebras, Proc. Ed-
inb. Math. Soc. (2) 56 (2013), no. 2, 575 -- 597.
M. Ionescu and A. Kumjian, Hausdorff measures and KMS states, Indiana
Univ. Math. J. 62 (2013), no. 2, 443 -- 463.
A. Julien and J. Savinien, Embeddings of self-similar ultrametric Cantor sets,
Topology and its Applications 158 (2011), 2148 -- 2157.
[JS11a]
[IK13]
[JS11b]
[KP00]
[KPS15]
, Transverse Laplacians for substitution tilings, Comm. Math. Phys.
301 (2011), no. 2, 285 -- 318.
A. Kumjian and D. Pask, Higher rank graph C ∗-algebras, New York J. Math.
6 (2000), 1 -- 20.
A. Kumjian, D. Pask, and A. Sims, On twisted higher-rank graph C ∗-algebras,
Trans. American Math. Soc. 367 (2015), no. 7, 5177 -- 5216.
[Oka02]
[Oka03]
[Mic93]
[Nes13]
[OP78]
[PB09]
[McN15]
[LLN+15] M. Laca, N.S. Larsen, S. Neshveyev, A. Sims, and S.B.G. Webster, Von
Neumann algebras of strongly connected higher-rank graphs, Math. Ann. 363
(2015), no. 1-2.
R. McNamara, KMS states of graph algebras with a generalised gauge dynam-
ics, Ph.D. thesis, University of Otago, 2015.
G. Michon, Mesures de gibbs sur le Cantor r´eguliere, Annales de L'I.H.P. 58
(1993), 267 -- 285.
S. Neshveyev, KMS states on the C ∗-algebra of non-principal groupoids, J.
Operator Theory 70 (2013), 513 -- 530.
Rui Okayasu, Cuntz-Krieger-Pimsner algebras associated with amalgamated
free product groups, Publ. Res. Inst. Math. Sci. 38 (2002), no. 1, 147 -- 190.
R. Okayasu, Type III factors arising from Cuntz-Krieger algebras, Proc. Amer.
Math. Soc. 131 (2003), no. 7, 2145 -- 2153.
D. Olesen and G.K. Pedersen, Some C ∗-dynamical systems with a single KMS
state, Math. Scand. 42 (1978), no. 1, 111 -- 118.
J. Pearson and J. Bellissard, Noncommutative Riemannian geometry and dif-
fusion on ultrametric Cantor sets, J. Noncommut. Geom. 3 (2009), 447 -- 480.
J. Renault, A groupoid approach to C ∗-algebras, Lecture Notes in Mathemat-
ics, vol. 793, Springer-Verlag, 1980.
C.A. Rogers, Hausdorff measures, Cambridge University Press, London-New
York, 1970.
G. Robertson and T. Steger, Affine buildings, tiling systems and higher rank
Cuntz-Krieger algebras, J. Reine Angew. Math. 513 (1999), 115 -- 144.
D.I. Robertson and A. Sims, Simplicity of C ∗-algebras associated to higher-
rank graphs, Bull. Lond. Math. Soc. 39 (2007), 337 -- 344.
E. Ruiz, A. Sims, and A. P. W. Sørensen, UCT-Kirchberg algebras have nuclear
dimension one, Adv. Math. 279 (2015), 1 -- 28.
I. Raeburn, A. Sims, and T. Yeend, Higher-rank graphs and their C ∗-algebras,
Proc. Edinburgh Math. Soc. 46 (2003), 99 -- 115.
A.J. Schwenk, Tight bounds on the spectral radius of asymmetric nonnegative
matrices, Linear Algebra Appl. 75 (1986), 257 -- 265.
K. Thomsen, KMS weights on groupoid and graph C ∗-algebras, J. Funct. Anal.
266 (2014), 2959 -- 2988.
D. Yang, Endomorphisms and modular theory of 2-graph C ∗-algebras, Indiana
Univ. Math. J. 59 (2010), 495 -- 520.
[RSY03]
[RSS15]
[RS99]
[RS07]
[Ren80]
[Rog70]
[Sch86]
[Tho14]
[Yan10]
[Yan12]
, Type III von Neumann algebras associated with 2-graphs, Bull. Lond.
Math. Soc. 44 (2012), no. 4, 675 -- 686.
GENERALIZED GAUGE ACTIONS AND HAUSDORFF STRUCTURE
35
[Yan17]
[Yeo15]
, Factoriality and type classification of k-graph von Neumann algebras,
Proc. Edinb. Math. Soc. (2) 60 (2017), no. 2, 499 -- 518.
D. Yeo, Perturbation of eigenvectors, https://eventuallyalmosteverywhere.wordpress.com/
2015/11/09/perturbation-of-eigenvectors/, 2015.
|
0911.4978 | 6 | 0911 | 2013-09-17T05:46:29 | Product between ultrafilters and applications to the Connes' embedding problem | [
"math.OA",
"math.GR"
] | In this paper we want to apply the notion of product between ultrafilters to answer several questions which arise around the Connes' embedding problem. For instance, we will give a simplification and generalization of a theorem by Radulescu; we will prove that ultraproduct of hyperlinear groups is still hyperlinear and consequently the von Neumann algebra of the free group with uncountable many generators is embeddable into $R^{\omega}$. This follows also from a general construction that allows, starting from an hyperlinear group, to find a family of hyperlinear groups. We will introduce the notion of hyperlinear pair and we will use it to give some other characterizations of hyperlinearity. We will prove also that the cross product of a hyperlinear group via a profinite action is embeddable into $R^{\omega}$. | math.OA | math |
PRODUCT BETWEEN ULTRAFILTERS AND APPLICATIONS
TO THE CONNES′ EMBEDDING PROBLEM
V. CAPRARO - L. P AUNESCU1
Abstract. In this paper we want to apply the notion of product between ultrafilters
to answer several questions which arise around the Connes' embedding problem. For
instance, we will give a simplification and generalization of a theorem by Radulescu; we
will prove that ultraproduct of hyperlinear groups is still hyperlinear and consequently the
von Neumann algebra of the free group with uncountable many generators is embeddable
into Rω. This follows also from a general construction that allows, starting from an
hyperlinear group, to find a family of hyperlinear groups. We will introduce the notion of
hyperlinear pair and we will use it to give some other characterizations of hyperlinearity.
We shall prove also that the cross product of a hyperlinear group via a profinite action is
embeddable into Rω.
1 Preliminaries
We start by introducing the notion of product between ultrafilters. It is already known in
Model Theory (see, for example, [DiNa-Fo]), but it seems nobody applied it to Operator
Algebras.
Definition 1.1. Let U ,V be two ultrafilters respectively on I and J. The tensor product
U ⊗ V is the ultrafilter on I × J defined by setting
X ∈ U ⊗ V ⇔ {i ∈ I : {j ∈ J : (i, j) ∈ X} ∈ V} ∈ U
Observe that this is indeed a maximal filter, i.e. an ultrafilter.
1Second author is supported by the Marie Curie Research Training Network MRTN-CT-2006-031962
EU-NCG.
1
Remark 1.2. This definition is equivalent to the following one:
X ∈ U ⊗ V ⇔ ∃A ∈ U s. t. ∀i ∈ A, πJ (X ∩ π−1
I (i)) ∈ V
where πI , πJ are the projections of I × J on the first and second component.
We prefer this second definition since it is easier to apply to prove the following
Theorem 1.3. Let {xj
i}(i,j)∈I×J ⊆ R bounded. Then
limi→U limj→Vxj
i = lim(i,j)→U ⊗Vxj
i
Proof. Let x = limi→U limj→V xj
i . Fixed ε > 0, we notice from the definitions that
A = {i ∈ I : limj→V xj
i − x <
and
i − limj→Vxj
Combining this two (by triangle inequality) we get
Ai = {j ∈ Jxj
i <
ε
2} ∈ U
ε
2} ∈ V
X = {(i, j) ∈ I × J : i ∈ A, j ∈ Ai} ⊆ {(i, j) ∈ I × J : xj
i − x < ε}
Since X ∈ U ⊗ V and ε was arbitrary, it follows the thesis.
Notation 1.4. By ω, ω′ we shall denote free ultrafilters on N. R stands for the hyperfinite
type II1 factor. We shall use the classical notation Rω for the ultrapower of R with regard
to ω and denote by τ its trace. By L(G) we denote the von Neumann group algebra of G.
2 Main result and immediate consequences
The main result is actually an easy consequence of Th.1.3, but it gives a tool to pass by
the limit on representations. We shall give some applications of this procedure.
Proposition 2.1. Let ω, ω′ two ultrafilters on N. Then
(Rω)ω′ ∼= Rω⊗ω′
Proof. Those von Neumann algebras have the same algebraic structure. So we only have
to prove that they have the same trace. It is just a consequence of Th.1.3.
2
We want to apply this result to hyperlinear groups. In order to fully benefit from it
we will introduce the notion of hyperlinear pair.
Definition 2.2. By a central pair we mean (G, ϕ), where G is a group and ϕ : G → C is
a positive defined function, central (i.e. constant on conjugacy classes) and ϕ(e) = 1. Let
Cen(G) be the set of those functions on G.
Remark 2.3. An important element of Cen(G) is the function δe, defined by setting
δe(g) = 0,∀g 6= e.
Remark 2.4. If (G, ϕ) is a central pair then we have a canonical bi-invariant and bounded
metric induced on G by:
d(g, h)2 = 2 − ϕ(g−1h) − ϕ(h−1g)
∀g, h ∈ G.
We recall that one can define the notion of ultraproduct of groups with bi-invariant
metric (see [Pe]). We can use this definition for our particular case of central pairs.
Definition 2.5. Let (Gn, ϕn)n∈N a sequence of central pairs and ω an ultrafilter. By the
ultraproduct of the family we mean the central pair:
(G, ϕ) = (ΠnGn/N, limωϕn),
where ΠnGn is just the cartesian product and N = {(gn) ∈ ΠnGn : limωϕn(gn) → 1}.
We shall denote by Πω(Gn, ϕn) the ultraproduct of central pairs.
Note 2.6. It is easy to recognize that our definition of N coincides with the classical one:
N = {(gn)n ∈ ΠGn : limωdn(gn, en) → 0}.
Definition 2.7. A central pair
(G, ϕ)
homomorphism θϕ : G → U (Rω) such that
is called hyperlinear
if
there exists an
τ (θϕ(g)) = ϕ(g)
∀g ∈ G.
Let Hyp(G) = {φ ∈ Cen(G) : (G, φ) is a hyperlinear pair}
Remark 2.8. We recall the original definition by Radulescu: a countable i.c.c. group
G is called hyperlinear if there exists a monomorphism G → U (Rω). It happens if and
3
only if δe ∈ Hyp(G) (see [Ra], Prop.2.5). Countability and i.c.c. properties are not
necessary, but they come from the reason of this definition: to study when the group
algebra is embeddable into Rω. This problem, well-known as Connes' embedding problem
for groups, regard only separable type II1 group factor.
Remark 2.9. If (G, ϕ) is a hyperlinear pair, then (G, ϕ) is also a central pair and the
induced distance is just the distance in norm 2 in Rω.
We can now use Prop. 2.1 in order to get the following
Proposition 2.10. Ultraproduct of hyperlinear pairs is a hyperlinear pair.
Proof. Take a sequence (Gn, ϕn) of hyperlinear pairs and just embed each pair in an Rω.
The ultraproduct of the family with respect to ω′ will sit inside (Rω)ω′ ∼= Rω⊗ω′
.
In case we cannot find a "good" ω for all hyperlinear pairs, we just need to adapt our
notion of product between two ultrafilters to a notion of ultraproduct of ultrafilters. We
shall not do this, as it is just a technical trick and assuming continuum hypothesis this
Rω are isomorphic between themselves anyway.
In order to give some information on the structure of Hyp(G), we recall that a monoid
is a set with a binary associative operation admitting a neutral element.
If (X,·) is a
monoid, an element x ∈ X is called annihilator if x · y = y · x = x,∀y ∈ X. The set
of annihilators of X is denoted by 0(X). Clearly Cen(G) is a monoid with respect the
pointwise product and δe ∈ 0(Cen(G)).
Proposition 2.11. Hyp(G) is a submonoid of Cen(G). It is closed under ultralimits and
convex combinations. Moreover, for G countable, 0(Hyp(G)) = {δe} if and only if G is
hyperlinear in the classical sense of Radulescu.
Proof. The constant function 1 forms with G a hyperlinear pair via the trivial
representation. Hyp(G) is closed under pointwise multiplication because Rω ⊗ Rω ⊂ Rω,
τ (x ⊗ y) = τ (x)τ (y) and so θϕ·ψ = θϕ ⊗ θψ will do the work.
For the second part note that (G, limωϕn) ⊂ Πω(G, ϕn) and use our last proposition.
For convex combination define an homomorphism of G in Rω ⊕ Rω with the same convex
combination of traces.
The last part is an easy consequence of the following Prop.2.13.
4
Corollary 2.12. An i.c.c. group G embeds in U (Rω) if and only if L(G) embeds into Rω.
Proof. If L(G) ⊆ Rω then clearly G ⊂ U (Rω). Conversely, let θ : G → U (Rω) an
embedding. Let τ be the normalized trace on Rω. Then ϕ(g) = τ (θ(g)) < 1 for
any g 6= e (since G is i.c.c.) and ϕ ∈ Hyp(G). Because of Prop. 2.11 we have that
ϕn ∈ Hyp(G) and limn→ωϕn ∈ Hyp(G).
Now ϕ(g) < 1 for g 6= e so limnϕ(g)n = 0. This means that limnϕn = δe, so
δe ∈ Hyp(G). This is equivalent to L(G) embeds in Rω.
This is a simplification of the initial proof given by Radulescu in [Ra] and also note
that our proof doesn't need the contability of G.
Proposition 2.13. A countable group G is hyperlinear if and only if for any g ∈ G \ {e}
there is a hyperlinear pair (G, ϕg) such that ϕg(g) < 1.
Proof. The only if part is trivial. Conversely, we need to show that δe ∈ Hyp(G). Take
G = Sn Fn, with Fn increasing sequence of finite subsets of G. Define ϕFn = Πg∈Fnϕg.
According to Prop.2.11 ϕFn ∈ Hyp(G) and by the same proposition so is ϕ = limn→ωϕFn.
Now because of the hypothesis ϕg(g) < 1 and because of Fn is an increasing
sequence we deduce ϕ(g) < 1. As in the above corollary we now have δe = limnϕn,
so δe ∈ Hyp(G).
We end this section by presenting a motivation for our definition of Hyp(G). Let F∞
be the free group with countable many generators.
Proposition 2.14. If Cen(F∞) = Hyp(F∞) then every countable group is hyperlinear.
Proof. Let G be a countable group. Let H be a normal subgroup of F∞ such that
G ∼= F∞/H. Let ϕH : F∞ → C be the characteristic function of H. We shall prove
that ϕH ∈ Cen(F∞). It is easy to see that δe ∈ Hyp(G) if and only if ϕH ∈ Hyp(F∞).
This will finish the proof.
Now H is normal in F∞. So for any g, h ∈ F∞ h ∈ H if and only if ghg−1 ∈ H. This
prove that ϕH is central. To prove that it is also positive defined take g1, . . . , gn ∈ F∞.
Consider the matrix {ϕH (g−1
i gj)}i,j and notice that is the matrix of an equivalence relation
5
on a set with n elements (because H is a subgroup). By permuting elements (gi)i we can
assume that is a block matrix. This means that Pn
i gj) is nonnegative. So
ϕH is positive defined.
i,j=1 λiλjϕ(g−1
Note 2.15. Our sets Cen(G) and Hyp(G) can be generalized to a type II1 factor instead
of just group algebras. Let M be such a factor and consider B = {xn}n∈N ⊂ M a
basis in L2(M, tr). Suppose that x0 = id. We shall consider now ϕ : B → C such
that ϕ(x0) = 1 and the linear extension of ϕ to M is positive and tracial (may not
be faithful). The problem is that such a linear extension may not be well defined. We
formalize this as follows: ϕ ∈ Cen(M ) iff whenever ϕ(x∗x) is well defined then so is ϕ(xx∗)
and ϕ(x∗x) = ϕ(xx∗) ≥ 0.
For ϕ ∈ Cen(M ) we can define Mϕ by the GN S-construction. We define ϕ ∈ Hyp(M )
iff this Mϕ is embedable in Rω. As we saw, for M = L(G) and ϕH for H a normal subgroup
of G then L(G)ϕH = L(G/H).
As another example we may take the crossed product M = L∞(X) ⋊ G of a non-free
measure preserving action. Take {fi : i ∈ N} a basis for L∞(X) and B = {fiug : i ∈
N, g ∈ G}. Define ϕ(fiug) = RXg
fi where Xg = {x ∈ X : gx = x}. Then Mϕ = M (EG),
the Feldmann-Moore construction for the equivalence relation induced by G on X.
3 Other applications
3.1 Construction of uncountable hyperlinear groups
Now we want to present a construction that, starting from an hyperlinear group G,
allows to construct a family of countable and uncountable hyperlinear groups. An easy
application of this construction is that the von Neumann algebra of the free group with
uncountable many generators Fℵc is embeddable into Rω. The Hilbert-Schmidt distance
between two distinct universal unitaries of Fℵc will be equal to √2, giving another proof
of the non-separability of Rω.
Definition 3.1. Let G be a countable group with generators g1, g2, .... Let ℑ be a
family of infinite subsets of N such that F1, F2 ∈ ℑ implies F1 ∩ F2 is finite. Now let
F = {f1, f2, ...} ∈ ℑ, define the sequence (gF
n modulo
n )n = gfn. Let gF be the sequence gF
6
ω. We can multiply gF1, gF2 component-wise, by using the relations on G. The group
generated by the elements gF is denoted by G(ω,ℑ).
Notice that G(ω,ℑ) does not depend only on ω and ℑ, but also on the set of generators
chosen.
Remark 3.2. The generators gF of G(ω,ℑ) are different elements in G(ω,ℑ). This is
because gF1
n holds only for a finite number of indexes, by the definition of ℑ. Since
a free ultrafilter does not contain finite sets, gF1 and gF2 must be different.
n = gF2
Remark 3.3. G(ω,ℑ) can be countable (if the family ℑ is countable), but also
uncountable. Indeed one can use the Zorn's lemma to prove the existence of an uncountable
family ℑ which verifies the property F1, F2 ∈ ℑ implies F1∩F2 is finite. An elegant example
privately suggested by Ozawa is the following: take t ∈ [ 1
10 , 1), for example t = 0, 132483...,
define
It = {1, 13, 132, 1324, 13248, 132483, ...}
i.e. It is the set of the approximation of t. Then {It}t∈[ 1
subsets of N such that It ∩ Is is finite for all t 6= s.
Proposition 3.4. If G is hyperlinear, then also G(ω,ℑ) is hyperlinear.
10 ,1) is an uncountable family of
Proof. We want to prove that G(ω,ℑ) ⊂ Πω(G, δ) and the last is a hyperlinear pair
because of Prop.2.10. Moreover we shall prove that if in an ultraproduct of central pairs
just δe appears, then the central positive defined function of the ultraproduct will also be
δe. This two affirmations will show that δe ∈ Hyp(G(ω,ℑ)), i.e. G(ω,ℑ) is hyperlinear.
Recall that Πω(Gn, ϕn) = (ΠnGn/N, limωϕn), where ΠnGn is just the cartesian
product and N = {(gn) ∈ ΠnGn : limωϕn(gn) → 1}. So let Gn a copy of G and ϕn = δe
for each n. Then limωϕn ∈ {0, 1}. If this limit is 1 for some element, then that element is
in N i.e. it is the identity in the ultraproduct. So indeed limωδe = δe proving our second
affirmation.
Now from the construction of G(ω,ℑ) we see that G(ω,ℑ) ⊂ ΠnGn. If an element
g = (gn)n of G(ω,ℑ) is in N then limωδe(gn) = 1 meaning that gn = e in G for any n
in a set in ω. From the definition of G(ω,ℑ) this means that g = e. We proved that
G(ω,ℑ) ⊂ Πω(G, δ).
7
It is well known that F∞, free group with countable many generators is hyperlinear.
We shall denote with Fℵc the free group with ℵc many generators (set of continuum power).
Corollary 3.5. Fℵc is hyperlinear. In particular Rω is not separable.
Proof. If Card(ℑ) = ℵc then F∞(ω,ℑ) = Fℵc, and we can apply the previous proposition.
Representing L(Fℵc) on Rω, the Hilbert-Schmidt distance between two elements of
Fℵc will be √2. Separability in the weak or in the strong topology is the same and the
last one coincide with the Hilbert-Schmidt topology on the bounded sets (see [Jo]).
Note 3.6. Non-separability of Rω is already well-known. The first proof is probably due
to Feldman (see [Fel]); S. Popa proved in [Po] that every MASA in Rω is not separable.
Anyway, we want to underline the importance of non-separability of Rω around the Connes'
embedding conjecture: every separable type II1 factor can be embedded into Rω (see [Co]).
This conjecture imply the existence of a universal type II1 factor. If a factor embeds in
Rω then it embeds in any Rω′
. We are grateful to Pestov for communicating this fact to
us. Ozawa proved in [Oz] that such a universal factor cannot be separable, also proved by
Nicoara, Popa and Sasyk in [Ni-Po-Sa]). So, if Rω was been separable, Connes embedding
conjecture would be false.
Problem 3.7. What kind of groups have the shape G(ω,ℑ)? Is it true that if {Ra}a∈A
is the set of distinct relations on G and B ⊆ A, then there exist ω and ℑ such that the
set of relations of G(ω,ℑ) is {Ra}a∈B?
3.2 Cross product via profinite actions
We want to apply Prop.2.1 also to some other type II1 factors than group algebras. For
this we ask ourselves when the crossed product L∞(X) ⋊α G for a free action α embeds
in Rω. Of course when this happens G has to be hyperlinear. We shall prove the converse
in the easy case in which α is profinite.
Definition 3.8. Let α be an action of a group G on a von Neumann algebra P . Then
α is called profinite if there is an increasing sequence of finite dimensional G-invariant
subalgebras A1 ⊂ A2 ⊂ . . . such that P = (Sn An)′′.
8
Proposition 3.9. Let G be a hyperlinear group and α be a profinite action of G on X.
Then L∞(X) ⋊α G is embeddable into Rω.
Proof. The crossed product is generated on L2(X) ⊗ l2G by the operators α(g) ⊗ λ(g) for
g ∈ G and mf ⊗ 1 for f ∈ L∞(X) (here λ is the regular representation of G on l2G and
mf is the multiplication operator).
Let L∞(X) = (Sn An)′′ with An G-invariant and finite dimensional. We can then
form An ⋊α G and L∞(X) ⋊α G = (Sn An ⋊α G)′′. Looking at the above definition of
crossed product we can deduce that An ⋊α G ⊂ Mkn ⊗ L(G). Here entered the fact that
An is finite dimensional. Now, because G is hyperlinear Mkn ⊗ L(G) ⊂ R⊗ Rω ⊂ Rω. We
can than embed Sn An ⋊α G in (Rω)ω′
so that L∞(X) ⋊α G ⊂ Rω⊗ω′
.
4 Acknowledgements
It is our pleasure to thanks professor Florin Radulescu for many useful discussions and
remarks on the topics presented in this paper.
References
[Co] A. Connes, Classification of injective factors, Ann. of Math. 104 (1976), 73-115.
[DiNa-Fo] M. Di Nasso - M. Forti, Hausdorff ultrafilters, Proc. Amer. Math. Soc. 134
(2006), 1809-1818.
[Fa-Ha-Sh] I. Farah - B.Hart - D. Sherman, Model theory of operator algebras I: Stability,
arXiv:math/0908.2790
[Fe] J. Feldman, C. Moore, Ergodic equivalence Relations, Cohomology, and Von Neumann
Algebras II, Trans. Amer. Math. Soc. Vol. 234, No. 2 (1977) pp.325-359.
[Fel] J. Feldman, Nonseparability of certain finite factors, Proc. Amer. Math. Soc. 7 (1956),
23 -- 26.
[Ge-Ha] L. Ge - D. Hadwin, Ultraproducts of C ∗-algebras, Oper. Theory Adv. Appl. 127
(2001), 305-326.
9
[Io] A.
Ioana, Cocyle Superrigidity for Profinite Actions of property (T) Groups,
arXiv:0805.2998 (2008)
[Jo] V.R. Jones, von Neumann algebras, notes from a course.
[Ni-Po-Sa] , R. Nicoara - S. Popa - R. Sasyk, on type II1 factors arising from 2-cocycles
of w-rigid groups, J. Funct. Anal. 242 (2007), no.1, 230-246.
[Oz] N. Ozawa, There is no separable II1 universal factor, Proc. Amer. Math. Soc. 132
(2) (2004), arXiv:math/0210411v2.
[Pe] V.
Pestov,
Hyperlinear
and
Sofic
Groups:
A
Brief
Guide,
arXiv:math/0804.3968v8(2008).
[Po] S. Popa, On a problem of R. V. Kadison on maximal abelian *-subalgebras in factors,
Invent. Math. 65 (1981/82), 269281.
[Ra] F. Radulescu, The von Neumann algebras of the non-residually finite Baumslag group
< a, bab3a−1 = b2 > embeds into Rω, arXiv:math/0004172v3 (2000).
V.
CAPRARO, UNIVERSIT `A DI ROMA TOR VERGATA e-mail:
[email protected]
L.
P AUNESCU,
UNIVERSIT `A
DI
ROMA
TOR
VERGATA
and
INSTITUTE of MATHEMATICS "S. Stoilow" of the ROMANIAN ACADEMY email:
[email protected]
10
|
1604.06290 | 4 | 1604 | 2017-06-22T18:34:19 | A look at the inner structure of the $2$-adic ring $C^*$-algebra and its automorphism groups | [
"math.OA",
"math.GR"
] | We undertake a systematic study of the so-called $2$-adic ring $C^*$-algebra $\mathcal{Q}_2$. This is the universal $C^*$-algebra generated by a unitary $U$ and an isometry $S_2$ such that $S_2U=U^2S_2$ and $S_2S_2^*+US_2S_2^*U^*=1$. Notably, it contains a copy of the Cuntz algebra $\mathcal{O}_2=C^*(S_1, S_2)$ through the injective homomorphism mapping $S_1$ to $US_2$. Among the main results, the relative commutant $C^*(S_2)'\cap \mathcal{Q}_2$ is shown to be trivial. This in turn leads to a rigidity property enjoyed by the inclusion $\mathcal{O}_2\subset\mathcal{Q}_2$, namely the endomorphisms of $\mathcal{Q}_2$ that restrict to the identity on $\mathcal{O}_2$ are actually the identity on the whole $\mathcal{Q}_2$. Moreover, there is no conditional expectation from $\mathcal{Q}_2$ onto $\mathcal{O}_2$. As for the inner structure of $\mathcal{Q}_2$, the diagonal subalgebra $\mathcal{D}_2$ and $C^*(U)$ are both proved to be maximal abelian in $\mathcal{Q}_2$. The maximality of the latter allows a thorough investigation of several classes of endomorphisms and automorphisms of $\mathcal{Q}_2$. In particular, the semigroup of the endomorphisms fixing $U$ turns out to be a maximal abelian subgroup of ${\rm Aut}(\mathcal{Q}_2)$ topologically isomorphic with $C(\mathbb{T},\mathbb{T})$. Finally, it is shown by an explicit construction that ${\rm Out}(\mathcal{Q}_2)$ is uncountable and non-abelian. | math.OA | math |
A look at the inner structure of the 2-adic ring C
∗
-algebra
and its automorphism groups
Valeriano Aiello†, Roberto Conti
♯
, Stefano Rossi
♮*
† Dipartimento di Matematica e Fisica
Universit`a Roma Tre
Largo S. Leonardo Murialdo 1, 00146 Roma, Italy.
♯
Dipartimento di Scienze di Base e Applicate per l'Ingegneria
Sapienza Universit`a di Roma
Via A. Scarpa 16, I-00161 Roma, Italy.
♮
Dipartimento di Matematica,
Universit`a di Roma Tor Vergata
Via della Ricerca Scientifica 1, I -- 00133 Roma, Italy.
Abstract
-algebra Q2. This is the
S2 and
2
We undertake a systematic study of the so-called 2-adic ring C ∗
2 U ∗
2 + U S2S∗
-algebra generated by a unitary U and an isometry S2 such that S2U = U
universal C ∗
S2S∗
through the injective homomorphism mapping S1 to U S2. Among the main results, the rela-
= 1. Notably, it contains a copy of the Cuntz algebra O2 = C ∗(S1, S2)
tive commutant C ∗(S2)′ ∩ Q2 is shown to be trivial. This in turn leads to a rigidity property
enjoyed by the inclusion O2 ⊂ Q2, namely the endomorphisms of Q2 that restrict to the
identity on O2 are actually the identity on the whole Q2. Moreover, there is no conditional
expectation from Q2 onto O2. As for the inner structure of Q2, the diagonal subalgebra D2
a thorough investigation of several classes of endomorphisms and automorphisms of Q2. In
particular, the semigroup of the endomorphisms fixing U turns out to be a maximal abelian
and C ∗(U) are both proved to be maximal abelian in Q2. The maximality of the latter allows
subgroup of Aut(Q2) topologically isomorphic with C(T, T). Finally, it is shown by an
explicit construction that Out(Q2) is uncountable and non-abelian.
Contents
1 Introduction
2 First results
2.1 The gauge-invariant subalgebra . . . . . .
. . . . . .
2.2 The canonical representation .
. . . . .
. . . . .
. . . . . .
. . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
3 Structure results
. . . . . .
3.1 Two maximal abelian subalgebras
3.2
. . . . . .
3.3 The relative commutant of the generating isometry . . . . . .
Irreducible subalgebras . . . .
. . . .
. . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
2
4
5
6
9
9
12
13
*E-mail: [email protected], [email protected], [email protected]
1
4 Extending endomorphisms of the Cuntz algebra
4.1 Uniqueness of the extensions .
. . . . . .
4.2 Extensible Bogoljubov automorphisms . .
. . . . .
. . . . .
. . . . . .
. . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
5 Outer automorphisms
5.1 Gauge automorphisms and the flip-flop . .
5.2 A general result
. . . . . .
. . . . .
. . .
. . . . .
. . . . .
. . . . . .
. . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
6 Notable endomorphisms and automorphism classes
6.1 Endomorphisms and automorphisms α such that α(S2) = S2 .
6.2 Automorphisms α such that α(U) = U .
6.3 Automorphisms α such that α(U) = zU .
. . . . . .
. . . . . .
. . . . .
. . . . .
f(z2) = f(z)2
A The functional equation f(z2) = f(z)2
f(z2) = f(z)2
on the torus
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
15
15
18
20
21
22
24
24
25
28
30
1 Introduction
∗
Ever since their formal debut in the most cited paper [17], the Cuntz algebras have received a great
deal of attention. The reasons are so many they resist any attempt to be only briefly accounted,
and this introduction will be no exception. Therefore, we cannot but draw a rather quick and
incomplete outline of the later developments until the present state of the art, if only to better
frame the scope of our work. For many authors who have focused their interest on more and more
general constructions inspired by the Cuntz algebras, there are as many authors who have devoted
themselves to as thorough as possible a study of the concrete Cuntz algebras. This study includes,
in particular, an in-depth investigation of endomorphisms and automorphisms. Cuntz is among
those who have undertaken both the tasks. As for the first, he and other authors have written a long
series of works where increasingly broad classes of C ∗
-algebras associated with algebraic objects
such as rings are contrived. In particular, in [18] he introduced a C ∗
-algebra QN associated with
the ax + b-semigroup over the natural numbers. A few years later, Larsen and Li [23] considered
its 2-adic version which, accordingly, they denoted by Q2. The main object of our interest in the
present paper, this novel C
-algebra is in fact naturally associated with the semidirect product
semigroup of the additive group Z acted upon by multiplication with non-negative powers of 2. It
did appear before elsewhere, cf. [23] and the references therein, but it is in the above-mentioned
work of Larsen and Li that it was studied systematically for the first time. After recalling that Q2
is a nuclear C ∗
-algebra, they prove, among other things, that Q2 is also a purely infinite simple
C ∗
-algebra. They give two proofs of this fact. Notably, one is a straightforward application of
Q2 being a Cuntz-Pimsner algebra, to which general results of Exel, an Huef and Raeburn [20]
apply. From our viewpoint, this lucky circumstance is well worth mentioning. Indeed, very little
is known about the general structure of endomorphisms or automorphisms for general Pimsner
algebras, cf. [28, 19, 7]. Therefore, as should follow from some of the main results announced in
the abstract, a good way to look at Q2 might be to regard it as a felicitous example of a Pimsner
algebra for which a far-reaching study is not that prohibitive. Far be it from us, however, to
allege we have done all that could be done. Rather, our hope is that further research may stem
from this work. That for a Pimsner algebra a thorough comprehension of the properties of all
its endomorphisms is a virtually impossible task should be no surprise. Indeed, already for the
Cuntz algebras On the problem, at least in its full generality, has turned out to be well beyond
the reach of current research. For instance, all endomorphisms of On are known to come from
unitary elements of On via a correspondence first pointed out by Takesaki, see e.g. [16], and yet
it is notoriously difficult to find non-tautological and effective characterizations of those unitaries
that yield automorphisms, although a number of intriguing if partial results about the structure of
Aut(On) have recently been achieved in [14, 11, 5, 12, 6, 8, 10], see also [9, 13] for an informative
2
account. Without further ado, we can now move on to the basic definitions needed throughout the
paper. While being a Pimsner algebra, the 2-adic ring C ∗
-algebra is perhaps best described as the
universal C ∗
-algebra Q2 generated by a unitary U and an isometry S2 such that
S2U = U
2
S2
and S2S
∗
2 + U S2S
∗
2 U
∗
= 1
1 + X2X ∗
The reader interested in its description in terms of Pimsner algebras is again referred to [23].
However, in this paper we will never need to resort to that picture, which is why we may as well
dispense with it. What we do need to observe is that Q2 contains a copy of the Cuntz algebra O2.
Indeed, the latter is by definition the universal C ∗
-algebra generated by two isometries X1 and X2
such that X1X ∗
2 = 1. Therefore, the map taking X1 to U S2 and X2 to S2 extends by
universality to a homomorphism from O2 to Q2, which is injective thanks to the simplicity of O2.
Accordingly, as of now it will be convenient to think of O2 as being a subalgebra of Q2. To us
the rather explicit description of the inclusion O2 ⊂ Q2 was in fact among the strongest motiva-
tions to carry out the present study of Q2, especially as far as the extension problem is concerned.
This asks whether an endomorphism of O2 extends to Q2. It turns out that this is not always the
case. For instance, as soon as Bogoljubov automorphisms are looked at, easy examples are found
of non-extensible automorphisms. More precisely, we find that the only extensible Bogoljubov
automorphisms are the flip-flop, the gauge automorphisms and their products. In addition, facing
the extension problem in general leads to an interesting rigidity property enjoyed by the inclusion
O2 ⊂ Q2, namely if an endomorphism of Λ of Q2 restricts to O2 trivially, then it is the identity
automorphism. To the best of our knowledge, the pair(O2, Q2) is the only known example of a
non-trivial inclusion of Pimsner algebras that fulfills the rigidity condition. In this respect, it is
also worth mentioning that there is no conditional expectation from the larger onto the smaller of
the two.
∗
Once these questions have been answered, it is natural to go on to study endomorphisms and
automorphisms of Q2 irrespective of whether they leave O2 globally invariant or not. Asking ques-
tions of this sort is of course motivated by the overwhelming literature written on similar issues for
the Cuntz algebras. However, this entails a preliminary study of the inner structure of Q2. In this
-algebra generated by U and the diagonal subalgebra D2 ⊂ O2
regard, we prove that both the C
are maximal abelian. It came as a surprise to us to learn that not as many results as one would
expect are known on maximal abelian subalgebras for general C ∗
-algebras. Apparently, that of
maximal abelian subalgebra is a notion far more relevant to von Neumann algebras. Yet Q2 seems
of selected classes of automorphisms, many of which are, incidentally, quasi-free in the sense of
Dykema-Shlyakhtenko and Zacharias, see [19, 28]. Notably, we show that the semigroup of the
to be one of the few exceptions, for its theory does benefit from C ∗(U) being such a subalgebra.
Indeed, we exploit the maximality of C ∗(U) to derive a number of results on the general form
endomorphisms of Q2 that fix U is in fact a maximal abelian subgroup of Aut(Q2) isomorphic
with C(T, T), the group of all continuous T-valued functions defined on the one-dimensional torus
in [16]. Moreover, they indicate that it is not unconceivable to regard C(T, T) ⊂ Aut(Q2) as a
T understood as the spectrum of U . These results are in fact in same spirit as those expounded
generalized maximal torus, although not connected, for which a kind of infinite-dimensional Weyl
theory might well worth attempting, as is done in [6, 8] for Cuntz algebras.
Since the group of equivalence classes of outer automorphisms of O2 is known to be so large
as to contain most locally compact groups, our investigation also addresses Out(Q2). We have
partial evidence to hold that Out(Q2) is not as large, not least because, as a drawback of the
embedding locally compact groups into Out(Q2) as we would do with Out(O2). At any rate, we
prove that Out(Q2) is still an uncountable non-abelian group. This is done in two steps. First,
aforementioned result on Bogoljubov automorphisms, we no longer have a general procedure for
3
we prove that both the flip-flop and the gauge automorphisms are mutually non-equivalent outer
automorphisms. Second, we provide a broad class of outer automorphisms that do not commute
in Out(Q2) with the flip-flop. Even so, the non-commutativity thus exhibited is admittedly of a
extent Out(Q2) is non-abelian.
rather mild form. We do believe that it is an interesting, albeit difficult, problem to say to what
A few words on the organization of the material are in order. The various results of the paper
are scattered throughout several sections, which more or less follow the order in which the topics
developed have been introduced above, as to allow the reader to find them more easily. For conve-
nience, here follows a description of the content of the several sections, to be also understood as
a short guide to the main results. Section 2 is preparatory in character, as it sets the stage for our
subsequent considerations. Indeed, all the needed definitions and basic properties are to be found
here. In Section 3.1 both C ∗(U) and D2 are shown to be maximal abelian subalgebras of Q2,
see Theorems 3.4 and 3.9, respectively. In Section 3.2 the extended canonical endomorphism is
proved to be a shift on Q2, Theorem 3.14. Moreover, there is no conditional expectation from Q2
onto O2, Theorem 3.16. The main result of Section 3.3 is Theorem 3.20, where the relative com-
mutant C ∗(S2)′ ∩ Q2 is shown to be trivial. Section 4 is focused on the uniqueness of extensions
of automorphisms from O2 to Q2, which is proved in Theorem 4.5, and the non-extendability
of general Bogoljubov automorphisms, which is proved in Theorem 4.14. Section 5 deals with
the outerness of the gauge automorphisms and the flip-flop, but it also includes a general result,
see Theorems 5.1, 5.8, 5.9, respectively. Finally, Section 6 provides a complete description of
AutC∗(U)(Q2), see Theorem 6.13. Moreover, this group is shown to be isomorphic with C(T, T)
and maximal abelian in Aut(Q2), see Theorems 6.14 and 6.16, respectively. The last two results
along with Theorem 6.9, which states that the outer automorphism group is non-abelian, should
be regarded as the main results of the present paper.
Throughout the paper, all endomorphisms are assumed to be unital and ∗-preserving. Finally,
for endomorphisms of the Cuntz algebra O2 we adopt the well-established notations to be found
in the wide literature of the field (see the beginning of Section 4 for a very short description of the
Cuntz-Takesaki correspondence). This is certainly the case for the symbols αz, ϕ, λf introduced
in full detail in the next section.
2 First results
As we observed in the introduction, the 2-adic ring C ∗
-algebra contains a copy of the Cuntz al-
gebra O2, as the C ∗
-subalgebra generated by S2 and S1 ≐ U S2. Since the theory of the latter
has been enriched by a deeper and deeper knowledge of distinguished classes of endomorphisms
as well as automorphisms, problems to do with their extensions to Q2 are undeniably among the
most natural things to initiate a study with. We will see in Section 4.1 that as soon as too much
generality is allowed, these problems begin to be intractable for all practical purposes. If one asks
a bit more specific questions, a great many partial results do start cropping out. At any rate, we
will also obtain a general result, namely that whenever extensions exist they are unique, which
is yet another way to state the rigidity property of the inclusion O2 ⊂ Q2 we explained in the
introduction. The precise statement of this fact, too, is contained in Section 4.1. In the present
section, we limit ourselves to three remarkable examples that are easily dealt with. The first is the
canonical shift. The second is the flip-flop. The third are the gauge automorphisms.
The canonical shift is explicitly defined on every x ∈ O2 as ϕ(x) = S1xS∗
if we set
1 +S2xS∗
2 , therefore
ϕ(x) = U S2xS
∗
2 U
∗ + S2xS
∗
2
for any x ∈ Q2
4
we still define an endomorphism of Q2, which restricts to O2 as the usual shift. The intertwin-
ing rules Six = ϕ(x)Si for any x ∈ Q2 with i = 1, 2 still hold true. Moreover, a straight-
forward computation shows that ϕ(U) = U 2
ϕ(f(U)) = f(U 2), which is true for any continuous function f . It goes without saying that the
. Since the continuous functional calculus of a
normal operator commutes with any endomorphism, the above equality can also be rewritten as
same equality retains its validity with any Borel function whenever Q2 is represented on some
Hilbert space. We shall avail ourselves of this useful fact later on.
As is well known, the flip-flop is the involutive automorphism λf ∈ Aut(Q2) that switches
S1 and S2 with each other. The flip-flop extends as well, although the proof is less obvious and
needs an argument. This is done here below.
Proposition 2.1. ([15]) The flip-flop automorphism of O2 extends to an automorphism of Q2.
≐ U ∗
and S′
Proof. If we set U ′
2 ≐ U S2, then the identity S′
2S′∗
2 +U ′S′
2S′∗
2 U ∗
checked. By universality of Q2, there exists a unique endomorphism λf ∈ End(Q2) such that
λf(U) = U ′
and λf(S2) = S′
∗) = U and λf(U S2) = S2, the image of α must be the whole Q2, that is
is simple. Since λf(U
to say λf is an automorphism. Finally, it is obviously an extension of the flip-flop.
2 = U S2. This endomorphism is necessarily injective as Q2
= U ∗
□
= 1 is immediately
As of now, the above extension will be referred to simply as the flip-flop of Q2 and will be
denoted by λf .
automorphisms αz given by αz(Si) = zSi, with z ∈ T. We can also prove the following extension
It is also well known that the Cuntz algebra O2 is acted upon by T through the so-called gauge
result concerning gauge automorphisms.
Proposition 2.2. The gauge automorphisms of O2 can all be extended to automorphisms of Q2.
′
′
∗
S
′
2S
′∗
2 U
≐ U and S
′∗
2 + U
′
2S
Proof. Now we set U
to 1 . As we still have S
′
2 ≐ zS2, where z is any complex number of absolute value equal
= 1, there exists an automorphism αz ∈ Aut(Q2)
such that αz(U) = U and αz(S2) = zS2. To conclude, all that we are left to do is note that
αz(S1) = αz(U S2) = αz(U)αz(S2) = U zS2 = zS1.
With a slight abuse of terminology, the automorphisms αz obtained above will be referred
to as the gauge automorphisms. To conclude, it is worth noting that the flip-flop and the gauge
automorphisms commute.
□
2.1 The gauge-invariant subalgebra
The gauge-invariant subalgebra of O2, usually denoted by F2, is known to be isomorphic with
the CAR algebra. The corresponding gauge-invariant subalgebra of Q2, which throughout this
paper will be denoted by QT
-algebra.
However, it can be described far more conveniently as the closure of a suitable linear span. To do
so, we need to point out the following simple but useful result. In order to state it as clearly as
possible, let us first set some notation. As in [17], we denote by W2 the set of all multi-indices
2 , can no longer be identified with such a remarkable C ∗
µ = (µ1, µ2, . . . , µn) with µi ∈ {1, 2} and n ∈ N; the integer n is commonly referred to as the
length of the multi-index µ and is denoted by∣µ∣. For any such multi-index µ =(µ1, µ2, . . . , µn),
Proposition 2.3. Q2 = span{SµS∗
ν U k ∶ µ, ν ∈ W2, k ∈ Z}.
we denote by Sµ the monomial Sµ1 Sµ2 . . . Sµn .
5
Proof. In order to prove the equality above all we have to do is observe that the following relations
from the left to the right side of any monomial of the form SµS ∗
allow us to take both U and U ∗
ν .
• U S1 = S2U
2 U
2 U 2
• U S2 = S1
1 = S∗
• U S∗
2 = S∗
• U S∗
• U ∗S1 = S2
• U ∗S2 = S1U ∗
• U
∗
S
• U
∗
S
∗
1 = S
∗
2 = S
∗)2
∗
1(U
∗
1 U
∗
The relations themselves are immediately verified by direct computation instead.
□
The gauge automorphisms yield a conditional expectation įE from Q2 onto QT
the action itself on T, that is for any x ∈ Q2 we have įE(x) = ∫T αz(x)dz, with dz being the
normalized Haar measure of T. Now since įE(SµS∗
ν U k) = 0 if and only if ∣µ∣ ≠ ∣ν∣. This helps to prove the description alluded to
have įE(SµS∗
ν U k) = SµS∗
0 ei(∣µ∣−∣ν∣)θ dθ
ν U k ∫ 2π
2 by averaging
, we also
2π
above.
Proposition 2.4. The equalities below hold:
Q
T
2 = spanSµS
∗
ν U
k ∶ µ, ν ∈ W2,∣µ∣ =∣ν∣, k ∈ Z = C
∗(U, F2) ⊂ Q2
2 ⊂ span{SµS∗
Proof. The second equality is obvious. We focus then on the first, for which we only have to worry
about the inclusion QT
x ∈ QT
ν U k ∶ ∣µ∣ = ∣ν∣}, the other being immediately checked. If
2 , then x = įE(x). Now pick a sequence {xn} in the algebraic linear span of the set
ν U k ∶ µ, ν ∈ W2, k ∈ Z} such that ∥xn − x∥ tends to zero. As E is a bounded map,
{SµS∗
∥įE(xn) − įE(x)∥ = ∥įE(xn) − x∥ tends to zero as well. The conclusion follows easily now
because įE(xn) ∈ span{SµS∗
∗(D2, U) is the Bunce-Deddens algebra of type
ν U k ∶∣µ∣ =∣ν∣} by the remark we made above.
We should also mention that C
, see [3, Remark 2.8].
∗(F2, U) = C
2∞
□
2.2 The canonical representation
In this section we gather as much information as we need about a distinguished representation of
Q2, which will actually play a major role in most of what follows here and in the next sections.
As far as we know, it was first exhibited in [23], where it is called the canonical representation.
Therefore, from now on it will always be referred to as the canonical representation. After a brief
review of its main properties, we discuss a number of results where the canonical representation
proves to be rather useful.
The canonical representation acts on ℓ2(Z) through the operators S2, U ∈ B(ℓ2(Z)) given by
S2ek ≐ e2k and U ek ≐ ek+1, where{ek ∶ k ∈ Z} is the canonical orthonormal basis of ℓ2(Z), i.e.
ek(m) = δk,m. The very first thing to note is that 1 is the only eigenvalue of S2, corresponding
to the one-dimensional eigenspace generated by e0. This simple observation enables us to give a
short proof that the canonical representation is irreducible. Since we do not know of any reference
where this possibly known fact is explicitly pointed out, we do include an independent proof for
the reader's convenience.
6
Proposition 2.5. The canonical representation of Q2 is irreducible.
If P is the associated orthogonal
projection, then P ∈ Q′
2. In particular, S2P = P S2, and so S2P e0 = P e0. As the eigenspace of
S2 corresponding to the eigenvalue 1 is spanned by e0, we must have either P e0 = e0 or P e0 = 0.
Proof. Let M ⊂ ℓ2(Z) be a Q2-invariant closed subspace.
In the first case, e0 ∈ M , and therefore C ∗(U)e0 ⊂ M , which says that M = ℓ2(Z) because e0
is a cyclic vector for C ∗(U). In the second, e0 ∈ M
= ℓ2(Z), i.e. M = 0. Note, however, that O2 does not act irreducibly on ℓ2(Z),
for the closed span of the set{ek ∶ k = 0, 1, . . . ,} is obviously a proper O2-invariant subspace. □
instead. As above, M
being Q2-invariant
too, we have M
⊥
⊥
⊥
However, the canonical representation restricts to O2 as a reducible representation, which we
denote by π. More precisely, it is a direct sum of two inequivalent irreducible representations of
O2. To see this, let us define H+, H− ⊂ ℓ2(Z) as the closed subspaces given by
H+ ≐ span{ek ∶ k ≥ 0}
H− ≐ span{ek ∶ k < 0}
and
The Hilbert space ℓ2(Z) is immediately seen to decompose into the direct sum of these subspaces,
i.e. ℓ2(Z) = H+ ⊕ H−. Furthermore, both H+ and H− are O2-invariant, and finally they may be
checked to be O2-irreducible too. This last statement should be a well-known fact. Even so, we
give the proof for the sake of self-containedness.
Proposition 2.6. The subspaces H± are both O2-irreducible.
⊥
Proof. We only need to worry about H+, for H− is dealt with in much the same way. Exactly
as above, if M ⊂ H+ is an O2-invariant subspace, then either M or its orthogonal complement
must contain e0. The proof is thus complete if we can show that an O2-invariant subspace
M
containing e0, say N , is the whole H+, and this is proved once we show ek ∈ N for every k ≥ 0.
This is in turn easily achieved by induction on k. Suppose we have proved{el ∶ l = 0, 1, . . . , k} ⊂
N . For the inductive step we have two cases, according as k + 1 is even or odd. If it is even, then
. In either cases we see that ek+1 is in N , as wished.
ek+1 = S2e k+1
□
; if it is odd, then ek+1 = S1e k
2
2
Denoting by π± the restriction of π to H± respectively, the decomposition into irreducible
representations π = π+ ⊕ π− has just been proved to hold. Now, as what we are really interested
, we also need to observe that π+ and π− are disjoint. This is done
in is the commutant π(O2)′
here below.
Lemma 2.7. If π+ and π− are the irreducible representations defined above, then π+ ⫰ π−.
Proof. It is enough to note that π+(S2) has 1 in its point spectrum, whereas π−(S2) does not. □
To state the next result as clearly as possible some notation is needed, so let us denote by E±
the orthogonal projections onto H± respectively.
Proposition 2.8. The commutant of π(O2) is given by π(O2)′
Proof. According to the decomposition described above we have π(O2)′
But because π+ and π− are disjoint, we can go a bit further and write
= CE+ + CE−.
=(π+(O2)⊕π−(O2))′
.
(π+(O2) ⊕ π−(O2))′
=(π+(O2))′ ⊕(π−(O2))′
= CE+ ⊕ CE−
where the last equality is due to the irreducibility of π±.
□
7
This immediately leads to the following corollary, which needs no proof.
Corollary 2.9. In the canonical representation of Q2 the bicommutant of O2 is given by
π(O2)′′
={T ∈ B(ℓ2(Z)) ∶ T = T+ ⊕ T− ∶ T± ∈ B(H±)}
′′
D
The information we have gathered about π will actually turn out to be vital in tackling the
In passing, we also take the opportunity to exploit the canonical representation to show that
expect. However, the relative proof is not as obvious as the statement. In fact, we still lack some
basic ingredients. In particular, we need to observe that the basis vectors ek are all cyclic and
separating for U . Therefore, the W ∗
problem as to whether C ∗(U)′ ∩ O2 ⊂ Q2 is trivial. This is the case indeed, as anyone would
-algebra W ∗(U) generated by U is a maximal abelian von
Neumann algebra of B(ℓ2(Z)). Furthermore, it is common knowledge that W ∗(U) is isomorphic
with L∞(T, µ), where µ is the Haar measure of T.
2 ⊂ B(ℓ2(Z)) is a maximal abelian subalgebra as well. This will in turn be vital to conclude that
D2 is a maximal abelian subalgebra of Q2. Henceforward we shall denote by ℓ∞(Z) the atomic
MASA of B(ℓ2(Z)) acting through diagonal operators with respect to the canonical basis.
2 = ℓ∞(Z).
Proof. Since ℓ∞(Z) is a MASA, it is enough to prove that D′′
2 = ℓ∞(Z), which will be imme-
2)n} ⊂ D2 strongly converges to
2(S
closure of D2. To begin with, we note that the sequence {Sn
2)nU −k ∶ n ∈ N} strongly converges to Ek. The conclusion
E0. But then the sequence{U kSn
now follows from the fact that D2 is globally invariant under ad(U).
∗(U)′ ∩ O2 is trivial. This will in
diately checked once we have proved that the projections Ek onto Cek all belong to the strong
Proposition 2.10. In the canonical representation we have D′
turn result from a straightforward application of the next proposition, where much more is proved.
We now have all the necessary tools to get to prove that C
2(S∗
□
∗
Proposition 2.11. We have W ∗(U) ∨ π(O2)′
= B(ℓ2(Z)).
∗(U) ∨ π(O2)′
. From P E+ =
Proof. Let P be an orthogonal projection in the commutant of W
E+P and P E− = E−P , H± are straightforwardly seen to be both P -invariant. In particular, P e0
e0 must be an eigenvector of P . As such, we have either P e0 = 0 or P e0 = e0. In the first case
□
must take the form P e0 = ∑k≥0 akek. For the same reason, P e−1 is in H−, but it is also given by
P e−1 = P U ∗e0 = U ∗P eo = ∑k≥0 akek−1, which means ak = 0 for every k > 0. In other words,
P = 0, whilst in the second P = 1, because P f(U)e0 = f(U)P e0 for every f ∈ L∞(T).
Corollary 2.12. In the canonical representation W ∗(U) ∩ π(O2)′′
C ∗(U)′ ∩ O2 = C1.
ever, note that the rank-one orthogonal projections onto Cen, with n ∈ Z, all belong to W ∗(U) ∨
π(O2)′
= Espan{ek∶ k≥n} for any n ∈ Z. In light of this, the above
group Γ the von Neumann algebra on ℓ2(Γ) generated by λ(Γ) and the multiplication operators
Mf , with f ∈ c0(Γ), is the whole B(ℓ2(Γ)). Obviously, our case corresponds to Γ = Z.
: indeed, we have U nE+(U ∗)n
Remark 2.13. We have included an elementary proof of 2.11 for the reader's convenience. How-
proposition is well worth comparing with the more far-reaching classical result that for any discrete
= C1. As a consequence,
At this stage, there is another step to take to improve our knowledge of the C
ated by U . Indeed, we are yet to prove that C
Since this task requires some technical work, we postpone the proof to the next section.
∗(U) is a maximal abelian subalgebra of Q2 as well.
-algebra gener-
∗
8
3 Structure results
3.1 Two maximal abelian subalgebras
The goal of the present section is to tackle two structure problems for Q2, namely that both C ∗(U)
and D2 are maximal abelian subalgebras. We start with C ∗(U). The relative result is easily
more refined tools such as conditional expectations from B(H) onto a maximal subalgebra. As
guessed, and yet its proof is unfortunately far from being straightforward, in that it needs some
is known, the proof of the existence of such conditional expectations can be traced back to the
classic work of Kadison and Singer [21], where the authors first described a general procedure to
obtain them. Nowadays, the existence of conditional expectations of this sort is preferably seen
as an immediate consequence of the injectivity of abelian von Neumann algebras. Even so, we do
sketch the original procedure, not least because we make use of it in the proof of Theorem 3.6.
This runs as follows. Given T ∈ B(H), set T∣P
≐ P T P +(I − P)T(I − P) for any projection P
in W ∗(U). If{Pi} is a generating sequence of projections of W ∗(U), then every cluster point of
the sequence{T∣P1∣P2∣...∣Pn} lies in W ∗(U)′
= W ∗(U), as explained in [21]. This enables us to
define a conditional expectation from B(H) onto W ∗(U) associated with each ultrafilter p ∈ βN
simply by taking the strong limit of the subnet corresponding to that ultrafilter.
k
∗
Sk
2 U = U 2
2] = E[(S
clarity, our proof is in turn divided into a series of preliminary lemmata.
We next show as a key lemma to achieve our result that, for any conditional expectation E
Proof. First note that the second equality is a straightforward consequence of the first thanks
β] is at worst a monomial in U . For the sake of
2)k] = 0 hold for k ∈ N.
, which holds for every T ∈ B(H). The commutation rules
2], we see
E[Sk
2] = 0 and we are done.
from B(H) onto W ∗(U), we have that E[SαS∗
Lemma 3.1. With the notations set above, the equalities E[Sk
to the fact that E[T
that f(z)(z − z2k) = 0, hence f(z) = 0 for every z ∈ T. This says E[Sk
2(S
Lemma 3.2. We have that E[Sk
α] = U
. If we now set f(U) ≐ E[Sk
2)k] = 2
E[S∣α∣2 (S
∗] = E[T]∗
2]U = U 2
= E[S∣α∣2 (S
2 give E[Sk
Proof. To begin with, we observe that
2] = E[Sk
for every non-negative integer k.
2)∣α∣]
2]U 2
Sk
−h
□
−k
∗
∗
k
k
h
∗
E[SαS
2)∣α∣]U
for some positive integer h. Since 1 = ∑∣α∣=k SαS∗
α] = 2
E[SαS
∗
∗
k
α we have that
E[S∣α∣2 (S
∗
2)∣α∣].
.
□
1 = =
∣α∣=k
2)k] = 2−k
2(S∗
2(S∗
This implies that E[Sk
Lemma 3.3. We have that E[Sk
2)m] = 0 for k, m ≠ 0, k ≠ m.
2)m]U
2)m], we obtain the functional equation(z2
now set f(U) ≐ E[Sk
clearly implies f(z) = 0 for every z ∈ T. This shows that E[Sk
2(S∗
2(S
E[Sk
Sk
2 we get U
2(S∗
2 U = U
2k
We now have all the necessary information to carry out our proof of C
Proof. Thanks to the last two lemmas, it suffices to show the statement for k > m > 0. By using
the commutation rules Sk
2k
∗
m
∗
k−2
−2m
2)m]. If we
2(S
= E[Sk
− 1)f(z) = 0, which
2)m] = 0 and we are done. □
∗(U) being a maximal
abelian algebra.
9
-subalgebra of Q2.
Theorem 3.4. C ∗(U) is a maximal abelian C ∗
Proof. As C ∗(U)′ ∩ Q2 = W ∗(U) ∩ Q2, it is enough to prove that given f ∈ L∞(T) with
f(U) in Q2, then f is in fact a continuous function. Now if f(U) belongs to Q2, it is also the
norm limit of a sequence {xk} ⊂ Q2 with each xk taking the form ∑α,β,h cα,β,hSαS∗
E ∶ B(H) → W ∗(U) is any of the conditional expectations considered above, we have f(U) =
E(f(U)) = limk E(xk). But then each E(xk) is of the form ∑α,β,h cα,β,hU h+kα,β , where U kα,β
is nothing but E(SαS∗
β). Consequently, there exists a sequence of Laurent polynomials pk such
that∥f(U) − pk(U)∥ tends to zero, that is f(U) ∈ C ∗(U), as maintained.
β U h
. If
□
Among other things, it is interesting to note that the former proof yields a distinguished condi-
tional expectation from Q2 onto C ∗(U), which is simply obtained by restricting any of the afore-
∗(U)
said conditional expectations to Q2. Although there are conditional expectations onto W
aplenty, as proved in [21], the above computations also show that E is in fact unique, a fact worth
a statement of its own.
Theorem 3.5. The conditional expectation E ∶ Q2 → C ∗(U) is unique.
To complete the picture, we next show that E is faithful. This is actually a straightforward
consequence of a general well-known result due to Tomiyama [27], whose proof in our setting
is nevertheless included for the sake of completeness, being utterly independent of Tomiyama's
work to boot.
Proposition 3.6. The unique conditional expectation E above is faithful.
is α-coercive as well regardless of the projection P . In particular, if T ∈
Proof. By uniqueness it is enough to make sure that the conditional expectation yielded by the
Kadison-Singer procedure is faithful. Now if T is an α-coercive operator, i.e. (T x, x) ≥ α∥x∥2
with α > 0, then T∣P
B(H) is a coercive operator, then E[T] cannot zero, being by definition a weak limit of coercive
and ε > 0 is any real number with ε < ∥T∥, the spectral theorem provides us with an orthogonal
T ↾Nε is ε-coercive. The remark we started our proof with allows to conclude that E[T] is not
decomposition H = Mε ⊕ Nε with Mε and Nǫ both T -invariant and such that the restriction
positive operators all with the same constant as T . If now T is any non-zero positive operator
zero either.
□
We can now move on to D2. Again, the techniques we employ make a rather intensive use of
conditional expectations. Before we start, it is worth mentioning that this result can be understood
as a generalization of the well-known property of D2 being maximal in O2. We start attacking the
problem with the following couple of lemmas, for which we first need to set some notation. We
still denote by E the unique faithful conditional expectation from B(H) onto ℓ∞(Z). As known,
this is simply given by(E[T]ei, ej) =(T ei, ej)δi,j.
Lemma 3.7. The following relations hold:
• E[U k] = δk,0I,
α] = SαS∗
• E[SαS∗
• if∣α∣ ≠∣β∣, E[SαS∗
• if∣α∣ =∣β∣, E[SαS∗
α,
β U h] is either 0 or Ei,
β U h] is either 0 or SαS∗
β U h
.
10
Proof. The first two equalities need no proof. For the third, without harming generality, we may
suppose that∣α∣ <∣β∣
E[SαS
∗
β U
h] = E[U
= E[U
∗
h(α)(S2)∣α∣(S
h(α)(S2)∣α∣(S
h(α)(S2)∣α∣(S
2)∣β∣U
2)∣α∣U
2)∣α∣U
h−h(β)]
−h(α)U
−h(α)E[U
∗
∗
= U
h(α)(S
h(α)(S
∗
2)∣β∣−∣α∣U
2)∣β∣−∣α∣U
∗
h−h(β)]
h−h(β)].
where h(α) and h(β) are positive integers. Accordingly, we are led to compute E[U h(S2)kU l],
where h ∈ Z. Now the condition U h(S2)kU lei = ei implies that i = 2k(i + l) + h, and because
the former equation has a unique solution, we get the thesis. Finally, for the fourth we have that
E[SαS
∗
β U
∗
∗
= U
h(α)(S2)∣α∣(S
h] = E[U
h(α)(S2)∣α∣(S
= E[U
h(α)(S2)∣α∣(S
= δh−h(β)+h(α),0U
= δh−h(β)+h(α),0U
= δh−h(β)+h(α),0SαS
h−h(β)]
2)∣β∣U
−h(α)U
2)∣α∣U
2)∣α∣U
−h(α)E[U
h(α)(S2)∣α∣(S
h(α)(S2)∣α∣(S
∗
β U
∗
∗
h
.
h−h(β)+h(α)]
h−h(β)+h(α)]
−h(α)
2)∣α∣U
2)∣α∣U
h−h(β)
∗
□
In order to make our proof work, we also need to take into account the conditional expec-
tation Θ from Q2 onto D2 described in [23]. We recall that this is uniquely determined by
2, where f is in F2. Moreover, it is there shown to
be faithful too.
2)iU −lf U l′
2) ≐ δi,i′δl,l′(S∗
Θ((S∗
Si′
Lemma 3.8. If∣α∣ = ∣β∣, Θ[SαS∗
2)iU −lf U lSi
β U h] is either 0 or SαS∗
on monomials SαS∗
β U h
with∣α∣ =∣β∣.
β U h
. In particular, Θ and E coincide
Proof. By direct computation. Indeed, we have
Θ[SαS
∗
β U
h(α)(S2)∣α∣(S
h] = Θ(U
= δh(α),−h+h(β)U
2)∣α∣U
h(α)(S2)∣α∣(S
h−h(β))
2)∣α∣U
∗
∗
h−h(β)
□
Theorem 3.9. The diagonal subalgebra D2 ⊂ Q2 is a maximal abelian subalgebra.
2∩Q2 = ℓ∞(Z)∩Q2
Proof. As usual, all we have to do is make sure that the relative commutant D′
reduces to D2. Let x ∈ ℓ∞(Z) ∩ Q2, then there exists a sequence {xk} converging normwise to
. As above, x = E(x) = limk E(xk). Thanks
x with each of the xk of the form ∑ cα,β,hSαS∗
to the former lemmata, we can rewrite E(xk) as dk + fk, where dk ∈ D2 and fk are all diagonal
finite-rank operators. Now, being dk = Θ(xk), we see that dk must converge to some d ∈ D2. But
Q2, hence k = 0, being K(H) ∩ Q2 ={0}, and x = d ∈ D2.
then fk converge normwise to a diagonal compact operator, say k, which means k = x − d is in
□
β U h
The former result readily leads to D2 being a Cartan subalgebra of Q2. An anonymous referee
pointed out an alternative approach to show this by using the works of N. Larsen and X. Li, [23],
and J. Renault, [26]. We believe this approach is very elegant and we give a brief account of it.
In the first place, one should see the 2-adic ring C ∗
-algebra
-algebra Q2 as a reduced groupoid C ∗
11
Z
(see [23, Section 5]). Indeed, if G =(Z[ 1
2]⋊2Z)⋉Q2 is the transformation groupoid, where Z[ 1
2]
r(F) ≃ Q2.
act on Q2 by addition and multiplication, respectively, and F ∶= G∣Z2 , then C
and 2
Now, the claim follows at once by using [26, Proposition 3.1] after showing that the set of isotropy
points of Z2 is countable, and thus that the set of points with trivial isotropy is dense (that is, the
corresponding action is topologically free [26, Sec. 6.1]). The canonical abelian subalgebra of
a crossed product with respect to a group is considered as the most typical example of MASA,
provided that the action is topologically free. In the present case, although Q2 is only a crossed
product by a semigroup, the situation is almost the same because it is just a corner of a crossed
product by a group. One advantage of this description of Q2 as a crossed product makes more
natural to recognize D2 as the canonical Cartan subalgebra (and the canonical expectation).
∗
As for O2, later on we will prove that there is no conditional expectation from Q2 onto O2.
3.2 Irreducible subalgebras
In order to take a step further towards the study of Q2, especially as far as the properties of the
inclusion O2 ⊂ Q2 are concerned, it is worthwhile to recall a useful result proved by Larsen
It says that a representation ρ of O2 extends to a
and Li in their aforementioned paper [23].
representation of Q2 if and only if the unitary part of the Wold decomposition of ρ(S1) and ρ(S2)
are unitarily equivalent. Accordingly, once the unitary parts of the Wold decompositions have
been proved to be unitarily equivalent, the isometries are unitarily equivalent too because of the
relation U S1 = S2U . As remarked by the authors themselves, this allows us to think of Q2 as a
sort of symmetrized version of O2. Notably, the result applies to those representations π of O2 in
which both π(S1) and π(S2) are pure. Moreover, in such cases the sought extension is unique, as
pointed out in [23, Remark 4.2]. For the reader's convenience, though, we do single out this as a
separate statement.
Proposition 3.10. Let π be a representation of O2 on the Hilbert space Hπ such that π(S1)
and π(S2) are both pure. Then there exists a unique unitary įU ∈ B(Hπ) such that π(S2)įU =
įU 2π(S2) and π(S1) = U π(S2).
The recalled theorem by Larsen and Li would actually be enough to prove that the Cuntz
algebra O2 is irreducible in Q2, as might be expected. Even so, this can also be derived from the
much stronger result that even the UHF subalgebra F2 has trivial relative commutant, which is
shown below.
Theorem 3.11. The UHF subalgebra F2 ⊂ Q2 is irreducible, i.e. F ′
Proof. As D2 is a subalgebra of F2, we have F ′
depends upon D2 being a maximal abelian subalgebra of Q2. Therefore, we find F ′
F
2 ∩ Q2 = D2, where the last equality
2 ∩ Q2 =
□
2 ∩ Q2 ⊂ D′
2 ∩ Q2 = C1.
′
2 ∩ D2 ⊂ F
′
2 ∩ F2 = Z(F2) = C1.
While immediately derived from the above theorem, the following couple of corollaries do
deserve to be highlighted.
Corollary 3.12. The relative commutant of O2 in Q2 is trivial, i.e. O′
2 ∩ Q2 = C1.
Corollary 3.13. The relative commutant of the gauge-invariant subalgebra QT
2)′ ∩ Q2 = C1.
(QT
enjoy the so-called shift property, i.e. ⋂k ϕk(Q2) = C1, whence its name. This important prop-
Interestingly, the irreducibility of F2 also applies to the canonical shift, which turns out to
erty should have first been singled out by R. T. Powers, who called it strong ergodicity, but we do
not have a precise reference for the reader. The canonical shifts of On are of course known to be
strongly ergodic, see [22] for a full coverage of the topic.
2 is trivial, i.e.
12
Theorem 3.14. The canonical endomorphism ϕ of Q2 is a shift, i.e. ⋂k ϕk(Q2) = C1.
β ϕk(x), x ∈ Q2, is
γ , the equality ϕk(x)SαS∗
Proof. Since ϕk(x) = ∑γ∶∣γ∣=k Sγ xS∗
straightforwardly checked to hold true for every pair of multi-indices α and β with∣α∣ =∣β∣ = k.
ν ∶ ∣µ∣ =
2 ≐ span{SµS∗
2)′ ∩ Q2 for every k, where F k
This says that ϕk(Q2) is contained in(F k
∣ν∣}. But then we have the chain of inclusions ⋂k ϕk(Q2) ⊂ ⋂k(F k
2 ∩ Q2 = C1. □
Remark 3.15. The former theorem says in particular that ϕ is not surjective in a rather strong
sense. We can be more precise by observing that U is not in ϕ(Q2). Indeed, by maximality of
∗(U), any inverse image of U should lie in C
∗(U) does not
∗(U), but the restriction of ϕ to C
2)′ ∩ Q2 ⊂ F ′
C
yield a homeomorphism of T.
β = SαS∗
We can now go back to the announced result that conditional expectations onto the Cuntz
algebra O2 do not exist.
Theorem 3.16. There is no unital conditional expectation from Q2 onto O2.
Proof. Suppose that such a conditional expectation does exist. We want to show that this leads
to E(U) being U , which is obviously absurd. We shall work in any representation in which S1
and S2 are both pure, for instance the one described in [4]. If we compute E on the operator
U Sn
by using the commutation rule Sn
1 we easily get to the equality
2 U = U Sn
1 S2S∗
2(S∗
1)n
But on the other hand we also have that U Sn
and U must coincide on the direct sum of the subspaces Mn ≐ Sn
easily seen to be the whole Hπ.
1 S2S∗
= Sn
2 S1S∗
∗
2(S
n
2 S1S
∗
= S
1)n
2(S∗
1)n
∗
1)n
1)n
2(S∗
. Accordingly, E(U)
1)nHπ, which can be
2(S∗
1 S2S∗
□
E[U]S
n
1 S2S
∗
2(S
3.3 The relative commutant of the generating isometry
This section is entirely devoted to proving that C
is the same as proving that C
For this, we still need some preliminary definitions and results.
∗(S1)′ ∩ Q2 is trivial, merely because ad(U
2 ≐ span{SαS∗
∗(S2)′ ∩ Q2 is trivial. We first observe that this
∗(S1).
β U h ∣∣α∣ =∣β∣ = k, h ∈ Z}. For every k we have
∗(S2)) = C
∗)(C
, which readily follows from the Cuntz relation 1 = S1S∗
1 + S2S∗
2 .
Given any k ∈ N, we set Bk
the inclusion Bk
2 ⊂ Bk+1
2
Lemma 3.17. Let x ∈ Bk
2 , then
1 ∈ C ∗(U);
1)kxSk
1. (S∗
2. the sequence{(S
∗
1)mxSm
1 } stabilizes to a scalar cx ∈ C.
Proof. Without loss of generality, suppose that x = SαS
. We have that
∗
β U h
(S
∗
1)k
xS
k
1 = δα,1S
∗
β U
h
k
1 .
S
If h > 0, then by using the relation S2U = U S1, we see that the r.h.s. of the last expression is
given by
δα,1S
∗
β SγU
l(h) = δα,1δβ,γ U
l(h)
where γ is a multi-index of length k. If h = 0, we obtain
δα,1S
∗
β S
k
1 = δα,1δβ,1.
13
When h < 0, by using the relations U ∗S1 = S2 and U ∗S2 = S1U ∗
we obtain
δα,1S
∗
β U
h
k
1 = δα,1U
S
l(h)S
∗
γ S1 = δα,1δγ,1U
l(h)
and this fact will be important in the sequel. For the second part of the thesis, we may suppose
where γ is a multi-index of length k. We observe that in these cases we always have∣l(h)∣ ≤∣h∣,
that m > k +∣h∣ + 1. The needed computations can be made faster in the canonical representation
(for brevity we write l instead of l(h)):
1)m
1 ej =(S
e2mj+2m−1 =(S
e2mj+2m−1+l.
1)m
1)m
(S
U
U
S
m
∗
∗
∗
l
l
The expression above is non-zero if and only if 2mj + 2m − 1 + l = 2mi + 2m − 1 for some i, that
is to say l = 2m(i − j). But m > k + h + 1 and∣l∣ ≤∣h∣, therefore we finally get i = j and l = 0.
□
Proposition 3.18. Let x ∈ QT
∗
2 = span{SαS
h (S
β U h∣∣α∣ =∣β∣, h ∈ Z}. Then
1)h
h
1 ∈ C
lim
xS
∗
f(k)
Proof. By hypothesis there exists a sequence xk ∈ B
of natural numbers i and j. For any k ∈ N sufficiently larger than f(i) and f(j), by the former
lemma we have that (S∗
1 =∶ cj ∈ C. The sequence ci is convergent
1 =∶ ci,(S∗
that tends normwise to x. Choose a pair
2
since
1)kxiSk
∣ci − cj∣ =∥(S
∗
1)k
xiS
k
1)kxj Sk
1 −(S
1)k
1)hxSh
∗
∗
We denote by c the limit. Now the sequence(S
xj S
k
1∥ ≤∥xi − xj∥.
1 tends to c as well.
□
(1)
(2)
For any non-negative integer i we now define the linear maps Fi ∶ Q2 → QT
2 given by
∗
Fi(x) ≐ ET αz[x(S
F−i(x) ≐ ET αz[S
1)i]dz
1x]dz.
i
We observe that
i
1(S
Fi(x) = Fi(x)S
1)i
F−i(x).
1)i
F−i(x) = S
1(S
∗
∗
i
Before proving the main result of the section, we also need to recall the following lemma, whose
proof can be adapted verbatim from the original [17, Proposition 1.10], where it is proved for the
Cuntz algebras instead.
Proposition 3.19. Let x ∈ Q2 be such that Fi(x) = 0 for all i ∈ Z. Then x = 0.
Theorem 3.20. Let w ∈ U(Q2) such that wS1w∗
Now we have all the tools for completing our proof.
= S1. Then w ∈ T1.
14
Proof. First of all we observe that we also have wS∗
1 w∗
Fi(w). Indeed, for i ≥ 0,
= S∗
1 . We have that S∗
1 Fi(w)S1 =
S
∗
1 Fi(w)S1 = S
S
∗
1 F−i(w)S1 = S
= ET αz[S
∗
∗
∗
∗
∗
∗
S
1 ET αz[w(S
1)i]dz S1 = ET
1 αz[w(S
1)i]S1dz
1)i]dz = Fi(w)
S1]dz = ET αz[w(S
1)i
1 w(S
1w]dz S1 = ET
1 ET αz[S
1 αz[S
1w]S1dz
1w]dz = F−i(w)
1wS1]dz = ET αz[S
∗
1 S
S
∗
∗
∗
i
i
i
i
= ET αz[S
By Proposition 3.18 we obtain that for each i ∈ Z one has
lim(S
∗
1)k
Fi(w)S
k
1 = Fi(w) ∈ C.
Equation (1)-(2) together imply that for i ≠ 0 we have Fi(w) = 0. Now Proposition 3.19 applied
to w − F0(w) gives the claim.
□
Exactly as for the Cuntz algebras, we can also state a slight generalization of the former result,
which says an inner automorphism of Q2 cannot send Si to a scalar multiple of it.
Proposition 3.21. Let φ ∈ Aut(Q2) be such that φ(Si) = zSi for some z ∈ T\{1} and i = 1 or
i = 2. Then φ is an outer automorphism.
The proof is straightforwardly obtained by recasting Theorem 3.20 in terms of a unitary w such
= zS1. This is done for On in [24], from which some of the techniques deployed in
that wS1w∗
this section have actually been taken.
4 Extending endomorphisms of the Cuntz algebra
For what follows it may be convenient to recall that associated with any unitary V ∈ U(O2)
there is an endomorphism λV of O2, defined by λV(Si) = V Si, i = 1, 2. Notably, the other
way round is also true, i.e. any endomorphisms λ ∈ O2 comes from a unique V . The bijective
correspondence thus obtained is often named after Cuntz and Takesaki. To take just one important
example, the flip-flop λf is nothing but the automorphism corresponding to f ≐ S1S∗
2 + S2S∗
1 .
4.1 Uniqueness of the extensions
This section is mostly concerned with the problem of extending endomorphisms of O2 to endo-
morphisms of Q2. More precisely, we spot necessary and sufficient conditions for an extension
to exist. Before entering into the details, some comments are in order.
Indeed, the equations
we get are hardly ever easy to verify save for the endomorphisms we already know of to ex-
tend. Notwithstanding their intrinsic difficulty, they do provide general information when applied
to idO2. We can now go on with our discussion. To begin with, if V is a unitary of O2 such
that the associated endomorphism λV extends to an endomorphism λ of Q2, then one must have
V U S2 = V S1 = λV(S1) = λ(S1) = λ(U)λV(S2) = λ(U)V S2. Therefore, λ(U)V S2 = V U S2
and thus U ∗V ∗λ(U)V S2 = S2. Thus, setting W = U ∗V ∗λ(U)V , it holds W S2 = S2 and
λ(U) = V U W V ∗
. We now examine whether such extensions exist. We always have
V S2S
∗
2 V
∗ + V U W V
∗
V S2S
∗
∗
2 V
V W
∗
∗
V
U
∗
∗
2 + U S2S
∗
2 U
∗
= V V
= 1
∗
∗)V
= V(S2S
15
and we must have
or, equivalently,
V S2V U W V
∗
=(V U W V
∗)2
V S2
λ(U) = λ(U
2)V S2
V S2 = V U W V
∗
V U W V
∗
V S2 = V U W U W S2 = V U W U S2 .
We have thus shown the following result.
λV extends to an endomorphism of Q2 if and only if there exists a unitary W ∈ Q2 such that
W S2 = S2 and S2V U W V ∗
= U W U S2. For any such W , we have an extension λ = λV,W with
Proposition 4.1. Let V ∈ U(O2) and let λV ∈ End(O2) be the associated endomorphism. Then
λ(U) = V U W V ∗
Furthermore, the endomorphism λ is necessarily injective because of Q2 being simple. Moreover,
if λV is an automorphism of O2, then λ is surjective if and only if the associated W is contained
in λ(Q2). Moreover, for the extensions built above the following composition rule holds:
As shown later, the W defined above is uniquely determined in all the cases we have examined.
.
λV,W ◦ λV ′,W ′ = λλV(V ′)V,W V ∗λV,W(W ′)V
As an example, if ϕ is the canonical shift introduced in Sect. 2, we have
where θ = ∑2
i,j =1 SiSjS∗
i S∗
j ∈ U(F 2
2) is the self-adjoint unitary flip.
ϕ = λθ,U ∗θU 2θ
It is interesting to note that the extensions of the gauge automorphisms we have considered all
work with W = 1. This is not a case. In fact, the converse also holds true.
Proposition 4.2. Let V ∈ U(O2). If the associated endomorphism λV ∈ End(O2) extends to
λV,1, that is to say the choice W = 1 does yield an extension, then V = z1, for some z ∈ T.
= U 2S2. But
= U , i.e. V commutes with
□
Proof. If we put W = 1 in the equality S2V U W V ∗
U 2S2 = S2U , and so we must have S2V U V ∗
= S2U . Hence V U V ∗
= U W U S2, we get S2V U V ∗
= 1
= S2S∗
2 + U S2S∗
2 U ∗
2 + U W S2S∗
2 W ∗U ∗
U W , W S2 = S2 and W S1 = S1W . Indeed, S2S∗
Extensions of the identity map of O2, which obviously correspond to V = 1, may be looked
S1W = W S1, as stated. Obviously, the trivial choice W = 1 corresponds to the trivial extension.
U . Since V is a unitary, we also have V ∈ C ∗(U)′ ∩ O2, which concludes the proof.
at more closely. If we define W ≐ U ∗λ(U), we find that W is a unitary in Q2 such that λ(U) =
and S2U W = (U W)2S2, so that U 2S2W = U W U W S2 and thus U S2W = W U S2. Hence,
Proposition 4.3. If W ∈ U(Q2) is such that W S2 = S2 and W S1 = S1W , then W = 1.
Proof. This is in fact a straightforward application of the fact that C ∗(S1)′ ∩ Q2 = C1. However,
produced in [4]. This acts on the Hilbert space H = L2([−1, 1]) through the pure isometries
S1, S2 ∈ B(H) given by the formulas:
Ó2f(2t − 1)
(S1f)(t) = 0
(S2f)(t) = Ó2f(2t + 1)
we can also give an alternative if longer proof that only depends on the theorem of Larsen and
Li we have quoted. The computations are easily made in the irreducible representation of O2
for −1 ≤ t ≤ 0
for 0 < t ≤ 1
for −1 ≤ t ≤ 0
for 0 < t ≤ 1
0
16
:
1 and(S∗
1)n
f(2nt − ∑k
i=0 2i)
Ó2
2 n
f)(t) =
(S
n
n
Ó2
1 f)(t) = 0
1)n
((S
∗
and
for −1 ≤ t ≤ 1 − 1
2n−1
for 1 − 1
2n−1 < t ≤ 1
2 f)(t) = Ó2
2
1 f)(t) = Ó2
2
f t+1
f t−1
2 and (S∗
S1 and S2, namely U S2 = S1U . By virtue of the result of Larsen and Li we mentioned above,
we can then regard this representation as a representation of Q2 as well, which allows us to think
2 for every f ∈ H. The unitary
Note that (S∗
operator U ∈ U(H) given by(U f)(t) = f(−t) for every t ∈ [−1, 1] is an intertwiner between
of Q2 as a subalgebra of B(H). In order to prove the proposition, we will actually show even
more: any unitary W ∈ B(H) such that W S2 = S2 and W S1 = S1W must be the identity
by Pn ≐ (S1)nS2S∗
every n and PnPm = PmPn = 0 for every m ≠ n. Therefore Qn ≐ ∑n
operator I on H. To accomplish this task, we define a sequence of orthogonal projections given
for each n ∈ N. It is straightforwardly checked that W Pn = Pn for
k=0 Pk is still an orthogonal
projection such that W Qn = Qn. Accordingly, the conclusion will be easily achieved once we
have proved that Qn converges to I in the strong operator topology. As easily recognized, we have
the following explicit formulas for Sn
1)n
2(S∗
We can use them to see that Pn is the projection corresponding to the multiplication operator by
f t + 2n − 1
2n, i.e. the characteristic function of the interval1− 1
χ1− 1
Qn is nothing but the projection associated with χ−1,1− 1
topology, which was to be proved.
2n−1 ,1− 1
2n
2n−1 , 1− 1
2n. As a consequence,
2n. Hence Qn → 1 in the strong operator
□
Remark 4.4. The representation of Q2 recalled in the proof of the above result is not equivalent
to the canonical representation, merely because its restriction to O2 is still irreducible, as proved
in [4], whereas the restriction of the canonical representation to O2 is not, as we remarked.
We are at last in a position to prove the following result that says that a non-trivial endomor-
phism of Q2 cannot fix O2 pointwise.
Theorem 4.5. If Λ ∈ End(Q2) is such that Λ ↾O2
= idO2 , then Λ = idQ2.
Proof. A straightforward application of the former proposition.
□
Remark 4.6. Actually, the theorem just obtained strengthens the information that the relative
commutant O′
2 ∩ Q2, then ad(u) is an automorphism fixing O2
pointwise. As such, ad(u) is trivial, hence u is a central element. Since Q2 is simple, u must be a
2 ∩ Q2 is trivial. For, if u ∈ O′
multiple of the identity, i.e. O′
2 ∩ Q2 = C1.
As a simple corollary, we can also get the following property of the inclusion O2 ⊂ Q2.
Corollary 4.7. If Λ1 ∈ Aut(Q2) and Λ2 ∈ End(Q2) are such that Λ1 ↾O2
Λ1 = Λ2. In particular, Λ2 is an automorphism as well.
= Λ2 ↾O2 , then
Proof. Just apply the above theorem to the endomorphism Λ
O2.
−1
1 ◦ Λ2, which restricts trivially to
□
In particular, the extensions of both the flip-flop and the gauge automorphisms are unique.
Of course there are automorphisms of Q2 that do not leave O2 globally invariant. The most el-
= S2 =
ementary example we can come up with is probably ad(U). Indeed, ad(U)S1 = U S1U ∗
S2 Hence, ad(U)(O2) is not contained in O2, be-
is not in O2. Even more can be said. Indeed, ad(U)(F2) is not contained in O2 either.
ad(U)S2 = U S2U
U S1,
cause S1U
= S1U
= U
∗
∗
∗
∗
17
2) = U S1S∗
2 U ∗
= S2U S∗
2 U ∗
= S2U S∗
∗
1 + Λ(S2)S∗
were in O2, then U = S∗
2 S2U S∗
2 ≐ span{SαS
2 belongs to F2. Given that U S1S∗
2 U ∗
O2 although S1S∗
if U S1S∗
2 U ∗
easy computations involving the projections of Dk
2 . Furthermore, Λ leaves O2 globally invariant if and only if uΛ ∈ O2.
does not belong to
= S2U S∗
1 U U ∗
1 ,
1 S1 would in turn be in O2, which is not. Even so,
This is seen as easily as before, since for instance ad(U)(S1S∗
ad(U) does leave the diagonal subalgebra D2 globally invariant. This can be shown by means of
α s.t.∣α∣ = k} for every k ∈ N.
We would like to end this section by remarking that for each Λ ∈ End(Q2) there still exists
a unique uΛ ∈ U(Q2) such that Λ(S2) = uΛS2 and Λ(S1) = uΛS1, which is simply given by
uΛ = Λ(S1)S∗
This allows us to regard the map End(Q2) ∋ Λ → uΛ ∈ U(Q2) as a generalization of the well-
U(Q2) such that the correspondence S1 → uS1, S2 → uS2 do not extend to any endomorphism
of Q2. Examples of such u are even found in U(O2), as we are going to see in the next section,
the time being we observe that if a unitary u ∈ U(Q2) does give rise to an endomorphism Λu,
the equation uU S2 = Λu(U)uS2 must be satisfied. This says that Λu(U) = uU W u∗
W ∈ U(Q2) such that W S2 = S2 and S2uU W u∗
hitherto unmanageable description of End(Q2). At any rate, our guess is that the above equations
are hardly ever verified unless u is of a very special form, such as u = v ϕ(v∗) for any v ∈ U(Q2),
for some
= U W U S2. By the same computations as at
the beginning of the section, the converse is also seen to be true. Hence we obtain a complete if
known Cuntz-Takesaki correspondence. Nevertheless, this map is decidedly less useful for Q2
than it is for O2, not least because it is not surjective. In other words, there exist unitaries u in
where we shall give a complete description of the extensible Bogoljubov automorphisms. For
2 U ∗ + U S2S∗
corresponding to inner automorphisms, u = z1, corresponding to the gauge automorphisms αz, or
u = S2S∗
2 , corresponding to the flip-flop. In fact, this prediction is partly supported
by the result in the negative obtained in the next section. Moreover, it is still not clear at all whether
the map End(Q2) ∋ Λ → uΛ is injective, although its restriction to Aut(Q2) certainly is.
4.2 Extensible Bogoljubov automorphisms
We have seen some remarkable classes of automorphisms of O2 that extend to Q2. However,
there is no a priori reason to expect every endomorphism of O2 to extend automatically to an
endomorphism of Q2. In fact, we next give rather elementary examples of automorphisms of O2
that do not extend. Indeed, if we denote by ηα,β the automorphism of O2 defined by ηα,β(S1) =
αS1 and ηα,β(S2) = βS2 for any given α, β ∈ T, we have the following proposition.
Proposition 4.8. The automorphisms ηα,β ∈ Aut(O2) defined above extend to endomorphisms
of Q2 if and only if α = β.
Proof. Since S1 and S2 are unitarily equivalent in Q2, their images αS1 and βS2 would be uni-
tarily equivalent as well if an extension of ηα,β existed.
In particular, we would find {α} =
σp(αS1) = σp(βS2) = {β}, where σp denotes the point spectrum. This is absurd unless α = β,
in which case the corresponding endomorphism does extend being but a gauge automorphism. □
It is no surprise that the same proof as above covers the case of the so-called anti-diagonal
automorphisms. These are simply given by ρα,β(S1) = αS2 and ρα,β(S2) = βS1 for any given
α, β ∈ T. Again, an automorphism ρα,β extends precisely when α = β. To complete the picture,
we shall presently determine which Bogoljubov automorphisms of O2 extend to endomorphisms
of Q2. A suitable adaptation of some of the techniques developed by Matsumoto and Tomiyama
in [24] will be again among the ingredients to concoct the proof of the main result of this section.
This says that the extensible Bogoljubov automorphisms are precisely the flip-flop, the gauge
automorphisms, and their products, which altogether form a group isomorphic with the direct
18
product T × Z2. To this aim, consider a unitary matrix
A = a b
c d ∈ U(C
2).
and let α = λA be the corresponding automorphism of O2, i.e. α(S1) = aS1 + cS2 =(aU + c)S2
and α(S2) = bS1+dS2 =(bU +d)S2. Set f(U) =(bU +d) for short. The condition S2U = U
implies that f(U)S2α(U) = α(U)2
f(U)S2. Suppose that α is extensible and denote by α such
an extension. Finally, set U ≐ α(U), S2 ≐ α(S2), S1 ≐ α(S1). From now on we shall always be
focusing on the case where a, b, c, d are all different from zero. That said, the first thing we need
to prove is the extension is unique provided that it exists.
S2
2
Lemma 4.9. If λA extends, then its extension is unique.
1( S∗
Proof. By Proposition 3.10 all we need to check is that S1 is pure as an isometry acting on
1( S∗
1)n]. As Mn+1 ⊂ Mn, we have that E⋂n Mn
1)n] = {0}. To this aim, let us set Mn ≐
ℓ2(Z). This entails ascertaining that ⋂n Ran[ Sn
Ran[ Sn
= 0. For this it is enough to prove limn∥EMn ek∥ = 0 for every k. Now
with n1(α)
being the number of 1's occurring in α and n2(α) the number of 2's occurring in α. We set
L ≐ max{∣a∣,∣c∣} and observe that L < 1 by the hypotheses on the unitary matrix A. We have
1 = ∑∣α∣=n cαSα, where cα ≐ an1(α)cn2(α) ∈ C
to show limn EMn
the powers of S1 are given by Sn
= lim EMn strongly. Thus we are led
∗
that
∗
∥( S
1)n
ek∥ =∣cα(k)∣ ≤ L
n
→ 0 n → +∞
for a unique coefficient cα(k) that depends on k (this is actually a consequence of the fact that
∑∣α∣=k SαS∗
α = 1). This in turn implies the claim.
□
In light of the previous result, it is a very minor abuse of notation to denote by λA also its
extension to Q2 when existing.
Lemma 4.10. If λA extends, then U ∈ QT
2 .
Proof. Suppose that U is not in QT
automorphism αz such that αz( U) ≠ U . By applying αz to both sides of the equalities S2
2 . Then by definition there must exist a non-trivial gauge
U =
U 2 S2 and S2
S∗
2 + U S2
S∗
2
U ∗
= 1, we also get
αz( U)2 S2 = S2 αz( U)
2 + αz( U) S2
S2
S
S
∗
∗
2 αz( U)∗
= 1
which together say there exists an endomorphism Λ ∈ End(Q2) such that Λ(S2) = S2 and
Λ(U) = αz( U). Now Λ(S1) = Λ(U S2) = Λ(U)Λ(S2) = αz( U) S2 = ¯z αz( U S2) = ¯z αz( S1) =
S1 = λA(S1). A contradiction is thus arrived at because Λ and λA are different maps.
Now we introduce some lemmas to prove that α(U) is actually contained in C ∗(U).
, where ∣α∣ = ∣β∣ = k. If h ≥ 0, we have that (S
1 =
1 is a polynomial in U . The case h ≤ 0 can be handled with similar computations.
2 , the element(S
Lemma 4.11. For any x ∈ Bk
Proof. Suppose that x = SαS
2)kxSk
2)kxSk
∗(U).
1 belongs to C
∗
β U h
□
∗
∗
(S∗
2)kSαS∗
□
β U h Sk
Lemma 4.12. Let x ∈ QT
z ∈ C ∗(U).
2 such that the sequence (S∗
2)kxSk
1 converges to an element z. Then
19
k
∗
yk S
follows from the following inequality
∥z −(S
2)k
Proof. Let {yk}k≥0 be a sequence such that yk ∈ Bk
1∥ +∥(S
xS
1∥ ≤∥z −(S
2)k
∗(U).
Lemma 4.13. We have that U ∈ C
Proof. By applying įλA to the identity U Sk
that( S∗
1 = Sk
∗
k
2 U we get U Sk
1 = U . Therefore, U is in C ∗(U) thanks to Lemma 4.12, applied to x = U .
2)k U Sk
We have verified that α(U) = g(U) for some g ∈ C(T), which turns out to be vital in proving
Theorem 4.14. If α ∈ Aut(O2) is a Bogoljubov automorphism, then α extends to Q2 if and only
if α is the flip-flop, a gauge automorphism, or a composition of these two.
the following result.
U . For all k ∈ N we have
□
1 = Sk
2
Proof. By the discussion at the beginning of this section it is enough to consider the case in which
a, b, c, d are all different from zero. All the computations are henceforth made in the canonical
2 and yk → x normwise. Then the thesis
∗
2)k(x − yk)S
k
1∥ .
□
representation. The condition f(U)S2α(U) = α(U)2
f(U)S2g(U) = g(U)2
f(U)g(U
f(U)S2 yields
f(U)S2
2)S2 = f(U)g(U)2
S2.
Since the point spectrum of U is empty, f(U) is always injective, unless b = d = 0, in which
case A is not unitary. Thus g(U 2)S2 = g(U)2S2. At the function level we must then have
g(z2) = g(z)2
Appendix. Therefore g(U) = U l
for every z ∈ T. By continuity, we find that g(z) = zl
. We have that α(S1) = aS1 + cS2 = bU l+1
the
S2 + dU lS2. If we
compute the above equality on the vectors em, we get
, for this see e.g.
ae2m+1 + ce2m = be2m+l+1 + de2m+l.
which is to be satisfied for each m ∈ Z. Therefore, there are only two possibilities to fulfill these
conditions:
1. l = 1, and a = d ≠ 0, b = c = 0;
2. l = −1, and b = c ≠ 0, a = d = 0.
The first corresponds to gauge automorphisms, whilst the second to the flip-flop and its composi-
tions with gauge automorphisms.
□
5 Outer automorphisms
In this section the group Out(Q2) is shown to be non-trivial. More precisely, it turns out to be a
non-abelian uncountable group. A thorough description of its structure, though, is still missing.
As far as we know, it might well be chimerical to get.
20
5.1 Gauge automorphisms and the flip-flop
Below the flip-flop and non-trivial gauge automorphisms are proved to be outer.
In fact, this
parallels analogue known results for O2. Since gauge automorphisms are more easily dealt with,
we start our discussion focusing on them first. The next result (partly) follows from Proposition
3.21, however we give an alternative proof because it will shed further light on the properties of
the gauge automorphisms, see Remark 5.2.
Theorem 5.1. The extensions to Q2 of the non-trivial gauge automorphisms of O2 are still outer
automorphisms (and they are not weakly inner in the canonical representation).
namely U = V U V ∗
Proof. We shall argue by contradiction. From now on Q2 will always be thought of as a concrete
subalgebra of B(ℓ2(Z)) via the canonical representation. We will actually prove: no unitary
V ∈ B(ℓ2(Z)) can implement a non-trivial gauge automorphism. Indeed, let z ∈ T different
from 1, and let Λz ∈ Aut(Q2) the corresponding gauge automorphism. If V is a unitary operator
on ℓ2(Z) such that Λz = ad(V) ↾Q2, then in particular we must have Λz(U) = ad(V)(U),
take the form V = f(U), for some f ∈ L∞(T); in particular it belongs to W ∗(U) too. But we
also have zS2 = Λz(S2) = ad(V)(S2) = V S2V ∗
∗(U) and e0 is a separating vector for W
, that is to say zS2V = V S2. If we now compute
this identity between operators on the vector e0, we get zS2V e0 = V e0, i.e. S2V e0 = ¯zV e0. As
a consequence, we also have V e0 = 0, merely because 1 is the only eigenvalue of S2. Since
V ∈ W
which is clearly a contradiction.
∗(U) as well, we finally find that V is zero,
. This shows that V commute with U . Since U generates a MASA, V must
□
Indeed, the non-trivial gauge
Remark 5.2. Actually, the proof given above says a bit more.
automorphisms of Q2 ⊂ B(ℓ2(Z)) are not weakly continuous. For if they were, they should
clearly extend to an automorphism of B(ℓ2(Z)), but this is absurd because the automorphisms of
B(ℓ2(Z)) are all inner.
Among other things, we also gain the additional information that Out(Q2) is an uncountable
group, in that different gauge automorphisms give raise to distinct classes in Out(Q2). Indeed,
= αw(x) for every x ∈ Q2. For the sake of complete-
any unitary u ∈ Q2 such that u αz(x)u∗
if αz and αw are two different gauge automorphisms, then by Proposition 3.21 there cannot exist
ness we should also mention that every separable traceless C ∗
uncountable many outer automorphisms [25].
-algebra is actually known to have
Remark 5.3. Notably, the former result also provides a new and simpler proof of the well-known
fact that the gauge automorphisms on O2 are outer. However, the case of a general On cannot be
recovered from our discussion, and must needs be treated separately, as already done elsewhere.
As for the flip-flop, instead, we start our discussion by showing it is a weakly inner automor-
phism, which is the content of the next result.
Proposition 5.4. The extension of the flip-flop to Q2 is a weakly inner automorphism.
plemented by a unitary in π(Q2)′′
V ∈ U(ℓ2(Z)) is the self-adjoint unitary given by V ek ≐ e−k−1, the equalities V S1V ∗
Proof. By definition, we only have to produce a representation π of Q2 such that λf is im-
. The canonical representation does this job well. For if
= S2
= S1 are both easily checked. Since the canonical representation is irreducible, the
□
and V S2V
proof is thus complete.
∗
This result should also be compared with a well-known theorem by Archbold [1] that the
flip-flop is weakly inner on O2.
21
Remark 5.5. The unitary V as defined above can be rewritten as V = PU = U ∗P, where P is the
self-adjoint unitary given by Pek = e−k, k ∈ Z. Obviously, V is in Q2 if and only if P is. We shall
=
2 = V f .
2)∗
prove that P is not in Q2 in a while. At any rate, we observe the equality V(S1V S∗
Finally, it is worth noting that U = λZ(1), and that P is the canonical intertwiner between λZ and
r(Z) and the copy
ρZ. In this picture, Q2 is thus the concrete C*-algebra on ℓ2(Z) generated by C ∗
1 ≐ f ∈ O2, which is immediately checked, and f V = S1V S∗
1 + S2V S∗
1 +S2V S∗
2 + S2S∗
of O2 provided by the canonical representation.
S1S∗
In spite of being weakly inner, λf is an outer automorphism, as is its restriction to O2. To
prove that, we first need to show that the unitary V above is up to multiplicative scalars the unique
operator in B(ℓ2(Z)) that implements the flip-flop.
Proposition 5.6. If W ∈ B(ℓ2(Z)) is such that ad(W) ↾Q2
Proof. First note that we must have ad(W 2) = idB(ℓ2(Z)) since the flip-flop is an involutive
automorphism and Q′′
is a multiple of 1, and therefore there is no loss
of generality if we also assume that W 2
. From the relation W S1W = S2, we
get S1W e0 = W e0. Hence W e0 = λe−1 for some λ ∈ T. From e0 = W 2e0 = λW e−1 we get
W e−1 = ¯λe0. We now show that either λ = 1 or λ = −1. Indeed, from U W = W U ∗
it follows
that W e−1 = W U ∗e0 = U W e0 = U(λe−1) = λe0 = ¯λe0, which in turn implies that λ is real. Of
2 = B(ℓ2(Z)). Hence W 2
course, we only need to deal with the case λ = 1. The conclusion is now obtained at once if we
use the equality W U = U ∗W inductively, for W ek+1 = W U ek = U ∗W ek = U ∗e−k−1 = e−k−2,
□
as maintained.
= λf , then W = λV for some λ ∈ T.
= 1, i.e. W = W ∗
Remark 5.7. Of course, the uniqueness of V could also have been obtained faster merely by
irreducibility of Q2. However, the proof displayed above has the advantage of showing how we
came across the operator V .
Here finally follows the theorem on the outerness of the flip-flop.
Theorem 5.8. The extension of the flip-flop is an outer automorphism.
Proof. Thanks to the former result, all we have to prove is that P is not in Q2, which en-
tails checking that P cannot be a norm limit of a sequence xk of operators of the form xk =
∑k ckSαk Sβ∗
for some finite sum of the kind ∑α,β,h cα,β,hSαSβU h
U hk . Indeed, if this were the case, we should have ε >∥P −∑α,β,h cα,β,hSαS∗
, with ε > 0 as small as needed. If so, we
β U h∥
k
would also find the inequality
∥e−n − =
α,β,h
cα,β,hSαS
h
∗
β U
en∥ =∥Pen − =
α,β,h
cα,β,hSαS
h
∗
β U
en∥ < ε
This inequality, though, becomes absurd as soon as ε < 1 and n is sufficiently large, i.e. n is
bigger than the largest value of ∣h∣, for we would have ∥e−n − ∑α,β,h cα,β,hSαS∗
1 +∥∑α,β,h cα,β,hSαS
≥ 1, as ∑α,β,h cα,β,hSαS
β U hen∥2
∗
β U hen ∈ H+.
β U hen∥2
=
□
∗
Now, as we know λf is outer, we would also like to raise the question whether there exists a
representation of Q2 in which λf is not unitarily implemented. The answer would indeed complete
our knowledge of λf itself.
5.2 A general result
We saw above that the flip-flop is an outer automorphism. This is not an isolated case, for every
automorphism that takes U to its adjoint must in fact be outer. This is the content of the next result.
22
∗
∗
= U
. To
phism.
= U ∗
= U , which
is an outer automor-
such that W U W ∗
, then we have W U W ∗
that such a W cannot be in Q2.
Proof. All we have to prove is that there is no unitary W ∈ Q2 such that W U W
= PU P, hence PW U(PW)∗
If f is a continuous function, there is not much to say, for
is in Q2 as well, which we
know is not the case. The general case of an essentially bounded function is dealt with in much
Theorem 5.9. Every automorphism α ∈ Aut(Q2) such that α(U) = U ∗
this aim, we will be working in the canonical representation. If W ∈ B(H) is a unitary operator
says that PW commutes with U . Therefore, PW = f(U) for some f ∈ L∞(T) with∣f(z)∣ =
1 a.e. with respect to the Haar measure of T, hence W = Pf(U). Then we need to show
Pf(U) ∈ Q2 would immediately imply that P = Pf(U)f(U)∗
the same way apart from some technicalities to be overcome. Given any f ∈ L∞(T) and ε > 0,
thanks to Lusin's theorem we find a closed set Cε ⊂ T such that µ(T\ Cε) < ǫ and f ↾Cε is
continuous. This in turn guarantees that there exists a continuous function gε ∈ C(T, T) that
coincides with f on Cε by an easy application of the Tietze extension theorem. If Pf(U) is in
Q2, then Pf(U)gε(U)∗
whose support is contained in T\Cε. In particular, we can rewrite Pf(U)gε(U)∗
as P +Phε(U).
If the latter operator were in Q2, then we could find an operator of the form ∑α,β,h cα,β,hSαS∗
such that∥P + Phε(U) − ∑α,β,h cα,β,hSαS∗
β U h∥ < ε. If N is any natural number greater than
the maximum value of∣h∣ as h runs over the set the above summation is performed on, we should
β U heN∥ < ε, namely
have∥PeN + Phε(U)eN − ∑α,β,h cα,β,hSαS
∥e−N + Phε(U)eN − =
is also in Q2. Note that f ¯gε = 1 + hε, where hε is a suitable function
eN∥ < ε
cα,β,hSαS
β U h
∗
β U
∗
h
α,β,h
But then we should also have
∥e−N +Phε(U)eN − =
α,β,h
cα,β,hSαS
h
∗
β U
eN∥ ≥∥e−N − =
α,β,h
cα,β,hSαS
h
∗
β U
eN∥−∥Phε(U)eN∥
Hence
ε >∥e−N + Phε(U)eN − =
α,β,h
cα,β,hSαS
h
∗
β U
eN∥ ≥ 1 −∥Phε(U)eN∥
1
f
2 ≤ 2ε
−1)z
−N∣2
dµ(z)) 1
2 and is accordingly smaller than
are all outer, sending U in U ∗
2 . The above inequality becomes absurd as soon as ε is taken small enough. □
As a straightforward consequence, we can immediately see that the class of the flip-flop in
The conclusion is now got to if we can show that∥Phε(U)eN∥ is also as small as needed. But the
norm ∥Phε(U)eN∥ is easily computed in the Fourier transform of the canonical representation,
where it takes the more workable form(∫∣hε(z
2µ(T\ Cε) 1
Out(Q2) do not coincide with any of the classes of the gauge automorphisms. In other terms,
the automorphisms αz ◦ λ−1
by χ−1 the automorphism such that χ−1(S2) ≐ S2 and χ−1(U) ≐ U
result also applies to χz ≐ χ−1 ◦ αz, which are outer as well by the same reason. Also note that
χz(S2) = zS2. More interestingly, if z is not 1, the corresponding χz yields a class in Out(Q2)
other than the one of the flip-flop. To make sure this is true, we start by noting that χ−1 and λf do
not commute with each other. Even so, they do commute in Out(Q2), in that they even yield the
same conjugacy class. Indeed, we have ad(U ∗) ◦ χ−1 = λf , or equivalently λf ◦ χ−1 = ad(U ∗),
which in addition says that λf ◦ χ−1 has infinite order in Aut(Q2) while being the product of
two automorphisms of order 2. From this our claim follows easily. For, if χz ◦ λf is an inner
automorphism, the identity χz ◦ λf = χz ◦ χ−1 ◦ λf = χz ◦ ad(U) implies at once that χz is inner
as well, which is possible for z = 1 only. However, the classes [χz] and [λf] do commute in
2). In order to prove that Out(Q2) is not abelian,
Out(Q2), because χz ◦ λf ◦ χ
we still need to sort out a new class of outer automorphisms. This will be done in the next sections.
z ◦ λf = ad(U
. Furthermore, if we now denote
, then the above
= U
−1
−1
∗
23
6 Notable endomorphisms and automorphism classes
2k+1
isfies the two defining relations of Q2. This means that the map that takes S2 to itself and U
extends to an endomorphism of Q2, which will be denoted by χ2k+1. Trivially, this
to U
α(S2) = S2
6.1 Endomorphisms and automorphisms ααα such that α(S2) = S2
α(S2) = S2
For any odd integer 2k + 1, whether it be positive or negative, the pair (S2, U 2k+1) still sat-
endomorphism extends the identity automorphism of C ∗(S2). A slightly less obvious thing to
deed, χ2k+1(S1) = χ2k+1(U S2) = U
note is that these endomorphisms cannot be obtained as extensions of endomorphisms of O2. In-
S2 is not in O2: if it were, we would
also find that S∗
= U k
would be in O2, which is not.
In this way we get a class endomorphisms χ2k+1, k ∈ Z with χ1 = id and χ−1 being clearly
an automorphism of order two. All these endomorphisms commute with one other and we have
χ2k+1 ◦ χ2h+1 = χ(2k+1)(2h+1) for any k, h ∈ Z. Phrased differently, the set {χ2k+1 ∶ k ∈ Z}
is a semigroup of proper endomorphisms of Q2. One would like to know if the endomorphisms
singled out above give the complete list of the endomorphisms of Q2 fixing S2. In other words,
the question is whether the set
2 U ∗U U 2kS2 = S∗
1 U 2k+1S2 = S∗
2k+1
2 S2U k
2k+1
S2, and U
U2 ≐{V ∈ U(Q2)∣ V
2
S2 = S2V, S2S2
∗ + V S2S
∗
2 V
∗
= 1}
contains elements other than the U 2k+1
with k ∈ Z above. As a matter of fact, answering this
question in its full generality is not an easy task. An interesting if partial result does surface,
though, as soon as we introduce an extra assumption. Going back to the endomorphisms χ2k+1,
we next show they are all proper apart from χ±1. The proof cannot be considered quite elementary,
in that it uses the maximality of C ∗(U).
Proposition 6.1. None of the endomorphisms χ2k+1 is surjective if 2k + 1 ≠ ±1.
Proof. Let A ⊂ Q2 the C ∗
-subalgebra of those x ∈ Q2 such that χ2k+1(x) ∈ C ∗(U). We clearly
have C ∗(U) ⊂ A. By simplicity of Q2 the endomorphism χ2k+1 is injective, which means A is
still commutative. Therefore, by maximality of C ∗(U), we must have A = C ∗(U). From this it
now follows that U is not in the range of χ2k+1, for the restriction χ2k+1 ↾C∗(U) is induced, at the
spectrum level, by the map T ∋ z ↦ z2k+1
∈ T.
□
In addition, in the canonical representation the endomorphisms χ2k+1 cannot be implemented
by any unitary W ∈ B(ℓ2(Z)). Indeed, we can state the following result.
B(ℓ2(Z)) such that V S2 = S2V and V U V ∗
= U 2k+1
.
Proposition 6.2. Let 2k + 1 be an odd integer different from 1. Then there is no unitary V in
Proof. The proposition is easily proved by reductio ad absurdum. Let V be such a unitary as in
the statement. From V S2e0 = S2V e0 we deduce that V e0 is an eigenvector of S2 with eigenvalue
1. Without loss of generality, we may assume that V e0 = e0. Now,
V eh = V U
h
e0 = U(2k+1)h
V e0 = U(2k+1)h
e0 = e(2k+1)h, h = 0, 1, 2, . . .
and, similarly,
V e−h = e−(2k+1)h, h = −1, −2, . . . .
To conclude, it is now enough to observe that V is not surjective whenever 2k + 1 ≠ −1, whereas
the case of 2k + 1 = −1 leads to V being equal to P, which does not belong to Q2.
□
Rather than saying what the whole U2 is, we shall focus on its subset U2 ∩ C ∗(U) instead,
which is more easily dealt with. This task is accomplished by the next result.
24
Theorem 6.3. The set U2∩C ∗(U) is exhausted by the odd powers of U , i.e. U2 ∩ C ∗(U) =
{U
2k+1 ∶ k ∈ Z}.
Proof. Let W ∈ U2∩C ∗(U). Then there exists a function f ∈ C(T, T) such that W = f(U).
The condition S2W = W 2S2 can be rewritten as S2f(U) = f(U)2S2. On the other hand, we also
have S2f(U) = f(U 2)S2. Therefore f(U 2)S2 = f(U)2S2, that is f(U 2) = f(U)2
ingly, the function f must satisfy the functional equation f(z2) = f(z)2
function f must be of the form f(z) = zn
. Accord-
. Being continuous, our
for some integer n ∈ Z, see the Appendix. This means
= 1, we finally
□
that W = U n
find that n is forced to be an odd number, say n = 2k + 1.
. If we also impose the condition on the ranges S2S∗
2 + U nS2S∗
2 U −n
The result obtained above can also be stated in terms of endomorphisms of Q2. With this in
mind, we need to introduce a bit of notation. In particular, we denote by EndC∗(S2)(Q2, C ∗(U))
the semigroup of those endomorphisms of Q2 that fix S2 and leave C ∗(U) globally invariant.
Corollary 6.4. The semigroup EndC∗(S2)(Q2, C ∗(U)) identifies with {χ2k+1 ∶ k ∈ Z}. As a
result, we also have
AutC∗(S2)(Q2, C
∗(U)) ={id, χ−1} ≅ Z2
back to elsewhere.
The foregoing result might possibly be improved by dropping the hypothesis that our endo-
morphisms leave C ∗(U) globally invariant also. This is in fact a problem we would like to go
α(U) = U
6.2 Automorphisms ααα such that α(U) = U
α(U) = U
In this section we study those endomorphisms and automorphisms Λ ∈ End(Q2) such that
Λ(U) = U . Of course, the problem of describing all of them amounts to determining the set
W = 1, W U = U
W, W W
2
∗ + U W W
∗
U
∗
S2 ≐{W ∈ Q2 ∶ W
∗
= 1} .
Curiously enough, it turns out that S2 can be described completely, which is what this section is
chiefly aimed at. We start our discussion by sorting out quite a simple class of automorphisms
the automorphism of Q2 given by
of that sort. Given a function f ∈ C(T, T) we denote by βf
βf(U) = U and βf(S2) = f(U)S2, which is well defined because the pair (f(U)S2, U) still
satisfies the two defining relations of Q2. Note that βf ◦ βg
subgroup of AutC∗(U)(Q2) and that a constant function f(z) = w gives back the gauge auto-
morphism αw. Furthermore, we have the following result, which gives a sufficient condition on f
for the corresponding βf
f ↦ βf
to be outer. As the condition is not at all restrictive, the correspondence
, which is one to one, provides plenty of outer automorphisms.
so that we obtain an abelian
= βf ⋅g
Proof. If V ∈ Q2 is a unitary such that βf
Proposition 6.5. If f ∈ C(T, T) is such that f(1) ≠ 1, then βf
= ad(V), then V commutes with U and therefore it
is of the form g(U) for some g ∈ C(T, T) by maximality of C ∗(U). The condition βf(S2) =
, that is f(U)S2 = g(U)g(U 2)∗S2. But
ad(V)(S2) yields the equation f(U)S2 = g(U)S2g(U)∗
then g and f satisfy the relation f(z) = g(z)g(z2) for every z ∈ T. In particular, the last equality
says that f(1) = g(1)g(1) = 1.
is an outer automorphism.
□
However, the condition spotted above is not necessary. This will in turn result as a consequence
of the following discussion. We will be first concerned with the problem as to whether an automor-
phism βf
may be equivalent in Out(Q2) to a gauge automorphism. If so, there exist z0 ∈ T and
W ∈ U(Q2) such that W U W
that W = h(U) for some h ∈ C(T, T), which makes the second into h(U)f(U)h(U
= z0S2. As usual, the first relation says
= z0,
= U and W f(U)S2W
2)∗
∗
∗
25
that is h satisfies the functional equation h(z)f(z)h(z2) = z0. The latter says in particular that z0
is just f(1). We next show that there actually exist many continuous functions f for which there
is no continuous h that satisfies
h(z)h(z
2) = f(z)f(1) ≐ Ψ(z) .
(3)
n
1
, then we have the following.
not equivalent with any of the gauge au-
π + ε ≤ θ ≤ 2π − ε with 0 < ε ≤ π
4
Here is our result, which provides examples of βf
Note that Ψ(1) = 1 and that there is no loss of generality if we assume h(1) = 1 as well. By
evaluating (3) at z = −1 we find h(−1) = 1 provided that f(−1) = f(1).
2) = Ψ(z) are com-
Remark 6.6. By density, the continuous solutions of the equation h(z)h(z
pletely determined by the values they take at the 2n−th roots of unity. Furthermore, the value of
such an h at a point z with z2
k=0 Ψ(z2k) .
∏n−1
= 1 is simply given by the interesting formula h(z) =
The latter is easily got to by induction starting from the relation h(z2) = h(z)Ψ(z).
tomorphisms. Let f ∈ C(T, T) be such that f(eiθ) = 1 for 0 ≤ θ ≤ π and f(eiθ) = −1 for
Proposition 6.7. If f ∈ C(T, T) is a function as above, then the associated βf
Proof. With the above notations, suppose that h(z) is a solution of (3) such that h(1) = h(−1) =
1, which is not restrictive. Since Ψ(i) = Ψ(ei π
2) = 1 and h(i2) = h(−1) = 1, we immediately see
that h(i) = 1. As Ψ(z) = 1 for 0 ≤ θ ≤ π, by using the functional equation we find that h(ei π
2n) =
4)) = h(i) = 1,
which in turn gives h(e5i π
2n+2 = −1. This proves that any
We can now devote ourselves to answering the question whether Out(Q2) is abelian. It turns
functions f do not commute in Out(Q2) with the flip-flop. To begin with, if βf
with λf in Out(Q2), then there must exist a unitary V ∈ Q2 such that λf ◦ βf ◦λf = ad(V) ◦ βf
Exactly as above, the unitary V is then a continuous function of U , say V = h(U). In addition,
4) = −1. By induction we also see he
corresponding to suitable
does commute
.
out that it is not. Our strategy is merely to show that automorphisms βf
solution h of the functional equation with f as in the statement cannot be continuous.
4) = −1 and h(e2(5i π
1 for any n ∈ N. Consider then z = e5i π
4 . We have Ψ(e5i π
to any gauge automorphism.
is not equivalent
□
5πi
we also have
f(U
∗)S2 = h(U)f(U)S2h(U)∗
= h(U)f(U)h(U
2)∗
S2
apart from a sufficiently small neighborhood of z = ei 9π
8 .
the unknown function h, with f being given instead. We next exhibit a wide range of continuous
functions f for which the corresponding h does not exist. To state our result as clearly as possible,
and so we find that f and h satisfy the equation f(¯z) = h(z)f(z)h(z2) for every z ∈ T, which
can finally be rewritten as f(¯z)f(z) = h(z)h(z2), to be understood as an equation satisfied by
8 ) = i, and f(z) = 1 everywhere
we fix some notation first. Let f ∈ C(T, T) be such that f(ei 9π
Proposition 6.8. If f ∈ C(T, T) is as above, then βf
Out(Q2).
that he
Theorem 6.9. The group Out(Q2) is not abelian.
Notably, this also yields the announced result on Out(Q2).
Proof. Repeat almost verbatim the same argument as in the foregoing proposition, now verifying
2n = 1 first and then he
2n+3 = −1.
does not commute with the flip-flop in
9πi
□
πi
26
∗
= s
Now we have got a better guess of what S2 might be, we can finally prove that it is in fact
still requires some preliminary work. First observe that given any s ∈ S2, a straightforward
Proof. We start with the equality s
substitute the above expressions. This leads to s
exhausted by isometries of the form f(U)S2, where f is a continuous function onto T. This
computation shows that both s∗S2 and s∗S1 commute with U , but then by maximality of C ∗(U)
we can rewrite them as h(U) and g(U) respectively, with h and g being continuous functions.
Lemma 6.10. There exists a continuous function f such that s = f(U)S2.
1 +(s
S1)S
2 , that is s = S1g(U)∗ +
1 + h(U)S
= U S2g(U)∗ + h(U 2)∗S2 = (U g(U 2)∗ + h(U 2)∗)S2. Therefore, our claim is true
S2h(U)∗
with f(z) = zg(z∗) + h(z2).
Lemma 6.11. With the notations set above, for every z ∈ T we have∣h(z)∣2 +∣g(z)∣2
Lemma 6.12. With the notations set above, for every z ∈ T we have zh(z)g(z) + g(z)h(z) = 0.
2) =(s
= g(U)S
Proof. It is enough to rewrite the equality s
Proof. Once again it is enough to rewrite the equality s∗U s = 0, which is merely the orthogonality
□
relation between s and U s, in terms of h and g.
s = 1 in terms of h and g.
∗
2 , in which we
∗(S1S
S2)S
= 1.
∗
1 + S2S
∗
□
∗
∗
∗
□
∗
∗
∗
∗
We are at last in a position to prove the main result on S2.
Theorem 6.13. If s ∈ S2, then there exists a f ∈ C(T, T) such that s = f(U)S2.
Proof. At this stage, all we have to do is prove that∣f(z)∣2
2) + ¯zg(z
2)h(z
∣f(z)∣2
As an immediate consequence, we finally gain full information on AutC∗(U)(Q2).
=g(z2)z + h(z2)(g(z
2) = 1 + zg(z2)h(z
2)z + h(z
= 1. But
Theorem 6.14. The equalities hold
2) = 1
□
EndC∗(U)(Q2) = AutC∗(U)(Q2) ={β
f ∶ f ∈ C(T, T)}
pointwise convergence.
Remark 6.15. The bijective correspondence f ↔ βf
In particular, the semigroup EndC∗(U)(Q2) is actually a group isomorphic with C(T, T).
is also a homeomorphism between C(T, T)
equipped with the uniform convergence topology and AutC∗(U)(Q2) endowed with the norm
We end this section by proving that AutC∗(U)(Q2) is in addition a maximal abelian subgroup
of Aut(Q2).
Theorem 6.16. The group AutC∗(U)(Q2) is maximal abelian in Aut(Q2).
Proof. We have to show that if α ∈ Aut(Q2) commutes with any element of AutC∗(U)(Q2)
then α is itself an element of the latter group. Now, the equality α ◦ ad(U) = ad(U) ◦ α gives
ad(α(U)) = ad(U). Therefore, α(U) = zU for some z ∈ T by simplicity of Q2. The conclusion
is then achieved if we show that actually z = 1. Exactly as above, we also have ad(α(g(U))) =
ad(g(U)) for any g ∈ C(T, T). Again, thanks to simplicity we see that g(zU) = g(α(U)) =
α(g(U)) = λg(U) for some λ ∈ T, possibly depending on g. In terms of functions we find the
equality g(zw) = λgg(w), which can hold true for any g ∈ C(T, T) only if z = 1. Indeed, when
2(T) as(Φzf)(w) ≐ f(zw). Finally, the case of a z that is a root of unity is dealt
z ≠ 1 is not a root of unity the characters wn
Φz acting on L
with similarly.
are the sole eigenfunctions of the unitary operator
□
27
Remark 6.17. The findings above are worth comparing with a result obtained in [16] that the
group of automorphisms of O2 fixing the diagonal D2 is maximal abelian too.
Moreover, the theorem enables to thoroughly describe the automorphisms that send U to its
adjoint, which have been shown to be automatically outer.
Theorem 6.18. If α is an automorphism of Q2 such that α(U) = U ∗
a suitable f ∈ C(T, T).
Proof. Just apply the former result to λf ◦ α.
, then α(S2) = f(U)S2 for
□
Finally, the automorphisms βf
can also be characterized in terms of the Cuntz-Takesaki gener-
alized correspondence we discussed at the end of Section 4.1. Indeed, they turn out to be precisely
those Λ ∈ End(Q2) for which the corresponding W ≐ U
Λ(U)u equals 1, where u stands
= U . Therefore by maximality there exists a function f ∈ C(T, T) such that u = f(U),
for uΛ for brevity. For W = 1 we find in fact the equality S2uU u∗
uU u∗
that is Λ = βf
= U 2S2 = S2U , whence
u
∗
∗
.
The following discussion addresses the problem of studying those automorphisms Λ of Q2 such
α(U) = zU
6.3 Automorphisms ααα such that α(U) = zU
α(U) = zU
that Λ(U) = zU , with z ∈ T. We start tackling the problem by defining two operators acting on
ℓ2(Z). The first is the isometry S′
which is given by Uzek ≐ zkek. The following commutation relations are both easily verified:
zek ≐ zke2k. The second is the unitary Uz,
z, which is given by S′
• UzU = zU Uz
• UzS2 = S′
zUz
The first relation can also be rewritten as ad(Uz)(U) = zU . We caution the reader that at this
level ad(Uz) makes sense as an automorphism of B(ℓ2(Z)) only, because we do not know yet
whether Uz sits in Q2. If it does, the first relation says, inter alia, that Q2 also contains a copy
of the noncommutative torus Az in a rather explicit way, which is worth mentioning. In order to
decide what values of z do give a unitary Uz belonging to Q2, the first thing to note is that if Uz is
in Q2, then it must be in the diagonal subalgebra D2, as shown in the following lemma.
Lemma 6.19. If Uz is in Q2, then Uz ∈ D2.
Proof. A straightforward application of the equality D
2 = ℓ∞(Z). □
The second thing to note is that the unitary representation T ∋ z ↦ Uz ∈ U(B(ℓ2(Z))) is only
z ↦ Uz is only strongly continuous, which means the set {Uz}z∈T is not separable with respect
strongly continuous. This implies that not every Uz is an element of Q2. For the representation
to the norm topology, whereas Q2 obviously is. The next result provides a first answer to the
question whether Uz belongs to D2 . More than that, it also gives an explicit formula for Uz.
′
2 ∩ Q2 = D2, as Uz is in D
′
n
= 1 for some natural number n, then Uz is in D2.
Proposition 6.20. If z ∈ T satisfies z2
Proof. Obviously only primitive roots have to be dealt with. But for such roots, say z = ei2π/2
,
the unitary Uz may in fact be identified to the sum ∑2k−1
j =0 zj Pj , where the projection Pj belongs
to D2, being more explicitly given by Pi1i2...ik , where the multi-index(i1, i2, . . . , ik) ∈{1, 2}k
is
the j-th with respect to the lexicographic order in which 2 < 1 and the multi-index itself is read
□
from right to left.
k
28
The automorphisms obtained above are of course of finite order. More precisely, the order of
ad(Uz) is just the same as the order of the corresponding z. In other words, what we know is
that the automorphism of C ∗(U) induced by the rotation on T by a 2n
-th root of unity extends to
an inner automorphism of Q2, whose order is still finite being just 2n
. Due to the lack of norm
continuity of the representation z → Uz, though, the case of a general z is out of the reach of the
foregoing proposition and must needs be treated separately with different techniques. To begin
with, we recall a result whose content should be well known. Nevertheless, we do include a proof
not only for the sake of completeness but also to set some notations we shall need in the following
considerations.
Proof. It is convenient to realize D2 as the concrete C ∗
Lemma 6.21. Any projection P ∈ D2 is in the linear algebraic span of{SαS∗
given by the Tychonoff product{1, 2}N
the characteristic function of a cylinder Cα = {x ∈ K ∶ x(k) = αk for any k = 0, 1, . . . ,∣α∣}
. If we do so, the projections of D2 are immediately seen
to identify with the characteristic functions of the clopens of K, and these are clearly the cylinder
sets in the product space. The conclusion now follows noting that for any multi-index α ∈ W2
-algebra C(K), with the spectrum K being
α}α∈W2.
corresponds indeed to SαS
□
∗
α.
At this point, it remains to show that Uz does not belong to D2 for any other values of z ∈ T.
Although this could be done by means of explicit computations, as it was in an early version of
this paper, we prefer to present a rather elegant method suggested by the referee.
The Cantor set K = {0, 1}N
n=0 xn2n
n=0 ↔ ∑∞
Z2 is by definition the completion of Z under the metric induced by the 2-adic absolute value, any
can also be realized as the ring of 2-adic integer numbers Z2 via
∈ Z2. In this picture the former
the bijective correspondence K ∋ x = {xn}∞
digit 2 has to be replaced by 0. Accordingly, as of now we think of D2 as C(Z2). Furthermore, as
f ∈ C(Z2) is uniquely determined by its restriction to Z ⊂ Z2. This gives an isometric inclusion
of D2 ≅ C(Z2) into C(Z) ⊂ ℓ∞(Z), which is nothing but the canonical representation of D2
on ℓ2(Z). To see this, it is enough to note that the generating projections SαS∗
α are indeed the
characteristic functions of the subsets{2nk + l ∶ k ∈ Z} ⊂ Z, where n =∣α∣ and l = ∑n−1
j =0 αj2j
.
Lemma 6.22. Let f ∈ ℓ∞(Z) ⊂ B(ℓ2(Z)). Then f is in D2 if and only if f ∶ Z → C extends to
Phrased differently, we have obtained the following useful characterization.
a continuous function f ∶ Z2 → C.
We are now in a position to state and prove the main result of the present subsection.
Theorem 6.23. Let z ∈ T. Then Uz ∈ D2 if and only if z is a root of unity of order a power of 2.
Proof. Thanks to Lemma 6.22, it is enough to make it plain when Z ∋ k ↦ zk
∈ C extends to a
continuous function of Z2. It is easily seen that this is the case if and only if z is a dyadic root of
□
unity.
For those z ∈ T such that Uz lies in Q2 we can say a bit more.
βf
Proposition 6.24. Let z ∈ T be a dyadic root of unity and let α ∈ Aut(Q2) be such that α(U) =
zU . Then there exists a f ∈ C(T, T) such that α(S2) = f(zU)S
Proof. By its very definition ad(Uz−1) ◦ α(U) = U . Therefore, we must have ad(Uz−1) ◦ α =
for some f ∈ C(T, T). But then f(U)S2 = βf(S2) = Uz−1 α(S2)Uz, i.e. α(S2) =
Uzf(U)S2Uz−1 = f(zU)UzS2Uz−1 = f(zU)S
Remark 6.25. We have already seen that if Uz ∈ Q2 then S′
fact, one can easily observe that Uz = S∗
z ∈ Q2. The converse, too, is true. In
z hence the claim follows. In particular, whenever Uz
is not in Q2, the corresponding ad(Uz) understood as an automorphism of the whole B(ℓ2(Z))
does not even leave Q2 globally invariant.
2 S′
′
z.
□
′
z.
29
f(z2) = f(z)2
A The functional equation f(z2) = f(z)2
f(z2) = f(z)2
This appendix presents a self-contained treatment of the functional equation f(z2) = f(z)2
, of
which we made an intensive use in the previous sections. Although the following facts might all
be well known, we do include complete arguments, because their proofs are not to be easily found
in the literature, however carefully examined.
on the torus
Proposition A.1. Let f be a continuous function from T to T such that f(z2) = f(z)2
z ∈ T. Then there exists a unique n ∈ Z such that f(z) = zn
.
for every
Proof. Thanks to the compactness of T and the continuity of f , the winding number of f is a
well-defined integer n ∈ Z, for details see e.g. Arveson's book [2, Chapter 4, pp 114-115]. The
new function g(z) ≐ z−nf(z) still satisfies our equation. Furthermore, the winding number of g
is zero by construction. Therefore, there exists h ∈ C(T, R) such that g(z) = e2πih(z) for every
z ∈ T. Rephrasing the equation in terms of h, we find that h(z2) − 2h(z) must be an integer for
generality if we also assume h(z
2) − 2h(z) = 0 for every z ∈ T. Being bounded, the function h
is then forced to be identically zero, which finally leads to f(z) = zn
every z ∈ T. By connectedness, the function h is thus a constant. Obviously there is no lack in
for every z ∈ T.
□
The above proposition can be regarded as a one-variable description of the characters of the
one-dimensional torus. It is worth pointing out, though, that it no longer holds true as soon as T is
replaced by the additive group R. In other words, there do exist continuous functions f ∶ R → T
, which are obtained by exponentiating non-linear
such that f(2x) = f(x)2
continuous functions g ∶ R → R such that g(2x) = 2g(x) for every x ∈ R. However, any such g
other than ft(x) ≐ eitx
cannot be everywhere differentiable with continuous derivative at 0.
for every n ∈ N and z ∈ T.
.
The proof given above can be further simplified if we assume that f satisfies a stronger func-
tional equation, i.e. f(zn) = f(z)n
Proposition A.2. If f ∈ C(T, T) satisfies f(zn) = f(z)n
exists a unique k ∈ Z such that f(z) = zk
Proof. Obviously, it is enough to prove that f(zw) = f(z)f(w) for any z, w ∈ T. Let z be a fixed
element of T, and let Cz ⊂ T be the set Cz ≐ {w ∈ T ∶ f(zw) = f(z)f(w)}. From the equality
f(zn) = f(z)n
we may note that Cz contains the set{zn ∶ n = 0, 1, 2, . . .}. By continuity of f
words, for such z, we have f(zw) = f(z)f(w) for any w ∈ T. The full conclusion is now easily
we have in addition that Cz is closed. Hence Cz = T if z = e2iπθ
for every z ∈ T and n ∈ N, then there
, with θ being irrational. In other
got to by density.
□
The function f that is 1 on R and f(z) = z on its complement still satisfies f(zn) = zn
Remark A.3. The continuity assumption cannot be left out in either the above propositions. To
see this, let R ⊂ T be the set ∪nHn, where Hn ⊂ T is the subgroup of the nth roots of unity.
for every
n ∈ N, as easily verified. Due to the density of R in the torus, this function is nowhere continuous.
Nevertheless, it is equal to the character z almost everywhere. This seems to indicate that any
measurable solution of the equation might equal a character almost everywhere. At any rate, it
do not enjoy automatic
is worth pointing up that the solutions of the equation f(zn) = f(z)n
continuity, unlike the solutions of the equation f(zw) = f(z)f(w), which are of course even
automatically differentiable.
Acknowledgments We would like to take this opportunity to thank L´aszl´o Zsid´o for a fruitful
conversation about the functional equation discussed in the Appendix. We are also grateful to
Nicolai Stammeier for his valuable comments on the first draft of this paper. Finally, we owe the
30
referee a debt of gratitude for his or her particularly attentive perusal of the manuscript, which
resulted in many improvements not only to notation and presentation but even to the proof of the
main result in the Appendix.
References
[1] R. J. Archbold, On the "flip-flop"automorphism of C ∗(S1, S2), Quart. J. Math. Oxford Ser.
(2) 30 (1979), 129-132.
[2] W. Arveson, A Short Course on Spectral Theory, Springer-Verlag, Graduate Texts in Mathe-
matics 209 (2002).
[3] S. Barlak, T. Omland, and N. Stammeier, On the K-theory of C ∗
-algebras arising from
integral dynamics, arXiv:1512.04496, to appear in Ergodic Theory Dynam. Systems.
[4] M. D. Choi and F. Latr´emoli`ere, Symmetry in the Cuntz algebra on two generators, J. Math.
Anal. Appl. 387 (2012), 1050 -- 1060.
[5] R. Conti, Automorphisms of the UHF algebra that do not extend to the Cuntz algebra, J.
Aust. Math. Soc. 89 (2010), 309 -- 315.
[6] R. Conti, J. H. Hong, and W. Szyma´nsky, The restricted Weyl group of the Cuntz algebra and
shift endomorphisms, J. Reine Angew. Math. 667 (2012), 177-191.
[7] R. Conti, J. H. Hong, and W. Szyma´nsky, Endomorphisms of graph algebras, J. Funct. Anal.
263 (2012), 2529-2554.
[8] R. Conti, J. H. Hong, and W. Szyma´nski, The Weyl group of the Cuntz algebra, Adv. Math.
231 (6) (2012), 3147-3161.
[9] R. Conti, J. H. Hong, and W. Szyma´nski, Endomorphisms of the Cuntz Algebras, Banach
Center Publ. 96 (2012), 81 -- 97.
[10] R. Conti, J. H. Hong, and W. Szyma´nski, On conjugacy of maximal abelian subalgebras
and the outer automorphism group of the Cuntz algebra, Proc. Roy. Soc. Edinburgh 145 A
(2015), 269 -- 279.
[11] R. Conti, J. Kimberley, and W Szyma´nski, More localized automorphisms of the Cuntz alge-
bras, Proc. Edinburgh Math. Soc. (Series 2) 53 (2010), 619 -- 631.
[12] R. Conti, M. Rørdam and W. Szyma´nski, Endomorphisms of On which preserve the canoni-
cal UHF-subalgebra, J. Funct. Anal. 259 (2010), 602 -- 617.
[13] R. Conti and W. Szyma´nski, Automorphisms of the Cuntz algebras, Progress in operator
algebras, noncommutative geometry, and their applications, 1 -- 15, Theta Ser. Adv. Math.,
15, Theta, Bucharest, 2012.
[14] R. Conti and W. Szyma´nski, Labeled Trees and Localized Automorphisms of the Cuntz Alge-
bras, Trans. Amer. Math. Soc. 363 (2011) no. 11, 5847 -- 5870.
[15] R. Conti and W. Szyma´nski, Unpublished.
[16] J. Cuntz, Automorphisms of certain simple C*-algebras,
in Quantum fields-algebras-
processes, ed. L. Streit, Springer, 1980.
31
[17] J. Cuntz, Simple C*-algebras generated by isometries, Commun. Math. Phys. 57 (1977),
173 -- 185.
[18] J. Cuntz, C ∗
-algebras associated with the ax + b-semigroup over N, K-theory and noncom-
mutative geometry, 201215, EMS Ser. Congr. Rep., Eur. Math. Soc., Z urich, 2008.
[19] K. Dykema and D. Shlyakhtenko, Exactness of Cuntz-Pimsner C ∗
-algebras, Proc. Edin-
burgh Math. Soc. (Series 2) 44 (2001), 425 -- 444.
[20] R. Exel, A. an Huef, and I. Raeburn, Purely infinite C ∗
-algebras associated to integer dila-
tion matrices, Indiana Univ. Math. J. 60 (2011), 1033 -- 1058.
[21] R. V. Kadison and I. M. Singer, Extensions of pure states, Amer. J. Math. 81 (1959), 383 -- 400.
[22] M. Laca, Endomorphisms of B(H) and Cuntz algebras, J. Operator Theory 30 (1993), 85 --
108.
[23] N. S. Larsen and X. Li, The 2-adic ring C*-algebra of the integers and its representations, J.
Funct. Anal. 262 (4) (2012), 1392 -- 1426.
[24] K. Matsumoto and J. Tomiyama, Outer automorphisms on Cuntz algebras, Bull. London
Math. Soc. 25 (1) (1993), 64 -- 66.
[25] J. Phillips, Outer Automorphisms of separable C ∗
-algebras, J. Funct. Anal. 70 (1987), 111 --
116.
[26] J. Renault, Cartan subalgebras in C ∗
algebras, Irish Math. Soc. Bull. (2008), 61, 29 -- 63.
[27] J. Tomiyama, On Some Types of Maximal Abelian Subalgebras, J. Funct. Anal. 10 (1972),
373 -- 386.
[28] J. Zacharias, Quasi-free automorphisms of Cuntz-Krieger-Pimsner algebras, C
∗
-algebras
(Munster, 1999), 262-272, Springer, Berlin, 2000.
32
|
1608.03986 | 1 | 1608 | 2016-08-13T14:25:49 | Classification of certain inductive limit actions of compact groups on AF algebras | [
"math.OA"
] | Let $A=\underrightarrow{\lim}{A_n}$ be an AF algebra, $G$ be a compact group. We consider inductive limit actions of the form $\alpha=\underrightarrow{\lim}{\alpha_n}$, where $\alpha_n\colon G\curvearrowright A_n$ is an action on the finite dimensional C*-algebra $A_n$ which fixes each matrix summand. If each $\alpha_n$ is inner, such actions are classified by equivariant K-theory by Handelman and Rossmann. However, if the actions $\alpha_n$ are not inner, we show that such actions are not classifiable by equivariant K-theory. We give a complete classification of such actions using twisted equivariant K-theory. | math.OA | math |
Classification of certain inductive limit actions of
compact groups on AF algebras
Qingyun Wang
May 13, 2019
Abstract
Let A = lim
−→
An be an AF algebra, G be a compact group. We consider
inductive limit actions of the form α = lim
αn, where αn : G y An is an
−→
action on the finite dimensional C*-algebra An which fixes each matrix
summand. If each αn is inner, such actions are classified by equivariant K-
theory in [4]. However, if the actions αn are not inner, we show that such
actions are not classifiable by equivariant K-theory. We give a complete
classification of such actions using twisted equivariant K-theory.
1 Introduction
Throughout the whole paper, we shall restrict our attention to compact group
actions, although some of the definitions and results are valid for more general
groups.
Following the convention in [11], if α : G y A is an action, we shall also use
(G, A, α) to denote this action and call A a G-C*-algebra. All actions are
assumed to be continuous in the point-norm topology, i.e. for each a ∈ A, the
map g → αg(a) is continuous.
Definition 1.1. Let A = lim
An be a C*-algebra, G be a compact group.
−→
Let αn : G y An be a sequence of actions such that the connecting maps
φn,m : An → Am are all equivariant. Then there is an induced action α : G y A.
We call α an inductive limit action in this case. We shall write
(G, A, α) = lim−→ (G, An, αn).
We say α is locally representable if each αn is induced by a representation
G → U (An). We say α is locally spectrally trivial
if each αn induces trivial
action on the spectrum of An.
We shall say that α is an inductive limit action on an AF algebra if each An
is finite dimensional in the above definition.
The classification of C*-algebras admitting an inductive limit structure has been
extensively studied. However, the equivariant version is far less well-understood.
1
For example, it's not known how to classify all inductive limit actions of a finite
group on a general AF algebra, except when the group is finite cyclic; see [3]
and [8]. There is a number of results on classification of locally representable
inductive limit actions, see e.g.
[4], [7], [1] and [2]. On the other direction,
inductive limits of actions on certain homogeneous algebras, which acts trivially
on the fiber but non-trivially on the spectrum, has been classified in [13].
The major invariant used in aforementioned results is the equivariant K-
theory. For the class of actions we shall consider, we will show that equivariant
K-theory is not enough for classification. We shall introduce the twisted equiv-
ariant K-theory for compact group actions on C*-algebras, which generalize the
ordinary equivariant K-theory. The main result of this paper is a complete clas-
sification using this new invariant. We expect that this new invariant will be
useful in future classifications of group actions on C*-algebras.
2 Actions on full matrix algebras
We start our analysis from actions on matrix algebras. we shall soon see that
one has to go beyond representable actions and look for more invariants than
just equivariant K-theory. For the definition of equivariant K-theory and it's
properties, our standard reference is [11].
Let Mn be the n × n complex matrix algebra. Let α : G y Mn be an action.
Its easy to see that Aut(Mn) is isomorphic to P Un as groups, where P Un =
Un/T is the complex projective unitary group. One can check that they are
actually isomorphic as topological groups, where P Un has the standard quotient
topology. Note that for n > 1, there is no continuous section from P Un to Un,
so a continuous map G → P Un does not necessarily lift to a continuous map
G → Un. However, by a result of Dixmier, there is always a Borel section from
P Un to Un, so any continuous map G → P Un admits a lift G → Un which is
Borel.
The above argument shows that there is a Borel family of unitaries Ug such
that αg = Ad Ug for each g ∈ G. Now use the identities αgαh = αgh and
(αgαh)αk = αg(αhαk), we can show that there is a Borel function λ : G×G → T
such that:
(1) UgUh = λg,hUgh,
(2) λg,1 = λ1,g = 1,
λg,hλgh,k = λg,hkλh,k.
Definition 2.1. A Borel function λ : G×G → T satisfying the second condition
above is called a 2-cocycle. A Borel map π : G → Un satisfying UgUh = λg,hUgh
for some 2-cocycle λ is called a λ-representation, or a cocycle representation
(also called projective representation in some literature) if we do not want to
stress λ.
If λ is trivial (i.e. λg,h = 1 for all g, h), we will simply say π is a
representation.
Remark 2.2. What we have just defined should be "unitary" λ-representation.
Since we will not talk about non-unitary ones, we shall make this convention for
2
the rest of the paper. Most concepts for representations carry over to cocycle
representations naturally, like direct sum, tensor product, unitary equivalence
etc, we shall use these terminologies freely. We refer to [6] for basic facts of
cocycle representations.
Now suppose {Vg} is another Borel family of unitaries inducing α, let λ be the
corresponding 2-cocycle. Then there exist a Borel function µ : G → T such that
Ug = µgVg
λg,h =
µgµh
µgh
λg,h
A Borel function s : G × G → T of the form µgµh/µgh for some µ : G → T is
called a 2-coboundary. In this case, we shall write
s = d µ.
Two cocycles differ by a coboundary is said to be cohomologous. Let Z 2(G, T) be
the set of 2-cocycles and B2(G, T) be the set of 2-coboundaries. Its easy to see
that B2(G, T) becomes a subgroup of Z 2(G, T) under pointwise multiplication.
If λ ∈ Z 2(G, T), we shall use either ¯λ or λ−1 to denote the inverse of λ, where
¯λ is the complex conjugate of λ.
Definition 2.3. ([9]) The second (measurable) cohomology group of G, denoted
by H 2
m(G, T), is the quotient group Z 2(G, T)/B2(G, T).
By the above analysis, for any action α : G y Mn , there is a well-defined
element [λα] ∈ H 2
m(G, T). We can see that [λα] = [λβ] if α and β are conjugate,
and α is inner (induced by a representation) if and only if [λα] is the identity
element. On the other hand, any λ-representation induces an action. The
induced action is continuous because a Borel homomorphism between two Polish
groups is automatically continuous, by a classical theorem of Banach.
In general, H 2
m(G, T) could be non-trivial, even for finite abelian groups G. It is
known that for any λ ∈ Z 2(G, T), there is at least one λ-representation. So if we
pick some λ-representation such that [λ] is not the identity element of H 2
m(G, T),
then it will induce an action which is pointwisely inner but not inner. Using
cocycle representations, we can actually produce examples of two actions with
the same equivariant K-theory, but are not conjugate.
Example 2.4. Let G = Z/3Z ⊕ Z/3Z. Denote the generators of G by e1 and
e2. Let ω be a primitive 3rd root of unity. Set
A1 =
0
1
0 ω
0
0
0
0 ω2
, A2 =
0
1
0 ω2
0
0
0
0
ω
, B =
0
0
1
1 0
0 1
0 0
.
For i = 1, 2, let πi : G → U3 be defined by
πi(er
1es
2) = Ar
i Bs,
for 0 ≤ r, s ≤ 2.
Straightforward computation shows that πi is a cocycle representation corre-
sponding some 2-cocycle λi . Denote the trivial cocycle by 1, then one can check
3
m(G, T) (Actually H 2
that [1], [λ1] and [λ2] are all different in H 2
m(G, T) = Z/3Z).
For i = 1, 2, let αi : G y M3 be the action induced by πi. Then α1 and α2 are
not conjugate. But their equivariant K-theories K G
0 (A, αi) are isomorphic. To
see this, we first check that both crossed products A ⋊αi G are isomorphic to
M9. This could be proved either by elementary methods or referring to Example
3.11 in the next section. By Julg's theorem, K G
0 (A, α2) as groups.
For abelian group actions, one can identify the representation ring R(G) with
Z( G), and the module structure on the equivariant K0-groups is induced by
the dual action G y A ⋊αi G. Since both crossed products are isomorphic to
a matrix algebra, the dual group G acts trivially on K0(A ⋊αi G). Therefore
K G
0 (A, α2) as modules over the representation ring.
0 (A, α1) ∼= K G
0 (A, α1) ∼= K G
A counter-example for actions on infinite dimensional C*-algebras could also
be obtained, but the construction is more sophisticated.
Example 2.5. We still let G = Z/3Z ⊕ Z/3Z and let πi, λi, αi be constructed
as in Example 2.4. Let A be the the UHF-algebra M3∞ . As a consequence of
[14], one can find two product-type actions (infinite tensors of inner actions on
M3 ) β1, β2 : G y A such that β1 has the tracial Rokhlin property but not the
Rokhlin property, while β2 has the Rokhlin property. Let γi = αi ⊗ βi be the
tensor product action, for i = 1, 2. Then K G
0 (A, γ2) but γ1 and γ2
are not conjugate.
0 (A, γ1) ∼= K G
Viewing γi as an inductive limit action on M3 ⊗A ∼= A in the natural way, we
can see that at each finite stage, γi is induced by some λi-representation. Hence
the crossed products are both isomorphic to some full matrix algebra. Arguing
as in Example 2.4, and using the fact that equivariant K-theory is sequentially
continuous:
K G
0 (A, α) = lim−→ K G
0 (An, αn)
One can show that K G
is an immediate consequence of the following fact:
0 (A, γ1) ∼= K G
0 (A, γ2). That γ1 and γ2 are not conjugate
Proposition 2.6. Let α : G y A be an action of a finite group G on a simple
C*-algebra A. Let β : G y Mn be an arbitrary action. Then α has the tracial
Rokhlin property if and only if α ⊗ β has the tracial Rokhlin property. α has
the Rokhlin property if and only if α ⊗ β has the Rokhlin property.
Proof. We shall only deal with the tracial Rokhlin property part. The proof of
the Rokhlin property part is similar. One direction is easy: if α has the tracial
Rokhlin property then so does α ⊗ β. The converse is true if β is an inner
action, by Lemma 3.9 of [12]. For arbitrary β, suppose it is induced by some
λ-representation π. Let ¯β be the action induced by ¯π, the contragredient of π.
We see that ¯π is a ¯λ-representation, therefore π ⊗ ¯π is a genuine representation.
Now assume α ⊗ β has the tracial Rokhlin property, then so does α ⊗ β ⊗ ¯β. But
now β ⊗ ¯β is an inner action, by Lemma 3.9 of [12], α has the tracial Rokhlin
property.
4
3 Twisted equivariant K-theory
The example in previous section suggest us to look for additional invariant in
order to obtain classification result. Before we describe the new invariant, we
need some basic facts about cocycle representations.
Definition 3.1. Let G be a compact group, let λ be a 2-cocycle. We let V λ(G)
be the set of equivalent classes of λ-representations of G. It becomes an abelian
semigroup under direct sum of representations. We use Rλ(G) to denote the
Grothendieck completion of V λ(G), i.e. Rλ(G) is the group of formal differences
of equivalent classes of λ-representations.
We shall call Rλ(G) the λ-representation group. We also use the term cocycle
representation group when the 2-cocycle λ is not specified. Rλ(G) becomes an
ordered abelian group with V λ(G) being the positive cone. When λ is trivial,
i.e. λ(g, h) = 1 for any g, h ∈ G, we simply write Rλ(G) by R(G), which is just
the representation ring of G.
Unlike ordinary representations, when λ is nontrivial, the group Rλ(G) does
not form a ring under tensor product. This is because if ρi is a λi -representation,
for i = 1, 2, then ρ1 ⊗ ρ2 is a λ1λ2 -representation. Instead, tensor product gives
us a bunch of pairings
Rλ1 (G) × Rλ2(G) → Rλ1λ2 (G),
which are commutative and associative in the sense that
(1) [ρ1] ⊗ [ρ2] = [ρ2] ⊗ [ρ1].
(2) ([ρ1] ⊗ [ρ2]) ⊗ [ρ3] = [ρ1] ⊗ ([ρ2] ⊗ [ρ3]).
Proposition 3.2. If λ1 and λ2 are cohomologous, then Rλ1 (G) and Rλ2(G)
are isomorphic as ordered abelian group.
Proof. Since λ1 and λ2 are cohomologous, there is some Borel function µ : G →
T such that λ2 = dµ λ1. Let ρ : G → U (H) be a λ1 -representation. Let
L(ρ) : G → U (H) be defined by L(ρ)(g) = µ(g)ρ(g). Then L(ρ) is a λ2-
representation.
Its an easy exercise to check that L defines an isomorphism
of V λ1 (G) and V λ2 (G), with inverse defined by R(ρ)(g) = µ−1(g)ρ(g). Hence
it defines an isomorphism of the ordered abelian groups.
Note however that there is no natural choice of µ in the above proof, and
different choices may lead to different isomorphisms. Now we are ready to de-
fine our invariant, which is certain twisted version of equivariant K-theory. The
main idea is to generalize the construction of equivariant K-theory by using
cocycle representations of the group instead of ordinary representations. In the
following we shall fix an action α : G y A of a compact group on a C*-algebra A.
5
Definition 3.3. (Definition 2.4.1, [11]) Let λ be a 2-cocycle. Let (G, H1, ρ1)
and (G, H2, ρ2) be two λ-representations. Let L(H1, H2) be the set of all linear
maps from H1 to H2 . If dim(H1) = m and dim(H2) = n, then we can identify
L(H1, H2) with Mm,n(C). One checks that
g · t = ρ2(g) · t · ρ1(g)−1
defines an action of G on L(H1, H2).
We identify L(H1, H2)⊗ A with a set of A-module homomorphisms from H1 ⊗ A
to H2 ⊗ A, by formula
(t ⊗ a)(ξ ⊗ b) = tξ ⊗ ab,
t ⊗ a ∈ L(H1, H2) ⊗ A, ξ ⊗ b ∈ H1 ⊗ A.
When A is unital, we can also identify L(H1, H2) ⊗ A with Mm,n(A).
Give L(H1, H2) ⊗ A the diagonal action, so that g · (t ⊗ a) = (gt ⊗ αg(a)),
for t ∈ L(H1, H2) and a ∈ A. Define the product of s ∈ L(H2, H1) ⊗ A and
t ∈ L(H3, H2) ⊗ A to be their product as A-module homomorphisms. Then
one has (s ⊗ a)(t ⊗ b) = (st ⊗ ab). We also define an adjoint operation from
L(H1, H2)⊗A to L(H2, H1)⊗A by (t⊗a)∗ = (t∗⊗a∗). We see that multiplication
and adjoint are preserved by the G-action.
We write L(H) for L(H, H).
Definition 3.4. Let (G, A, α) be a G-C*-algebra. Let λ be a 2-cocycle. We de-
fine P λ(G, A, α) be the the set of pairs (p, π), where (G, H, π) is a λ-representation
and p is an invariant projection of L(H) ⊗ A (The action on L(H) is induced
by π). If (p1, π1) and (p2, π2) are two such pairs, we define an addition
(p1, π1) ⊕ (p2, π2) : = (diag{p1, p2}, π1 ⊕ π2)
Two pairs (p1, π1) and (p2, π2) are called Murray-von Neumann equivalent if
there is an G-invariant element u ∈ L(H1, H2) ⊗ A such that u∗u = p and
uu∗ = q.
We let V λ(G, A, α) be the set of equivalence classes in P λ(G, A, α). The addition
in P λ(G, A, α) respect the equivalence relation, therefore defines an addition on
V λ(G, A, α). Just like ordinary K-theory, one can show that V λ(G, A, α) is an
abelian semigroup with identity.
If φ : (G, A, α) → (G, B, β) is an equivariant homomorphism of G-C*-algebras,
we define
φ∗ : V λ(G, A, α) → V λ(G, B, β) by φ∗([(p, π)]) = [((idL(H) ⊗ φ)(p), π)].
Lemma 3.5. (Lemma 2.4.3, [11]) V λ(∗) is a covariant functor from G-C*-
algebras to abelian semigroups with identity.
Definition 3.6. If A is unital, we define K λ
0 (G, A, α) to be the Grothendieck
If A is non-unital, Let A be the unitalization of
completion of V λ(G, A, α).
A and α be the natural extension of α. Let p : (G, A, α) → (G, A, α) be the
canonical projection. Then K λ
0 (G, A, α) is defined to be ker p∗. If G and A are
clear from the context, we shall simply write K λ
0 (α)+
for the image of V λ(G, A, α) in K λ
0 (α). We shall write K λ
0 (α) under the canonical map.
6
When A is stably finite, it becomes an ordered abelian group with K λ
0 (α)+
0 (∗) is a covariant functor from
being the positive cone. One checks that K λ
stably finite G-C*-algebras to ordered abelian groups.
Now for any pairs of 2-cocycles λ1 and λ2 , there is a natural map:
Rλ1 (G) × K λ2
0 (α) → K λ1λ2
0
(α),
induced by
[(G, H1, π1)] × [(p, π2)] → [(p ⊗ 1, π2 ⊗ π1)],
Which becomes an element in
Hom(Rλ1 (G), Hom(K λ2
0 (α), K λ1λ2
0
(α))).
We shall call this map the partial action of Rλ1 (G) on K λ2
0 (α). The reason for
using this terminology is that, if we take the direct sum of K λ
0 (α) for different
2-cocycle λ, then Rλ1(G) indeed gives an action. As a special case, when λ1 is
trivial, the partial action of Rλ1 (G) = R(G) gives K λ2
0 (α) a module structure
over the representation ring.
One can check that this map is compatible with the pairing between cocycle
representation groups in the sense that the following diagram commutes:
Rλ1(G) × Rλ2(G) × K λ3
0 (α)
Rλ1λ2 (G) × K λ3
0 (α)
Rλ1(G) × K λ2λ3
0
(α)
K λ1λ2λ3
0
(α)
When A is unital, there is one special element lives in K0(α) (λ is trivial)
which is represented by (1A, πt), where πt is the trivial representation. We shall
denote this element by [1α].
Let φ : (G, A, α) → (G, B, β) be an equivariant homomorphism. Then we can
check that the induced map
φ∗ : K λ
0 (α) → K λ
0 (β)
is compatible with the partial actions of the cocycle representation groups in
the sense that the following diagram commutes:
K λ2
0 (α)
Rλ1 (G)
K λ1λ2
0
(α)
φ∗
φ∗
K λ2
0 (β)
Rλ1 (G)
K λ1λ2
0
(β)
Just like equivariant K-theory, there is also a module picture of the twisted
equivariant K-theory. One can also talk about K1, but we will not need it in
7
this paper.
The following theorem could be proved by mimicking the proof of the original
Julg's theorem. We refer the reader to [10] for the definition of twisted crossed
product and its basic properties.
Theorem 3.7. (Julg's theorem) Let (G, A, α) be an G-C*-algebra, where G is
compact. Then:
K λ
0 (α) ∼= K0(A ⋊
α,¯λ G),
where ¯λ is the complex conjugate of λ (viewed as a complex function), and
A ⋊
α,¯λ G is the twisted crossed product.
The K λ-groups are sequentially continuous and commute with direct sums:
Proposition 3.8. If (G, A, α) = lim−→ (G, An, αn), then for any 2-cocycle λ, we
have
0 (α) ∼= lim−→ K λ
K λ
0 (αn).
Moreover, the partial actions of Rλ(G) commutes with inductive limit.
Proposition 3.9. Let (G, A, α) and (G, B, β) be two G-C*-algebras, then for
any 2-cocycle λ, we have
K λ
0 (G, A ⊕ B, α ⊕ β) ∼= K λ
0 (α) ⊕ K λ
0 (β),
which are compatible with the partial actions of the cocycle representation
groups.
We now give some examples of twisted equivariant K-theory.
Proposition 3.10. Let α : G y Mn be an action induced by a λ1-representation
ρ. Then there is an ordered group isomorphism between K λ2
0 (α) and Rλ1λ2 (G),
for each 2-cocycle λ2. Furthermore, under these isomorphisms, the partial ac-
tion of Rλ3 (G) on K λ2
0 (α) becomes the pairing between Rλ3(G) and Rλ1λ2 (G)
(induced by tensor product). The special element [1α] becomes [ρ].
Proof. The map between K λ2
0 (α) and Rλ1λ2(G) is induced by:
[(p, π)] → [(π ⊗ ρ) p].
Its routine to check that this is a well-defined group homomorphism which
preserves the positive cone.
Its also easy to see that this map is injective.
To show that this map is surjective, we need only to show that any λ1λ2-
representation π′ is in the image. Let ¯ρ be the contragredient representation of
ρ, then ¯ρ ⊗ ρ is an ordinary representation. It contains the trivial representation
πt as a subrepresentation, because
hχ ¯ρ⊗ρ, χπti = ZG
χ ¯ρ(g)χρ(g) dg = hχρ, χρi > 0.
8
Let e be the 1-dimensional projection correspond to a trivial subrepresenta-
tion of ¯ρ ⊗ ρ. Now consider [(1 ⊗ e, π′ ⊗ ¯ρ)], which is evidently an element of
K λ2
Finally, it following from the definition of the isomorphism that [1α] corresponds
to [ρ].
0 (α) being mapped to [π′].
Example 3.11. Let G = Z/3Z ⊕ Z/3Z. Let πi, λi and αi be constructed as in
Example 2.4. Then Rλi (G) ∼= Z, generated by [πi].
Proof. See Lemma 2.4 of [5].
4 Equivariant Classification
Definition 4.1. Let α : G y A be an action, where G is compact. The scale
of K0(α) is
D(α) = {[(p, πt)] ∈ K0(α) p ∈ A, πt is the trivial representation}.
We use ξα to denote [1α] if A is unital, and the scale D(α) if A is non-unital.
Following the convention in the Elliott's classification program, we will write
the invariant of (G, A, α) by Ell(G, A, α), or Ell(α) for short if G and A is clear
from the context.
It consists of the collection of the ordered abelian groups
K λ
Let α : G y A and β : G y B be two actions. A homomorphism
0 (α) together with ξα, and all the partial actions by Rλ(G).
is understood to be a collection of ordered group homomorphisms
T : Ell(α) → Ell(β)
T λ : (K λ
0 (α), K λ
0 (α)+) → (K λ
0 (β), K λ
0 (β)+),
which are compatible with the partial actions of the cocycle representations
groups. That is, the following diagram is commutative for any 2-cocycles λ1
and λ2:
K λ2
0 (α)
T λ2
K λ2
0 (β)
Rλ1 (G)
Rλ1 (G)
K λ1λ2
0
(α)
T λ1
K λ1λ2
0
(β)
We say T is unital if T sends [1α] to [1β]. We say T is contractive if it preserves
the scales. We say T is an isomorphism if each T λ is an isomorphism and it
sends ξα to ξβ.
Now we are ready to state the main theorem of this paper:
Theorem 4.2. Let G be a compact group. Let A, B be two AF algebras.
Let α : G y A and β : G y B be two inductive limit actions which are locally
spectrally trivial. Suppose A, B are either both unital or both non-unital. Then
α and β are conjugate if and only if
Ell(α) ∼= Ell(β).
9
Just like classification of AF algebras, the proof is divided into two steps.
The first step is to establish an existence theorem and a uniqueness theorem
for actions on finite dimensional C*-algebras. The second step is to use the
intertwining argument to get the desired isomorphisms. Since we are assuming
that the actions on finite dimensional C*-algebras are spectrally trivial, they are
direct sum of actions on each simple summand. So let's look at what happens
for actions on matrix algebras first.
Proposition 4.3. Let α : G y Mn and β : G y Mk be two actions. Suppose
that they are induced by some cocycle representations πα and πβ, respectively.
Then there is a nontrivial equivariant homomorphism between the two G-C*-
algebras if and only if there is some cocycle representation πγ and some nonzero
projection p ∈ Mk, invariant under πβ, such that πβ p is equivalent to πα ⊗ πγ
.
Proof. One direction is easy: if such cocycle representation exist, then the map
a → U p(a ⊗ 1)pU ∗ is equivariant, where U is a unitary implementing the equiv-
alence between πβ p and πα ⊗ πγ. For the other direction, let T be a nontrivial
equivariant homomorphism. Set p = T (1), which is a nonzero projection in Mk.
It's easy to check that p is invariant under β. Restricting the action to β pMkp,
we may without loss of generality assume that T is unital.
Let A = T (Mn)′ , the commutant of T (Mn). Using the fact that T is equiv-
ariant, we can show that A is a subalgebra of Mk invariant under β. Consider
the map S : A ⊗ Mn → Mk , induced by S(a ⊗ b) = aT (b). This map is an
equivariant homomorphism if we define the action on A ⊗ Mn by (β A) ⊗ α.
Now it's elementary to check that A is actually a matrix algebra, say isomorphic
to Mr, such that Mr ⊗ Mn ∼= Mk . (In particular, S is an isomorphism). Let
πγ be a cocycle representation on Mr inducing β A under the isomorphism of
A and Mr. Let γ be the action on Mr induced by πγ . Then up to conjugate,
β and α ⊗ γ are the same. Hence
πβ = µgU (πα ⊗ πγ)U ∗,
for some U ∈ Mk and µg ∈ T.
Replacing πγ by µπγ, we get the desired cocycle representation.
Proposition 4.4. Let α : G y Mn and β : G y Mk be two actions. Then for
any unital homomorphism Γ : Ell(α) → Ell(β), there is an unital equivariant
homomorphism T : Mn → Mk such that T ∗ = Γ. Furthermore, if T1 and T2 are
two unital equivariant homomorphism such that T ∗
2 , then T1 and T2 are
conjugate by some unitary.
1 = T ∗
Proof. Suppose α is induced by a λα-representation πα, β is induced by a λβ
-representation πβ. By Proposition 3.10, there is an ordered group isomorphism
0 (α) and Rλλα (G), for each 2-cocycle λ. Furthermore, under these
between K λ
isomorphisms, the partial action of Rλ
0 (α) becomes the pairing be-
tween Rλ(G) and Rλλα(G) (induced by tensor product). The special element
0 (G) on K λ
10
[1α] becomes [πα]. Similar conclusion holds for β. We now claim that, un-
der these isomorphisms, there is some λαλβ -representation πγ, such that the
homomorphisms Γ becomes
Γλ : Rλλα(G) → Rλλβ (G), Γλ([π]) = [π ⊗ πγ]
To see this, consider the map
Γλα : Rλαλα (G) = R(G) → Rλαλβ (G).
Let πt be the trivial representation. Then [πt] ∈ R(G). Choose πγ such that
[πγ] = Γλα[πt]. Now let [π] ∈ Rλλα (G) be an arbitrary element. Since the
homomorphisms Γλ is compatible with the partial actions, we have
Γλ([π]) = [π]Γλ([πt]) = [π][πγ ] = [π ⊗ πγ],
which completes the proof of our claim.
Since [πα ⊗ πγ] = [πβ], we see that πα ⊗ πγ and πβ are equivalent. Now let
γ : G y Mr be the action induced by πγ . We have Mn ⊗ Mr ∼= Mk. The map
Mn → Mn ⊗ Mr,
a → a ⊗ 1
induces an equivariant map T : Mn → Mk. From the construction of T we see
that T ∗ is induced by tensor with [πγ], hence T ∗ = Γ.
Now suppose T1 and T2 are two equivariant maps such that T ∗
2 . From
the above proof, we may assume that Ti is induced by tensoring with some
cocycle representations πγi , for i = 1, 2. Since T ∗
2 ([1α]), we have
that πα ⊗ πγ1 and πα ⊗ πγ2 are equivalent. Therefore T1 and T2 are conjugate
by some unitary.
1 ([1α]) = T ∗
1 = T ∗
Remark 4.5. From the proof of the above proposition, we can actually see that
for any homomorphism Γ : Ell(G, Mn, α) → Ell(G, B, β), the homomorphism
is entirely determined by the image of [πt], where [πt] is the equivalent class
of the trivial representation, viewed as an element of K λα(G, Mn, α) under the
isomorphism K λα(G, Mn, α) ∼= R(G).
One can also get a non-unital version from the above proof.
Corollary 4.6. Let α : G y Mn and β : G y Mk be two actions. Let
Γ : Ell(α) → Ell(β)
be a contractive homomorphism. Then there is an equivariant homomorphism
T : Mn → Mk such that T ∗ = Γ. Furthermore, if T1 and T2 are two equivariant
homomorphisms such that T ∗
2 , then T1 and T2 are conjugate by some
unitary.
1 = T ∗
We can now extend Proposition 4.4 to actions on finite dimensional C*-
algebras which are spectrally trivial:
11
Proposition 4.7. Let A, B be two finite dimensional C*-algebras, let α : G y
A and β : G y B be two spectrally trivial actions. Then for any contractive
homomorphism Γ : Ell(α) → Ell(β), there is an equivariant homomorphism
T : A → B such that T ∗ = Γ. We can choose T to be unital if Γ is unital.
Furthermore, if T1 and T2 are two equivariant homomorphism such that T ∗
1 = T ∗
2
, then T1 and T2 are conjugate by some unitary.
Proof. A finite dimensional C*-algebra is a direct sum of matrix algebras, and
a spectrally trivial action is a direct sum of actions on each matrix algebra. Let
A = A1 ⊕ A2 ⊕ · · · Al , let B = B1 ⊕ B2 ⊕ · · · Bs , where Ai = Mni and Bj = Mkj
are matrix algebras. Let αi be the induced action on Ai and βj be the induced
action on Bj . In view of Proposition 3.9, we may write
Ell(α) ∼= ⊕l
i=1Ell(G, Ai, αi), Ell(β) ∼= ⊕s
j=1Ell(G, Bj, βj).
Now let Γi,j be the partial maps from Ell(G, Ai, αi) to Ell(G, Bj, βj) induced by
Γ. Then by Corollary 4.6, there exists equivariant homomorphisms φi,j : Ai →
Bj such that φ∗
i,j = Γi,j. Since T is contractive, we have
⊕l
i=1Γi,j([1αi ]) = [(p, πt)],
where πt is the trivial representation and p ∈ Bj is invariant under βj. Under
the identification of K λ(G, Bj , βj) and Rλλβj (G), we can see that Γi,j([1αi ])
are equivalent to subrepresentations of πβj p such that the direct sum is equal
to πβj p. Conjugating suitable unitaries if necessary, we can arrange that the
homomorphisms φi,j have orthogonal range. Then the partial maps φi,j will give
a homomorphism φ : A → B such that φ∗ = Γ. It is unital if Γ is. Arguing as in
Proposition 4.4, one can see that if T1 and T2 are two equivariant homomorphism
such that T ∗
2 , then T1 and T2 are conjugate by some unitary.
1 = T ∗
We can now use the intertwining argument to prove our main result Theo-
rem.
Proof. (Proof of Theorem 4.1) Let G be a compact group. Suppose that
(G, A, α) = lim
−→
(G, An, αn)
(G, B, β) = lim
−→
(G, Bn, βn),
where An, Bn are finite dimensional C*-algebras and the actions αn and βn are
spectrally trivial. Let Γ : Ell(α) → Ell(β) be an isomorphism. By Proposition
3.8, we have
Ell(α) ∼= lim−→ Ell(αn), Ell(β) ∼= lim−→ Ell(βn).
Set n0 = 1. Consider the homomorphism Γn0 : Ell(αn0) → Ell(β) which is
the composition of the connecting map and Γ. We want to show that this
homomorphism pulls back to finite stage. Since (G, An0 , αn0 ) is a finite direct
sum of actions on matrix algebras, without loss of generality we may assume that
An0 is a matrix algebra. Suppose αn0 is induced by some λn0 -representation
12
πn0 . From Remark 4.5, we see that Γn0 is determined by the image of [πt]. Here
[πt] is the equivalent class of the trivial representation, viewed as an element of
K λn0 (α) under the isomorphism
K λn0 (α) ∼= Rλn0 λn0 (G) = R(G).
The image of [πt] can certainly be pulled back to finite stage: there is some m0
and certain element c ∈ K λn
0 (β)
is Γλn0 ([πt]).
Now for any 2-cocycle λ, we can define a homomorphism
0 (G, Bm0 , βm0) such that the image of c in K λ
Θλ : K λ
0 (G, An0 , αn0 ) → K λ
0 (G, Bm0 , βm0),
θ([π]) = [π]c.
It's then easy to verify that {Θλ} defines homomorphisms which are compatible
with the partial actions by Rλ(G).
If A, B are unital, enlarge m0 if necessary, we may assume that Θ([1αn0 ]) =
If A, B are both non-unital, since Γn0[1αn0 ] ∈ D(β), enlarge m0 if
[1βm0 ].
necessary, we may assume that Θ([1αn0 ]) = [(p, πt)], for some p ∈ Bm0 (πt
is the trvial representation). In any case, we have the following commutative
diagram.
Ell(αn0)
Γn0
Θ
Ell(βm0)
Ell(β)
While the homomorphisms in the above diagram are all contractive. Similarly,
the map Γ−1
m0 : Ell(βm0) → Ell(α) can be pulled back to some Ell(αn1). By
Remark 4.5, if γ : G y Mn is an action on matrix algebra and θ : G y C is an
arbitrary action, then any homomorphism (not necessarily preserve the special
element)
Ell(G, Mn, γ) → Ell(G, C, θ)
is determined by the image of the trivial representation. Enlarge n1 if necessary,
we can match the images of the trivial representations in the following diagram,
hence making it commutative:
Ell(αn0)
Ell(αn1)
Ell(αn1)
Γ−1
m0
Ell(βm0)
We can further assume that all homomorphisms are contractive by enlarging n1
if necessary.
13
Continuing this way, we get the following commutative diagram:
Ell(αn0)
Ell(αn1) · · ·
Ell(α)
Ell(βm0)
Ell(αm1) · · ·
Ell(β)
By Proposition 4.7 we can lift the commutative diagram to the diagram
(G, An0 , αn0 )
(G, An1 , αn1) · · ·
(G, A, α)
(G, Bm0 , βm0)
(G, Bm1, αm1 ) · · ·
(G, B, β)
By Proposition 4.7 again, the above diagram commutes up to conjugation by
unitaries. Hence we can make it truly commutative by conjugating suitable
unitaries. Let T : (G, A, α) → (G, B, β) be the induced homomorphism. Then
it's easy to see that T is an equivariant isomorphism such that T ∗ = Γ.
5 Simplification of the invariant
Let α : G y A be an action, where G is compact. When we define the invariant
Ell(α), we made use of all possible 2-cocycles λ. There is some redundancy that
we can get rid of. First, we note the following:
Proposition 5.1. Let λ and λ1 be two cohomologous 2-cocycle. Then there is
an isomorphism between K λ(α) and K λ1 (α).
Proof. Let µ : G → T be a Borel map such that λ1 = (d µ)λ. Let [(p, π)]
be an element in V λ(α). Let π1 = µπ. It's easy to check that π1 defines a λ1-
representation. Since π1 and π induce the same action, we see that p is invariant
under α ⊗ Adπ1. Hence [(p, π1)] defines an element in V λ1 (α). Let us denote
this map [(p, π)] → [(p, π1)] by fµ. One can check from the definition that
fµ is a semigroup homomorphism, with the obvious inverse map fµ−1 . Hence
fµ defines an semigroup isomorphism, which induces an isomorphism between
K λ(α) and K λ1(α).
Now we can show that, in order to have a homomorphism for the invariants,
it's enough to use a complete set of representatives of 2-cocycles containing the
trivial 2-cocycle.
14
Proposition 5.2. Let (G, A, α) and (G, B, β) be two G-C*-algebras, where G
is compact. Let {λi i ∈ I} be a complete set of representatives of H 2
m(G, T).
Suppose we have a collection of homomorphisms
Γλi : (K λi
0 (α), K λi
0 (α)+) → (K λi
0 (β), K λi
0 (β)+),
which are compatible with the partial actions of all cocycle representation groups
of the form Rλ−1
i λj (G):
K λi
0 (α)
Γλi
K λi
0 (β)
−1
i
Rλ
λj (G)
K λj
0 (α)
−1
i
Rλ
Γλj
λj (G)
K λj
0 (β)
Then there is a homomorphism T : Ell(α) → Ell(β) such that T λi = Γλi , for
each i ∈ I.
Proof. Any two 2-cocycle λ is cohomologous to some λi. Pick any Borel function
µ such that λ = (d µ)λi. Let the map fµ be defined as in the proof of 5.2. We
define a map:
T λ : (K λ
0 (α), K λ
0 (α)+) → (K λ
0 (β), K λ
0 (β)+)
by T λ = fµ ◦ Γλi ◦ fµ−1 . We want to show that these homomorphisms are
compatible with the partial actions. So let ωi, ωj be two 2-cocycles such that
ωi = (d µi)λi and ωj = (d µj )λj. Let T ωi, T ωj be defined as above. Let π be
a ω−1
i λj-representation. We
then have the following diagram:
i ωj-representation. Let π = µiµ−1
j π, which is a λ−1
K λi
0 (α)
Γλi
[π]
T ωi
K λj
0 (α)
K ωi
0 (β)
[π]
Γλj
f
µ
−1
i
K ωi
0 (α)
[π]
f
µ
−1
j
K ωj
0 (α)
T ωi
K ωj
0 (β)
K λi
0 (β)
fµi
[π]
K λj
0 (β)
fµj
The top and bottom squares are commutative by the definition of T λ, the
back square is commutative by assumption. We now check that the right square
is commutative. Let [p, π0] ∈ K λi
0 (β). Then we have:
[π]fµi ([(p, π0)]) = [π][(p, µiπ0)] = [p ⊗ 1, (µiπ0) ⊗ π].
15
Therefore
fµj ([π][(p, π0)]) = fµj ([(p ⊗ 1, π0 ⊗ π)])
= [(p ⊗ 1, µj(π0 ⊗ µiµ−1
= [(p ⊗ 1, (µiπ0) ⊗ π)] = [π]fµi ([(p, π0)]).
j π))]
Hence the right square is commutative. Similar computation shows that the
left square is commutative. Hence the front square is also commutative, which
finishes the the proof.
References
[1] O. Bratteli, George A Elliott, D E Evans, and A. Kishimoto, On the clas-
sification of inductive limits of inner actions of a compact group, Current
topics in operator algebras (1991), 13 -- 24.
[2] Andrew J Dean, Inductive limits of inner actions on approximate interval
algebras generated by elements with finite spectrum, J. Ramanujan Math.
Soc. 24 (2009), no. 4, 323.
[3] George A. Elliott and Hongbing Su, K-theoretic classification for inductive
limit Z2 actions on AF algebras, Journal canadien de math´ematiques 48
(1996), no. 5, 946 -- 958.
[4] David Handelman and Wulf Rossmann, Actions of compact groups on AF
C ∗-algebras, Illinois J. Math. 29 (1985), no. 1, 51 -- 95.
[5] Russell J. Higgs, Projective representations of abelian groups, J. Algebra
242 (2001), 769 -- 781.
[6] Gregory Karpilovsky, Projective representations of finite groups, vol. 94,
Marcel Dekker Inc, 1985.
[7] A. Kishimoto, Actions of finite groups on certain inductive limit C*-
algebras, Internat. J. Math. 01 (1990), no. 03, 267 -- 292.
[8] Zhiqiang Li and Wenda Zhang, A K-theoretic classification for certain Z/pZ
actions on AF algebras, arXiv preprint arXiv:1304.0813 (2013).
[9] Calvin C Moore, Group extensions and cohomology for locally compact
groups. III, Trans. Amer. Math. Soc. 221 (1976), no. 1, 1 -- 33.
[10] Judith A Packer and Iain Raeburn, Twisted crossed products of C*-algebras,
Math. Proc. Cambridge, vol. 106, Cambridge Univ Press, 1989, pp. 293 --
311.
[11] N. C. Phillips, Equivariant K-theory and freeness of group actions on C*-
algebras, Springer-Verlag, 1987.
16
[12] N Christopher Phillips, The tracial Rokhlin property for actions of finite
groups on C*-algebras, Amer. J. Math. 133 (2011), no. 3, 581 -- 636.
[13] Hongbing Su, K-theoretic classification for certain inductive limit Z2 ac-
tions on real rank zero C*-algebras, Trans. Amer. Math. Soc. 348 (1996),
no. 10, 4199 -- 4230.
[14] Qingyun Wang, Characterization of product-type actions with the Rokhlin
properties, Indiana Univ. Math. J. 64 (2015), 295 -- 308.
Qingyun Wang, Department of Mathematics, University of Toronto
E-mail address: [email protected]
17
|
1212.4587 | 1 | 1212 | 2012-12-19T07:24:58 | On the fusion algebras of bimodules arising from Goodman-de la Harpe-Jones subfactors | [
"math.OA"
] | By using Ocneanu's result on the classification of all irreducible connections on the Dynkin diagrams, we show that the dual principal graphs as well as the fusion rules of bimodules arising from any Goodman-de la Harpe-Jones subfactors are obtained by a purely combinatorial method. In particular we obtain the dual principal graph and the fusion rule of bimodules arising from the Goodman-de la Harpe-Jones subfactor corresponding to the Dynkin diagram $E_8$. As an application, we also show some subequivalence among $A$-$D$-$E$ paragroups. | math.OA | math |
On the fusion algebras of bimodules
arising from Goodman-de la Harpe-Jones subfactors
By Satoshi Goto ∗
Abstract
By using Ocneanu's result on the classification of all irreducible con-
nections on the Dynkin diagrams, we show that the dual principal graphs
as well as the fusion rules of bimodules arising from any Goodman-de la
Harpe-Jones subfactors are obtained by a purely combinatorial method.
In particular we obtain the dual principal graph and the fusion rule of
bimodules arising from the Goodman-de la Harpe-Jones subfactor corre-
sponding to the Dynkin diagram E8. As an application, we also show some
subequivalence among A-D-E paragroups.
1
Introduction
Since V. F. R. Jones initiated the index theory for subfactors in [15], intensive
studies on the classification of subfactors have been made by many people. The
classification of subfactors of the AFD type II1 factor with index less than 4 has
been completed by many people's contribution ([2, 13, 14, 15, 16, 21], see also
[9]) after A. Ocneanu's announcement [18].
Goodman-de la Harpe-Jones subfactors (abbreviated as GHJ subfactors) [11]
are known as a series of interesting non-trivial examples of irreducible subfactors
with indices greater than 4, though some of them have indices less than 4. The
indices of all GHJ subfactors are given in [11]. They are constructed from the
commuting squares arising from the embeddings of type A string algebras into
other string algebras of type ADE. (See [9, Chapter 11] for the construction of
GHJ subfactors from a viewpoint of string algebra embedding.) The principal
∗This manuscript is typeset by LATEX2e
2010 Mathematics Subject Classification. 46L37.
1
graphs of these subfactors are easily obtained by a simple method but the dual
principal graphs as well as their fusion rules are much more difficult to compute.
(Okamoto first computed their principal graphs in [20].)
One of the most important examples of GHJ subfactor has index 3 + √3 and
it is constructed from the embedding of the string algebra of A11 into that of E6.
In this particular case it happens that it is not very difficult to compute the dual
principal graph (see [17], [9, Section 11.6]). But it is more difficult to determine its
fusion rules. Actually D. Bisch has tried to compute the fusion rule just from the
graph but there were five possibilities and it turned out that the fusion rule cannot
be determined from the graph only [3]. Some more information is needed and Y.
Kawahigashi obtained the fusion rule as an application of paragroup actions in
[17].
In his lectures at The Fields Institute A. Ocneanu introduced a new algebra
called double triangle algebra by using the notion of essential paths and extension
of Temperley-Lieb recoupling theory of Kauffman-Lins [19]. He also announced a
solution to the problem of determining the dual principal graphs and their fusion
rules of the GHJ subfactors as one of some applications of his theory. But the
details have not been published.
After A. Ocneanu's works, F. Xu and J. Bockenhauer-D. E. Evans have re-
vealed a relation between the GHJ subfactors and conformal inclusions ([22], [5],
[6], [7]) and J. Bockenhauer, D. E. Evans and Y. Kawahigashi ([8]) obtained es-
sentially the same fusion algebras of GHJ subfactors of type D2n, E6, E8 by using
conformal field theory and the Cappelli-Itzykson-Zuber's classification of modular
invariant [10].
In this paper we give detailed computations of the dual principal graphs and
the fusion rules for any GHJ subfactors by a purely combinatorial method. For
this purpose we will make the most use of Ocneanu's result on the classification
of all irreducible connections on the Dynkin diagrams (See [19]. Our method here
is based on the observation in [12]). Especially we will make use of Figures 21'36,
which were first found by A. Ocneanu [19]. Our result does not rely on either
conformal field theory or the classification of modular invariant.
2 Correspondence between system of connections and system of bi-
modules
Let K and L be two connected finite bipartite graphs. A bi-unitary connection
on four graphs is called a K-L bi-unitary connection if it has the graph K as an
2
upper horizontal graph and the graph L as a lower horizontal graph as in Figure
1.
If we have a K-L connection, we can construct a subfactor N ⊂ M by choosing
a distinguished vertex ∗K of the upper graph K and applying string algebra
construction to the connection. (See [9, Section 11].) This construction seems to
depend on the choice of the vertex ∗K. But it is well-known that the subfactors
constructed from this connection does not depend on the choice of the vertex ∗K,
that is, they become all isomorphic because of the relative McDuff property [4].
∗K
K
w
L
· · · −→
· · · −→
N
∩
M
Figure 1:
On the one hand as a paragroup of the subfactor N ⊂ M obtained from the
connection w as above, we obtain the system of 4-kinds of bimodules, i.e. N-N,
N-M, M-N, M-M bimodules, by taking irreducible decomposition of alternating
relative tensor products of N MM and its conjugate bimodule M MN as usual. (See
[9] for details.)
On the other hand we also get the system of 4-kinds of connections, i.e. K-K,
K-L, L-K, L-L bi-unitary connections, by taking irreducible decomposition of
alternating compositions of the connection w and its conjugate L-K connection
¯w.
Now the problem is the relation between the system of bimodules and the
system of connections obtained as above. We can easily see that those two systems
become the same paragroup for N ⊂ M if the subfactor N ⊂ M has finite depth.
To see this it is enough to see the relation among a usual paragroup based
on bimodules, a system of generalized open string bimodules and a system of
bi-unitary connections. The details of these relations are found in [1]. Note
that when we consider a system of bi-unitary connections forms a paragroup, we
need the notion of intertwiners between two connections. For this purpose, we
need to fix distinguished vertices ∗K and ∗L of both even and odd part of the
graphs K and L, then we identify all the bi-unitary connections of the system
3
as the generalized open string bimodules constructed from those connections.
Then we define the intertwiners between two connections by those between the
corresponding two generalized open string bimodules. Now from the argument
in [1], the intertwiners between two connections can naturally be identified with
the intertwiners between the correponding 4 kinds of bimodules, i.e. N-N, N-M,
M-N, M-M bimodules arising from the usual paragroup. See Theorem 4 in [1]
for more details.
Hence we obtain the following theorem.
Theorem 2.1 If the subfactor N ⊂ M constructed from a K-L connection KwL
has finite depth, the system of 4-kinds of connections obtained from KwL and
the system of 4-kinds of bimodules obtained from the subfactor N ⊂ M have the
same fusion rules. Moreover these two systems defines the same paragroup for
N ⊂ M via the correspondence between connections and generalized open string
bimodulesD
Remark 2.2 As we mentioned above, the subfactor constructed from a connec-
tion KwL does not depend on the choice of the distinguished vertex ∗K. In the
same way we need to fix two vertices ∗K and ∗L in order to construct a generalized
open string bimodule from a connection. But the above theorem holds true for
arbitrary choice of two distinguished vertices ∗K and ∗L of the graphs K and L
respectively.
The above theorem provide us a purely combinatorial method to compute
fusion rules for the subfactor obtained from a connection KwL. Actually we can
compute the fusion rules of a system of connections by looking at the composition
and decomposition of their vertical graphs.
3 The (dual) principal graphs and their fusion rules of the Goodman-
de la Harpe-Jones subfactor
Let A be the Dynkin diagram An and K one of the ADE Dynkin diagrams
with the same Coxeter number. The subfactors constructed from the commut-
ing square as in Figure 2 are called the Goodman-de la Harpe-Jones subfactors
(abbreviated as GHJ subfactors). Here the construction depends only on the
graph K and the vertex ∗K = x. (See [11] for details.) We denote this subfactor
GHJ(K,∗K = x). We remark that the vertical graphs G and G′ as in Figure 2 are
4
easily obtained from the dimension of essential paths on the graph K (Figures
21'30). Here we note that the graphs G and G′ may be disconnected.
∗A
G
∗K
A
w
K
· · · −→
G′
· · · −→
N
∩
M
Figure 2:
We use the next two propositions to compute the fusion rule of the Goodman-
de la Harpe-Jones subfactors.
Proposition 3.1 ([12, Proposition 5.6]) Let A, K, G and G′ be the four graphs
connected as in Figure 3. Suppose there is a bi-unitary connection on the four
graphs. Then the connecting vertical graphs G and G′ are uniquely determined
by the initial condition, i.e., the condition of edges connected to the distinguished
vertex of the graph A (see Figure 4). Moreover such a connection is unique up to
vertical gauge choice.
G
A
K
G′
Figure 3:
∗A
x
A
w
K
Figure 4:
5
Proposition 3.2 (Frobenius reciprocity) ([12, Proposition 3.21]) Let K, L
and M be three connected finite bipartite graphs with the same Perron-Frobenius
eigenvalue. Let KαL, LβM and KγM be three irreducible bi-unitary connections
which are K-L, L-M and K-M respectively. If γ appears n times in the composite
connection αβ, then α appears n times in γ ¯β and β appears n times in ¯αγ.
3.1. The fusion rules of four kinds of connections arising from GHJ subfactors
The system of connections arising from a GHJ subfactor consists of four kinds
of connections, i.e. A-A, A-K, K-A and K-K connections. So the fusion rules
consist of the following 8 kinds of multiplication table.
(1) A-A × A-A −→ A-A
(2) A-A × A-K −→ A-K
(2)′ K-A × A-A −→ K-A
(3) A-K × K-K −→ A-K
(3)′ K-K × K-A −→ K-A
(4) K-K × K-K −→ K-K
(2)′′ A-K × K-A −→ A-A
(3)′′ K-A × A-K −→ K-K
Among these multiplication tables, (2)′ and (3)′ are obtained by taking conju-
gation of (2) and (3) respectively. The tables (2)′′ and (3)′′ are also obtained from
(2) and (3) respectively by Frobenius reciprocity. So it is enough to determine
four multiplication table (1), (2), (3) and (4).
3.1.1. The fusion rules of (1) A-A × A-A −→ A-A and (2) A-A × A-K −→
A-K and the principal graphs
We put the labels 0, 1, 2,· · · , m − 1 of vertices of the Dynkin diagram Am as
in Figure 5. We denote the unique irreducible A-A connection with the "initial
edge" connected to the vertex n in the lower graph Am by AnA (Figure 6). We also
denote the unique irreducible A-K connection with the "initial edge" connected
to the vertex x in the lower graph K by AxK (Figure 6).
Dynkin diagram Am
0
∗A
2
4
m − 3
m − 1
· · ·
1
3
m − 2
Figure 5: The label of vertices of the Dynkin diagram Am.
6
∗A
n
A
AnA
A
∗A
x
A
AxK
K
Figure 6:
Then the fusion rules of (1) A-A × A-A −→ A-A and (2) A-A × A-K −→
A-K can be obtained by composition and decomposition of the (left) vertical
edges of the two connections AnA and AxK as in Figure 7. So we have only to
count the vertical edges of the connection AxK in order to determine the fusion
tables of (1) and (2).
∗A
∗A
A
[AnA]
A
[AxK]
n
k l
∗A
A
∗A
A
∼= k
[AyK]
⊕ l
[AzK]
x y
z
K
y
K
z
K
Figure 7: The fusion rule of A-A × A-K −→ A-K
Because we need the notion of essential paths on graphs in order to describe
these fusion rules, we review the definition here for readers convenience. Please see
[19, section 32.2, page 254 -- 256] for more details and the proof of the moderated
Pascal rule.
Definition 3.3 A space of essential paths of a graph G with length n is defined
by EssPath(n)G = pn · HPath(n)G. Here pn = 1 − e1 ∨ e2 ∨ · · · ∨ en−1 is the
Wenzl projector and ek is the k-th Jones projection. We denote the space of
essential paths of a graph G with length n, with starting point x and end point y
by EssPath(n)
x,yG.
The dimensions of spaces of essential paths of length n is easily obtained by
7
using the following moderated Pascal rule.
dim EssPath(n+1)
a,x G = X
ξ∈Edge G,r(ξ)=x
dim EssPath(n)
a,s(ξ)G − dim EssPath(n−1)
a,x G
Now we continue the description of the fusion rules (1) and (2). Because
the connection AxK comes from the inclusion of the string algebras String∗A ⊂
StringxK, the number of vertical edges of this inclusion coincides with the dimen-
sion of essential paths from the vertex x to y of K with length n. (See Figures
21'30 for the dimension of essential paths.) Hence we get the fusion tables of (1)
and (2) as follows.
y∈VertK
AnA · AxK ∼= M
K ¯xA · AnA ∼= M
AyK · K ¯xA ∼= M
y∈VertK
n∈VertA
(dim EssPath(n)
x,yK) AyK
(dim EssPath(n)
x,yK) K ¯yA
(dim EssPath(n)
x,yK) AnA
Since the principal graph is obtained from the fusion rule of A-A × A-K −→
A-K, we can easily see that the principal graph of GHJ(K,∗K = x) coincides with
the connected component of the vertical edges of the connection AxK including the
distinguished vertex ∗A. This principal graph can be obtained easily by counting
the dimension of essential path. It follows from this fact that the even vertices of
the the principal graph of GHJ(K,∗K = x) coincides with (possibly a subset of)
the even vertices of the Dynkin diagram Am.
3.1.2. The fusion rules of
graphs
(3) A-K × K-K −→ A-K and the dual principal
We denote the unique irreducible A-K connection with the "initial edge"
connected to the vertex x in the lower graph K by AxK as before and an irreducible
K-K connection by KwiK (Figure 8). Here KwiK is one of the connections of all
K-K connection system (Figures 31'36). (See [12, section 5.3, pages 244 -- 252]
for details.) In this case the fusion rule of (3) A-K × K-K −→ A-K is also
obtained by composition and decomposition of the (left) vertical edges of the two
connections AxK and KwiK as in Figure 9. We can get the fusion table of (3) by
8
counting the vertical edges of the connection KwiK in the same way as subsection
3.1.1.
∗A
x
A
AxA
K
K
KwiK
K
∗A
x
k
y
A
[AxK]
K
[KwiK]
K
l
z
Figure 8:
∗A
A
∗A
A
∼= k
[AyK]
⊕ l
[AzK]
y
K
z
K
Figure 9: The fusion rule of A-K × K-K −→ A-K
This time the method of counting dimensions of essential paths does not work
in order to get the vertical edges of the connection KwiK. But we can compute
them by using Ocneanu's classification of all irreducible K-K connections and
their fusion rules ([12, section 5.3, pages 244 -- 252]).
For example, the vertical edges of all K-K connections are given in Figures
37'47 in the case of K = A3, A4, A5, A6, D4, D5, D6, E6, E7, E8. Here in the case of
E6, E7, E8, we give list of incidence matrices of vertical graphs instead of graphs
themselves because it is complicated to draw them all.
Now we get the fusion rule of A-K × K-K −→ A-K as follows.
AxK · KwiK ∼= M
KwiK · K ¯xA ∼= M
y∈VertK
y∈VertK
n(wi)x,y AyK
n(wi)x,y K ¯yA
9
K ¯xA · AyK ∼= M
wi∈K ZK
n(wi)x,y KwiK
Here KZK represents the system of all K-K connections which is isomorphic
to the fusion algebras of the center of K-K double triangle algebra ([12, Theorem
4.1, Corollary 4.5]). And n(wi)x,y means the number of vertical edges of the K-K
connection KwiK connecting the vertices x and y.
Now we can get the dual principal graph from the fusion rule of (3) A-K ×
K-K −→ A-K. It is the connected component of the fusion graph of (3) which
contains the connection AxK.
3.1.3. The fusion rules of (4) K-K × K-K −→ K-K
This is the fusion rule of the system of all K-K connections obtained by
Ocneanu (Figures 31'36, [12, section 5.3, pages 244 -- 252]).
It is isomorphic to
the fusion algebras of the center of K-K double triangle algebra (KZK,·) with
dot product (vertical product) "·". We know that this fusion algebra (KZK,·) is
generated by chiral left part and chiral right part which are isomorphic to the
fusion algebra of connections arising from corresponding ADE subfactor and that
the chiral left and right part are relatively commutative [12, Theorem 5.16]. So
we can compute the fusion rule of (KZK,·) from the above facts.
We remark that the commutativity of the chiral left and right part is proved at
the same time when we draw the diagrams of all K-K connections (Figures 31'36).
The proof is based on coset decomposition, fusion rules of chiral left (right) part
and indices of irreducible connections. We refer readers to [12, section 5.3, pages
244 -- 252] for details.
The fusion tables of (KZK,·), i.e. the system of all K-K connections for K
= E6, E7 and (a part of) E8 is given in Figures 48'50. We note that these fusion
tables is expressed in product form. For example in the table 49, we can read
(3) · 4=(1)2(3)3(5), which means the fusion rule w(3) · w4=2w(1) + 3w(3) + w(5)
holds.
3.2. The fusion rules of even vertices of the (dual) principal graphs of GHJ(K,∗K =
x)
Let N ⊂ M be the Goodman-de la Harpe-Jones subfactor GHJ(K,∗K = x).
Here we will compute the fusion rules of even vertices of the (dual) principal
graphs of GHJ(K,∗K = x), that is, the fusion rules of N-N bimodules and M-M
10
bimodules of the subfactor N ⊂ M.
The system of N-N bimodules are isomorphic to the system of A-A connec-
tions generated by AxK and this is the same as AZ even
A , i.e. the fusion algebra of
even part of AZA. So the fusion algebra of N-N bimodules are isomorphic to the
fusion algebra Aeven, i.e. the fusion algebra of even vertices of the Jones' type
A subfactor. Hence it turns out that the fusion algebra of N-N bimodules are
always commutative for any GHJ subfactors.
The system of M-M bimodules are similarly isomorphic to the system of K-K
connections generated by AxK and this is the same as (a part of) KZ even
K , i.e. the
fusion algebra of even part of KZK. So we have only to compute the fusion rule
of KZ even
K .
Here the fusion rule of KZ even
and the vertical edges of irreducible K-K con-
nections can be summarized as in the Table 1. As we mentioned above, we can
compute the fusion rule of KZ even
in detail from the fusion graph of all K-K
connections as in Figures 31'36 and 48'50.
K
K
In the following table, ε represents the index 1 D2n-D2n connection which cor-
responds to the flip of two tails of D2n. Because D2nZD2n has coset decomposition
D2n∪D2n·ε and ε2 = id as shown in Figures 32, 41 and 43, we can easily compute
the fusion rule for D2nZD2n.
Graph K fusion rule of KZ even
K
vertical edges of K-K connections
An
D2n
D2n+1
E6
E7
E8
commutative
EssPathAn (Figures 37'40)
non-commutative
EssPathD2n + ε (Figures 41 and 43)
commutative
commutative
commutative
commutative
EssPathD2n+1 (Figure 42)
Figure 44
Figure 45
Figures 46, 47
Table 1: The fusion rule of KZ even
K
and vertical edges of K-K connections
11
4 The structure of Goodman-de la Harpe-Jones subfactors
4.1. Goodman-de la Harpe-Jones subfactors of type An
Let N ⊂ M be the Jones' subfactor of type An and N ⊂ M ⊂ M1 ⊂ M2 ⊂
· · · ⊂ Mk ⊂ be the Jones tower. We label the vertices of the Dynkin diagram
An by a0, a1,· · · , an−1 as in Figure 10. Then the Goodman-de la Harpe-Jones
subfactor GHJ(An,∗ = am) is isomorphic to pN ⊂ pMm−1p, where p is a minimal
projection in Proj(N ′ ∩ Mm−1) corresponding to the vertex am. Hence in this
case the principal graph and the dual principal graph coincide and fusion rule of
even vertices of both graphs becomes Aeven
.
n
a4
∗A =a0
a2
a1
a3
an−3
an−1
· · ·
an−2
Figure 10: The label of vertices of the Dynkin diagram An.
4.2. Goodman-de la Harpe-Jones subfactors of type D2n+1
d2n−1, d′
2n−1 as in Figure 11.
We label the vertices of the Dynkin diagram D2n+1 by d0, d1, d2,· · · , d2n−2,
The Goodman-de la Harpe-Jones subfactor GHJ(D2n+1,∗K = d0) is isomor-
phic to the unique index 2 subfactor N ⊂ N ⋊ Z2.
If the vertex ∗K 6= d0, d2n−1, d′
2n−1, GHJ(D2n+1,∗K) has nontrivial intermedi-
ate subfactor as in Figure 12 because we have the decomposition of connections
AdkD = Ad0D · D[k]D for k = 1, 2, . . . , 2n − 2. Here D[k]D is the D2n+1-D2n+1
connection corresponding to the vertex [k] as in Figures 33 and 42.
The (dual) principal graphs of GHJ(D2n+1,∗K) are given in Figures 51'73 for
The incidence matrices of the (dual) principal graphs of GHJ(Dodd,∗K) are
n = 2, 3, 4, 5.
also given in Figure 102.
12
d0
d2
d4
d2n−4
d2n−2
· · ·
d1
d3
d2n−3
d2n−1 d′
2n−1
Figure 11: The label of vertices of the Dynkin diagram D2n+1.
· · · −→
· · · −→
N
∩
∃P
∩
M
A4n−1
∗
D2n+1
D2n+1
d0
dk
The fusion rule
of even vertices
(N ⊂ P ) ∼= (N ⊂ N ⋊ Z2)
index = 2
P 6= M
· · · −→
The principal graph ∼= The dual principal graph
Aeven
4n−1
=
Aeven
4n−1
Figure 12:
4.3. Goodman-de la Harpe-Jones subfactors of type D2n
d2n−2, d′
2n−2 as in Figure 13.
We label the vertices of the Dynkin diagram D2n by d0, d1, d2,· · · , d2n−3,
The Goodman-de la Harpe-Jones subfactor GHJ(D2n,∗K = d0), GHJ(D4,∗K =
2) are isomorphic to the unique index 2 subfactor
d2) and GHJ(D4,∗K = d′
N ⊂ N ⋊ Z2.
If n > 2 and the vertex ∗K 6= d0, GHJ(D2n,∗K) has nontrivial intermedi-
ate subfactor as in Figure 12 because we have the decomposition of connections
AdkD = Ad0D · D[k]D for k 6= 0. Here D[k]D is the D2n-D2n connection correspond-
ing to the vertex [k] as in Figures 32 and 43.
The (dual) principal graphs of GHJ(D2n,∗K) are given in Figures 74'101 for
The incidence matrices of the (dual) principal graphs of GHJ(Deven,∗K) are
n = 3, 4, 5, 6.
also given in Figures 103 and 104.
13
d0
d2
d4
d2n−6
d2n−4
d2n−2 d′
2n−2
· · ·
d1
d3
d2n−5
d2n−3
Figure 13: The label of vertices of the Dynkin diagram D2n.
A4n−3
∗
D2n
D2n
d0
dk
The number
of even vertices
The fusion rule
of even vertices
· · · −→
· · · −→
N
∩
∃P
∩
M
(N ⊂ P ) ∼= (N ⊂ N ⋊ Z2)
index = 2
P 6= M
· · · −→
The principal graph 6∼= The dual principal graph
2n − 1
Aeven
4n−3
commutative
6=
6=
Figure 14:
2n + 2
D2nZ even
D2n
non-commutative
Example 4.1 From these computations for GHJ(Dn,∗K) as above, the (dual)
principal graphs of GHJ(Dn,∗ = triple point) can be obtained for general n as in
Figure 105.
4.4. Goodman-de la Harpe-Jones subfactors of type E6
We label the vertices of the Dynkin diagram E6 by e0, e1, e2,· · · , e5 as in
Figure 15.
The Goodman-de la Harpe-Jones subfactor GHJ(E6,∗K = e0), has index 3 +
√3 and it has the same principal and dual principal graph. But the fusion rules
of the two graphs are different. This subfactor is known as the example which
has the smallest index among such subfactors. The (dual) principal graphs of
GHJ(E6,∗K) are given in Figures 106'109.
If the vertex ∗K 6= e0, e4, GHJ(E6,∗K) has nontrivial intermediate subfactor
as in Figure 16 because we have the decomposition of connections Aek E = Ae0E ·
14
E[wk]E for k = 1, 2, 3, 5. Here E[wk]E is the E6-E6 connection corresponding to
the vertex [k] as in Figures 34 and 44.
e0
e2
e4
e1
e3
e5
Figure 15: The label of vertices of the Dynkin diagram E6.
∗
A11
E6
E6
e0
ek
· · · −→
· · · −→
N
∩
∃P
∩
M
(N ⊂ P ) ∼= GHJ(E6,∗K = e0)
index = 3 + √3
P 6= M
· · · −→
The principal graph ∼= The dual principal graph
The fusion rule
of even vertices
Aeven
11
commutative
6=
E6Z even
commutative
E6
Figure 16:
4.5. Goodman-de la Harpe-Jones subfactors of type E7
Figure 17.
We label the vertices of the Dynkin diagram E7 by e0, e1, e2,· · · , e6 as in
The Goodman-de la Harpe-Jones subfactor GHJ(E7,∗K = e0), has index A17
E7
which is approximately 7.759. Here A17 and E7 represents the "total mass" of
the graph A17 and E7 respectively, i.e. the sum of squares of normalized Perron-
Frobenius eigenvalues over all the vertices of the graph. The (dual) principal
graphs of GHJ(E7,∗K) are given in Figures 110'116.
If the vertex ∗K 6= e0, e4, e5, GHJ(E7,∗K) has nontrivial intermediate subfac-
tor as in Figure 18 because we have the decomposition of connections Aek E =
15
Ae0E · E[wk]E for k = 1, 2, 3 and Ae6E = Ae0E · E[w(5)]E. Here E[wk]E is the E7-E7
connection corresponding to the vertex [k] (k = 1, 2, 3) and (5) as in Figures 35
and 45.
e0
e2
e4
e6
e1
e3
e5
Figure 17: The label of vertices of the Dynkin diagram E7.
∗
A17
E7
E7
e0
ek
The number
of even vertices
The fusion rule
of even vertices
· · · −→
· · · −→
N
∩
∃P
∩
M
(N ⊂ P ) ∼= GHJ(E7,∗K = e0)
index ; 7.759
P 6= M
· · · −→
The principal graph 6∼= The dual principal graph
9
Aeven
17
commutative
=
6=
9
E7Z even
commutative
E7
Figure 18:
4.6. Goodman-de la Harpe-Jones subfactors of type E8
Figure 19.
We label the vertices of the Dynkin diagram E8 by e0, e1, e2,· · · , e7 as in
The Goodman-de la Harpe-Jones subfactor GHJ(E8,∗K = e0), has index A29
E8
which is approximately 19.48. Here A29 and E8 represents the "total mass" of
the graph A29 and E8 respectively. The (dual) principal graphs of GHJ(E8,∗K)
are given in Figures 117'124.
If the vertex ∗K 6= e0, GHJ(E7,∗K) has nontrivial intermediate subfactor as in
16
Figure 20 because we have the decomposition of connections Aek E = Ae0E · E[wk]E
for k 6= 0. Here E[wk]E is the E8-E8 connection corresponding to the vertex [k]
as in Figures 36, 46 and 47.
e0
e2
e4
e6
e1
e3
e5
e7
Figure 19: The label of vertices of the Dynkin diagram E8.
∗
A29
E8
E8
e0
ek
The number
of even vertices
The fusion rule
of even vertices
· · · −→
· · · −→
N
∩
∃P
∩
M
(N ⊂ P ) ∼= GHJ(E8,∗K = e0)
index ; 19.48
P 6= M
· · · −→
The principal graph 6∼= The dual principal graph
15
Aeven
29
commutative
6=
6=
16
E8Z even
commutative
E8
Figure 20:
5 An application to subequivalence on paragroups
Let K be one of the Dynkin diagrams D2n(n ≥ 3), E6, E8 and Al the Dynkin
diagram of type A with the same Perron-Frobenius eigenvalue as K. We can
choose the vertex ∗K so that the GHJ subfactor GHJ(K,∗K) does not have index
2. Let N ⊂ M be the GHJ subfactor GHJ(K,∗K) chosen as above, then the fusion
algebra of N-N bimodules is isomorphic to Aeven
and the fusion algebra of M-M
bimodules is isomorphic to KZ even
contains K even as its strict
fusion subalgebra, the paragroup of type K becomes a strictly subequivalent to
K . Because KZ even
l
K
17
that of type Al. Here we use the terminology strictly subequivalent in the sense
that a fusion algerba A is subequivalent but not equivalent to B. And in such a
case, we denote A ≻ B.
In the case of D4, we can choose the direct sum of 3 connections for GHJ(D4,∗K)
(∗K = d0, d2, d′
2) as a connection for subequivalence between A5 and D4 para-
groups.
Hence we get the following subequivalence of paragroups.
Theorem 5.1 The paragroups of Jones' type A subfactors have the following
strictly subequivalent paragroups.
A4n−3 ≻ D2n (n ≥ 2), A11 ≻ E6, A29 ≻ E8.
Figure 21: Essential paths on the Coxeter graph A4
Figure 22: Essential paths on the Coxeter graph A5
Figure 23: Essential paths on the Coxeter graph D4
18
Figure 24: Essential paths on the Coxeter graph D5
Figure 25: Essential paths on the Coxeter graph D6
19
Figure 26: Essential paths on the Coxeter graph E6
20
Figure 27: Essential paths on the Coxeter graph E7 (1)
Figure 28: Essential paths on the Coxeter graph E7 (2)
21
Figure 29: Essential paths on the Coxeter graph E8 (1)
22
Figure 30: Essential paths on the Coxeter graph E8 (2)
23
Figure 31: Chiral symmetry for the Coxeter graph An
Figure 32: Chiral symmetry for the Coxeter graph Deven
24
Figure 33: Chiral symmetry for the Coxeter graph Dodd
Figure 34: Chiral symmetry for the Coxeter graph E6
25
Figure 35: Chiral symmetry for the Coxeter graph E7
Figure 36: Chiral symmetry for the Coxeter graph E8
26
Figure 37: Vertical graphs for connections on the Coxeter graph A3
Figure 38: Vertical graphs for connections on the Coxeter graph A4
Figure 39: Vertical graphs for connections on the Coxeter graph A5
27
Figure 40: Vertical graphs for connections on the Coxeter graph A6
Figure 41: Vertical graphs for connections on the Coxeter graph D4
28
Figure 42: Vertical graphs for connections on the Coxeter graph D5
Figure 43: Vertical graphs for connections on the Coxeter graph D6
29
Figure 44: The incidence matrices of the vertical edges of E6-E6 connections
30
Figure 45: The incidence matrices of the vertical edges of E7-E7 connections
31
Figure 46: The incidence matrices of the vertical edges of E8-E8 even connections
32
Figure 47: The incidence matrices of the vertical edges of E8-E8 odd connections
33
Figure 48: The fusion table of E6-E6 connections
34
Figure 49: The fusion table of E7-E7 connections
35
Figure 50: A part of the fusion table of E8-E8 connections
36
Figure 51: The (dual) principal graph of GHJ(D5,∗ = d1).
Figure 52: The (dual) principal graph of GHJ(D5,∗ = d2).
37
Figure 53: The (dual) principal graph of GHJ(D5,∗ = d3).
Figure 54: The (dual) principal graph of GHJ(D7,∗ = d1).
38
Figure 55: The (dual) principal graph of GHJ(D7,∗ = d2).
Figure 56: The (dual) principal graph of GHJ(D7,∗ = d3).
39
Figure 57: The (dual) principal graph of GHJ(D7,∗ = d4).
Figure 58: The (dual) principal graph of GHJ(D7,∗ = d5).
40
Figure 59: The (dual) principal graph of GHJ(D9,∗ = d1).
Figure 60: The (dual) principal graph of GHJ(D9,∗ = d2).
41
Figure 61: The (dual) principal graph of GHJ(D9,∗ = d3).
Figure 62: The (dual) principal graph of GHJ(D9,∗ = d4).
42
Figure 63: The (dual) principal graph of GHJ(D9,∗ = d5).
Figure 64: The (dual) principal graph of GHJ(D9,∗ = d6).
43
Figure 65: The (dual) principal graph of GHJ(D9,∗ = d7).
Figure 66: The (dual) principal graph of GHJ(D11,∗ = d1).
44
Figure 67: The (dual) principal graph of GHJ(D11,∗ = d2).
Figure 68: The (dual) principal graph of GHJ(D11,∗ = d3).
45
Figure 69: The (dual) principal graph of GHJ(D11,∗ = d4).
Figure 70: The (dual) principal graph of GHJ(D11,∗ = d5).
46
Figure 71: The (dual) principal graph of GHJ(D11,∗ = d6).
Figure 72: The (dual) principal graph of GHJ(D11,∗ = d7).
47
Figure 73: The (dual) principal graph of GHJ(D11,∗ = d8).
Figure 74: The (dual) principal graph of GHJ(D6,∗ = d1).
48
Figure 75: The (dual) principal graph of GHJ(D6,∗ = d2).
Figure 76: The (dual) principal graph of GHJ(D6,∗ = d3).
49
Figure 77: The (dual) principal graph of GHJ(D6,∗ = d4).
50
Figure 78: The (dual) principal graph of GHJ(D8,∗ = d1).
51
Figure 79: The (dual) principal graph of GHJ(D8,∗ = d2).
52
Figure 80: The (dual) principal graph of GHJ(D8,∗ = d3).
53
Figure 81: The (dual) principal graph of GHJ(D8,∗ = d4).
54
Figure 82: The (dual) principal graph of GHJ(D8,∗ = d5).
55
Figure 83: The (dual) principal graph of GHJ(D8,∗ = d6).
56
Figure 84: The (dual) principal graph of GHJ(D10,∗ = d1).
57
Figure 85: The (dual) principal graph of GHJ(D10,∗ = d2).
58
Figure 86: The (dual) principal graph of GHJ(D10,∗ = d3).
59
Figure 87: The (dual) principal graph of GHJ(D10,∗ = d4).
60
Figure 88: The (dual) principal graph of GHJ(D10,∗ = d5).
61
Figure 89: The (dual) principal graph of GHJ(D10,∗ = d6).
62
Figure 90: The (dual) principal graph of GHJ(D10,∗ = d7).
63
Figure 91: The (dual) principal graph of GHJ(D10,∗ = d8).
64
Figure 92: The (dual) principal graph of GHJ(D12,∗ = d1).
65
Figure 93: The (dual) principal graph of GHJ(D12,∗ = d2).
66
Figure 94: The (dual) principal graph of GHJ(D12,∗ = d3).
67
Figure 95: The (dual) principal graph of GHJ(D12,∗ = d4).
68
Figure 96: The (dual) principal graph of GHJ(D12,∗ = d5).
69
Figure 97: The (dual) principal graph of GHJ(D12,∗ = d6).
70
Figure 98: The (dual) principal graph of GHJ(D12,∗ = d7).
71
Figure 99: The (dual) principal graph of GHJ(D12,∗ = d8).
72
Figure 100: The (dual) principal graph of GHJ(D12,∗ = d9).
73
Figure 101: The (dual) principal graph of GHJ(D12,∗ = d10).
74
Figure 102: The incidence matrices of the (dual) principal graphs of GHJ(Dodd).
75
Figure 103: The incidence matrices of the principal graphs of GHJ(Deven).
76
Figure 104: The incidence matrices of the dual principal graphs of GHJ(Deven).
77
The incidence matrices of the (dual) principal graphs of
GHJ(D,∗ = triple point).
Figure 105:
78
Figure 106: The (dual) principal graph of GHJ(E6,∗ = e0).
Figure 107: The (dual) principal graph of GHJ(E6,∗ = e1).
79
Figure 108: The (dual) principal graph of GHJ(E6,∗ = e2).
Figure 109: The (dual) principal graph of GHJ(E6,∗ = e3).
80
Figure 110: The (dual) principal graph of GHJ(E7,∗ = e0).
81
Figure 111: The (dual) principal graph of the GHJ subfactor corresponding to
(E7,∗ = e1).
82
Figure 112: The (dual) principal graph of the GHJ subfactor corresponding to
(E7,∗ = e2).
83
Figure 113: The (dual) principal graph of the GHJ subfactor corresponding to
(E7,∗ = e3).
84
Figure 114: The (dual) principal graph of the GHJ subfactor corresponding to
(E7,∗ = e4).
85
Figure 115: The (dual) principal graph of the GHJ subfactor corresponding to
(E7,∗ = e5).
86
Figure 116: The (dual) principal graph of the GHJ subfactor corresponding to
(E7,∗ = e6).
87
Figure 117: The (dual) principal graph of the GHJ subfactor corresponding to
(E8,∗ = e0).
88
Figure 118: The (dual) principal graph of the GHJ subfactor corresponding to
(E8,∗ = e1).
89
Figure 119: The (dual) principal graph of the GHJ subfactor corresponding to
(E8,∗ = e2).
90
Figure 120: The (dual) principal graph of the GHJ subfactor corresponding to
(E8,∗ = e3).
91
Figure 121: The (dual) principal graph of the GHJ subfactor corresponding to
(E8,∗ = e4).
92
Figure 122: The (dual) principal graph of the GHJ subfactor corresponding to
(E8,∗ = e5).
93
Figure 123: The (dual) principal graph of the GHJ subfactor corresponding to
(E8,∗ = e6).
94
Figure 124: The (dual) principal graph of the GHJ subfactor corresponding to
(E8,∗ = e7).
95
References
[1] Asaeda, M. and U. Haagerup, Exotic subfactors of finite depth with Jones
indices (5 +√13)/2 and (5 +√17)/2, Comm. Math. Phys. 202 (1999), 1 -- 63.
[2] Bion-Nadal, J., Subfactor of the hyperfinite II1 factor with Coxeter graph
E6 as invariant, J. Operator Theory 28 (1992), 27 -- 50.
[3] Bisch, D., Principal graphs of subfactors with small Jones index, Math. Ann.
311 (1998), 223 -- 231.
[4] Bisch, D., On the existence of central sequences in subfactors, Trans. Amer.
Math. Soc. 321 (1990), 117 -- 128.
[5] Bockenhauer, J. and D. E. Evans, Modular Invariants, Graphs and α-
Induction for Nets of Subfactors I, Comm. Math. Phys. 197 (1998), 361 -- 386.
[6] Bockenhauer, J. and D. E. Evans, Modular Invariants, Graphs and α-
Induction for Nets of Subfactors II, Comm. Math. Phys. 200 (1999), 57 -- 103.
[7] Bockenhauer, J. and D. E. Evans, Modular Invariants, Graphs and α-
Induction for Nets of Subfactors III, Comm. Math. Phys. 205 (1999), 183 --
228.
[8] Bockenhauer, J., D. E. Evans and Y. Kawahigashi, Chiral structure of mod-
ular invariants for subfactors, Comm. Math. Phys. 210 (2000), 733 -- 784.
[9] Evans, D. E. and Y. Kawahigashi, Quantum symmetries on operator alge-
bras, Oxford University Press, Oxford, 1998.
[10] Cappelli, A., C. Itzykson, and J.-B. Zuber, The A-D-E classification of
conformal invariant theories, Comm. Math. Phys. 113
minimal and A(1)
1
(1987), 1 -- 26.
[11] Goodman, F., P. de la Harpe and V. F. R. Jones, Coxeter graphs and towers
of algebras, MSRI Publications, 14, Springer, Berlin, 1989.
[12] Goto, S., On Ocneanu's theory of double triangle algebras for subfactors
and classification of irreducible connections on the Dynkin diagrams, Expos.
Math. 28 (2010), 218 -- 253.
96
[13] Izumi, M., Application of fusion rules to classification of subfactors, Publ.
Res. Inst. Math. Sci. 27 (1991), 953 -- 994.
[14] Izumi, M., On flatness of the Coxeter graph E8, Pacific J. Math. 166 (1994),
305 -- 327.
[15] Jones, V. F. R., Index for subfactors, Invent. Math. 72 (1983), 1 -- 15.
[16] Kawahigashi, Y., On flatness of Ocneanu's connections on the Dynkin dia-
grams and classification of subfactors, J. Funct. Anal. 127 (1995), 63 -- 107.
[17] Kawahigashi, Y., Classification of paragroup actions on subfactors, Publ.
Res. Inst. Math. Sci. 31 (1995), 481 -- 517.
[18] Ocneanu, A., Quantized group string algebras and Galois theory for algebras,
in "Operator algebras and applications, Vol. 2 (Warwick, 1987)," London
Math. Soc. Lect. Note Series Vol. 136, Cambridge University Press, (1988),
119 -- 172.
[19] Ocneanu, A., Paths on Coxeter diagrams: from Platonic solids and singu-
larities to minimal models and subfactors, (Notes recorded by S. Goto), in
Lectures on operator theory, (ed. B. V. Rajarama Bhat et al.), The Fields In-
stitute Monographs, Providence, Rhode Island: AMS Publications. (2000),
243 -- 323.
[20] Okamoto, S., Invariants for subfactors arising from Coxeter graphs, Current
Topics in Operator Algebras, World Scientific Publishing, (1991), 84 -- 103.
[21] Sunder, V. S. and A. K. Vijayarajan, On the non-occurrence of the Coxeter
graphs β2n+1, E7, D2n+1 as principal graphs of an inclusion of II1 factors,
Pacific J. Math. 161 (1993), 185 -- 200.
[22] Xu, F., New braided endomorphisms from conformal inclusions, Comm.
Math. Phys. 192 (1998), 349 -- 403.
97
|
1004.0128 | 3 | 1004 | 2010-08-04T08:05:44 | Maximal injective and mixing masas in group factors | [
"math.OA"
] | We present families of pairs of finite von Neumann algebras $A\subset M$ where $A$ is a maximal injective masa in the type $\mathrm{II}_1$ factor $M$ with separable predual. Our results make use of the strong mixing and the asymptotic orthogonality properties of $A$ in $M$ and are borrowed from ideas of S. Popa who proved that if $G$ is a non abelian free group and if $a$ is one of its generators, then the von Neumann algebra generated by $a$ is maximal injective in the factor $L(G)$. Our results apply to pairs $H<G$ where $H$ is an infinite abelian subgroup of a suitable amalgamated product group $G$. | math.OA | math | Maximal injective and mixing masas in group
factors
Paul Jolissaint
October 30, 2018
Abstract
We present families of pairs of finite von Neumann algebras A ⊂ M where A is a
maximal injective masa in the type II1 factor M with separable predual. Our results
make use of the strong mixing and the asymptotic orthogonality properties of A in M .
Our method is directly borrowed from S. Popa's original result where he proved that if
G is a non abelian free group and if a is one of its generators, then the von Neumann
algebra generated by a is maximal injective in the factor L(G). Our results apply to
pairs H < G where H is an infinite abelian subgroup of a suitable amalgamated product
group G.
Mathematics Subject Classification: Primary 46L10; Secondary 20E06.
Key words: Type II1-factors, maximal abelian subalgebras, injective algebras.
0
1
0
2
g
u
A
4
]
.
A
O
h
t
a
m
[
3
v
8
2
1
0
.
4
0
0
1
:
v
i
X
r
a
1 Introduction
Inspired by [5] and [7], we introduced in [4] the notions of weakly mixing and strongly mixing
masas in finite von Neumann algebras, and we presented several families of examples coming
essentially from pairs of groups. The purpose of the present paper is to use these mixing
properties in order to give examples of masas that are maximal injective in the ambiant
factor. In fact, we will present a slight generalization of the main result in S. Popa's article
[6], and our point of view is very similar to the exposition of Popa's theorem as presented in
A. Sinclair's and R. Smith's monograph [8].
Before stating our main results, let us recall our notations. In this article, M will always
denote a finite von Neumann algebra with separable predual (equivalently, it admits a count-
able dense set for the strong operator topology), and τ will be some fixed finite, faithful,
normal, normalized trace on M. If B is a unital von Neumann subalgebra of M, then EB
denotes the τ -preserving conditional expectation onto B, and we let M ⊖ B be the set of all
x ∈ M such that EB(x) = 0; equivalently, it is the set of all x that satisfy τ (xb) = 0 for all
b ∈ B. For future use, we observe that, for all x, y ∈ M and every u ∈ B, one has:
(∗) EB((x − EB(x))u(y − EB(y))) = EB(xuy) − EB(x)uEB(y).
If Φ is a linear map on M, we set
kΦk∞,2 = sup
x∈M,kxk≤1kΦ(x)k2.
If ω is a free ultrafilter on N, then M ω denotes the associated ultrapower algebra; M embeds
into M ω in a natural way, and, if M is a factor, we say that it has Property Γ of Murray
1
and von Neumann if the relative commutant M ′ ∩ M ω is non-trivial. If it is the case, it is
automatically diffuse. If M does not have Property Γ, it is called a full factor. For all of this,
see for instance Appendix A in [8].
Let G be a countable group. We denote by L(G) the von Neumann algebra generated by
the left regular representation of G on ℓ2(G); we denote simply by gξ the action of g ∈ G on
ξ ∈ ℓ2(G) defined by (gξ)(h) = ξ(g−1h) for every h ∈ G. As is well known, L(G) is a finite
von Neumann algebra, every x ∈ L(G) has a unique Fourier expansion Pg∈G x(g)g which
converges in the k·k2-sense and Pg x(g)2 = kxk2
2. Furthermore, L(G) is a factor if and only
if G is an ICC group. If H is a subgroup of G, then its associated von Neumann algebra L(H)
embeds into L(G) by setting x(g) = 0 for all g ∈ G \ H if x ∈ L(H).
Let now M be a type II1 factor with separable predual and let A be a unital, abelian von
Neumann subalgebra of M. Following [4] and [2], we say that A is weakly mixing in M if
there exists a sequence of unitary operators (un) ⊂ A such that
(∗∗)
n→∞kEA(xuny)k2 = 0 ∀x, y ∈ M ⊖ A.
lim
Note that by (∗), the latter is equivalent to limn kEA(xuny) − EA(x)unEA(y)k2 = 0 for all
x, y ∈ M. We say that A is strongly mixing in M if (∗∗) holds for all sequences of unitary
operators (un) ⊂ A such that limn→∞ un = 0 in the weak operator topology. If G is a countable
ICC group and if H is an abelian subgroup of G, if we set M = L(G) and A = L(H), then
it follows from [4] that A is weakly mixing in M if and only if the pair H < G satisfies the
so-called condition (SS): for every finite subset C ⊂ G \ H, there exists h ∈ H such that
g1hg2 /∈ H for all g1, g2 ∈ C. Similarly, A is strongly mixing in M if and only if the pair
H < G satisfies the so-called condition (ST): for every finite subset C ⊂ G \ H, there exists
a finite subset E ⊂ H such that g1hg2 /∈ H for all h ∈ H \ E and all g1, g2 ∈ C.
Remarks. (1) Let A be a masa in a type II1 factor with separable predual M. The main
theorem of [9] states that A is a singular masa if and only if it is weakly mixing in M. See
also Theorem 11.1.2 of [8].
(2) Let H < G be a pair of groups as above. It is easy to see that it satisfies condition (ST)
if and only if, for every g ∈ G\ H, the intersection H ∩ gHg−1 is a finite group. In particular,
when the intersection is trivial, H is said to be malnormal in G. Thus, if a subgroup H of a
group G which satisfies condition (ST) is also called almost malnormal in G.
Inspired by [6], the authors of [1] introduced the following property for a pair A ⊂ M where
A is abelian and M is a type II1 factor: one says that A has the asymptotic orthogonality
property if there is a free ultrafilter ω on N such that x(1)y1 ⊥ y2x(2) in L2(M ω) whenever
x(1), x(2) ∈ A′ ∩ M ω with EAω (x(i))) = 0, and y1, y2 ∈ M with EA(yi) = 0 for i = 1, 2. Then
the authors of [1] prove in Corollary 2.3 that if A is a singular masa which has the asymptotic
orthogonality property, then it is maximal injective in M.
As we will see, strongly mixing masas provide a central decomposition of intermediate
algebras that strengthens maximal injectivity. Our first result extends to arbitrary pairs
A ⊂ M Theorem 14.2.1 of [8] which was stated for group algebras; our proof differs partly
from that in [8].
Theorem 1 Let M be a type II1 factor with separable predual and let 1 ∈ A ⊂ M be a
strongly mixing abelian von Neumann subalgebra of M. If N is a von Neumann subalgebra
of M which contains A, then there exists a partition of the unity (ek)k≥0 in the center Z of
N such that Ne0 = Ae0 and, for every k ≥ 1 such that ek 6= 0, Nek is a type II1 factor and
(N ′ ∩ Aω)ek has a non-zero atomic part.
2
The strong mixing assumption in Theorem 1 is essential, as was kindly communicated to
us by S. White. Indeed, let A0 ⊂ M0 be a weakly, but not strongly mixing masa in a type II1
factor. Then A = A0⊗A0 is obviously a weakly mixing masa in the factor M = M0⊗M0, but
taking N = A0⊗M0, we have Z = A0 ⊗ 1 which has no atoms.
The second result is essentially Theorem 14.2.5 of [8] where it was stated in the special
case of the free group factors. We recall it for the sake of completeness and future use.
Theorem 2 Let A be a strongly mixing masa that satisfies the asymptotic orthogonality prop-
erty in a type II1 factor M with separable predual, let N be an intermediate von Neumann
subalgebra and let (ek)k≥0 ⊂ Z(N) be the corresponding partition of the unity as in Theorem
1. Then for every k ≥ 1 such that ek 6= 0, Nek is a full type II1 factor. In particular, A is a
maximal injective subalgebra of M.
As will be seen, pairs of groups can provide examples of such algebras. Thus, for the rest
of the present section, let G be an ICC countable group and let H be an abelian subgroup of
G. Put M = L(G) and A = L(H). We assume that the pair H < G satisfies the following
hypotheses:
There exists a sequence (Wm)m≥1 of subsets of G \ H such that
(H1) Wm ⊂ Wm+1 for every m ≥ 1 and Sm Wm = G \ H ;
(H2) if Vm denotes the complementary set of Wm ∪ W −1
m in G\ H, then for all g1, g2 ∈ G\ H,
there exists a positive integer m1 = m1(g1, g2) such that g1Vm ∩ Vmg2 = ∅ for every
m > m1;
(H3) there exist an integer m0 > 0 and an element h ∈ H such that, for every m > m0, one
can find an integer im > 0 such that hiWmh−i ∩ Wm = ∅ for every i ≥ im.
Theorem 3 Let H < G be a pair of groups such that G is countable and ICC, H is abelian,
and assume that the pair satisfies condition (ST) on the one hand, and conditions (H1),
(H2) and (H3) on the other hand. Then L(H) is a strongly mixing masa that satisfies the
asymptotic orthogonality property in L(G). Thus, if N is a von Neumann subalgebra of M
which contains A, then there exists a partition of the unity (ek)k≥0 in the center of N such
that Ne0 = Ae0 and, for every k ≥ 1 such that ek 6= 0, Nek is a full type II1 factor. In
particular, A is a maximal injective subalgebra of M.
The next section is devoted to the proof of Theorems 1, 2 and 3. In Section 3, we provide a
family of examples of pairs H < G that satisfy all conditions of Theorem 2, hence which gives
examples of maximal injective masas in group factors; it is given by amalgamated products
G = H ∗Z K where H is infinite and abelian, Z is finite and different from K, and G is ICC.
Acknowledgements. I am grateful to Stuart White for helpful comments and for having
detected a mistake in the first version of Theorem 1.
2 Proofs of the main results
Let M be a type II1 factor with separable predual and let A be a masa in M. Before proving
Theorem 1, we present an auxiliary result of independent interest. See also Section 4 in [2]
for related results.
3
Proposition 4 Let N be a finite von Neumanna algebra with separable predual and let A ⊂ N
be a strongly mixing abelian von Neumann algebra in N. Then:
(1) the von Neumann algebra N ′ ∩ Aω has a non-zero atomic part;
(2) for every diffuse von Neumann subalgebra 1 ∈ B ⊂ A and for every unitary u ∈ U(N),
one has
In particular, NN (B)′′ ⊂ A.
kEB − EuBu∗k∞,2 ≥ ku − EA(u)k2.
Proof. (1) We assume that N ′ ∩ Aω is diffuse and we will get a contradiction. Choose a Haar
unitary u ∈ U(N ′ ∩ Aω); this means that τω(uk) = 0 for all integers k 6= 0. The algebra N
being separable with respect to the k · k2-topology, choose an increasing sequence of finite
subsets
E1 ⊂ E2 ⊂ . . .{x ∈ N ⊖ A : kxk ≤ 1}
so that Sn En is dense in {x ∈ N ⊖ A : kxk ≤ 1}. Similarly, choose an increasing sequence of
finite subsets F1 ⊂ F2 . . . of the unit ball of A which is k · k2-dense. Let us write u = [(un)n]
with un ∈ U(A). The unitaries uk being pairwise orthogonal in Aω, one has, by Parceval's
inequality
hence
Xk∈Z
τω(xuk)2 ≤ kxk2
2
lim
k→∞
τω(xuk) = 0
for every x ∈ Aω. Thus, for every integer n ≥ 1, there exists an integer kn > 0 such that
a∈Fn τω(auk) <
max
1
2n ∀k ≥ kn.
As moreover xuk = ukx for every x ∈ N and every integer k, we infer that, for every n > 0,
there exists a positive integer ℓn such that
τ (aukn
ℓn ) <
Thus the sequence (ukn
for every x ∈ N ⊖ A
1
n ∀a ∈ Fn
and kukn
ℓn xu−kn
ℓn − xk2 <
1
n ∀x ∈ En.
ℓn ) ⊂ U(A) converges to 0 in the weak operator topology, and we get
n→∞kEA(x∗ukn
lim
ℓn x)k2 = 0.
However, if x 6= 0 is orthogonal to A, we have
0 < kEA(x∗x)k2 ≤ kEA(x∗(x − ukn
ℓn xu−kn
ℓn ))k2 + kEA(x∗ukn
ℓn x)k2 → 0,
which gives the desired contradiction.
(2) Fix a diffuse von Neumann subalgebra B of A, a unitary operator u ∈ U(N), and let us
consider x = u∗ − EA(u∗) and y = u − EA(u) ∈ N ⊖ A and let ε > 0. One has for every
v ∈ U(B):
EB(xvy) = EB(u∗vu) − EB(EA(u∗)vEA(u))
As B is diffuse, there exists a sequence of unitaries (vn) ⊂ U(B) which converges to 0 with
respect to the weak operator topology. Since A is strongly mixing in N, there exists a positive
4
integer n such that kEA(xvny)k2 ≤ ε. As EB = EBEA, we have kEB(xvny)k2 ≤ ε as well. The
above computation gives
kEB(u∗vnu)k2 ≤ kEB(EA(u∗)vnEA(u))k + ε ≤ kEA(u)k2 + ε.
We get then:
kEB − EuBu∗k2
∞,2 ≥ kvn − EuBu∗(vn)k2
2
2
= ku∗vnu − EB(u∗vnu)k2
= 1 − kEB(u∗vnu)k2
≥ 1 − (kEA(u)k2 + ε)2
= 1 − kEA(u)k2
= ku − EA(u)k2
2
2 − 2εkEA(u)k2 − ε2
2 − 2εkEA(u)k2 − ε2.
As ε is arbitrary, we get the conclusion.
(cid:3)
Remark. Let A be a masa in the type II1 factor M. Let us say that it satisfies condition
(D) if, for every diffuse von Neumann subalgebra 1 ∈ B ⊂ A, the normalizer NM (B) is equal
to the unitary group U(A). Statement (2) of Proposition 4 above shows that if A is strongly
mixing in M then it satisfies condition (D), and one may ask whether the converse holds true.
In fact, here is a partial positive answer: suppose that G is an ICC group and that H is
an infinite abelian subgroup of G. Set A = L(H) and M = L(G). Then if A ⊂ M satisfies
condition (D), A is strongly mixing in M. Indeed, suppose that it is not the case; by Theorem
3.5 in [4], the pair H < G would not satisfy condition (ST), hence there exists an element
g ∈ G \ H such that H ∩ gHg−1 is an infinite subgroup of H. Set B = L(H ∩ gHg−1), which
is a diffuse subalgebra of A. Then g ∈ NM (B) but it does not belong to U(A).
Proof of Theorem 1. Our proof is strongly inspired by that of theorem 14.2.1 in [8]. Let e0 ∈ Z
be the largest projection such that Ne0 = Ae0 and set e = 1 − e0. We will prove that eZ is
atomic. Assume on the contrary that it is not. Then there exists a projection q ∈ Z such that
0 6= q ≤ e and qZ is a diffuse algebra and (e− q)Z is atomic. Set B = (1− q)A + qZ. As A is
maximal abelian and since Z commutes with A, we have Z ⊂ A hence B ⊂ A with qB = qZ
diffuse, hence B is diffuse as well. By the previous proposition, one has NM (B)′′ ⊂ A, and in
particular
q(B′ ∩ M) ⊂ A. As in the proof of Theorem 14.2.1 in [8], we get
qN ⊂ q(Z ′ ∩ M) = q(B′ ∩ M) ⊂ qA,
hence qA = qN which is diffuse, and this contradicts maximality of e0 since q ≤ 1 − e0.
Finally, as observed in [2], for every non-zero projection e ∈ A, the masa Ae is still strongly
mixing in eMe. Hence Aek is strongly mixing in ekMek, and we deduce that (N ′ ∩ Aω)ek has
atoms by the previous proposition.
(cid:3)
Proof of Theorem 2. We reproduce the proof of Theorem 14.2.5 in [8] for the reader's con-
venience. If A ⊂ N ⊂ M are as in Theorem 2, let (ek)k≥0 ⊂ Z(N) be the partition of the
unity provided by Theorem 1: Ne0 = Ae0, and for every k > 0 such that ek 6= 0, the von
Neumann algebra Nek is a type II1 factor such that the von Neumann algebra (N ′∩Aω)ek has
a non-zero atomic part. If there is some k > 0 such that Nek has Property Γ, then the relative
commutant (Nek)′ ∩ (Nek)ω is diffuse, hence (N ′ ∩ Aω)ek ( (Nek)′ ∩ (Nek)ω. As in the proof
of Theorem 14.2.5 of [8], we choose some non-zero x ∈ (Nek)′ ∩ (Nek)ω such that EAω (x) = 0
5
and some unitary w ∈ Nek such that EA(w) = 0. By the asymptotic orthogonality property
of A ⊂ M applied to x(1) = x(2) = x and y1 = y2 = w, we get
2kxk2
2 = kwxk2
2 = 0
2 + kxwk2
2 = kwx − xwk2
which is a contradiction.
(cid:3)
From now on, we consider a countable, ICC group G and an abelian, infinite subgroup H
of G and we assume that the pair H < G satisfies the three conditions (H1) to (H3) in Section
1, and we set as before A = L(H) ⊂ M = L(G). For convenience, we recall some notations
from [8]: If W ⊂ G, let pW be the orthogonal projection of ℓ2(G) onto the subspace ℓ2(W ):
pW (x) = Xg∈W
x(g)g
if x = Xg∈G
x(g)g.
We remind the reader that for all V, W ⊂ G, one has pV pW = pV ∩W (thus in particular,
pV pW = 0 if V and W are disjoint), that
pgW g−1(x) = gpW (g−1xg)g−1
and pW (x)∗ = pW −1(x∗)
for all g ∈ G and x ∈ ℓ2(G), and that, if V ⊂ W , then kpV (x)k2 ≤ kpW (x)k2 for every x.
The proof of Theorem 3 follows immediately from part (iii) of the following lemma whose
proof is similar to those of Lemmas 14.2.3 and 14.2.4 in [8]. However, we give a proof for the
sake of completeness.
Lemma 5 Let G, H satisfy conditions (H1) and (H2), and let (Wm)m≥1 be the corresponding
sequence of subsets of G \ H. Assume also that it satisfies the following weaker variant of
condition (H3):
(H3') there exists an integer m0 > 0 such that, for every m > m0, one can find elements
h1,m, . . . , hnm,m ∈ H such that nm → ∞ as m → ∞ and
hi,mWmh−1
i,m ∩ hj,mWmh−1
j,m = ∅ ∀i 6= j.
(i) Let ε > 0, let m > m0 and h1,m, . . . , hnm,m be as in condition (H3') and let x ∈ ℓ2(G),
kxk2 ≤ 1, be such that
Then
khj,mxh−1
j,m − xk2 ≤ ε ∀j = 1, . . . , nm.
kpWm∪W −1
m
(x)k2
2 ≤ 4(ε2 + n−1
m ).
(ii) If y ∈ ℓ2(G) is such that EA(y) = 0, then
m→∞ky − pWm(y)k2 = 0.
lim
(iii) The abelian algebra A satisfies the asymptotic orthogonality property in M, namely, let
ω be a free ultrafilter on N, x(1), x(2) ∈ A′ ∩ M ω and y1, y2 ∈ M be such that EAω (x(j)) =
EA(yj) = 0 for j = 1, 2. Then y1x(1) ⊥ x(2)y2 in M ω and
ky1x(1) − x(2)y2k2
ω,2 = ky1x(1)k2
ω,2 + kx(2)y2k2
ω,2.
6
Proof. (i) Using (α + β)2 ≤ 2(α2 + β2) for arbitrary real numbers α and β, we get for every
m > m0 and every 1 ≤ j ≤ nm:
kpWm(x)k2
2 ≤ (kpWm(x − h−1
j,mxhj,m)k2 + kpWm(h−1
j,mxhj,m)k2)2
≤ 2kx − h−1
j,mxhj,mk2
≤ 2ε2 + 2kphj,mWmh−1
j,m
(x)k2
2.
2 + 2kpWm(h−1
j,mxhj,m)k2
2
Using condition (H3'), phj,mWmh−1
over 1 ≤ j ≤ nm, we get:
j,m
(x) is orthogonal to phi,mWmh−1
i,m
(x) for all i 6= j. Summing
nm
nmkpWm(x)k2
2 ≤ 2nmε2 + 2
Xj=1
Xj=1
≤ 2nmε2 + 2k
≤ 2nmε2 + 2.
nm
kphj,mWmh−1
j,m
(x)k2
2
phj,mWmh−1
j,m
(x)k2
2
Hence
As x∗ satisfies the same conditions as x, using pWm(x∗) = pW −1
(x)∗, we have
m
kpWm(x)k2
2 ≤ 2(ε2 + n−1
m ).
kpWm∪W −1
m
(x)k2
2 = kpWm(x)k2
2 + kpW −1
≤ kpWm(x)k2
2 + kpW −1
≤ 4(ε2 + n−1
m ).
m
m \Wm(x)k2
2
(x)k2
2
This proves claim (i).
Claim (ii) follows immediately from condition (H1).
Let us prove claim (iii): We assume that kx(i)k,kyik ≤ 1 for i = 1, 2. Furthermore, x(i) =
[(x(i)
r k ≤ 1 for
every r ∈ N and i = 1, 2. Suppose first that yj = gj ∈ G \ H for j = 1, 2.
r ), we assume that EA(x(i)
r ) = 0 and kx(i)
r )r], and, replacing x(i)
r − EA(x(i)
r by x(i)
Let ε > 0 be fixed; we will show that
τω(g1x(1)[x(2)g2]∗) ≤ 6ε.
Let us choose m > 0 large enough so that g1Vm ∩ Vmg2 = ∅ and that n−1
g1pVm(x(1)
r )g2 for every r.
m ≤ ε2. Thus,
r ) ⊥ pVm(x(2)
Next set
T = {r ∈ N : khj,mx(i)
r − x(i)
r hj,mk2 ≤ ε, ∀1 ≤ j ≤ nm and i = 1, 2}
which belongs to ω since each x(i) ∈ A′ ∩ M ω. By part (ii), we have for r ∈ T and i = 1, 2,
kx(i)
r − pVm(x(i)
r )k2
2 = kpWm∪W −1
m
(x(i)
r )k2
2 ≤ 4(cid:18)ε2 +
1
nm(cid:19) ≤ 8ε2.
(We used pH(x(i)
r ) = 0, hence x(i)
r = pVm(x(i)
r ) + pWm∪W −1
m
(x(i)
r ).)
7
≤ kx(1)
r − pVm(x(1)
+τ (g1pVm(x(1)
r − pVm(x(1)
+τ ([g1pVm(x(1)
≤ kx(1)
≤ 2√8ε < 6ε
r )k2 + τ (g1pVm(x(1)
2 pVm(x(2)∗
r )g−1
r )k2 + kx(2)
r )][pVm(x(2)
r − pVm(x(2)
r )g2]∗)
))
r
r )k2
r )g2]∗) = 0. Thus
T ⊂ T ′ := {r ∈ N : τ (g1x(1)
r [x(2)
r [x(2)
r g2]∗) ≤ 6ε}
since τ ([g1pVm(x(1)
r )][pVm(x(2)
For the same values of r, we get:
r g2]∗) ≤ τ (g1(x(1)
τ (g1x(1)
r [x(2)
r − pVm(x(1)
r ))g−1
2 x(2)∗
r
) + τ (g1pVm(x(1)
r )g−1
2 (x(2)∗
r )g−1
2 x(2)∗
r − pVm(x(2)∗
r
r
)
)))
and T ′ ∈ ω, hence τω(g1x(1)
Next, using linearity, τω(y1x(1)[x(2)y2]∗) = 0 for all y1, y2 ∈ M with finite support and such
that EA(y1) = EA(y2) = 0, and using density and the same kind of arguments as above, we
get τω(y1x(1)[x(2)y2]∗) = 0 for arbitrary y1, y2 ∈ M ⊖ A.
r g2]∗) ≤ 6ε.
Finally, the equality
ky1x(1) − x(2)y2k2
ω,2 = ky1x(1)k2
ω,2 + kx(2)y2k2
ω,2
comes from y1x(1) ⊥ x(2)y2.
3 Examples
(cid:3)
Before discussing our first family of examples, we need to recall some facts on length-functions
on groups taken from [3]; a length-function on a group Γ is a map ℓ : Γ → R+ satisfying:
(i) ℓ(gh) ≤ ℓ(g) + ℓ(h) for all g, h ∈ Γ;
(ii) ℓ(g−1) = ℓ(g) for every g ∈ Γ;
(iii) ℓ(e) = 0, where e denotes the identity in Γ.
Typical and important examples of length-functions are provided by word length-functions in
finitely generated groups: if Γ is a finitely generated group and if S is a finite, symmetric set
of generators of Γ, then the associated word length-function is defined by
ℓS(g) = min{n ∈ N : g = s1 · · · sn, si ∈ S}.
If S′ is another finite generating set then ℓS and ℓS ′ are equivalent in the sense that there
exist positive numbers a, a′ such that
ℓS ′(g) ≤ aℓS(g) and ℓS(g) ≤ a′ℓS ′(g)
for every g ∈ Γ. When the generating set S is fixed, one often write g instead of ℓS(g).
Let now G = H1∗Z H2 be an amalgamated product where H1 and H2 are finitely generated
groups, H1 is infinite and abelian, Z is a common finite subgroup of H1 and H2, Z 6= H2, and
we assume that G is an ICC group. We choose sets of representatives R1 ∋ e and R2 ∋ e of
left Z-cosets in H1 and H2 respectively, and, because Z is a finite group, we choose length-
functions ℓ1 and ℓ2 on H1 and H2 respectively with the following properties (cf [3], Section
2.2):
8
(a) ℓ1 and ℓ2 take integer values and are equivalent to the word length-functions on H1 and
H2 respectively;
(b) for all z, w ∈ Z, j = 1, 2 and all h ∈ Hj, one has ℓj(zhw) = ℓj(h);
(c) {h ∈ Hj : ℓj(h) = 0} = Z for j = 1, 2.
We set hereafter h = ℓj(h) for j = 1, 2 and h ∈ Hj and we observe that, for every h ∈ Hj, if
h = rz denotes its decomposition with r ∈ Rj and z ∈ Z, then h = r. We recall that every
g ∈ G has a unique normal form
g = r1 · · · rnz
with n ≥ 0, z ∈ Z, and, if n > 0, then rj ∈ Rij \ {e} and ij 6= ij+1 for every j = 1, . . . , n − 1.
For such a g, put
this defines a length-function on G that is equivalent to the word length-function.
g = r1 + · · · + rn;
Thus we define for every m ≥ 1 :
Wm = (H2 \ Z) ∪ {g = r1 · · · rnz : n ≥ 2, r1 ∈ R1,r1 < m}.
Let us check that Wm and Vm = (Wm ∪ W −1
m )c satisfy conditions (H1) to (H3) of the first
section. Indeed, (H1) is obviously satisfied. For (H2), notice first that the normal form of
every g ∈ Vm is of the following type:
(∗) g = g1g2 · · · gkz
where k ≥ 3, g1, gk ∈ R1, g1,gk ≥ m and z ∈ Z. Thus, if γ, γ′ ∈ (H1 ∗Z H2) \ H1 are fixed
and if n > 0 is such that γ,γ′ < n then for every m > 2n and all g, g′ ∈ Vm, the element
γg cannot start as in (∗) and g′γ′ cannot end as in (∗) either. Hence they cannot be equal.
Finally, let us check that (H3) holds true. Since H1 is infinite, abelian and finitely gener-
ated, it follows from the structure of such groups that one can choose an element h ∈ H1 \ Z
of infinite order. Moreover, if i 6= j are integers, hi and hj belong to different cosets mod Z
(otherwise Z would contain an element of infinite order). Hence we can assume that hj ∈ R1
for every j ∈ Z. One also has limj→∞ hj = ∞. Let jm > 0 be large enough so that hj > 2m
for every j > jm; then one has hjWmh−j ∩ Wm = ∅ for all such j's. This proves that the
sequence (Wm) satisfies condition (H2).
Thus we get, using the fact that the pair H1 < G satisfies also condition (ST) from
Proposition 4.1 of [4]:
Corollary 6 Let G = H1 ∗Z H2 be an amalgamated product as above. If L(H1) ⊂ N ⊂ L(G)
is an intermediate von Neumann algebra, there exists partition of the unity (e)k≥0 in the center
of N such that L(H1)e0 = Ne0, and, for each k such that ek 6= 0, Nek is a full factor. In
particular, L(H1) is strongly mixing and maximal injective in L(G).
Similarly, let G = K ∗ L be a free product group such that K ≥ 2 and L contains an
element β of order at least 3. Let α be some non-trivial element of K and set H = hαβi.
Then, by Corollary 4.5 of [4], the pair H < G satisfies condition (ST) and it is easy to see
that it satisfies also conditions (H1) to (H3) of Section 1. Thus we get:
9
Corollary 7 With H < G = K ∗ L as above, and let L(H) ⊂ N ⊂ L(G) is an intermediate
von Neumann algebra. Then there exists partition of the unity (e)k≥0 in the center of N such
that L(H)e0 = Ne0, and, for each k such that ek 6= 0, Nek is a full factor. In particular,
L(H) is strongly mixing and maximal injective in L(G).
The following proposition is straightforward.
Proposition 8 Let G1 be a countable ICC group and let H < G1 be an infinite abelian
subgroup. Assume that G1 \ H contains a sequence (Wm) of subsets which satisfies conditions
(H1), (H2) and (H3) of Section 1. Let G2 be an arbitrary, at most countable, non-trivial
group and let G = G1 ∗ G2 be the corresponding free product. For every m > 0, let W ′
m be the
set of reduced words w = g1 · · · gn ∈ G1 ∗ G2 \ H such that either g1 ∈ Wm or w = hkg2 · · · gn
with 0 ≤ k < m and g2 ∈ G2 \ {e}. Then the sequence (W ′
m) satisfies conditions (H1), (H2)
and (H3).
As a consequence of Proposition 3.7 of [4], if L(H) is strongly mixing in L(G), then it is
also strongly mixing in the free product factor L(G1 ∗ G2) = L(G1) ∗ L(G2), thus we get:
Corollary 9 If H < G1 and G2 are as in Proposition 8 and if L(H) is strongly mixing in
L(G1), then L(H) is maximal injective in L(G1) and in L(G1 ∗ G2).
References
[1] J. Cameron, J. Fang, M. Ravichandran, and S. White. The radial masa in a free group
factor is maximal injective. ArXive:math.OA/0810.3906 v1, 2008.
[2] J. Cameron, S. Fang, and K. Mukherjee. Mixing subalgebras of finite von Neumann
algebras. arXiv: math. OA/1001.1069 v1, 2009.
[3] P. Jolissaint. Rapidly decreasing functions in reduced C ∗-algebras of groups. Trans. Amer.
Math. Soc., 317:167 -- 196, 1990.
[4] P. Jolissaint and Y. Stalder. Strongly singular MASAs and mixing actions in finite von
Neumann algebras. Ergod. Th. & Dynam. Sys., 28:1861 -- 1878, 2008.
[5] S. Popa. Strong rigidity of II1 factors arising from malleable actions of w-rigid groups, I.
arXiv: math. OA/0305306 v13, 2005.
[6] S. Popa. Maximal injective subalgebras in factors associated with free groups. Adv. Math.,
50:27 -- 48, 1983.
[7] G. Robertson, A. M. Sinclair, and R. R. Smith. Strong singularity for subalgebras of finite
factors. Int. J. Math., 14:235 -- 258, 2003.
[8] A. Sinclair and R. Smith. Finite von Neumann Algebras and Masas. Cambridge University
Press, Cambridge, 2008.
10
[9] A. M. Sinclair, R. R. Smith, S. A. White, and A. Wiggins. Strong singularity of singular
masas in II1 factors. Illinois J. Math., 51:1077 -- 1084, 2007.
Université de Neuchâtel,
Institut de Mathémathiques,
Emile-Argand 11
Case postale 158
CH-2009 Neuchâtel, Switzerland
[email protected]
11
|
1108.3208 | 2 | 1108 | 2011-10-16T16:36:40 | Spectral synthesis for operator space projective tensor product of $C^*$-algebras | [
"math.OA"
] | We study the spectral synthesis for the Banach *-algebra $A\oop B$, the operator space projective tensor product of $C^*$-algebras $A$ and $B$. It is shown that if $A$ or $B$ has finitely many closed ideals, then $A\oop B$ obeys spectral synthesis. The Banach algebra $A \oop A$ with the reverse involution is also studied. | math.OA | math |
SPECTRAL SYNTHESIS FOR OPERATOR SPACE
PROJECTIVE TENSOR PRODUCT OF C ∗-ALGEBRAS
RANJANA JAIN AND AJAY KUMAR
Abstract. We study the spectral synthesis for the Banach ∗-algebra A b⊗B,
the operator space projective tensor product of C ∗-algebras A and B. It is
shown that if A or B has finitely many closed ideals, then A b⊗B obeys
spectral synthesis. The Banach algebra A b⊗A with the reverse involution is
also studied.
1. Introduction and notations
For operator spaces V and W , and u ∈ V ⊗ W , the operator space projective
tensor norm is defined as
kuk∧ = inf{kαkkvkkwkkβk : u = α(v ⊗ w)β},
where α ∈ M1,pq, β ∈ Mpq,1, v ∈ Mp(V ) and w ∈ Mq(W ), p, q ∈ N being
arbitrary, and v ⊗ w = (vij ⊗ wkl)(i,k),(j,l) ∈ Mpq(V ⊗ W ). The operator space
The algebraic tensor product V ⊗W is complete with respect to k·k∧-norm if and
only if either V or W is finite dimensional. To see this, if V is finite dimensional,
projective tensor product V b⊗W is the completion of V ⊗ W under k · k∧-norm.
then as a Banach space it is isomorphic to Cn, thus V b⊗W is Banach space
for all n ∈ N. The sequence (un) in V ⊗ W defined as un =Pn
It is known that for C ∗-algebras A and B, Ab⊗B is a Banach ∗-algebra under
isomorphic to the direct sum of n-copies of W , which is complete. Also, if V
and W are both infinite dimensional, then one can choose two sequences {en}
and {fn} of linearly independent vectors in V and W such that kenk = kfnk = 1
i=1 2−iei ⊗ fi is a
Cauchy sequence with respect to k · k∧-norm, but is not convergent in V ⊗ W .
natural involution ([14]).
The notion of spectral synthesis has been studied extensively for commu-
tative and unital Banach algebras, for L1-group algebras and for Banach ∗-
algebras [20, 6, 7, 13]. Spectral synthesis for Banach space projective tensor
product of commutative Banach algebras and for the Haagerup tensor product
of C ∗-algebras has also been explored([13, 8, 1, 7]). Roughly speaking spectral
synthesis holds for a Banach ∗-algebra X if every closed ideal of X is the in-
tersection of primitive ideals containing it. Spectral synthesis for Banach space
projective tensor product of commutative Banach algebras has already been
explored([13]). For commutative C ∗-algebras A and B, the natural contractive
homomorphism of Ab⊗B into A ⊗h B is an isomorphism whose inverse has norm
2000 Mathematics Subject Classification. 46L06, 46L07, 47L25, 43A45.
Key words and phrases. C ∗-algebras, operator space projective tensor norm, spectral syn-
thesis, hull-kernel topology.
1
2
R. JAIN AND A. KUMAR
equal to Grothendieck constant. Thus, for countable locally compact Hausdorff
([8, 11.2.1],[13]).
spaces X and Y , C0(X)b⊗C0(Y ) has spectral synthesis. However, for cantor set
or any infinite compact group D, C(D)b⊗C(D) does not have spectral synthesis
In Section 2, we define the concept of spectral ideals in Ab⊗B, and prove
that the Banach ∗- algebra Ab⊗B has spectral synthesis if and only if each
closed ideal of Ab⊗B is spectral. This result is then used to produce plenty
of spectral ideals in Ab⊗B. We also discuss few cases where Ab⊗B obeys spec-
ideals, then Ab⊗B has spectral synthesis. Thus, the Banach ∗-algebras like
C0(X)b⊗B(H), B(H)b⊗K(H) and B(H)b⊗B(H) all obey spectral synthesis, X
separable Hilbert space. In section 3, the algebra Ab⊗B with the reverse involu-
tion is discussed. It is shown that with this involution the algebra is symmetric
and ∗-semisimple only in the trivial cases.
being a locally compact topological space and H being an infinite dimensional
tral synthesis. In particular, we prove that if A or B has finitely many closed
For a Banach algebra X, we denote the set of closed (two-sided) ideals of
X by Id(X), the set of proper closed ideals of X by Id′(X) and the set of all
prime ideals by P rime(X). If X is a Banach ∗-algebra, then P rim(X) stands
for the set of primitive ideals of X, that is, the set of all kernels of irreducible
∗-representations of X on Hilbert space. There is a topology τw on Id(X) which
is generated by the sub-basic open sets of the form
ZJ := {I ∈ Id(X) : I + J}, J ∈ Id(X).
We throughout use the notation qJ for the quotient map qJ : A → A/J. Recall
that, for closed ideals M and N of C ∗-algebras A and B, the map qM ⊗ qN :
and qM ⊗min qN : A ⊗min B → A/M ⊗min B/N .
A ⊗ B → A/M ⊗ B/N extends to quotient maps qMb⊗qN : Ab⊗B → A/Mb⊗B/N
Let A and B be C ∗-algebras. Define a map Φ : Id(A) × Id(B) → Id(Ab⊗B)
as
Φ(M, N ) = Ab⊗N + Mb⊗B.
The map Φ is well defined by [12, Proposition 3.2].
topological properties listed as below:
Proposition 1.1. Let A and B be C ∗-algebras and Φ : Id(A) × Id(B) →
It satisfies many nice
Id(Ab⊗B) be defined as above. Then
(i) Φ maps P rime(A) × P rime(B) onto P rime(Ab⊗B).
(ii) Φ maps P rim(A)×P rim(B) into P rim(Ab⊗B). If A and B are separable,
then Φ maps P rim(A) × P rim(B) onto P rim(Ab⊗B).
(iii) Φ maps Id′(A) × Id′(B) into Id′(Ab⊗B) injectively.
(iv) The mapping Φ is τw-continuous.
(v) The restriction of Φ to Id′(A)×Id′(B) is a homeomorphism onto its image
(vi) The restriction of Φ to P rime(A) × P rime(B) is a homeomorphism onto
in Id′(Ab⊗B).
P rime(Ab⊗B).
Proof: (i) and (ii) follow from Theorems 3.1 and 3.2 of [11], respectively.
For (iii), note that, for proper closed ideals M and N of A and B, the
isomorphism of A/Mb⊗B/N onto (Ab⊗B)/(Ab⊗N + Mb⊗B) ([11, Lemma 2.2])
SPECTRAL SYNTHESIS FOR O. S. PROJECTIVE TENSOR PRODUCT
3
M1 ⊆ M2, N1 ⊆ N2. To see this, consider an m ∈ M1, so that for an arbitrary
assures that Ab⊗N + Mb⊗B is also proper in Ab⊗B. Further, for M1, M2 ∈
Id′(A), N1, N2 ∈ Id′(B), Ab⊗N1 + M1b⊗B ⊆ Ab⊗N2 + M2b⊗B if and only if
b ∈ B, m ⊗ b ∈ ker(qM1b⊗qN1) ⊆ ker(qM2b⊗qN2), giving qM2(m) = 0, that is
m ∈ M2, and similarly N1 ⊆ N2. Thus, Φ is injective.
(iv)-(vi) can be proved exactly on the same lines of their counterparts in
Haagerup tensor product as discussed in Lemma 1.4 and Theorem 1.5 of [2]. (cid:3)
Throughout this paper A and B represent C ∗-algebras, until otherwise spec-
ified.
2. Spectral synthesis
We first give the standard definition of spectral synthesis for a Banach ∗-
algebra that appear in the literature. Let X be a Banach ∗-algebra. For each
E ⊆ P rim(X), there associates a closed ideal kernel of E defined as
Also, for each M ⊆ X, hull of M is defined as
k(E) := ∩P ∈EP.
hX (M ) := {P ∈ P rim(X) : P ⊇ M }.
We shall denote the hull of M by h(M ), when there is no confusion with X.
Equip P rim(X) with the hull-kernel topology (or, hk-topology), where for every
E ⊆ P rim(X), its closure is E = h(k(E)). Similarly, one can talk about the
hk-topology on P rime(X). Note that, if E ⊆ P rime(X), then the relative
τw-topology on E coincides with the hull-kernel topology.
Definition 2.1. A closed subset E of P rim(X) is called spectral if k(E) is the
only closed ideal in X with hull equal to E. A Banach ∗-algebra X is said to
have spectral synthesis if every closed subset of P rim(X) is spectral.
A closed ideal of Banach ∗-algebra X is said to be semisimple if it is an
intersection of all the primitive ideals of X containing it.
Proposition 2.2. Let X be a Banach ∗-algebra having Wiener property. Then
X has spectral synthesis if and only if for every J ∈ Id(X), J = k(h(J)), or,
in other words, every closed ideal of X is semisimple.
Proof: Let us consider a proper closed ideal J of X. Since X has Wiener
property, there exists an irreducible ∗-representation, say π, of X which annihi-
lates J, that is, J ⊆ ker π, so that E = h(J) is non empty. We claim that E is
closed in the hk-topology. Let Q ∈ E = h(k(E)), then k(E) ⊆ Q. Since J ⊆ P
for all P ∈ E we have J ⊆ k(E) ⊆ Q, so that Q ∈ E. which gives that E is
closed. Since X obeys spectral synthesis, and E = h(J), we have J = k(E),
that is, J is the intersection of primitive ideals containing it. Also, note that
since X has Wiener property, the empty set φ is spectral, so that X = k(h(X)).
(cid:3)
Converse follows easily from the given condition.
Corollary 2.3. Let X be a Banach ∗-algebra having Wiener property. Then
X has spectral synthesis if and only if there is a one-one correspondence between
the closed ideals of X and the τw-open subsets of P rim(X) (or, P rime(X)).
4
R. JAIN AND A. KUMAR
Proof: Let X have spectral synthesis. For J ∈ Id(X), recall ZJ := {P ∈
P rim(X) : P + J} = P rim(X) \ h(J) is an open subset of P rim(X) under
the relative τw-topology, so that we have a well defined correspondence J 7→ ZJ
between the closed ideals of X and τw-open subsets of P rim(X). For K, L ∈
Id(X), it is clear from Proposition 2.2 that K = k(h(K)), and L = k(h(L)).
Thus, it can be easily seen that
K ⊆ L if and only if ZK ⊆ ZL,
which shows that the correspondence in one-one. Now consider a τw-open subset
G of P rim(X), and set J := k(P rim(X) \ G). Since P rim(X) \ G is closed
under the hull-kernel topology,
ZJ = P rim(X) \ h(k(P rim(X) \ G)) = P rim(X) \ (P rim(X) \ G) = G,
which proves that this correspondence is surjective.
Conversely, for every closed ideal I of X, since h(I) = h(k(h(I)), we have
ZI = Zk(h(I)). Using the given condition, this gives I = k(h(I)). Result now
follows from Proposition 2.2.
(cid:3)
Remark 2.4. For C ∗-algebras A and B, since Ab⊗B has Wiener property ([12,
Theorem 4.1]), Ab⊗B has spectral synthesis if and only if every closed ideal J
of Ab⊗B is semisimple. In particular, if Ab⊗B has spectral synthesis then every
closed ideal J of Ab⊗B is the intersection of prime ideals containing J.
The next two results connect the spectral synthesis of a Banach ∗-algebra
with that of its ideal and the corresponding quotient algebra. The first result
follows on the similar lines as that in [7, Proposition 1.16]. However, we present
here a proof for the sake of completion.
Proposition 2.5. Let X be a Banach ∗-algebra with Wiener property, and J be
a closed ∗-ideal of X having bounded approximate identity and Wiener property.
If J and X/J both have spectral synthesis (as Banach ∗-algebras), then X has
spectral synthesis.
Proof: By Corollary 2.3, it is sufficient to show that for I, K ∈ Id(X), I = K,
whenever hX(I) = hX (K). Note that, since X has Wiener property, X/J
also has Wiener property, so by Proposition 2.2, every closed ideal of J and
X/J is semisimple. For P ∈ hX/J (qJ (I)), P = ker π with π(qJ (I)) = {0},
π : X/J → B(H) being an irreducible ∗-representation. Then π0 := π ◦ qJ is an
irreducible ∗-representation of X on H with π0(I) = 0. Since hX (I) = hX(K),
π0 ∈ hX (K), which further gives P ∈ hX/J (qJ (K)). Thus, hX/J (qJ (I)) =
hX/J (qJ (K)). Since J has an approximate identity, by [5, Proposition 2.4],
I + J and K + J are closed in X, so that qJ (I) and qJ (K) are closed ideals of
X/J. Since X/J obeys spectral synthesis, by Proposition 2.2, qJ (I) = qJ (K).
Further, for any closed ideal L of J, it is routine to check that there is a one-
one correspondence between the sets {P ∈ hX(L) : J * P } and hJ (L) via
P 7→ P ∩ J. Xlso, hX(I ∩ J) = hX (I) ∪ hX (J) = hX(K) ∪ hX (J) = hX(K ∩ J).
Thus, it can be easily seen that hJ (I ∩ J) = hJ (K ∩ J). Since J has spectral
synthesis, this gives, I ∩ J = K ∩ J.
SPECTRAL SYNTHESIS FOR O. S. PROJECTIVE TENSOR PRODUCT
5
Now, consider x ∈ I, then qJ (x) = qJ (y) for some y ∈ K, so that a := x− y ∈
J. Let Ja be the smallest closed ideal of J containing a. Since J obeys spectral
synthesis, Ja = ∩{P ∈ P rim(J) : Ja ⊆ P }. Clearly JaJ ⊆ Ja. Now consider
P ∈ P rim(J) such that JaJ ⊆ P . Since P is prime being primitive, this gives
a ∈ P which shows that Ja ⊆ P . Thus JaJ = ∩{P ∈ P rim(J) : JaJ ⊆ P } =
∩{P ∈ P rim(J) : Ja ⊆ P } = Ja. So
x − y ∈ J(x − y)J ⊆ JxJ − JyJ
⊆ JIJ − JKJ
⊆ I ∩ J − K ∩ J
= K ∩ J.
So, x = y − (y − x) ∈ K + (K ∩ J) = K, which gives I ⊆ K. Similarly, K ⊆ I,
which proves the claim.
(cid:3)
In fact, the converse of the above statement is also true as presented below.
Proposition 2.6. Let X be a Banach ∗-algebra with a closed ∗-ideal J such
that X and J both possess Wiener property. If X obeys spectral synthesis, then
so does J and X/J.
Proof: By Proposition 2.2, it is enough to check that for a closed ideal L of J,
L = k(hJ (L)). Since every closed ideal of X is semisimple, and every primitive
ideal is prime, from [7, Proposition 1.14], L is also a closed ideal of X, so that
by Proposition 2.2, L = k(hX (L)).
It can be easily verified that there is a
one-one correspondence between the sets {P ∈ hX (L) : J * P } and hJ (L) via
P 7→ P ∩ J. So, we have
(P ∩ J)
L = L ∩ J = \P ∈hX (L)
= (cid:18) \P ∈hX (L)
= (cid:0) \P ′∈hJ (L)
J*P
(P ∩ J)(cid:19)
(P ∩ J)(cid:19) ∩(cid:18) \P ∈hX (L)
P ′(cid:1) ∩ J
J ⊆P
Thus, J obeys spectral synthesis.
= k(hJ (L))
Next, consider a closed ideal K of X/J. Since X/J has Wiener property, it is
enough to check that K ⊇ k(hX/J (K)). Consider an element x ∈ k(hX/J (K)),
where x = y + J ∈ X/J. Note that K = I/J for some closed ideal I of
X containing J. Using the one-one correspondence between P rim(X/J) and
{P ∈ P rim(X) : J ⊆ P }, one can check that y ∈ k(hX (I)). Since X has
spectral synthesis, I = k(hX (I)), so that y ∈ I, which shows that x ∈ K.
Hence the result.
(cid:3)
We are now prepared to discuss spectral synthesis for operator space projec-
tive tensor product Ab⊗B of C ∗-algebras A and B. Allen, Sinclair and Smith,
in [1], defined the concept of spectral synthesis for the Haagerup tensor product
6
R. JAIN AND A. KUMAR
of C ∗-algebras in a somewhat different flavor. In the same spirit, using the ter-
It is known that for any C ∗-algebras A and B, the canonical ∗-homomorphism
minologies of [1], we give another definition for the spectral synthesis of Ab⊗B.
i : Ab⊗B → A ⊗min B is injective ([10, Corollary 1]), so that we can regard Ab⊗B
as a ∗-subalgebra of A ⊗min B. Consider a closed ideal J of Ab⊗B and let Jmin
be the closure of i(J) in A ⊗min B, in other words, Jmin is the min-closure of
J in A ⊗min B. Now we associate two closed ideals, namely the upper and the
lower ideals, with J as:
Jl = closure of span of all elementary tensors of J in Ab⊗B,
J u = Jmin ∩ (Ab⊗B).
Clearly Jl ⊆ J ⊆ J u for any closed ideal J of Ab⊗B.
Definition 2.7. A closed ideal J of Ab⊗B is said to be spectral if Jl = J = J u.
The main aim of this section is to show that Ab⊗B has spectral synthesis if
and only if its every closed ideal is spectral. We first characterize the upper
ideals in terms of primitive ideals.
Lemma 2.8. For closed ideals M and N of A and B,
ker(qMb⊗qN ) = ker(qM ⊗min qN ) ∩ Ab⊗B.
zk∧ = 0, then
Proof: For z ∈ Ab⊗B, let {zn} be a sequence in A ⊗ B such that limn kzn −
which shows that the sequence {(qMb⊗qN )(zn)} is convergent to (qMb⊗qN )(z) in
k(qMb⊗qN )(zn) − (qMb⊗qN )(z)kmin ≤ kqMb⊗qN kkzn − zk∧,
A ⊗min B. Also, kzn − zkmin ≤ kzn − zk∧, so that limn kzn − zkmin = 0, which
further gives
(qM ⊗min qN )(zn) min−→ (qM ⊗min qN )(z).
proving the given relation.
Since both the mappings qMb⊗qN and qM ⊗min qN agree on A⊗ B, by continuity,
we have (qM ⊗min qN )(z) = (qMb⊗qN )(z), and this is true for all z ∈ Ab⊗B,
Proposition 2.9. For a closed ideal J of Ab⊗B, J = J u if and only if J is
semisimple.
(cid:3)
Proof: Let us first assume that J = J u. Since, in a C ∗-algebra every closed
ideal is semisimple, Jmin = ∩{ker πα : Jmin ⊆ ker πα}, where each πα is an
irreducible ∗-representation of A ⊗min B on some Hilbert space. Set πα :=
is larger than {ker πα : πα = πα ◦ i}, it is easy to check that J is actually the
Using some routine calculations, and the fact that J = J u one can prove that
πα ◦ i, then each πα is an irreducible ∗-representation of Ab⊗B annihilating J.
J = ∩ ker πα. Note that, although the collection {P ∈ P rim(Ab⊗B) : J ⊆ P }
intersection of all the primitive ideals of Ab⊗B containing J.
Conversely, let J = ∩J ⊆PαPα, Pα being primitive ideals of Ab⊗B. Let, if
ideals M and N in A and B, respectively, such that Pα = Ab⊗N + Mb⊗B.
possible, there exist an element x ∈ J u such that x /∈ J. Then x /∈ Pα for
some α. Since Pα is primitive, by [11, Theorem 3.2], there exist closed (prime)
SPECTRAL SYNTHESIS FOR O. S. PROJECTIVE TENSOR PRODUCT
7
Now, consider the bounded homomorphisms qMb⊗qN : Ab⊗B → A/Mb⊗B/N ,
and qM ⊗min qN : A ⊗min B → A/M ⊗min B/N with ker(qMb⊗qN ) = Pα ([12,
Proposition 3.5]. By Lemma 2.8, x /∈ ker(qM ⊗min qN ), which by Hahn Banach
Theorem gives a φ ∈ (A ⊗min B)∗ such that φ(x) 6= 0, and φ(ker(qM ⊗min
qN )) = {0}. The relation J ⊆ Pα ⊆ ker(qM ⊗min qN ) gives Jmin ⊆ ker(qM ⊗min
qN ), which further shows that φ(Jmin) = 0. Thus x /∈ Jmin, which gives a
contradiction to the fact that x ∈ J u. Hence the result.
(cid:3)
Using Propositions 2.2 and 2.9, we have a following characterization for spec-
tral synthesis in terms of upper ideals.
to prove the same. We first need an elementary result.
Lemma 2.11. Let Ji and Ki be closed ideals of C ∗-algebras Ai, i = 1, 2. Then
Theorem 2.10. The Banach ∗-algebra Ab⊗B has spectral synthesis if and only
if J = J u, for every closed ideal J of Ab⊗B.
We now prove that the Banach ∗-algebra Ab⊗B has spectral synthesis if and
only if every closed ideal of Ab⊗B is spectral. We borrow some ideas from [15]
J1b⊗J2 ⊆ A1b⊗K2 + K1b⊗A2 if and only if either J1 ⊆ K1 or J2 ⊆ K2.
Theorem 2.12. For C ∗-algebras A and B, the Banach ∗-algebra Ab⊗B has
spectral synthesis if and only if every closed ideal of Ab⊗B is spectral.
Proof: We just need to prove that for every closed ideal J of Ab⊗B, J = Jl,
if Ab⊗B has spectral synthesis. Using Corollary 2.3, it is sufficient to show that
ZJ ⊆ ZJl, where ZJ := {P ∈ P rime(Ab⊗B) : P + J}. Set X := P rime(Ab⊗B)
and consider an element P of ZJ . Since ZJ is an open subset of X and Φ :
P rime(A) × P rime(B) → X is continuous, there exist open subsets U1, U2 of
P rime(A) and P rime(B) such that Φ(U1 × U2) ⊆ ZJ and P ∈ Φ(U1 × U2).
Let J1 ∈ Id(A), J2 ∈ Id(B) be the corresponding closed ideals such that
Ui = ZJi, i = 1, 2. We claim that ZJ1
b⊗J2 = Φ(U1 × U2) = Φ(ZJ1 × ZJ2).
For any Q ∈ ZJ1
(Proposition 1.1), there exists Q1 ∈ P rime(A), Q2 ∈ P rime(B) such that
b⊗J2. So,
Note that Φ(K1, K2) ∈ X, thus by the definition, Φ(K1, K2) ∈ ZJ1
ZJ1
This implies that Qi ∈ ZJi = Ui, so that Q = Φ(Q1, Q2) ∈ Φ(U1 × U2). Thus,
ZJ1
b⊗J2 ⊆ Φ(U1×U2). For the other containment, consider Φ(K1, K2) ∈ Φ(ZJ1 ×
b⊗J2, by definition, J1b⊗J2 * Q. Since Q ∈ X, and Φ is onto
Ab⊗Q2 + Q1b⊗B = Φ(Q1, Q2) = Q. By Lemma 2.11, J1 * Q1 and J2 * Q2.
ZJ2). Since Ki ∈ ZJi, we have Ji * Ki, so by Lemma 2.11, J1b⊗J2 * Φ(K1, K2).
b⊗J2 ⊆ ZJ , which further gives, J1b⊗J2 ⊆ J. But, the definition of Jl says that
J1b⊗J2 ⊆ Jl. This means that ZJ1
Remark 2.13. In other words, if Ab⊗B obeys spectral synthesis, then every
closed ideal J of Ab⊗B is the closure of the sum of all product ideals J1b⊗J2 ⊆ J,
The Banach ∗-algebra Ab⊗B contains plenty of spectral ideals as demon-
b⊗J2,
this gives P ∈ ZJl. Thus ZJ ⊆ ZJl, which proves that J ⊆ Jl, and hence the
result.
(cid:3)
b⊗J2 ⊆ ZJl. Since, P ∈ Φ(U1 × U2) = ZJ1
strated in the following and some later examples.
where J1 ∈ Id(A), J2 ∈ Id(B).
8
R. JAIN AND A. KUMAR
Proposition 2.14. For I ∈ Id(A) and J ∈ Id(B), the closed ideal Ab⊗J + Ib⊗B
of Ab⊗B is spectral. In particular, every closed maximal ideal, primitive ideal
and prime ideal of Ab⊗B is spectral.
Proof: Set K := Ab⊗J +Ib⊗B = ker(qIb⊗qJ ), then it is clear from the definition
that K = Kl. Consider an element u ∈ K u. Let, if possible, u /∈ K, then by
Lemma 2.8, u /∈ ker(qI ⊗min qJ ). Now, K ⊆ ker(qI ⊗min qJ ) implies Kmin ⊆
ker(qI ⊗min qJ ), giving u /∈ Kmin, a contradiction. Thus K is spectral. Rest
follows from the fact that every maximal, primitive and prime ideal can be
expressed as an ideal of this form ([12, Theorem 3.10], [11, Theorem 3.1, 3.2]).
(cid:3)
Next, we prepare the ingredients to prove that for an infinite dimensional
need some elementary results regarding the lower and upper ideals of a closed
ideal.
separable Hilbert space H, B(H)b⊗B(H) obeys spectral synthesis. We first
Proposition 2.15. For closed ideals J and K in Ab⊗B, we have:
(a) Jl ⊆ Kl and J u ⊆ K u, if J ⊆ K;
(b) (JK)l = JlKl = Jl ∩ Kl = (J ∩ K)l, if Jl or Kl has a bounded approximate
identity;
(c) (J ∩ K)u ⊆ J u ∩ K u, with equality if J = J u, K = K u.
Proof: (a) is trivial. For (b), we first show that Jl ∩ Kl ⊆ JlKl. Let x ∈
Jl ∩ Kl and assume that Jl has bounded approximate identity. By Cohen's
Factorization Theorem, there exist y, z ∈ Jl such that x = yz and z belongs
to the closed left ideal generated by x in Jl. Clearly, z ∈ Jl ∩ Kl, so that
x ∈ JlKl. Thus, Jl ∩ Kl ⊆ JlKl. Now, for an elementary tensor x in J ∩ K,
i=1 xi ⊗yi ∈ Jl and
j=1 zj ⊗ wj ∈ Kl, clearly ab ∈ (JK)l, being an elementary tensor of JK.
Since Jl and Kl are both generated by elementary tensors, routine calculations
show that JlKl ⊆ (JK)l. Thus, we have
clearly x ∈ Jl ∩Kl, giving (J ∩K)l ⊆ Jl ∩Kl. Also, for a =Pn
b =Pm
(J ∩ K)l ⊆ Jl ∩ Kl ⊆ JlKl ⊆ (JK)l ⊆ (J ∩ K)l,
which gives the required equality. For (c), using the fact that (J ∩ K)min ⊆
Jmin ∩ Kmin, we get
(J ∩ K)u ⊆ Jmin ∩ Kmin ∩ Ab⊗B = J u ∩ K u.
Following are some direct consequences of the above proposition.
(cid:3)
having bounded approximate identity. Then I ∩ J is spectral, whenever I and J
are spectral.
Corollary 2.16. If I and J are closed ideals of Ab⊗B with at least one of them
Corollary 2.17. Every product ideal of Ab⊗B is spectral. In particular, for
closed ideals I and J of A and B, Ib⊗J = (I ⊗min J) ∩ Ab⊗B.
Proof: For a product ideal Ib⊗J of Ab⊗B, using [11, Proposition 2.4], we can
write
Ib⊗J = (Ab⊗J) ∩ (Ib⊗B).
SPECTRAL SYNTHESIS FOR O. S. PROJECTIVE TENSOR PRODUCT
9
Also, from [12, Lemma 3.1], Ab⊗J and Ib⊗B both possess bounded approximate
identities. Thus, from Proposition 2.14 and Corollary 2.16, Ib⊗J is spectral.
Clearly,
Ib⊗J = (Ib⊗J)u = (Ib⊗J)min ∩ Ab⊗B = (I ⊗min J) ∩ Ab⊗B.
(cid:3)
spectral synthesis.
product ideal and thus is spectral by Corollary 2.17. Using Theorem 2.12, we
(cid:3)
Corollary 2.18. If either A or B is a simple C ∗-algebra, then Ab⊗B obeys
Proof: Let A be simple. By [12, Theorem 3.8], every closed ideal of Ab⊗B is a
get Ab⊗B obeys spectral synthesis.
In particular, for any C ∗-algebra A, the Banach ∗-algebras Ab⊗C ∗
r (F2), Ab⊗A∞
and Ab⊗K(H) obey spectral synthesis, where C ∗
r (F2) is the C ∗-algebra associ-
ated to the left regular representations of the free group F2 on two generators,
A∞ is the Glimm algebra ([18]) and K(H) is the C ∗-algebra of compact opera-
tors on an infinite dimensional separable Hilbert space H.
Theorem 2.19. For an infinite dimensional separable Hilbert space H, the Ba-
follows from Theorem 2.12.
(cid:3)
Proof: From [12, Theorem 3.11], we know that the only non trivial closed
nach ∗-algebra B(H)b⊗B(H) obeys spectral synthesis.
ideals of B(H)b⊗B(H) are K(H)b⊗K(H), B(H)b⊗K(H), K(H)b⊗B(H) and B(H)b⊗
K(H) + K(H)b⊗B(H). Using Proposition 2.14 and Corollary 2.17, we can see
that all the proper closed ideals of B(H)b⊗B(H) are spectral. The result now
many closed ideals. Then Ab⊗B obeys spectral synthesis.
Proposition 2.20. Let A and B be C ∗-algebras such that A or B has finitely
Proof: Without loss of generality, we may assume that B has finitely many
closed ideals say n, where n ≥ 2. We prove the result by induction on n.
For n = 2, B is simple and the result follows from Corollary 2.18. Let the
result be true for all C ∗-algebras with at most (n − 1) ideals. Let B have
n > 2 closed ideals. Since there are finitely many closed ideals of B, there
exists a minimal (non-trivial) closed ideal, say K, of B, which is clearly simple.
Corollary 2.18, it is clear that J has spectral synthesis. Note that, by [11,
Consider the closed ∗-ideal J := Ab⊗K of X := Ab⊗B. Since K is simple, using
Lemma 2.2(1)], X/J is isomorphic to Ab⊗(B/K) and the latter has spectral
synthesis by induction hypothesis, since B/K has atmost (n − 1) closed ideal.
So, X/J also has spectral synthesis. Moreover, J and X/J both have Wiener
property ([11, Theorem 4.1]), and J has bounded approximate identity ([12,
Lemma 3.1]), the result now follows from Proposition 2.5.
(cid:3)
a separable infinite dimensional Hilbert space.
Thus, for any C ∗-algebra A, Ab⊗B(H) obey spectral synthesis, where H is
In particular, C0(X)b⊗B(H),
B(H)b⊗B(H) and B(H)b⊗K(H) obey spectral synthesis, where X is a locally
compact Hausdorff space. For more examples of C ∗-algebras with finitely many
closed ideals, see [16].
10
R. JAIN AND A. KUMAR
Corollary 2.21. If A and B both have finite number of closed ideals, then
every closed ideal of Ab⊗B is a finite sum of product ideals.
Proof: It follows from Proposition 2.20 and Remark 2.13.
(cid:3)
Remark 2.22. Let A and B be C ∗-algebras. If A is subhomogeneous, then by
[3, Proposition IV.1.4.6], every subhomogeneous C ∗-algebra is bidual type I, so
that A∗∗ is a type I von Neumann algebra, which is then nuclear by Corollary
IV.2.2.10 and Theorem IV.3.1.5 of [3]. Consider the Gelfand-Naimark semi
norm on A ⊗ B defined as
γ(x) = sup{kT (x)k},
where the supremum runs over all the ∗-representations T of A ⊗ B on Hilbert
spaces. Since the k · kγ-norm on A ⊗ B is continuous with respect to '∧′-norm,
it can be extended to Ab⊗B, and thus A ⊗ B is k · kγ-dense in (Ab⊗B, k · kγ).
By [17, Proposition 10.5.20], Ab⊗B is ∗-regular, whenever A ⊗ B is so. Since
A is nuclear, by [9, Corollary 2.7], Ab⊗B is ∗-regular. It is also Hermitian ([11,
Theorem 4.6] and is ∗-semisimple (follows from [11, Theorem 4.1]). Hence, the
definition of spectral synthesis in [6] is equivalent to our definition in this case.
In the case of commutative separable C ∗-algebras the ideals which are not
singly generated fail to be spectral. A similar result also holds true in the non-
commutative situation. The following can be proved exactly on the same lines
of [1, Theorem 6.12].
Proposition 2.23. Let A and B be separable C ∗-algebras, and J be a non-zero
closed ideal of Ab⊗B. Then J is singly generated if it is spectral.
3. Reverse Involution
Let A be a C ∗-algebra. On the Banach algebra A ⊗ A (with usual multi-
plication), define the involution as (a ⊗ b)∗ = b∗ ⊗ a∗ for all a, b ∈ A. Then
it extends to an isometric involution on Ab⊗A and Ab⊗A forms a Banach ∗-
algebra with this involution, which we denote by Ab⊗rA. Regarding the closed
∗-ideals of Ab⊗rA, note that the closed ideals of Ab⊗rA coincide with the ones
in Ab⊗A; however, the closed ∗-ideals differ. We do not know whether a closed
ideal of Ab⊗A is a ∗-ideal or not, but in Ab⊗rA a closed ideal need not be a
∗-ideal. For example, in the space B(H)b⊗rB(H), the closed ideals K(H)b⊗B(H)
and B(H)b⊗K(H) are not ∗-ideals. In fact, it has only two non-trivial closed
∗-ideals, namely K(H)b⊗rK(H) and B(H)b⊗K(H) + K(H)b⊗B(H).
We know that with natural involution Ab⊗A has a faithful ∗-representation
is not the case with Ab⊗rA.
Proposition 3.1. Let A be a unital C ∗-algebra. Then, Ab⊗rA has a faithful
Proof: Let π be a faithful ∗-representation of Ab⊗rA on a Hilbert space H.
and is always ∗-semisimple for any C ∗-algebra A. However, we show that this
∗-representation if and only if A = CI, I being the unity of A.
Define π1(a) := π(1 ⊗ a) and π2(a) := π(a ⊗ 1) for all a ∈ A. Then π1 and π2
SPECTRAL SYNTHESIS FOR O. S. PROJECTIVE TENSOR PRODUCT
11
are both bounded representations of A on B(H), with π(a ⊗ b) = π1(a)π2(b) =
π2(b)π1(a) for all a, b ∈ A. Also
(1)
π1(a∗) = π(1 ⊗ a∗) = π((a ⊗ 1)∗) = (π(a ⊗ 1))∗ = π2(a)∗
for all a ∈ A. It is known that an element h ∈ A is self adjoint if and only if
k exp ithk = 1 for all t ∈ R. For a self adjoint element h ∈ A, using the facts
that π is contractive and that k · k∧-norm is a cross norm, we have
k exp itπ1(h)k = kπ(exp it(h ⊗ 1))k
≤ k exp it(h ⊗ 1)k∧
= lim
= lim
= lim
m (cid:13)(cid:13)(cid:13)
mXn=1
m (cid:13)(cid:13)(cid:13)(cid:16) mXn=1
m (cid:13)(cid:13)(cid:13)
mXn=1
intn(hn ⊗ 1)
intnhn
n!
(cid:13)(cid:13)(cid:13)∧
n! (cid:17) ⊗ 1(cid:13)(cid:13)(cid:13)∧
n! (cid:13)(cid:13)(cid:13)
intnhn
= k exp ithk = 1,
and this is true for all t ∈ R. Thus, k exp itπ1(h)k = 1 for all t ∈ R, which shows
that π1(h) is a self adjoint element of B(H). This, combined with equation (1),
gives π1(h) = π2(h), that is π(1⊗h) = π(h⊗1). Since π is faithful, 1⊗h = h⊗1.
So, for any φ ∈ A∗, φ(1)h = φ(h)1, which further gives h ∈ CI, and this is true
for any self adjoint element h of A. Since any a ∈ A can be written as a = h+ik,
h and k being self adjoint elements of A, we obtain the required result.
(cid:3)
Corollary 3.2.
(i) Ab⊗rA is ∗-semisimple if and only if A = CI.
Proof: (i) Follows easily from the fact that a semisimple Banach ∗-algebra
possesses a faithful ∗-representation [19, Corollary 4.7.16].
(ii) Ab⊗rA is symmetric if and only if A = CI
(ii) Let Ab⊗rA be symmetric. Using the same argument as in [1, Proposition
5.16], one can show that the radical of Ab⊗rA is {0}. By [19, Theorem 4.7.15],
∗-radical of Ab⊗rA coincides with its radical. Thus Ab⊗rA is ∗-semisimple, which
using above part implies A = CI.
(cid:3)
References
[1] S. D. Allen, A.M. Sinclair and R. R. Smith. The ideal structure of the Haagerup tensor
product of C ∗-algebras, J. Reine Angew. Math. 442(1993), 111-148.
[2] R. J. Archbold, E. Kaniuth, G. Schlichting and D. W. B. Somerset. Ideal space of the
Haagerup tensor product of C ∗-algebras. Internat. J. Math. 8 (1997), 1-29.
[3] B. Blackadar. Operator algebras: Theory of C ∗-algebras and von-Neumann algebras.
Springer-Verlag Berlin Heidelberg, 2006.
[4] P. J. Cohen. Factorization in group algebras. Duke Math. J. 26 (1959), 199-205.
[5] P. G. Dixon. Non closed sums of closed ideals in Banach algebras. Proc. Amer. Math.
Soc. 128 (2000), 3647-3654.
[6] J. F. Feinstein, E. Kaniuth and D. W. B. Somerset. Spectral synthesis and topologies on
ideal spaces for Banach ∗-algebras. J. Funct. Anal. 196 (2002), 19-39.
12
R. JAIN AND A. KUMAR
[7] J. F. Feinstein and D. W. B. Somerset. Spectral synthesis for Banach algebras II. Math.
Zeit. 239 (2002), 183-213.
[8] C. C. Graham and O. C. McGehee. Essays in commutative harmonic analysis. Springer-
Verlag, New York, 1979.
[9] W. Hauenschild, E. Kaniuth and A. Voigt. ∗-regularity and uniqueness of C ∗-norm for
tensor products of ∗-algebras. J. Funct. Anal. 89 (1990), 137-149.
[10] R. Jain and A. Kumar. Operator space tensor products of C ∗-algebras. Math. Zeit. 260
(2008), 805-811.
[11] R. Jain and A. Kumar. Ideals of operator space projective tensor product of C ∗-algebras.
To appear in J. Aust. Math. Soc., arxiv:1106.3143v1 [math.OA].
[12] R. Jain and A. Kumar. Operator space projective tensor product: Embedding into second
dual and ideal structure. arXiv:1106.2644v1 [math.OA].
[13] E. Kaniuth. A course in commutative Banach algebras. Springer-Verlag, 2009.
[14] A. Kumar. Operator space projective tensor product of C ∗-algebras. Math. Zeit. 237
(2001), 211-217.
[15] A. J. Lazar. The space of ideals in the minimal tensor product of C ∗-algebras. Math.
Proc. Camb. Phil. Soc. 148 (2010), 243-252.
[16] H. Lin. Ideals of multiplier algebras of simple C ∗-algebras. Proc. Amer. Math. Soc. 104
(1988), 239-244.
[17] T. W. Palmer. Banach algebras and the general theory of ∗-algebras II. Cambridge Uni-
versity Press, 2001.
[18] G. K. Pedersen. C ∗-algebras and their automorphism groups. LMS monographs, Academic
press, 1978.
[19] C. E. Rickart. General theory of Banach algebras. van Nostrand, 1974.
[20] D. W. B. Somerset. Spectral synthesis for Banach algebras. Quart. J. Math. Oxford Ser.
49(2) (1998), 501-521.
[21] M. Takesaki. Theory of operator algebras I. Springer-Verlag, Berlin Heidelberg New York,
1979.
Department of Mathematics, Lady Shri Ram College for Women, New Delhi-
110024, India.
E-mail address: ranjanaj [email protected]
Department of Mathematics, University of Delhi, Delhi-110007, India.
E-mail address: [email protected]
|
1507.01831 | 2 | 1507 | 2015-11-17T19:54:01 | Uniqueness results for noncommutative spheres and projective spaces | [
"math.OA",
"math.QA"
] | It is known that, under strong combinatorial axioms, $O_N\subset O_N^*\subset O_N^+$ are the only orthogonal quantum groups. We prove here similar results for the noncommutative spheres $S^{N-1}_\mathbb R\subset S^{N-1}_{\mathbb R,*}\subset S^{N-1}_{\mathbb R,+}$, the noncommutative projective spaces $P^{N-1}_\mathbb R\subset P^{N-1}_\mathbb C\subset P^{N-1}_+$, and the projective orthogonal quantum groups $PO_N\subset PO_N^*\subset PO_N^+$. | math.OA | math |
UNIQUENESS RESULTS FOR NONCOMMUTATIVE SPHERES AND
PROJECTIVE SPACES
TEODOR BANICA AND SZABOLCS M´ESZ ´AROS
Abstract. It is known that, under strong combinatorial axioms, ON ⊂ O∗
N are
the only orthogonal quantum groups. We prove here similar results for the noncommu-
tative spheres SN −1
⊂
P N −1
R,+ , the noncommutative projective spaces P N −1
N ⊂ P O+
N .
, and the projective orthogonal quantum groups P ON ⊂ P O∗
⊂ P N −1
⊂ SN −1
R,∗ ⊂ SN −1
R
C
+
R
N ⊂ O+
The concept of half-liberation goes back to [5], [6]. According to an old result of Brauer
[11], for the orthogonal group ON , with fundamental representation u, we have:
Introduction
Hom(u⊗k, u⊗l) = span(cid:16)Tπ(cid:12)(cid:12)(cid:12)
π ∈ P2(k, l)(cid:17)
Here P2(k, l) is the set of pairings between an upper row of k points, and a lower row of l
points, and the action of pairings on the tensors over CN is as follows, with the Kronecker
symbols δπ ∈ {0, 1} being 1 when all the strings of π join pairs of equal indices:
Tπ(ei1 ⊗ . . . ⊗ eik) = Xj1...jl
δπ(i1...ik
j1...jl )ej1 ⊗ . . . ⊗ ejl
A similar result holds for O+
N . This quantum group, introduced by Wang in [16], and
satisfying the axioms of Woronowicz in [17], [18], is given by:
C(O+
N ) = C ∗(cid:16)(uij)i,j=1,...,N(cid:12)(cid:12)(cid:12)
u = ¯u, ut = u−1(cid:17)
In other words, the passage ON → O+
N is obtained by assuming that the standard
coordinates uij no longer satisfy the commutation relations ab = ba. Now since these
commutation relations read T/\ ∈ End(u⊗2), removing them amounts in "removing the
crossings" from the corresponding set of pairings. We are therefore led to an analogue of
the Brauer formula, with P2 being replaced by the set of noncrossing pairings NC2.
2000 Mathematics Subject Classification. 46L65 (46L54).
Key words and phrases. Half-liberation, Noncommutative sphere.
1
2
TEODOR BANICA AND SZABOLCS M ´ESZ ´AROS
This phenomenon, reminding the liberation philosophy in free probability theory [7],
[13], [14], [15], was investigated in [5], [6], one of the findings there being that an inter-
mediate object O∗
N can be inserted, according to the following scheme:
O+
N
O∗
N
ON
∅
abc = cba
ab = ba
∅
/\
/\
NC2
P ∗
2
P2
To be more precise, O∗
N appears by assuming that the standard coordinates uij
satisfy the "half-commutation" relations abc = cba. These relations are equivalent to
T/\ ∈ End(u⊗3), and the uniqueness result, proved in [6], states that the corresponding
category P ∗
2 =< /\ > is the unique intermediate one NC2 ⊂ P ⊂ P2. See [5], [6].
N ⊂ O+
We will formulate and prove here a number of similar results, concerning some related
geometric objects. We will first discuss the case of noncommutative spheres, with the
statemement that, under strong axioms, the spheres SN −1
R,+ constructed
in [4] are the only ones. Then we will discuss the passage from the affine to the projective
setting, with uniqueness results both for the associated noncommutative projective spaces
P N −1
R ⊂ P N −1
The paper is organized as follows: in §1-2 we discuss the spheres and projective spaces,
, and for their quantum isometry groups P ON ⊂ P O∗
R,∗ ⊂ SN −1
R ⊂ SN −1
C ⊂ P N −1
+
N ⊂ P O+
N .
and in §3-4 we discuss the associated quantum isometry groups.
Acknowledgements. This work was done during the graduate Summer school "Topo-
logical quantum groups", Bedlewo 2015, and we would like to express our gratitude to
the organizers, Uwe Franz, Adam Skalski and Piotr So ltan. Also, we would like to thank
Malte Gerhold, Jan Liszka-Dalecki, and the referee, for several useful suggestions.
1. Noncommutative spheres
According to [4], which was inspired from Wang's paper [16], and from [5], the free and
half-liberated analogues of the unit sphere SN −1
R ⊂ RN are constructed as follows:
Definition 1.1. Associated to any N ∈ N is the following universal C ∗-algebra:
C(SN −1
R,+ ) = C ∗ x1, . . . , xN(cid:12)(cid:12)(cid:12)
xi = x∗
i ,Xi
x2
i = 1!
The quotient of this algebra by the relations xixjxk = xkxjxi is denoted C(SN −1
R,∗ ).
Observe that the above two algebras are indeed well-defined, because the quadratic
i = 1 show that we have xi ≤ 1, for any C ∗-norm. Thus the biggest
C ∗-norm is bounded, and the enveloping C ∗-algebras are well-defined.
relations Pi x2
O
O
O
O
NONCOMMUTATIVE SPHERES AND PROJECTIVE SPACES
3
We use the convention that the category of "noncommutative compact spaces" is the
category of unital C ∗-algebras, with the arrows reversed. Given such a space X =
Spec(A), its classical version Xclass, which is a usual compact space, is the Gelfand spec-
trum Xclass = Spec(A/I), where I ⊂ A is the commutator ideal. We have then:
Proposition 1.2. We have inclusions of noncommutative compact spaces
SN −1
R ⊂ SN −1
R,∗ ⊂ SN −1
R,+
and SN −1
R
is the classical version of both spaces on the right.
Proof. According to the Gelfand and the Stone-Weierstrass theorems, the algebra of con-
tinuous functions on the real sphere has the following description:
C(SN −1
R
) = C ∗
comm x1, . . . , xN(cid:12)(cid:12)(cid:12)
R,+ ) → C(SN −1
xi = x∗
i ,Xi
R,∗ ) → C(SN −1
R
x2
i = 1!
), with the second map
(cid:3)
Thus we have quotient maps C(SN −1
being obtained by dividing by the commutator ideal, and this gives the result.
We can axiomatize our spheres, by using the following notion, from [1]:
Definition 1.3. A monomial sphere is a subset S ⊂ SN −1
R,+ obtained via relations of type
xi1 . . . xik = xiσ(1) . . . xiσ(k), ∀(i1, . . . , ik) ∈ {1, . . . , N}k
with σ ∈ Sk being certain permutations, of variable size k ∈ N.
Equivalently, consider the inductive limit S∞ =Sk≥0 Sk, with the inclusions Sk ⊂ Sk+1
being given by σ ∈ Sk =⇒ σ(k + 1) = k + 1. To any σ ∈ S∞ we can then associate the
relations xi1 . . . xik = xiσ(1) . . . xiσ(k), for any (i1, . . . , ik) ∈ {1, . . . , N}k, with k ∈ N being
such that σ ∈ Sk. Observe that these relations are indeed unchanged when replacing
k → k + 1, because by using Pi x2
xi1 . . . xik xik+1 = xiσ(1) . . . xiσ(k)xik+1 =⇒ xi1 . . . xik x2
i = 1 we can always "simplify" at right:
ik+1 = xiσ(1) . . . xiσ(k)x2
ik+1
xi1 . . . xik x2
xiσ(1) . . . xiσ(k)x2
ik+1
=⇒ Xik+1
ik+1 =Xik+1
=⇒ xi1 . . . xik = xiσ(1) . . . xiσ(k)
With this convention, a monomial sphere is a subset S ⊂ SN −1
R,+ obtained via relations
xi1 . . . xik = xiσ(1) . . . xiσ(k) as above, associated to certain elements σ ∈ S∞.
Observe that the basic 3 spheres are all monomial, with the permutations producing
SN −1
R
, SN −1
R,∗ being the standard crossing and the half-liberated crossing:
◦
◦
❀❀❀❀❀❀❀❀
✄✄✄✄✄✄✄✄
◦
◦
◦
◦
●●●●●●●●●●●
✇✇✇✇✇✇✇✇✇✇✇
◦
◦
◦
◦
4
TEODOR BANICA AND SZABOLCS M ´ESZ ´AROS
Here, and in what follows, we agree to represent the permutations σ ∈ Sk by diagrams
between two rows of k points, acting by definition downwards.
With these notions in hand, we can now formulate our main classification result:
Theorem 1.4. The spheres SN −1
R ⊂ SN −1
R,∗ ⊂ SN −1
R,+ are the only monomial ones.
Proof. We follow the approach in [1], where the result was conjectured. We fix a monomial
sphere S ⊂ SN −1
R,+ , and we associate to it subsets Gk ⊂ Sk, as follows:
Gk =nσ ∈ Sk(cid:12)(cid:12)(cid:12)
xi1 . . . xik = xiσ(1) . . . xiσ(k), ∀(i1, . . . , ik) ∈ {1, . . . , N}ko
Since the relations of type xi1 . . . xik = xiσ(1) . . . xiσ(k) can be composed and reversed,
each Gk is a group. Moreover, since we have σ ∈ Gk =⇒ σ ∈ Gk+1, we can form the
increasing union G = (Gk), which is a subgroup of the increasing union S∞ = (Sk).
Since the relations xi1 . . . xik = xiσ(1) . . . xiσ(k) can be concatenated as well, our group
G = (Gk) is "filtered", in the sense that it is stable under the operation (π, σ) → π ⊗ σ.
Moreover, G is stable under two more diagrammatic operations, as follows:
(1) Removing outer strings. Indeed, by summing over a, we have:
Xa = Y a =⇒ Xa2 = Y a2 =⇒ X = Y
aX = aY =⇒ a2X = a2Y =⇒ X = Y
(2) Removing neighboring strings. Indeed, once again by summing over a, we have:
XabY = ZabT =⇒ Xa2Y = Za2T =⇒ XY = ZT
XabY = ZbaT =⇒ Xa2Y = Za2T =⇒ XY = ZT
The problem is that of proving that the only such groups are {1} ⊂ S∗
∞ is the group associated to the half-liberated sphere SN −1
S∗
G ⊂ S∞, assumed non-trivial, G 6= {1}, and satisfying the above conditions.
∞ ⊂ S∞, where
R,∗ . So, consider a filtered group
Step 1. Our first claim is that G contains a 3-cycle. For this purpose, we use a stan-
dard trick, stating that if π, σ ∈ S∞ have support overlapping on exactly one point,
say supp(π) ∩ supp(σ) = {i}, then the commutator σ−1π−1σπ is a 3-cycle, namely
(i, σ−1(i), π−1(i)). Indeed the computation of the commutator goes as follows:
π
σ
π−1
σ−1
=
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
❖❖❖❖❖❖❖❖❖❖
⑦⑦⑦⑦⑦
♦♦♦♦♦♦♦♦♦♦
❅❅❅❅❅
◦
◦
◦
◦
◦
•
•
•
•
•
◦
♦♦♦♦♦♦♦♦♦♦
❅❅❅❅❅
◦
◦
❖❖❖❖❖❖❖❖❖❖
⑦⑦⑦⑦⑦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
NONCOMMUTATIVE SPHERES AND PROJECTIVE SPACES
5
Now let us pick a non-trivial element τ ∈ G. By removing outer strings at right and at
left we obtain permutations τ ′ ∈ Gk, τ ′′ ∈ Gs having a non-trivial action on their right/left
leg, and by taking π = τ ′ ⊗ ids−1, σ = idk−1 ⊗ τ ′′, the trick applies.
Step 2. Our second claim is G must contain one of the following permutations:
◦
◦
◦
✳✳✳✳✳✳✳
✳✳✳✳✳✳✳
✂✂✂✂✂✂✂✂
◦
◦
◦
◦
◦
◦
◦
❁❁❁❁❁❁❁❁
✳✳✳✳✳✳✳
✈✈✈✈✈✈✈✈✈✈✈
◦
◦
◦
◦
◦
◦
◦
❁❁❁❁❁❁❁❁
✳✳✳✳✳✳✳
✈✈✈✈✈✈✈✈✈✈✈
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
♣♣♣♣♣♣♣♣♣♣♣♣♣♣
❁❁❁❁❁❁❁❁
❁❁❁❁❁❁❁❁
◦
◦
◦
◦
◦
Indeed, consider the 3-cycle that we just constructed. By removing all outer strings,
and then all pairs of adjacent vertical strings, we are left with these permutations.
Step 3. Our claim now is that we must have S∗
∞ ⊂ G. Indeed, let us pick one of the
permutations that we just constructed, and apply to it our various diagrammatic rules.
From the first permutation we can obtain the basic crossing, as follows:
◦
◦
◦
◦
♦♦♦♦♦♦♦♦♦♦
❅❅❅❅❅
❅❅❅❅❅
♦♦♦♦♦♦♦♦♦♦
❅❅❅❅❅
❅❅❅❅❅
◦
◦
◦
◦
◦
◦
◦
◦
→
◦
◦
✴✴✴✴✴✴✴✴✴✴
✎✎✎✎✎✎✎✎✎✎
◦
◦
◦
◦
✴✴✴✴✴✴✴✴✴✴
✎✎✎✎✎✎✎✎✎✎
◦
◦
→
◦
◦
✴✴✴✴✴✴✴✴✴✴
✎✎✎✎✎✎✎✎✎✎
◦
◦
Also, by removing a suitable /\ shaped configuration, which is represented by dotted
lines in the diagrams below, we can obtain the basic crossing from the second and third
permutation, and the half-liberated crossing from the fourth permutation:
◦
◦
◦
◦
✳✳✳✳✳✳✳
✈✈✈✈✈✈✈✈✈✈✈
◦
◦
◦
◦
◦
◦
◦
✳✳✳✳✳✳✳
✈✈✈✈✈✈✈✈✈✈✈
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
♣♣♣♣♣♣♣♣♣♣♣♣♣♣
❁❁❁❁❁❁❁❁
◦
◦
◦
◦
◦
Thus, in all cases we have a basic or half-liberated crossing, and so S∗
∞ ⊂ G, as claimed.
Step 4. Our last claim, which will finish the proof, is that there is no proper intermediate
subgroup S∗
∞ ⊂ S∞ is the
subgroup of "parity-preserving" permutations (i even =⇒ σ(i) even). Equivalently,
S∗
∞ ⊂ G ⊂ S∞. In order to prove this, we recall from [1] that S∗
∞ ⊂ S∞ is the subgroup generated by the transpositions (1, 3), (2, 4), (3, 5), . . .
Now let us pick an element σ ∈ Sk − S∗
k, with k ∈ N. We must prove that the group
∞, σ > equals the whole S∞. In order to do so, we use the fact that σ is not
G =< S∗
parity preserving. Thus, we can find i even such that σ(i) is odd.
In addition, up to passing to σ, we can assume that σ(k) = k, and then, up to passing
one more time to σ, we can further assume that k is even.
Since both i, k are even we have (i, k) ∈ S∗
k, and so σ(i, k)σ−1 = (σ(i), k) belongs to
G. But, since σ(i) is odd, by deleting an appropriate number of vertical strings, (σ(i), k)
reduces to the basic crossing (1, 2). Thus G = S∞, and we are done.
(cid:3)
6
TEODOR BANICA AND SZABOLCS M ´ESZ ´AROS
Summarizing, we have now a complete axiomatization for the basic 3 spheres. In what
follows we will prove some similar results, for some related geometric objects.
2. Projective spaces
We discuss here a "projective version" of the classification results in section 1. Our
starting point is the following functional analytic description of P N −1
R
, P N −1
C
:
Proposition 2.1. We have presentation results as follows,
C(P N −1
R
C(P N −1
C
) = C ∗
) = C ∗
comm(cid:16)(pij)i,j=1,...,N(cid:12)(cid:12)(cid:12)
comm(cid:16)(pij)i,j=1,...,N(cid:12)(cid:12)(cid:12)
p = ¯p = pt = p2, T r(p) = 1(cid:17)
p = p∗ = p2, T r(p) = 1(cid:17)
for the algebras of continuous functions on the real and complex projective spaces.
Proof. This follows indeed from the Gelfand and Stone-Weierstrass theorems, by using
the fact that P N −1
are the spaces of rank one projections in MN (R), MN (C). (cid:3)
, P N −1
R
C
The above result suggests the following definition:
Definition 2.2. Associated to any N ∈ N is the following universal algebra,
whose abstract spectrum is called "free projective space".
C(P N −1
+
) = C ∗(cid:16)(pij)i,j=1,...,N(cid:12)(cid:12)(cid:12)
p = p∗ = p2, T r(p) = 1(cid:17)
Observe that we have embeddings of noncommutative compact spaces P N −1
R ⊂ P N −1
P N −1
+
, and that the complex projective space P N −1
C
is the classical version of P N −1
+
C ⊂
.
Given a closed subset X ⊂ SN −1
R,+ , its projective version is by definition the quotient
space X → P X determined by the fact that C(P X) ⊂ C(X) is the subalgebra generated
by the variables pij = xixj. We have then the following result, from [4]:
Proposition 2.3. The projective versions of the 3 spheres are given by
SN −1
R
SN −1
R,∗
SN −1
R,+
P N −1
R
/ P N −1
C
/ P N −1
+
where P N −1
+
is a certain noncommutative compact space, contained in P N −1
+
.
/
/
/
/
/
/
NONCOMMUTATIVE SPHERES AND PROJECTIVE SPACES
7
Proof. The assertion at left is true by definition. For the assertion at right, we have to
prove that the variables pij = zizj over the free sphere SN −1
R,+ satisfy the defining relations
for C(P N −1
), from Definition 2.2, and the verification here goes as follows:
+
(p∗)ij = p∗
ji = (zjzi)∗ = zizj = pij
ziz2
kzj = zizj = pij
(p2)ij = Xk
T r(p) = Xk
pikpkj =Xk
pkk =Xk
z2
k = 1
Regarding now the middle assertion, stating that we have P SN −1
, the inclusion
"⊂" follows from the relations abc = cba, which imply abcd = cbad = cbda. In the other
sense now, the point is that we have a matrix model representation, as follows:
R,∗ = P N −1
C
π : C(SN −1
R,∗ ) → M2(C(SN −1
C
))
:
xi →(cid:18) 0 zi
0(cid:19)
¯zi
But this gives the missing inclusion "⊃", and we are done. See [4].
(cid:3)
The inclusion P N −1
is known to be quite similar to an isomorphism, algebrically
speaking. To be more precise, when performing the GNS construction with respect to the
canonical integration functionals, P N −1
becomes an isomorphism. See [4], [6].
+ ⊂ P N −1
+ ⊂ P N −1
+
+
We can axiomatize our noncommutative projective spaces, as follows:
Definition 2.4. A monomial space is a subset P ⊂ P N −1
+
obtained via relations of type
pi1i2 . . . pik−1ik = piσ(1)iσ(2) . . . piσ(k−1)iσ(k), ∀(i1, . . . , ik) ∈ {1, . . . , N}k
with σ ranging over a certain subset of Sk∈2N Sk, stable under σ → σ.
Observe the similarity with Definition 1.3. The only subtlety in the projective case is
the stability under σ → σ, which in practice means that if the above relation associated
to σ holds, then the following relation, associated to σ, must hold as well:
pi0i1 . . . pikik+1 = pi0iσ(1)piσ(2)iσ(3) . . . piσ(k−2)iσ(k−1)piσ(k)ik+1
As an illustration, the basic projective spaces are all monomial:
Proposition 2.5. The 3 projective spaces are all monomial, with the permutations
◦
◦
❀❀❀❀❀❀❀❀
✄✄✄✄✄✄✄✄
◦
◦
◦
◦
producing respectively the spaces P N −1
R
, P N −1
C
.
❆❆❆❆❆❆❆❆❆
⑥⑥⑥⑥⑥⑥⑥⑥⑥
❆❆❆❆❆❆❆❆❆
⑥⑥⑥⑥⑥⑥⑥⑥⑥
◦
◦
◦
◦
◦
◦
8
TEODOR BANICA AND SZABOLCS M ´ESZ ´AROS
Proof. We must divide the algebra C(P N −1
in the statement, as well as those associated to their shifted versions, given by:
) by the relations associated to the diagrams
+
◦
◦
◦
◦
✵✵✵✵✵✵✵
✍✍✍✍✍✍✍
◦
◦
◦
◦
◦
◦
◦
◦
❆❆❆❆❆❆❆❆❆
⑥⑥⑥⑥⑥⑥⑥⑥⑥
❆❆❆❆❆❆❆❆❆
⑥⑥⑥⑥⑥⑥⑥⑥⑥
◦
◦
◦
◦
◦
◦
◦
◦
(1) The basic crossing, and its shifted version, produce the relations pab = pba and
pabpcd = pacpbd. Now by using these relations several times, we obtain:
pabpcd = pacpbd = pcapdb = pcdpab
Thus, the space produced by the basic crossing is classical, P ⊂ P N −1
one more time the relations pab = pba we conclude that we have P = P N −1
C
R
, and by using
, as claimed.
(2) The fattened crossing, and its shifted version, produce the relations pabpcd = pcdpab
and pabpcdpef = padpebpcf . The first relations tell us that the projective space must be
classical, P ⊂ P N −1
. Now observe that with pij = zi ¯zj, the second relations read:
C
za ¯zbzc ¯zdze ¯zf = za¯zdze ¯zbzc ¯zf
Since these relations are automatic, we have P = P N −1
C
, and we are done.
(cid:3)
We can now formulate our projective classification result, as follows:
Theorem 2.6. The projective spaces P N −1
R ⊂ P N −1
C ⊂ P N −1
+
are the only monomial ones.
Proof. We follow the proof from the affine case. Let Rσ be the collection of relations
associated to a permutation σ ∈ Sk with k ∈ 2N, as in Definition 2.4. We fix a monomial
projective space P ⊂ P N −1
, and we associate to it subsets Gk ⊂ Sk, as follows:
+
Gk =({σ ∈ SkRσ hold over P }
(k even)
{σ ∈ SkRσ hold over P } (k odd)
As in the affine case, we obtain in this way a filtered group G = (Gk), which is sta-
ble under removing outer strings, and under removing neighboring strings. Thus the
computations in the proof of Theorem 1.4 apply, and show that we have only 3 possible
situations, corresponding to the 3 projective spaces in Proposition 2.3 above.
(cid:3)
3. Quantum isometries
We discuss now the quantum isometry groups of the spheres and projective spaces.
N , with standard coordinates denoted
N obtained by assuming that the standard
Consider the free orthogonal quantum group O+
uij. Consider as well the subgroup O∗
coordinates uij satisfy the half-commutation relations abc = cba. See [5], [6].
N ⊂ O+
Given a closed subgroup G ⊂ O+
N , its projective version G → P G is by definition
given by the fact that C(P G) ⊂ C(G) is the subalgebra generated by the variables
NONCOMMUTATIVE SPHERES AND PROJECTIVE SPACES
9
wij,ab = uiaujb. In the classical case we recover in this way the usual projective version,
P G = G/(G ∩ ZN
2 ). It is also known that we have P O∗
N = P UN . See [6].
We use the following action formalism, inspired from [12], [16]:
Definition 3.1. Consider a closed subgroup G ⊂ O+
N , and a closed subset X ⊂ SN −1
R,+ .
C ∗-algebras Φ : C(X) → C(G) ⊗ C(X).
(1) We write G y X when the formula Φ(zi) = Pa uia ⊗ za defines a morphism of
(2) We write P G y P X when the formula Φ(zizj) = Pa uiaujb ⊗ zazb defines a
morphism of C ∗-algebras Φ : C(P X) → C(P G) ⊗ C(P X).
Observe that the above morphisms Φ, if they exist, are automatically coaction maps.
Observe also that an affine action G y X produces a projective action P G y P X.
Finally, let us mention that given an algebraic subset X ⊂ SN −1
R,+ , it is routine to prove
that there exist universal quantum groups G ⊂ O+
N acting as (1), and as in (2).
We have the following result, with respect to the above notions:
Theorem 3.2. The quantum isometry groups of the spheres and projective spaces are
ON
O∗
N
O+
N
P ON
/ P UN
/ P O+
N
with respect to the affine and projective action notions introduced above.
Proof. The fact that the 3 quantum groups on top act affinely on the corresponding 3
spheres is known since [4], and is elementary. By restriction, the 3 quantum groups on
the bottom follow to act on the corresponding 3 projective spaces.
We must prove now that all these actions are universal. At right there is nothing to
prove, so we are left with studying the actions on SN −1
R
, SN −1
R,∗ and on P N −1
R
, P N −1
C
.
SN −1
R
. Here the fact that the action ON y SN −1
R
is universal is known from [8], and
follows as well from the fact that the action P ON y P N −1
R
is universal, proved below.
SN −1
R,∗ . The situation is similar here, with the universality of O∗
N
in [2], and following as well from the universality of P UN y P N −1
C
R,∗ being proved
y SN −1
, proved below.
P N −1
R
. In terms of the projective coordinates wia,jb = uiaujb and pij = zizj, the coaction
map is given by Φ(pij) =Pab wia,jb ⊗ pab, and we have:
Φ(pij) = Xa<b
Φ(pji) = Xa<b
(wij,ab + wij,ba) ⊗ pab +Xa
(wji,ab + wji,ba) ⊗ pab +Xa
wij,aa ⊗ paa
wji,aa ⊗ paa
/
/
/
/
/
/
10
TEODOR BANICA AND SZABOLCS M ´ESZ ´AROS
By comparing these two formulae, and then by using the linear independence of the
variables pab = zazb for a ≤ b, we conclude that we must have:
wij,ab + wij,ba = wji,ab + wji,ba
Let us apply now the antipode to this formula. For this purpose, observe first that
we have S(wij,ab) = S(uiaujb) = S(ujb)S(uia) = ubjuai = wba,ji. Thus by applying the
antipode we obtain wba,ji + wab,ji = wba,ij + wab,ij, and by relabelling, we obtain:
wji,ba + wij,ba = wji,ab + wij,ab
Now by comparing with the original relation, we obtain wij,ab = wji,ba. But, with
wij,ab = uiaujb, this formula reads uiaujb = ujbuia. Thus our quantum group G ⊂ O+
N
must be classical, G ⊂ ON , and so we have P G ⊂ P ON , as claimed.
The idea here will be that of using the formula pabpcd = padpcb. We have:
. Consider a coaction map, written as Φ(pij) =Pab uiaujb ⊗ pab, with pab = za ¯zb.
P N −1
C
Φ(pijpkl) = Xabcd
Φ(pilpkj) = Xabcd
uiaujbukculd ⊗ pabpcd
uiauldukcujb ⊗ padpcb
The terms at left being equal, and the last terms at right being equal too, we deduce
that, with [a, b, c] = abc − cba, we must have the following formula:
uia[ujb, ukc, uld] ⊗ pabpcd = 0
Xabcd
Now since the quantities pabpcd = za ¯zbzc ¯zd at right depend only on the numbers
{a, c}, {b, d} ∈ {1, 2}, and this dependence produces the only possible linear relations
between the variables pabpcd, we are led to 2 × 2 = 4 equations, as follows:
(1) uia[ujb, uka, ulb] = 0, ∀a, b.
(2) uia[ujb, uka, uld] + uia[ujd, uka, ulb] = 0, ∀a, ∀b 6= d.
(3) uia[ujb, ukc, ulb] + uic[ujb, uka, ulb] = 0, ∀a 6= c, ∀b.
(4) uia[ujb, ukc, uld] + uia[ujd, ukc, ulb] + uic[ujb, uka, uld] + uic[ujd, uka, ulb] = 0, ∀a 6= c,
∀b 6= d.
We will need in fact only the first two formulae. Since (1) corresponds to (2) at b = d,
we conclude that (1,2) are equivalent to (2), with no restriction on the indices. By
multiplying now this formula to the left by uia, and then summing over i, we obtain:
[ujb, uka, uld] + [ujd, uka, ulb] = 0
We use now the antipode/relabel trick from [8]. By applying the antipode we obtain
[udl, uak, ubj] + [ubl, uak, udj] = 0, and by relabelling we obtain:
[uld, uka, ujb] + [ujd, uka, ulb] = 0
NONCOMMUTATIVE SPHERES AND PROJECTIVE SPACES
11
Now by comparing with the original relation, we obtain [ujb, uka, uld] = [ujd, uka, ulb] =
(cid:3)
0. Thus our quantum group is half-classical, G ⊂ O∗
N , and we are done.
say that a closed subgroup G ⊂ U +
The above results can be probably further improved. As an example here,
let us
N acts projectively on P X when we have a coaction
jb ⊗ zazb. Then the above proof can be adapted, by putting ∗
signs where needed, and so Theorem 3.2 still holds, under this more general formalism.
However, establishing the most general universality results, involving arbitrary subgroups
H ⊂ P O+
map Φ(zizj) = Pab uiau∗
N , looks like a quite non-trivial question, and we have no results here.
4. Projective easiness
We discuss here the analogues of the classification results in sections 1-2, for the quan-
tum groups introduced in section 3. First, we have the following key result, from [3]:
Proposition 4.1. We have the following results:
(1) The quantum group inclusion ON ⊂ O∗
(2) The group inclusion P ON ⊂ P UN is maximal.
N is maximal.
Proof. The idea here is that (2) can be obtained by using standard Lie group tricks, and
(1) follows then from it, via standard algebraic lifting results. See [3].
(cid:3)
Our claim now is that, under suitable assumptions, O∗
ON ⊂ G ⊂ O+
N , and P UN is the only intermediate object P ON ⊂ G ⊂ P O+
formulate a precise statement here, we recall the following notion, from [5]:
N is the only intermediate object
N . In order to
Definition 4.2. An intermediate quantum group ON ⊂ G ⊂ O+
N is called easy when
span(NC2(k, l)) ⊂ Hom(u⊗k, u⊗l) ⊂ span(P2(k, l))
comes via Hom(u⊗k, u⊗l) = span(D(k, l)), for certain sets of pairings D(k, l).
As explained in [5], by "saturating" the sets D(k, l), we can assume that the collection
D = (D(k, l)) is a category of pairings, in the sense that it is stable under vertical and
horizontal concatenation, upside-down turning, and contains the semicircle. See [5].
In the projective case now, we have the following related definition:
Definition 4.3. A projective category of pairings is a collection of subsets
stable under the usual categorical operations, and satisfying σ ∈ E =⇒ σ ∈ E.
NC2(2k, 2l) ⊂ E(k, l) ⊂ P2(2k, 2l)
As basic examples here, we have the categories NC2 ⊂ P ∗
2 ⊂ P2, where P ∗
2 is the
category of matching pairings. This follows indeed from definitions.
Now with the above notion in hand, we can formulate:
12
TEODOR BANICA AND SZABOLCS M ´ESZ ´AROS
Definition 4.4. A quantum group P ON ⊂ H ⊂ P O+
N is called projectively easy when
span(NC2(2k, 2l)) ⊂ Hom(v⊗k, v⊗l) ⊂ span(P2(2k, 2l))
comes via Hom(v⊗k, v⊗l) = span(E(k, l)), for a certain projective category E = (E(k, l)).
Observe that, given any easy quantum group ON ⊂ G ⊂ O+
N , its projective version
N is projectively easy in our sense. In particular the quantum groups
N are all projectively easy, coming from NC2 ⊂ P ∗
2 ⊂ P2.
P ON ⊂ P G ⊂ P O+
P ON ⊂ P UN ⊂ P O+
We have in fact the following general result:
Proposition 4.5. We have a bijective correspondence between the affine and projective
categories of partitions, given by G → P G at the quantum group level.
Proof. The construction of correspondence D → E is clear, simply by setting:
Conversely, given E = (E(k, l)) as in Definition 4.3, we can set:
E(k, l) = D(2k, 2l)
D(k, l) =(E(k, l)
(k, l even)
{σ : σ ∈ E(k + 1, l + 1)} (k, l odd)
Our claim is that D = (D(k, l)) is a category of partitions. Indeed:
(1) The composition action is clear.
Indeed, when looking at the numbers of legs
involved, in the even case this is clear, and in the odd case, this follows from:
σ, σ′ ∈ E =⇒ σ
τ ∈ E =⇒ σ
τ ∈ D
(2) For the tensor product axiom, we have 4 cases to be investigated. The even/even
case is clear, and the odd/even, even/odd, odd/odd cases follow respectively from:
σ, τ ∈ E =⇒ στ ∈ E =⇒ στ ∈ D
σ, τ ∈ E =⇒ σ, τ ∈ E =⇒ στ ∈ E =⇒ στ ∈ E =⇒ στ ∈ D
σ, τ ∈ E =⇒ σ, τ ∈ E =⇒ στ ∈ E =⇒ στ ∈ E =⇒ στ ∈ D
(3) Finally, the conjugation axiom is clear from definitions.
Now with these definitions in hand, both compositions D → E → D and E → D → E
(cid:3)
follow to be the identities, and the quantum group assertion is clear as well.
Now back to the uniqueness issues, we have here:
Theorem 4.6. We have the following results:
N is the only intermediate easy quantum group ON ⊂ G ⊂ O+
N .
(1) O∗
(2) P UN is the only intermediate projectively easy quantum group P ON ⊂ G ⊂ P O+
N .
Proof. The assertion regarding ON ⊂ O∗
P ON ⊂ P UN ⊂ P O+
N ⊂ O+
N follows from it, and from the duality in Proposition 4.5.
N is from [6], and the assertion regarding
(cid:3)
NONCOMMUTATIVE SPHERES AND PROJECTIVE SPACES
13
There are of course a number of finer results waiting to be established, regarding the
, with inspiration from Proposition 4.1. The
inclusions SN −1
results in [9], [10] provide in principle useful tools in dealing with such questions.
R ⊂ SN −1
R,∗ and P N −1
R ⊂ P N −1
C
References
[1] T. Banica, Liberations and twists of real and complex spheres, J. Geom. Phys. 96 (2015), 1 -- 25.
[2] T. Banica, Half-liberated manifolds, and their quantum isometries, Glasg. Math. J., to appear.
[3] T. Banica, J. Bichon, B. Collins and S. Curran, A maximality result for orthogonal quantum groups,
Comm. Algebra 41 (2013), 656 -- 665.
[4] T. Banica and D. Goswami, Quantum isometries and noncommutative spheres, Comm. Math. Phys.
298 (2010), 343 -- 356.
[5] T. Banica and R. Speicher, Liberation of orthogonal Lie groups, Adv. Math. 222 (2009), 1461 -- 1501.
[6] T. Banica and R. Vergnioux, Invariants of the half-liberated orthogonal group, Ann. Inst. Fourier
60 (2010), 2137 -- 2164.
[7] H. Bercovici and V. Pata, Stable laws and domains of attraction in free probability theory, Ann. of
Math. 149 (1999), 1023 -- 1060.
[8] J. Bhowmick and D. Goswami, Quantum isometry groups: examples and computations, Comm.
Math. Phys. 285 (2009), 421 -- 444.
[9] J. Bichon, Half-liberated real spheres and their subspaces, preprint 2015.
[10] J. Bichon and M. Dubois-Violette, Half-commutative orthogonal Hopf algebras, Pacific J. Math.
263 (2013), 13 -- 28.
[11] R. Brauer, On algebras which are connected with the semisimple continuous groups, Ann. of Math.
38 (1937), 857 -- 872.
[12] D. Goswami, Existence and examples of quantum isometry groups for a class of compact metric
spaces, Adv. Math. 280 (2015), 340 -- 359.
[13] A. Nica and R. Speicher, Lectures on the combinatorics of free probability, Cambridge Univ. Press
(2006).
[14] R. Speicher, Multiplicative functions on the lattice of noncrossing partitions and free convolution,
Math. Ann. 298 (1994), 611 -- 628.
[15] D.V. Voiculescu, K.J. Dykema and A. Nica, Free random variables, AMS (1992).
[16] S. Wang, Free products of compact quantum groups, Comm. Math. Phys. 167 (1995), 671 -- 692.
[17] S.L. Woronowicz, Compact matrix pseudogroups, Comm. Math. Phys. 111 (1987), 613 -- 665.
[18] S.L. Woronowicz, Tannaka-Krein duality for compact matrix pseudogroups. Twisted SU(N) groups,
Invent. Math. 93 (1988), 35 -- 76.
T.B.: Department of Mathematics, Cergy-Pontoise University, 95000 Cergy-Pontoise,
France. [email protected]
S.M.: Department of Mathematics, Central European University, 1051 Budapest, Hun-
gary. [email protected]
|
1512.02296 | 2 | 1512 | 2015-12-14T03:25:52 | Continous-trace $k$-graph $C^*$-algebras | [
"math.OA"
] | A characterization is given for directed graphs that yield graph $C^*$-algebras with continuous trace. This is established for row-finite graphs with no sources first using a groupoid approach, and extended to the general case via the Drinen-Tomforde desingularization. Partial results are given to characterize higher-rank graphs that yield $C^*$-algebras with continuous trace. | math.OA | math |
CONTINUOUS-TRACE k-GRAPH C ∗-ALGEBRAS
DANNY CRYTSER
Abstract. A characterization is given for directed graphs that yield graph
C ∗-algebras with continuous trace. This is established for row-finite graphs
with no sources first using a groupoid approach, and extended to the general
case via the Drinen-Tomforde desingularization. Partial results are given to
characterize higher-rank graphs that yield C ∗-algebras with continuous trace.
1. Introduction
To any directed graph E one can affiliate a graph C ∗-algebra C ∗(E), generated
by a universal family of projections and partial isometries that satisfy certain Cuntz-
Krieger relations. Many C ∗-algebraic properties of C ∗(E) are governed by graph-
theoretic properties of E. For example, C ∗(E) is an AF algebra exactly when E
has no directed cycles. To any graph E there is also affiliated a path groupoid
GE, an ´etale groupoid which models the shift dynamics of infinite paths in E.
This groupoid provides an alternate model for C ∗(E), in the sense that C ∗(E) ∼=
C ∗(GE). This isomorphism allows for the use of tools from the theory of groupoid
C ∗-algebras to study graph algebras.
In this paper we give an example of this
approach, characterizing continuous-trace graph C ∗-algebras by applying the main
result of [9] to the path groupoid of a directed graph.
The path groupoid is easiest to use if the graph is non-singular, in the sense
that each vertex is the range of a finite non-empty set of edges. We first work in
the non-singular case, and then use desingularization to extend to the general case.
Desingularization takes a non-regular graph E and returns a non-singular graph E
such that the affiliated graph C ∗-algebras are Morita equivalent (so that continuity
of trace is preserved). As an application we use a result from [14] to characterize
continuous trace AF algebras in terms of their Bratteli diagrams.
In the last section we consider higher-rank graph C ∗-algebras with continuous
trace. Higher-rank graphs are categories which generalize the category of finite
directed paths within a directed graph. They have C ∗-algebras defined along the
same lines as graph C ∗-algebras. We include necessary background on the theory
of higher-rank graph algebras. Again, the use of groupoids is crucial. The higher-
rank case is more complicated and we are only able to give partial results.
In
particular, giving a combinatorial description of when the isotropy groups vary
continuously for a k-graph path groupoid seems out of reach, so we focus instead
on the principal/aperiodic case. We note a simple necessary condition on a higher-
rank graph for its associated C ∗-algebra to have continuous trace, a corollary of a
result from [3].
While this paper was in preparation we were made aware of a related paper of
Hazlewood ([5]) which contains similar results to ours.
In particular, Theorems
6.5.22, 6.2.13 and 6.4.11 in [5] are in some sense cycle-free/principal versions of
1
2
DANNY CRYTSER
Theorem 3.5 and Theorem 5.12. The results in [5] also show that some desingular-
ization ´a la Section 4 in the present paper is possible for the k-graph case (although
it seems that resolving infinite receivers is somewhat more difficult).
2. Continuous-trace C ∗-algebras; graph algebras; groupoids
For an element a in a C ∗-algebra A, and a unitary equivalence class s = [π] ∈ A,
we define the rank of s(a) to be the rank of π(a). We say that s(a) is a projection
if and only if π(a) is a projection.
Definition 2.1. [12, Def. 5.13] Let A be a C ∗-algebra with Hausdorff spectrum A.
Then A is said to have continuous trace (or be continuous-trace) if for each t ∈ A
there exist an open set U ⊂ A containing t and an element a ∈ A such that s(a) is
a rank-one projection for every s ∈ U .
For an introduction to graphC ∗- algebras, please see [11]. The reader who is
already familiar with graph C ∗-algebras may pass over the following standard def-
initions.
Definition 2.2. A (directed) graph E is an ordered quadruple E = (E0, E1, r, s),
where the E0 and E1 are countable sets called the vertices and edges, and r, s :
E1 → E0 are maps called the range and source maps.
A vertex v is called an infinite receiver if there are infinitely many edges in E1
with range v; a vertex is called a source if it receives no sources. A vertex is regular
if it is neither an infinite receiver or a source; otherwise, it is called singular. A
graph is row-finite if it has no infinite receivers and has no sources if every vertex
receives an edge. A cycle in a directed graph is a path λ ∈ E ∗ \E0 with r(λ) = s(λ);
a simple cycle is a cycle λ which does not contain another cycle. An entrance to
the cycle λ = e1 . . . en is an edge e with r(e) = ek and e 6= ek.
The finite path space E ∗ consists of all finite sequences e1 . . . en in E1 such that
s(ei) = r(ei+1) for i = 1, . . . , n − 1. The range of the path e1 . . . en is defined to
be r(e1) and its source is s(en). If µ = e1 . . . en is a finite path, then we define the
length to be n and write µ = n. The vertices are included in the finite path space
as the paths of length zero. If λ = e1 . . . en and µ = f1 . . . fm are finite paths with
s(λ) = r(µ), we can concatenate them to from λµ = e1 . . . enf1 . . . fm ∈ E ∗. The
infinite path space is E∞ = {e1e2 . . . s(ei) = r(ei+1)∀i ≥ 1}. If λ = e1 . . . en ∈ E ∗
and x = f1 . . . ∈ E∞, then λx = e1 . . . enf1 . . . ∈ E∞. The range of x = e1e2 . . . ∈
E∞ is defined as r(x) := r(e1). The shift map σ : E∞ → E∞ removes the first
edge from an infinite path: σ(e1e2 . . .) = e2e3 . . .. Composing σ with itself yields
powers σ2, σ3, . . ..
Definition 2.3. Let E be a directed graph. Then the graph C ∗-algebra of E,
denoted C ∗(E), is the universal C ∗-algebra generated by projections {pv : v ∈ E0}
and partial isometries {se : e ∈ E1} satisfying the following Cuntz-Krieger relations:
ese = ps(e) for any e ∈ E1;
e ≤ pr(e) for any e ∈ E1;
(1) s∗
(2) ses∗
(3) s∗
(4) If v is regular, then pv = Pr(e)=v ses∗
e.
We also include the basic definitions for groupoids. A concise definition of a
esf = 0 for distinct e, f ∈ E1;
groupoid is a small category with inverses; we include a more detailed definition.
CONTINUOUS-TRACE k-GRAPH C ∗-ALGEBRAS
3
A groupoid is a set G along with a subset G(2) ⊂ G × G of composable pairs and
a two functions: composition ◦ : G(2) → G (written (α, β) → αβ) and an involution
−1 : G → G (written γ → γ−1) such that the following hold:
(i) γ(ηζ) = (γη)ζ whenever (γ, η), (η, ζ) ∈ G(2);
(ii) (γ, γ−1) ∈ G(2) for all γ ∈ G, and γ−1(γη) = η and (γη)η−1 = γ for
(γ, η) ∈ G(2).
Elements satisfying g = g2 ∈ G are called units of G and the set of all such units is
denoted G(0) ⊂ G and called the unit space of G. There are maps r, s : G → G(0)
defined by
r(γ) = γγ−1
s(γ) = γ−1γ
that are called, respectively, the range and source maps. These maps orient G as
category, with units serving as objects: (α, β) ∈ G(2) if and only if s(α) = r(β). For
a given unit u ∈ G(0) there is an associated group G(u) = {γ ∈ G : r(γ) = s(γ) =
u}; this is called the isotropy or stabilizer group of u. The union of all isotropy
groups in G forms a subgroupoid of G called Iso(G), the isotropy bundle of G. A
groupoid is called principal (or an equivalence relation) if Iso(G) = G(0); that is, if
no unit has non-trivial stabilizer group.
A topological groupoid is a groupoid G endowed with a topology so that the
composition and inversion operations are continuous (the domain of ◦ is equipped
with the relative product topology). A topological groupoid is ´etale if the topology
is locally compact and the range and source maps are local homeomorphisms. All
the groupoids we encounter in this paper will be Hausdorff and second countable.
Note that if G is ´etale then each range fiber r−1(u) is a discrete in the relative
topology (likewise for source fibers). Hence a compact subset of G meets a given
range fiber at most finitely many times.
In order to define a C ∗-algebra from an ´etale groupoid G, it is necessary to
specify a ∗-algebra structure on Cc(G). This is given by
(f ∗ g)(γ) = X
(α,β)∈G(2):αβ=γ
f (α)g(β);
compactness of supports ensures that this sum gives a well-defined element of Cc(G).
(Really the important thing here is that the counting measures on range fibers form
a Haar system, which is necessary for any topological groupoid to define a C ∗-
algebra; see [13].) We do not include all details on how to put a norm on Cc(G),
these can be found in [13]. In brief, there are two distinguished C ∗-norms · , · r
on Cc(G) and completing in these yields the full groupoid C ∗-algebra C ∗(G) and
the reduced groupoid C ∗
r (G), respectively. Our interest in groupoid C ∗-algebras
will strictly be as an alternate model for graph C ∗-algebras.
Definition 2.4. Let E be a graph and let E∞ denote its infinite path space. Then
the path groupoid of E is
GE = {(x, n, y) ∈ E∞ × Z × E∞ : ∃p, q ∈ N such that σpx = σqy and p − q = n}
The groupoid operations are (x, n, y)(y, m, z) = (x, m + n, z) and (x, n, y)−1 =
(y, −n, x). The unit space is identified with E∞ via the mapping x 7→ (x, 0, x), so
that the range and source maps are given by r(x, n, y) = x and s(x, n, y) = y.
4
DANNY CRYTSER
Remark. If G = GE, then the isotropy group of an infinite path x is either trivial
(σpx = σqx implies p = q) or infinite cyclic (in which case x = α(λ∞) for some
finite path α and cycle λ).
The topology on GE is generated by basic open sets of the form
Z(α, β) = {(αz, α − β, βz) ∈ GE : r(z) = s(α)}
where α, β ∈ E ∗ with s(α) = s(β). The topology defined above restricts to the
relative product topology on G(0)
E = E∞ ⊂ QN E1, if we treat E1 as a discrete
space. We will refer to this topology on E∞ using basic compact-open sets of the
form Z(α) = {αxx ∈ E∞, r(x) = s(α)}.
It is noted in [7] that Z(α, β) ∩ Z(γ, δ) = ∅ unless (α, β) = (γǫ, δǫ) or vice versa.
This topology makes GE into an ´etale groupoid ([7, Prop. 2.6]), because the restric-
tion of the range map to the basic sets is a homeomorphism, and furthermore each
basic set is compact. Thus GE has a canonical Haar system {λx}x∈E∞ consisting
of counting measures on the source fibers. Because GE is an ´etale groupoid, it has
an groupoid C ∗-algebra C ∗(GE) ([13]). The following theorem has been modified
from its original statement to fit our orientation convention.
Theorem 2.5 ([7, Thm. 4.2]). For any row-finite graph with no sources E, we
have C ∗(E) ∼= C ∗(GE) via an isomorphism carrying se to 1Z(e,s(e)) ∈ Cc(G) and
pv to 1Z(v,v).
(It is a fact that GE is always an amenable groupoid, so that we have C ∗(GE) =
r (GE).) To describe graph C ∗-algebras with continuous trace, we need to know
C ∗
when the isotropy groups G(u) vary continuously with respect to the unit u ∈ G(0).
First the topology on the set of isotropy groups has to be defined.
Definition 2.6 ([12]). Let X be a topological space. Consider the collection F (X)
of all closed subsets of X; the Fell topology on F (X) is defined by the requirement
that a net (Yi)i∈I ⊂ F (X) converges to Y ∈ F (X) exactly when
(1) if elements yi are chosen in Yi such that yi → z, then z belongs to Y , and
(2) for any element y ∈ Y there is a choice of elements yi ∈ Yi (possibly taking
a subnet of (Yi) and relabeling) such that yi → y.
We say that G has continuous isotropy if the isotropy map G(0) → F (G) defined
by x 7→ G(x) is continuous.
We will not need to handle the Fell topology directly in this paper, thanks to
the following result which describes continuous isotropy for graph algebras.
Theorem 2.7 ([4]). Let E be a row-finite graph with no sources. Then GE has
continuous isotropy if and only if no cycle in E has an entrance.
Definition 2.8. A topological groupoid G is proper if the orbit map ΦG : G →
G(0) × G(0) given by g → (r(g), s(g)) is proper (where the codomain is equipped
with the relative product topology).
Definition 2.9. Let G be a groupoid. Let πR : G → G(0) × G(0) be given by
πR(g) = (r(g), s(g)). Then the orbit groupoid of G, denote by RG = R, is the
image of πR, where the groupoid operations are
(u, v)(v, w) = (u, w)
(u, v)−1 = (v, u).
CONTINUOUS-TRACE k-GRAPH C ∗-ALGEBRAS
5
The unit space of R is identified with the unit space of G. The range and source
maps are naturally identified with the projections onto the first and second factors.
Definition 2.10. The topology on R = RG is the quotient topology induced by
the above map πR : G → R.
Remark. If the groupoid G is principal, in the sense that G(x) = {x} for every
unit x ∈ G(0), then the map πR is a groupoid isomorphism. If G is an arbitrary
topological groupoid, it need not be the case that the quotient topology makes
R into a topological groupoid. However, for graph and k-graph groupoids (we
will see these later on), this issue never arises and R will always be a topological
groupoid. The following commutative diagram serves to keep the relevant groupoids
and spaces distinct. Note that as a set map, ΦR is just an inclusion. However,
R carries a different topology from the product topology on G(0) × G(0), so we
distinguish between the two.
G
π
R
ΦG
ΦR
G(0) × G(0)
Remark. In some sources (such as [9]) the orbit groupoid is denoted by R = G/A, to
indicate that it is the quotient of a groupoid by the (in this case, abelian) isotropy
bundle. By Σ(0) we denote the collection of closed subgroups of G, equipped with
the Fell topology.
Theorem 2.11. [9, Thm 1.1] Let G be a second-countable locally compact Hausdorff
groupoid with unit space G(0), abelian isotropy, and Haar system {λu}u∈G(0) . Then
C ∗(G, λ) has continuous trace if and only if
(1) the stabilizer map u 7→ G(u) is continuous from G(0) to Σ(0);
(2) the action of R on G(0) is proper.
3. Continuous-trace graph algebras
The path groupoid of a directed graph E is made of infinite paths, and the
It is unsurprising then that the
open sets are described by finite path prefixes.
characterization of proper path groupoids is stated in terms of a certain finiteness
condition on paths. For this section, the standing assumption is that E is
a row-finite graph with no sources.
Definition 3.1. Let v and w be vertices in a directed graph E. An ancestry pair for
v and w is a pair of paths (λ, µ) such that r(λ) = v, r(µ) = w, and s(λ) = s(µ). A
minimal ancestry pair is an ancestry pair (λ, µ) such that (λ, µ) = (λ′ν, µ′ν) implies
that ν = s(λ) = s(µ). An ancestry pair(λ, µ) is cycle-free if neither path contains a
cycle. The graph E has finite ancestry if for every pair of vertices (not necessarily
distinct) v, v′ has at most finitely many cycle-free minimal ancestry pairs.
6
DANNY CRYTSER
Remark. Note that it is not necessary that any two vertices of E have an ancestry
pair in order for E to have finite ancestry.
Lemma 3.2. Let (α, β) and (γ, δ) be two cycle-free minimal ancestry pairs. Then
πR(Z(α, β)) ∩ πR(Z(γ, δ)) = ∅ unless there is a simple cycle λ and either factor-
izations α = γα′, δ = βδ′ and λ = α′δ′ or factorizations γ = αγ ′, β = δβ′ and
λ = γ ′β′. If (α, β) is a minimal cycle-free ancestry pair, and if no cycle of E has
an entrance, then there are finitely many minimal cycle-free ancestry pairs (γ, δ)
such that πR(Z(α, β)) ∩ πR(Z(γ, δ)) 6= ∅
Proof. We have that πR(Z(α, β)) = {(αz, βz)} ⊂ E∞ × E∞ and πR(Z(γ, δ)) =
{(γw, δw)}. Suppose that (αz, βz) = (γw, δw) as pairs of infinite paths. If α ≥ γ
and β ≥ δ, then α = γν and β = δν ′, where ν and ν ′ are initial segments of the
infinite path w. If ν = ν ′, this contradicts minimality of the ancestry pair (α, β). If
ν 6= ν ′, then (assuming WLOG that ν ⊂ ν ′) we can note that s(ν) = s(α) = s(β) =
s(ν ′), so that ν ′ contains a cycle, hence that β contains a cycle, contradicting the
assumption that (α, β) is a cycle-free ancestry pair.
We must have WLOG that α > γ and β < δ. We can factor α = γα′ and
δ = βδ′. Then λ = α′δ′ is a cycle with range/source equal to s(γ) = s(δ). This
cycle must be simple because neither α′ ⊂ α nor δ′ ⊂ δ contains a cycle. This
establishes the first claim in the lemma.
If no cycle has an entry then any vertex is the source of at most one simple cycle.
If (α, β) is a cycle-free minimal ancestry pair, and no cycle of E has an entrance,
then there is at most one simple cycle at s(α) = s(β). Thus there are only finitely
many factorizations of the above form.
(cid:3)
Lemma 3.3. Let (λ, µ) be an ancestry pair. Then there is a unique minimal
ancestry pair (α, β) such that Z(λ, µ) ⊂ Z(α, β) in GE. If (λ, µ) is cycle-free, then
so is (α, β).
Proof. Let ǫ be maximal (with respect to length) such that (α, β) = (α′ǫ, β′ǫ).
Then Z(α, β) ⊂ Z(α′, β′) 6= ∅. As mentioned in [7], Z(λ, µ) ∩ Z(γ, δ) implies that
(λ, µ) factors as (γǫ, δǫ) or vice versa. Thus Z(α, β) is contained in Z(α′, β′) and
not in Z(γ, δ) for any other minimal ancestry pair (γ, δ). The second claim follows
from the construction.
(cid:3)
Lemma 3.4. Let E be a directed graph in which no cycle has an entrance, and
let (α, β) be an ancestry pair such that α or β contains a cycle. Then there is a
cycle-free minimal ancestry pair (λ, µ) such that πR(Z(λ, µ)) contains πR(Z(α, β)).
Proof. Suppose that α contains a cycle, so that α = α′λλ′, where α′ does not
contain a cycle, λ is a cycle, and λ′ is an initial segment of λ. Let λ = λ′λ′′. Note
that α must factor in this way because of the condition that no cycle has an entrance.
Then πR(Z(α, β)) = {(αx, βx) : x ∈ E∞, r(x) = s(α)}. Again by the condition
that no cycle has an entrance, we have that the only infinite path with range equal
to s(α) is λ′′(λ∞). Thus πR(Z(α, β)) = πR(Z(α′λ′, β)). We can perform a similar
operation to get rid of any cycles from β, obtaining a cycle free ancestry pair
(α′, β′) such that πR(Z(α, β)) = πR(Z(α′, β′)). By the previous lemma we can find
a cycle-free minimal ancestry pair (λ, µ) such that Z(α′, β′) ⊂ Z(λ, µ). The lemma
is established on applying πR to this containment.
(cid:3)
Theorem 3.5. Let E be a row-finite directed graph with no sources. Then C ∗(E)
has continuous trace if and only if both
CONTINUOUS-TRACE k-GRAPH C ∗-ALGEBRAS
7
(1) no cycle of E has an entrance, and
(2) E has finite ancestry.
Proof. Suppose no cycle of E has an entrance and E has finite ancestry. We show
that ΦR is proper. The collection of sets of the form Z(v)×Z(w) ⊂ E∞×E∞ form a
compact-open cover for E∞ × E∞. Thus it suffices to prove that Φ−1
R (Z(v) × Z(w))
is compact for any vertices v and w. It is not hard to see that
G (Z(v) × Z(w)) = [
Φ−1
(α,β)∈M
Z(α, β),
where M is the set of all minimal ancestry pairs for v and w. We can partition these
pairs into two families: C, the set of minimal ancestry pairs for v and w containing
cycles and D, the set of cycle-free minimal ancestry pairs.
By definition, we have that Φ−1
R (Z(v) × Z(w)) = πR(Φ−1(Z(v) × Z(w))). Thus
we can write
πR(Φ−1(Z(v) × Z(w))) = πR
[
(α,β)∈C
Z(α, β)
∪ πR
[
(λ,µ)∈D
Z(λ, µ)
.
But for each (α, β) ∈ C, we have that πR(Z(α, β)) ⊂ πR(Z(λ, µ)) for some
(λ, µ) ∈ D. Thus Φ−1
R (Z(v) × Z(w)) = ∪(λ,µ)∈DπR(Z(λ, µ)). Each πR(Z(λ, µ))
is compact by the continuity of πR and compactness of Z(λ, µ), and D is finite
by the assumption that E has finite ancestry. Thus ΦR is a proper map and the
C ∗-algebra has continuous trace by [9].
Now suppose that C ∗(E) has continuous trace. Then the isotropy groups of G
must vary continuously so no cycle of E has an entrance. So we only need to show
that E has finite ancestry. Suppose that v and w are two vertices and let {(αk, βk) :
k ∈ A} be an enumeration of the cycle-free minimal ancestry pairs for v and w. As
in the proof of sufficiency, we can write Φ−1
R (Z(v) × Z(w)) = ∪kπR(Z(αk, βk)). By
properness of ΦR we must be able to extract a finite subcover. However for any finite
subset B ⊂ A, the set B′ = {k ∈ A : πR(Z(αk, βk)) ∩ (∪j∈B πR(Z(αj , βj))) 6= ∅} is
finite by Lemma 1. This implies that A is finite. Thus E has finite ancestry.
(cid:3)
4. Arbitrary graphs
The previous theorem is given only in the context of row-finite graphs with no
sources. In this section we will remove the requirement that all graphs be row-finite
and have no sources, by use of the Drinen-Tomforde desingularization.
Definition 4.1. A tail at the vertex v is an infinite path with range v.
Briefly, the Drinen-Tomforde desingularization adds a tail to each singular ver-
tex. If the singular vertex v is an infinite receiver, it takes all the edges with range
v and redirects each to a different vertex on the infinite tail. This produces a new
graph E which has no singular vertices. For details, see [2] or [11]. Note that
we have reversed the edge orientation of [2], to fit with the higher-rank graphs
considered in the next section.
Theorem 4.2. [2, Thm 2.11] Let E be an (arbitrary) directed graph. Let E be a
desingularization for E. Then C ∗(E) embeds in C ∗( E) as a full corner, so that
C ∗(E) is Morita equivalent to C ∗( E).
8
DANNY CRYTSER
The basic technical lemma needed is a bijection between finite paths in a singular
graph and certain finite paths in its desingularization. (We've omitted the part
about infinite paths.)
Lemma 4.3. [2, Lemma 2.6] Let E be a directed graph and let E be a desingular-
ization. Then there is a bijection
φ : E ∗ → {β ∈ E ∗ : s(β), r(β) ∈ E0}.
The map φ preserves source and range.
Lemma 4.4. Let E be a directed graph and let E be a desingularization for E.
Then no cycle of E has an entrance if and only if no cycle of E has an entrance.
Proof. Suppose that no cycle of E has an entrance. Let λ = e1 . . . en be a cycle in
E and let λ = φ(λ) = f1 . . . fm be the corresponding path in E, with r(λ) = r(λ)
and s(λ) = s(λ). Suppose that e is an edge in E with r(e) = r(ek) and yet e 6= ek.
Then e = φ(e) is a path in E with r(e) = r(φ(ek)) and yet e 6= ek (here we are
using the fact that φ is a bijection). Thus e is an entrance to the cycle λ.
Suppose that no cycle of E has an entrance, and let µ = f1f2 . . . fn be a cycle
in E. If s(µ) belongs to E0, then φ−1(µ) is a cycle in E. Furthermore, we know
that no vertex of E0 on φ−1(µ) can be singular, because then the cycle φ−1(µ)
would have an entrance. Thus µ consists solely of edges in E and does not meet
any singular vertices or tails. The only edges in E that meet µ are images under φ
of edges from E, and we know that µ has no entrances in E, so µ has no entrances.
Now we show that, under the assumption that no cycle of E has an entrance,
no cycle of E can have source on an infinite tail. Suppose that µ is a cycle in
E with source on an infinite tail. Because no infinite tail contains a cycle, we
can write µ = f1 . . . fkd1 . . . dj, where d1 . . . dj is the largest path in the infinite
tail containing s(µ) such that d1 . . . dj is contained in µ. Then r(d1) must be the
vertex to which the infinite tail is attached, i.e. r(d1) ∈ E0. Consider the cycle
µ′ = d1 . . . djf1 . . . fk. This begins and ends in E0, so it equals φ(λ) for some
cycle λ in E. This cycle cannot meet any singular vertices in E (or else it would
have an entrance), so it must be the case that λ = φ(λ). But s(µ) belongs to λ,
contradicting our assumption that s(µ) belongs to an infinite tail. Combining this
with the previous part shows that if no cycle of E has an entrance, then no cycle
of E has an entrance.
(cid:3)
Lemma 4.5. Let E be a directed graph and let E be a desingularization of E. Then
E has finite ancestry if and only if E has finite ancestry.
Proof. Suppose that E has finite ancestry and let v be a vertex of E. We show
that v has finitely many cycle free minimal ancestry pairs by defining an injection
from minimal ancestry pairs of v to minimal ancestry pairs of some vertex in E.
If v belongs to E, then it must be the case that s(α) = s(β) belongs to E0 for any
minimal ancestry pair (α, β). For otherwise s(α) lies on an infinite tail, with only
one edge f leaving s(α), and we could factor a common edge (α, β) = (α′f, β′f ).
Thus in the case that v belongs to E, we can map (α, β) 7→ (φ−1(α), φ−1(β)). This
carries cycle-free minimal ancestry pairs to cycle-free minimal ancestry pairs, so v
must have finitely many cycle free minimal ancestry pairs in F .
If v belongs to an infinite tail, let d1 . . . dj be the path from v to the singu-
if (α, β) is a min-
it must terminate in a vertex of E. Define a map
lar vertex w to which the infinite tail is attached. Again,
imal ancestry pair for v,
CONTINUOUS-TRACE k-GRAPH C ∗-ALGEBRAS
9
(α, β) 7→ (φ−1(d1 . . . djα, d1 . . . djβj)). As above, this will give us an injection into
minimal cycle free ancestry pairs for w in E. Hence v has finitely many minimal
cycle free ancestry pairs.
If E has finite ancestry then E trivially has finite ancestry by using φ.
(cid:3)
Theorem 4.6. Let E be an arbitrary graph. Then C ∗(E) has continuous trace if
and if both
(1) no cycle of E has an entrance, and
(2) E has finite ancestry.
Proof. Let us begin by fixing a desingularization E of E. If no cycle of E has an
entrance and E has finite ancestry, then Lemmas 8 and 9 tell us that the same is
true of E. Then Theorem 4 says that C ∗( E) has continuous trace. Theorem 5 and
the well-known fact that the class of continuous-trace C ∗-algebras is closed under
Morita equivalence then give that C ∗(E) has continuous trace.
Now suppose that C ∗(E) has continuous trace. Then C ∗( E) has continuous
trace as in the previous part of the proof. By Theorem 4, we see that no cycle of
E has an entrance and E has finite ancestry. Lemmas 8 and 9 again give us that
E satisfies the same conditions.
(cid:3)
Corollary 4.7. If E is a graph with no cycles, then C ∗(E) has continuous trace if
and only if E has finite ancestry.
This is useful for studying AF algebras. Drinen showed that every AF algebra
arises as the C ∗-algebra of a locally finite pointed directed graph ([1]). Tyler gave
a useful complementary result, showing that if E is Bratteli diagram for an AF
algebra A, then there is a Bratteli diagram KE for A such that (treating the
diagrams as directed graphs) C ∗(KE) contains A and C ∗(E) as complementary
full corners ([14]). Thus in particular A and C ∗(E) are Morita equivalent.
Corollary 4.8. Let A be an AF algebra and let E be a Bratteli diagram for A,
treated as a directed graph. Then A has continuous trace if and only if E has finite
ancestry.
Example 4.9. Let A = N∞
Bratteli diagram for A (after decoration with labels) is
n=1 M2(C) be the UHF algebra of type 2∞. The familiar
e1
f1
v1
e2
f2
v2
. . .
v3
This graph fails to have finite ancestry: for each k, we have the cycle-free min-
imal ancestry pair (f1e2f3 . . . e2k, e1f2e3 . . . f2k) for v1, v1. Thus it does not have
continuous trace. (As is well-known, we can actually reach a stronger conclusion,
namely that A does not have Hausdorff spectrum; see [4].)
5. Higher-rank graphs
In this section we partially extend the results of the previous section to the
realm of higher-rank graphs. We have not completely described which higher-
rank graph C ∗-algebras have continuous trace. However, we do characterize the
aperiodic higher-rank graphs which yield continuous-trace C ∗-algebras. The jump
10
DANNY CRYTSER
in combinatorial complexity from the graph to the k-graph case is noteworthy. In
addition, we provide some negative results regarding the generalized cycles of [3].
In particular, a generalized cycle with entry causes the affiliated vertex projection
to be infinite, which cannot happen if the algebra has Hausdorff spectrum.
Remark. The semigroup Nk is treated as a category with a single object, 0.
Definition 5.1. A higher-rank graph (or k-graph) is a countable category Λ equipped
with a degree functor d : Λ → Nk which satisfies the following factorization prop-
erty: if d(λ) = m + n for some m, n ∈ Nk, then λ = µν for some unique µ, ν such
that d(µ) = m and d(ν) = n. The vertices Λ0 of Λ are identified with the objects.
The elements of Λ are referred to as paths. For fixed degree n ∈ Nk, the paths
of degree n are denoted by Λn. We refer to paths of degree 0 as vertices in the
k-graph (each path in Λ) has a well-defined range vertex and source vertex.
We can affiliate a C ∗-algebra to a higher-rank graph but some additional hy-
potheses have to be added in order to ensure the result is not trivial. The hypothe-
ses we use are not the weakest set which defines a meaningful C ∗-algebra; however,
they allow us to use the groupoid machinery easily.
Definition 5.2. A higher-rank graph is row-finite if each vertex v ∈ Λ0 and degree
n ∈ Nk, there are only finitely many paths of degree n with range v. It is said to
have no sources if for all v ∈ Λ0 and n there is some path λ with d(λ) = n and
r(λ) = v.
Definition 5.3. Let Λ be a row-finite k-graph with no sources. Then the higher-
rank graph C ∗-algebra of Λ, denoted C ∗(Λ), is the universal C ∗-algebra generated
by a family of partial isometries {sλ}λ∈Λ satisfying:
λsλ = ss(λ);
(1) {sv : v ∈ Λ0} is a family of mutually orthogonal projections;
(2) if λ, µ ∈ Λ with s(λ) = r(µ), then sλsµ = sλµ;
(3) s∗
(4) for any v ∈ Λ0 and any degree n ∈ Nk, we have sv = Pλ∈Λn:r(λ)=v sλs∗
λ.
Just as in the graph case, we study continuous-trace higher-rank graph C ∗-
algebras by studying an affiliated groupoid. The following k-graph is used to define
infinite paths in k-graphs.
Definition 5.4. Let Ωk be the category of all pairs {(m, n) : m ≤ n}, where m ≤ n
if mi ≤ ni for i = 1, . . . , k. The composition is given by (m, n)(n, p) = (m, p). The
degree functor is given by d(m, n) = n − m. The objects are all pairs of the form
(m, m). If Λ is a k-graph, then an infinite path in Λ is a degree preserving functor
x : Ωk → Λ. The collection of infinite paths in Λ is denoted Λ∞.
Let Λ be a k-graph and let x be an infinite path in Λ. For any p ∈ Nk, we
define σpx to be the infinite path given by σpx(m, n) = x(m + p, n + p). The range
of an infinite path x ∈ Λ∞ is defined to be x(0, 0). If λ ∈ Λ and x ∈ Λ∞ with
s(λ) = r(x), then there is a unique path y = λx ∈ Λ∞ such that σd(λ)y = x and
y(0, d(λ)) = λ.
Now we can define the higher-rank version of the path groupoid. As noted in
[6], by the no sources assumption we know that for every vertex v ∈ Λ0, there is at
least one x ∈ Λ∞ with r(x) = v.
CONTINUOUS-TRACE k-GRAPH C ∗-ALGEBRAS
11
Definition 5.5 ([6]). Let Λ be a k-graph. Then the path groupoid of Λ is
GΛ = {(x, n, y) ∈ Λ∞ × Zk × Λ∞ : ∃p, q ∈ Nk such that σpx = σqy and p − q = n}.
The groupoid operations are given by (x, n, y)(y, m, z) = (x, m+n, z) and (x, n, y)−1 =
(y, −n, x).
The topology on GΛ is defined in the same way as the topology on GE, for E a
graph. The basic open sets are
Z(α, β) = {(x, n, y) ∈ GΛ : σd(α)(x) = σd(β)(y), d(α) − d(β) = n}.
The topology on GΛ generated by these sets makes it into a locally compact
Hausdroff ´etale groupoid over unit space Λ∞. (See [6, Prop. 2.8].) The relative
topology on the unit space can be described by open sets of the form Z(α) = {x ∈
Λ∞ : x(0, d(α)) = α}.
Theorem 5.6. [6, Cor. 3.5] Let Λ be a row-finite k-graph with no sources and let
GΛ be its path groupoid. Then C ∗(Λ) ∼= C ∗(GΛ).
Therefore we can decide questions about C ∗(Λ) by studying GΛ. Deciding when
the path groupoid GΛ has continuously varying isotropy groups is substantially
harder than the graph case.
Definition 5.7. Let Λ be a row-finite k-graph with no sources. Then Λ is said
to be tight if whenever g = (x, n, x) ∈ GΛ(x) for some x ∈ Λ∞, there must exist
α, β ∈ Λ such that
(i) g ∈ Z(α, β) (so that in particular r(α) = r(β) = r(x) and s(α) = s(β);
(ii) if y ∈ Λ∞ and r(y) = s(α), then αy = βy.
Remark. It is not difficult to see that a 1-graph E is tight if and only if no cycle of
E has an entrance.
Lemma 5.8. Let Λ be a row-finite k-graph with no sources. Then GΛ has contin-
uously varying stabilizers if and only if Λ is tight.
Proof. It is shown in [8] that an ´etale groupoid has continuously varying isotropy if
and only if the isotropy subgroupoid is open. The lemma follows immediately then
from this and the description of the topology on GΛ.
(cid:3)
We modify our definition of ancestry pair to the k-graph situation.
Definition 5.9. Let Λ be a row-finite k-graph with no sources and let v, w ∈ Λ0
be two vertices. Then an ancestry pair for v, w is a pair (λ, µ) ∈ Λ × Λ such that
r(λ) = v, r(µ) = w, and s(λ) = s(µ) = w. An ancestry pair (λ, µ) is minimal if
(λ, µ) = (λ′ν, µ′ν) implies ν = s(λ). We say that Λ has strong finite ancestry if
each pair of vertices has at most finitely many minimal ancestry pairs.
Remark. Strong finite ancestry implies finite ancestry in the 1-graph case. In fact,
a 1-graph E having strong finite ancestry is equivalent to E having finite ancestry
and E having no directed cycles.
By RΛ we denote the orbit groupoid of GΛ (again equipped with the quotient
topology from γ 7→ (r(γ), s(γ))).
Lemma 5.10. Suppose that Λ is a row-finite k-graph with no sources. If Λ has
strong finite ancestry, then RΛ is proper.
12
DANNY CRYTSER
Proof. We adopt the notation of Theorem 3.5. As in the proof of Theorem 3.5, we
see that
Φ−1
R (Z(v) × Z(w)) = ∪(α,β)∈MπR(Z(α, β))
where M is the set of minimal finite ancestry pairs for v and w. Strong finite
ancestry then implies that ΦR is proper, so that RΛ is proper.
(cid:3)
The following lemma is used to show that finite ancestry is necessary for a
(strongly aperiodic) k-graph to yield a C ∗-algebra with continuous trace. A k-
graph is called strictly aperiodic if for any x ∈ Λ∞ and p, q ∈ Nk, σpx = σqx implies
p = x (that is, if there are no periodic infinite paths in Λ). This implies that every
isotropy group in GΛ is trivial, so that the map GΛ 7→ RΛ is an isomorphism.
Lemma 5.11. Let Λ be a strictly aperiodic row-finite k-graph with no sources.
Let (α, β) and (γ, δ) be two distinct minimal ancestry pairs in Λ. Then Z(α, β) ∩
Z(γ, δ) = ∅.
Proof. Claim: It suffices to show that if Z(α, β) ∩ Z(γ, δ) 6= ∅, then either d(α) ≥
d(γ) and d(β) ≥ d(δ), or d(γ) ≥ d(α) and d(δ) ≥ d(β). For suppose that
(αz, n, βz) = (γw, n, δw), d(α) ≥ d(γ) and d(β) ≥ d(δ). Then αz = γα′z = γw, so
that α′z = w. We also have βz = δβ′z = δw, so that β′z = w. If d(α′) = d(β′),
then this shows that α′ = β′, so that (α, β) = (γα′, βα′), contradicting minimal-
ity. If d(α′) 6= d(β′), then σd(α′)w = z = σd(β ′)w, so that w is periodic, against
hypothesis. The other case follows by symmetry. This establishes the claim.
If the intersection is nonzero then as above we have (αz, n, βz) = (γw, n, δw) for
some z, w ∈ Λ∞. Moreover, n = d(α) − d(β) = d(γ) − d(δ), so that d(α) − d(γ) =
d(β) − d(δ). Thus if d(α) ≥ d(γ), we also have d(β) ≥ d(δ), reducing to the
claim. Thus we assume that d(α)1 − d(γ)1 = d(β)1 − d(δ)1 > 0 and d(γ)2 −
d(α)2 = d(δ)2 − d(β)2 > 0 (*). From the equation αz = γw and the inequality
d(α)1 > d(γ)1, we obtain α(d(γ)1, d(α))z = γ(d(γ)1, d(γ))w; call this path x1.
Similarly we have β(d(β)1, d(β))z = δ(d(δ1), d(δ))w. The conditions (*) imply that
d(α) − d(γ)1 + d(δ) − d(δ)1 = d(β) − d(δ)1 + d(γ) − d(γ)1.
Now iterating the shift we obtain
σd(δ)−d(δ)1 z = σd(α)−d(γ)1+d(δ)−d(δ)1x1 = σd(β)−d(δ)1+d(γ)−d(γ)1x1 = σd(β)−d(δ)1w.
Similarly we obtain
Now we can write
σd(α)−d(α)2w = σd(γ)−d(γ)2z.
σd(δ)−d(δ)1σd(α)−d(α)2w = σd(δ)−d(δ)1σd(γ)−d(γ)2z
= σd(γ)−d(γ)2+d(δ)−d(δ)1z = σd(γ)−d(γ)2σd(β)−d(δ)1w.
Now by strict aperiodicity we must have d(δ) − d(δ)1 + d(α) − d(α)2 = d(γ) −
d(γ)2+d(β)−d(δ)1. Compare second coordinates in the degrees; we must have d(δ)2
on the left and d(β)2 on the right. This gives d(δ)2 = d(β)2, a contradiction.
(cid:3)
Theorem 5.12. Let Λ be a row-finite k-graph with no sources.
(i) if Λ has strong finite ancestry and is tight, then C ∗(Λ) has continuous trace.
(ii) If C ∗(Λ) has continuous trace and Λ is strictly aperiodic, then Λ has strong
finite ancestry.
CONTINUOUS-TRACE k-GRAPH C ∗-ALGEBRAS
13
Proof. (i): As Λ is tight, we have that GΛ has continuous isotropy as in Lemma
5.8. Lemma 5.10 implies that RΛ is proper. Thus [9, Thm. 1.1] implies that C ∗(Λ)
has continous trace.
(ii): Because Λ is strictly aperiodic, we can identify the groupoids GΛ and RΛ.
Let v and w be two vertices of Λ; then
Φ−1(Z(v) × Z(w)) = [
(α,β)∈M
Z(α, β)
as in the proof of Theorem 3.5. Lemma 5.11 implies that the sets Z(α, β) are
pairwise disjoint and open. Thus M must be finite by compactness of Φ−1(Z(v) ×
Z(w)), which implies that Λ has strong finite ancestry.
(cid:3)
Remark. Theorem 5.12 is not as complete as Theorem 3.5; a complete description
of those k-graphs which define continuous-trace C ∗-algebras like Theorem 3.5 seems
out of reach. This is because the definition of "cycle-free" path in a k-graph does
not readily carry over or generalize to the k-graph case.
Generally the conditions discussed on k-graphs in the literature (such as the
aperiodicity condition in [6]) imply that the interior of the isotropy bundle of GΛ
coincides with G(0)
Λ (the interior of the isotropy bundle always includes the unit
space). Our definition of tight k-graphs is in some sense the opposite of this -- if Λ is
tight then the interior of the isotropy bundle is the entire isotropy bundle. Strict
aperiodicity is amounts to saying that the isotropy bundle coincides with the unit
space.
Desingularization is in general more complicated for higher-rank graphs, so it
seems perhaps unlikely that this could be easily extended to higher-rank graphs
with sources. However, we can give some necessary conditions for a k-graph to
satisfy in order that its C ∗-algebra have continuous trace. The following definition
is modified somewhat from [3]
Definition 5.13 ([3]). Let Λ be a row-finite graph with no sources. Then a pair
(λ, µ) ∈ Λ×Λ is a generalized cycle if r(λ) = r(µ) and s(λ) = s(µ) and Z(λ) ⊂ Z(µ).
We say that a generalized cycle (λ, µ) has an entrance if (µ, λ) is not a generalized
cycle. (That is, if Z(λ) ( Z(µ).)
Recall that a projection p in a C ∗-algebra A is infinite if there exists v ∈ A with
v∗v = p and vv∗ < p; that is, if it is Murray-von Neumann equivalent to a proper
subprojection of itself.
Lemma 5.14. [3, Cor. 3.8] If Λ contains a generalized cycle with entrance, then
C ∗(Λ) contains an infinite projection.
The following simple observation is probably not new, but is proven here for ease
of reference.
Lemma 5.15. If A is a C ∗-algebra containing an infinite projection, then A does
not have continuous trace.
Proof. Let p be a projection in A with a proper subprojection q such that p ∼ q.
Take an irreducible representation π : A → B(H) such that π(p − q) 6= 0. Then
π(q) < π(q) are equivalent projections in B(H). All compact projections are finite
rank, so it cannot be the case that the range of π lies within the compacts. As every
irreducible representation of a C ∗-algebra with continuous trace has range within
14
DANNY CRYTSER
the compact operators ([10, Thm. 6.1.11]), we see that A does not have continuous
trace.
(cid:3)
Corollary 5.16. If Λ is a row-finite k-graph with no sources that contains a gen-
eralized cycle with entrance, then C ∗(Λ) does not have continuous trace.
It is somewhat unsatisfactory that the question of when a higher-rank graph
yields a continuous-trace C ∗-algebra should have such a partial answer in com-
parison with the graph case. However this is somewhat in line with the case of
other C ∗-algebraic properties: it is difficult to decide when a k-graph yields an AF
algebra or a purely infinite (simple) C ∗-algebra, whereas for the graph case it is
straightforward.
References
1. D. Drinen. Viewing AF algebras as graph algebras. Proc. Amer. Math. Soc., 128(7):1991 -- 2000,
1999.
2. D. Drinen and M. Tomforde. The C ∗-algebras of arbitrary graphs. Rocky Mountain J. Math.,
35:105 -- 135, 2005.
3. D. G. Evans and A. Sims. When is the Cuntz-Krieger algebra of a higher-rank graph approx-
imately finite-dimensional? J. Funct. Anal., 263(1):183 -- 215, 2012.
4. G. Goehle. Groupoid C ∗-algebras with Hausdorff spectrum. Bull. Aust. Math. Soc., 88:232 --
242, 2013.
5. R. Hazlewood. Categorising the operator algebras of groupoids and higher-rank graphs. PhD
thesis, The University of New South Wales, 2013.
6. A. Kumjian and D. Pask. Higher-rank graph C ∗-algebras. New York J. Math., 6:1 -- 20, 2001.
7. A. Kumjian, D. Pask, I. Raeburn, and J. Renault. Graphs, groupoids, and Cuntz-Krieger
algebras. J. Funct. Anal., 144:505 -- 541, 1997.
8. S. Lalonde and D. Milan. Amenability and uniqueness for groupoids associated with inverse
semigroups. arXiv:1511.01517. November 2015.
9. P. Muhly, J. Renault, and D. Williams. Continuous-trace groupoid C ∗-algebras, III. Transac-
tions of the American Mathematical Society, 348(9):3621 -- 3641, 1996.
10. G.K. Pedersen. C ∗-Algebras and their Automorphism Groups. Academic Press, 1979.
11. I. Raeburn. Graph Algebras. CBMS Lecture Notes. American Mathematical Society, 2005.
12. I. Raeburn and D.P. Williams. Morita Equivalence and Continuous-Trace C ∗-Algebras, vol-
ume 60 of Mathematical Surveys and Monographs. American Mathematical Society, 1998.
13. J. Renault. A Groupoid Approach to C ∗-Algebras. Lecture Notes in Mathematics. Springer
Verlag, 1980.
14. Jason Tyler. Every AF-algebra is Morita equivalent to a graph algebra. Bull. Aust. Math.
Soc., 69:237 -- 240, 2004.
|
1903.02796 | 2 | 1903 | 2019-03-21T02:23:26 | The homotopy groups of the automorphism groups of Cuntz-Toeplitz algebras | [
"math.OA"
] | The Cuntz-Toeplitz algebra $E_{n+1}$ for $n\geq1$ is the universal C*-algebra generated by $n+1$ isometries with mutually orthogonal ranges. In this paper, we investigate the automorphism groups of the Cuntz-Toeplitz algebras and determine their homotopy groups. | math.OA | math |
The homotopy groups of the automorphism groups of Cuntz-Toeplitz
algebras
Taro Sogabe
Graduate School of Science
Kyoto University
Sakyo-ku, Kyoto 606-8502, Japan ∗
March 22, 2019
Abstract
The Cuntz-Toeplitz algebra En+1 for n ≥ 1 is the universal C*-algebra generated by n + 1 isometries with mutually
orthogonal ranges. In this paper, we determine the homotopy groups of the automorphism group of En+1.
1 Introduction
The Cuntz-Toeplitz algebra En+1 for n ≥ 1 is the universal C*-algebra generated by n + 1 isometries with mutually
orthogonal ranges. In this paper, we investigate the automorphism groups of the Cuntz-Toeplitz algebras and determine
their homotopy groups.
The homotopy groups of the automorphism groups are necessary to classify the continuous fields of C*-algebras.
However, there are only few classes of C*-algebras whose homotopy groups of the automorphism groups are determined.
To the best knowledge of the author's, the homotopy groups are known only for Kirchberg algebras [6, 12, 4], strongly
self-absorbing C*-algebras [9], and simple AF-algebras [22, 30]. The rough strategy of computation of the homotopy
groups in the previous work is as follows. First, we show the weak homotopy equivalence between the automorphism
group and the endomorphism semi-group. Then we compute the homotopy groups of the endomorphism semi-group
from the K-theoretic or KK-theoretic data of the C*-algebra.
We illustrate the strategy in the case of Kirchberg algebras where we have a powerful tool, Kirchberg-Phillips'
classification theorem. Regarding a continuous map ρ : X → End(A ⊗ K) as an element in Hom(A ⊗ K, C(X) ⊗ A ⊗ K),
we can associate a KK-class KK(ρ) ∈ KK(A, C(X) ⊗ A) to ρ. Therefore the homotopical data of End(A ⊗ K)
are recovered from the KK-theoretic data, and we can directly compute the general homotopy sets by the map
[X, Aut A ⊗ K] → KK(A, C(X) ⊗ A) (see [6, Proposition 5.8, Theorem 4.6]).
In general, there is no such powerful tool and the homotopy groups are computed for only exceptional classes of
C*-algebras. The strongly self-absorbing C*-algebras are such examples. Dadarlat and Pennig show in [9, Theorem
2.3] that the automorphism group Aut D is contractible for every unital strongly self-absorbing C*-algebra D. Using
a certain fibration, they determine the homotopy groups of Aut(D ⊗ K). In this paper, we use similar fibrations in
the case mentioned above.
Let {Ti}n+1
i=1 be the canonical generators of En+1 and let End0 En+1 be the path component of idEn+1 of the
i . Then the
i ∈ UEn+1(1 − e) is a homeomorphism where UEn+1 is the unitary group of
En+1. Our proof of the main result is based on the the fact that the map UEn+1 → UEn+1 (1 − e) defined by the right
multiplication by 1 − e gives a fibration with a fibre S1.
semi-group of unital endomorphisms of En+1. We denote by e the minimal projection 1 − Pn+1
map End0 En+1 ∋ ρ 7→ Pn+1
i=1 ρ(Ti)T ∗
i=1 TiT ∗
Theorem 1.1. The homotopy groups of Aut En+1 are as follows :
π1(Aut En+1) = Zn,
π2k+1(Aut En+1) = Z,
π2k(Aut En+1) = 0,
k ≥ 1.
To prove Theorem 1.1, we show that the inclusion map Aut En+1 → End0 En+1 is a weak homotopy equivalence
(Theorem 3.14).
∗[email protected]
1
Corollary 1.2. Let X be a compact CW complex. The following sequence is an exact sequence of pointed sets and
the first 4 terms give an exact sequence of groups :
H 1(X) → K 1(X) → [X, Aut En+1] → H 2(X) → [X, BAute En+1] → [X, BAut En+1] → H 3(X).
The group Aute En+1 is the subgroup of all automorphisms that fix the minimal projection e ∈ En+1.
The original motivation of this work is to investigate the structure of continuous fields of the Cuntz algebras beyond
Dadarlat's work [10] using the Cuntz-Toeplitz extensions, and we will hopefully come back to this subject in the near
future. We discuss the group structure of the homotopy sets [X, Aut En+1] and [X, Aut On+1] in [16].
We organize this paper as follows. In section 2, we give some preliminaries to compute the homotopy groups. We
introduce several fibration sequences with the help of [9, Lemma 2.8, 2.16 and Corollary 2.9]. As a consequence, the
homotopy groups of the connected component of the endomorphism semi-group, denoted by End0 En+1, are obtained.
In section 3, we show the weak homotopy equivalence of End0 En+1 and Aut En+1. The main ingredient of the
proof is Pimsner -- Popa -- Voiculescu's non-commutative Weyl -- von Neumann type theorem.
2 Preliminaries
2.1 Notation and the basic facts of the theory of C*-algebras
Let A be a unital C*-algebra and let UA be the group of unitary elements in A. We denote by U0A the path component
of 1A of UA. For a non-unital C*-algebra B, we denote its unitization by B∼. The K-groups of A are denoted by
Ki(A), i = 0, 1. We denote by [p]0 the class of the projection p in K0(A), and denote by [u]1 the class of the unitary
u in K1(A). Let SA be the suspension of A, the set of A-valued functions on [0, 1] that vanish at 0 and 1. For the
K-theory, we refer to [1, 17].
For a topological space Y and two elements y0 and y1, we denote y0 ∼h y1 in Y if there is a continuous path from
y0 to y1. Two unitaries u, v ∈ UA are homotopy equivalent if u ∼h v in UA. There is a natural map UA/ ∼h→ K1(A)
from the set of homotopy classes of unitaries to the K1-group. We say A is K1-injective if the map is injective. For
the non-unital C*-algebra B, it is K1-injective if the natural map UB∼ / ∼h→ K1(B) is injective. For example, the
algebra A ⊗ K is K1-injective by the definition of the K1-group.
We denote by K the algebra of compact operators of infinite dimensional separable Hilbert space H. For A ⊗ K,
we denote by M(A ⊗ K) the multiplier algebra of A ⊗ K, and denote by Q(A ⊗ K) its quotient by A ⊗ K. We denote
the quotient map by π. We remark that Q(A ⊗ K) is K1-injective, (see [21, Section 1.13]). We identify M(K) with
B(H) where B(H) is the algebra of the bounded operators on H. For A = C(X), we denote by C b
s∗(X, B(H)) the
set of B(H)-valued bounded continuous functions on X with respect to the strong* operator topology (abbreviated to
SOT*). This is a realization of the multiplier algebra M(C(X) ⊗ K) (see [29, Proposition 2.57]).
We refer to [13, Theorem 1] for the K-theory of the multiplier algebra, and a generalization of Kuiper's theorem.
Theorem 2.1. Let A be a unital C*-algebra. Then UM(A⊗K) is contractible with respect to the norm topology, and
we have Ki(M(A ⊗ K)) = 0, i = 1, 2.
Let A, B and C be C*-algebras. An extension C of A by B ⊗ K is an exact sequence
0 → B ⊗ K → C → A → 0,
and the Busby invariant of the extension is the induced map τ : A → Q(B ⊗ K). We refer to [1] for the definition
of the Busby invariant. The extension is called trivial if the above exact sequence splits. The extension is called
essential if τ is injective, and called unital if τ is unital. We refer to [1] for the basic facts of the theory of extensions
of C*-algebras. There are two equivalence relations of unital extensions, the strong unitary equivalence and the weak
unitary equivalence.
Definition 2.2. Let A and B be C*-algebras. Two Busby invariants τi : A → Q(B ⊗ K), i = 1, 2 are said to be
strongly unitarily equivalent if there exists a unitary U ∈ UM(B⊗K) satisfying τ1 = Adπ(U ) ◦ τ2. They are said to be
weakly unitarily equivalent if there exists a unitary u ∈ UQ(B⊗K) with τ1 = Adu ◦ τ2. We denote the strong unitary
equivalence by ∼s.u.e and denote the weak unitary equivalence by ∼w.u.e. We denote τ1 ∼s τ2 if there exists two trivial
extensions ρ1 and ρ2 satisfying τ1 ⊕ ρ1 ∼s.u.e τ2 ⊕ ρ2. We denote by Ext(A, B ⊗ K) the set of the equivalence classes
of the Busby invariants with respect to the equivalence relation ∼s.
We note that the weak unitary equivalence, ∼w.u.e induces the equivalence ∼s(see [1, Proposition 5.6.4]). In this
paper, we deal with the extensions of the Cuntz algebras by C(X) ⊗ K. One has a universal coefficient theorem of
Ext-groups.
2
Theorem 2.3 ([1, Theorem 23.1.1]). Let A and B be separable C*-algebras, with A in the bootstrap class. Then there
is an unnaturally splitting short exact sequence
0 → Mi=0,1
Ext1
Z(Ki(A), Ki(B)) → Ext(A, B ⊗ K) → Mi=0,1
Hom(Ki(A), Ki+1(B)) → 0.
If Li=0,1 Hom(Ki(A), Ki+1(B)) = 0, we have an isomorphism Ext(A, B ⊗ K) → Li=0,1 Ext1
Z(Ki(A), Ki(B)) that
sends a class of Busby invariant [τ ] of an extension 0 → B ⊗ K → Cτ → A → 0 to the class of group extension of the
commutative groups [Ki(B) → Ki(Cτ ) → Ki(A)] ∈ Ext1
Z(Ki(A), Ki(B)) for i = 0, 1.
{Ti}n+1
i=1 TiT ∗
i
i=1 (see [26, Section 1]). Then one has K ⊂ C ∗({Ti}n+1
Let En+1 be the universal C*-algebra generated by n + 1 isometries with mutually orthogonal ranges and let
It is called the Cuntz-Toeplitz algebra. The closed two-sided ideal
i=1 be the canonical generator of En+1.
generated by the minimal projection e : = 1 − Pn+1
is isomorphic to the compact operators K, which is
known to be the only closed non-trivial two-sided ideal. Consider the full Fock space F(Cn+1) and the left creations
{Ti}n+1
i=1 ) = En+1 ⊂ B(F(Cn+1)). In this paper, we frequently
identify K∼ with K + C1En+1 ⊂ B(F(Cn+1)). Let π : En+1 → On+1 be the quotient map by the ideal K, and let
Si : = π(Ti). The quotient algebra On+1 is the universal simple C*-algebra generated by n + 1 isometries with
the relation : S∗
i . We denote by O∞ the universal C*-algebra generated by the countably
infinite isometories with mutually orthogonal ranges. The algebras On+1 and O∞ are called the Cuntz algebras, whose
K-groups are the following :
j Si = δij , 1 = Pn+1
i=1 SiS∗
K0(On+1) = Zn, K1(On+1) = 0, K0(O∞) = Z, K1(O∞) = 0.
See [4, Theorem 3.7, 3.8, Corollary 3.11].
The Cuntz algebras are the Kirchberg algebras, and they tensorially absorb O∞, On+1 ⊗ O∞ ∼= On+1. The algebra
that tensorially absorbs O∞ has K1-injectivity by the lemma below.
Lemma 2.4 ([25, Lemma 2.1.7]). Let A be a unital C*-algebra. Then the natural map UA⊗O∞ / ∼h→ K1(A ⊗ O∞)
is bijective. In particular, every unital C*-algebra that tensorially absorbs O∞ is K1-injective.
Definition 2.5. We denote by τ0 the Busby invariant of the extension
0 → K → En+1 → On+1 → 0.
The inclusion map C(X) ⊗ K ֒→ C(X) ⊗ En+1 induces the Busby invariant τ = idC(X) ⊗ τ0 : C(X) ⊗ On+1 ֒→
Q(C(X) ⊗ K) of the unital essential extension
0 → C(X) ⊗ K → C(X) ⊗ En+1
idC(X)⊗π
−−−−−−→ C(X) ⊗ On+1 → 0.
Since K and En+1 are KK-equivalent to C (see [26, Theorem 4.4]) the above exact sequence induces the following
6-term exact sequence :
K0(C(X))
−n
/ K0(C(X))
ρ
/ K0(C(X) ⊗ On+1)
ind
K1(C(X) ⊗ On+1)
ρo
K1(C(X))
K1(C(X)).
−n
For a pointed topological space (X, x0), we denote by ΣX its reduced suspension with the base point x0. For pointed
topological spaces (X, x0), (Y, y0), we denote the set of the continuous maps from X to Y by Map(X, Y ) and denote
the set of base point preserving continuous maps by Map0(X, Y ). We denote the homotopy set Map(X, Y )/ ∼h by
[X, Y ] and denote Map0(X, Y )/ ∼h by [X, Y ]0. We remark that if Y is an H-space, the natural map [X, Y ]0 → [X, Y ]
is bijective.
Lemma 2.6. Let (X, x0) be a based compact Hausdorff space. Then the natural map
is a surjective isomorphism.
U(C0(X,x0)⊗On+1)∼ / ∼h→ K1(C0(X, x0) ⊗ On+1)
Proof. Since K1(On+1) = 0, we have K1(C0(X, x0) ⊗ On+1) = K1(C(X) ⊗ On+1). By Lemma 2.4, the natural map
is an isomorphism. Since UOn+1 is an H-space, we have
[X, UOn+1 ] = UC(X)⊗On+1 / ∼h→ K1(C(X) ⊗ On+1)
U(C0(X,x0)⊗On+1)∼ / ∼h= [X, UOn+1 ]0 = [X, UOn+1 ].
Therefore we have the conclusion.
3
/
/
O
O
o
o
o
Lemma 2.7 ([2, Proposition 6.6]). Let A be a C*-algebra and let I be a two-sided closed ideal of A. If A/I and I are
K1-injective and the natural map US(A/I)∼ → K1(S(A/I)) is surjective, then A is K1-injective.
We refer to [28] for the surjectivity, the properly infinite full projections and the properly infiniteness of the
C*-algebras.
Lemma 2.8 ([28, Exercise 8.9]). Let A be a unital properly infinite C*-algebra. Then the natural map UA/ ∼h→ K1(A)
is surjective.
Lemma 2.9 ([28, Exercise 4.9]). Let A be a unital C*-algebra, and let p and q be properly infinite full projections.
Then there exists a partial isometry v with p = vv∗, q = v∗v, if and only if [p]0 = [q]0 in K0(A).
We show that the algebra C(X) ⊗ En+1 is K1-injective.
Proposition 2.10. Let X be a compact Hausdorff space. Then, the map
UC(X)⊗En+1 / ∼h→ K1(C(X) ⊗ En+1)
is an isomorphism.
Proof. Surjectivity follows from the fact that C(X)⊗En+1 is properly infinite and Lemma 2.8. We identify SC0(X, x0)⊗
On+1 with C0(ΣX, x0) ⊗ On+1. Since C(X) ⊗ On+1 is K1-injective by Lemma 2.4 and C(X) ⊗ K is K1-injective,
it is sufficient to prove the surjectivity of the natural map U(SC(X)⊗On+1)∼ → K1(SC(X) ⊗ On+1). For the space
Y : = [0, 1] × X/({0} × X ⊔ {1} × X), we have SC(X) = C0(Y, y0). So we have the conclusion by Lemma 2.7.
Let End En+1 be the semi-group of unital ∗-endomorphisms of En+1. We topologize End En+1 by the point-wise
norm topology, and let End0 En+1 be the path component of idEn+1 in End En+1. We denote by Ende En+1 (resp.
Aute En+1) the subset of End En+1 (resp. Aut En+1) consisting of all elements fixing the minimal projection e. Every
automorphism of En+1 preserves the ideal of compact operators and induces an automorphism of On+1. For α in
Aut En+1, we denote by α the induced automorphism of On+1. This gives a group homomorphism Aut En+1 →
Aut On+1.
Lemma 2.11. The set End0 En+1 equals to a subset {ρ ∈ End En+1 ρ(e) is a minimal projection of K} of
End En+1, and the map
End0 En+1 ∋ ρ 7→ uρ : =
n+1
Xi=1
is a homeomorphism.
ρ(Ti)T ∗
i ∈ UEn+1 (1 − e) : = {u(1 − e) ∈ En+1 u ∈ UEn+1 }
Proof. First, we show End0 En+1 ⊂ {ρ ∈ End En+1 ρ(e) : minimal projection} because the converse is trivial.
Consequently, the map End0 En+1 ∋ ρ 7→ uρ ∈ UEn+1 (1 − e) is well-defined. If ρ(e) is a minimal projection, there
exists a partial isometry v with vv∗ = ρ(e), v∗v = e. Then the unitary v + uρ is in the path component of 1En+1 by
the K1-injectivity of En+1. We take a norm continuous path of unitaries {ut}t∈[0,1] in UEn+1 from v + uρ to 1En+1 ,
and we have the continuous path ρt : Ti 7→ utTi from ρ to idEn+1 .
Second, we show the map End0 En+1 ∋ ρ 7→ uρ ∈ UEn+1 (1−e) is a homeomorphism. For every w ∈ UEn+1(1−e), we
have the map ρw : Ti 7→ wTi by the universality of En+1. The map UEn+1 (1 − e) ∋ w 7→ ρw ∈ End0 En+1 is continuous
because {Ti}n+1
i=1 is the generator of En+1. This gives the inverse of the map End0 En+1 ∋ ρ 7→ uρ ∈ UEn+1(1 − e).
2.2 Section algebras and the theory of extensions of C*-algebras
We use the following elementary fact.
Lemma 2.12. Let A be a unital C*-algebra, and let X be a compact metrizable space. Let P1 and P2 be principal
Aut A bundles over X. Let A1 and A2 be the section algebras of the associated bundles of P1 and P2 with fibre A
respectively. Then P1 and P2 are isomorphic if and only if there exists a C(X)-linear isomorphism ϕ : A1 → A2.
Let π : M(C(X) ⊗ K) → Q(C(X) ⊗ K) be the quotient map by the ideal C(X) ⊗ K. We need the following technical
theorem of the theory of extensions of C*-algebras.
Theorem 2.13 ([27, Theorem 2.10]). Let X be a finite CW complex. Let A be a separable simple unital C*-algebra,
and let µ : A → M(C(X) ⊗ K) and σ : A → Q(C(X) ⊗ K) be unital ∗-homomorphisms. Then σ ⊕ π ◦ µ and σ are
strongly unitarily equivalent.
The theorem above is a special case of [27, Theorem 2.10]. Since A is simple, the assumptions for it are satisfied.
Lemma 2.14. Let X be a finite CW complex. Let A be a separable simple unital C*-algebra. Suppose that A has a
unital essential trivial extension π ◦ µ where µ : A ֒→ M(C(X) ⊗ K) is a unital embedding. Then two unital essential
extensions τ1 and τ2 are weakly unitarily equivalent if and only if [τ1] = [τ2] in Ext(A, C(X) ⊗ K).
4
Proof. We show that [τ1] = [τ2] implies τ1 ∼w.u.e τ2 as the other implication is always the case. By definition, there
exists a trivial extension π ◦ ρi such that τ1 ⊕ π ◦ ρ1 ∼s.u.e τ2 ⊕ π ◦ ρ2. Adding π ◦ µ to the both side, we may assume
that ρi(1A) is a properly infinite full projection in M(C(X) ⊗ K). Since K0(M(C(X) ⊗ K)) = 0 and ρi(1A) is properly
infinite full, there exists an isometry Vi with ViV ∗
2 ◦ρ2).
It follows from Theorem 2.13 that τi ∼s.u.e τi ⊕ π ◦ (AdV ∗
i = ρi(1A). Now we have τ1 ⊕π◦(AdV ∗
i ◦ ρi), and we have the conclusion.
1 ◦ρ1) ∼w.u.e τ2 ⊕π◦(AdV ∗
We have the following theorem of Paschke and Valette.
Theorem 2.15 ([32, Proposition 3], [23, Theorem 6]). Let A and B be unital separable C*-algebras, and assume
that A is nuclear. Let µ : A → M(K) be a unital embedding with µ(A) ∩ K = {0}. For the unital ∗-homomorphism
τ : = π(1B ⊗ µ), we have an isomorphism
ατ : K1(τ (A)′ ∩ Q(B ⊗ K)) → Ext(SA, B ⊗ K)
which sends the class of a unitary u ∈ τ (A)′ ∩ Q(B ⊗ K) to the class of extension
τu : SA ∋ (e2πit − 1)a 7→ (u − 1)τ (a) ∈ Q(B ⊗ K).
The following theorem holds from the argument of [24, Section 1, Theorem 1.5].
Theorem 2.16 ([24, Section 1]). Let τ1 and τ2 be unital extensions of On+1 by K. Then τ1 ∼s.u.e τ2 if and only if
τ1 ∼h τ2.
Proposition 2.17. Let X be a compact Hausdorff space with Tor(K0(C(X)), Zn) = 0, and let σ : On+1 → Q(C(X) ⊗
K) be an arbitrary unital extension. Then every element of K1(σ(On+1)′ ∩ Q(C(X) ⊗ K)) is a n-torsion element, and
the set U(σ(On+1)′ ∩Q(C(X)⊗K)) is contained in the path component of 1 of UQ(C(X)⊗K).
Proof. By Theorem 2.3, all elements of Ext(SOn+1, C(X) ⊗ K) are n-torsion elements. We define τ : = π ◦ (1C(X) ⊗ µ)
where µ : On+1 → M(K) is a unital embedding. Since On+1 is simple, we have µ(On+1) ∩ K = {0}. By Theorem 2.15,
we have
K1(τ (On+1)′ ∩ Q(C(X) ⊗ K)) ∼= Ext(SOn+1, C(X) ⊗ K).
So we have [σ⊕n] = n[σ] = [τ ] = 0, and Lemma 2.14 gives a unitary w ∈ UQ(C(X)⊗K) with σ⊕n = Adw ◦ τ . We have
an isomorphism
Adw : (σ⊕n(On+1)′ ∩ Q(C(X) ⊗ K)) ∼= (τ (On+1)′ ∩ Q(C(X) ⊗ K)).
We also have (σ(On+1)′ ∩ Q(C(X) ⊗ K)) ⊗ Mn ∼= (σ⊕n(On+1)′ ∩ Q(C(X) ⊗ K)). So we have
K1(σ(On+1)′ ∩ Q(C(X) ⊗ K)) ∼= K1(τ (On+1)′ ∩ Q(C(X) ⊗ K)).
Since K0(C(X)) = K1(Q(C(X) ⊗ K)) has no n-torsion, we have [w]1 = 0 ∈ K1(Q(C(X) ⊗ K)) for every w ∈
U(σ(On+1)′ ∩Q(C(X)⊗K)). So we have the conclusion by K1-injectivity of Q(C(X) ⊗ K) (see [21, Section 1.13]).
As an application of Lemma 2.14, we show in Proposition 2.22 that the group Aut En+1 is path connected. The
straightforward computation yields the lemma below.
Lemma 2.18. We have the following isomorphisms of K-groups and Ext-groups :
evpt∗ : Ext(On+1, C(S2m−1) ⊗ K) → Ext(On+1, K),
K1(evpt) : K1(Q(C(S2m−1) ⊗ K)) → K1(Q(K)),
K1(evpt) : K1(Q(C([0, 1]) ⊗ K)) → K1(Q(K)),
for m ≥ 1.
We need the following lemma.
Lemma 2.19 ([19, Lemma 2.3]). Let τ0 : On+1 → Q(K) be the Busby invariant in Definition 2.5. If a unitary u in
UM(K) commutes with En+1 up to compact operators (i.e. [u, d] ∈ K for every d in En+1), there exists a self adjoint
element h in Q(K) such that e2πih = π(u) and [h, a] = 0 for every a in On+1.
Corollary 2.20. The group N : = {u ∈ UM(K) [u, En+1] ⊂ K} is path connected.
Lemma 2.21. Let α be an automorphism of En+1, and Uα : = Pi α(ei1)ve1i be an implementing unitary of α ↾K
where v is a partial isometry with vv∗ = α(e), v∗v = e. Then we have α = AdUα ↾En+1
Proof. We show AdUα ↾En+1 = α. Let F ⊂ K be the set of all finite rank projections. Since α is an automorphism,
the image α(K) = K contains a net {α(p)}p∈F that weakly converges to 1. For every d ∈ En+1, we have α(p)α(d) =
α(pd) = AdUα(pd) = α(p)AdUα(d), and α = AdUα ↾En+1 holds.
5
Proposition 2.22. The group Aut En+1 is path connected.
Proof. Let α be an automorphism of En+1 and let α be an induced automorphism of On+1. Since Aut On+1 is path
connected, we take a path ht with h0 = α, h1 = idOn+1 . We take two unital essential extensions
τ1 : = idC[0,1] ⊗ τ0 : C[0, 1] ⊗ On+1 ∋ f (t) 7→ f (t) ∈ Q(C[0, 1] ⊗ K)
τ2 : = τ1 ◦ h : C[0, 1] ⊗ On+1 ∋ f (t) 7→ ht(f (t)) ∈ Q(C[0, 1] ⊗ K),
where τ0 is the Busby invariant in Definition 2.5, and we regard h : C[0, 1] ⊗ On+1 → C[0, 1] ⊗ On+1 as a C[0, 1]-
linear isomorphism. Since τ1 ∼h τ2, we have [τ1] = [τ2] in Ext(C[0, 1] ⊗ On+1, C[0, 1] ⊗ K). We have [τ1(1C[0,1] ⊗
idOn+1 )] = [τ2(1C[0,1] ⊗idOn+1 )] in Ext(On+1, C[0, 1]⊗K). By Lemma 2.14, there exists a unitary v ∈ UQ(C[0,1]⊗K) with
τ2(1C[0,1] ⊗ idOn+1 ) = Adv ◦ τ1(1C[0,1] ⊗ idOn+1 ). Since C[0, 1] is in the center of Q(C[0, 1] ⊗ K), we have τ2 = Adv ◦ τ1.
We show [v]1 = 0 in K1(Q(C[0, 1] ⊗ K)). By the construction of τ2 and v, the unitary v1 is in τ0(On+1)′ ∩ Q(K).
By Proposition 2.17, we have [v1]1 = 0 in K1(Q(K)). Since the map ev1∗ : K1(Q(C[0, 1] ⊗ K)) → K1(Q(K)) is an
isomorphism from Lemma 2.18, we have [v]1 = 0.
We take a unitary lift V ∈ UM(C[0,1]⊗K) of v. It follows that AdV is a C[0, 1]-linear isomorphism of C[0, 1] ⊗ En+1.
Therefore the map [0, 1] ∋ t 7→ AdVt ∈ Aut En+1 is continuous. Let Uα : = Pi α(ei1)ve1i be an implementing unitary
of α restricted to K where v is a partial isometry satisfying vv∗ = α(e11), v∗v = e11, and {eij } is a system of matrix
units. By Lemma 2.21, we have AdUα ↾En+1 = α. We have Adπ(V0) = h0 = α = Adπ(Uα), and it follows that
V ∗
0 Uα commutes with En+1 up to compact operators. By Corollary 2.20, the automorphism AdV ∗
0 Uα is in the path
component of idEn+1 in Aut En+1. Similary there is a continuous path from idEn+1 to AdV1 in Aut En+1. Therefore
we have
α ∼h AdV1 ◦ AdV ∗
0 Uα ∼h AdV1 ∼h idEn+1 .
2.3
Implementing unitaries of AutC(X)(C(X) ⊗ En+1)
Let AutC(X)(C(X)⊗En+1) be the group of C(X)-linear automorphisms of C(X)⊗En+1. We remark that the homotopy
set [X, Aut En+1] is identified with the set of homotopy equivalence classes of the elements of AutC(X)(C(X) ⊗ En+1).
Let G be a compact topological group. We denote by B G its classifying space, and denote by E G the universal
principal G-bundle over B G. One realization of those spaces is as follows. For a contractible space X equipped with
a free G action, the quotient map X → X/G gives the universal bundle. We refer to [14] for the basic facts about the
classifying spaces.
Let H1 be the set vectors of norm 1 in a separable Hilbert space with the norm topology. We identify H1 with the
set {f ∈ L2[0, 1] f 2 = 1}. There is a map ht : H1 × [0, 1] → H1 that sends (f, t) to (1[0,t]f + 1[t,1])/1[0,t]f + 1[t,1]2
where 1[a,b] is the characteristic function of [a, b]. This gives the deformation retraction to the set {1[0,1]}, and the
space H1 is contractible, (see [29]). The group S1 freely acts on H1 by the scalar multiplication. Therefore we
can adopt H1 as a model of E S1. We identifies B S1 with the set consisting of all minimal projections, and the map
E S1 = H1 ∋ ξ 7→ ξ ⊗ξ∗ ∈ B S1 gives the universal bundle where we denote by ξ ⊗η∗ the operator H ∋ x 7→ hx, ηiξ ∈ H
for ξ, η ∈ H. The space B S1 is the Eilenberg-Maclane space K(Z, 2) and we identifies the homotopy set [X, B S1]
with H 2(X) via the Chern classes of the line bundles.
Proposition 2.23. Let X be a compact Hausdorff space, and let α : X → Aut En+1 be a continuous map. Let η be
η∗−→ [X, B S1] is zero, then there
the map Aut En+1 ∋ α 7→ α(e) ∈ B S1. If the image of [α] by the map [X, Aut En+1]
exists a unitary U in UM(C(X)⊗K) such that AdUx = αx.
Proof. Let ξ0 be a norm 1 eigenvector corresponding to the minimal projection e. By assumption, there exists a norm
continuous section ξ : X → H1 with ξx ⊗ ξ∗
x = αx(e). Using a system of matrix units {1C(X) ⊗ eij } with e = e11 : =
ξ0 ⊗ ξ∗
0 e1i. Since ξx is norm continuous, U : X ∋ x 7→ Ux ∈ M(K)
is SOT-continuous. In particular, U : X → UM(K) is SOT*-continuous and U ∈ UM(C(X)⊗K). Lemma 2.21 shows
AdUx ↾En+1 = αx for every x ∈ X.
0 , we have a unitary Ux : = Pi αx(ei1)ξx ⊗ ξ∗
Lemma 2.24. Let X be a compact Hausdorff space and let α be an element of Map(X, Aut On+1). Then the map
α : C(X) ⊗ En+1 → C(X) ⊗ En+1 induces the identity map of the K-groups, Ki(α) = idKi(C(X)⊗En+1), i = 1, 2.
Proof. Since α is C(X)-linear, we have the commutative diagram below :
C(X)
C(X)
C(X) ⊗ En+1
α /
/ C(X) ⊗ En+1.
6
We have the conclusion from the KK-equivalence of C and En+1.
Let r : Aut En+1 → Aut K be the restriction map. Then we have a commutative diagram below
[X, Aut En+1]
r∗
/ [X, Aut K]
η∗
'❖❖❖❖❖❖❖❖❖❖❖
η∗
[X, B S1].
We remark that the map η : Aut K → B S1 gives the homotopy equivalence (see [9, Lemma 2.8]).
Lemma 2.25. The map η∗ : [Sk, Aut En+1] → [Sk, B S1] is the zero map for k ≥ 1. Hence the map r∗ : [Sk, Aut En+1] →
[Sk, Aut K] is also zero.
Proof. If k 6= 2, we have [Sk, B S1] = H 2(Sk) = 0. We show the statement in the case of k = 2. For every α in
Map(S2, Aut En+1), the map K0(α) : K0(C(S2) ⊗ En+1) → K0(C(S2) ⊗ En+1) is the identity by Lemma 2.24. Since
the map K0(C(S2) ⊗ K) −n−−→ K0(C(S2) ⊗ En+1) is injective, [e]0 = [α(e)]0 in K0(C(S2) ⊗ K). The group K0C(S2) is
generated by the Bott element, the class of the tautological line bundle (see [1, Section 9.2.10]), and the tautological
line bundle is also a generator of H 2(S2). Therefore we have η∗([α]) = 0 in H 2(S2).
2.4 Some fibration sequences
In this section, we introduce several fibrations to compute the homotopy groups of Aut En+1. We refer to [5, Chap.
6] for the definition and the basic facts about fibrations.
Definition 2.26. Let X, Y and Z be the topological spaces, and let π : X → Y be the continuous map. The map π
has the homotopy lifting property (abbreviated to HLP) for Z, if for every commuting diagram
{0} × Z
[0, 1] × Z
g
f
X
π
/ Y,
there exists a continuous map g : [0, 1] × Z → X such that g(0, z) = g(z) for every z in Z and π ◦ g = f .
The map π : X → Y is a Serre fibration, if π has HLP for every n-disc, Dn.
We remark that a Serre fibration has HLP for every CW complex. A fibration gives a long exact sequence of
homotopy sets. We denote by ΩX or Ωx0 X the loop space of the pointed set (X, x0).
Theorem 2.27. Let (Z, z0) be a pointed CW complex. Let π : (X, x0) → (Y, y0) be a Serre fibration with the fibre
F := π−1(y0). Then, there is a long exact sequence of groups (i ≥ 1), and exact sequence of pointed sets (i ≥ 0)
→ [Z, ΩiF ]0 → [Z, ΩiX]0 → [Z, ΩiY ]0 → · · · → [Z, F ]0 → [Z, X]0 → [Z, Y ]0.
In particular, we have the long exact sequence of the homotopy groups in the case of Z = {z0}.
We have the following fact.
Proposition 2.28 ([5, Theorem 6.42]). Let (X, x0), (Y, y0) be the pointed topological spaces. Then the natural map
[ΣX, Y ]0 → [X, ΩY ]0 is a bijection.
By the theorem of Hurewicz, every principal G-bundle is a fibration. Therefore we use the long exact sequence to
compute the homotopy groups of the topological group G. We refer to the argument in [9, Lemma 2.8, 2.16, Corollary
2.9] for the proof of the following 4 lemmas.
Lemma 2.29. Let p : UEn+1 → UEn+1 (1 − e) be the multiplication by 1 − e. Then, the map p is a principal S1-bundle
that has the S1 action by the right multiplication of (1 − e) + ze, z ∈ S1.
Lemma 2.30. Let H1 be the set of vectors of norm 1 of the Hilbert space H, and ξ0 ∈ H1 be a vector corresponding
the minimal projection e. The map q : UEn+1 → H1 that sends a unitary u to uξ0 is a fibration with the fiber
U(1−e)En+1(1−e).
Remark 2.31. Since H1 is contractible, it is follows from the long exact sequence of the homotopy groups induced by
the fibration of Lemma 2.30 that the map
U(1−e)En+1(1−e) ∋ w 7→ w + e ∈ UEn+1
is the weak homotopy equivalence. Hence the map Ende En+1 ∋ ρ 7→ e + Pi ρ(Ti)T ∗
equivalence.
i ∈ UEn+1 is the weak homotopy
7
'
/
/
/
/
Lemma 2.32. Let η : End0 En+1 → B S1 be the map that sends α to α(e) and let Aute En+1 be the stabilizer subgroup
of the minimal projection e. Then there is a principal Aute En+1-bundle
Aute En+1 → Aut En+1
η
−→ B S1.
Remark 2.33. Since the map η∗ : [Sk, Aut En+1] → [Sk, B S1] is the zero map by Lemma 2.25, it follows from
Lemma 2.32 that for every α in Map(Sk, Aut En+1), there exists α′ in Map(Sk, Aute En+1) that is homotopic to α in
Map(Sk, Aut En+1).
Lemma 2.34. The following sequence gives a fibration :
Ende En+1 → End0 En+1
η
−→ B S1.
Remark 2.35. In section 3, we show that the map Aut En+1 → End0 En+1 is a weak homotopy equivalence. Hence
the groups Aute En+1 and Ende En+1 are weakly homotopy equivalent from the long exact sequences and 5-lemma.
Then the map Aute En+1 ∋ α 7→ e +Pi α(Ti)T ∗
By the fibration in Lemma 2.29, we know the homotopy groups of End0 En+1.
i ∈ UEn+1 is the weak homotopy equivalence by the Remark 2.31.
Theorem 2.36. The homotopy groups of End0 En+1 are as follows:
π1(End0 En+1) = Zn, π2k+1(End0 En+1) = Z,
π2k(End0 En+1) = 0, where k ≥ 1.
Proof. By Lemma 2.11, it is sufficient to compute the homotopy groups of UEn+1 (1 − e). By the fibration sequence
we have the long exact sequence of the homotopy groups
S1 → UEn+1
p
−→ UEn+1 (1 − e),
· · · → 0 → πk(UEn+1 ) → πk(UEn+1 (1 − e)) → · · · → π1(S1) → π1(UEn+1 ) → π1(UEn+1 (1 − e)) → 0.
The map S1 → UEn+1 sends a complex number z to a unitary 1−e+ze. We have [Sk, UEn+1]0 = [Sk, UEn+1 ] = K 1(Sk)
by K1-injectivity of C(Sk) ⊗ En+1. The map
is the multiplication by −n, and so we have the conclusion.
Z = [S1, S1]0 ∋ [z] 7→ [1 − e + ze] ∈ [S1, UEn+1 ]0 = Z
We remark that a generator of π1(End0 En+1) = Zn is the canonical gauge action of S1 that is λz : Ti 7→ zTi for
every z ∈ S1.
3 The main result
3.1 The homotopy groups of Aut En+1
In this section, by using the theory of extensions, we show that the inclusion map Aut En+1 → End0 En+1 is a weak
homotopy equivalence. First, we show [S2m, Aut En+1] is trivial, for m ≥ 1. Second, we show the surjectivity of the
map [S2m−1, Aut En+1] → [S2m−1, End0En+1] for m ≥ 1. Finally, we show the injectivity of the map.
Let X be a compact Hausdorff space. It is well-known in homotopy theory that every principal Aut En+1 bundle
P over X comes from the classifying map X → BAut En+1 [14, Section 4, Proposition 10.6]. So we identifies the
equivalence class of a principal bundle [P] with the homotopy equivalence class of its classifying map and denote
[P] ∈ [X, BAut En+1]. For a bundle P, the section algebra of the associated bundle P ×Aut En+1 En+1 is a locally
trivial continuous C(X)-algebra Γ(X, P ×Aut En+1 En+1).
Let k be a natural number. For every α ∈ Map(Sk, Aut En+1), there is a principal Aut En+1-bundle Pα representing
the class [α] in [Sk, Aut En+1] ∼= [Sk+1, BAut En+1]. We construct a continuous field of En+1 over Sk+1 corresponding
. We view Sk+1 as a non-reduced suspension
to Pα as follows. We denote the interior of the closed k + 1-disc by
k+1
◦
D
of Sk, that is,
◦
D
k+1
∪ Sk ∪
◦
D
k+1
, and view α a clutching function on Sk of two trivial bundles over
k+1
◦
D
∪ Sk and
k+1
◦
D
Sk ∪
. By the following lemma, we have [Sk, Aut En+1] = [Sk+1, BAut En+1].
8
Lemma 3.1 ([14, Corollary 8.3]). Let G be a path connected group. Let X be a topological space, and let SX be its
non-reduced suspension. Then the map
[X, G] ∋ [α] 7→ [Pα] ∈ [SX, B G]
is bijective.
Definition 3.2. We identify the section algebra of Pα ×Aut En+1 En+1 with the following algebra :
Bα := {(F1, F2) ∈ (C([0, 1] × Sk) ⊗ En+1)⊕2 Fi(0) ∈ 1C(Sk ) ⊗ En+1, F1(1) = α(F2(1)) ∈ C(Sk) ⊗ En+1},
and denote by Cα the essential ideal
Cα := {(F1, F2) ∈ (C([0, 1] × Sk) ⊗ K)⊕2 Fi(0) ∈ 1C(Sk) ⊗ K, F1(1) = α(F2(1)) ∈ C(Sk) ⊗ K}.
Let Aα be the quotient algebra of Bα by Cα :
Aα := {(a1, a2) ∈ (C([0, 1] × Sk) ⊗ On+1)⊕2 ai ∈ 1C(Sk) ⊗ On+1, a1(1) = α(a2(1)) ∈ C(Sk) ⊗ On+1}.
The algebra Aα is isomorphic to the section algebra Γ(Sk+1, P α ×Aut On+1 On+1), where α is the induced map in
Map(Sk, On+1). We remark that C(Sk+1) is identified with the algebra
{(f1, f2) ∈ (C([0, 1] × Sk))⊕2 fi(0) ∈ C, f1(1) = f2(1) ∈ C(Sk)}.
which is the center of Bα. Since the map [Sk, Aut En+1] → [Sk, Aut K] is zero map by Lemma 2.25, the associated
K is trivial. We fix a trivialization and obtain θα : Cα → C(Sk+1) ⊗ K. Thus we get a unital
bundle Pα ×Aut En+1
essential extension τθα
Cα
θα
πα
Bα
θα
Aα
τθα
C(Sk+1) ⊗ K
/ M(C(Sk+1) ⊗ K)
π
/ Q(C(Sk+1) ⊗ K)
where the isomorphism θα : Cα → C(Sk+1) ⊗ K depends on the trivialization of the bundle Pα ×Aut En+1
Lemma 3.3. Let m ≥ 1 be a natural number. Then we have [S2m, Aut En+1] = 0.
K.
Proof. Since [S2m, Aut On+1] = 0 by [12, Theorem 7.4], there is a trivialization ϕα : C(S2m+1) ⊗ On+1 → Aα that is
C(S2m+1)-linear isomorphism for every α ∈ Map(S2m, Aut On+1). Consider two extensions of On+1 by C(S2m+1) ⊗ K
:
σα : = τθα ◦ ϕα(1C(S2m+1) ⊗ idOn+1 ),
σ : = 1C(S2m+1) ⊗ τ0.
where the map τ0 is the Busby invariant in Definition 2.5. It follows from the construction that [evpt ◦ σα] = [evpt ◦ σ]
in Ext(On+1, K). By Lemma 2.18, we have [σα] = [σ] in Ext(On+1, C(S2m+1) ⊗ K), and Lemma 2.14 yields that
there exists a unitary w in UQ(C(S2m+1)⊗K) satisfying Adw ◦ σ = σα. There is another unitary U in UM(K) with
Adπ(U ) ◦ evpt ◦ σ = evpt ◦ σα by Theorem 2.16. By Proposition 2.17, we have [evpt(w)]1 = [evpt(w)π(U ∗)]1 = 0
in K1Q(K), and Lemma 2.18 yields that [w]1 = 0. Therefore we have a unitary W that is a lift of w, and the
map AdW : C(S2m+1) ⊗ En+1 → θα(Bα) is a C(S2m+1)-linear isomorphism. From Lemma 2.12, the bundle Pα is
isomorphic to the trivial bundle, and we have [α] = 0 in [S2m, Aut En+1] by Lemma 3.1 and Proposition 2.22.
Lemma 3.4. Let m ≥ 1 be a natural number. Then the map
[S2m−1, Aut En+1] ∋ [α] 7→ [ α] ∈ [S2m−1, Aut On+1]
is surjective.
Proof. Let τ be the map idC(S2m−1) ⊗ τ0 where τ0 is the Busby invariant in Definition 2.5. We show that for every
γ ∈ Map(S2m−1, Aut On+1) there exists a lift Γ ∈ Map(S2m−1, Aut On+1) with Γx = γx for every x ∈ S2m−1. We
recall the notation that Γx is an induced automorphism of On+1 from Γx. For every γ in Map(S2m−1, Aut On+1), we
regard γ as an element of AutC(S2m−1)(C(S2m−1) ⊗ On+1), and there are two extensions of On+1
σγ : = τ ◦ γ(1C(S2m−1) ⊗ idOn+1 ),
σ : = τ (1C(S2m−1) ⊗ idOn+1 ).
9
/
/
/
/
/
/
For every x ∈ S2m−1, the map γx is homotopic to idOn+1 in Aut On+1 because Aut On+1 is path connected by [12,
Theorem 1.1]. Hence we have evx ◦ σγ ∼s.u.e evx ◦ σ by Theorem 2.16 because evx ◦ σγ ∼h evx ◦ σ. From Lemma
2.18, we have [σγ ] = [σ] in Ext(On+1, C(S2m−1) ⊗ K), and σγ ∼w.u.e σ by Lemma 2.14. We have two unitaries
v ∈ Q(C(S2m−1) ⊗ K) and V ∈ M(K) satisfying σγ = Adv ◦ σ and evpt ◦ σγ = Adπ(V ) ◦ evpt ◦ σ. So we have
[evpt(v)]1 = [π(V )∗evpt(v)]1 = 0 by Proposition 2.17, and Lemma 2.18 yields [v]1 = 0 ∈ K1Q(C(S2m−1) ⊗ K).
Therefore we have σγ ∼s.u.e σ, and there is a unitary Uγ ∈ UM(C(S2m−1)⊗K) with
We have the following commutative diagram
Adπ(Uγ )(τ (1 ⊗ a) = τ (γ(1 ⊗ a)), a ∈ On+1.
C(S2m−1) ⊗ En+1
AdUγ /
/ C(S2m−1) ⊗ En+1
π
π
C(S2m−1) ⊗ On+1
γ
/ C(S2m−1) ⊗ On+1.
The map Γ : S2m−1 ∋ x 7→ Ad(Uγ )x ∈ Aut En+1 is continuous and it is a lift of the map γ.
For α′
in Map(S2m−1, Aute En+1) with m ≥ 1, we take the map θα′ as follows. Let Uα′ be a unitary in
UM(C(S2m−1)⊗(1−e) K(1−e)) of the form Uα′ = Pi6=1 α′(1C(S2m−1) ⊗ ei1)(1C(S2m−1) ⊗ e1i) where {eij } is a system of
matrix units with e11 = e. By Theorem 2.1, there is a norm continuous path from 1−e to Uα′ in UM(C(Sk)⊗(1−e) K(1−e) ).
Adding the projection 1C(S2m−1) ⊗ e to the path, we have a norm continuous path U ∈ C[0, 1] ⊗ M(C(S2m−1) ⊗ K)
satisfying
Ut ∈ UM(C(S2m−1)⊗K), AdU1 ↾En+1 = α′, U0 = 1, eUt = Ute = e, t ∈ [0, 1],
where we write 1C(S2m−1) ⊗ e simply by e. We define two C(S2m)-algebras M and Mα′ :
M : = {(F1, F2) ∈ M(C([0, 1] × S2m−1) ⊗ K)⊕2 Fi(0) ∈ 1 ⊗ M(K), F1(1) = F2(1) ∈ M(C(S2m−1) ⊗ K)},
Mα′ : = {(F1, F2) ∈ M(C([0, 1] × S2m−1) ⊗ K)⊕2 Fi(0) ∈ 1 ⊗ M(K), F1(1) = AdU1(F2(1)) ∈ M(C(S2m−1) ⊗ K)}.
The algebras M and Mα′ are C(S2m)-linearly isomorphic to M(C(S2m)⊗ K) and we identify M with M(C(S2m)⊗ K).
Definition 3.5. We define a map θα′ : Mα′ → M by the C(S2m)-linear isomorphism
θα′ (F1, F2) := (F1, AdU (F2)), Fi ∈ M(C(S2m−1) ⊗ K).
The algebras Bα′ and Cα′ defined in Definition 3.2 are subalgebras of Mα′ .
We denote by l the constant map S2m−1 → {idEn+1 } and denote by l the induced map S2m−1 → {idOn+1 }. If
α′ is homotopic to l in Map(S2m−1, Ende En+1), then [ α′] = 0 in [S2m−1, Aut On+1] because of [S2m−1, End On+1] =
[S2m−1, Aut On+1] by [12, Proposition 6.1], and there is a trivialization ϕα′ : C(S2m) ⊗ On+1 → Aα′
. We can
explicitely construct ϕα′ from the homotopy between α′
Definition 3.6. Let α′ be an element of Map(S2m−1, Aute En+1) homotopic to l in Map(S2m−1, Ende En+1), and let
ht : [0, 1] × S2m−1 → Aut On+1 be a path from l = h0 to α′
= h1. Then we define the map ϕα′ as a C(S2m)-linear
isomorphism of the form
and l.
−1
−1
ϕα′ : C(S2m) ⊗ On+1 ∋ (a1(s), a2(t)) 7→ (a1(s), ht(a2(t))) ∈ Aα′ , s, t ∈ [0, 1],
where C(S2m) ⊗ On+1 is identified with the algebra
Al = {(a1, a2) ∈ (C([0, 1] × S2m−1) ⊗ On+1)⊕2 ai(0) ∈ 1C(S2m−1) ⊗ On+1, a1(1) = a2(1) ∈ C(S2m−1) ⊗ On+1}.
The map τθα′ ◦ ϕα′ is the Busby invariant of a unital essential extension of C(S2m) ⊗ On+1. The following lemma
says that τα′ ◦ ϕα′ ∼w.u.e τ = idC(S2m) ⊗ τ0 where τ0 is the Busby invariant in Definition 2.5.
Lemma 3.7. Let m ≥ 1 be a natural number and let α′ be an element of Map(S2m−1, Aute En+1) which is homotopic
to l in Map(S2m−1, Ende En+1). Let τα′ ◦ ϕα′ and τ be as above. Let i : C(S2m) ∋ (f1, f2) 7→ (f1, f2) ∈ Bα′ be the
canonical unital embedding and let j : C(S2m) ⊗ K ⊂ θα′ (Bα′ ) be the inclusion map. Then the following hold.
(1) (θα′ ◦ i)∗ : K0(C(S2m)) ∼= K0(θα′ (Bα′ )).
We denote g1 := (θα′ ◦ i)∗([1C(S2m )]0) and g2 := (θα′ ◦ i)∗(b1), where b1 is a generator of K0C(S2m).
(2) We have j∗([1C(S2m) ⊗ e]0) = −ng1, and there exists a generator b2 ∈ K0(C(S2m) ⊗ K) with j∗(b2) = −ng2.
In particular, it follows that [τα′ ◦ ϕα′ ] = [τ ] in Ext(C(S2m) ⊗ On+1, C(S2m) ⊗ K).
10
/
Proof. We identify the sphere S2m with the space (
2m
◦
D
∪ S2m−1 ∪
2m
◦
D
) where
2m
◦
D
is the interior of the 2m-disc, and
we identify C0(
2m
◦
D
) with the algebra {F ∈ C0[0, 1) ⊗ C(S2m−1) F (0) ∈ C1C(S2m−1)}. Let x0 ∈ S2m−1 be the base
point of S2m and S2m−1. The map ι0 : C0(
2m
◦
D
) ∋ F 7→ (F, 0) ∈ C0(S2m, x0) induces an isomorphism of K-groups.
An element b1 is the generator of K0(C0(Sk+1, x0)). Let ι : (C0(
embedding, and let r : Bα′ ∋ (F1, F2) 7→ F1(1) ∈ C(S2m−1) ⊗ En+1 be the restriction map.
First, we show (1). We have the following commutative diagram
2m
◦
D
) ⊗ En+1)⊕2 ∋ (F1, F2) 7→ (F1, F2) ∈ Bα′ be an
2m
◦
D
(C0(
) ⊗ En+1)⊕2
ι
/ Bα′
r
/ C(S2m−1) ⊗ En+1
.
i
2m
◦
D
)⊕2
C0(
C(S2m)
C(S2m−1)
From the KK-equivalence of En+1 and C, the vertical maps (id
⊗ 1En+1 )⊕2 and idC(S2m−1) ⊗ 1En+1 induce
isomorphisms of K-groups. Therefore the map K0(i) : Ki(C(S2m)) → Ki(Bα′ ) is an isomorphism by 6-term exact
sequences and the 5-lemma.
C0(
2m
◦
D
)
Second, we find b2. We denote by ι1 the inclusion C0(
consider the following commutative diagram
◦
D
2m
) ⊗ En+1 ∋ F1 7→ (F1, 0) ∈ (C0(
◦
D
2m
) ⊗ En+1)⊕2. We
2m
◦
D
C0(
) ⊗ K
ι1
2m
◦
D
(C0(
) ⊗ K)⊕2
θα′ ◦ι
C0(S2m, x0) ⊗ K
2m
◦
D
C0(
) ⊗ En+1
ι1 /
/ (C0(
◦
D
2m
) ⊗ En+1)⊕2
θα′ ◦ι
/ θα′ (Bα′ )
j
id⊗1En+1
2m
◦
D
)
C0(
ι0
θα′ ◦i
C0(S2m, x0).
Since θα′ ◦ ι ◦ ι1 = ι0 ⊗ idK and K0(ι0) is an isomorphism of K-groups, from diagram chasing we can find a generator
) ⊗ K) that sent to −nb1 ∈ K0(C0(S2m, x0)) by the map K0(θα′ )−1 ◦ K0(j) ◦ K0(θα′ ◦ ι ◦ ι1). Hence
◦
D
2m
b′
2 ∈ K0(C0(
we have b2 : = K0(θα′ ◦ ι ◦ ι1)(b′
2).
Third, we show j∗([1 ⊗ e]0) = −ng1. From the assumptions, there exists the map h′ : [0, 1] × Sk → Ende En+1
with h′
1 = α′, h′
0 = l. We have the unital ∗-homomorphism
η : Bα′ ∋ (F1(s), F2(t)) 7→ (F1(s), h′
t(F2(t)) ∈ Bl = C(S2m) ⊗ En+1
which sends (e, e) ∈ Bα′ to (e, e) ∈ Bl = C(S2m) ⊗ En+1. We have θ−1
(e, e) = 1C(S2m ) ⊗ e ∈ Bl = C(S2m) ⊗ En+1, and the following commutative diagram holds
α′ (j(1C(S2m ) ⊗ e)) = (e, e) ∈ Bα′ and
K0(Bα′ )
K0(η)
/ K0(Bl)
.
K0(i)
K0(id⊗1En+1
)
K0(C(S2m))
K0(C(S2m))
We have
K0(η)([θ−1
α′ (j(1C(S2m ) ⊗ e))]0) = [(e, e)]0 = −n[1Bl ]0 = K0(η)([1Bα′ ]0).
Since i∗ is an isomorphism, the map η∗ is also an isomorphism, and we have [j(1C(Sk+1)⊗e)]0 = −n[1θα′ (Bα′ )]0 = −ng1.
Finally, we show that [τα′ ◦ ϕα′ ] = [τ ]. By Theorem 2.3, we identify Ext(C(S2m) ⊗ On+1, C(S2m) ⊗ K) with
Z(K0(C(S2m) ⊗ On+1), K0(C(S2m) ⊗ K)) because k is an odd number. The element [τα′ ◦ ϕα′ ] is identified with
Ext1
the class of extension
[K0(C(S2m) ⊗ K) → K0θα′ (Bα′ ) → K0(C(S2m) ⊗ On+1)].
By the computation above, it is equal to the class [Z⊕2 −n−−→ Z⊕2 → Z⊕2
n ] = [τ ].
11
/
/
/
/
O
O
/
/
O
O
O
O
/
/
/
/
/
/
/
O
O
O
O
/
O
O
O
O
We have the Busby invariants of two extensions of On+1 by C(S2m) ⊗ K :
σα′ :=τα′ ◦ ϕα′ (1C(S2m ) ⊗ idOn+1 ),
σ :=1C(S2m ) ⊗ τ0.
From the lemma above, we have
[σα′ ] = (1 ⊗ idOn+1 )∗([τα′ ◦ ϕα′ ]) = (1 ⊗ idOn+1 )∗([τ ]) = [σ]
in Ext(On+1, C(S2m) ⊗ K). Hence there exists a unitary wα′ in UQ(C(S2m)⊗K) satisfying σα′ = Adwα′ ◦ σ by Lemma
2.14.
For a unital essential extension ν : On+1 → Q(C(S2m) ⊗ K), we take a unitary Vν introduced in [24, Section 1] :
Vν = (cid:18) 0
π(w)
ν(S)
0n+1
(cid:19) ∈ Mn+2(Q(C(S2m) ⊗ K)).
We denote (S1, · · · , Sn+1) by S. We claim that, for the above σ, we have ind([Vσ]1) = −[1C(S2m ) ⊗ e]0 ∈ K0(C(S2m) ⊗
K). Indeed, there is a unitary lift
0
w
On+2
1C(S2m ) ⊗ T
1C(S2m) ⊗ e
0n+1
0
01
1C(S2m) ⊗ T∗
0
0n+1
w∗
0n+1
σ , where T = (T1, · · · , Tn+1) and the element w is of the form 1C(S2m) ⊗ w0 and w0 : H → H ⊕n+1 is a
of Vσ ⊕ V ∗
unitary operator. The element w ∈ Mn+1,1(M(C(S2m) ⊗ K)) is a partial isometry with ww∗ = 1n+1 and w∗w = 1
in Mn+1(M(C(S2m) ⊗ K)). From direct computation of the index map, we have ind([Vσ]1) = −[1C(S2m) ⊗ e]0 ∈
K0(C(S2m) ⊗ K). Direct computation yields
Vσα′ V ∗
σ =(
n+1
Xi=1
wα′ σ(Si)w∗
α′ σ(S∗
i )) ⊕ 1n+1
S∗ (cid:19)(cid:18) w∗
Hence we have [Vσα′ ]1 − [Vσ]1 = −n[wα′ ]1 in K1(Q(C(S2m) ⊗ K)).
= (wα′ ⊕ 1n+1)(cid:18) S
0n+1
01
0
α′ ⊗ 1n+1
0
01 (cid:19)(cid:18) S∗
01
0n+1
S (cid:19) .
Recall the path h : [0, 1] × S2m−1 → Aut On+1 from l = h0 to α′
write a unitary Pn+1
We show [wα′ ]1 = 0 ∈ K1(Q(C(S2m) ⊗ K)) in Theorem 3.11, and we need the following three lemmas for that.
= h1 in Definition 3.6. Here and subsequently, we
i ∈ U(C([0,1]×S2m−1)⊗On+1) by v where we denote 1C([0,1]×S2m−1) ⊗ Si by 1 ⊗ Si
i=1 h(1 ⊗ Si)1 ⊗ S∗
for simplicity. We denote
On+2
−1
0
0
w
0n+1
W =
On+2
01
0
w∗
0n+1
∈ M2n+4(M ).
By the definition of τθα′ and ϕα′ , the following lemma holds.
Lemma 3.8. Let yα′ be an element of the form
On+2
S
yα′ : =
01
0
0n+1
On+2
01
S∗
0
0n+1
,
01
0
vS
0n+1
On+2
On+2
01
0
S∗v∗
0n+1
∈ M2n+4(Aα′ )
where we write 1C([0,1]×S2m−1) ⊗ S simply by S. Then we have
Vσα′ ⊕ V ∗
σα′ = π(W ) + τθα′ ⊗ idM2n+4 (yα′ ).
In the lemma below, we regard an element x ∈ C([0, 1] × S2m−1) ⊗ En+1 as a C(S2m−1) ⊗ En+1 valued continuous
function on [0, 1] and denote by xt, t ∈ [0, 1], and frequently write 1C(S2m−1)⊗En+1 by 1C(S2m−1) for simplicity.
12
Lemma 3.9. Let V ∈ UC([0,1], C(S2m−1)⊗En+1) be a unitary with V0 = 1C(S2m−1)⊗En+1. Then we can choose a unitary
V ∈ UC([0,1], C(S2m−1)⊗K)∼ satisfying the following
V0 =1C(S2m−1)⊗En+1 ,
1C(S2m−1)⊗En+1 −Vt ∈ C(S2m−1) ⊗ K,
V ∗
t Vt(1C(S2m−1) ⊗ e) =(1C(S2m−1) ⊗ e)V ∗
t Vt = (1C(S2m−1) ⊗ e), t ∈ [0, 1].
Proof. There is a partition 0 = t0 < t1 < · · · < tm = 1 satisfying,
Vt(1C(S2m−1) ⊗ e)V ∗
t − Vtk (1C(S2m−1) ⊗ e)V ∗
tk < 1, t ∈ [tk, tk+1].
We construct the unitary V by induction. For t ∈ [t0, t1], we have a polar decomposition
Vt(1C(S2m−1) ⊗ (1 − e))V ∗
t (1C(S2m−1) ⊗ (1 − e)) = w0
t Vt(1C(S2m−1) ⊗ (1 − e))V ∗
t (1C(S2m−1) ⊗ (1 − e))
(1)
for t ∈ [t0, t1], and there exists a unitary
V 0
t : = (cid:26) w0
w0
t + Vt(1C(S2m−1) ⊗ e), t ∈ [t0, t1]
t1 + Vt1 (1C(S2m−1) ⊗ e), t ∈ [t1, tm].
with V 0
1C(S2m−1) − V 0
0 = 1C(S2m−1). Since π(Vt(1C(S2m−1) ⊗ (1 − e))V ∗
t (1C(S2m−1) ⊗ (1 − e))) = 1C(S2m−1)⊗On+1 and (1), we have
t ∈ C(S2m−1) ⊗ K. The unitary V 0∗
V ∈ UC([0,1], C(S2m−1)⊗En+1) satisfies the following
∗
V 0
t
Vt(1C(S2m−1) ⊗ e) = (1C(S2m−1) ⊗ e)V 0
t
∗
Vt = (1C(S2m−1) ⊗ e), t ∈ [t0, t1],
∗
V 0
t
Vt(1C(S2m−1) ⊗ e)V ∗
t V 0
t − V 0
tk
∗
Vtk (1C(S2m−1) ⊗ e)V ∗
tk V 0
∗
V 0
0
V0 = 1C(S2m−1)⊗En+1,
tk < 1, t ∈ [tk, tk+1], m − 1 ≥ k ≥ 0.
The condition (2) is satisfied by the computation below
V 0
t
=(cid:26)
∗
Vt(1C(S2m−1) ⊗ e)V ∗
t V 0
∗
Vtk (1C(S2m−1) ⊗ e)V ∗
t − V 0
tk
0, t ∈ [t0, t1] ∩ [tk, tk+1]
t − Vtk (1C(S2m−1) ⊗ e)V ∗
tk V 0
tk
AdV 0
t1
∗(Vt(1C(S2m−1) ⊗ e)V ∗
tk ), t ∈ [t1, tm] ∩ [tk, tk+1].
Let l be a number with m − 1 ≥ l ≥ 0. Assume that there exist unitaries V 0, · · · , V l satisfying
1C(S2m−1)⊗En+1 − V i
U l
t
∗
(1C(S2m−1) ⊗ e) = (1C(S2m−1) ⊗ e)U l
t
t ∈ C(S2m−1) ⊗ K, V i
0 = 1C(S2m−1)⊗En+1, l ≥ i ≥ 0,
= (1C(S2m−1) ⊗ e), t ∈ [t0, tl+1],
∗
∗
U l
t
(1C(S2m−1) ⊗ e)U l
t − U l
tk
∗
(1C(S2m−1) ⊗ e)U l
tk < 1, t ∈ [tk, tk+1], m − 1 ≥ k ≥ 0,
where we denote U l
t : = V ∗
t V 0
t · · · V l
t. Now we construct a unitary V l+1 satisfying
1C(S2m−1)⊗En+1 − V l+1
t
∈ C(S2m−1) ⊗ K, V l+1
t
V l+1
∗
∗
∗
U l
t
∗
(1 ⊗ e) = (1 ⊗ e)V l+1
(1 ⊗ e)U l
tk V l+1
t
∗
∗
U l
t
tk < 1, t ∈ [tk, tk+1], m − 1 ≥ k ≥ 0,
0 = 1C(S2m−1)⊗En+1 ,
= (1 ⊗ e), t ∈ [t0, tl+2],
V l+1
t
∗
∗
U l
t
(1 ⊗ e)U l
t V l+1
t − V l+1
tk
U l
tk
(2)
(3)
(4)
(5)
(6)
where we write 1C(S2m−1) ⊗ e simply by 1 ⊗ e. By the assumption (3), we have a partial isometry wl+1 of a polar
decomposition
(1C(S2m−1) − U l
t
∗
(1 ⊗ e)U l
t )(1C(S2m−1) − U l
tl+1
∗
(1 ⊗ e)U l
tl+1 )
=wl+1
t
(1C(S2m−1) − U l
t
∗
Let V l+1 be a unitary of the form
(1 ⊗ e)U l
t )(1C(S2m−1) − U l
tl+1
∗
(1 ⊗ e)U l
tl+1 ), t ∈ [tk+1, tk+2].
(7)
V l+1
t
: =
wl+1
t + U l
t
+ U l
wl+1
tl+2
tl+2
∗
∗
1C(S2m−1), t ∈ [0, tl+1]
(1C(S2m−1) ⊗ e)U l
(1C(S2m−1) ⊗ e)U l
tl+1 , t ∈ [tl+1, tl+2]
tl+1 , t ∈ [tl+2, tm].
13
Since π((1C(S2m−1) − U l
t
construction of V l+1, we have
∗
(1 ⊗ e)U l
t )(1C(S2m−1) − U l
∗
(1 ⊗ e)U l
tl+1)) = 1C(S2m−1)⊗On+1 and (7), we have (4). By the
tl+1
V l+1∗
t U l
t
∗
(1 ⊗ e) =(1 ⊗ e)V l+1∗
∗
t U l
t
t ∈ [t0, tl+1],
=(1 ⊗ e),
V l+1∗
t U l
t
∗
(1 ⊗ e) =U l
tl+1
∗
(1 ⊗ e)U l
t U l
t
∗
(1 ⊗ e)
=(1 ⊗ e)
=(1 ⊗ e)U l
tl+1
∗
=(1 ⊗ e)V l+1
t
∗
∗
t U l
t
(1 ⊗ e)U l
U l
t
,
∗
t ∈ [tl+1, tl+2],
and V l+1 satisfies (5). For every k, m − 1 ≥ k ≥ 0, direct computation yields
∗
t U l
t
(1 ⊗ e)U l
t − V l+1∗
tk+1 U l
tk
∗
(1 ⊗ e)U l
tk V l+1
tk+1
t V l+1
0,
V l+1∗
=(cid:26)
AdV l+1
tl+2 (U l
t
∗
(1 ⊗ e)U l
t − U l
tk
(1 ⊗ e)U l
tk ), t ∈ [tl+2, tm] ∩ [tk, tk+1],
∗
t ∈ [t0, tl+2] ∩ [tk, tk+1]
and the condition (6) is satisfied by (3). Now we have a sequence of unitaries V 0, · · · , V m−1 by induction, and a
unitary V : = V 0 · · · V m−1 satisfies the assertion of the lemma.
In the sequal, we denote by uα′−1 the element Pn+1
because α′ ∈ Map(S2m−1, Aute En+1). We remark that
i=1 α′−1(Ti)T ∗
i . The element uα′ −1 is in UC(S2m−1)⊗(1−e)En+1(1−e)
π(uα′−1 ) = Xi
h1(1C(S2m−1) ⊗ Si)1C(S2m−1) ⊗ S∗
i = v1 ∈ C(S2m−1) ⊗ On+1.
We also note v ∈ U0(C([0,1]×S2m−1)⊗On+1) because v0 = 1C(S2m−1).
Lemma 3.10. There exists a unitary V ∈ UC([0,1]×S2m−1)⊗En+1 satisfying the following
π(V ) =v, V0 = 1, V1 = uα′ −1 + 1C(S2m−1) ⊗ e,
Vt(1C(S2m−1) ⊗ e) =(1C(S2m−1) ⊗ e)Vt = (1C(S2m−1) ⊗ e), t ∈ [0, 1].
In particular, an element Yα′ of the form
Yα′ : =
01
0
T
0n+1
On+2
e
0
01
T∗
0
0n+1
0
0n+1
,
01
0
V T
0n+1
On+2
e
0
01
0
0n+1
0
T∗V ∗
0n+1
∈ M2n+4(Bα′ )
is send to yα′ by the quotient map πα′ : Bα′ → Aα′ where we write 1C([0,1]×S2m−1) ⊗ T and 1C([0,1]×S2m−1) ⊗ e by T
and e respectively for simplicity.
Proof. Since v ∈ U0(C([0,1]×S2m−1)⊗On+1), one has a unitary lift V ′ ∈ U0(C(S2m−1)⊗En+1) of v with V ′
3.9, we may assume the following
0 = 1. By Lemma
t (1C(S2m−1) ⊗ e) = (1C(S2m−1) ⊗ e)V ′
V ′
t = 1C(S2m−1) ⊗ e, t ∈ [0, 1]
V ′
0 = 1C(S2m−1)⊗En+1
(8)
(9)
∗ is a unitary
Now we show that we can get the unitary V by a compact perturbation of V ′. By (8), an element uα′ −1 V ′
1
∗) = 1C(S2m−1)⊗On+1 . Since α′ is homotopic to l in Map(S2m−1, Ende En+1),
in U(C(S2m−1)⊗(1−e) K(1−e))∼ with π(uα′−1 V ′
the unitary uα′ −1 is in U0(C(S2m−1)⊗(1−e)En+1(1−e) ). Hence we have uα′ −1 +1C(S2m−1) ⊗e ∈ U0(C(S2m−1)⊗En+1). Recall
that the map K1(C(S2m−1) ⊗ K) ֒→ K1(C(S2m−1) ⊗ En+1) is injective because Tor(K1(C(S2m−1)), Zn) = 0 and the
map
1
K1(C(S2m−1) ⊗(1−e) K(1−e)) ∋ [u]1 7→ [u + 1 ⊗ e]1 ∈ K1(C(S2m−1) ⊗ K)
is an isomorphism. Therefore we have [uα′ −1 V ′
1
c : [0, 1] → U(C(S2m−1)⊗(1−e) K(1−e))∼ from uα′ −1 V ′
1
K(1−e). For every t ∈ [0, 1], we have λ(t) ∈ S1 with
∗]1 = 0 in K1(C(S2m−1) ⊗(1−e) K(1−e)), and we get a continuous path
∗ to 1C(S2m−1) ⊗ (1 − e) by the K1-injectivity of C(S2m−1) ⊗(1−e)
λ(0) = λ(1) = 1, λ(t)1C(S2m−1) ⊗ (1 − e) − c(t) ∈ C(S2m−1) ⊗(1−e) K(1−e)
14
∗) = 1C(S2m−1)⊗On+1 , and the function λ : [0, 1] → S1 is continuous. Now we get the element
by (9) and π(uα′−1 V ′
1
V : = (¯λc+1C([0,1]×S2m−1)⊗e)V ′ ∈ U(C([0,1]×S2m−1)⊗En+1) satisfying the assertion of the lemma. Since V1(1C(S2m−1)⊗
T) = uα′−1 (1C(S2m−1) ⊗ T) = α′−1(1C(S2m−1) ⊗ T) and α′(1C(S2m−1) ⊗ e) = (1C(S2m−1) ⊗ e), direct computation yields
α′ ⊗ idM2n+4
01
0
V T
0n+1
On+2
e
0
01
0
0n+1
0
T∗V ∗
0n+1
=
01
0
T
0n+1
On+2
e
0
01
T∗
0
0n+1
0
0n+1
.
Hence Yα′ is an element of M2n+4(Bα′ ) that is sent to yα′ by the quotient map πα′ : Bα′ → Aα′ .
We remark that Yα′ is a partial isometry. We have θα′ (Yα′ )θα′ (Yα′ )∗ = 1⊕0n+1 ⊕01 ⊕1n+1 and θα′ (Yα′ )∗θα′ (Yα′ ) =
01 ⊕ 1n+1 ⊕ 1 ⊕ 0n+1. Recall that W is a partial isometry with W W ∗ = 01 ⊕ 1n+1 ⊕ 1 ⊕ 0n+1 and W ∗W =
1 ⊕ 0n+1 ⊕ 01 ⊕ 1n+1. Therefore an element W + θα′ (Yα′ ) is a unitary in UM2n+4(M ).
Theorem 3.11. Let m ≥ 1 be a natural number. Let α′ be an element of Map(S2m−1, Aute En+1) that is homotopic to l
in Map(S2m−1, Ende En+1). Let wα′ , Yα′ and Vσα′ be as mentioned above. Then we have [wα′ ]1 = 0 in K1(Q(C(S2m)⊗
K)).
Proof. Recall the following commutative diagram
Bα′
Mα′
θα′
M
πα′
/ Aα′
τθ
α′
π
/ Q(C(S2m) ⊗ K).
By Lemma 3.8 and Lemma 3.10, we have
Vσα′ ⊕ V ∗
σα′ :=π ⊗ idM2n+4 (W ) + τθalpha′ ⊗ idM2n+4 (yα′ )
=π ⊗ idM2n+4 (W + θα′ ⊗ idM2n+4 (Yα′ )),
so W + θα′ (Yα′ ) is a unitary lift of Vσα′ ⊕ V ∗
σα′ . Let P be a projection of the form
P : = (W + θα′ (Yα′ ))(1n+2 ⊕ 0n+2)(W + θα′ (Yα′ ))∗.
σ ]1 = ind[Vσα′ ]1 − ind[Vσ]1 = [P ]0 + [1C(S2m ) ⊗ e]0 − [1n+2]0 ∈ K0((C(S2m) ⊗ K)∼) and we show
We have ind[Vσα′ V ∗
that the index is 0. Recall Vt(1C(S2m−1) ⊗ e) = (1C(S2m−1) ⊗ e)Vt = (1C(S2m−1) ⊗ e) and Ut(1C(S2m−1) ⊗ e) =
(1C(S2m−1) ⊗ e)Ut = (1C(S2m−1) ⊗ e) by Definition 3.5. Direct computation yields
P =W (1n+2 ⊕ 0n+2)W ∗ + θα′ ⊗ idM2n+4 (Yα′ (1n+2 ⊕ 0n+2)Y ∗
α′ )
=01 ⊕ 1n+1 ⊕ 0n+2 + (1 − e) ⊕ 0n+1 ⊕ 01 ⊕ 0n+1
Now we have P = (1 − e) ⊕ 1n+1 ⊕ 0n+2 and get ind[Vσα′ V ∗
have −n[wα′ ]1 = [Vσα′ V ∗
σ ]1 = 0 and this proves the theorem because Tor(K1(Q(C(S2m) ⊗ K)), Zn) = 0.
σ ]1 = [P ]0 + [1C(S2m ) ⊗ e]0 − [1n+2]0 = 0. Therefore we
Corollary 3.12. For the above α′, we have [α′] = 0 in [S2m−1, Aut En+1].
Proof. Since [wα′ ]1 = 0, there is a unitary Wα′ in UM(C(S2m )⊗K) that is a lift of wα′ . Therefore we have C(S2m)-linear
isomorphism AdWα′ : Bl → Bα′ , and Pl ∼= Pα′ from Lemma 2.12. By Lemma 3.1, we have [α′] = [l] = 0.
Lemma 3.13. Let m ≥ 1 be a natural number. Let α be an element in Map(S2m−1, Aute En+1).
Map(S2m−1, End0 En+1), then there exists α′ in Map(S2m−1, Aute En+1) satisfying the following :
If α ∼h l in
α′ ∼h l in Map(S2m−1, Ende En+1), α ∼h α′ in Map(S2m−1, Aut En+1).
15
/
/
Proof. It follows from Lemma 2.29 that there is an exact sequence
[S2m, B S1] → [S2m−1, Ende En+1] → [S2m−1, End0 En+1] → [S2m−1, B S1],
and by Remark 2.31 the map [S2m−1, Ende En+1] → [S2m−1, UEn+1 ] which sends [α] to [uα] := [e +Pn+1
i ] ∈
[S1, UEn+1 ] is an isomorphism. Since B S1 is K(Z, 2) space, if m ≥ 2 and α ∼h l in Map(S2m−1, End0 En+1), then we
have [α] = 0 in [S2m−1, Ende En+1] because [S2m−1, Ende En+1] = [S2m−1, End0 En+1], m ≥ 2.
Hence it is sufficient to show the claim in the case of m = 1. Let α be an element of Map(S1, Aute En+1) with α ∼h l in
Map(S1, End0 En+1). Computation in Theorem 2.36 yields that there exists d ∈ Z with [uα] = −nd ∈ [S1, UEn+1 ] = Z.
We define ρd by
i=1 αx(Ti)T ∗
ρd : = Ad(¯zdT1T ∗
1 + (1 − T1T ∗
1 )) ∈ Map(S1, Aute En+1).
By Lemma 2.10, there is a continuous path from (¯zdT1T ∗
1 )) to ¯zd in UC(S1)⊗En+1, and ρd ∼h Ad¯zd = l
in Map(S1, Aut En+1). We have [uρdα]1 = [ρd(uα)]1 + [uρd ]1 in K1(C(S1) ⊗ En+1) = [S1, UEn+1]. By Lemma 2.24, it
follows that [ρd(uα)]1 = [uα]1. Hence we have [uρdα]1 = −nd + [uρd ]1. The following computations yield [uρd ]1 = nd :
1 + (1 − T1T ∗
n+1
uρd = e +
Xi=1
= (¯zdT1T ∗
(¯zdT1T ∗
1 + (1 − T1T ∗
1 ))Ti(zdT1T ∗
1 + (1 − T1T ∗
1 ))T ∗
i )
1 + (1 − T1T ∗
1 ))(e +
n+1
Xi=1
Ti(zdT1T ∗
1 + (1 − T1T ∗
1 ))T ∗
i ),
(cid:18) e +Pn+1
=(cid:18) T
i=1 Ti(zdT1T ∗
1 + (1 − T1T ∗
0
1 ))T ∗
i )
0
1n+1 (cid:19)
0n+1 T∗ (cid:19)(cid:18) (zdT1T ∗
e
0
1 + (1 − T1T ∗
1 )) ⊗ 1n+1
0
1 (cid:19)(cid:18) T∗
e
0n+1
T (cid:19) .
Therefore we have [uρdα] = 0 in [S1, UEn+1 ]. By Remark 2.31, we have the isomorphism [S1, Ende En+1] ∋ ρdα 7→
[uρdα] ∈ [S1, UEn+1 ], and α′ : = ρdα satisfies all assumptions of the lemma.
We show the weak homotopy equivalence.
Theorem 3.14. The inclusion map Aut En+1 → End0 En+1 is a weak homotopy equivalence.
Proof. By Lemma 3.3 and Theorem 2.36, we consider only the case of odd homotopy groups. Let k be an odd number.
First, we show the map [Sk, Aut En+1] → [Sk, End0 En+1] is injective. If α in Map(Sk, Aut En+1) is homotopic to
l in Map(Sk, End0 En+1), we may assume that there exists α′ ∈ Map(Sk, Aute En+1) homotopic to α by Remark 2.33.
From Lemma 3.13, we may assume that α′ ∼h l in Map(Sk, Ende En+1), and we have [α′] = [α] = 0 in [Sk, Aut En+1]
by Corollary 3.12. Therefore the map [Sk, Aut En+1] → [Sk, End0 En+1] is injective.
Second, we show the surjectivity. The following commutative diagram holds
[Sk, Aut En+1]
Lem 3.4
[Sk, Aut On+1]
[Sk, End0 En+1]
/ [Sk, End On+1].
In the case of k = 1, we have [S1, End0 En+1] = [S1, End On+1] = Zn because the generators of the both groups are
constructed from canonical gauge actions of S1 that are of the form λz : Ti 7→ zTi and λz : Si 7→ zSi. Therefore the
surjectivity follows from Lemma 3.4.
In the case of k ≥ 3, the map [Sk, UEn+1 ] → [Sk, End0 En+1] = Z in Theorem 2.36 is an isomorphism. Therefore
the map Z = [Sk, End0 En+1] → [Sk, End On+1] = [Sk, UOn+1 ] = Zn is the quotient by nZ. Hence the image of the
map [Sk, Aut En+1] → [Sk, End0 En+1] = Z contains an element nd + 1 for some d ∈ Z by Lemma 3.4.
On the other hand, we show that the image contains nZ. For every V ∈ UC(Sk)⊗En+1 , there exists V ′ ∈
UC(Sk)⊗En+1 with V ′(1C(Sk) ⊗ e) = (1C(Sk) ⊗ e)V ′ = (1C(Sk ) ⊗ e) which is homotopic to V in UC(Sk)⊗En+1 by
Remark 2.31.
Since the isomorphism [Sk, UEn+1 ] → [Sk, End1
0 En+1] sends
−n[V ]1 = [1C(Sk ) ⊗ e +
V ′(1 ⊗ Ti)V ′∗(1 ⊗ T ∗
i )]1
n+1
Xi=1
16
/
/
/
/
/
to [AdV ′] = [AdV ], the subset
is mapped onto the subset
{[AdV ] ∈ [Sk, Aut En+1] V ∈ UC(Sk)⊗En+1}
{−n[V ]1 ∈ K1(C(Sk) ⊗ En+1) V ∈ U(C(Sk)⊗En+1 } = nZ ⊂ [Sk, End0 En+1] = Z.
Therefore the image contains nd + 1 and nZ, and we have the conclusion.
3.2 An exact sequence of homotopy sets
We have the principal Aute En+1-bundle Aute En+1
bundle and denote by r the restriction map Aut En+1 → Aut K. In this section, we show the following theorem.
η
−→ B S1. We denote by f the classifying map of the
i−→ Aut En+1
Theorem 3.15. Let X be a compact CW-complex. Then we have the following exact sequence of the pointed set where
first 4-terms gives the exact sequence of the groups :
H 1(X) → K 1(X) → [X, Aut En+1]
η∗−→ H 2(X)
f∗−→ [X, BAute En+1]
Bi∗−−→ [X, BAut En+1]
Br∗−−→ H 3(X).
It follows that Im η∗ ⊂ Tor(H 2(X), Zn) and Im Br∗ ⊂ Tor(H 3(X), Zn).
The following lemma is well known in the homotopy theory. We refer to [20, Chap 3, Section 6]
Lemma 3.16. Let X be a CW-complex. Let G be a topological group and let H be a subgroup of G such that H →
G → G/H is a principal H-bundle. Suppose that G/H has a homotopy type of a CW-complex. Let f : G/H → B H
be its classifying map. Then we have the exact sequence of pointed sets :
[X, G] → [X, G/H]
f∗−→ [X, B H] → [X, B G].
Since B S1 has a homotopy type of a CW-complex, we can apply the above lemma to Aute En+1 → Aut En+1 →
B S1.
Lemma 3.17. Let X be a CW-complex. The following sequence of pointed sets is exact :
[X, BAute En+1]
Bi∗−−→ [X, BAut En+1]
Br∗−−→ [X, BAut K].
Proof. The group Aute K is identified with the group UM(K) by the map taking the implementing unitary Uα =
K. Hence it is contractible, and [X, BAute K] = {pt}. From the commutative diagram
Pi6=1 α(ei1)e1i for α ∈ Aute11
below,
[X, BAute En+1]
[X, BAut En+1]
/ [X, BAut K]
(❘❘❘❘❘❘❘❘❘❘❘❘❘
6♥♥♥♥♥♥♥♥♥♥♥♥
[X, BAute K],
Br∗ ◦ Bi∗ is trivial. Therefore it is sufficient to prove that for every P ∈ Map(X, BAut En+1) with the trivial
associated bundle P ×Aut En+1 Aut K, the structure group of P is reduced to Aute En+1. Let P ∈ Map(X, BAut En+1)
be a principal Aut En+1-bundle with the trivial associated bundle P ×Aut En+1 Aut K. We take an open covering {Ui}
of X giving a local trivialization of P, and denote by φji : Uj ∪ Ui → Aut En+1 the transition function. By the
assumption, there exists the map hi : Ui × Aut K → Ui × Aut K that is compatible with the transition functions, and
is equivariant with respect to the right multiplication of Aut K. The diagram below holds
Ui ∩ Uj × Aut K
r(φji)
Uj ∩ Ui × Aut K
hi
hj
/ Ui ∩ Uj × Aut K
/ Uj ∩ Ui × Aut K.
We also denote by φji the map
Ui ∩ Uj × Aut K ∋ (x, α) 7→ (x, r(φji(x))α) ∈ Uj ∩ Ui × Aut K.
We denote hi(x) := P ri(hi(x, id)) where P ri : Ui ×Aut K → Aut K. Since hi is equivariant, we have h−1
We have hj (x)r(φji(x))h−1
an appropriate refinement of {Ui}, we may assume that for every i, there exists xi ∈ Ui satisfying h−1
h−1
i
(x) = hi(x)−1.
If we take
(x)(e) −
i (x) that is the sum of partial isometries constructed from the polar
(xi)(e) < 1, x ∈ Ui. There is a unitary V ′
(x)(e) = e because hj ◦ φji ◦ h−1
(x, id) = (x, id) for every x ∈ Uj ∩ Ui.
i
i
i
i
17
/
/
(
/
6
/
/
decomposition of h−1
holds. We fix a unitary Wi ∈ UK1 with WieW ∗
Vi(x)eVi(x)∗ = h−1
i = h−1
i
(x)(e). The correction of the map
(xi)(e) and (1 − h−1
(x)(e)h−1
i
i
i
(x)(e))(1 − h−1(xi)(e)), and V ′
(xi)(e)V ′
(xi)(e). Then we have a unitary Vi(x) = V ′
i (x)h−1
i
i (x)∗ = h−1
(x)(e)
i (x)Wi ∈ UK∼ with
i
i
gives the following :
ui : Ui × Aut En+1 ∋ (x, α) 7→ (x, AdViα) ∈ Ui × Aut En+1
Ui ∩ Uj × Aut En+1
φji
Uj ∩ Ui × Aut En+1
ui
uj
Ui ∩ Uj × Aut En+1
φji
Uj ∩ Ui × Aut En+1,
φji : (x, α) 7→ (x, AdVj(x)∗φji(x)AdVi(x)α).
Uj ∩ Ui ∋ x 7→ AdVj(x)∗φji(x)AdVi(x) ∈ Aute En+1
where φji is of the form
We have the transition function
by the computation below :
AdVj(x)∗φj i(x)AdVi(x)(e) =AdVj (x)∗φji(x)h−1
j (x)(e)
j (x)(e)
=AdVj (x)∗h−1
=AdVj (x)∗AdVj (x)(e)
=e.
Therefore the structure group of P is reduced to Aute En+1.
Lemma 3.18. Let X be a compact Hausdorff space. The map Aute En+1 ∋ α 7→ e +Pn+1
a group isomorphism [X, Aute En+1] → [X, UEn+1 ] = K 1(X).
i=1 α(Ti)T ∗
i ∈ UEn+1 induces
Proof. By Remark 2.35 and Theorem 3.14, the map is bijective. So we show that it is a group homomorphism. Let
i ∈
α and β be elements of Map(X, Aute En+1), and we denote uα : = 1C(X) ⊗ e + Pn+1
UC(X)⊗En+1 . We show [uαβ]1 = [uα]1 + [uβ]1. Since α and β fix e, direct computation yields
i=1 α(1C(X) ⊗ Ti)1C(X) ⊗ T ∗
uαβ =α(e +Xi
=α(uβ)uα.
β(1C(X) ⊗ Ti)(1C(X) ⊗ T ∗
i ))(e +Xi
α(1C(X) ⊗ Ti)(1C(X) ⊗ T ∗
i ))
By Lemma 2.24, we have [α(uβ)]1 = K1(α)([uβ]1) = [uβ]1.
We need the following fact to determine the second cohomology group of Aut En+1. See Allen Hatcher's unpublished
book [15, Proposition 5.11].
Proposition 3.19. Let X be a path connected space with finite homotopy groups. Then its homology group Hn(X) is
finite for all n > 0.
Lemma 3.20. We have the following cohomology groups :
H 2(Aut En+1) = Zn, H 3(BAut En+1) = Zn.
Proof. Two spaces Aut En+1 and Aut On+1 are path connected and the map Aut En+1 → Aut On+1 gives
So we have
πi(Aut En+1) ∼=πi(Aut On+1) i = 0, 1, 2
π3(Aut En+1) ։π3(Aut On+1).
Hi(Aut En+1) ∼=Hi(Aut On+1) i = 0, 1, 2
H3(Aut En+1) ։H3(Aut On+1).
by Whitehead's theorem (see [5, Corollary 6.69]). By the universal coefficient theorem, we have
H 2(Aut En+1) ∼= f ree(H2(Aut En+1)) ⊕ Tor(H1(Aut En+1))
where f ree(H2(Aut En+1)) is the free part of the homology group. By Proposition 3.2, the homology groups of
Aut On+1 are finite, and f ree(H2(Aut En+1)) = 0. So we have H 2(Aut En+1) ∼= Tor(H1(Aut On+1)). Hurewicz'
theorem ( [5, Theorem 6.66]) yields H1(Aut On+1) = π1(Aut On+1) = Zn. Similarly, we have H 3(BAut En+1) =
Zn.
18
o
o
o
o
Now we prove Theorem 3.15.
Proof of Theorem 3.15. By Lemma 3.16 and the long exact sequences of the principal bundle Aute En+1
B S1, we have an exact sequence of pointed set where first 4-terms gives the exact sequence of the groups :
i−→ Aut En+1
η
−→
H 1(X) = [X, S1] → [X, Aute En+1] → [X, Aut En+1]
η∗−→ H 2(X)
f∗−→ [X, BAute En+1]
Bi∗−−→ [X, BAut En+1].
By Lemma 3.18, and Lemma 3.17, we have the exact sequence :
H 1(X) → K 1(X) → [X, Aut En+1]
η∗−→ H 2(X)
f∗−→ [X, BAute En+1]
Bi∗−−→ [X, BAut En+1]
Br∗−−→ H 3(X),
where we identify H 3(X) with [X, BAut K] because BAut K is the K(Z, 3)-space. We identify [Aut En+1, B S1] with
H 2(Aut En+1). For every [α] ∈ [X, Aut En+1], it follows that η∗([α]) = α∗([η]), and the element α∗([η]) is in the image
of the map
α∗ : H 2(Aut En+1) = [Aut En+1, B S1] ∋ [η] 7→ [η ◦ α] ∈ [X, B S1] = H 2(X).
Therefore we have Im η∗ ⊂ Tor(H 2(X), Zn) from Lemma 3.20. Similar argument yields Im Br∗ ⊂ Tor(H 3(X), Zn).
Acknowledgements
The author would like to show his greatest appreciation to his supervisor Prof. Masaki Izumi who gave many insightful
comments and suggestions, and patiently checked his arguments.
References
[1] B. Blackadar, K-theory for operator algebras, 2nd ed., Math. Sci. Inst. Publ., vol. 5, Cambridge University Press,
Cambridge, 1998.
[2] L. G. Brown and G. K. Pedersen, Non-stable K-theory and extremally rich C*-algebras, J. Functional Analysis
267 (2014), 262 -- 298.
[3] J. Cuntz, K-theory for cerain C*-algebras, Ann. of Math. (2) 113 : 1 (1981), 181 -- 197.
[4] J. Cuntz, On the homotopy groups of the space of endomorphisms of a C*-algebra (with applications to topological
Markov chains), Operator algebras and group representations, vol. I (Neptun, 1980), 124 -- 137, Monogr. Stud.
Math. 17, Pitman, Boston, MA, 1984.
[5] J. F. Davis and P. Kirk, Lecture notes in algebraic topology, Graduate Studies in Mathematics, 35. Amer. Math.
Soc, Providence, RI, 2001.
[6] M. Dadarlat, The homotopy groups of the automorphism groups of Kirchberg algebras, J. Noncommut. Geom. 1
(2007), no. 1, 113 -- 189.
[7] J. Dixmier, Les C*-algebres et leurs representations, 2nd ed., Cahiers Scientifiques 29, Gauthier-Villars, Paris,
1969. Reprinted by Editions Jacques Gabay, Paris, 1996. Translated as C*-algebras, North-Holland, Amsterdam,
1977.
[8] J. Dixmier and A. Douady, Champs continus d'espaces hilbertiens et de C*-algebres, Bull. Soc. Math. France 91
(1963),227 -- 284.
[9] M. Dadarlat and U. Pennig, A Dixmier-Douady theory for strongly self-absorbing C*-algebras, J. Reine Angew.
Math. 718 (2016), 153 -- 181.
[10] M. Dadarlat, The C*-algebra of a vector bundle, J. Reine Angew. Math. 670 (2012), 121 -- 143.
[11] M. Dadarlat and W. Winter, On the KK-theory of strongly self-absorbing C*-algebras, Math. Scand. 104 (2009),
no. 1, 95 -- 107.
[12] M. Dadarlat, Continuous fields of C*-algebras over finite dimensional spaces, Adv. Math. 222 (2009), no. 5,
1850 -- 1881.
[13] N. Higson and J. Cuntz, Kuiper's theorem for Hilbert modules, Contemporary Mathematics 62, 1987.
[14] D. Husemoller, Fibre bundles third edition., Grad. Texts Math. 20, Springer-Verlag, New York 1994.
[15] A. Hatcher, Spectral sequences, preprint.
[16] M. Izumi and T. Sogabe, The group structure of the homotopy set whose target is the automorphism group of
the Cuntz algebra, preprint.
19
[17] M. Karoubi, K-theory, Grundl. Math. Wiss. 226, Springer-Verlag, Berlin 1978.
[18] T. Katsura, On C*-algebras associated with C*-correspondences, Journal of Functional Analysis. 217 (2004),
366 -- 401.
[19] H. Lin, On the classification of C*-algebras of real rank zero with zero K1, J. Operator theory 35 (1996), 147 -- 178.
[20] M. Mimura and H. Toda, Topology of Lie groups I and II, Translations of Mathematical Monographs, vol 91.
Amer. Math. Soc, 1978.
[21] J. A. Mingo, K-theory and multipliers of stable C*-algebras, Trans. Amer. Math. Soc. 299 : 1 (1987), 397 -- 411.
[22] V. Nistor, On the homotopy groups of the automorphism groups of AF-C*-algebras, J. Operator theory 19 (1988),
319 -- 340.
[23] W. L. Paschke, K-theory for commutants in the Calkin algebra, Pacific J. Math. 95 : 2 (1981), 427 -- 434.
[24] W. L. Paschke and Salinas, Matrix algebras over On+1, Michigan Math. J. 26 : 1 (1979), 3 -- 12.
[25] N. C. Phillips, A classification theorem for nuclear purely infinite simple C*-algebras, Doc. Math. 5 (2000),
49 -- 114.
[26] M. Pimsner, A class of C*-algebras generalizing both Cuntz-Krieger algebras and crossed products by Z, Fields
Inst. Commun. 12 (1997), 189 -- 212.
[27] M. Pimsner, S. Popa and D. Voiculescu, Homogenious C*-extensions of C(X) ⊗ K(H). Part I, J.Operator theory
1 (1979), 55 -- 108.
[28] M. Rørdam, F. Larsen and N. J. Laustsen, An introduction to K-theory for C*-algebras, London Mathematical
Society Student Texts, vol. 49, Cambridge University Press, Cambridge, 2000.
[29] I. Raeburn and Dana P. Williams, Morita wquivalence and continuous-trace C*-algebras, Mathematical Surveys
and Monographs, 60. American Mathematical Society, Providence, RI, 1998.
[30] K. Thomsen, The homotopy type of the group of automorphisms of a UHF-algebra, J. Functional Analysis 72
(1987), 182 -- 207.
[31] A. S. Toms and W. Winter, Strongly self-absorbing C*-algebras, Trans. Amer. Math. Soc. 359 (2007), no. 8,
3999 -- 4029.
[32] A. Valette, A remark on the Kasparov groups Ext(A, B), Pacific J. Math. 109 : 1 (1983), 247 -- 255.
[33] W. Winter, Strongly self-absorbing C*-algebras are Z-stable, J. Noncommut. Geom. 5 (2011), no. 2, 253 -- 264.
20
|
1001.1268 | 2 | 1001 | 2010-12-08T17:33:33 | Minimal dynamics and Z-stable classification | [
"math.OA",
"math.DS"
] | Let X be an infinite compact metric space, \alpha : X \to X a minimal homeomorphism, u the unitary implementing \alpha in the transformation group C*-algebra, and S a class of separable nuclear C*-algebras that contains all unital hereditary C*-subalgebras of C*-algebras in S. Motivated by the success of tracial approximation by finite dimensional C*-algebras as an abstract characterization of classifiable C*-algebras and the idea that classification results for C*-algebras tensored with UHF algebras can be used to derive classification results up to tensoring with the Jiang-Su algebra Z, we prove that the transformation group C*-algebra tensored with a UHF algebra is tracially approximately S if there exists a y in X such that a certain C*-subalgebra is tracially approximately S. If the class S consists of finite dimensional C*-algebras, this can be used to deduce classification up to tensoring with Z for C*-algebras associated to minimal dynamical systems where projections separate tracial states. This is done without making any assumptions on the real rank or stable rank of either the transformation group C*-algebra or the C*-subalgebra, nor on the dimension of X. The result is a key step in the classification of C*-algebras associated to uniquely ergodic minimal dynamical systems by their ordered K-groups. It also sets the stage to provide further classification results for those C*-algebras of minimal dynamical systems where projections do not necessarily separate traces. | math.OA | math |
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
KAREN R. STRUNG AND WILHELM WINTER
α
α
Abstract. Let X be an infinite compact metric space, α : X → X a minimal
homeomorphism, u the unitary implementing α in the transformation group
C ∗-algebra C(X) ⋊
Z, and S a class of separable nuclear C ∗-algebras that
contains all unital hereditary C ∗-subalgebras of C ∗-algebras in S. Motivated
by the success of tracial approximation by finite dimensional C ∗-algebras as an
abstract characterization of classifiable C ∗-algebras and the idea that classifi-
cation results for C ∗-algebras tensored with UHF algebras can be used to derive
classification results up to tensoring with the Jiang -- Su algebra Z, we prove
that (C(X) ⋊
Z) ⊗ Mq∞ is tracially approximately S if there exists a y ∈ X
such that the C ∗-subalgebra (C ∗(C(X), uC0(X \ {y}))) ⊗ Mq∞ is tracially
approximately S. If the class S consists of finite dimensional C ∗-algebras, this
can be used to deduce classification up to tensoring with Z for C ∗-algebras
associated to minimal dynamical systems where projections separate tracial
states. This is done without making any assumptions on the real rank or sta-
ble rank of either C(X) ⋊
Z or C ∗(C(X), uC0(X \{y})), nor on the dimension
of X. The result is a key step in the classification of C ∗-algebras associated
to uniquely ergodic minimal dynamical systems by their ordered K-groups. It
also sets the stage to provide further classification results for those C ∗-algebras
of minimal dynamical systems where projections do not necessarily separate
traces.
α
1. Introduction
The two subjects of C∗-algebras and dynamical systems have long been close
allies. On the one hand, dynamical systems provide a rich source of elegant and
fundamental examples for C∗-algebra theory, see [5], [7] and [4], to name but a
few. On the other hand, the techniques of C∗-algebras have been used to make
progress in distinguishing dynamical systems, most notably in the work of Gior-
dano, Putnam, and Skau on minimal Cantor systems [10]. The class of C∗-algebras
associated to actions on infinite compact metric spaces has also played an inter-
esting role in Elliott's classification program for C∗-algebras. The classification
program was initiated when Elliott showed that the class of approximately finite
dimensional (AF) C∗-algebras is classified by K-theory. Following this success, El-
liott conjectured that simple separable nuclear C∗-algebras might be classified by a
certain K-theoretic invariant, now called the Elliott invariant, in the sense that if
A and B are two simple separable nuclear C∗-algebras with isomorphic invariants,
then A and B are ∗-isomorphic as C∗-algebras. Moreover, the isomorphism may
be chosen in such a way as to induce the isomorphism of Elliott invariants.
Date: November 20, 2018.
2000 Mathematics Subject Classification. 46L85, 46L35.
Key words and phrases. minimal homeomorphisms, classification, Z-stability.
Supported by: EPSRC First Grant EP/G014019/1.
1
2
KAREN R. STRUNG AND WILHELM WINTER
Evidence in support of the conjecture was successfully gathered by way of clas-
sification results for various classes of C∗-algebras, including theorems for those
C∗-algebras associated to minimal dynamical systems. Many of these results re-
quire the presentation of the C∗-algebra as a direct limit, showing it is of a form
known to be classifiable. For example, the irrational rotation algebras, which are
the C∗-algebras associated to the dynamical system (T, α) where α is an irrational
rotation of the circle T, were shown by Elliott and Evans to be AT algebras with
real rank zero [7], and hence classifiable by their Elliott invariants. Similarly, the
crossed products associated to minimal homeomorphisms of the Cantor set were
shown, using results by Putnam, to be AT algebras with real rank zero and thus
classifiable [6].
In the past decade, efforts have been made to avoid the use of specific direct
limit presentations as a means of obtaining classification theorems and more ab-
stract methods have been considered. To this end, Lin introduced the first notion of
tracial approximation of C∗-algebras with his concept of tracially approximately fi-
nite dimensional (TAF) C∗-algebras [15]. In the case of a simple unital C∗-algebra,
these may be thought of as being approximated by finite dimensional C∗-algebras
in trace. Similarly, one may consider C∗-algebras that are tracially approximately
interval (TAI) algebras. These are C∗-algebras that can be approximated by in-
i=1 Mmi(C(Xi)) where Xi is a
terval algebras, that is, C∗-algebras of the form Ln
single point or Xi = [0, 1], see [14].
Tracial approximation was also considered by Elliott and Niu in [8], where they
studied approximation by splitting interval algebras (TASI). In the same paper
they consider the concept of tracial approximation in a more general sense, that
is, classes of TAS C∗-algebras where S is an arbitrary class of C∗-algebras (see
Definition 2.1 below). In particular, they look at which properties of the class S
pass to the class TAS.
The concept of tracial approximation has proven to be extremely useful in pro-
viding classification results. One of the most notable results has been by Lin and
Phillips in [20], where classification results for C∗-algebras with tracial rank zero
were successfully applied to many C∗-algebras of minimal dynamical systems. Let
X be an infinite compact metric space, α : X → X a minimal homeomorphism and
u the unitary implementing α in C(X) ⋊α Z. Lin and Phillips proved that under
certain conditions, it is enough to verify that there exists a point y ∈ X such that
the C∗-subalgebra C∗(C(X), uC0(X \ {y})) has tracial rank zero.
In this paper, we generalize the results in [20] by following the strategy for
classification up to Z-stability as outlined in [35]. Here, Z denotes the Jiang -- Su
algebra, introduced in [12]. There are many characterizations of this algebra, both
abstract and concrete, which single it out as a universal object playing a role as
fundamental as that of the Cuntz algebra O∞, cf. [30] and [37]. Z-stability (i.e.,
the property of absorbing Z tensorially) is an important structural property for
C∗-algebras; it has recently been shown to be closely related to other topological
and algebraic regularity properties, such as finite topological dimension and strict
It is remarkable that all nuclear C∗-algebras
comparison of positive elements.
classified so far by their Elliott invariants are in fact Z-stable.
In [35], the second named author derived classification up to Z-stability from
classification results up to UHF stability. The latter are usually much easier to es-
tablish, since UHF stability ensures a wealth of projections. For any q ∈ N \ {1}, let
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
3
Mq∞ denote the UHF algebra N∞
n=1 Mq. Our main theorem says that if S is a class
of separable unital C∗-algebras such that the property of being a member of S passes
to unital hereditary C∗-subalgebras, then (C(X) ⋊α Z) ⊗ Mq∞ is TAS whenever
there exists a y ∈ X such that the C∗-subalgebra C∗(C(X), uC0(X \ {y})) ⊗ Mq∞
is TAS. We tensor with the UHF algebra to alleviate some of the restrictions on
C(X) ⋊α Z and C∗(C(X), uC0(X \ {y})) that are found in [20] and to take advan-
tage of results showing that classification results for C∗-algebras tensored with UHF
algebras can be used to derive classification results up to tensoring with the Jiang --
Su algebra, as shown in [35], [16] and [19]. When S is the set of finite dimensional
C∗-algebras, our result shows that the C∗-algebras associated to minimal dynam-
ical systems of infinite compact metric spaces whose projections separate tracial
states are classified by their K-theory, up to tensoring with Z. Toms and the sec-
ond named author show in [33] (see also [34]) that in the case where the base space
has finite topological dimension the crossed product is Z-stable. Together with the
results of the present paper this completes the classification of C∗-algebras asso-
ciated to uniquely ergodic minimal finite dimensional dynamical systems by their
ordered K-groups. We are optimistic that our result will also help provide further
classification results for those C∗-algebras of minimal dynamical systems of infinite
compact metric spaces where projections do not necessarily separate traces.
As an application, we obtain that the C∗-algebras associated to uniquely ergodic
minimal homeomorphisms of odd spheres as considered by Connes in [4] are all iso-
morphic. In the smooth case, this was already shown in [38], using the inductive
limit decomposition of [22]. The latter is technically very advanced; our method
provides a somewhat easier path since the inductive limit structure of the subal-
gebras C∗(C(X), uC0(X \ {y})) can be established with much less effort. In the
not necessarily uniquely ergodic case, one expects the space of tracial states to be
the classifying invariant; building on our present results, this will be pursued in a
subsequent article.
The paper is organized as follows.
In Section 2 we give the definition for a
C∗-algebra that is TAS and outline the strategy for arriving at our main theorem
in Section 3. In Section 4 we state and prove our main technical results and in
Section 5 we apply these to derive classification results. We also discuss some
examples and outline the strategy for further results relating to the classification
of transformation group C∗-algebras.
2. Preliminaries
We begin with a definition of what is meant by tracial approximation, cf. [13]
and [8].
Definition 2.1. Let S denote a class of separable unital C∗-algebras. Let A be
a simple unital C∗-algebra. Then A is tracially approximately S (or TAS) if the
following holds. For every finite subset F ⊂ A, every ǫ > 0, and every nonzero
positive element c ∈ A, there exists a projection p ∈ A and a unital C∗-subalgebra
B ⊂ pAp with 1B = p and B ∈ S such that:
(i) kpa − apk < ǫ for all a ∈ F ,
(ii) dist(pap, B) < ǫ for all a ∈ F ,
(iii) 1A − p is Murray -- von Neumann equivalent to a projection in cAc.
4
KAREN R. STRUNG AND WILHELM WINTER
For a compact metric space X and a minimal homeomorphism α : X → X, put
A = C(X) ⋊α Z. We denote
A{y} = C∗(C(X), uC0(X \ {y})).
A{y} is a unital C∗-subalgebra of A, a generalization of those introduced by Putnam
in [25]. This algebra carries much of the information contained in A while at the
same time is significantly more tractable. In particular, by Theorem 4.1(3) of [24],
its K0-group is isomorphic to that of A, and it can be written as an inductive limit
of subhomogeneous algebras in a straightforward manner, see Section 3 of [21].
There are natural bijections between the set of α-invariant probability measures on
X, the set of tracial states on A and the set of tracial states on A{y} ([21], Theorem
1.2).
We recall the notion of strict comparison of positive elements for a C∗-algebra
A. For two positive elements a, b ∈ A, write a . b if there exists rj ∈ A such that
j = a. For any C∗-algebra A, denote by T (A) the tracial state space of A.
limj rjbr∗
For τ ∈ T (A), define
dτ (a) = lim
n→∞
τ (a1/n)
(a ∈ A+).
If A is exact, then the set {dτ τ ∈ T (A)} coincides with the set of normalized
lower semicontinuous dimension functions of A ([28] and [11]).
If dτ (a) < dτ (b)
for all τ ∈ T (A) implies a . b, then we say that A has strict comparison (of
positive elements). It was recently shown by Toms ([32], Corollary 5.5) that if X
is a compact smooth connected manifold and α : X → X a diffeomorphism, then
C(X) ⋊α Z has strict comparison. His result relies on the direct limit structure
of C(X) ⋊α Z given in [22]; such a structure is not known in the general case.
However, because we have tensored with a UHF algebra, we are able to use results
of Rørdam to circumvent this issue. For a compact metric space X and a minimal
homeomorphism α : X → X, let A = C(X) ⋊α Z. Then, for any q ∈ N \ {1} and
any y ∈ X, the C∗-algebras A ⊗ Mq∞ and A{y} ⊗ Mq∞ have strict comparison by
Theorem 5.2 of [28].
3. Strategy
Our aim is to show that when A = C(X) ⋊α Z is the C∗-algebra arising from
a minimal dynamical system of an infinite compact metric space such that the
subalgebra A{y} ⊗ Mq∞ ⊂ A ⊗ Mq∞ is TAS, then A ⊗ Mq∞ is TAS. One can
generally only expect such a passage from A{y} to A after tensoring with a UHF
algebra. We also need to assume the class S of Definition 2.1 to be closed with
respect to unital hereditary C∗-subalgebras.
Our result (and method) is a generalization of the tracial rank zero case (with-
out tensoring with a UHF algebra), which was obtained by Lin and Phillips in [20].
A key lemma for their proof is showing that the finite dimensional C∗-subalgebra
in the definition of tracial rank zero can be replaced by a sufficiently large simple
unital C∗-subalgebra of tracial rank zero. Since we are concerned with A ⊗ Mq∞ ,
we show that for any simple unital C∗-algebra A, after tensoring with the UHF
algebra, A ⊗ Mq∞ is TAS when the subalgebra B ∈ S in Definition 2.1 is re-
placed by a simple unital subalgebra of A ⊗ Mq∞ that is TAS. With the aim of
using this result (Lemma 4.5 below), the key step is finding a suitable projection
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
5
p in the C∗-subalgebra A{y} ⊗ Mq∞ . Under the assumption that S is closed un-
der taking unital hereditary C∗-subalgebras, the unital hereditary C∗-subalgebra
B := p(A{y} ⊗ Mq∞ )p is also TAS; this follows from Lemma 2.3 of [8]. The idea
then is to find a projection such that B will be nearly invariant under conjugation
by u ⊗ 1Mq∞ and large enough for us to conclude that all of A ⊗ Mq∞ is TAS. This
is achieved in Lemma 4.4.
We impose no additional requirements on A{y} ⊗ Mq∞ in Lemma 4.5. Notice
in particular, that by virtue of having the simple stably finite C(X) ⋊α Z tensored
with a UHF algebra, we do not require that the C∗-subalgebra A{y} has stable rank
one, as is necessary for Lemma 4.2 of [20]; this follows from Corollary 6.6 of [27]. In
addition, Lemma 4.2 of [20] requires the assumption that A{y} has real rank zero.
This is used to pick out an initial small projection using Lemma 4.1 of [20]. This
projection is used to find a sequence of orthogonal projections, where conjugation
by u ⊗ 1Mq∞ acts as a shift, which are then perturbed using Berg's technique [1].
When S is the set of finite dimensional C∗-algebras, then TAS is simply the class
of C∗-algebras with tracial rank zero, in which case real rank zero is automatic
by Theorem 3.4 of [15]. However if we are interested in the class of TAI algebras,
for example, we can no longer assume this to be the case, as was shown in [17].
Thus in an effort to allow the class S to be as general as possible, we remove the
assumption that A ⊗ Mq∞ has real rank zero. Our Lemma 4.1 allows us to find an
initial small projection while avoiding the real rank zero restriction. We then find
a projection p ∈ A{y} ⊗ Mq∞ satisfying the same conditions, (i) -- (iii) of Lemma
4.5. Here we do not get an initial sequence of projections all lying in A{y} ⊗ Mq∞ .
However the projections lie close enough to this C∗-subalgebra that we are able to
push them inside. After applying Berg's technique, we once again end up outside
A{y} ⊗ Mq∞ , but close enough to push the resulting loop inside, eventually ending
up with the desired result.
4. Main Results
Lemma 4.1. Let X be an infinite compact metric space and α : X → X a minimal
homeomorphism. Let y ∈ X, and set A{y} = C∗(C(X), uC0(X \ {y})), where u is
the unitary in C(X) ⋊α Z implementing α. Let q ∈ N \ {1}.
Then for any η > 0 and any open set V ⊂ X containing y, there exists an open
set W ⊂ V with y ∈ W , functions g0 ∈ C0(W ), g1 ∈ C0(V ), 0 ≤ g0, g1 ≤ 1 and a
projection q0 ∈ C0(V )A{y}C0(V ) ⊗ Mq∞ such that
g0(y) = 1,
g1W = 1,
and kq0(g1 ⊗ 1) − g1 ⊗ 1k ≤ η.
Proof. We claim that there is a non-zero projection in C0(V )A{y}C0(V ) ⊗ Mq∞ .
The set V is non-empty since y ∈ V . Thus C0(V ) is non-zero and hence we
can find a non-zero positive contraction in (C0(V )A{y}C0(V )) ⊗ Mq∞ , call it e.
Since A{y} is simple by Proposition 2.5 of [20], so is A{y} ⊗ Mq∞ . Thus every
tracial state τ ∈ T (A{y} ⊗ Mq∞) is faithful, and in particular we have τ (e) > 0 for
every tracial state τ . Since A{y} ⊗ Mq∞ is unital, T (A{y} ⊗ Mq∞) is compact. Thus
minτ ∈T (A{y}⊗Mq∞ ) τ (e) > 0. Furthemore, dτ (e) > τ (e) so the previous observations
imply that minτ ∈T (A{y}⊗Mq∞ ) dτ (e) > 0.
6
KAREN R. STRUNG AND WILHELM WINTER
Since A{y} ⊗ Mq∞ has projections that are arbitrarily small in trace, there is a
projection p ∈ A{y} ⊗ Mq∞ satisfying
max
τ ∈T (A{y}⊗Mq∞ )
τ (p) <
min
τ ∈T (A{y}⊗Mq∞ )
dτ (e).
By the above, for the projection p and any τ ∈ T (A{y} ⊗ Mq∞ ) we have dτ (p) =
n → p.
τ (p) < dτ (e), so by strict comparison there are xn ∈ A{y} ⊗ Mq∞ with xnex∗
Let
an = e1/2x∗
nxne1/2 ∈ (C0(V )A{y}C0(V )) ⊗ Mq∞ .
Then an is self-adjoint and
kan − a2
nk → 0.
Disregarding any an such that kan − a2
nk ≥ 1/4, we obtain a sequence of projections
bn satisfying kbn−ank ≤ 2kan −a2
nk → 0 (Lemma 2.5.5 of [13]). Thus we obtain, for
large enough n, a projection b = bn contained in C0(V )A{y}C0(V ) ⊗ Mq∞ , proving
the claim. Moreover, b is Murray -- von Neumann equivalent to p, so minτ τ (b) =
minτ τ (p).
Let W be an open set contained in V such that y ∈ W and small enough so that
for every function f ∈ C0(W ) with 0 ≤ f ≤ 1 we have dτ (f ⊗ 1Mq∞ ) ≤ 1
2 minτ τ (b)
for every τ ∈ T (A{y} ⊗ Mq∞ ). Choose g0, g1 ∈ C0(W ) such that 0 ≤ g0, g1 ≤ 1,
g0(y) = 1 and g1g0 = g0. Then dτ (g1 ⊗1) < dτ (b) for every τ ∈ T (A{y}⊗Mq∞ ), and
so by the comparison of positive elements we have (g1 ⊗ 1) . b in A{y} ⊗ Mq∞ and
hence also in C0(V )A{y}C0(V ) ⊗ Mq∞ . Since A{y} ⊗ Mq∞ has stable rank one and
C0(V )A{y}C0(V )⊗Mq∞ is a full hereditary C∗-subalgebra of A{y}⊗Mq∞ , it also has
stable rank one by Theorem 3.6 of [26] with Theorem 2.8 of [3]. By Proposition 2.4 of
[28], for η/2 > 0 there is a unitary v in (C0(V )A{y}C0(V ) ⊗ Mq∞ )+ (the unitization
of C0(V )A{y}C0(V )⊗Mq∞ ) such that (g1 ⊗1−η/2)+ ≤ vbv∗ in (C0(V )A{y}C0(V )⊗
Mq∞)+, and hence in C0(V )A{y}C0(V ) ⊗ Mq∞ . Put q0 = vbv∗. Then
kq0(g1 ⊗ 1) − (g1 ⊗ 1)k < kq0(g1 ⊗ 1 − η/2)+ − (g1 ⊗ 1)k + η/2
= k(g1 ⊗ 1 − η/2)+ − (g1 ⊗ 1)k + η/2
< η.
(cid:3)
We will use the previous lemma to choose an initial projection in A{y} ⊗ Mq∞ .
However, since this projection actually only approximates the properties we would
like it to have, we require the following easy lemma that pushes orthogonal projec-
tions into a C∗-subalgebra. The proof is straightforward and hence omitted.
Lemma 4.2. Given ǫ > 0 and a positive integer n, there is a δ > 0 with the
following property. Let A be a C∗-algebra, B a C∗-subalgebra of A. Suppose that
p1, . . . , pn are mutually orthogonal projections in A, the first k, 0 ≤ k ≤ n, of which
are contained in B, and that ak+1, . . . , an are self adjoint elements of B such that
kpi − aik < min(1/2, δ),
i = k + 1, . . . , n.
Then there are mutuallly orthogonal projections q1, . . . , qn in B, where qi = pi for
1 ≤ i ≤ k, and for k + 1 ≤ i ≤ n we have
Moreover, if A is unital then there are unitaries ui ∈ A such that qi = uipiu∗
i .
kqi − pik < ǫ.
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
7
Lemma 4.3. Let X be an infinite compact metric space with a minimal homeo-
morphism α : X → X. Let A = C∗(X) ⋊α Z and A{y} = C∗(C(X), uC0(X \ {y})).
Then
K0(A{y} ⊗ Mq∞ ) ∼= K0(A ⊗ Mq∞ )
as ordered groups, with the isomorphism induced by the inclusion ι : A{y} → A.
Proof. Since the K1-group of the UHF algebra Mq∞ is {0}, we have
K∗(B) ⊗ K∗(Mq∞ ) = (K0(B) ⊗ K0(Mq∞ )) ⊕ (K1(B) ⊗ K0(Mq∞ ))
where B denotes A or A{y}.
Let ι : A{y} → A be the inclusion map. We have the associated inclusion map
ι ⊗ idMq∞ : A{y} ⊗ Mq∞ → A ⊗ Mq∞ whence we get the group homomorphisms
K0(ι) : K0(A{y}) → K0(A) and K0(ι ⊗ idMq∞ ) : K0(A{y} ⊗ Mq∞) → K0(A ⊗ Mq∞).
Since K∗(Mq∞ ) is torsion free, by the Kunneth Theorem for Tensor Products, we
have homomorphisms
α1 : K∗(A{y}) ⊗ K∗(Mq∞ ) → K∗(A{y} ⊗ Mq∞)
and
α2 : K∗(A) ⊗ K∗(Mq∞ ) → K∗(A ⊗ Mq∞ ),
which are of degree 0, thus giving maps of the K0 groups, which we will also call
α1 and α2. We get the following diagram
0
0
/ K0(A{y}) ⊗ K0(Mq∞ )
K0(ι)⊗idK0 (Mq∞ )
/ K0(A) ⊗ K0(Mq∞ )
α1
α2
/ K0(A{y} ⊗ Mq∞ )
K0(ι⊗idMq∞ )
/ K0(A ⊗ Mq∞ )
0
/ 0
which commutes by naturality. It follows from Theorem 4.1(3) of [24] that the map
K0(ι) ⊗ idK0(Mq∞ ) is an isomorphism. We conclude that K0(ι ⊗ idMq∞ ) is also an
isomorphism.
Clearly K0(ι ⊗ idMq∞ )([1 ⊗ 1Mq∞ ]) = [1 ⊗ 1Mq∞ ], and thus preserves order units.
It remains to show K0(ι ⊗ idMq∞ ) preserves the order structure and hence is an
order isomorphism.
Let η ∈ K0(A⊗ Mq∞ )+. Then there are projections p and q in M∞(A{y} ⊗ Mq∞)
such that K0(ι ⊗ idMq∞ )([p] − [q]) = η. We show that [p] − [q] ∈ K0(A{y} ⊗ Mq∞)+.
Let τ ′ be the unique tracial state on Mq∞ . Then, by Theorem 1.2 (4) of [21], for
σ ∈ T (A{y} ⊗ Mq∞ ), we have σ = (τ ◦ ι) ⊗ τ ′ for some τ ∈ T (A). By simplicity of
A ⊗ Mq∞, we have K0(τ ⊗ τ ′)(η) > 0. It follows that (τ ⊗ τ ′)(ι ⊗ idMq∞ )(p − q) > 0,
hence (τ ◦ ι) ⊗ (τ ′ ◦ idMq∞ )(p) > (τ ◦ ι) ⊗ (τ ′ ◦ idMq∞ )(q) and finally, σ(p) > σ(q).
Since A{y} ⊗ Mq∞ has strict comparison, we must have q . p, so [p] − [q] > 0, as
desired.
(cid:3)
The following lemma generalizes Lemma 4.2 of [20], due to H. Lin and N. C. Phillips.
Lemma 4.4. Let X be an infinite compact metric space, α : X → X a minimal
homeomorphism, y ∈ X and q ∈ N \ {1}. Let A = C(X) ⋊α Z and A{y} =
C∗(C(X), uC0(X \ {y})), where u is the unitary implementing α in A. Then,
for any finite subset F ⊂ A ⊗ Mq∞ and every ǫ > 0, there is a projection p in
A{y} ⊗ Mq∞ such that
/
/
/
/
/
/
/
8
KAREN R. STRUNG AND WILHELM WINTER
(i) kpa − apk < ǫ for all a ∈ F ,
(ii) dist(pap, p(A{y} ⊗ Mq∞ )p) < ǫ for all a ∈ F ,
(iii) τ (1A⊗Mq∞ − p) < ǫ for all τ ∈ T (A ⊗ Mq∞ ).
Proof. Let ǫ > 0. We first show that there exists a projection satisfying properties
(i) -- (iii) of the lemma when F is assumed to be of the form
F = (G ⊗ {1Mq∞ }) ∪ {u ⊗ 1Mq
∞ }
where G is a finite subset of C(X).
Let N0 ∈ N such that π/(2N0) < ǫ/4.
Let δ0 > 0 with δ0 < ǫ/4 and sufficiently small so that for all g ∈ G we have
kg(x1) − g(x2)k < ǫ/8 as long as d(x1, x2) < 4δ0.
Choose δ > 0 with δ < δ0 and such that d(α−n(x1), α−n(x2)) < δ0 whenever
d(x1, x2) < δ and 0 ≤ n ≤ N0.
Since α is minimal, there is an N > N0 + 1 such that
Let R ∈ N be sufficiently large so that
d(αN (y), y) < δ.
R > (N + N0 + 1)/ min(1, ǫ).
Minimality of α also implies that there is an open neighbourhood U of y such
that
α−N0(U ), α−N0+1(U ), . . . , U, α(U ), . . . , αR(U )
are all disjoint. Making U smaller if necessary, we may assume that each αn(U ),
−N0 ≤ n ≤ R has diameter less than δ. To apply Berg's technique, we only need
Un for −N0 ≤ n ≤ N , however we require R to be larger in order to satisfy property
(iii) of the lemma.
Let λ = max{kgk g ∈ G}, and choose
0 < ǫ0 < min(1/2, ǫ/(2(N + 3N0 + 1)), ǫ/(32N (λ + ǫ/4)),
and
0 < ǫ1 < min(ǫ0/8, δǫ0,N /16)
0 < η < min(2ǫ, δǫ1,N0+N +1)
where δǫ1,N0+N +1 is given by Lemma 4.2 with respect to ǫ0 and N0 + N + 1 in place
of ǫ and n, respectively; similarly for δǫ0,N .
Let f0 : X → [0, 1] be continuous with supp(f0) ⊂ U , and f0V = 1 for some
open set V ⊂ U containing y.
By Lemma 4.1, there is an open set W ⊂ V containing y, functions g0 ∈ C0(W ),
g1 ∈ C0(V ), 0 ≤ g0, g1 ≤ 1 and a projection q0 ∈ C0(V )A{y}C0(V ) ⊗ Mq∞ such
that
g0(y) = 1, g1W = 1 and kq0(g1 ⊗ 1) − g1 ⊗ 1k < η/2.
Consequently, (f0 ⊗ 1)q0 = q0 = q0(f0 ⊗ 1) and kq0(g0 ⊗ 1) − g0 ⊗ 1k < η/2.
For −N0 ≤ n ≤ N , set
qn = (un ⊗ 1)q0(u−n ⊗ 1),
fn = unf0u−n = f0 ◦ α−n
and Un = αn(U ).
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
9
Then supp(fn) ⊂ Un and
(fn ⊗ 1)qn = ((unf0u−n)⊗ 1)(un ⊗ 1)q0(u−n ⊗ 1) = (un ⊗ 1)(f0 ⊗ 1)q0(u−n ⊗ 1) = qn.
Similarly, qn(fn ⊗ 1) = qn. Since the fn have disjoint support, it follows that the
projections
are mutually orthogonal.
q−N0, . . . , q−1, q0, q1, . . . , qN
We claim that q−N0, . . . , q−1, q0 ∈ A{y} ⊗ Mq∞ and that there are self-adjoint
c1, . . . , cN ∈ A{y} ⊗ Mq∞ such that kqn − cnk < η for 1 ≤ n ≤ N .
Let 1 ≤ n ≤ N0 and consider q−n. We have (uf−n ⊗ 1) ∈ A{y} ⊗ Mq∞ for
1 ≤ n ≤ N0 since U−n ∩ U0 = ∅. Let an = f n
0 un ⊗ 1. Then
an = f n
0 un ⊗ 1 = (uu−1f0u2u−2f0u3u−3 · · · unu−nf0un) ⊗ 1
= (uf−1 ⊗ 1)(uf−2 ⊗ 1) · · · (uf−n ⊗ 1) ∈ A{y} ⊗ Mq∞ .
From this it follows that
q−n = (u−n ⊗ 1)q0(un ⊗ 1)
= (u−n ⊗ 1)(f n
= a∗
0 ⊗ 1)q0(f n
nq0an ∈ A{y} ⊗ Mq∞ .
0 ⊗ 1)(un ⊗ 1)
Note that q0(g0 ⊗ 1) = q0(g1g0 ⊗ 1), since g1W = 1 and g0 ∈ C0(W ). Thus
k(g1g0 ⊗ 1) − q0(g0 ⊗ 1)k = k(g1 ⊗ 1 − q0(g1 ⊗ 1))( g0 ⊗ 1)k
< η/2.
Also, g1f0 = g1 since f0V = 1 and g1 ∈ C0(V ). Hence
k(q0 − g1 ⊗ 1)(f0 ⊗ 1 − g0 ⊗ 1) − (q0 − g1 ⊗ 1)k
= kq0(f0 ⊗ 1) − q0(g0 ⊗ 1) − (g1f0) ⊗ 1 + (g1g0) ⊗ 1 − q0 + g1 ⊗ 1k
= k(g1g0 ⊗ 1) − q0(g0 ⊗ 1)k
< η/2.
Since f0(y) = 1 = g0(y), we have that u(f0 − g0) ⊗ 1 ∈ A{y} ⊗ Mq∞ . Set
c1 = (u(f0 − g0) ⊗ 1)(q0 − g1 ⊗ 1)(u(f0 − g0) ⊗ 1)∗ + (ug1u∗ ⊗ 1).
Then c1 is a self-adjoint element in A{y} ⊗ Mq∞ and
kq1 − c1k
= k(u ⊗ 1)q0(u∗ ⊗ 1) − (u(f0 − g0) ⊗ 1)(q0 − g1 ⊗ 1)(u(f0 − g0) ⊗ 1)∗
−(u ⊗ 1)(g1 ⊗ 1)(u∗ ⊗ 1)k
= kq0 − ((f0 − g0) ⊗ 1)(q0 − g1 ⊗ 1)((f0 − g0) ⊗ 1) − g1 ⊗ 1k
≤ k(q0 − g1 ⊗ 1) − (q0 − g1 ⊗ 1)((f0 − g0) ⊗ 1)k
+ k(q0 − g1 ⊗ 1)((f0 − g0) ⊗ 1) − ((f0 − g0) ⊗ 1)(q0 − g1 ⊗ 1)((f0 − g0) ⊗ 1)k
≤ k(q0 − g1 ⊗ 1) − (q0 − g1 ⊗ 1)((f0 − g0) ⊗ 1)k
+ k(q0 − g1 ⊗ 1) − ((f0 − g0) ⊗ 1)(q0 − g1 ⊗ 1)k k((f0 − g0) ⊗ 1)k
< η.
10
KAREN R. STRUNG AND WILHELM WINTER
For 2 ≤ n ≤ N , define
cn = ((ufn−1 · · · uf1) ⊗ 1)c1((ufn−1 · · · uf1) ⊗ 1)∗.
The cn are self-adjoint elements in A{y} ⊗ Mq∞ since fn−1, . . . , f1 all vanish at y.
Furthermore,
kqn − cnk = k(un ⊗ 1)q0(u−n ⊗ 1) − cnk
0
= (cid:13)(cid:13)((unf n−1
= (cid:13)(cid:13)((unf n−1
0
) ⊗ 1)q0((f n−1
u−1) ⊗ 1)q1((uf n−1
u−n) ⊗ 1) − cn(cid:13)(cid:13)
0
0
= k((ufn−1 · · · uf1) ⊗ 1)q1((ufn−1 · · · uf1) ⊗ 1)∗ − cnk
≤ kq1 − c1k
< η.
u−n) ⊗ 1) − cn(cid:13)(cid:13)
This proves the claim.
We now apply Lemma 4.2 to obtain projections p1, . . . , pN in A{y} ⊗ Mq∞ such
that
q−N0 , . . . , q−1, q0, p1, . . . , pN
are mutually orthogonal, and, for 1 ≤ n ≤ N , we have
and unitaries yn such that
kpn − qnk < ǫ1
pn = ynqny∗
n.
Since pn ∼ qn and qn ∼ q0, we have [pN ] = [q0] in K0(A ⊗ Mq∞ ). Since
K0(ι⊗idMq∞ ) : K0(A{y}⊗Mq∞) → K0(A⊗Mq∞) is an isomorphism by Lemma 4.3,
we also have [pN ] = [q0] in K0(A{y} ⊗ Mq∞ ). Moreover, simplicity of A{y} ([20],
Proposition 2.5) implies A{y} ⊗ Mq∞ has stable rank one by Corollary 6.6 of [27].
Thus projections in matrix algebras over A{y} ⊗Mq∞ satisfy cancellation, and there
is a partial isometry w ∈ A{y} ⊗ Mq∞ such that w∗w = q0 and ww∗ = pN .
For t ∈ R, set
v(t) = cos(πt/2)(q0 + pN ) + sin(πt/2)(w − w∗).
Then v(t) is a unitary in the corner (q0 + pN )(A{y} ⊗ Mq∞ )(q0 + pN ). The matrix
of v(t) with respect to the obvious decomposition is
(cid:18)cos(πt/2) − sin(πt/2)
cos(πt/2) (cid:19) .
sin(πt/2)
wk = (u−k ⊗ 1)v(k/N0)(uk ⊗ 1).
w′
k = (ak + bk)∗v(k/N0)(ak + bk)
For 0 ≤ k ≤ N0, define
Also, let
where
ak = (f k
bk = (f k
0 uk) ⊗ 1 = (uf−1 · · · uf−k) ⊗ 1 (as above)
N uk) ⊗ 1 = (ufN −1 . . . ufN −k) ⊗ 1.
Both ak and bk are in A{y} ⊗ Mq∞ , hence
w′
k ∈ A{y} ⊗ Mq∞ .
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
11
We show what wk is close to w′
k. Define
xk = (u−k ⊗ 1)(q0 + qN )v(k/N0)(q0 + qN )(uk ⊗ 1).
We have that
kwk − xkk = k(u−k ⊗ 1)(q0 + pN )v(k/N0)(q0 + pN )(uk ⊗ 1)
−(u−k ⊗ 1)(q0 + qN )v(k/N0)(q0 + qN )(uk ⊗ 1)k
= kq0v(k/N0)pN − q0v(k/N0)qN + pN v(k/N0)q0 − qN v(k/N0)q0
+pN v(k/N0)pN − qN v(k/N0)qN k
≤ 4kpN − qN k
< 4ǫ1.
Also,
kw′
k − (ak + bk)∗(q0 + qN )v(k/N0)(q0 + qN )(ak + bk)k
≤ kak + bkk2k(q0 + pN )v(k/N0)(q0 + pN ) − (q0 + qN )v(k/N0)(q0 + qN )k
≤ 4kpN − qN k
< 4ǫ1.
But
(ak + bk)∗(q0 + qN )v(k/N0)(q0 + qN )(ak + bk)
0 uk + f k
N uk) ⊗ 1)∗(q0 + qN )v(k/N0)(q0 + qN )(f k
= (f k
= (u−k ⊗ 1)(q0 + qN )v(k/N0)(q0 + qN )(uk ⊗ 1)
= xk.
0 uk + f k
N uk) ⊗ 1)
Thus
kwk − w′
kk < 8ǫ1.
Comparing wk to wk+1 conjugated by u ⊗ 1 we have
k(u ⊗ 1)wk+1(u−1 ⊗ 1) − wkk = kv((k + 1)/N0) − v(k/N0)k ≤ π/(2N0) < ǫ/4
for 0 ≤ k < N0. Define projections
e0 = q0, en = pn for 1 ≤ n < N − N0
en = wN −nqn−N w∗
N −n for N − N0 ≤ n ≤ N.
dn = qn for 0 ≤ n < N − N0
and
Also define
and
dn = xN −nqn−N x∗
N −n for N − N0 ≤ n ≤ N.
Note that this gives dN = x0q0x∗
0 = (q0 + qN )v(0)q0v(0)∗(q0 + qN ) = q0 = d0 and
and eN = v(0)q0v(0)∗ = q0 = e0. We also have that the x∗
kxl = 0 when k 6= l. This
follows from the fact that, if k 6= l, then q−k, qN −k, q−l and qN −l, 0 ≤ k 6= l ≤ N0,
are mutually orthogonal and
(q0 + qN )(uk ⊗ 1)(u−l ⊗ 1)(q0 + qN )
= (uk ⊗ 1)(q−k + qN −k)(q−l + qN −l)(u−l ⊗ 1)
= 0.
12
KAREN R. STRUNG AND WILHELM WINTER
Also, if 0 < m < N − N0 and N − N0 ≤ n ≤ N then
qm(u−(N −n) ⊗ 1)(q0 + qN ) = qm(q−(N −n) + qn)(u−(N −n) ⊗ 1) = 0,
and similarly
(q0 + qN )(uN −n ⊗ 1)qm = 0.
From this it follows that dmdn = 0 for 0 ≤ m 6= n ≤ N .
For 1 ≤ n ≤ N − N0 − 1 we have
ken − dnk = kpn − qnk < ǫ1
and
ke0 − d0k = 0.
If N − N0 ≤ n ≤ N , then
ken − dnk = kwN −nq−(N −n)w∗
= kwN −nq−(N −n)w∗
−xN −nq−(N −n)x∗
N −n − xN −nq−(N −n)x∗
N −n − wN −nq−(N −n)x∗
N −nk
N −nk
N −n + wN −nq−(N −n)x∗
N −n
N −nk + kwN −n − xN −nk
N −n − x∗
≤ kw∗
< 4ǫ1 + 4ǫ1
= 8ǫ1.
We now show that conjugating the dn by u ⊗ 1 acts approximately as a cyclic
shift. For 1 ≤ n ≤ N − N0 − 1 we have (u ⊗ 1)dn−1(u ⊗ 1)∗ = dn since dn = qn.
If n = N − N0, then
dN −N0 = xN0 q−N0x∗
N0
= (u−N0 ⊗ 1)(q0 + qN )v(1)(q0 + qN )q0(q0 + qN )v(−1)(q0 + qN )(uN0 ⊗ 1)
= (u−N0 ⊗ 1)(q0 + qN )pN (q0 + qN )(uN0 ⊗ 1)
= (u−N0 ⊗ 1)qN pN qN (uN0 ⊗ 1).
Thus
k(u ⊗ 1)dN −N0−1(u∗ ⊗ 1) − dN −N0k
= kqN −N0 − (u−N0 ⊗ 1)qN pN qN (uN0 ⊗ 1)k
= k(u−N0 ⊗ 1)qN (uN0 ⊗ 1) − (u−N0 ⊗ 1)qN pN qN (uN0 ⊗ 1)k
≤ kqN − pN k
< ǫ1.
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
13
When N − N0 < n ≤ N , first consider what happens to the en using the
estimation made on the wk above. We have
k(u ⊗ 1)en−1(u∗ ⊗ 1) − enk
= k(u ⊗ 1)wN −(n−1)(u∗ ⊗ 1)qn−N (u ⊗ 1)w∗
N −(n−1)(u∗ ⊗ 1)
−wN −nqn−N w∗
N −nk
≤ k(u ⊗ 1)wN −(n−1)(u∗ ⊗ 1)qn−N (u ⊗ 1)w∗
N −(n−1)(u∗ ⊗ 1) −
(u ⊗ 1)wN −(n−1)(u∗ ⊗ 1)qn−N w∗
+k(u ⊗ 1)wN −(n−1)(u∗ ⊗ 1)qn−N w∗
N −nk
N −(n−1)(u∗ ⊗ 1) − w∗
N −nk
≤ k(u ⊗ 1)w∗
N −n − wN −nqn−N w∗
N −nk
+k(u ⊗ 1)wN −(n−1)(u∗ ⊗ 1) − wN −nk
< ǫ/2.
From this we have
k(u ⊗ 1)dn−1(u∗ ⊗ 1) − dnk < ken−1 − dn−1k + ken − dnk + ǫ/2
< 16ǫ1 + ǫ/2.
Now we use the fact that the wk are almost in A{y} ⊗ Mq∞ to find projections
in A{y} ⊗ Mq∞ that lie close to the en and hence also close to the dn. When
0 ≤ n ≤ N − N0 − 1, we have en = pn ∈ A{y} ⊗ Mq∞. Also, since eN = q0, we only
need to find projections when N − N0 ≤ n ≤ N − 1. In this case, we have
ken − w′
N −nq−(N −n)(w′
= kwN −nq−(N −n)w∗
< 16ǫ1.
N −n)∗k
N −n − w′
N −nq−(N −n)(w′
N −n)∗k
Since w′
projections rn ∈ A{y} ⊗ Mq∞ with
N −nq−(N −n)(w′
N −n)∗ ∈ A{y} ⊗ Mq∞ , by Lemma 4.2 we find orthogonal
krn − enk < ǫ0 and rn = znenz∗
n
for unitaries zn ∈ A ⊗ Mq∞ . This also implies that
krn − dnk < ǫ0 + 8ǫ1 < 2ǫ0.
For 1 ≤ n ≤ N − N0 − 1 put rn = en = pn and put rN = eN = q0. Then set
r =
N
Xn=1
rn
and
p = 1 − r.
We verify that the projection p ∈ A{y} ⊗ Mq∞ satisfies properties (i) -- (iii) of the
lemma.
Let d = PN
n=1 dn. Note that
d − (u ⊗ 1)d(u ⊗ 1)∗ =
N
Xn=N −N0
((u ⊗ 1)dn−1(u ⊗ 1)∗ − dn).
14
KAREN R. STRUNG AND WILHELM WINTER
For N − N0 ≤ m 6= n ≤ N , we have
((u ⊗ 1)dn−1(u ⊗ 1)∗ − dn)((u ⊗ 1)dm−1(u ⊗ 1)∗ − dm)
= (u ⊗ 1)dn−1dm−1(u ⊗ 1)∗ − (u ⊗ 1)dn−1(u ⊗ 1)∗dm
−dn(u ⊗ 1)dm−1(u ⊗ 1)∗ + dndm
= −(u ⊗ 1)xN −(n−1)qn−1−N x∗
N −(n−1)(u ⊗ 1)∗xN −mqm−N x∗
−xN −nqn−N x∗
N −n(u ⊗ 1)xN −(m−1)qm−1−N x∗
N −m
N −(m−1)(u ⊗ 1)∗
= −(u ⊗ 1)xN −(n−1)qn−1−N (u ⊗ 1)∗x∗
N −nxN −mqm−N x∗
−xN −nqn−N x∗
N −nxN −m(u ⊗ 1)qm−1−N x∗
N −m
N −(m−1)(u ⊗ 1)∗
= 0.
Thus the terms in the sum are mutually orthogonal with norm at most 16ǫ1 + ǫ/2,
hence
kd − (u ⊗ 1)d(u ⊗ 1)∗k < 16ǫ1 + ǫ/2.
Now
kp − (u ⊗ 1)p(u ⊗ 1)∗k
= k((u ⊗ 1)r(u∗ ⊗ 1) − r) − ((u ⊗ 1)d(u∗ ⊗ 1) − d) + ((u ⊗ 1)d(u∗ ⊗ 1) − d)k
≤ 2kr − dk + 16ǫ1 + ǫ/2
<
N −N0−1
Xn=1
2kpn − qnk +
N −1
Xm=N −N0
2krm − dmk + 16ǫ1 + ǫ/2
< 2(N − N0 − 1)ǫ1 + 4N0ǫ0 + 16ǫ1 + ǫ/2
< (N − N0 − 1)ǫ0 + 4N0ǫ0 + 2ǫ0 + ǫ/2
< ǫ.
Since g1(y) = 1 it follows that u(1 − g1) ⊗ 1 ∈ A{y} ⊗ Mq∞ . Thus we also have
that p(u ⊗ 1)((1 − g1) ⊗ 1)(1 − q0)p ∈ A{y} ⊗ Mq∞ . Note that p ≤ 1 − q0. Using
this and the fact that kg1 ⊗ 1 − (g1 ⊗ 1)q0k < η/2 < ǫ, it follows that
kp(u ⊗ 1)p − p(u ⊗ 1)((1 − g1) ⊗ 1)(1 − q0)pk
= kp(u ⊗ 1)p − p(u ⊗ 1)p + p(u ⊗ 1)(g1 ⊗ 1)(1 − q0)pk
≤ kp(u ⊗ 1)(g1 ⊗ 1)p − p(u ⊗ 1)(g1 ⊗ 1)q0pk
< ǫ.
This proves (i) and (ii) for the element u ⊗ 1 ∈ F .
Now consider g ⊗ 1 ∈ F , where g ∈ C(X).
Since d(αN (y), y) < δ, we have d(αn(y), αn−N (y)) < δ0 for N − N0 ≤ n ≤ N . It
follows that Un−N ∪ Un has diameter less than 2δ + δ0 ≤ 3δ0. The function g ∈ G
varies by at most ǫ/8 on sets of diameter less than 4δ0, and since the sets
U1, U2, . . . , UN −N0−1, UN −N0 ∪ U−N0, UN −N0+1 ∪ U−N0+1, . . . , UN ∪ U0
are open and pairwise disjoint, there is g ∈ C(X) which is constant on each of
these sets and satisfies kg − gk < ǫ/4. Let the values of g on these sets be λ1 on U1
through to λN on UN ∪ U0.
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
15
For 0 ≤ n ≤ N − N0 − 1 we have
k(fn ⊗ 1)rn − rnk = k(fn ⊗ 1)rn − (fn ⊗ 1)qn + qn − rnk
≤ 2kqn − pnk
< 2ǫ1.
Thus
k(g ⊗ 1)rn − λn · rnk ≤ k(g ⊗ 1)rn − (g ⊗ 1)(fn ⊗ 1)rnk
+k(g ⊗ 1)(fn ⊗ 1)rn − λn · rnk
< 4kgkǫ1.
For N −N0 ≤ n ≤ N , we have that (fn−N +fn)xN −n = xN −n, since we may write
xN −n = (qn−N + qn)(un−N ⊗ 1)v((N − n)/N0)(uN −n ⊗ 1)(qn−N + qn). Similarly,
x∗
N −n(fn−N + fn) = x∗
N −n. Thus (fn−N + fn)dn = dn = dn(fn−N + fn). It follows
that
k(fn−N + fn)rn − rnk = k(fn−N + fn)rn − (fn−N + fn)dn + dn − rnk
< 4ǫ0.
Thus, similar to the above, k(g ⊗ 1)rn − λn · rnk < 8kgkǫ0.
Hence
k(g ⊗ 1)p − p(g ⊗ 1)k < k(g ⊗ 1)p − p(g ⊗ 1)k + ǫ/2
≤
N
Xn=1
k(g ⊗ 1)rn − λn · rn + λn · rn − rn(g ⊗ 1)k + ǫ/2
≤ 2N (8kgkǫ0) + ǫ/2
< ǫ.
This shows property (i) of the lemma for g ⊗ 1, g ∈ G. The second condition is
immediate since g ⊗ 1 is an element of A{y} ⊗ Mq∞ .
It remains to verify the third condition.
Since the sets α−N0(U ), α−N0+1(U ), . . . , U, α(U ), . . . , αR(U ) are all disjoint and
R > (N + N0 + 1)/ min(1, ǫ), it follows that
R
Xn=−N0
unf0u−n = f −N0
0
+ · · · + f0 + · · · + f R
0 ≤ 1
and hence τ1(f0) ≤ τ1(1)/(R + N0 + 1) < ǫ/(N + N0 + 1) for every τ1 ∈ T (A). Since
any τ ∈ T (A ⊗ Mq∞ ) is of the form τ = τ1 ⊗ τ2 for τ1 ∈ T (A) and τ2 the unique
tracial state on Mq∞, we have τ (q0) ≤ τ (f0 ⊗ 1) = τ1(f0) < ǫ/(N + N0 + 1). For
1 ≤ n ≤ N − N0 − 1 each rn is just q0 conjugated by a unitary so τ (rn) = τ (q0).
16
KAREN R. STRUNG AND WILHELM WINTER
For N − N0 ≤ n ≤ N , rn = znenz∗
N −nz∗
n. Thus
τ (rn) = τ (znwN −nq−(N −n)w∗
n = znwN −nq−(N −n)w∗
N −nz∗
n)
N −n)
= τ (wN −nq−(N −n)w∗
= τ (v((N − n)/N0)q0v((N − n)/N0)∗)
= τ ((q0 + pN )q0)
= τ (q0).
Thus
τ (1 − p) =
N
Xn=1
τ (rn) =
N
Xn=1
τ (q0) < N ǫ/(N + N0 + 1) < ǫ.
This proves the case where F is of the form (G ⊗ {1Mq∞ }) ∪ {u ⊗ 1Mq∞ }.
For the general case, let F ⊂ A⊗Mq∞ be a finite subset. Using the identification
A ⊗ Mq∞ ∼= A ⊗ Mqr ⊗ Mq∞ ∼= A ⊗ Mq∞ ⊗ Mqr ,
for r ∈ N, we may assume that the finite set is of the form
({1A} ⊗ {1Mq∞ } ⊗ B) ∪ ( G ⊗ {1Mq∞ } ⊗ {1Mqr }) ∪ ({u} ⊗ {1Mq∞ } ⊗ {1Mqr })
where r ∈ N, B is a finite subset of Mqr and G is a finite subset of C(X).
We may further assume that 1X = 1A ∈ G and also that 1Mqr ∈ B. Then
F = (G ⊗ {1Mq∞ }) ∪ {u ⊗ 1Mq∞ } and F = F ⊗ B.
Let ǫ > 0. By the above, there exists a projection p ∈ A{y} ⊗ Mq∞ satisfying
properties (i) -- (iii) of the lemma for the finite set F = G ⊗ {1Mq∞ } ∪ {u ⊗ 1Mq∞ },
with ǫ/ max({kbk b ∈ B}, 1) in place of ǫ.
Define p := p ⊗ 1Mqr ∈ A{y} ⊗ Mq∞ ⊗ Mqr . We now show that p satisfies
properties (i) -- (iii) of the lemma for F and ǫ.
Let a ∈ F . Then a = a ⊗ b for some a ∈ F and some b ∈ B. We have
kpa − apk = k(p ⊗ 1)(a ⊗ b) − (a ⊗ b)(p ⊗ 1)k
= k(pa) ⊗ b − (ap) ⊗ bk
= k(pa − ap) ⊗ bk
= kpa − apkkbk
< ǫ.
By the special case above, for every a ∈ F , there is some x ∈ p(A{y} ⊗ Mq∞ )p
such that kpap − xk < ǫ/(maxb∈B kbk). Thus x ⊗ 1 ∈ p(A{y} ⊗ Mq∞ ⊗ Mqr )p. It
is clear that p(1 ⊗ b)p ∈ p(A{y} ⊗ Mq∞ ⊗ Mqr )p for any b ∈ B, and so x ⊗ b ∈
p(A{y} ⊗ Mq∞ ⊗ Mqr )p. It follows that
kp(a ⊗ b)p − p(x ⊗ b)pk = kp(a ⊗ 1)(1 ⊗ b)p − p(x ⊗ 1)(1 ⊗ b)pk
= kp(a ⊗ 1)p(1 ⊗ b) − p(x ⊗ 1)p(1 ⊗ b)k
= k(pap − pxp) ⊗ 1kkbk
< ǫ.
This shows that (i) and (ii) hold.
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
17
To prove (iii), simply observe that τ ∈ T (A ⊗ Mq∞ ⊗ Mqr ) is of the form τ1 ⊗ τ2
where τ1 ∈ T (A ⊗ Mq∞ ) and τ2 ∈ T (Mqr ). Then
τ (1 − p) = τ (1 ⊗ 1 − p ⊗ 1) = τ ((1 − p) ⊗ 1) = τ1(1 − p)τ2(1) < ǫ.
(cid:3)
For the following lemma, we do not need any assumptions on the properties of
the class S of separable unital C∗-algebras. However, for Theorem 4.6, we cut
down a TAS C∗-subalgebra B ⊂ A ⊗ Mq∞ by a projection; there we must make the
additional requirement that the property of being a member of S passes to unital
hereditary subalgebras.
Lemma 4.5. Let S be a class of separable unital C∗-algebras. Let A be a simple
unital C∗-algebra and q ∈ N\{1}. Suppose that for every finite subset F ⊂ A⊗Mq∞ ,
every ǫ > 0, and every nonzero positive c ∈ A ⊗ Mq∞ , there exists a projection
p ∈ A ⊗ Mq∞ and a simple unital C∗-subalgebra B ⊂ p(A ⊗ Mq∞ )p which is TAS,
satisfies 1B = p and
(i) kpa − apk < ǫ for all a ∈ F ,
(ii) dist(pap, B) < ǫ for all a ∈ F ,
(iii) 1A − p is Murray -- von Neumann equivalent to a projection in c(A ⊗ Mq∞ )c.
Then A ⊗ Mq∞ is TAS.
Proof. Although a TAS C∗-algebra may not have property (SP), the C∗-algebra
A ⊗ Mq∞ always will, since A ⊗ Mq∞ has strict comparison (cf. [28]) and con-
tains nonzero projections which are arbitrarily small in trace. After noting this,
the proof is essentially the same as that of Lemma 4.4 of [20], replacing the C∗-
subalgebra of tracial rank zero with the TAS C∗-subalgebra B, and replacing the
finite dimensional C∗-subalgebra with a C∗-subalgebra from the class S.
(cid:3)
Theorem 4.6. Let S be a class of separable unital C∗-algebras such that the prop-
erty of being a member of S passes to unital hereditary C∗-subalgebras. Let X be
an infinite compact metric space, α : X → X a minimal homeomorphism, let u be
the unitary implementing α in A := C(X) ⋊α Z and q ∈ N \ {1}. Suppose there is
a y ∈ X such that A{y} ⊗ Mq∞ is TAS. Then A ⊗ Mq∞ is TAS.
Proof. We show that A ⊗ Mq∞ satisfies the conditions of Lemma 4.5.
Let ǫ > 0, F a finite subset of A ⊗ Mq∞ and a positive nonzero element c in
A⊗Mq∞ be given. Use Lemma 4.4 to find a projection p ∈ A{y} ⊗Mq∞ with respect
to F , c, and ǫ0 = min(ǫ, minτ ∈T (A⊗Mq∞ ) τ (c)). Put B = p(A{y} ⊗ Mq∞ )p. It is a
unital simple C∗-subalgebra of p(A ⊗ Mq∞)p and is TAS by the assumptions made
on S and Lemma 2.3 of [8]. Conditions (i) and (ii) of Lemma 4.5 are satisfied by the
choice of p. Since τ (1A − p) < minσ∈T (A⊗Mq∞ ) σ(c) < τ (c) for every tracial state
τ ∈ T (A ⊗ Mq∞ ), it follows from Theorem 5.2(a) of [28] that 1A − p is Murray -- von
Neumann equivalent to a projection in c(A ⊗ Mq∞ )c. Thus A ⊗ Mq∞ is TAS by
Lemma 4.5.
(cid:3)
5. Classification. Outlook.
Recall that for a separable simple unital stably finite nuclear C∗-algebra A, the
Elliott invariant of A is given by
((K0(A), K0(A)+, [1A]), K1(A), T (A), rA : T (A) → S(K0(A)))
18
KAREN R. STRUNG AND WILHELM WINTER
consisting of the ordered K-groups, the Choquet simplex of tracial states T (A) and
rA : T (A) → S(K0(A)),
the canonical affine map to the state space of (K0(A), K0(A)+, [1A]) given by
rA(τ )([p]) = τ (p) [29]. Since we are interested in applying our results to Elliott's
classification program, the most immediate application for Theorem 4.6 is when
S is the set of finite dimensional C∗-algebras, where we are able to apply Lin's
classification theorem for C∗-algebras of tracial rank zero. A simple unital C∗-
algebra with tracial rank zero always has real rank zero ([15], Theorem 3.4). When
A has real rank zero, the map rA is bijective, and as such the invariant becomes
the ordered K-theory [29]. Applying Theorem 4.6 to this special case, we have the
following classification result up to tensoring with the Jiang -- Su algebra Z.
Corollary 5.1. Let A denote the class of C∗-algebras with the following properties.
(i) A ∈ A is of the form C(X) ⋊α Z for some infinite compact metric space X
and minimal homeomorphism α.
(ii) The projections in A separate T (A).
Let A, B ∈ A and suppose there is a graded order isomorphism φ : K∗(A ⊗ Z) →
K∗(B ⊗ Z). Then there is a ∗-isomorphism Φ : A ⊗ Z → B ⊗ Z inducing φ.
Proof. Let A = C(X) ⋊α Z and B = C(Y ) ⋊β Z be in A and suppose φ : K∗(A ⊗
Z) → K∗(B ⊗ Z) is a graded order isomorphism. Let q be any prime number, let
x ∈ X and y ∈ Y ; define the C∗-subalgebras A{x} := C∗(C(X), uC0(X \ {x})) and
B{y} := C∗(C(Y ), vC0(Y \ {y})), where u and v are the unitaries implementing
α and β in C(X) ⋊α Z and C(Y ) ⋊β Z, respectively. It follows from Section 3 of
[21] and Corollary 2.2 of [23] that both A{x} ⊗ Mq∞ and B{y} ⊗ Mq∞ have locally
finite decomposition rank. By Lemma 4.3 above and Theorem 1.2 (4) of [21], the
K0-groups and tracial state spaces of A{x} ⊗ Mq∞ and A ⊗ Mq∞ (respectively
B{y} ⊗ Mq∞ and B ⊗ Mq∞ ) are identical. But then our assumptions on the class
A imply that A{x} ⊗ Mq∞ and B{y} ⊗ Mq∞ have projections separating tracial
states. Since A{x} ⊗ Mq∞ and B{y} ⊗ Mq∞ are approximately divisible (cf. [29]),
we deduce they have real rank zero by the results of [2], whence tracial rank zero
by Theorem 2.1 of [36]. Applying Theorem 4.6 (with S being the class of finite
dimensional C∗-algebras), A ⊗ Mq∞ and B ⊗ Mq∞ have tracial rank zero. Since
this is true for any q, we may employ Theorem 5.4 of [19] to conclude that there
is a ∗-isomorphism Φ : A ⊗ Z → B ⊗ Z inducing φ. Note that [19] also requires A
and B to satisfy the UCT, which automatically holds in our situation.
(cid:3)
Most notably, this is the missing link between existing classification results and
the work of Toms and the second named author in [33]. There it is shown that
if X has finite covering dimension, then the resulting crossed product is Z-stable
([33], Theorem 4.4); in this case Corollary 5.1 becomes Theorem 0.1 of [33]; see
also Theorem A of [34].
In particular this solves the classification problem for
C∗-algebras associated to uniquely ergodic minimal finite dimensional dynamical
systems.
A compelling set of examples are the C∗-algebras arising from minimal homeo-
morphisms of odd spheres Sn for n ≥ 3 odd. These were first considered in Section
5 of [4] for n = 3 and α a minimal diffeomorphism. It follows from Corollary 3
of Section 5 of [4] that the crossed product of C(Sn) by Z induced by a minimal
diffeomorphism has no nontrivial projections. The K0-group of such a C∗-algebra
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
19
A is shown in Example 4.6 of [24] to be K0(A) = Z2 with order given by (m, n) ≥ 0
if and only if n > 0 or (m, n) = (0, 0). These examples are covered by Corollary 5.1
precisely in the uniquely ergodic case (since otherwise 5.1 (ii) is not satisfied), so
they are classified up to Z-stability. The main result of [33] (which in turn makes
heavy use of [38]) then shows that Z-stability is automatic, so in the uniquely er-
godic case Connes' odd spheres are all isomorphic. The respective statement in the
smooth case was already derived in [38], using the inductive limit decomposition of
[22].
To use Corollary 5.1 to classify crossed products with more general tracial state
spaces, at the current stage only Lin's classification of C∗-algebras which are TAI
after tensoring with UHF algebras is available [18]. The algebras covered by this
are all rationally Riesz, i.e., their ordered K0-groups become Riesz groups after
tensoring with the K-theory of a UHF algebra. However, for general transformation
group C∗-algebras it is not clear when they are rationally Riesz, and one cannot
hope for TAI classification to be a sufficient tool in this case. The strategy would
then be to identify a suitable class S of unital C∗-algebras such that A{y} ⊗ Mq∞
can be verified to be TAS and such that TAS algebras can be classified in a similar
manner as TAF or TAI algebras. At least for odd spheres it is not hard to see that
they are rationally Riesz, so it only remains to verify that they are indeed TAI up
to tensoring with UHF algebras; the result would then be that they are entirely
determined by their tracial state spaces, by virtue of [19] and our Corollary 5.1. At
least in the case of finitely many ergodic measures we are confident that our results
will lead to a complete solution; this will be pursued in a subsequent paper.
We wish to point out that Corollary 5.1 does not require any condition on the
dimension of the underlying space and that, without such a condition, classification
up to Z-stability is probably the best for which one can hope. The examples con-
structed by Giol and Kerr in [9] suggest that counterexamples to the general case of
the Elliott conjecture as exhibited by Toms in [31] can also occur as transformation
group C∗-algebras. One would still expect classification up to Z-stability in this
setting, which would then also imply that at least the crossed products stabilized
by Z have finite topological dimension. Conversely, if the underlying space is fi-
nite dimensional, then Z-stability is automatic by [33]. In this sense, our result is
analogous to [36], which in the real rank zero case also provided classification up
to Z-stability under otherwise mild structural conditions.
Acknowledgments
We would like to thank the referee for a number of helpful comments.
Supported by: EPSRC First Grant EP/G014019/1.
References
[1] I. D. Berg, On approximation of normal operators by weighted shifts, Michigan Math. J. 21
(1975), 377 -- 383.
[2] B. Blackadar, A. Kumjian, and M. Rørdam, Approximately central matrix units and the
structure of non-commutative tori, K-theory 6 (1992), 267 -- 284.
[3] L. G. Brown, Stable isomorphism of hereditary subalgebras of C ∗-algebras, Pacific J. Math.
71 (1977), 335 -- 348.
[4] A. Connes, An analogue of the Thom isomorphism for crossed products of a C ∗-algebra by
an action of R, Advances in Math. 39 (1981), 31 -- 55.
[5] J. Cuntz, Simple C ∗-algebras generated by isometries, Comm. Math. Phys. 57 (1977), 173 --
185.
20
KAREN R. STRUNG AND WILHELM WINTER
[6] G. A. Elliott, On the classification of C ∗-algebras of real rank zero, J. Reine Angew. Math.
443 (1993), 179 -- 219.
[7] G. A. Elliott and D. E. Evans, The structure of the irrational rotation C ∗-algebras, Ann. of
Math. 138 (1993), no. 2, 477 -- 501.
[8] G. A. Elliott and Z. Niu, On tracial approximation, J. Funct. Anal. 254 (2008), no. 2, 396 --
440.
[9] J. Giol and D. Kerr, Subshifts and perforation, J. Reine Angew. Math. 639 (2010), 107 -- 119.
[10] T. Giordiano, I. F. Putnam, and C. F. Skau, Topological orbit equivalence and C ∗-crossed
products, J. Reine Angew. Math. 469 (1995), 51 -- 111.
[11] U. Haagerup, Every quasi-trace on an exact C ∗-algebra is a trace, Preprint, 1991.
[12] X. Jiang and H. Su, On a simple unital projectionless C ∗-algebra, Amer. J. Math. 121 (1999),
no. 2, 359 -- 413.
[13] H. Lin, An introduction to the classification of amenable C ∗-algebras, World Scientific, Sin-
gapore, 2001.
[14] H. Lin, The tracial topological rank of C ∗-algebras, Proc. London Math. Soc. (3) 83 (2001),
no. 1, 199 -- 234.
[15] H. Lin, Tracially AF C ∗-algebras, Trans. Amer. Math. Soc. 353 (2001), 693 -- 722.
[16] H. Lin, Localizing the Elliott conjecture at strongly self-absorbing C ∗-algebras -- an appendix,
arXiv preprint math.OA/0709.1654, 2007.
[17] H. Lin, Simple nuclear C ∗-algebras of tracial topological rank one, J. Funct. Anal. 251 (2007),
no. 2, 601 -- 679.
[18] H. Lin, Asymptotically unitary equivalence and classification of simple amenable C ∗-algebras,
arXiv preprint math.OA/0806.0636, 2009.
[19] H. Lin and Z. Niu, Lifting KK-elements, asymptotic unitary equivalence and classification
of simple C ∗-algebras, Adv. Math. 219 (2008), no. 5, 1729 -- 1769.
[20] H. Lin and N. C. Phillips, Crossed products by minimal homeomorphisms, J. Reine Angew.
Math. 641 (2010), 95 -- 122.
[21] Q. Lin and N. C. Phillips, Ordered K-theory for C ∗-algebras of minimal homeomorphisms,
Operator algebras and operator theory (Shanghai, 1997), Contemp. Math., vol. 228, Amer.
Math. Soc., Providence, RI, 1998, pp. 289 -- 314.
[22] Q. Lin and N. C. Phillips, Direct limit decomposition for C ∗-algebras of minimal diffeomor-
phisms, Advanced Studies in Pure Mathematics 38 "Operator Algebras and Applications",
Math. Soc. Japan, Tokyo, 2004, pp. 107 -- 133.
[23] P. W. Ng and W. Winter, A note on subhomogeneous C ∗-algebras, C. R. Acad. Sci. Canada
28 (2006), 91 -- 96.
[24] N. C. Phillips, Cancellation and stable rank for direct limits of recursive subhomogeneous
algebras, Trans. Amer. Math. Soc. 359 (2007), no. 10, 4625 -- 4652.
[25] I. F. Putnam, The C ∗-algebras associated with minimal homeomorphisms of the Cantor set,
Pacific J. Math. 136 (1989), no. 2, 329 -- 353.
[26] M. Rieffel, Dimension and stable rank in the K-theory of C ∗-algebras, Proc. London Math.
Soc. 46 (1983), no. (3), 301 -- 333.
[27] M. Rørdam, On the structure of simple C ∗-algebras tensored with a UHF-algebra, J. Funct.
Anal. 100 (1991), 1 -- 17.
[28] M. Rørdam, On the structure of simple C ∗-algebras tensored with a UHF-algebra, II, J.
Funct. Anal. 107 (1992), 255 -- 269.
[29] M. Rørdam, Classification of nuclear, simple C ∗-algebras, Classification of nuclear C ∗-
algebras. Entropy in operator algebras, Encyclopaedia Math. Sci., vol. 126, Springer, Berlin,
2002, pp. 1 -- 145.
[30] M. Rørdam and W. Winter, The Jiang -- Su algebra revisited, J. Reine Angew. Math. 642
(2010), 129 -- 165.
[31] A. S. Toms, On the independence of K-theory and stable rank for simple C ∗-algebras, J.
Reine Angew. Math. 578 (2005), 185 -- 199.
[32] A. S. Toms, Comparison theory and smooth minimal C ∗-dynamics, Comm. Math. Phys. 289
(2009), no. 2, 401 -- 433.
[33] A. S. Toms and W. Winter, Minimal dynamics and K-theoretic rigidity: Elliott's conjecture,
arXiv preprint math.OA/0903.4133, 2009.
[34] A. S. Toms and W. Winter, Minimal dynamics and the classification of C ∗-algebras, Proc.
Natl. Acad. Sci. USA 106 (2009), no. 40, 16942 -- 16943.
MINIMAL DYNAMICS AND Z-STABLE CLASSIFICATION
21
[35] W. Winter, Localizing the Elliott conjecture at strongly self-absorbing C ∗-algebras, arXiv
preprint math.OA/0708.0283v3, with an appendix by H. Lin, 2007.
[36] W. Winter, Simple C ∗-algebras with locally finite decomposition rank, J. Funct. Anal. 243
(2007), 394 -- 425.
[37] W. Winter,
Strongly
self-absorbing C ∗-algebras
are Z-stable,
arXiv
preprint
math.OA/0905.0583; to appear in J. Noncomm. Geom., 2009.
[38] W. Winter, Decomposition rank and Z-stability, Invent. Math. 179 (2010), no. 2, 229 -- 301.
School of Mathematical Sciences, University of Nottingham, University Park, Not-
tingham NG7 2RD, United Kingdom
E-mail address: [email protected]
E-mail address: [email protected]
|
1303.1216 | 3 | 1303 | 2013-04-01T13:39:02 | Hodge theory for elliptic complexes over unital Banach $C^*$-algebras | [
"math.OA",
"math.AP",
"math.KT"
] | We introduce a notion of ellipticity of complexes of linear pseudodifferential operators acting on sections of $A$-Hilbert bundles over smooth manifolds, $A$ being a Banach $C^*$-algebra. We prove that the cohomology groups of an $A$-elliptic pseudodifferential complex in finitely generated projective $A$-Hilbert bundles over a compact manifold are norm complete finitely generated $A$-modules if the images of the associated Laplacians are closed. This establishes a Hodge theory for these structures. | math.OA | math |
Hodge theory for elliptic complexes over unital
Banach C ∗-algebras
Svatopluk Kr´ysl ∗
Charles University in Prague, Faculty of Mathematics and Physics
September 25, 2018
Abstract
For a unital Banach C∗-algebra A, we prove that the cohomology groups of
A-elliptic complexes of pseudodifferential operators in finitely generated projec-
tive A-Hilbert bundles over compact manifolds are norm complete topological
vector spaces and finitely generated A-modules provided the images of certain
extensions of the so called associated Laplacians are closed. This establishes a
Hodge type theory for these structures.
Key words: Hodge theory, elliptic complexes, A-Hilbert bundles
Math. Subj. Class.: 35J05, 46L87, 58A14, 58J10
1 Introduction
In this paper, we deal with fields having their values in possibly infinite rank
vector bundles and systems of pseudodifferential equations for these fields. To
be more precise, we focus our attention at complexes of pseudodifferential oper-
ators acting between smooth sections of finitely generated projective A-Hilbert
bundles over compact manifolds. We recall a definition of an A-elliptic complex
and prove a Hodge type theory for a certain subclass of them. By a Hodge type
theory, we mean not only constructing an A-linear isomorphism between the
cohomology groups and the spaces of harmonic elements of the complex, but
also giving a description of the the cohomology groups from a topological point
of view.
Let A be a unital Banach C∗-algebra. A finitely generated projective A-
Hilbert bundle is, roughly speaking, a fiber bundle the total space of which is
a Banach manifold, and the fibers of which are finitely generated projective
Hilbert A-modules. A Hilbert A-module is a module over A which is first,
equipped with a positive definite A-sesquilinear map with values in A, and
second, it is a complete topological vector space with respect to the norm derived
from the A-sesquilinear map and from the Banach norm in A. Our reference
∗E-mail address: [email protected], Tel./Fax: + 420 222 323 221/ + 420 222
323 394
1
for general Hilbert A-modules is Lance [6], and for the finitely generated and
projective ones the monograph of Solovyov, Troitsky [12].
One of the basic basic steps in proving the so called Mishchenko-Fomenko
index theorem (Fomenko, Mishchenko [3]) for A-elliptic pseudodifferential oper-
ators acting on sections of finitely generated projective A-Hilbert bundles over
a given compact manifold is a construction of certain pseudoinverses to exten-
sions of such operators to the Sobolev type completions of the space of smooth
sections. See Fomenko, Mishchenko [3] for this construction. In [3], not only
a paramatrix for A-elliptic pseudodifferential operators is constructed, but the
authors also prove that such operators are A-Fredholm. In particular, the ker-
nels of such operators are finitely generated projective Hilbert A-modules. Let
us notice, that we use the theorems and notions mentioned in this paragraph in
the form in which they appear in Solovyov, Troitsky [12].
i Di + Di−1D∗
For a chain complex (Di, Γi)i∈N0 of pre-Hilbert A-modules and adjointable
pre-Hilbert A-module homomorphisms, we can form a sequence of the so called
associated Laplacians △i, i ∈ N0, without supposing any topology on the mod-
ules. Namely, one sets △i = D∗
i−1 simply. Assume that each △i
possesses a parametrix, i.e., there exist pre-Hilbert A-module homomorphisms
gi : Γi → Γi and pi : Γi → Γi such that parametrix equations 1 = △igi + pi,
1 = gi△i + pi hold, and pi maps Γi into the kernel of △i. We call complexes
with such behaved Laplacians parametrix possessing. It is quite interesting, and
we prove that in this case, the pseudoinverses gi are necessarily chain homomor-
phisms, i.e., gi+1Di = Digi. This is at least implicitly known also in the finite
rank situation, but we derive this property in a purely algebraic way. Know-
ing this, it is not hard to show that the cohomology groups of a parametrix
possessing complex are A-linearly isomorphic to the kernels of the appropriate
Laplacians, i.e., to the spaces of harmonic elements. This establishes an abstract
or, say, an algebraic Hodge theory. At this level, a topological characterization
of the cohomologies is missing.
By an A-elliptic complex, we mean a complex of pseudodifferential oper-
ators acting on smooth sections of A-Hilbert bundles the associated symbol
sequence of which is exact out of the zero section of the cotangent bundle. We
prove that the symbols of the Laplacians associated to an A-elliptic complex
are isomorphisms out of the zero section. From now on till the end of this
paragraph, all mentioned bundles are supposed to be finitely generated pro-
jective A-Hilbert bundles over compact manifolds. We use a generalization of
the Sobolev embedding theorem for sections of A-Hilbert bundles, which we
prove in this text, to derive a regularity of A-elliptic pseudodifferential opera-
tors. If we moreover, assume that the images of certain extensions, specified
in the paper, of each of the associated Laplacians of an A-elliptic complex are
closed, we are able to construct parametrix equations for the Laplacians using
the regularity. Thus, we prove that such A-elliptic complexes are parametrix
possessing. Especially, their cohomology groups are isomorphic to the spaces of
harmonic elements as A-modules. This establishes the algebraic Hodge theory
for these structures. Consequently, the mentioned A-Fredholm property of A-
elliptic pseudodifferential operators implies that the cohomology groups of this
2
subclass of A-elliptic complexes are finitely generated A-modules. Using some
basic topological reasoning, we prove that the cohomologies of these complexes
are Banach topological vector spaces.
The regularity of an A-elliptic operator in finitely generated projective A-
Hilbert bundles over compact manifolds was already proved, e.g., in [12] (Theo-
rem 2.1.145). We don't give a substantially new proof of this assertion, but we
try to write an as much self-contained proof of this fact as possible. We also no-
tice that although we could have considered general Banach C∗-algebras at least
until Lemma 5 inclusively, we decided to suppose that all Banach C∗-algebras
are unital from the beginning.
Further, let us remark that there exist generalizations of the classical Hodge
theory in directions different from that one described here. See, e.g., Bartholdi
et al. [1] and Smale et al. [11] for a generalization to complete separable metric
spaces endowed with a probabilistic measure, and also for further references
given there. We do not attempt to give a full reference to the topic of complexes
of pseudodifferential operators in A-Hilbert bundles, and refer the reader to
Troitsky [13], Solovyov, Troitsky [12] and Schick [10]. We develop the presented
theory mainly in order to enable a description of solutions to equations for
operator or Hilbert module valued fields similar to the equations appearing in
Quantum field theory. Our motivation comes, however, from geometric and
deformation quantization via the so called symplectic spinor fields. See Kostant
[5], Fedosov [2] and Habermann [4] for this context.
In the second section, we prove a theorem on the homotopy properties of
the parametrix possessing complexes (Theorem 3), and in Theorem 4, we derive
the algebraic form of the Hodge theory for them. In the third part, we recall
the definition of a finitely generated projective A-Hilbert bundle, Sobolev type
completions of smooth sections, and the definition of the Fourier transform in A-
Hilbert bundles. The embedding theorem is stated as Lemma 5. The regularity
for A-elliptic operators is proved in Theorem 7, and the smooth pseudoinverses
for the A-elliptic and self-adjoint ones are constructed in Theorem 8. In the
fourth section, we recall the notion of an A-elliptic complex and prove that its
cohomology groups are finitely generated A-modules and Banach topological
vector spaces under the mentioned condition on the images of the extensions of
the associated Laplacians (Theorem 11).
Preamble: In the whole text when not said otherwise, manifolds, fibra-
tions (bundle projections, total and base spaces) and sections of fibrations are
assumed to be smooth. Further, if an index exceeds its allowed range, the object
labeled by this index is supposed to be zero.
2 Parametrix possessing complexes
To fix a terminology, we recall some notions from the theory of Hilbert A-
modules. Let A be a unital Banach C∗-algebra. For a pre-Hilbert A-module
(U, (, )U ), let U denote the associated norm on U defined by u = p(u, u)U A,
u ∈ U, where A denotes the Banach norm on A. If (U, U ) is a complete
3
normed space, we call (U, (, )U ) a Hilbert A-module. If U is a pre-Hilbert A-
module or a Hilbert A-module, the A-valued map (, )U : U × U → A is called an
A-product or a Hilbert A-product, respectively. For definiteness, we consider left
pre-Hilbert A-modules, and the A-products are supposed to be conjugate linear
in the first variable. For pre-Hilbert A-modules U, V, the space of continuous
A-module homomorphisms between U and V is denoted by HomA(U, V ) and its
elements are called pre-Hilbert A-module homomorphisms. If U, V are Hilbert
A-modules, we omit the prefix pre. The notion of continuity is meant with
respect to the norms U and V . Further, we say that u, v ∈ U are orthogonal
if (u, v)U = 0. When we write a finite direct sum, the summands are supposed to
be mutually orthogonal pre-Hilbert A-modules. The adjoints of maps between
pre-Hilbert A-modules are always thought with respect to the considered A-
products. Let us remark that there exist continuous maps on a Hilbert A-module
(U, (, )U ) which are adjointable with respect to a Hilbert space scalar product
U × U → C on U, the induced norm of which is equivalent to the norm U , but
which do not posses an adjoint with respect to (, )U . For it, see, e.g., Solovyov,
Troitsky [12]. A Hilbert A-module U is called projective if there exist n ∈ N0
and a Hilbert A-module V such An = U ⊕ V, where An denotes the direct sum
of n copies of the standard Hilbert A-module A. It is called finitely generated if
there exists a finite number of elements u1, . . . , uk ∈ U so that for each u ∈ U
i=1 aiui. Let us notice,
that the last mentioned Hilbert A-modules are sometimes called algebraically
finitely generated. For these notions, we refer the reader to Paschke [9] and
Solovyov, Troitsky [12].
there exist elements ai ∈ A, i = 1, . . . , k, such that u = Pk
Let us recall the following statement generalizing the rank-nullity theorem
from linear algebra.
Theorem 1: Let U, V be Hilbert A-modules and L ∈ HomA(U, V ) be an
adjointable map. If the image of L is closed, then also the image of L∗ is closed
and Rng L = (Ker L∗)⊥, Rng L∗ = (Ker L)⊥ and U = Ker L ⊕ Rng L∗.
Proof. See Lance [6] and a reference to the original proof there. (cid:3)
Remark 1: Under the same assumptions as in Theorem 1, the equality
Ker L∗ = (Rng L)⊥ follows immediately.
Lemma 2: Let U, V, W be pre-Hilbert A-modules and
U D
/ V
D′
/ W
be a sequence of adjointable pre-Hilbert A-module homomorphisms. Then for
△ = D′∗D′ + DD∗, we have
Ker △ = Ker D′ ∩ Ker D∗.
Proof. From the definition of △, we get the inclusion Ker D′ ∩ Ker D∗ ⊆
Ker △. For each v ∈ V, we may write (v, △v)V = (v, D′∗D′v + DD∗v)V =
(v, D′∗D′v)V + (v, DD∗v)V = (D′v, D′v)W + (D∗v, D∗v)U . From this, the op-
posite inclusion follows using the positive definiteness of the A-products on W
and U. (cid:3)
4
/
/
To each complex D• = (Di, Γi)i∈N0 of pre-Hilbert A-modules and adjointable
pre-Hilbert A-module homomorphisms, we attach the sequence
△i = Di−1D∗
i−1 + D∗
i Di, i ∈ N0,
of associated Laplacians, where we assume D−1 = 0 (according to the preamble).
Theorem 3: Let Γi, i = 1, . . . , 5, be pre-Hilbert A-modules, and let
Γ1 D1
/ Γ2 D2
/ Γ3 D3
/ Γ4 D4
/ Γ5
be a complex of adjointable pre-Hilbert A-module homomorphisms. Suppose
that for i = 1, . . . , 4, there exist elements gi, pi ∈ HomA(Γi, Γi) such that
1Γi = gi△i + pi,
1Γi = △igi + pi and
△ipi = 0.
Then for i = 1, 2, 3,
pi+1Di = 0 and Digi = gi+1Di.
(1)
(2)
(3)
(4)
Proof. We use the relation Dipi = 0, i = 1, . . . , 4, repeatedly, which follows
from Lemma 2 and from (3). For i = 1, 2, 3 and u ∈ Γi, we may write
pi+1Diu = pi+1Di(△igiu + piu)
i )Digiu
i Di)giu
i−1 + D∗
= pi+1Di△igiu + pi+1(Dipi)u
= pi+1Di(Di−1D∗
= pi+1(DiD∗
= pi+1△i+1Digiu
= (1 − △i+1gi+1)△i+1Digiu
= △i+1Digiu − △i+1gi+1△i+1Digiu
= △i+1Digiu − △i+1(1 − pi+1)Digiu
= △i+1Digiu − △i+1Digiu + △i+1pi+1Digiu
= 0,
where we used △i+1pi+1 = 0, i = 1, 2, 3, in the last step. Notice that in the
above computation (rows 2 and 5), we derived also the relation Di△i = △i+1Di,
which we use in what follows. For i = 1, 2, 3 and u ∈ Γi, we have
gi+1Diu = gi+1Di(△igi + pi)u
= gi+1Di△igiu + gi+1(Dipi)u
= gi+1△i+1Digiu
= (1 − pi+1)Digiu = Digiu,
where at the last step, we used the relation pi+1Di = 0 derived above. (cid:3)
Remark 2:
5
/
/
/
/
1) We call equations (1) and (2) the parametrix equations, and any map gi
satisfying (1), (2) and (3) the pseudoinverse of △i. In the special case
A = C, the name Green's functions is also used.
2) The second relation in row (4) says that g• = (gi)i=1,2,3 is a chain map.
Remark 3:
1) Notice that from (1), we get piKer △i
= 1Ker △i
fact and relation (3), we see that pi maps Γi onto Ker △i.
, i = 1, . . . , 4. Using this
2) Suppose now that (1), (2) and (3) are valid and the relation
gipi = 0
(5)
is satisfied for i = 1, . . . , 4. Applying pi on (2) from the right, we get
p2
i = pi, i.e., pi is an idempotent in HomA(Γi, Γi), i = 1, . . . , 4.
Definition 1: We call a complex D• = (Di, Γi)i∈N0 of pre-Hilbert A-
modules and adjointable pre-Hilbert A-module homomorphisms parametrix pos-
sessing if for each i ∈ N0, there exist elements gi, pi ∈ HomA(Γi, Γi) such that
relations (1), (2) and (3) are satisfied.
For a complex D• = (Di, Γi)i∈N0 of pre-Hilbert A-modules and pre-Hilbert
A-module homomorphisms, we consider the cohomology groups
H i(D, A) =
Ker(Di : Γi → Γi+1)
Rng(Di−1 : Γi−1 → Γi)
, i ∈ N0,
and denote the nominator, the cycles of the complex, by Z i(D, A), and the
space of boundaries, Rng (Di−1 : Γi−1 → Γi), by Bi(D, A). The cohomology
groups come up with the canonical quotient A-module structure. We do not
speak about a pre-Hilbert A-module structure on them because we do not know
whether Bi(D, A) is a closed subspace of Γi. Further, let us denote the kernel
of △i : Γi → Γi, the A-module of harmonic elements, by K i(D, A).
Theorem 4: If D• = (Di, Γi)i∈N0 is a parametrix possessing complex, then
for each i ∈ N0,
H i(D, A) ≃ K i(D, A) as A-modules.
Proof. Because D• is parametrix possessing, there exist maps pi, gi ∈
HomA(Γi, Γi) satisfying (1), (2) and (3) for each i ∈ N0.
1) Consider the map Φi : K i(D, A) → H i(D, A) given by Φia = [a], a ∈
K i(D, A). This map is well defined. Indeed, due to Lemma 2, a ∈ Ker △i
implies a ∈ Ker Di. It is evident that Φi is an A-module homomorphism.
2) Let us consider the map Ψi : H i(D, A) → K i(D, A) given by Ψi[a] =
pia for each a ∈ Z i(D, A). It follows from Theorem 3 that Ψi is well
defined. Indeed, for a = Di−1b we have pia = piDi−1b = 0 due to the
first relation in row (4). Now, we prove that Ψi is inverse to Φi. For
6
a ∈ K i(D, A), we have Ψi(Φia) = Ψi[a] = pia = a. On the other hand,
Φi(Ψi[b]) = Φi(pib) = [pib] for any b ∈ Z i(D, A). The proof of the fact
that b and b = pib are cohomologous proceeds as follows. Using the
parametrix equation (1), we get b − b = b − pib = (1 − pi)b = gi△ib =
giDi−1D∗
i−1b), where we
used the second equation in row (4) from Theorem 3 and the fact that
b ∈ Ker Di. It is evident that Ψi is an A-module homomorphism. (cid:3)
i−1b = Di−1(gi−1D∗
i−1b + giD∗
i Dib = giDi−1D∗
3 A-elliptic pseudodifferential operators
Let p : E → M n be a locally trivial Banach bundle and A be a unital Banach
C∗-algebra. We call p an A-Hilbert bundle if
1) there exists a Hilbert A-module (U, (, )U ) (the typical fiber of p),
2) for each x ∈ M, the fiber Ex = p−1(x) is equipped with a Hilbert A-
module structure, and it is isomorphic to U as a Hilbert A-module,
3) the subset topology on p−1(x) ⊆ E is equivalent to the norm topology on
p−1(x) given by U , and
4) the local transition maps between the bundle charts are maps into the
group AutA(U ) of Hilbert A-module automorphisms of U.
We call a smooth map S : E1 → E2 between the total spaces of A-Hilbert
bundles p1 : E1 → M and p2 : E2 → M with typical fibers U1 and U2, respec-
tively, an A-Hilbert bundle homomorphism if it satisfies the equation p2 ◦S = p1
(defining a bundle homomorphism), and if it is a Hilbert A-module homomor-
phism in each fiber. Let S be an A-Hilbert bundle homomorphism. We call an
2 (x) is adjoint
A-Hilbert bundle homomorphism T : E2 → E1 adjoint to S if Tp−1
1 (x) → p−1
to the Hilbert A-module homomorphism Sp−1
2 (x) for each
x ∈ M. In this case, we write T = S∗. An A-Hilbert bundle is called finitely
generated projective A-Hilbert bundle if its typical fiber is a finitely generated
projective Hilbert A-module.
1 (x) : p−1
The vector space Γ(E) of smooth sections of p carries a structure of an A-
module given by (as)(x) = a(s(x)), where a ∈ A, s ∈ Γ(E) and x ∈ M. For a
compact manifold M and a Riemannian metric g on M, we fix an associated
volume element on M and denote it by volg. The volume element induces an
A-product on Γ(E) by the formula
(s, s′)Γ(E) = Zx∈M
(s(x), s′(x))U volg(x) ∈ A,
where s, s′ ∈ Γ(E). At the right hand side, we consider the Bochner integral.
In this way, Γ(E) gains a structure of a pre-Hilbert A-module. We denote the
induced norm by Γ(E). For each t ∈ Z, one defines a further A-product (, )t
7
on Γ(E) by setting
(s, s′)t = Zx∈M
((1 + △g)ts(x), s′(x))U volg(x) ∈ A,
where △g is minus the Laplace-Beltrami operator of (M, g) and s, s′ ∈ Γ(E).
The induced norm will be denoted by t, and the completion of Γ(E) with
respect to t by W t(E). In particular, Γ(E) is dense in W t(E). Due to the
construction, W t(E) together with the extended A-product (, )t form a Hilbert
A-module. We call any (W t(E), (, )t) a Sobolev type completion of Γ(E). See
Fomenko, Mishchenko [3] for more on the introduced A-products.
Because 0Γ(E) = Γ(E), the Hilbert A-module W 0(E) coincides with the
completion of Γ(E) with respect to Γ(E). For t ∈ Z and k ≥ t > 0, it is not
difficult to see that
Γ(E) ⊆ Ck(E) ⊆ W t(E) ⊆ W t−1(E).
A section of E belongs to Ck(E) if and only if it is at least k times continuously
differentiable. To be more precise, let us notice that we consider the elements of
Γ(E), Ck(E) and W t(E) modulo the relation of being zero almost everywhere.
For any O open in M, we use the symbol Ck(O, E) to denote the space of the
appropriate sections of the restricted bundle pO : p−1(O) → O.
Let us briefly recall a definition of the Fourier transform of local sections
of A-Hilbert bundles over compact Riemannian manifolds. For a local chart
(O ⊆ Rn, φ : O → φ(O) ⊆ M n), the Fourier transform of a section s of pφ(O),
which has its support in a compact subset of φ(O), is defined by
s(q) = Zx∈φ(O)
e−2πıhq,xis(x)dx,
where q ∈ Rn and h, i denotes the appropriate Euclidean product on Rn. We
use the same notation for a section as for its coordinate expression, hoping this
causes no confusion. Globally, one has to choose an atlas and a subordinate
partition of unity, make the local Fourier transforms, and apply the appropriate
gluing process at the end. Let us notice that the Fourier transform depends on
the choice of a particular partition, but it exists if the underlying Riemannian
manifold is compact independently of this choice. As in the classical case, one
Γ(E)(1+
q2)tdq]
2 , and q ∈ Rn. To obtain a global
version of this formula, one shall apply the same procedure as in the case of the
Fourier transform.
can show that the norm t is equivalent to the norm st = [Rq∈Rn s(q)2
0 , where s ∈ Γ(E), q = hq, qi
1
1
2 ∈ R+
In the following lemma, an A-Hilbert bundle analogue of the Sobolev em-
bedding theorem is proved.
Lemma 5: Let p : E → M n be an A-Hilbert bundle over a compact
manifold M. Then for each t > ⌊ n
2 ⌋ + 1 and 0 ≤ k < t − ⌊ n
2 ⌋ − 1,
s ∈ W t(E) implies s ∈ Ck(E).
8
Proof. Let t > ⌊ n
2 ⌋ − 1. If s is a section of pO,
O ⊆ M, we denote its coordinate expression by s as well. For α ∈ Nn
0 and
α ≤ k, we show that ∂αs ∈ C0(O, E). Obviously, it is sufficient to prove that
2 ⌋ + 1 and 0 ≤ k < t − ⌊ n
Rq∈Rn s(q)qαe2πıhx,qidq converges for all x ∈ O. Because s ∈ W t(O, E), we know
that st = [Rq∈Rn s(q)2
2 < ∞. Let us compute
1
s(q)Γ(E)qαdq =
Γ(E) (1 + q2)tdq]
s(q)Γ(E)qαdq ≤ Zq∈Rn
s(q)Γ(E)(1 + q2)t/2
Zq∈Rn
= Zq∈Rn
≤ (cid:18)Zq∈Rn
= st(cid:18)Zq∈Rn
s(q)2
qα
(1 + q2)t/2 dq ≤
Γ(E)(1 + q2)tdq(cid:19)1/2(cid:18)Zq∈Rn
(1 + q2)t dq(cid:19)
q2α
1/2
,
q2α
(1 + q2)t dq(cid:19)
1/2
=
0
2 − 1
Bochner integral. Therefore ∂αs ∈ C0(O, E) for each α ≤ k < t − ⌊ n
written integral is equivalent to the finiteness of R +∞
where we used the Cauchy-Schwartz inequality in L2(Rn) at the second last
step. Using polar coordinates in Rn, we see that the finiteness of the last
r2α+n−1(1 + r2)−tdr.
Near r = 0, the integrand is a bounded continuous function. At the infinity,
the integrand behaves like r2α+n−2t which is integrable over (C, +∞) for each
C > 0 if and only if 2α + n − 2t < −1. Thus, for each α < t − n
2 , the
integral Rq∈Rn s(q)qαe2πıhx,qidq converges due to the absolute convergence of the
2 ⌋ − 1. (cid:3)
A reference for proofs of the statements in this paragraph is Solovyov, Troit-
sky [12]. For A-Hilbert bundles p1 : E1 → M n and p2 : E2 → M, one defines
a symbol of a pseudodifferential operator of order r ∈ Z. Besided other proper-
ties, the symbol is an adjointable A-Hilbert bundle homomorphism. Using the
Fourier transform, to each symbol of a pseudodifferential operator of order r,
one associates a pseudodifferential operator of order r and vice versa. Let us
notice that any pseudodifferential operator D : Γ(E1) → Γ(E2) of order r is,
in particular, an adjointable pre-Hilbert A-module homomorphism between the
pre-Hilbert A-modules Γ(E1) and Γ(E2). The order of D∗ equals that of the
operator D. Further, it is known that an arbitrary t ∈ Z, each pseudodifferen-
tial operator D : Γ(E1) → Γ(E2) of order r admits a unique Hilbert A-module
homomorphism Dt : W t(E1) → W t−r(E2) extending D and that any of these
extensions is adjointable. Further, the symbol of D∗ equals (up to a constant
non-zero multiple) to the adjoint of the symbol of D.
One calls a pseudodifferential operator K : Γ(E1) → Γ(E2) A-elliptic if for
each x ∈ M and ξ ∈ T ∗
x M \ {0}, the symbol
σ(x, ξ) : (E1)x → (E2)x
of K at (x, ξ) is a Hilbert A-module isomorphism. Further, let U, V be Hilbert
A-modules, and the decompositions U = U0 ⊕ U1 and V = V0 ⊕ V1 hold,
9
where U1 and V1 are Hilbert A-modules, and U0 and V0 are finitely generated
projective Hilbert A-modules. Whenever Ki ∈ HomA(Ui, Vi), i = 0, 1, and K1
is an isomorphisms, then K = K0 ⊕ K1 is called an A-Fredholm operator. It
is immediately seen that Ker K = Ker K0 and Coker K = V0/Rng K0. Let us
notice that in general, the image of an A-Fredholm operator is not closed in
contrary to the case of A = C.
Now, we recall the theorem of Fomenko and Mishchenko on the existence of
smoothing pseudoinverses mentioned in the Introduction.
Theorem 6: Let p : E → M be a finitely generated projective A-Hilbert
bundle over a compact manifold M, and K : Γ(E) → Γ(E) be an A-elliptic
operator of order r. Then for each t ∈ Z, the extension K t of K is an A-
Fredholm operator and there exists a Hilbert A-module homomorphism gt−r :
W t−r(E) → W t(E) satisfying gt−rK t − 1 : W t(E) → W t+1(E).
Proof. See Solovyov, Troitsky [12] (Theorems 2.1.142 and 2.1.146). (cid:3)
Now, it is quite easy to prove the regularity of A-elliptic operators.
Theorem 7: Let p : E → M be a finitely generated projective A-Hilbert
bundle over a compact manifold M and K : Γ(E) → Γ(E) be an A-elliptic
operator of order r. Then for each t ∈ Z, the equality Ker K t = Ker K ⊆ Γ(E)
holds.
Proof. The inclusion Ker K ⊆ Ker K t is obvious. For t ∈ Z, let us prove
that Ker K t ⊆ Ker K. Due to Theorem 6, there exists a map gt−r : W t−r(E) →
W t(E) such that gt−rK t − 1 : W t(E) → W t+1(E). For s ∈ W t(E), we may
write
s = (gt−rK t)s − (gt−rK t)s + s
= gt−r(K ts) − (gt−rK t − 1)s.
Assuming K ts = 0 for s ∈ W t(E), we get s = gt−r(K ts) − (gt−rK t − 1)s =
−(gt−rK t − 1)s ∈ W t+1(E) (Theorem 6). By induction, s ∈ W t(E) for each
t ∈ Z. Using Lemma 5, we obtain s ∈ Tk∈N0 Ck(E) = Γ(E). (cid:3)
Using the previous theorem, we prove a smooth version of Theorem 6.
Theorem 8: Let p : E → M be a finitely generated projective A-Hilbert
bundle over a compact manifold M, and K : Γ(E) → Γ(E) be a self-adjoint
A-elliptic operator of order r. If Rng K r is closed in W 0(E), then there exist
pre-Hilbert A-module homomorphisms P : Γ(E) → Γ(E) and G : Γ(E) → Γ(E)
such that
1Γ(E) = GK + P, 1Γ(E) = KG + P and KP = 0.
Proof. Notice that K r : W r(E) → W 0(E) and K r∗ : W 0(E) → W r(E) ⊆
W −r(E).
1) For each a, b ∈ Γ(E), we have ((K r)∗a, b)r = (a, K rb)0 = (a, Kb)Γ(E) =
(a, K ∗b)Γ(E) = (Ka, b)Γ(E). Summing up, ((K r)∗a, b)r = (Ka, b)Γ(E) for
any a, b ∈ Γ(E). For each a ∈ Ker K ⊆ Γ(E) (Theorem 7), we thus get
((K r)∗a, b)r = 0 for any b ∈ Γ(E). Because (, )r is continuous and Γ(E)
is dense in W r(E), we obtain ((K r)∗a, b)r = 0 for each b ∈ W r(E). Thus
from the non-degeneracy of (, )r, K r ∗a = 0, and consequently, Ker K ⊆
10
Ker K r∗. If a ∈ Ker K r∗ ∩ Γ(E), then from the previous computation
and the non-degeneracy of (, )0, we get a ∈ Ker K. Therefore Ker K =
Ker K r∗ ∩ Γ(E). Because K r∗ is also an A-elliptic operator, Theorem
7 applies, and we may omit the intersection with Γ(E) in the previous
expression, obtaining
Ker K = Ker K r∗.
(6)
2) Because the image of K r : W r(E) → W 0(E) is closed and K r is ad-
jointable, we have W r(E) = Ker K r ⊕ Rng K r∗ due to Theorem 1. There-
fore, the projection pKer K r : W r(E) → Ker K r from W r(E) onto Ker K r
is a well defined Hilbert A-module homomorphism. Because Rng K r =
(Ker K r ∗)⊥ ⊆ W 0(E) (Theorem 1), there exists a bijective Hilbert A-
module homomorphism
δ = K r
(Ker K r)⊥ : (Ker K r)⊥ → (Ker K r∗)⊥.
Due to Banach's open mapping theorem, its inverse
γ : (Ker K r∗)⊥ → (Ker K r)⊥
is continuous. As any inverse of an A-module homomorphism, the map γ
is an A-module homomorphism as well. Extending γ by zero on Ker K r∗ =
(Rng K r)⊥ (Remark 1), we get a Hilbert A-module homomorphism
γ : W 0(E) → (Ker K r)⊥ ⊆ W r(E).
3) Due to the construction, we obtain a parametrix type equation 1W r(E) =
γK r+pKer K r for K r. Indeed, for a ∈ Ker K r, we have γK ra+pKer K r a =
0 + a = a. For a ∈ (Ker K r)⊥, we get γK ra + pKer K r a = γK ra + 0 = a.
Let us denote the restriction to Γ(E) of pKer K r by P. Thus P : Γ(E) →
Ker K r. Because Ker K r = Ker K ⊆ Γ(E) (Theorem 7), we have a pre-
Hilbert A-module homomorphism P : Γ(E) → Γ(E) at our disposal.
Further, let us set G = γΓ(E). For a ∈ Ker K r∗ ⊆ Γ(E), we have Ga =
γa = 0. For a ∈ (Ker K r∗)⊥ ∩ Γ(E), there exists b ∈ (Ker K r)⊥ (item 2)
such that a = δb. Since a ∈ Γ(E) and δ is a restriction of an extension of a
pseudodifferential operator of finite order, b ∈ Γ(E). We have Ga = γa =
γδb = b. Especially, Ga ∈ Γ(E). Thus, G : Γ(E) → Γ(E). Now, we may
restrict the parametrix type equation for K r to the space Γ(E) to obtain
1Γ(E) = GK + P.
4) Similarly as above, we prove that the parametrix type equation 1W 0(E) =
K r γ +pKer K r ∗ holds, where pKer K r ∗ denotes the projection from W 0(E)
onto Ker K r∗. Indeed, for a ∈ Ker K r ∗, we obtain K rγa+pKer K r ∗a = 0+
a = a. For a ∈ (Ker K r∗)⊥, we have K rγa + pKer K r ∗ a = a + 0 = a. Using
formula (6) from item 1, we see that (pKerK r ∗ )Γ(E) = (pKer K r )Γ(E).
Thus restricting 1W 0(E) = K rγ + pKer K r ∗ to Γ(E), we get 1Γ(E) =
KG + P.
11
5) The equation KP = 0 follows from the definition of P and the inclusion
Ker K r ⊆ Γ(E) (Thm. 7).
(cid:3)
Remark 4: Under the conditions of Theorem 8 and from the constructions
in its proof, we get that GP = 0 and P 2 = P. See also Remark 3 item 2.
4 Complexes of A-elliptic pseudodifferential op-
erators
In this section, we focus our attention at complexes of pseudodifferential opera-
tors rather than at one pseudodifferential operator only. Let M be a manifold,
E• = (pi : Ei → M )i∈N0 be a sequence of A-Hilbert bundles over M, and
D• = (Di, Γ(Ei))i∈N0 be a complex of pseudodifferential operators acting be-
tween sections of the appropriate bundles, A being a unital Banach C∗-algebra.
Let σi be the symbol of the pseudodifferential operator Di : Γ(Ei) → Γ(Ei+1),
i ∈ N0. For each i ∈ N0, we define σ′
i to be the restriction of the A-Hilbert
bundle homomorphism σi to π∗(Ei), where π : T ∗M \ 0 → M is the foot point
projection from the cotangent bundle with the image of the zero section removed
onto the manifold M . Let us denote the resulting restricted complex consisting
of A-Hilbert bundle homomorphisms σ′
i, i ∈ N0, by σ•′.
Definition 2: A complex D• = (Di, Γ(Ei))i∈N0 of pseudodifferential oper-
ators is called an A-elliptic complex if σ•′ is an exact complex in the category
of A-Hilbert bundles.
The following lemma is a generalization to the case of Hilbert A-modules of
a reasoning used in the classical Hodge theory. For the classical case, the reader
is referred, e.g., to Wells [14].
Lemma 9: Let U, V, W be Hilbert A-modules and σ and σ′ be adjointable
Hilbert A-module homomorphisms. If
U
σ
/ V
σ′
/ W
is an exact sequence and the image of σ′ is closed, then Σ = σσ∗ + σ′∗σ′ is a
Hilbert A-module automorphism of V .
Proof. Due to the assumption, the Hilbert A-module homomorphisms σ∗
and σ′∗ exist. Therefore, Σ = σσ∗ + σ′∗σ′ is a well defined Hilbert A-module
endomorphism. It is immediately seen that Rng σ′∗ ⊆ Ker σ∗. Indeed, for any
v ∈ Rng σ′∗ there exists an element w ∈ W such that σ′∗w = v. Thus σ∗v =
(σ∗σ′∗)w = (σ′σ)∗w = 0.
Now, we prove the injectivity of σσ∗ restricted to Rng σ. Suppose there
exists an element u ∈ U such that v = σu satisfies σσ∗v = 0. We may write
0 = (σσ∗v, v)V = (σ∗v, σ∗v)U which implies σ∗v = 0. Thus 0 = (σ∗v, u)U =
(v, σu)V = (v, v)V which implies v = 0. Now, we prove that σ′∗σ′ is injective
on Rng σ′∗. Indeed, let v = σ′∗w for an element w ∈ W and σ′∗σ′v = 0. Then
σ′∗σ′v = 0 implies that (σ′v, σ′v)W = (v, σ′∗σ′v)V = 0, which in turn implies
12
/
/
σ′v = 0. We may compute 0 = (σ′v, w)W = (v, σ′∗w)V = (v, v)V from which
v = 0 follows. Therefore σ′∗σ′ is injective on Rng σ′∗.
Rng σ
Further, due to Theorem 1, we have V = Ker σ′ ⊕ Rng σ′∗. Because Rng σ =
Ker σ′, we may write V = Rng σ ⊕ Rng σ′∗. Since Rng σ ⊆ Ker σ′ and Rng σ′∗ ⊆
Ker σ∗, we get ΣRng σ = σσ∗
Rng σ′ ∗ . Suppose
there exists v = (v1, v2) ∈ Rng σ ⊕ Rng σ′∗ such that Σv = 0. Then 0 = Σv =
σσ∗v1 + σ′∗σ′v2. Because Rng σ ∩ Rng σ′∗ = 0, we have σσ∗v1 = σ′∗σ′v2 = 0.
Due to the injectivity of the appropriate restrictions of the maps σσ∗ and σ′∗σ′,
the last two equations imply v1 = v2 = 0. Thus Σ is injective on V.
and ΣRng σ′ ∗ = σ′∗σ′
Since the images of σ and σ′ are closed, the images of σσ∗ and σ′∗σ′ are
closed as well. (For it, see the proof of Theorem 3.2 in Lance [6].) Because σσ∗
and σ′∗σ′ have closed images, their sum Σ has closed image as well. Obviously,
Σ is adjointable. Due to Theorem 1, Σ is surjective because V = Rng (Σ∗) ⊕
Ker (Σ) = Rng (Σ∗) = Rng (Σ). (cid:3)
Remark 5: Going through the proof of the previous theorem, we see that
under the same conditions on σ and σ′, we obtain that the map λσ∗σ + µσσ∗
is a Hilbert A-module automorphism for all λ, µ ∈ C \ {0} as well.
Corollary 10: Let M be manifold and D• = (Di, Γ(Ei))i∈N0 be an A-
elliptic complex in A-Hilbert bundles over M . Then for each i ∈ N0, the asso-
ciated Laplacian △i is A-elliptic.
Proof. Because D• is an A-elliptic complex, the restricted symbol complex
σ•′ is exact. From the exactness of σ•′, one concludes that the images of the
Hilbert A-module homomorphisms σ′
i evaluated at any (x, ξ) ∈ T ∗M \ 0 are
closed. Further, they are adjointable, being the symbols of pseudodifferential
operators. Using Lemma 9, the maps Σi = σ∗
i−1, i ∈ N0, are Hilbert
A-module automorphisms when evaluated at (x, ξ) ∈ T ∗M \ 0. Consequently,
the associated Laplacians △i, i ∈ N0, of D• are A-elliptic operators. (cid:3)
i σi + σi−1σ∗
We denote the order of the associated Laplacian △i by ri, i ∈ N0. Further, we
fix the topologies on the spaces which appear in the definition of the cohomology
groups of a complex D• = (Di, Γ(Ei))i∈N0 of pseudodifferential operators. We
consider the spaces Γ(Ei) with the topology given by the norm Γ(E i) and
the spaces Z i(D, A), Bi(D, A) ⊆ Γ(Ei) with the subset topologies. Notice that
Γ(Ei) is not a Banach space in general. We assume each cohomology group
H i(D, A) to be equipped with the quotient topology.
Theorem 11: Let D• = (Di, Γ(Ei))i∈N0 be an A-elliptic complex in finitely
If the
generated projective A-Hilbert bundles over a compact manifold M .
image of △ri
i
is closed in W 0(Ei), i ∈ N0, then for each i ∈ N0
1) the cohomology group H i(D, A) of D• is a finitely generated A-module,
2) the image of Di is closed in the norm topology on Γ(Ei), and
3) H i(D, A) is a Banach topological vector space.
Proof. Let us fix a non-negative integer i ∈ N0. According to Corollary 10,
the Laplacian △i is an A-elliptic operator. Since it is self-adjoint and Rng △ri
i ⊆
W 0(Ei) is closed, D• is a parametrix possessing complex according to Theorem
13
8. Let us denote the projection of Γ(Ei) onto K i(D, A) from Theorem 8 (E =
Ei, K = △i) by Pi.
1) Because D• is a parametrix possessing complex, K i(D, A) is A-linearly
isomorphic to H i(D, A) (Theorem 4). Since △i is A-elliptic, Ker △ri
i =
K i(D, A) due to the Theorem 7. From Theorem 6, we know that △ri
i
is A-Fredholm and thus, especially, Ker △ri
is finitely generated as an
i
A-module. Therefore H i(D, A) is finitely generated over A as well.
2) Now, we prove that H i(D, A) is a Hausdorff space. It is obvious that the
maps
Φi : K i(D, A) → H i(D, A) and Ψi : H i(D, A) → K i(D, A)
defined by Φia = [a] and Ψi[a] = Pia, a ∈ Z i(D, A) are continuous. For
the fact that they are mutually inverse, see item 2 in the proof of Theo-
rem 4. Thus, K i(D, A) is homeomorphic to H i(D, A). Because K i(D, A)
is a Hausdorff space, being a subspace of the Hausdorff space Γ(Ei),
H i(D, A) is a Hausdorff topological space as well. Since H i(D, A) =
Z i(D, A)/Bi(D, A) and Z i(D, A) is closed in Γ(Ei), being the kernel of a
continuous map, the space Bi(D, A) of boundaries in Γ(Ei) is closed in
Γ(Ei) as well.
3) Proving that K i(D, A) is a Banach space with respect to the norm topol-
ogy on Γ(Ei) we are done due to item 2 of this proof. Let us consider the
subset topology on Ker △0
i inherited from the topology of (W 0(Ei), 0).
Because Γ(E i) = 0Γ(E), any sequence in K i(D, A) which is cauchy
with respect to Γ(E i) is also cauchy with respect to 0 and thus, it has
a limit in W 0(Ei) since W 0(Ei) is complete. Because K i(D, A) equals
: W 0(Ei) → W −ri(Ei)) (Thm. 7), which is closed in W 0(Ei)
Ker (△0
i
being the kernel of a continuous map, the mentioned limit belongs also to
K i(D, A). Since the norms Γ(E i) and 0 (the latter restricted to Γ(Ei))
coincide, the limit is also a limit with respect to the norm Γ(E i).
(cid:3)
References
[1] Bartholdi, L., Schick, T., Smale, N., Smale, S., Hodge theory on metric
spaces, Found. Comput. Math. 12 (2012), No. 1, pp. 1 -- 48.
[2] Fedosov, B., A simple geometrical construction of deformation quantiza-
tion. J. Differential Geom. 40 (1994), No. 2, pp. 213 -- 238.
[3] Fomenko, A., Mishchenko, A., The index of elliptic operators over C∗-
algebras, Izv. Akad. Nauk SSSR, Ser. Mat. 43, No. 4, 1979, pp. 831 -- 859,
967.
14
[4] Habermann, K., Basic properties of symplectic Dirac operators, Comm.
Math. Phys. 184 (1997), No. 3, pp. 629 -- 652.
[5] Kostant, B., Symplectic spinors. Symposia Mathematica, Vol. XIV (Con-
vegno di Geometria Simplettica e Fisica Matematica, INDAM, Rome,
1973), pp. 139 -- 152. Academic Press, London, 1974.
[6] Lance, C., Hilbert C∗-modules. A toolkit for operator algebraists. Lon-
don Mathematical Society Lecture Note Series, 210, Cambridge University
Press, Cambridge, 1995.
[7] Meise, R., Vogt, D., Introduction to Functional Analysis, Clarendon Press,
Oxford (1997).
[8] Mishchenko, A., Banach algebras, pseudodifferential operators, and their
application to K-theory, Uspekhi Mat. Nauk 34, No. 6, 1979, pp. 67 -- 79.
[9] Paschke, W., Inner product modules over B∗-algebras, Trans. Amer. Math.
Soc., Vol. 182, 1973, pp. 443 -- 468.
[10] Schick, T., L2-index theorems, KK-theory, and connections, New York J.
Math. 11 (2005), pp. 387 -- 443.
[11] Smale, N., Smale, S., Abstract and classical Hodge-deRham theory, Anal.
Appl. (Singap.) 10 (2012), No. 1, pp. 91 -- 111.
[12] Solovyov, Y., Troitsky, E., C∗-algebras and elliptic operators in differen-
tial topology. Translations of Mathematical Monographs, 192. American
Mathematical Society, Providence, Rhode-Island, 2001.
[13] Troitsky, E., The index of equivariant elliptic operators over C -algebras,
Ann. Global Anal. Geom. 5 (1987), No. 1, pp. 3 -- 22.
[14] Wells, R., Differential analysis on complex manifolds, Graduate Texts in
Mathematics, Vol. 65, Springer, New York, 2008.
15
|
1206.6970 | 2 | 1206 | 2013-08-02T14:18:58 | Super Operator Systems, Strong Norms, and Operator Tensor Products | [
"math.OA"
] | A notion of super operator system is defined which generalizes the usual notion of operator systems to include certain unital involutive operator spaces which cannot be represented completely isometric as a concrete operator system on some Hilbert space. They can nevertheless be represented by bounded operators on a standard Z_2-graded Hilbert space equipped with a superinvolution. We apply this theory to investigate on the relation between certain tensor products defined for operator spaces and C^*-algebras, such as the projective tensor product, the Haagerup tensor product and the maximal C^*-tensor product. | math.OA | math |
SUPER OPERATOR SYSTEMS, STRONG NORMS
AND OPERATOR TENSOR PRODUCTS
U. HAAG
Abstract. A notion of super operator system is defined which generalizes the
usual notion of operator systems to include certain unital involutive operator
spaces which cannot be represented completely isometric as a concrete oper-
ator system on some Hilbert space. They can nevertheless be represented by
bounded operators on a standard Z2-graded Hilbert space equipped with a
superinvolution. We apply this theory to investigate on the relation between
certain tensor products defined for operator spaces and C ∗-algebras, such as
the projective tensor product, the Haagerup tensor product and the maximal
C ∗-tensor product.
0. Introduction
As the title indicates this paper is primarily concerned with the descrip-
tion and examination of certain operator spaces with some additional
structure we call super operator systems. If an operator space can be
regarded as the quantization of a Banach space, then a unital operator
space is sort of an operator space with a backbone. A super operator
system is a unital operator space which posesses an antilinear invo-
lution such that for each natural number n its composition with the
transposition operation on n×n-matrices is isometric on the n-fold am-
plification of the operator space (tensor product with Mn(C)). Also
the involution is required to fix the identity element. Because the term
super is sometimes overused we feel obliged to give some justification.
Following the general abstract characterization of an operator space
due to Ruan (cf.
[3]) abstract characterizations of a unital operator
space have been provided in [1] and [2]. One knows that given any
concretely represented unital operator space, its enveloping operator
system is uniquely determined up to complete ∗-isometry regardless
of the particular representation so that it is really determined by the
abstract structure of a unital operator space. Therefore a super oper-
ator system has a unique enveloping operator system which is in fact
Z2-graded. The (completely isometric ∗-linear) grading comes in by
Date: September 28, 2018.
1
2
U. HAAG
composing the antilinear involution with the antilinear adjoint opera-
tion of the enveloping operator system. It is then easy to see that any
abstract super operator system has a concrete realization on Z2-graded
Hilbert space which is ∗-linear with respect to the product of the grad-
ing automorphism plus the adjoint operation which is called the super
involution. On the other hand it is not always possible to get a ∗-linear
representation on ungraded Hilbert space (i.e. if and only if the super
operator system is in fact an operator system). Since there are inter-
esting super operator systems appearing quite naturally, for example
by considering certain operator tensor products of operator systems or
C ∗-algebras, which are not operator systems by themselves one might
say that Z2-gradings occur naturally in mathematics and especially in
operator theory. The term super operator system then appears quite
appropriate for this generalization of the concept of an operator system.
One is also reminded of certain concepts in mathematical physics, es-
pecially the notion of supersymmetry (c.f. the article [4] of R. Haag, J.
Lopuszanski and M. Sohnius). The author knows to little about these
matters to decide whether there might be some deeper applications of
the theory of super operator systems in physics but hopes that physi-
cists will feel challenged to consider this question in more detail. This
paper is self-contained to a large extent, using only standard results
from the theory of operator spaces which can all be found together
with further references to original papers in the books [3] by E. Effros
and Z.-J. Ruan and [5] by V. Paulsen, so we have chosen these two as
a general reference frame for this text. The article is organized into
three sections: the first gives the basic definitions and results for su-
per operator systems and super C ∗-algebras, including the notion of
superpositivity, the second section is an excursion to (super) operator
spaces equipped with what is called a strong matrix norm in our ter-
minology, it does not have any direct applications in the sequel so the
reader mainly interested in the results of the last section may well skip
the second one, but the feeling of the author is that some of the results
(and primarily the so called Strong Lemma) might have very interesting
applications in the future yet to be discovered (as was intended in the
writing) so we choose to maintain the section as it is leaving the judge-
ment whether or not it is worthwile reading to the recipient. Finally,
the third section may be considered as the main part since it gives some
of the more profound applications of the theory, most of them dealing
with operator tensor products of various kind, and centered around the
Super-Christensen-Effros-Sinclair Theorem (SCES-Theorem for short)
which gives some sufficient conditions for the action of a C ∗-algebra or
topological group on an operator system to be unitarily implementable
(strongly completely contractive).
1. Super operator systems and superpositivity
3
Let H denote a Hilbert space of a given finite or infinite dimension,
usually we assume that H is infinite but countably generated. The
standard Z2-graded Hilbert space is the direct sum of two copies of
H one of which is supposed to have even and the other odd grading,
matrix
bH = Heven ⊕ Hodd . Thus the grading operator on bH is given by the
0
ǫ =(cid:18)1
0 −1(cid:19)
with respect to this decomposition. Conjugating bounded operators
on bH by ǫ defines an order two automorphism compatible with the
adjoint operation. One then defines the superinvolution on B(bH) as
phism of B(bH) .
a norm-closed subalgebra of B(bH) which is invariant under the super-
product of the grading automorphism and the adjoint operation. It is
clear that the superinvolution is an isometric antilinear antiautomor-
In the following the notation x 7→ x∗ will always
allude to the superinvolution unless stated otherwise. To prevent mix-
ing up the super involution with the ordinary adjoint operation the
latter is denoted x 7→ x . By definition, a super-C ∗-algebra consists of
involution. Note that a super-C ∗-algebra need not be Z2-graded by
itself, i.e.
it is not required to be closed under the grading automor-
phism (and as a consequence also not by the ordinary adjoint opera-
tion). It is a certain type of operator algebra which in special cases
reduces to an ordinary C ∗-algebra (possibly with Z2-grading). Much
in the same way one defines a (concrete) super operator system to con-
element and is invariant under the superinvolution. Such a subspace is
naturally equipped with the induced operator matrix norms on the col-
sist of a norm-closed subspace X of B(bH) which contains the unit
lection of subspaces Mn(X) ⊆ Mn(B(bH)) . X may or may not contain
the grading operator ǫ . If it does we will refer to X as a complete super
operator system. Let A be a (for the time being concrete) super C ∗-
algebra. We want our representation to be as simple as possible. Given
any ∗-representation of A on a Z2-graded Hilbert space, consider the
enveloping Z2-graded C ∗-algebra bA and let A denote its underlying
trivially graded C ∗-algebra. Any (faithful) graded representation of bA
entails a (completely) isometric ∗-representation of A . Now assume
given a representation of A on ordinary ungraded Hilbert space. Let
4
U. HAAG
C1 denote the complex Clifford algebra on one odd generator. One has
C ∗-algebra M2( A) but with standard off-diagonal grading) determine
a natural Z2-graded isomorphism A ⊗ C1 ≃ bA ⊗ C1 so that the nat-
ural embeddings bA ֒→ bA ⊗ C1 and A ⊗ C1 ֒→ cM2( A) (cM2( A) is the
a Z2-graded representation of bA via the ungraded representation of
A . Then any hermitian (= superselfadjoint) element of bA takes the
form
(cid:18) a ib
ib a(cid:19)
with a and b selfadjoint. Let us call such a representation of standard
form. Given a contractive unital ∗-homomorphism of one unital super-
C ∗-algebra A into a unital super-C ∗-algebra B , both represented in
standard form, one obtains an induced contractive ∗-homomorphism
from A ⊗ C1 to B ⊗ C1 where the tensor product is endowed with
the obvious multiplication and product type superinvolution. Actually,
A ⊗ C1 is represented on the same Z2-graded Hilbert space, sending
the degree one (ordinary) self-adjoint generator F of C1 with F 2 = 1
to the matrix
(cid:18)0 1
1 0(cid:19)
∗-homomorphism of the enveloping Z2-graded C ∗-algebras bA , bB . In
owing to the fact that A is in standard form. Consider the super-C ∗-
algebras α(A) and α(B) (we will always let α denote the grading
automorphism on the ambient algebra of bounded operators). Since
α is an isometry one obtains a contractive ∗-homomorphism from
α(A) to α(B) . We would like to glue these two ∗-homomorphisms
together over the intersection (A ∩ α(A)) and extend to a Z2-graded
order to do this one must ensure that both homomorphisms agree on
the intersection which is clearly a Z2-graded C ∗-algebra by itself being
closed under superinvolution as well as the grading automorphism. A
necessary condition is that the restricted kernel of ϕ is invariant un-
der the grading automorphism. This turns out to be automatic from
the fact that the restricted kernel is necessarily a ∗-invariant ideal and
the Lemma below. Then the image of the intersection as above under
a contractive ∗-homomorphism is isometrically isomorphic with a Z2-
graded C ∗-algebra. A contractive ∗-representation of such an algebra
is also selfadjoint with respect to the ordinary involution and hence
graded by the following Lemma (this result is well known and included
only for convenience).
Lemma 1. Let ϕ : bA B(bH) be a contractive unital ∗-representation
(with respect to superinvolution) of a Z2-graded unital C ∗-algebra.
Then ϕ is a graded C ∗-representation.
5
Proof. Any unital contractive representation of an ordinary C ∗-algebra
is selfadjoint. This follows for example from the fact that an opera-
tor in B(H) is selfadjoint of norm less or equal to one if and only if
kx − it1k ≤ √1 + t2 for all t ∈ R (c.f. [3], Lemma A.4.2)
Consider the C ∗-algebra C ∗(1, ǫ) ≃ C ⊕ C generated by the unit and
the grading operator and let ω be any unitary operator of this algebra.
(cid:3)
One can define an isometric grading preserving involution on B(bH) by
the formula x 7→ xω := ω · x · ω where x 7→ x denotes the ordinary
adjoint operation. In particular x = x1 and x∗ = xǫ . An element
such that xω = x will be called ω-hermitian. There is an analogue of
the cited result for ω-hermitian elements, namely
Lemma 2. Let x be an operator in B(bH) with specified grading
operator ǫ as above. Then x is ω-hermitian of norm less or equal to
one if and only if for every t ∈ R the following inequality holds
kx − itωk ≤
√1 + t2 .
Proof. Let x be ω-hermitian of norm less or equal than one. Choose
a unitary square root γ of ω−1 , i.e. γ2 = ω−1 . With respect to the
matrix decomposition given by the grading one has
ω =(cid:18)ω0
0 ω1(cid:19) ,
1 d∗ . Multiplying the inequality kx − itωk ≤ √1 + t2
The condition that x is ω-hermitian means that a = ω2
ω0 ω1 c∗ , d = ω2
by γ from the left and from the right one obtains
x =(cid:18)a b
c d(cid:19) .
γ =(cid:18)γ0
0 a∗ , b =
0
γ1(cid:19) ,
0
0
√1 + t2
kex − it1k ≤
with ex = γ · x · γ an ordinary selfadjoint element whenever
by Lemma A.4.2 of [3]. The reverse direction is much the same
kxk ≤ 1
(cid:3)
Note however that the ω-involution is (anti)multiplicative and unit
preserving only if ω is selfadjoint.
In the applications we will usu-
ally have ω0 = 1 .
In view of Lemma 1 we may glue the two maps
as above over the common intersection of their domains to obtain a
Z2-graded ∗-linear map φ : (A + α(A)) → (B + α(B)) which restricted
6
U. HAAG
to either A or α(A) is a ∗-homomorphism. The extension is con-
tractive by [5], Proposition 2.12.. Since φ is also unital, it is positive
in the ordinary sense. The same construction carries over to matrices
of arbitrary size whenever ϕ is completely contractive. Then φ is
completely positive. Note that we have not made explicit use of the
fact that ϕ is an algebra homomorphism. Restricting to the ∗-linear
structure the same arguments yield for every ∗-invariant subspace X
of a super-C ∗-algebra canonically an enveloping Z2-graded operator
system bX . In particular starting with a Z2-graded operator system,
its structure as an operator system is uniquely determined by its struc-
ture as a super operator system. But the latter contrary to the former
also determines the Z2-grading and is thus finer than the operator sys-
tem structure which might be a bit of a surprise. We would like to
extend not only the ∗-linear, but also the multiplicative structure to
the graded setting. However the enveloping graded C ∗-algebra need
not be unique, but may depend on the particular representation.
It
is known that for any unital operator algebra A there is a minimal
choice of enveloping C ∗-algebra C ∗
min(A) which has the property that
any other C ∗-algebra generated by A in some particular representa-
tion canonically surjects onto C ∗
min(A) (Hamana's Theorem, compare
[5] Theorem 15.16). Correspondingly there is a maximal choice of en-
veloping C ∗-algebra which is obtained by taking the direct sum of all
(spatial equivalence classes of) completely contractive unital represen-
tations of A . Denote this algebra by C ∗
max(A) . This construction
has the advantage that it is functorial, i.e. given any completely con-
tractive homomorphism A → B one gets an induced ∗-homomorphism
C ∗
max(A) → C ∗
max(B) . The following simple example illustrates that
Cmax(A) might be different from C ∗
min(A) .
Example. Let
1
x = (cid:18) 1
−1 −1(cid:19) ∈ cM2(C)
and A0 be the commutative super-C ∗-algebra generated by x and
1 . Then the minimal enveloping graded C ∗-algebra is clearly equal to
cM2(C) . On the other hand, since x is nilpotent, there exists a non-
trivial ∗-homomorphism A0 → C that cannot be extended to cM2(C)
which is simple, so that bA0,max = C ∗
cM2(C) .
max(A0) must be an extension of
7
Let us return to the concept of super operator system for which one
would like to have some abstract characterization. There is some hid-
den structure which so far has remained unrevealed. This is connected
with the concept of superpositivity and of ǫ-positivity. As we have seen
one recaptures from a given super operator system X a Z2-graded
by positive elements. In a way the grading operator ǫ can be viewed
as a sort of second order unit with respect to an ordering by so-called
ǫ-positive elements. Define the graded spectrum of an element x ∈ X
operator system bX which entails the usual notion of matrix ordering
to be the subset bσ(x) ⊆ C such that λ ∈bσ(x) iff the operator x− λǫ
is not invertible in B(bH) . Of course this definition depends on the
particular representation. For example it is clear that the graded spec-
trum of any element in a standard representation of X is invariant with
respect to the operation λ 7→ −λ whereas if x is a positive element
which is homogenous of even degree, there exists a representation such
that bσ(x) coincides with the ordinary spectrum in that representation
which is purely positive. There are however some invariants. For ex-
ample it follows from the proof of Lemma 2 that multiplication with
the matrix
κ = (cid:18)1 0
0 i(cid:19)
from the left and right defines a complete isometry of a complete super
operator system X containing the grading operator with an ordinary
operator system X such that the grading operator ǫ transforms into
the unit element. Thus the norm of any hermitian element is given
by the spectral radius of its graded spectrum. For a given complete
super operator system X define a matrix order on the hermitian ele-
ments by the (convex Archimedean) cones Cǫ
n,+ ⊆ Mn(X) of elements
with positive graded spectrum (= ǫ-positive elements). From the com-
pletely isometric order isomorphism of X with X one deduces that
any hermitian element is the difference of two ǫ-positive elements, i.e.
Mn(X)h = Cǫ
n,+ = {0} . In the light of this
new perspective we have to rethink our definition of a super-C ∗-algebra,
which in any given representation can be completed to a complete su-
per operator system by adjoining the grading operator. Of course if
the algebra already contains the grading operator it is necessarily a
Z2-graded C ∗-algebra, but this requirement seems to restrictive.
In-
stead, one may think of the grading operator as a virtual graded unit
which is not part of the multiplicative algebraic structure of A , but
only of the ∗-linear structure with induced ǫ-ordering on the underlying
complete super operator system. This will be called a complete super
n,+ , and Cǫ
n,+ − Cǫ
n,+ ∩ −Cǫ
8
U. HAAG
C ∗-algebra as opposed to an ordinary super C ∗-algebra given by the
∗-invariant (unital) operator algebra A . The ǫ-order is not intrinsic
in the ∗-multiplicative structure of a given super-C ∗-algebra, and has
to be specified separately. We will now try to give a definition which
is independent of a particular representation. An abstract operator
space with specified element 1 which can be represented on Hilbert
space such that the specified element corresponds to the unit element
of B(H) will be called a unital operator space. It follows from [5], Prop.
3.5. that the enveloping operator system of a unital operator space is
well defined irrespective of the particular unital representation. Any
(ordinary) selfadjoint linear subspace of B(H) which does not necessar-
ily contain the unit element will be called a pre-operator system. If one
wants an abstract definition of a (complete) super operator system one
should first extract the relevant components of the structure of a given
concrete (complete) super operator system. One starts with a unital
operator space which is selfadjoint with respect to some isometric in-
volution and contains beside the unit another specified element ǫ , the
graded unit. The graded unit has several aspects. First of all, being
mines the grading automorphism on the completed Z2-graded operator
since ǫ · ǫ = 1 one has ǫ ∗ ǫ = 1 so that ǫ is a selfadjoint unitary of
property of being invertible is not necessarily inherent in the abstract
definition of an operator system (the spectrum of an element might
differ with the concrete representation). But, attached to the operator
the grading operator in B(bH) , it is a selfadjoint unitary and deter-
system bX which is obtained through conjugation with ǫ . Now the
system bX in a canonical manner, is its injective envelope I(bX) which
is a unital C ∗-algebra. If bX is represented on a Hilbert space H , then
bX ⊆ I(bX) ⊆ B(H) and there exists a completely positive projection
ϕ : B(H) → B(H) with range I(bX) such that ϕ2 = ϕ , and multipli-
cation in I(bX) is given by the formula x ∗ y = ϕ(x · y) . In particular,
even degree in I(bX) . The grading automorphism extends from bX to
I(bX) by letting α(x) = ϕ(α(x)) where α is any completely positive
on I(bX) from the fact that α2 = id on bX (compare [5], chap. 15).
is implemented in the abstract C ∗-algebra I(bX) through conjugation
From this and the identity ǫ ∗ x ∗ ǫ = ϕ(ǫ · x · ǫ) one gets that α
with ǫ . We gather these results in the following definition.
extension of α to B(H) (which exists from Arvesons extension theo-
rem); of course in a Z2-graded representation the extension is obvious.
The point is that no matter what extension, it is uniquely determined
9
(i) An abstract complete super operator system is a
Definition 1.
quadruple (X , ∗ , 1 , ǫ) where X is completely isometric to an abstract
operator system with order unit ǫ (the graded unit) and completely an-
tiisometric involution ∗ (which means that the involution composed
with the transposition of matrices is isometric for each Mn(X) ), de-
termining a matrix order given by a sequence of Archimedean cones of
ǫ-positive elements {Cǫ
n,+} ⊆ Mn(X)h , and containing another speci-
fied element 1 invariant under the involution, the unit of X . It then
follows that its injective envelope I(X, ǫ) is a C ∗-algebra with unit
ǫ and uniquely determined multiplication. It is assumed that 1 is a
selfadjoint unitary for this multiplicative structure. Any ∗-invariant
subspace of a complete abstract super operator system containing the
unit 1 is referred to as an (ordinary) super operator system.
A supermorphism of super operator systems is a (completely) bounded
∗-linear map s : X → Y . s is ǫ-unital, resp. unital, iff it maps the
graded unit, resp. unit, of X to the graded unit, resp. unit, of Y .
s is (completely) ǫ-positive iff s ( sn ) maps ǫ-positive elements into
ǫ-positive elements where sn : Mn(X) → Mn(Y) is the induced map.
(ii) An abstract (unital) complete super-C ∗-algebra is a pair (A, ǫ)
where A is a (unital) operator algebra equipped with a completely
antiisometric antimultiplicative involution, also referred to as an (ordi-
nary) abstract super-C ∗-algebra which together with the matrix norms
extends to the linear space A = A + C · ǫ such that A is a complete
super operator system with graded unit ǫ . A complete superhomo-
morphism φ : (A, ǫA) → (B, ǫB) is a (completely) bounded super-
morphism from A to B which restricts to a (completely) bounded
∗-homomorphism from A to B .
The case that ǫ is contained in A is not excluded in the above defi-
nition of a complete super-C ∗-algebra, but not assumed either. In (i)
the case that ǫ equals 1 is also not excluded. In this case X becomes
an operator system in the usual sense.
Example. As an example one checks that any two isometric standard
representations of a given super-C ∗-algebra A extend to an isometric
supermorphism of ǫ-hermitian parts of the completed concrete super-
C ∗-algebras. The task is, given two different standard ∗-representations
ρ , σ of A to extend the ∗-isometry ρ(A) → σ(A) to a ∗-isometry
ρ(A)h + R· ǫρ → σ(A)h + R· ǫσ . First of all, since both representations
are standard the grading operators ǫρ and ǫσ are not contained in the
respective images of A . Since the graded spectrum of each hermitian
element x is invariant by the inverse map, one easily calculates the
10
U. HAAG
norm of x + λǫ for each λ ∈ R and finds that it is independent of the
representation. Thus the natural extension of ρ(A)h → σ(A)h to the
hermitian parts of the completed super operator systems is an isometry,
and ǫ-unital, hence (bi-)ǫ-positive.
We next show that any abstract complete super operator system (and
abstract super-C ∗-algebra) can be represented as a concrete super op-
erator system on Z2-graded Hilbert space (concrete super-C ∗-algebra).
It is clear from the Choi-Effros representation theorem , that the struc-
ture of operator system with respect to the graded unit ǫ can be rep-
resented, so the problem is to find a representation which also suits the
structure of unital operator space with respect to the unit 1 . Thanks
to the discussion preceding the definition the proof is very simple (the
incomplete case is deferred until a better understanding of abstract
super operator system has been reached).
Theorem. (Complete super operator systems representation theorem)
(i) Any abstract complete super operator system (X , ∗ , 1 , ǫ) has a
faithful unital completely isometric ∗-representation on a Z2-graded
sponds to the grading operator.
It then follows that the enveloping
operator system of (X , 1) is well defined and decomposes as a direct
Hilbert space bH (with respect to superinvolution) such that ǫ corre-
sum bX = X0 ⊕ X1 , where X0 is the operator system generated by the
projections p0 :bX ։ X0 and p1 :bX ։ X1 are complete quotient map-
into bX .
set {x + x x = x∗, x ∈ X} , and the pre-operator system X1 is gen-
erated by the set {x − x x = x∗, x ∈ X} . Moreover, the two natural
pings with respect to the norms determined by the natural inclusions
(ii) Any abstract super-C ∗-algebra has a completely isometric unital ∗-
representation on Z2-graded Hilbert space.
Proof. Consider the Z2-graded C ∗-algebra I(X, ǫ) , the grading auto-
morphism being defined as conjugation with the image of 1 . It has
upon regarding the image of 1 as a grading operator inherits a Z2-
grading. Then consider the element κ = π ((1 + ǫ)/2 + i(1 − ǫ)/2) .
The correponding completely isometric representation of X ⊂ I(X)
thereby becomes a unital and ǫ-unital ∗-representation with respect
a unital faithful C ∗-representation π on some Hilbert space bH which
The map x 7→ κ · x · κ of B(bH) exchanges the images of 1 and ǫ .
to superinvolution in B(bH) , hence completely ǫ-positive. Consider its
enveloping Z2-graded operator system bX . To get the last statement
of (i) recall that any Z2-graded operator system can be represented in
standard form. Then any element is of the form
11
(cid:18)a b
b a(cid:19) .
It is clear that kak = inf b∈bX1 ka + bk and in the same manner kbk =
inf a∈bX0 ka + bk . So the natural projections p0 and p1 are complete
quotient mappings. This proves (i). For the first statement in (ii)
recall that by the Blecher-Ruan-Sinclair-Theorem A has a unital com-
pletely isometric representation on Hilbert space, in particular A being
graded by composing the antilinear completely antiisometric involution
on A with the antilinear completely antiisometric adjoint operation)
a unital operator space its enveloping operator system bX (naturally Z2-
is well defined with injective envelope I(bX, 1) a Z2-graded C ∗-algebra.
that the injection A ⊆bX ⊆ I(bX, 1) is a ∗-homomorphism. To get our
representation consider any unital graded ∗-representation of I(bX, 1) .
It is clear from the definition of multiplication on the injective envelope
Its restriction to A will give the desired representation
(cid:3)
The following result is well known (c.f.
[3], Corollary 5.2.3) and in-
cluded only for convenience of the reader since we will frequently use
it in the sequel.
Proposition 1. Let Φ : A → B be a bijective unital, linear complete
isometry of two C ∗-algebras A and B . Then Φ is a ∗-isomorphism (cid:3)
The draw-back of the ordering by ǫ-positive elements is that it is not
intrinsic in the ∗-algebraic structure of a given super-C ∗-algebra A .
In practice the completed super operator system of A will only be
determined by a particular representation on graded Hilbert space.
Here the ǫ-order is of some value in distinguishing between different
representations as demonstrated in the example preceding the repre-
sentation theorem. We will therefore call two completely isometric
∗-representations of an incomplete super operator system ǫ-equivalent
iff the induced complete isometry of the represented concrete super
operator systems extends to an ǫ-unital complete isometry on the com-
pleted super operator systems. We will now sketch how to derive an
intrinsic ordering which is independent of the representation. Think-
ing of how positive elements in a C ∗-algebra are defined one may be
tempted to define a matrix order on a given super-C ∗-algebra A by
considering the closure C∗
n,+(A) of the convex cone of positive linear
combinations of elements of the form xx∗ with x ∈ Mn(A) . In certain
12
U. HAAG
n,+∩−C∗
r,+ · α ⊆ C∗
cases this will indeed result in a matrix order on the given super-C ∗-
algebra. This means that the resulting cones should linearly generate
the sets of hermitian elements in Mn(A) , such that C∗
n,+ = {0} ,
and that for any hermitian element x there exists a positive number
r ≥ 0 with x + r1n ∈ C∗
n,+ . Also for a given matrix α ∈ Mn,r(C)
one must have α · C∗
n,+ . A super C ∗-algebra posessing this
property will be called a pure super C ∗-algebra. An example is the
algebra A0 generated by a single selfadjoint element x with x2 = 0
of the example above. The details are left to the reader. There are
however also rather striking counterexamples. The simplest is given
by the Z2-graded Clifford algebra C1 on one odd generator. In this
case the cones C∗
n,+ make up all of the hermitian elements. The same
grading for a trivially graded C ∗-algebra B . If we are not to stick with
the case of pure super-C ∗-algebras, a property which in general is hard
to check we must find some other way of introducing an intrinsic order.
A key observation in this context is given by the proof of the following
proposition.
holds for the Z2-graded C ∗-algebra cM2(B) with standard off-diagonal
associate a graded C ∗-algebra M bA which is a nontrivial continuous
field of C ∗-algebras over the unit circle with each fibre ∗-isomorphic to
Proposition 2. To any (unital) Z2-graded C ∗-algebra bA one can
bA (the Moebius strip over bA ).
Proof. We will introduce a continuous family of new products on bA ,
x = x0 + x1 and y = y0 + y1 of bA be given where always x0 denotes
indexed by the points of the unit circle S1 = {ω ∈ C ω = 1}
and show that the resulting continuous field A when equipped with
a suitable matrix norm becomes a unital C ∗-algebra. Let elements
the even part of x and x1 the odd part. Define the ∗ω-product of x
and y by the formula
x∗ωy = (x0y0 + ωx1y1) + (x0y1 + x1y0) .
a Z2-graded C ∗-algebra it is clear that the ∗-multiplication extends to
matrix algebras in a compatible way. Consider the algebra which is
Associativity of the product is readily checked. Since Mn(bA) is also
linearly isomorphic with C(S1)⊗ bA but with product at the fibre over
on the tensor product C(S1) ⊗ bA which is denoted ∗ to distinguish it
from the ordinary (tensor) product. Let x = x(ω) be an element of this
algebra. In order to obtain a C ∗-algebra we will have to renormalize
the point ω twisted according to the above formula. It is clear that
this local definition of the product coherently defines a global product
13
the fibres according to the following scheme. For each ω the fibre
Aω which is the image of our continuous field under the evaluation
homomorphism corresponding to the point ω is isomorphic with bA by
the map
x0 + x1 7→ x0 + √ω x1
ιω : Aω −→ bA ,
where the sign of √ω is deliberate (but fixed in order that √ω is
continuous in the parameter ω ). One may than use the map ιω to
renormalize the fibres Aω such that each becomes isometric and ∗-
continuous field into a (graded) C ∗-algebra. Note however that there
is no global trivialization which identifies the continuous field with
(cid:3)
isomorphic with bA . The (fibrewise) involution and grading make this
C(S1) ⊗ bA
In particular consider the twisted product at the point ω = −1 to-
gether with the corresponding map
,
x0 + x1 7→ x0 + i x1 .
ι : (bA , ∗−1) −→ (bA , ∗1)
Let A be an abstract unital super-C ∗-algebra with hermitian subspace
Ah . The enveloping Z2-graded operator system of A with injective
envelope equal to I(A) (which is a graded C ∗-algebra) is then well de-
injective envelope. Upon utilizing the map ι (viewed as a linear map
fined, so is its enveloping (minimal Z2-graded) C ∗-algebra bA inside this
of Mn(bA) to itself) the hermitian elements of Mn(A) are identified
with a unital subspace of the ordinary selfadjoint elements of Mn(bA)
which is the selfadjoint part of an operator system. Pulling back the
order structure of the latter to Mn(A) defines a matrix order on A
with corresponding cones Cs
n,+ ⊆ Mn(A)h of superpositive elements.
If X is any unital subspace of A invariant under the superinvolution,
it inherits a matrix order by taking the intersection of the cones of
superpositive elements of Mn(A) with Mn(X) . At first sight the defi-
nition of superpositive elements seems to be a simple trick, exchanging
the subspaces of hermitian and ordinary selfadjoint elements in a Z2-
graded frame by the linear map ι . What is not so clear is how the
norms of an element x and its image ι(x) are related (precisely). One
easily derives that
1/2 · kxk ≤ kι(x)k ≤ 2 · kxk ,
which seems to be the best possible estimate in general. We wish to
have an abstract notion of super operator system which is independent
of any particular representation or embedding into a complete super
operator system. One possibility is to follow the scheme of abstract
14
U. HAAG
operator systems by specifying a matrix order on the set of hermitian
elements. In the case of an operator system the norm of an element is
then uniquely determined by this order if one looks for a completely
isometric and completely order isomorphic representation. The propo-
sition below states that with super operator systems things are more
complicated (the norm of an element is determined only within a global
margin). Also, in practice, the order is usually induced by a represen-
tation and is difficult to build in an abstract setting. We will therefore
choose a different approach.
Definition 2. An (abstract) super operator system (X , ∗ , 1) is an
abstract unital operator space (X , 1) (abstract operator space with
specified unit element that admits a completely isometric unital repre-
sentation on Hilbert space) equipped with a completely antiisometric
antilinear involution ∗ fixing the unit.
An (abstract) super operator space (V , ∗) is an abstract operator
space V with a completely antiisometric antilinear involution ∗ .
Given an operator space V with to different matrix norms {k·k1
n}n∈N
and {k·k2
n}n∈N , these will be called completely equivalent iff there exists
a global constants C1 , C2 > 0 with
n ≤ kxk2
n ≤ C2 · kxk1
for all n .
C1 · kxk1
n
Theorem.
Let X be an abstract super operator system. There exists a Z2-graded
(Representation theorem for super operator systems)
Hilbert space bH and a completely isometric unital ∗-representation
ϕ : X → B(bH) .
Let V be an abstract super operator space. Then there exists a Hilbert
space H and a completely isometric ∗-representation ν : V → B(H)
of V as a pre-operator system.
Proof. Let ψ : X → B(H) be a completely isometric unital repre-
sentation of the unital operator space X . Define the representation
ψ : X → B(H) by composing ψ with the ∗-involution of X on the
right and the adjoint operation on B(H) on the left. Then ψ is another
define a grading on the direct sum by exchanging the two copies of H .
It is easily verified that the direct sum representation ψ ⊕ ψ has the
required properties. Much in the same way one proves the existence of
a completely isometric ∗-representation ν on a graded Hilbert space
in case of a super operator space. Now given a completely isometric
completely isometric unital representation of X . Put bH = H⊕H and
15
∗-representation bν with respect to superinvolution on graded Hilbert
space bH let κ be a square root of the grading operator ǫ as above
and put ν = κ ·bν · κ . Then ν defines a completely isometric ∗-
on B(bH)
representation of V with respect to the ordinary (adjoint) involution
(cid:3)
One may now define the matrix order on X by pulling back the cones
of superpositive elements in the concrete super operator system ϕ(X) .
The Proposition below shows that this does not depend on the par-
ticular representation. Thus the structure of abstract super operator
system uniquely determines the matrix order on X . On the other
hand given the order there might exist different matrix norms compat-
ible with the matrix order. Any two such matrix norms will then have
to be completely equivalent from the following Proposition.
Proposition 3. Let X , Y be super operator systems. If ϕ : X →
Y is a unital and (completely) contractive ∗-linear map, then it is
(completely) superpositive. A completely superpositive map ψ : X →
Y is completely bounded with kψkcb ≤ 8 · kψ(1)k . If ψ is unital then
kψkcb ≤ 4 .
Proof. Let ϕ be a unital, contractive ∗-linear map. Then it extends
to a graded contractive map on the enveloping graded operator systems
ϕ is superpositive. Let ψ be completely superpositive. Then the
bϕ : bX → bY . One has ϕ(x) = ι ◦ bϕ ◦ ι−1(x) and since bϕ is positive,
map ι ◦ ψ ◦ ι−1 : ι(X) → ι(Y) (viewed as subspaces of bX , bY ) is
completely positive, so that its completely bounded norm is given by
kι ◦ ψ ◦ ι−1(1)k ≤ 2 · kψ(1)k . Then the map ψ must be completely
bounded with completely bounded norm less or equal to 8·kψ(1)k . If
ψ is unital then also ι ◦ ψ ◦ ι−1 whence the result
Given the matrix order on an abstract super operator system one may
consider the set of equivalence classes of all completely superpositive
unital representations as opposed to completely contractive. Any such
representation is completely bounded by 4 . Let max X denote the im-
age of X under the direct sum of all such representations. Then max X
is a super operator system with the same underlying vector space, unit
and involution but equipped with a larger though completely equiva-
lent matrix norm. It is clear from construction that max X has the
additional property that any unital and completely superpositive map
into another super operator system is completely contractive. A su-
per operator system with this property will be called maximal. If one
(cid:3)
16
U. HAAG
considers the folium of all super operator systems with underlying vec-
tor space X and same matrix order (but different matrix norms) it
contains a unique element which is an operator system (namely the
image of X in some completely isometric ∗-representation under the
map ι ). This will be denoted by X0 . Given a super operator sys-
tem X one may consider its enveloping Z2-graded C ∗-algebra in the
direct sum of all (up to equivalence) unital completely contractive ∗-
representations on graded Hilbert space. This object will be called the
maximal enveloping graded C ∗-algebra of X , denoted C ∗
max(X) . This
definition slightly differs from the previously defined maximal envelop-
ing graded C ∗-algebra of a super-C ∗-algebra A which only takes into
account (multiplicative) ∗-representations of the operator algebra. As
before the construction is functorial in the following sense: given any
completely contractive unital ∗-linear map of super operator systems
X → Y there is a canonically associated graded C ∗-homomorphism
C ∗
max(X) → C ∗
max(Y) . If V is a super operator space one defines sim-
ilarly the universal graded representation of V by taking the direct
sum representation over all (up to equivalence) completely contrac-
the enveloping Z2-graded pre-operator system of V in this represen-
tation. One defines C ∗
max(V) to be the graded C ∗-algebra generated
tive ∗-representations of V on graded Hilbert space. Let bV denote
by bV . Again this definition slightly differs from the case of a super
operator system, but if V happens to be completely ∗-isometric with
a super operator system X then one has a natural graded surjective
C ∗-homomorphism C ∗
max(X) and the construction is func-
torial in the usual sense. Since any super operator space can be repre-
sented as a pre-operator system on ungraded Hilbert space one may also
consider the universal representation as a pre-operator system which is
defined analogously. Let C ∗
max,0(V) denote the enveloping (ungraded)
C ∗-algebra of the image of V in this representation. Also this con-
max(V) ։ C ∗
struction is functorial. For any element x of a graded C ∗-algebra bA
define the graded absolute value of x to be the unique superpositive
square root xs of x∗∗ x making use of the ∗-multiplication (resp. the
map ι ) as in the proof of Proposition 2. The graded absolute value is a
special case of graded continous function calculus which can be applied
to any element of bA which is normal with respect to ∗-multiplication.
Example. To see that super operator spaces arise quite naturally con-
sider the dual operator space X ∗ of an operator system X . The invo-
lution on X determines an isometric involution on X ∗ = CB(X , C)
by the formula f (x) = f (x∗) (we use the notation f rather than
f ∗ to avoid mistaking it with the dual map f ∗ ∈ CB(C∗ , X ∗) ).
If
φ is an element of Mn(X ∗) = CB(X , Mn(C)) then its completely
bounded norm is given by the ordinary norm of the n-fold amplifica-
tion φn = idMn(C)⊗φ (compare [5], Prop. 8.11) from which it is easy to
induce that the involution as above is in fact completely antiisometric.
Thus the dual X ∗ is naturally a super operator space.
17
2. Strong matrix norms
Let V , W be super operator spaces.
If φ : V → W is a ∗-linear
completely bounded map we will say that it is hermitian n-contractive,
if the restriction of φn to the hermitian elements Mn(V)h is contrac-
tive. One also defines the hermitian n-norm of a ∗-linear completely
bounded map of super operator spaces by
kφkh
n = sup{kφn(x)k x ∈ Mn(V) , x = x∗ , kxk ≤ 1} .
n
n = kφ∗kh
As with the ordinary norms it follows that one has kφkh
for each n where φ∗ : W∗ → V∗ is the dualized map. Finally, let
Hilbert space with grading operator ǫ (in case of an ungraded Hilbert
space put ǫ = 1 ). For x ∈ V put
V ⊆ B(bH) be a concrete operator space represented on a Z2-graded
kxks = sup(cid:8)hxξ , ǫξi (cid:12)(cid:12) ξ ∈ bH , kξk ≤ 1(cid:9) .
The resulting norm on V will be called the strong norm (with re-
spect to the representation). One checks that if V is a super op-
erator space and the representation is an isometric ∗-representation,
then the involution is also isometric for the strong norm. Another
nice property of the strong norm is that it is invariant under conjuga-
tion by unitaries of degree zero, i.e.
kU x U ∗ks = kxks . More generally, one defines the strong n-norm on
Mn(V) to be its strong norm with respect to the natural identification
n}
satisfies axiom
if U ∈ B(bH)0 is a unitary then
Mn(V) ⊆ Mn(B(bH)) ≃ B(bHn) . The sequence of strong norms {k·ks
kx ⊕ ykn+m = max{kxkn , kykm}
(M1)
of a matrix norm, but instead of
(M2)
for α ∈ Mn,r(C) , β ∈ Mr,n(C) one only has the weaker property
(ΣM2)
kα · x · βkn ≤ kαk kxkr kβk ,
kα · x · αks
n ≤ kαk2 · kxks
r
18
U. HAAG
where α ∈ Mn,r(C) and x ∈ Mr(V ) . A sequence of norms satisfying
(ΣM1) = (M1) and (ΣM2) will be called a strong matrix norm.
Given a Banach space V equipped with a strong matrix norm {k·ks
n}
for the sequence of spaces {Mn(V)} putting
so that
defines an (ordinary) matrix norm on V . Namely, suppose that x =
α · y · β with y ∈ Mr(V) , α ∈ Mn,r(C) , β ∈ Mr,n(C) and without
loss of generality that kαk ≥ kβk . Put eβ = kαk · kβk−1 · β and
ey = kβk · kαk−1 · y . Then
s
2n
kxkn := 2 (cid:13)(cid:13)(cid:13)(cid:18)0 x
0 0(cid:19)(cid:13)(cid:13)(cid:13)
0 eβ! ·(cid:18)0 ey
=(cid:13)(cid:13)(cid:13)(cid:18)0 x
x 0(cid:19)(cid:13)(cid:13)(cid:13)
0 0(cid:19) = α 0
0 eβ(cid:19)
(cid:18)0 x
0 0(cid:19) ·(cid:18)α 0
kxkn ≤ kαk · keykr · keβk = kαk · kykr · kβk .
2n ≤ 2 (cid:13)(cid:13)(cid:13)(cid:18)0 x
0 0(cid:19)(cid:13)(cid:13)(cid:13)
0 −x(cid:19)(cid:13)(cid:13)(cid:13)
2n
0
s
s
Property (M1) is even more trivial to check. Note that in any case
the following relation holds
n =(cid:13)(cid:13)(cid:13)(cid:18)x
s
2n
kxks
= kxkn .
It is also clear that in case of a concete operator space the matrix norm
thus obtained from the strong matrix norm coincides with the matrix
norm of the representation whereas there may exist representations of
the same underlying matrix norm giving rise to different strong norms.
Thus a strong matrix norm imposes a finer structure on an operator
space than an ordinary matrix norm. This leaves unanswered the ques-
tion which abstract strong matrix norms can actually be represented.
In case of a super operator space V and a ∗-representation on (graded)
Hilbert space one will certainly need at least one more axiom besides
demanding that the involution is strongly completely antiisometric. In
particular for each hermitian element x ∈ Mn(V)h one must have
(hΣM∗)
since for a selfadjoint (diagonalizable) element the strong norm coin-
cides with the usual norm. However since this property is valid only
for hermitian elements it does not seem sufficient to guarantee the
existence of a strongly completely isometric ∗-representation. A linear
map φ : V → W of concrete operator spaces will be called (completely)
strongly contractive if it is (completely) contractive with respect to the
strong norms. It will be called really strongly contractive if for each unit
n = 2 (cid:13)(cid:13)(cid:13)(cid:18)0 x
0 0(cid:19)(cid:13)(cid:13)(cid:13)
x ∈ Mn(V)h
kxks
2n
,
s
19
vector η in the representation space of W and each x ∈ V there ex-
ists a unit vector ξ in the representation space of V such that putting
α = hx ξ , ǫ ξi , β = hφ(x) η , ǫ ηi one has the relations
Re(α) ≥ Re(β) .
α ≥ β
The first condition just states that φ is strongly contractive. As for
the second it looks a bit awkward and difficult to check in a general
situation but we will see some instances in finite dimensional matrix
algebras where it can be sucessfully applied. If with notation as above
one also has that the sign of Re(α) is the same as the sign of Re(β)
the map is called signed really strongly contractive.
,
Let φ : V → W be a ∗-linear map of super oper-
Strong Lemma.
ator spaces and T ⊆ V a closed subspace with T + T ∗ = V . If there
exist isometric ∗-representations of V and W such that φ restricted
to T is really strongly contractive, then φ is strongly contractive (and
hence hermitian contractive).
Proof.
If V ⊆ B(bH) consider the strongly isometric cross-diagonal
embedding λ : V → M2(B(bH)) by the assignment
x 7→ (cid:18)0 x
x 0(cid:19) .
(cid:18)x
0 −x(cid:19) .
0
That it is strongly isometric follows from the fact that the image of x
can be rotated to
s
adjoining the grading operator (still denoted ǫ for simplicity) to λ(V) .
Now consider the operator subsystem eV of M2(B(bH)) obtained by
The same procedure applies to W . Also put eT = λ(T ) + C · ǫ . We
claim that the natural ǫ-unital extension eφ of φ is strongly contractive
when restricted to eT . If c ∈ C is any complex number, then
c · ǫ(cid:19)(cid:13)(cid:13)(cid:13)
(cid:13)(cid:13)(cid:13)(cid:18) c · ǫ φ(x)
= max(cid:8)kφ(x) + c · ǫks , kφ(x) − c · ǫks(cid:9) .
of fW such that the maximum of the strong norms of φ(x) ± γ ǫ
attained at η . Then there exists a unit vector ξ in bH such that
Let c = ωc · γ with γ ≥ 0 and ωc = 1 . Dividing by ωc will not
change anything about the strong norms so we may assume that c = γ
is a positive number. Let η be a unit vector in the representation space
is
hx ξ , ǫ ξi ≥ hφ(x) η , ǫ ηi
φ(x)
20
and
U. HAAG
Re(hx ξ , ǫ ξi) ≥ Re(hφ(x) η , ǫ ηi .
Now if α and β are complex numbers with α ≥ β and Re(α) ≥
Im(α) then also
max{α + γ , α − γ} ≥ max{β + γ , β − γ}
for any positive number γ . This proves the claim. Let then η be an
the functional
ρ(x) = hλ(φ(x)) η , ǫ ηi
Banach-Theorem it extends to an ǫ-unital and contractive functional
tional is hermitian contractive with respect to the ordinary norm. It is
also contractive on the larger real subspace generated by the hermitian
arbitrary unit vector in the representation space of fW and consider
on eT which is contractive for the strong norm. Then from the Hahn-
eρ on eV with respect to the strong norm which means that the func-
elements of eV and complex multiples of ǫ . To see this one utilizes the
with κ a square root of the grading operator on B(eH) to reduce to the
situation of linear combinations of selfadjoint elements in a pre-operator
system and complex multiples of the identity. Any such element is di-
agonalizable so that the strong norm coincides with its ordinary norm.
Now apply Lemma 2 above (resp. Lemma A.4.2 of [3]) to conclude
x 7→ κ · x · κ ,
strong isometry
ǫ 7→ 1
that
that eρ is selfadjoint and hermitian contractive, by which one deduces
holds for all x ∈ eV . This means upon taking the supremum over all
eρ(x) = heφ(x)η , ηi
unit vectors η that φ is hermitian contractive as stated
(cid:3)
Example. Consider a onedimensional projection in Mm(C) , say with
zero entries in all matrix entries except a 1 in the lower right corner
p =
...
. . .
0 · · · 0 0
...
...
0 · · · 0 0
0 · · · 0 1
.
and the projection ϕ with kernel generated by p and range the com-
plementary subspace Q of codimension one whose elements have zero
entry in the lower right corner
21
Q =
...
. . .
∗ · · · ∗ ∗
...
...
∗ · · · ∗ ∗
∗ · · · ∗ 0
.
Let T be the subspace of lower triangular matrices in Mm(C) . Then
the condition of the Sublemma above is easily seen to hold in this
situation, so that ϕ is hermitian contractive for any m . As a second
example consider the onedimensional subspace of Mm(C) spanned by
the element
e =
F =
...
. . .
0 · · · 0 0
0
...
...
...
0 · · · 0 0
0
0 · · · 0 1
0
0 · · · 0 0 −1
...
. . .
∗
∗ · · · ∗ ∗
...
...
...
∗ · · · ∗ ∗
∗
∗ · · · ∗ α β
∗ · · · ∗ γ α
.
and the projection ϕ with kernel generated by e and range the sub-
space of codimension one whose entries look like
Again one checks the condition of the Sublemma for the lower trian-
gular matrices with respect to this projection and concludes that it is
hermitian contractive. Similar examples of this kind can be found.
Besides the strong matrix norm as above there is another strong norm
for the C ∗-algebra of bounded operators on graded Hilbert space which
is of interest and will be called the strong σ-norm.
It is defined as
follows. Let x = x0 + x1 be an operator on a graded Hilbert space bH
with grading operator ǫ such that x0 is homogenous of degree zero
and x1 is of degree one. Define
kxkσ := sup{hx0ξ , ξi + hx1ξ , ǫξi (cid:12)(cid:12) ξ ∈ bH , kξk ≤ 1} .
Similarly one defines the strong σ-n-norm to be the σ-strong norm of
B(bHn) ≃ Mn(B(bH)) . Then it is clear how to define the strong matrix
σ-norm for a concete (super) operator space V represented on bH and
one checks that the involution is completely antiisometric also with
respect to the strong σ-norms. The strong matrix σ-norm still satisfies
22
U. HAAG
axioms (ΣM1) and (ΣM2) , but in this case the matrix norm given
by
kxkσ,n := 2(cid:13)(cid:13)(cid:13)(cid:18)0 x
0 0(cid:19)(cid:13)(cid:13)(cid:13)
σ
2n
need not coincide with the ordinary operator matrix norm of x ∈
Mn(V) . We may call it the matrix σ-norm. While the strong norm is
connected with the concept of ǫ-positivity, the strong σ-norm is related
hxξ , ǫξi ≥ 0
ǫ-positive if and only if
to superpositivity. Namely, one checks that an element x ∈ B(bH) is
for all ξ ∈ bH . Similarly, x = x0 + x1 is superpositive if and only if
for all ξ . To see this define the isometry κ : bH → bH by κ(ξ0 ⊕ ξ1) =
ξ0⊕iξ1 . Now x = x0 +x1 is superpositive iff ι(x) = x0 +ix1 is positive
in the ordinary sense and one has the identity
hx0ξ , ξi + hx1ξ , ǫξi ≥ 0
hι(x)ξ , ξi = hx0 κξ , κξi + hx1 κξ , ǫ κξi .
It is interesting to note that in case of a super operator system the ma-
trix σ-norm, which is the ordinary matrix norm of ι(X) , is determined
by its operator matrix norm (in particular it doesn't depend on the
representation). One checks that the proof of the Strong Lemma ap-
plies just the same if instead of ordinary (strong) matrix norms (strong)
matrix σ-norms are considered, so one only needs to replace the terms
"really strongly contractive" and "hermitian contractive" of the Lemma
by the corresponding terms "really strongly σ-contractive" and "her-
mitian σ-contractive" respectively.
Suppose given a super operator space V equipped with a strong matrix
norm {k · ks
n}n≥1 satisfying (ΣM1) and (ΣM2) and such that the
involution is strongly completely antiisometric. Such a super operator
space will be called an abstract h-strong super operator space if in ad-
dition it satisfies axiom (hΣM∗) . One may define two other abstract
strong matrix norms {rn , Rn}n≥1 from the given one by considering
the maximum of the real matrix seminorms
pω
n(x) =
i.e rn(x) = supω{pω
norms
1
2 kωx + ωx∗ks
n
, ω ∈ C , ω = 1
n(x) } and the maximum of the real matrix semi-
P ω
n (x) =
1
2 kωx + ωx∗kn
, ω ∈ C, ω = 1
23
i.e. Rn(x) = supω{P ω
∗-representable one gets kωx + ωx∗kn = kωx + ωx∗ks
n (x)} .
rn(x) = Rn(x) ≤ kxks
n .
If the original strong matrix norm is
n , and hence
The reverse inequality can also be derived in this case. For given x
choose a complex number ω of modulus one such that rn(x) = 1/2 ·
kωx + ωx∗ks
n . Since the strong matrix norm is representable there
exists a ∗-linear strongly contractive functional ϕ : Mn(V) → C with
ϕ(x) = kxks
n since in the induced representation of Mn(V) the strong
norm of x is given as the supremum over the absolute values of vector
states
ϕξ(x) = h xξ , ξ i .
Assume that the inequality above is strict, so there exists a real number
1/2 < r ≤ 1 with kxks
n . Identifying the range of ϕ
with the onedimensional subspace of Mn(V) generated by the element
ωx+ωx∗ assume that ϕ(x) = γr(ωx+ωx∗) with γ a complex number
of modulus one. Then
n = rkωx + ωx∗ks
ϕ(1/2(γx + γx∗)) = r(ωx + ωx∗)
contradicting the assumption that ϕ is strongly contractive. Since for
any super operator space its ordinary matrix norm is ∗-representable
from the representation theorem above this implies that a strong matrix
norm on a super operator space is representable if and only if it satisfies
the axiom
(ΣM∗)
A super operator space V equipped with a strong matrix norm satis-
fying (ΣM1) , (ΣM2) and (ΣM∗) will be called an abstract strong
super operator space. Note that the axiom also entails that the involu-
tion is completely antiisometric. The following theorem is a Corollary
of the preceding discussion
rn(x) = Rn(x) = kxks
n .
Theorem. (Representation theorem for strong super operator spaces)
Let V be an abstract strong super operator space. Then there exists
an (ungraded) Hilbert space H and a strongly completely isometric
∗-representation
µ : V −→ B(H) (cid:3)
Theorem. (Representation of h-strong super operator spaces) Let V
be an abstract h-strong super operator space. Then there exists for
24
U. HAAG
each n an (ungraded) Hilbert space Hn and a strongly contractive
and (strongly) hermitian isometric ∗-representation
ν : Mn(V) −→ B(Hn) .
Proof. For the given h-strong super operator space replace the strong
matrix norm by the abstract strong matrix norm rn(x) = Rn(x) . It is
easy to check that rn(x) ≤ kxks
n and that the induced matrix norms
are the same, since rn(x) = Rn(x) and
kxkRn = 2 R2n(cid:0)(cid:18)0 x
0 0(cid:19)(cid:1) = sup
ω (cid:8)(cid:13)(cid:13)(cid:13)(cid:18) 0
ωx∗
ωx
0(cid:19)(cid:13)(cid:13)(cid:13)2n(cid:9) = kxkn .
Since {rn} is representable from the previous theorem this then also
implies that rn(x) = kxks
n = kxkn in case that x = x∗ is hermitian,
since
kxkn = kxkrn = rn(x) ≤ kxks
n
and the reverse inequality is trivial. Suppose given x ∈ Mn(V) that
the maximum of the set {pω
n(x)} is taken at ω = 1 . From the Hahn-
Banach Theorem one chooses a contractive projection ϕ of Mn(V)
onto the onedimensional subspace generated by x + x∗ . Since the
subspace is selfadjoint one can assume the projection to be ∗-linear,
replacing it with 1/2(ϕ + ϕ) if necessary. Let ϕ(x) = α(x + x∗) with
α ∈ C . Since ϕ is hermitian one gets α + α = 1 .
If α is real
then it must be 1/2 and one puts f 1
If on the other
hand α is not a real number it follows that the maximum of the set
{pω
n(x)} cannot be taken at ω = 1 . One similarly constructs a ∗-linear
function f ω
x (x) =
x
n(x) provided that the function σ 7→ pσ
pω
n(x) attaines its maximum
at the point ω . Now take the direct sum over all onedimensional ∗-
representations {f ω
for each ω such that f ω
x }x∈Mn(V) to get the result
x (y) ≤ rn(y) and f ω
x (y) = ϕ(y) .
(cid:3)
3. Operator tensor products
We now turn to the subject of tensor products of operator spaces resp.
super operator systems. Since the operator spaces we are considering
carry some extra structure we would like this structure to be reflected
in the tensor product. For example, iff V and W are unital oper-
ator spaces it is natural to require that the tensor product V⊗γW
obtained as the completion of the algebraic tensor product by a certain
cross matrix norm γ also be unital with unit 1V ⊗ 1W . We make the
following convention: given an abstract operator space V together with
a specified element ∗V ∈ V of norm one (= pointed operator space),
25
one defines the abstract unital operator space or unification of (V , ∗V)
to be completely isometric with the image V of V under the direct
sum of all completely contractive unital representations of (V , ∗V) on
some Hilbert space up to equivalence. This space may or may not be
completely isometric with V . Since there always exists at least one
completely contractive unital map into C by the Hahn-Banach theo-
rem one sees that V is not empty and is a unital operator space. If
V happens to be a super operator space with specified element ∗V
of norm one then the pair (V , ∗V) is said to be a pointed super op-
erator space if the involution fixes the element ∗V . It is easy to see
(compare with the proof of Lemma 4 below) that the unification of a
pointed super operator space is a super operator system. Given two
unital operator spaces (V , 1V) and (W, 1W ) define the unital projec-
tive tensor product Vb⊗uW to be the abstract unital operator space
which is the unification of (Vb⊗W , 1V ⊗ 1W ) where Vb⊗W denotes
the usual projective operator tensor product. The Haagerup tensor
product V⊗hW and the injective tensor product V ⊗W of two unital
operator spaces are known to be unital in any case but in case of the
projective tensor product the author doesn't know if this is true. In
any case it is easy to see that the unital projective tensor product is
the maximal unital tensor product and surjects onto any other such
tensor product in a canonical way. Let (V , ∗V) , (W , ∗W ) be pointed
operator spaces. Then a completely bounded map ϕ : V → W is
said to be pointed iff ϕ(∗V) = ∗W . A unital operator space V is nat-
urally a pointed operator space with ∗V = 1V . The mapping space
of completely bounded linear maps of pointed operator spaces con-
tains the specified subset of completely contractive pointed maps de-
noted CCP(V , W) . Let V , W and X be unital operator spaces.
CCP (V × W , X ) is the set of pointed completely contractive bilin-
ear maps ϕ from V × W (with specified element 1V × 1W ) into X ,
i.e.
the maps are required to send 1V × 1W to 1X , and the c.b.
norm is defined as usual in the mapping space of all bilinear maps
by kϕkcb = sup{kϕp×q;r×sk p, q, r, s ∈ N} = sup{kϕp×p;p×p p ∈ N},
ϕp,q;r,s : Mp,q(V) × Mr,s(W) → Mp×r;q×s(V × W) ,
ϕp,q;r,s([vi,j] , [wk,l]) = [ϕ(vi,j, wk,l)]
and the matrix norm is obtained using the identification
Mn(CCP(V × W , X )) ⊆ Mn(CB (V × W , X )) = CB (V × W , Mn(X))
(compare [Effros-Ruan], chap. 7).
26
U. HAAG
Lemma 3.
natural completely isometric identifications
If V , W and X are unital operator spaces there are
CCP (Vb⊗uW , X ) ≃ CCP (V × W , X ) ≃ CCP (V , CB (W , X )) .
Proof Note that CCP (Vb⊗W , X ) ≃ CCP (Vb⊗u W , X ) with respect
to the pointed projective tensor product (Vb⊗W , 1V ⊗ 1W ) and com-
pare with the proof of [3], Prop. 7.1.2
(cid:3)
The completely bounded maps of the unital projective tensor product
are not that easily described, but the Lemma is sufficient for our pur-
poses. Assume now that V is a unital operator space represented on
some Hilbert space and let V be the adjoint operator space. The map-
ping v 7→ v is completely antiisometric in the sense of Definition 1.
Note that in the case where V is represented on a Z2-graded Hilbert
space its ordinary adjoint is of course completely isometric with its
superadjoint so there is nothing to be gained from the more general
setting, but in turn one finds that the first case of the Lemma below
also applies when V is a super operator space whence V is completely
isometric with V . Also let X , Y be two super operator systems with
corresponding units and superinvolutions.
Lemma 4. The unital projective tensor product Vb⊗uV , the Haagerup
tensor product V⊗hV , and the injective tensor product V ⊗V of a uni-
tal operator space with its adjoint space admit the structure of a super
operator system when equipped with the completely antiisometric in-
volution given by antilinear extension of v⊗w 7→ w⊗v . The projective
operator space with the corresponding involution. The unital projec-
tensor product (Vb⊗V , 1Vb⊗1V) admits the structure of a pointed super
tive tensor product (projective tensor product) Xb⊗uY ( Xb⊗Y ) and
the injective tensor product X ⊗Y of two super operator systems ad-
mit the structure of a super operator system (pointed super operator
space) equipped with the completely antiisometric involution given by
antilinear extension of x ⊗ y 7→ x∗ ⊗ y∗ .
Proof. Considering the projective tensor product Vb⊗V , the assign-
w)β ∈ Mn(Vb⊗V) with v ∈ Mp(V) , w ∈ Mq(V) , α ∈ Mn,p×q(C) , β ∈
ment v ⊗ w 7→ w ⊗ v extends to a completely antilinear and anti-
if x = α(v ⊗
isometric involution on the tensor product.
Mp×q,n(C) , then x∗ = β(w ⊗ v)α and kx∗k ≤ kxk from the definition
of the projective norm. By symmetry one gets kx∗k = kxk . So the
projective tensor product is naturally a pointed super operator space.
Indeed,
27
Now the unification of a pointed super operator space (V , ∗V) is a su-
per operator system. Let ρ : V → B(H) be any unital representation,
dropping to a representation of the unification. Then the composition
of the involution on V with the adjoint map of the algebra of bounded
operators on Hilbert space defines another completely contractive uni-
tal representation which therefore drops to the unification. Since this
is true for all such representations one finds that the given involution
drops to a completely antiisometric unital involution on the unification
of V which is therefore a super operator system.
In the same way
one verifies that the second involution defined on the projective tensor
product Xb⊗Y is completely antiisometric and drops to a completely
antiisometric antilinear involution on the unital projective tensor prod-
uct. The proof that the first of the outlined involutions is completely
antiisometric for the Haagerup tensor product is very similar. It can
also be derived using injectivity of the Haagerup tensor product on
embedding the unital operator space V into an enveloping C ∗-algebra
A and using the description of A⊗hA as a subspace of the free prod-
uct of two copies Ai , i = 1, 2 of A amalgamated over the identity by
linear extension of the assignment a ⊗ b 7→ a1 ∗ b2 due to Christensen-
Effros-Pisier-Sinclair (see [5], Theorem 17.6.). Then the involution of
first type corresponds to composition of the C ∗-involution on A1 ∗ A2
with the order two automorphism exchanging the two copies of A , the
first being completely antiisometric, and the second a completely iso-
metric map. In case of the (spatial) injective tensor product the proof
is obvious
(cid:3)
Besides the unital projective tensor product we wish to define another
tensor product which is closely related with the Haagerup tensor norm.
Unlike the former the Haagerup tensor product is not commutative so
that the product type involution on X ⊗ Y is not completely antiiso-
metric for the Haagerup tensor norm {k·kh,n}n∈N . Let x 7→ x∗ denote
the product type involution on the algebraic tensor product (extended
to matrices by composing it with the transpose mapping). Then one
defines a symmetrized Haagerup tensor norm putting
kxkh∗,n := max {kxkh,n , kx∗kh,n} .
It is easy to verify the axioms of a matrix norm. The correspond-
ing completion of the algebraic tensor product is denoted X⊗h∗Y .
One obtains a completely ∗-isometric unital representation of the re-
sulting pointed super operator space by taking the direct sum of a
completely isometric unital representation for the (unsymmetrized)
Haagerup norm and the completely contractive representation obtained
28
U. HAAG
by composing the first representation with the involution on the left and
the adjoint operation on the right with grading operator ǫ exchang-
ing the two copies of the same underlying Hilbert space, showing that
the symmetrized Haagerup tensor product is a super operator system.
Although the symmetrized Haagerup tensor product is commutative,
by the process of symmetrization associativity of the Haagerup ten-
(V ⊗h∗ W) ⊗h∗ X is in general different from
sor product is lost, i.e.
V ⊗h∗ (W⊗h∗ X) .
In the progress of our investigation of operator tensor products we
shall need a version of the Christensen-Effros-Sinclair-Theorem suited
for our purposes. Let A be a C ∗-algebra and V a super operator
space which is also an (left) A-module, which is to say that there exists
a completely bounded left action λ : A × V → V meaning that the
natural extensions idMn ⊗ λ : A× Mn(V) → Mn(V) induced by λ are
uniformly bounded for all n ∈ N . We only consider actions which are
nondegenerate in the following sense: given an approximate unit {uλ}λ
for A , the net {uλ · v}λ converges to v for any v ∈ V . It follows
from the nondegeneracy condition that the action extends naturally to
a completely bounded action of the C ∗-unitization eA by setting
(a, λ1) · v = a · v + λv
using associativity of the action. Therefore in the following we may
assume A to be unital without loss of generality. Putting v · a :=
(a∗·v∗)∗ defines a right action of A on V . We assume that the adjoint
right action of A is compatible with the left action in the sense that
a·(v·a′) = (a·x)·a′ (if V is a pointed super operator space resp. super
operator system we also require that a · ∗V = ∗V · a ). It then follows
directly from the definition of the adjoint actions that they are again
completely bounded. In our applications we shall usually need stronger
requirements -- that the action is completely contractive in a certain
sense. A condition which looks a bit farfetched at first but proves useful
in a technical surrounding is the following: given a completely bounded
action of A on V as above the action is said to be u-weakly completely
contractive if for every unitary u ∈ A the assignment
v 7→ u · v
is completely contractive (= completely isometric).
It is said to be
weakly completely contractive if for every a ∈ A with kak ≤ 1 the
assignment
v 7→ a · v
is completely contractive. The significance of the former notion is most
easily recognized in its proper context. Namely let U denote a topolog-
ical group (for example the unitary group of a C ∗-algebra). A (pointed)
super operator space V is said to be a weak U-bimodule, iff there is
defined a left action
29
U × V −→ V ,
(u , v) 7→ u · v
which is continuous, linear and completely isometric in V , and asso-
ciative, i.e. u1 · (u2 · v) = (u1u2) · v , and such that putting
v · u = (u−1 · v∗)∗
defines a right action of U on V commuting with the left action (and
if V is pointed u · ∗V = ∗V · u for any u ∈ U ). The actions extend
via matrix multiplication to completely bounded left and right actions
of matrix algebras Mn(A) (resp. U n ) on Mn(V) for each n . If the
so defined extended actions are contractive (i.e. if the natural induced
maps into CB(Mn(V) , Mn(V)) is contractive for each n ) the actions
will be called strongly completely contractive and V will be called a
strong A-bimodule (resp. strong U-bimodule), otherwise we say that
V is a (weak resp. u-weak) A-bimodule. For any super operator system
X2 (or super operator space) a matrix decomposition is synonymous for
a strongly completely contractive action of C ⊕ C . It will be called an
F -matrix decomposition if the strong action of C⊕C viewed as diagonal
of M2(C) with orthogonal projections e and 1 − e corresponding to
the left upper resp. lower right corner extends to an action of M2(C)
such that
F = (cid:18)0 1
1 0(cid:19)
X2 → M2(X)
acts completely isometric, and such that the element F = F · 1 of the
injective envelope I(X2) is a selfadjoint (even) unitary, and such that
putting X = (1 − e) · X2 · (1 − e) the natural map
is completely contractive. To make things just a little more complicated
we will also have to consider a notion which is slightly weaker than an
F -matrix decomposition. A weak F -matrix decomposition of X2 means
a matrix decomposition which extends to a completely bounded action
of M2(C) such that F as above acts completely isometric restricted
to any corner of the decomposition, and the element F = F · 1 is an
even selfadjoint unitary of I(X2) .
The following definition of an admissible action is designed to accom-
modate the main applications that we have in mind, in order to get the
30
U. HAAG
strongest possible results. Some of the conditions are of a seemingly
very special nature, and it may be that in a different situation an-
other perhaps more simple-minded set of assumptions is sufficient. For
example, if instead of a (weak) F -matrix decomposition one assumes
the structure of a strong M2(C)-bimodule the proof of the following
SCES-Theorem will be much shorter and more direct but this does
not fit with the situation encountered in our applications. If X is a
super operator system a bimodule action of A (resp. U ) on X will
be called admissible if it is u-weakly completely contractive and if the
following conditions hold:
for every unitary u ∈ A (resp. u ∈ U ),
putting U = u⊕ 1 ∈ A⊕ A , there exists a super operator system X2,U
and a unital complete ∗-contraction pU : X2,U ։ M2(X) with dense
image (it is in fact sufficient that pU is completely contractive when
restricted to any enlarged corner ∆ij
2,U as defined below) such that the
natural F -matrix decomposition of M2(X) lifts to a weak F -matrix
decompositon on X2,U , and there exists a subsystem
Y2,U = (cid:18) Y F · X
X (cid:19) , Y ⊆ F · X · F
X · F
with X = (1−e)X2,U (1−e) , such that the adjoint action of the unitary
U (with A ⊕ A viewed as subalgebra of diagonal elements in M2(A) )
on pU (Y2,U ) given by x 7→ adU (x) = U · x · U ∗ lifts to an action on
Y2,U which is compatible with the matrix decomposition and trivial on
the lower right corner X . Putting FU n = adU n(F ) it is also assumed
that the elements in the set { F · FU n n ∈ Z} are evenly graded
and adU -invariant and that F · FU n = F ◦ FU n (where ◦ denotes the
C ∗-product operation in the injective envelope). Let ∆ij
2,U denote the
corner determined by the i − th row and j − th column of the matrix
decomposition but with elements {FU n , F · FU n} adjoined, so that
F · ∆ij
2,U = ∆(i+1)j
2,U
, ∆ij
2,U · F = ∆i(j+1)
2,U + ∆22
2,U
with superscripts mod 2. Also put ∆2,U = ∆11
2,U and ∇2,U =
F · ∆2,U = ∆2,U · F . The action of F is still supposed to be completely
isometric intertwining the larger subsystems ∆2,U and ∇2,U (which is
consistent with our examples although probably some weaker hypothe-
ses will also do). The map adU is supposed to be completely isometric
when restricted to the intersection of Y2,U with either of these sub-
systems (but not necessarily on the whole space). It follows from the
condition F · FU n = F ◦ FU n etc. that the natural map
µ : X2,U ։ M2(X)
31
by identification of the corners via the F -action is completely isometric
when restricted to any of the subspaces ∆ij
2,U . The injective envelope
of Y2,U ∩∆2,U (or more precisely the subspace of the injective envelope
generated linearly by C ∗-products from elements of Y2,U ∩ ∆2,U with
the selfadjoint unitaries {FU n} from the left and right side) should
contain ∆2,U by some completely isometric ∗-linear extension of the
inclusion Y2,U ∩ ∆2,U ⊆ I(Y2,U ∩ ∆2,U ) . In addition there should be
two unital complete contractions
νl : X2,U ։ V 2,U,l ,
νr : X2,U ։ V 2,U,r
onto unital operator spaces with compatible (left and right) weak F -
matrix decomposition such that in the first case the left action of F
is given by C ∗-multiplication with F from the left in the injective
envelope (hence is a strong action), and in the second case the right
action of F is given by C ∗-multiplication from the right with F (in the
remaining cases one again assumes that the action of F is completely
isometric restricted to the images of ∆2,U and ∇2,U , such that if x ∈
Mn(X2,U ) is contained in a subspace ∆ij
2,U one has
kxkn = max{kνl(x)kn , kνr(x)kn} .
The map adU is supposed to drop to a completely isometric map on the
images of the intersections of Y2,U with the (enlarged) diagonal/off-
diagonal subspaces in both V 2,U,l and V 2,U,r (in particular the images
of the elements {FU n} still define selfadjoint unitaries in the injective
envelopes of both spaces which carry a unique structure of a unital
C ∗-algebra).
The action of A (or U ) on X will be called strongly admissible if it
is u-weakly completely contractive and if there exists a super operator
system X which is a strong A- (or U-)bimodule and a unital ∗-linear
complete contraction p : X ։ X with dense image commuting with the
corresponding bimodule structures. In particular a strong A-bimodule
X is strongly admissible taking X = X .
Example. Let V and W be super operator spaces with V a weak
A-module and consider the projective tensor product P = Vb⊗W .
From the proof of Lemma 4 P can be given the structure of a super
operator space with product involution of the involutions of V and
W respectively. Consider the natural left action of A on the algebraic
tensor product V ⊗ W equipped with the projective norm. We wish
to show that this action is weakly completely contractive so that it
extends to a weakly completely contractive action on the projective
32
U. HAAG
tensor product. Let a ∈ A and x ∈ Mn(V⊗ W) be given and suppose
that x has a product decomposition of the form α · (v ⊗ w) · β with
α ∈ Mn,p×q(C) , β ∈ Mp×q,n(C) and v ∈ Mp(V) , w ∈ Mq(W) . Then
the element a· x has a decomposition α· (((1p ⊗ a)· v)⊗ w)· β so that
the projective norm of a · x is smaller than kαk · k(1p ⊗ a) · vk · kwk ·
kβk ≤ kak · kαk · kvk · kwk · kβk and taking the infimum over all such
decompositions one gets ka · xk ≤ kak · kxk as desired. By taking the
adjoint right action one checks the compatibility conditions so that P
is a weak A-bimodule. If W also happens to be a weak B-bimodule,
then P is a weak A- and weak B-bimodule, and each of the four (left
and right) actions is compatible with each of the others.
Lemma 5. Suppose that the super operator space V is a u-weak A-
bimodule. Then the adjoint action of the unitary group of A extends
to a completely isometric and grading preserving action on its universal
enveloping graded pre-operator system bV .
Proof. The statement of the Lemma should be clear from the fact that
given any completely contractive ∗-linear representation ρ of V the
composition ρu = ρ ◦ adu for u ∈ A unitary defines another ∗-linear
representation
(cid:3)
(Super-Christensen-Effros-Sinclair-Theorem) Let A be
Theorem.
a C ∗-algebra (resp. U a topological group) and X a super operator
system which is a strongly admissible A-(resp. U-)bimodule. Then the
bimodule action extends to a graded action on its enveloping graded op-
X and X0 are strong A-(resp. U-)bimodules and there exists a graded
erator system bX which is strongly completely contractive. In particular
Hilbert space bH , a unital graded completely isometric ∗-representation
π : bX → B(bH) and a C ∗-representation λ : A → B(H)ev (resp. uni-
tary representation µ : U → B(H)ev ) into the subalgebra of elements
of even degree such that
π(a · x · a′) = λ(a)π(x)λ(a′)
or
π(u · x · u′) = µ(u)π(x)µ(u′)
respectively.
If X is an operator system which is an admissible A-bimodule (U-
bimodule), then X is a strong A-(resp. U-)bimodule (operator A-
system) and there exist compatible representations π of X and λ of
A (resp. µ of U ) on an (ungraded) Hilbert space as above.
33
Proof. We only write out the proof in case of the action by a C ∗-
algebra A since the argument in case of a toplological group U is
completely analogous. We begin with the second part so let X be an
operator system which is an admissible A-bimodule. Let u ∈ A be a
unitary and put
U = (cid:18)u 0
0 1(cid:19) ∈ M2(A) .
Then there exists a super operator system X2,U and a unital complete
∗-contraction with dense image pU : X2,U −→ M2(X) such that the
adjoint action of U on M2(X) lifts to a unital complete ∗-isometry
on Y2,U ⊆ X2,U (or its intersection with the subsystems ∆2,U , ∇2,U )
and admitting a compatible weak F -matrix decomposition. Any com-
pletely positive extension φ : I(X2,U ) → M2(I(X)) of pU to the injec-
tive envelopes will be a C ⊕ C-module map, since the strong action of
C ⊕ C extends uniquely by rigidity to a strong action on the injective
envelopes where it is implemented by C ∗-multiplication with elements
from the corresponding subalgebras which are identified by the com-
pletely positive map φ , so that in any case φ is compatible with the
matrix decomposition. By the same argument any extension φ is nec-
essarily a C ∗(FU n)-module map for each n ∈ Z . We wish to show that
adU acts completely positive on M2(X) . Since the adjoint action of U
on Y2,U is unital and completely isometric it follows that adU extends
uniquely to a graded C ∗-automorphism on I(X2,U ) = I(Y2,U ) as well
as I(∆2,U ) and I(∇2,U ) by rigidity and Proposition 1. On the other
hand the left action of u is completely isometric on X so the map
adU is unital and completely isometric restricted to the subsystem
SX = (cid:18)C1 X
X C1(cid:19) ⊆ M2(X) .
Then it extends to a unique C ∗-automorphism on the injective envelope
I(SX) which admits a matrix decomposition of the form
I(X)
I(X)
(cid:18) I11
I22 (cid:19) .
(cid:18)0 1
1 0(cid:19)
We claim that it is isomorphic to M2(I(X)) . Indeed it contains the
element
whose square is the identity element. Then define a product on the
subspace I(X) sitting in the right upper corner of the injective envelope
34
U. HAAG
of SX by the rule
(cid:18)0 x
0 0(cid:19) ∗(cid:18)0 y
0 0(cid:19) := (cid:18)0 x
0 0(cid:19) ◦(cid:18)0 1
1 0(cid:19) ◦(cid:18)0 y
0 0(cid:19)
where ◦ denotes the C ∗-multiplication in I(SX) . It is clear from the
definition that the new product is associative making I(X) into a unital
algebra, in fact a unital operator algebra. Being an injective operator
space it is equal to its own injective envelope. Since it is also unital it
has a unique structure of C ∗-algebra which is compatible with, hence
equal to, its operator algebra structure. Then one sees that I(SX)
contains (and is equal to) the subspace
(cid:18) 0
I(X)
I(X)
0 (cid:19) + (cid:18)0 1
1 0(cid:19) ◦(cid:18) 0
I(X)
I(X)
0 (cid:19)
which being a C ∗-algebra must be isomorphic with M2(I(X)) . There-
fore the map adU on SX has a unique (C ∗-automorphism) extension to
I(SX) ≃ M2(I(X)) which commutes with the matrix decomposition. It
remains to show that this extension restricted to M2(X) ⊆ M2(I(X))
is equal to adU . The completely isometric left and right action of F
mapping ∇2,U onto ∆2,U extends by rigidity to completely isometric
bijections lF and rF of the injective envelopes, so that putting
λF (x) := F lF (x)
,
ρF (x) := rF (x) F
defines, being unital bijective and completely isometric C ∗(F )-module
maps, (ungraded) C ∗-isomorphisms of I(∇2,U ) and I(∆2,U ) . One also
has the graded C ∗-automorphisms
Ad F (x) = F x F ,
adF (x) = F · x · F
on the injective envelopes of both of the two subsystems as above.
Each of these is a C ∗(F )-module map and respects the F -matrix de-
composition. Let α denote the grading automorphism of I(X2,U ) . Put
X2,U = α(X2,U ) and X = (1 − e) X2,U (1 − e) with enveloping graded
operator system bX = X + X . Also put lF = α◦ lF ◦ α , rF = α◦ rF ◦ α
etc. . The map
µ : (lF ◦ rF )(bX)
rF (bX)
bX ! ։ M2(bX)
lF (bX)
by identification of the corners via the F -matrix decomposition is again
completely isometric restricted to some enlarged corner. The domain
of µ needn't be invariant under adU but it is restricted to Y2,U . Of
course the same arguments apply with respect to the F -matrix decom-
position obtained from the maps lF and rF and replacing Y2,U by
35
2,U
Y
The map νl ⊕ νr is completely isometric restricted to the enveloping
operator system of any subspace ∆ij
2,U (since the unital complete isom-
etry extends to a selfadjoint complete isometry of injective envelopes).
Put AdFU n (x) = FU nadFU n (x)FU n . Also put
= α(Y2,U ) . The corresponding map to M2(bX) is denoted µ .
FU n lFU n (bX) FU n C (1 − e)! , S r,n
S l,n
By the general identity λF (α(x)) = α(ρF (x)) one gets
= (α ◦ AdFU n )(S l,n
lFU n (bX)
=
C e
) .
bX
bX
bX
lFU n (α(x)) = (α ◦ Ad FU n ◦ rFU n )(x) = (α ◦ AdFU n ◦ lFU n )(x)
bX
are operator systems with each of the (off diagonal) cor-
so that S l/r,n
ners completely isometric to the operator system bX . Their injective
envelopes are each naturally isomorphic with M2(I(bX)) by an argu-
ment as above and the same as the injective envelopes of the subspaces
and
and
0
T = (cid:18)C e
0 C (1 − e)(cid:19) ,
lFU n (bX)
U l,n
rFU n (bX) C (1 − e)(cid:19)
T = (cid:18) C e
Lr,n
0 C (1 − e)(cid:19) ,
T ,0 = (cid:18)C e
U l
rF (X) C (1 − e)(cid:19)
T ,0 = (cid:18) C e
Lr
lF (X)
0
of upper resp.
sponding subspaces
lower triangular matrices. We also consider the corre-
which are mutually adjoint to each other. Then consider the unital
operator spaces
T = ρ
(Lr,n
T )
FU n
eS n = λFU n (S l,n
bX
) = ρ
FU n
and their corresponding subspaces
T ) , eLr,n
bX
) , eU l,n
(S r,n
T = λFU n (U l,n
eLr,n
eU l,n
T ,0 ,
T ,0 .
Choose a completely isometric realization of the injective envelope
I(bX) in (1 − e) I(X2,U ) (1 − e) which is elementwise fixed under adU .
This is achieved starting from some arbitrary completely positive ex-
tension j to injective envelopes of the adU -covariant embedding
bX ֒→bX2,U
36
U. HAAG
by "averaging" over the adU -translates of j (compare with the dis-
′
2,U ) which must be a C ∗(F )-module
embedding of the latter into I(X
map) the lower right corner of the latter is an injective envelope of the
cussion below). Since bX ⊆ I(eS 0) (by any unital completely positive
former and may be assumed to agree with the specified copy of I(bX) as
above. In the same way bX ⊆ I(eS n) and one may choose both injective
envelopes to contain the same copy of I(bX) in the lower right corner.
One checks that each element FU n defines a selfadjoint unitary in the
injective envelope of U l
T ,0 and that any completely positive extension
of the inclusion into S l,n
to injective envelopes is a C ∗(FU n)-module
map. Then one gets the equality
eX′
λF (FU n) = FU n , ρF (FU n) = FU n ⇒ AdF (FU n) = FU n
FU n F = F FU −n ∈ I(bX) ⊕ F I(bX) F
by adU -invariance of any element in I(bX) and also
envelopes such that) (1− e) I(eU l,n
e) ⊆ I(bX) which means that adU leaves these subspaces elementwise
for every n . It follows that (one may choose corresponding injective
T ,0) (1−
T ,0) (1− e) ⊆ I(bX) , (1− e) I(eLr,n
fixed. This implies that
and since λFU n (FU k ) = FU k for every n and k the same holds for all
other corners, for example
(λFU n ◦ λF )((1 − e) I(eU l,0
(λFU n ◦ λF )(FU k (1 − e) I(eU l,0
so every I(eU l,n
T ,0) (1 − e)) ⊆ I(bX)
T ,0) (1 − e)) ⊆ FU k I(bX)
etc.
T ,0) is contained in the injective envelope of each
eS k , and all of these agree with the injective envelope of eS 0 which
in particular follows to be adU -invariant as a subspace of I(X2,U ) .
Then for x ∈ U l
T ,0 . We shall see that
there exists a completely isometric copy of U l
T ,0 such that its injective
envelope admits an adU -covariant realization as a subspace of the direct
sum of two copies of I(X2,U ) (with adU acting diagonally) and such
that its lower right corner with respect to the matrix decomposition is
T ,0 put exn = λFU n (x) ∈ eU l,n
contained in I(bX) ⊕ I(bX) (in particular it is elementwise fixed under
the action of adU ). If x ∈ U l
elements
T ,0 is an element consider the sequence of
(xm (cid:12)(cid:12)(cid:12) xm =
1
2m + 1
mXk=−mexk) .
( y
m (cid:12)(cid:12)(cid:12) y
=
m
1
2m + 1
mX−m eym) ,
(y) .
FU n
eym = ρ
l,m
Let U
so one gets a unital complete contraction
T ,0 be the subspace generated by the elements {xm x ∈ U l
T ,0}
37
σm : U l
T ,0
։ U
l,m
T ,0 ,
σm(x) = xm
for each m ∈ N . It is in fact completely bounded from below by the
image of the map νl . The idea is to construct a "halfsided" inverse
to the map σm which when composed with the map νl will be com-
pletely contractive. The complete contraction νl : X2,U ։ V 2,U,l has a
completely contractive adU -invariant extension
ψl : I(X2,U ) −→ I(V 2,U,l)
so that ψl(λFU n (x)) = ψl(x) for all x ∈ X2,U and n ∈ Z by the
of x under ψl , and the composition ψl ◦ σm equals νl on U l
T ,0 . In
particular kσm(x)kn ≥ kνl(x)kn . For y = x∗ ∈ Lr
T ,0 one analogously
constructs the sequence of elements
assumption on V 2,U,l . This means that each exk is sent to the image
This amounts to a complete contraction
σm : U l
T ,0
։ (Lr,m
T ,0)∗ ,
σm(x) = y∗
m
.
Using a similar argument gives kσm(x)kn ≥ kνr(x)kn . One concludes
that σm = σm ⊕ σm is completely isometric for each m . The se-
quence of maps {σm} must have an adherence value in the bounded
weak topology on the set of completely contractive maps of U l
T ,0 into
weak topology of completely contractive linear maps into B(H) for
[5], employing some adU -invariant completely positive projection of
the former with range equal to the latter space; to construct it one
first extends the map adU to a completely contractive map on B(H)
by injectivity and then uses a simple averaging process starting with
I(eS 0 ⊕ eS 0) ≃ M2(I(bX)) ⊕ M2(I(bX)) (induced by the so called bounded
a given representation space of M2(I(bX)) , compare with chap. 7 of
an arbitrary completely positive projection p onto M2(I(bX)) putting
pm =Pm
k=1 (adU −k ◦ p ◦ (adU )k) to obtain a sequence of approximately
adU -invariant projections as m → ∞ which again has some adU -
invariant limit point in the bounded weak topology) , which subset
is compact, and choosing a convergent subsequence {σmk k ∈ N} , the
limit map
σ = lim
k→∞
σnk
38
U. HAAG
is again a unital completely isometric and the image space σ(U l
adU -invariant, since for each element x
T ,0) is
kσm(adU (x)) − adU (σm(x))k → 0 , m → ∞ .
T ,0) admits an adU -covariant realization
such that its lower right corner is contained in I(bX) ⊕ I(bX) which
The injective envelope of σ(U l
means that each of its elements is fixed under adU , so the same must
be true for the injective envelope of U l
U l
T ,0 −−−→ SX
T ,0 . The composite map
pU
−−−→ SX
extends by a simple averaging process as above to a completely positive
map φ on injective envelopes which commutes with the maps adU .
Then φ is necessarily a C ∗(FU n)-module map for every n , so that if
x ∈ X one finds (approximately) a lift for
of the form
(cid:18)0 0
0 x(cid:19)
F (cid:18)0 x
0 0(cid:19) ∈ F U l
T ,0
(C ∗-multiplication in I(U l
T ,0) ). Since the lifted element is adU -invariant
and φ commutes with adU , then also its image must be adU -invariant.
Therefore, by rigidity, adU is the identity restricted to the subalgebra
0
(cid:18)C
0 I(X)(cid:19)
of M2(I(X)) and thus a C⊕ I(X)-module map. Following the proof of
Theorem 15.12. of [5] one gets that the map
A −→ I(X)
,
a 7→ a · 1
of A into the injective envelope of X is a C ∗-homomorphism imple-
menting the action of A . Then the action must be strongly completely
contractive so that X is an operator A-system which proves the second
assertion of the theorem (choosing a unital C ∗-representation of I(X) ).
Next assume that X is a super operator system with strongly admissi-
ble A-action and let X be the given super operator system with strong
A-action and ∗-linear complete contraction p : X ։ X with dense
image commuting with the A-actions. Then p extends naturally to a
graded map of enveloping graded operator systems
bp .bX −→ bX
39
which also is a complete contraction with dense image. We begin by
showing that the A-action extends to a graded strongly completely
map Ad U is unital, ∗-linear and completely contractive on Mn(X) ,
contractive action on bX . Let U ∈ Mn(A) be a unitary. Then the
hence it extends to a graded completely contractive map on Mn(bX) .
to a graded strongly completely contractive action on bX . This follows
One then gets that the left (and hence also right) action of A extends
since for U ∈ Mn(A) unitary putting
U = (cid:18)U 0
0 1(cid:19)
the map Ad U on M2n(X) extends (being unital, ∗-linear and com-
pletely contractive) to a Z2-graded completely contractive map on
M2n(bX) . But then the left action of U on bX must also be graded
and completely contractive. Then for any normal element X ∈ Mn(A)
consider the unitary
U = (cid:18)√1 − XX ∗
X ∗
X
−√1 − XX ∗(cid:19) ∈ M2n(A) .
One then deduces that the left action of A on X extends to a graded
joint elements, and then one easily extends to arbitrary elements so the
Since U acts completely contractive on M2n(bX) the element X acts
completely contractive on Mn(bX) . This holds in particular for selfad-
action of A on bX is strongly completely contractive and Z2-graded.
u-weakly completely contractive action on bX . To see this note that
on I(X) is certainly Z2-graded (in any case when restricted to bX ).
the left action of a unitary u extends (being completely isometric)
uniquely to a completely isometric map lu on the injective envelope
I(X) which is a unital Z2-graded C ∗-algebra. The corresponding map
Let π : I(X) → I(X) be any unital completely contractive extension
of p , and consider the composite map
I(X)
lu∗
−→ I(X)
π−→ I(X)
lu−→ I(X) .
It is another completely contractive extension of p , hence necessarily
selfadjoint for the ordinary adjoint operation. Since it is also selfadjoint
for the superinvolution (at least when restricted to X ) it must restrict
to a Z2-graded map from bX to bX . But then, since the maps lu∗ and
π are Z2-graded, so is the map lu :bX →bX . From this one infers that
the left action of A on X extends to a graded and u-weakly completely
40
U. HAAG
contractive action on its enveloping graded operator system. Clearly
this action is admissible so the result follows from the argument above
for the ungraded case
(cid:3)
The previous Theorem gives some information about the relation be-
tween the maximal C ∗-tensor product A⊗max B , which is the closure
of the algebraic tensor product in the direct sum over all joint com-
muting representations of A and B , and the symmetrized Haagerup
tensor product A⊗h∗B of two given (unital) C ∗-algebras A and B .
One can deduce that the former is completely ∗-isometric with the
largest operator system X0 which is smaller than the super operator
system A⊗h∗B and admits a u-weak A- and B-bimodule structure.
This follows since such an operator system is automatically admissible
for both actions, and hence is necessarily a strong A⊗maxB-bimodule
by the Theorem above. To see that it is admissible consider first the su-
per operator system Z = A⊗h∗B . From the very definition of the sym-
metrized Haagerup tensor product it is easy to see that the natural left
and right A-action is weakly completely contractive, this being the case
for the (unsymmetrized) Haagerup tensor product (the case of the B-
action follows by symmetry). A completely isometric ∗-representation
of the symmetrized Haagerup tensor product is constructed from a uni-
tal completely isometric representation of the unsymmetrized Haagerup
tensor product as indicated in the example above. On the other hand
a representation of the ordinary Haagerup tensor product is given by
embedding into the amalgamated free product A∗CB where an ele-
mentary tensor a ⊗ b is represented by the ordered product a ∗ b .
Then for the symmetrized Haagerup tensor product the elementary
tensor is represented by the element (a∗ b)b⊕(b∗ a) in the graded direct
sum Z = (A ∗C B)b⊕ (A ∗C B) with grading given by the flip of the
two summands. Let u ∈ A (resp. v ∈ B ) be a unitary and define an
ungraded C ∗-automorphism λu of Z by letting λu be equal to the
identity in the first direct summand as above and defining
λu(b1∗a1∗b2∗a2∗· · ·∗bn∗an) = u∗∗b1∗ua1u∗∗b2∗ua2u∗∗· · ·∗bn∗uan
on the second direct summand. Define the ungraded C ∗-automorphism
ρu to be equal to the identity on the second summand and given by
ρu(a1∗b1∗a2∗b2∗· · ·∗an∗bn) = a1u∗b1∗u∗a2u∗b2∗· · ·∗u∗anu∗bn∗u∗
on the first summand. Similarly one defines λv and ρv exchanging the
roles of the first and second summand. Then the left and right action
of the unitary group of A (resp. B ) on Z extends to a left and right
41
action on Z putting
lu(x) = u λu(x)
,
ru(x) = ρu(x) u .
Also note that with respect to the grading and superinvolution the
following identities hold
λu(x∗) = ρu∗(x)∗
,
λu(α(x)) = α(ρu∗(x)) .
Then consider the super operator systems ZA = M2(A) ⊗h∗ B in case
of the A-action, and similarly ZB = A⊗h∗ M2(B) for the B-action
with representations and enveloping graded C ∗-algebras Z A and Z B
as above. For a unitary u ∈ A and U = u ⊕ 1 the map adU can be
extended to a graded C ∗-automorphism of the enveloping C ∗-algebra
Z A . Consider in particular the pair of selfadjoint unitaries
and
0
E = (cid:18)1
0 −1(cid:19) ⊗ 1B ∈ M2(C)
F = (cid:18)0 1
1 0(cid:19) ⊗ 1B ∈ M2(C) .
(cid:8) λE(x) − x , ρE(x) − x (cid:12)(cid:12) x ∈ ZA(cid:9)
Let JE denote the graded C ∗-ideal of Z A generated by the relations
and check that it is invariant under adU , so that this automorphism
drops to the quotient which also admits a compatible matrix decompo-
sition by the strong action of C ∗(E) induced from C ∗-multiplication
(which restricts to the image super operator system ZA,E ). Then there
is a natural (bijective) complete contraction
ZA,E
։ M2(Z)
induced by dividing out the relations
(cid:8) λF (x) − x , ρF (x) − x (cid:12)(cid:12) x ∈ ZA,E(cid:9)
(check that λF and ρF pass to the quotient algebra since F anti-
commutes with E ) and commuting with the respective F -matrix de-
compositions and the maps adU , which renders a unital graded com-
plete contraction to M2(X0) , also compatible with F -matrix decom-
positions and adU . Put X2,U = ZA,E and X = Z and note that
(1 − e)X2,U (1 − e) is completely isometric to X . The condition that
the element FU n = adU n(F · 1) is an even selfadjoint unitary in the
injective envelope of X2,U for each n is easily deduced from the fact
that it defines an even selfadjoint unitary in the enveloping C ∗-algebra
42
U. HAAG
Z A,E = Z A/JE which is isomorphic with the graded direct sum of two
copies of the amalgamated free product
(cid:0)M2(A) ∗C ∗(E) (B ⊕ B)(cid:1) b⊕ (cid:0)M2(A) ∗C ∗(E) (B ⊕ B)(cid:1)
and Proposition 15.10 of [5]. Similarly one checks that F ·FU n = F FU n
for all n ∈ Z . Put ZA,E,0 = M2(A) ∗C ∗(E) (B ⊕ B) and Z0 = A∗C B ,
and let V 2,U,l and V 2,U,r be the subspaces linearly generated by ordered
products x ∗ y resp. y ∗ x with x ∈ M2(A) , y ∈ B ⊕ B . The last
postulate is first checked on the corresponding subspaces for the natural
C ∗-surjections
πl,r : Z A,E ։ M2(Z0)
by projection onto the left or right summand in the graded direct sum
as above. One checks that the map adU n naturally drops to the di-
agonal of M2(Z0) where it is given by adun in the left upper corner
and the identity map in the lower right corner. Then one defines adU n
to be equal to (the projection onto the left or right direct summand
of) the map lun as above in the upper right corner and equal to (the
corresponding projection of) ru−n in the lower left corner of M2(Z0)
and verifies that this map extends the (tautological) image of adU n
on M2(V l,r) such that the extension is completely isometric when re-
stricted to each corner. Since such an extension is necessarily unique
in the injective envelope, and since the subspace given by the linear
span of products of elements of M2(V l,r) with elements from the set
{FU n} in the injective envelope can be realized taking the C ∗-product
in any C ∗-representation of Z0 one gets that also this chosen exten-
sion of adU n restricted to the corresponding subspaces of M2(Z0) is
unique with this property and must be equal to the corresponding ex-
tension in the injective envelope. The composite adU n ◦ πl,r is equal
to πl,r ◦ adU n on the chosen subspaces which is an easy consequence
of the fact that adU is a left resp. right C ∗(FU n) − C ∗(FU n+1)-module
map for M2(V l) resp. M2(V r) which follows since the Haagerup ten-
sor product A⊗h B is a strong left A- and a strong right B-module,
so it is completely contractive and a C ∗(FU n)-module map for each
n ∈ Z completing the argument that the A-action on X0 is admissi-
ble. The case of B-action follows by symmetry. Since both actions are
strongly completely contractive from the Theorem above and commute
with each other, they combine to a strongly completely contractive ac-
tion of A⊗max B . Putting π0(x) = x · 1X0 renders a unital complete
contraction
π0 : A⊗max B −→ X0
43
which must be a complete isometry by the assumption that X0 is
maximal.
It is of course impossible that the operator system associated with
the symmetrized Haagerup tensor product itself should be completely
isometric with the maximal C ∗-tensor product except in trivial cases.
However the following Theorem gives a partial result in this direction.
Let A be a C ∗-algebra and consider the super operator system given
by the (ordinary) Haagerup tensor product A⊗h A with involution
defined on elementary tensors by
x ⊗ y
7→ y∗ ⊗ x∗ .
Let the unitary group UA of A act diagonally from the left by
lu(x ⊗ y) = ux ⊗ uy .
Then the left action commutes with the associated right action, so
that A⊗h A becomes a (weak) UA-bimodule in the sense defined above.
Moreover the action restricts to the subspace Y generated by elements
of the form {u ⊗ u u ∈ UA} which again is a super operator system.
Embedding it into the amalgamated free product C ∗-algebra A ∗C A
with Z2-grading exchanging the two natural copies of A denoted j(A)
and j(A) consider its quotient by the graded C ∗-ideal generated by the
odd elements
(cid:8)j(v)j(v) − j(v)j(v) v ∈ UA(cid:9) .
Let X0 be the image of Y in the quotient which is an operator sys-
tem with weakly completely contractive adjoint diagonal action of UA .
The argument used in the proof of the following Theorem then also
shows that any operator system which is smaller than X0 and ad-
mits a weakly completely contractive action of UA compatible with
the corresponding diagonal action on Y is a strong UA-bimodule. To
obtain a specific example we use a slightly different but related con-
struction. Namely, consider the amalgamated free product C ∗-algebra
Z = M2(A) ∗C M2(A) with grading automorphism given by exchang-
ing the two natural copies j(M2(A)) and j(M2(A)) of M2(A) in the
manner above. Let V ∈ M2(A) denote a unitary which is of the form
0
(cid:18)v
0 w(cid:19)
with v, w ∈ UA . Divide Z by the graded C ∗-ideal generated by the
images of the odd elements
{j(V )j(V ) − j(V )j(V )} .
44
U. HAAG
Putting Ej = j(E) , Ej = j(E) , Fj = j(F ) , Fj = j(F ) and F =
Fj Fj ( E and F are the selfadjoint unitaries of M2(C) as above)
the relations imply in particular that Ej and Ej commute with each
other. Then the elements p = ej ej = j(e) j(e) and q = (1−ej) (1−ej)
define orthogonal projections in the quotient Y . For u ∈ UA let
U = u⊕1 and adU be the free product of the inner C ∗-automorphisms
Ad j(U) on j(M2(A)) and Ad j(U) on j(M2(A)) by conjugation with
U leaving p and q invariant. We will also have to divide by the
relations
(cid:8)F U p − q F U , F U q − p F U u ∈ UA(cid:9)
with F U = adU (F ) in order that the unitary F intertwines the projec-
tions p and q . The image of the hereditary subalgebra (p+q) Y (p+q)
with unit p + q in the quotient C ∗-algebra is denoted X . It admits
a matrix decomposition by C ∗-multiplication with p and q . Then
consider the operator subsystem X2 of X which is generated by the
elements
(cid:26)(cid:18)Fj j(u1)j(u1) Fj
j(v2F ) j(v2F ) Fj j(v1)j(v1) Fj(cid:19)(cid:27)
j(u2F ) j(u2F )
with u1 , u2 , v1 , v2 ∈ UA .
Its lower right corner X is an operator
system with unit element q and corresponds to the upper right corner
by C ∗-multiplication with the unitary F from the left.
("Diagonal of tensor product" theorem) With notation
Theorem.
as above the operator system X is a strong UA-bimodule for the (left)
action given by
lu : q Fj j(v)j(v) Fj q 7→ q Fj j(uv)j(uv) Fj q .
Proof. The first aim is to exhibit that the action is well defined and
(weakly) completely isometric. To this end note that adU drops to a
C ∗-automorphism of X which leaves invariant the upper right corner
(and lower left corner) of X2 , so that by the composition
∼
adU
∼
X
−−−→ F X
−−−→ F X
−−−→ X
giving the left action of u one finds that lu is completely isometric.
In a second step we show that the action is admissible. Then one
may apply the SCES-Theorem to conclude that the action is strongly
completely contractive. One easily easily finds that the natural map
X0 ։ X is completely contractive. Let Z denote the super operator
system A⊗h M2(C) ⊗h A with the completely antiisometric involution
x ⊗ c ⊗ y
7→ y∗ ⊗ c∗ ⊗ x∗
45
and note that it can be embedded in the amalgamated free product
C ∗-algebra Z = A ∗C M2(C) ∗C A with grading automorphism given
by exchanging the two natural copies j(A) and j(A) of A . Using the
diagonal embedding of A into A ⊕ A ≃ A ⊗ C ∗(E) ( E and F are
the selfadjoint unitaries of M2(C) as above) consider the image Z E of
Z in the amalgamated free product (A ⊕ A) ∗C ∗(E) M2(C) ∗C ∗(E) (A ⊕
A) . Then Z E admits a matrix decomposition coming from the strong
graded action of C ∗(E) by C ∗-multiplication. One has two natural
projections
πr : Z E ։ (A⊕ A) ∗C ∗(E) M2(A)
πl : Z E ։ M2(A) ∗C ∗(E) (A⊕ A) ,
where the copy j(A) always maps to the left factor, and j(A) to the
right factor in the amalgamated free product (in the obvious way) and
a projection
π : Z E ։ M2(A ∗C A)
which factors over both πl and πr . Let Zr,l denote the C ∗-algebras
which are the surjective image of Z E under the ∗-homomorphisms πr
and πl respectively. Then the grading automorphism on Z E drops
to a ∗-isomorphism Zr ≃ Zl . For a given unitary u ∈ A one further
divides Z E by the graded C ∗-ideal generated by the elements
(cid:8)j(u) j(u) − j(u)j(u) ,
j(u) F − F j(u) ,
j(u) F − F j(u)(cid:9) ,
where u is shorthand for the diagonal element u ⊕ u ∈ A ⊕ A , which
quotient is denoted XU . Upon dividing Zl,r by the images of the
relations as above one obtains two corresponding projections
πU,l/r : XU ։ XU,l/r
induced by the grading on XU .
and a ∗-isomorphism XU,l ≃ XU,r
Let X U be the graded direct sum of two copies of XU with grading
given by the product of the grading on each summand plus the flip
exchanging the summands and πU,l/r be the (nonfaithful ungraded)
representations of X U given by composing πU,l and πU,r respectively
with the projection onto the left resp.
right direct summand in its
decomposition as a graded direct sum. Let adU denote the graded
C ∗-automorphism of X U defined by
p j(x) 7→ p j(uxu∗) ,
p j(x) 7→ p j(uxu∗) ,
q j(x) 7→ q j(x) ,
q j(x) 7→ q j(x) ,
F b⊕ 0 7→ (cid:0)p j(u) F j(u) + q j(u∗) F j(u∗)(cid:1) b⊕ 0
0b⊕ F 7→ 0b⊕ (cid:0)p j(u) F j(u) + q j(u∗) F j(u∗)(cid:1)
46
U. HAAG
plus the identity map of C ∗(E) and check that it is well defined. Then
consider the super operator system X2,U which is the subspace of X U
generated by elements
j(v2) F j(v2)
(cid:26)(cid:18)F j(u1)j(u1) F j(u2) F j(u2)
j(v1)j(v1) (cid:19)b⊕(cid:18)F j(u1)j(u1) F j(u2) F j(u2)
j(v1)j(v1) (cid:19)(cid:27)
and let X be its lower right corner. One defines the subsystems ∆2,U
and ∇2,U as in the definition of an admissible action above. Let rF be
the order two linear map which exchanges the columns on both sides
of the graded direct sum, i.e. the image of an element as above is sent
to
j(v2) F j(v2)
j(v1) F j(v1)
(cid:18)F j(u2)j(u2)F j(u1) F j(u1)
j(v2)j(v2) (cid:19)b⊕(cid:18)F j(u2)j(u2) F j(u1) F j(u1)
j(v2)j(v2) (cid:19) .
with C ∗-multiplication by the even selfadjoint unitary F = F b⊕ F
from the right, it is unital, and equal to the free product of the C ∗-
endomorphisms
One checks that it is well defined.
If ρF is the composition of rF
j(v1) F j(v1)
j(x) 7→ F j(x) F ,
j(x) 7→ j(x)
on the right side of the graded direct sum, whereas on the left side it
is given by the free product of the C ∗-endomorphisms
j(x) 7→ j(x) ,
j(x) 7→ F j(x) F
if restricted to the off-diagonal subsystem ∇2,U , while on the diagonal
subsystem ∆2,U it is the free product of the C ∗-endomorphisms
j(x) 7→ F j(x) F ,
j(x) 7→ j(x) .
hence ρF is completely isometric (though not graded) if restricted to
either of the subsystems ∇2,U or ∆2,U . Then also rF is completely
isometric on these subsystems. Similarly, let lF be the map exchanging
the rows on both sides, and λF the unital map with is the composite
of lF with C ∗-multiplication by F from the left. Then it is equal to
the map
j(x) 7→ j(x) ,
j(x) 7→ F j(x) F
on the left summand of the graded direct sum, whereas on the right
side it is given by
j(x) 7→ j(x) ,
j(x) 7→ F j(x) F
if restricted to the diagonal subsystem ∆2,U , and
j(x) 7→ F j(x) F ,
j(x) 7→ j(x)
47
for an element in ∇2,U . Put Y2,U = ∇2,U + X , and check that ∆2,U is
contained in the C ∗(F )-subbimodule generated by Y2,U ∩ ∆2,U in the
injective envelope of the latter subspace. It is readily seen that adU re-
stricts to a completely isometric ∗-linear bijection of Y2,U which leaves
the subspace X elementwise fixed. One also verifies the conditions that
the elements F · FU n = F FU n are evenly graded unitaries in the injec-
tive envelope (of either of the subspaces). Further one has two unital
complete contractions
νl : X2,U ։ V 2,U,l ,
νr : X2,U ։ V 2,U,r
onto unital operator spaces V 2,U,l/r which are defined to be the im-
ages of X2,U under π2,U,l/r respectively. The natural weak F -matrix
decomposition on the images which is compatible with the maps νl/r
is checked to have the additional property that the right F -action on
is implemented by C ∗-multiplication with F from the right,
V 2,U,r
while for V 2,U,l the left action of F coincides with C ∗-multiplication
(from the left). The map adU is checked to drop to a correspond-
ing map when restricted to the intersection of the images of Y2,U in
both quotients which is completely isometric on the images of the di-
agonal / off-diagonal subsystems.
In case of the diagonal subsystem
this map is simply given by conjugation with the image of the unitary
element j(U)j(U) , but in case of the off-diagonal subsystem one has
to make a distinction between the left and right case. In the first in-
stance one takes the composition of the free product of the identity
map on j(A⊕ A) and the map Ad j(U) on j(M2(A)) , composed with
conjugation by the unitary j(U) of the whole C ∗-algebra (leaving the
off-diagonal subsystem invariant), whereas in the second instance the
situation is reversed (mirrored). Finally one checks that restricted to
the image of an enlarged corner ∆ij
2,U the maps νl/r factor over the
surjection (up to complete isometry) X 2,U ։ M2(X U ) corresponding
to π : Z E ։ M2(A ∗C A) , so the relations
{j(u) F − F j(u) , j(u) F − F j(u)}
automatically drop out and X U is simply graded direct sum of two
copies of the image of A ∗C A divided by the C ∗-ideal generated by
the single commutator
{j(u) j(u) − j(u) j(u)} .
Then one easily verifies that the norm of an element x ∈ Mn(∆ij
satisfies
2,U )
kxkn = max{kνl(x)kn , kνr(x)kn} .
48
U. HAAG
To complete our argument we need to connect this construction to the
previous one by a ∗-linear map
pU : X2,U ։ M2(X)
which is compatible with the F -matrix decomposition and the map
adU and completely contractive restricted to an enlarged corner. But
then it is clear that the image of the subspace of A ∗C A modulo the
∗-ideal generated by the single relation above, generated by elements
{j(v) j(v) v ∈ UA}
surjects onto the corresponding subspace in the free product divided
by all relations of the form
{j(v) j(v) − j(v) j(v) v ∈ UA} ,
hence, a forteriori, onto X , while on the other hand the map to
M2(X U ) is completely isometric restricted to an enlarged corner so
its image is contained in (a copy of) M2(X) . The rest of the statement
is immediate. One concludes that the diagonal action of UA on X is
a strongly completely contractive, and is unitarily implemented in the
injective envelope
(cid:3)
Remark. The Theorem implies that X in fact carries the structure of
a unital C ∗-algebra UA which is linearly generated by the unitary group
of A (modulo {±1} ). Suppose that A admits a (unital) augmentation
∗-homomorphism ε : A ։ C . Then the free product of the identity
map of j(M2(A)) and ε on j(M2(A)) is compatible with the relations
for X which amounts to a completely positive surjection of UA onto
A which is a ∗-homomorphism since it induces a group isomorphism
In any case the assignment A UA is clearly
UA/{±1} → UA .
functorial with respect to ∗-homomorphisms.
The definition of the maximal C ∗-tensor product has a natural exten-
sion to super-C ∗-algebras which we record in the following definition.
(i) Let A , B be two (unital) super-C ∗-algebras. The
Definition 3.
maximal tensor product, denoted A⊗max B is defined to be the super-
C ∗-algebra which is the completion of the involutive algebra generated
by the algebraic tensor product of A and B with product type invo-
lution in the universal representation given by taking the direct sum
(up to equivalence) over all joint (unital) commuting ∗-representations
of A and B .
(ii) Let V and W be super operator spaces with dual super operator
spaces V∗ and W∗ respectively. One has a natural ∗-embedding of
49
the algebraic tensor product V⊗ W with product type involution into
the dual of the symmetrized Haagerup tensor product (V∗ ⊗h∗ W∗)∗ ,
which clearly defines a cross matrix norm on the algebraic tensor prod-
uct (since it is smaller than the projective, but larger than the injective
tensor norm). Its completion will be denoted V ⊗h∗ W and called the
dual symmetrized Haagerup tensor product.
Note that by commutativity of the symmetrized Haagerup tensor prod-
uct, also the dual symmetrized Haagerup tensor product is commuta-
tive. From this one easily infers that it is the largest commutative
tensor norm on the algebraic tensor product which is smaller than the
Haagerup tensor norm. In particular it must be larger than the maxi-
mal super-C ∗-tensor norm, which is also commutative (and associative)
and smaller than the Haagerup tensor norm.
Let C ∗(G) denote the full C ∗-algebra of the discrete group G and
∆n(G) ⊆ C ∗(G)b⊗ · · ·b⊗ C ∗(G) the super operator subspace of the n-
fold projective tensor product of C ∗(G) with itself which is the closure
of the linear span of diagonal elements {g ⊗ · · · ⊗ g g ∈ G} . From
the augmentation homomorphism e : C ∗(G) ։ C one gets a natural
complete contraction ∆n+1(G) ։ ∆n(G) for each n ∈ N , so that the
operator matrix norms imposed on the complex group ring C G by the
diagonal embedding into the n-fold projective tensor product yields an
increasing sequence of matrix norms which must necessarily converge
to a limit norm since it is constant for the generators {g} . The closure
with respect to this limit norm will be denoted by ∆(G) . It is easy
to see from the construction that this super operator space admits a
comultiplication map δ : ∆(G) → ∆(G)b⊗∆(G) , which is involutive
with respect to the product type involution on the projective tensor
product (which coincides on the diagonal with the involution given by
x ⊗ y 7→ y∗ ⊗ x∗ ), and a counit for this comultiplication given by the
composition
∆(G) −→ C ∗(G)
e−→ C .
We let AG = ∆(G)∗ denote the dual operator space of ∆(G) which
admits a dual completely antiisometric involution (compare with the
final example of section 1). The whole construction can also be carried
through with the Haagerup tensor product replacing the projective
tensor product. We denote the corresponding increasing sequence of
super operator spaces by {∆h
n(G)} converging up to the limit ∆h(G) ,
h
G for the dual super operator space ∆h(G)∗ . One has the
and write A
following result.
50
U. HAAG
Theorem 2. The super operator space AG admits the additional
structure of a unital and counital, commutative Hopf super C ∗-algebra
with multiplication (comultiplication) induced by comultiplication (mul-
tiplication) on ∆(G) . Here comultiplication of AG means a completely
contractive ∗-homomorphism ∆ of AG into the mapping space of com-
pletely bounded maps BG = CB(∆(G) , AG) which again is a super
C ∗-algebra naturally containing the spatial tensor product AG ⊗AG .
Then the term counital means that the dualization of the inclusions
satisfy the relations
1l , 1r : ∆(G) → ∆(G)b⊗∆(G) with 1l(x) = 1G ⊗ x , 1r(x) = x ⊗ 1G
1∗
l ◦ ∆ = idAG = 1∗
r ◦ ∆ .
h
G admits the additional structure of a
The super operator space A
unital commutative super C ∗-algebra with multiplication induced by
comultiplication of ∆h(G) . One has a natural completely contrac-
G → AG . Moreover there is an involutive
tive ∗-homomorphism A
and contractive homomorphism from the involutive Banach algebra
AG = C ∗(G)∗ which is the dual Banach space of C ∗(G) (with in-
volution given by φ∗(x) = φ(x∗) and multiplication induced by the
comultiplication of C ∗(G) ) into AG which factors over A
h
h
G .
Proof. We first outline the multiplication on AG which amounts to
the specification of a completely contractive ∗-linear map
AG ⊗h AG
m−→ AG
(for the natural involution x ⊗ y 7→ y∗ ⊗ x∗ on the Haagerup tensor
product) satisfying m ◦ (m ⊗ id) = m ◦ (id ⊗ m) . The comultiplica-
tion map δ of ∆(G) can be composed with the complete contraction
∆(G)b⊗∆(G) ։ ∆(G) ⊗h ∆(G) to yield a complete isometry
∆(G) −→ ∆(G) ⊗h∆(G)
which, if dualized, gives the multiplication map of AG by the compo-
sition
AG ⊗h AG −→ (∆(G) ⊗h ∆(G))∗
δ∗
−→ AG
the first map being completely isometric by selfduality of the Haagerup
tensor product (c.f. Theorem 9.4.7 of [3]). It is easily checked that m
is commutative and associative, secondly that the element of AG de-
termined by e is a left and right unit for m , and that (x ◦ y)∗ =
y∗ ◦ x∗ = x∗ ◦ y∗ (the last identity follows from commutativity). Thus
AG is a unital commutative abstract super C ∗-algebra. To construct
comultiplication one notes on approximating ∆(G) by the super oper-
ator spaces {∆n(G)} and using commutativity and associativity of the
projective tensor product that one gets an associative "multiplication"
map
51
induced by the multiplication of C ∗(G) . Dualizing this map gives the
desired comultiplication
∆(G)b⊗∆(G) −→ ∆(G)
∆−→ BG ≃ (∆(G)b⊗∆(G))∗
AG
(Corollary 7.1.5 and Proposition 8.1.2 of [3]). Using the same argu-
ments as above one finds that BG is a commutative super C ∗-algebra
h
and that ∆ and 1∗
G the
situation is much the same. The multiplication map is obtained from
the comultiplication
r are ∗-homomorphisms. Turning to A
l , 1∗
∆h(G)
δh
−−−→ ∆h(G)⊗h∆h(G)
which is defined just as before. The unit is again obtained from the
augmentation map
∆h(G) −→ C ∗(G)
e−→ C
and all stated properties can be checked by dualization of the corre-
sponding properties for ∆h(G) and ∆(G) (resp. C ∗(G) ). This also
renders the contractive ∗-homomorphisms AG −→ A
(cid:3)
G −→ AG
h
It is conceivable that there is also some kind of Hopf struc-
Remark.
h
G . The problem is however that the Haagerup tensor prod-
ture for A
uct is not commutative (and the symmetrized Haagerup tensor prod-
uct which gives the same metric on the diagonal is not associative).
Nevertheless it may be that one has some kind of comultiplication ho-
h
G itself or some super C ∗-algebra inbetween
momorphism on either A
h
G and AG possibly with range into the maximal super C ∗-tensor
A
product (resp. some enveloping super C ∗-algebra thereof). We write
h
Ah
G , and AG for the closure
G for the closure of the image of AG in A
of its image in AG , both of which are super C ∗-algebras. The situation
is very similar to the dual setting where one is dealing with different
kinds of group algebras, with AG taking the part of the Banach algebra
l 1(G) , then Ah
G corresponds in some sense to the full group C ∗-algebra
C ∗(G) , and AG vaguely to the reduced group C ∗-algebra C ∗
r (G) .
52
U. HAAG
That the completely antiisometric involution on the abstract operator
algebra AG as above is antimultiplicative is easy to check in this exam-
ple, but in fact is automatic. Namely, a unital operator algebra with a
given antilinear and completely antiisosmetric involution is in particu-
lar a super operator system, so its injective envelope is a Z/2Z-graded
C ∗-algebra (it is the same as the injective envelope of the enveloping
graded operator system), so the linear grading automorphism extends
by rigidity to an order two bijective unital completely isometric map
which is necessarily a C ∗-automorphism by Proposition 1. Then the
superinvolution which is the product of the adjoint map and the grad-
ing is antimultiplicative on the injective envelope. On the other hand
one knows that the embedding of the operator algebra into its injective
envelope is a homomorphism, so that the superinvolution must be anti-
multiplicative. This is a special feature of involutive operator algebras
and does not hold for Banach algebras with an isometric (antilinear)
involution.
Example. Let G be the finite cyclic group Z/nZ . By Pontrjagin du-
ality the dual abelian group bG is again isomorphic with G so that
C ∗(G) ≃ C(bG) ≃ C(G) and AG ≃ l 1(G) . As an algebra l 1(G) is
isomorphic with C ∗(G) the latter of which is the unique commutative
C ∗-algebra generated by n orthogonal projections. These projections
are of norm one already in l 1(G) so that if one is to obtain a con-
tractive representation of l 1(G) into B(H) the image is necessarily a
C ∗-algebra, and if one assumes the representation to be injective there
is only one such choice up to C ∗-isomorphism, which is C ∗(G) . So in
this case one must have that both AG and Ah
G are isomorphic with
C ∗(G) . Note however that the dual involution on AG (and AG , Ah
G )
does not coincide with the usual involution of l 1(G) but involves an
additional grading on these commutative algebras induced by (linear
extension of the assignment) sending a group element g to g−1 . So
even in this very simple case one gets (nontrivially) graded C ∗-algebras.
In the general case it cannot be expected that either AG or Ah
G should
be C ∗-algebras, only operator algebras with a completely antiisometric
involution.
As another example of (noncommutative) super C ∗-algebras, consider
the enveloping super C ∗-algebra of a superunitary representation of a
given topological (or locally compact) group. A superunitary is an el-
ement U of B(bH) satisfying UU ∗ = U ∗U = 1 . Since these elements
form a group under multiplication it makes sense to talk of a superuni-
tary representation meaning a group homomorphism into the group of
superunitaries which is continuous for a suitable topology. It should be
noted that the norm of a superunitary is always larger than one and
is equal to one if and only if the element is a (evenly graded) unitary.
Therefore since the norm function of a superunitary has no general
bound from above there is no such thing like a universal group super
C ∗-algebra, covering all possible superunitary representations of G .
53
References
[1] D. Blecher, M. Neal, Metric characterizations of isometries and of unital operator spaces and
systems, Math. arXiv (2008)
[2] D. Blecher, M. Neal, Metric characterizations II, Math arXiv (2012)
[3] E. Effros, Z.-J. Ruan, Operator Spaces, London Mathematical Sciences Monographs - New
Series 23, Oxford University Press, 2000.
[4] R. Haag, J. Lopuszanski, M. Sohnius, All possible generators of supersymmetries for the S-
matrix, Nuclear Physics B88 (1975), 257 -- 274.
[5] V. Paulsen, Completely Bounded Maps and Operator Algebras, Cambridge Studies in Ad-
vanced Mathematics 78, Cambridge University Press, 2002.
|
1210.3256 | 2 | 1210 | 2013-02-04T13:33:20 | On the Grothendieck Theorem for jointly completely bounded bilinear forms | [
"math.OA",
"math.FA"
] | We show how the proof of the Grothendieck Theorem for jointly completely bounded bilinear forms on C*-algebras by Haagerup and Musat can be modified in such a way that the method of proof is essentially C*-algebraic. To this purpose, we use Cuntz algebras rather than type III factors. Furthermore, we show that the best constant in Blecher's inequality is strictly greater than one. | math.OA | math |
ON THE GROTHENDIECK THEOREM FOR JOINTLY
COMPLETELY BOUNDED BILINEAR FORMS
TIM DE LAAT
Abstract. We show how the proof of the Grothendieck Theorem for jointly
completely bounded bilinear forms on C ∗-algebras by Haagerup and Musat
can be modified in such a way that the method of proof is essentially C ∗-
algebraic. To this purpose, we use Cuntz algebras rather than type III factors.
Furthermore, we show that the best constant in Blecher's inequality is strictly
greater than one.
1. Introduction
In [10], Grothendieck proved his famous Fundamental Theorem on the metric
theory of tensor products. He also conjectured a noncommutative analogue of this
theorem for bounded bilinear forms on C∗-algebras. This noncommutative Grothen-
dieck Theorem was proved by Pisier assuming a certain approximability condition
on the bilinear form [16]. The general case was proved by Haagerup [11]. Effros
and Ruan conjectured a "sharper" analogue of this theorem for bilinear forms on
C∗-algebras that are jointly completely bounded (rather than bounded) [9]. More
precisely, they conjectured the following result, with universal constant K = 1.
Theorem 1.1 (JCB Grothendieck Theorem). Let A, B be C∗-algebras, and let
u : A × B → C be a jointly completely bounded bilinear form. Then there exist
states f1, f2 on A and g1, g2 on B such that for all a ∈ A and b ∈ B,
2 g2(bb∗)
2 + f2(a∗a)
2 g1(b∗b)
u(a, b) ≤ Kkukjcb (cid:16)f1(aa∗)
1
1
2(cid:17) ,
1
1
where K is a constant.
We call this Grothendieck Theorem for jointly completely bounded bilinear forms
on C∗-algebras the JCB Grothendieck Theorem. It is often referred to as the Effros-
Ruan conjecture.
In [18], Pisier and Shlyakhtenko proved a version of Theorem 1.1 for exact op-
erator spaces, in which the constant K depends on the exactness constants of the
operator spaces. They also proved the conjecture for C∗-algebras, assuming that
at least one of them is exact, with universal constant K = 2
2 .
3
Haagerup and Musat proved the general conjecture (for C∗-algebras), i.e., The-
orem 1.1, with universal constant K = 1 [12]. They used certain type III factors
in the proof. Since the conjecture itself is purely C∗-algebraic, it would be more
satisfactory to have a proof that relies on C∗-algebras. In this note, we show how
the proof of Haagerup and Musat can be modified in such a way that essentially
only C∗-algebraic arguments are used. Indeed, in their proof, one tensors the C∗-
algebras on which the bilinear form is defined with certain type III factors, whereas
The author is supported by the Danish National Research Foundation through the Centre for
Symmetry and Deformation.
1
2
TIM DE LAAT
we show that it also works to tensor with certain simple nuclear C∗-algebras ad-
mitting KMS states instead. We then transform the problem back to the (classical)
noncommutative Grothendieck Theorem, as was also done by Haagerup and Musat.
Recently, Regev and Vidick gave a more elementary proof of both the JCB
Grothendieck Theorem for C∗-algebras and its version for exact operator spaces
[19]. Their proof makes use of methods from quantum information theory and has
the advantage that the transformation of the problem to the (classical) noncom-
mutative Grothendieck Theorem is more explicit and based on finite-dimensional
techniques. Moreover, they obtain certain new quantitative estimates.
For an extensive overview of the different versions of the Grothendieck Theorem,
as well as their proofs and several applications, we refer to [17].
This text is organized as follows. In Section 2, we recall two different notions
of complete boundedness for bilinear forms on operator spaces. In Section 3, we
recall some facts about Cuntz algebras and their KMS states. This is needed for
the proof of the JCB Grothendieck Theorem, which is given in Section 4 (with a
constant K > 1) by using (single) Cuntz algebras. We explain how to obtain K = 1
in Section 5. In Section 6, we show that using a recent result by Haagerup and
Musat on the best constant in the noncommutative little Grothendieck Theorem,
we are able to improve the best constant in Blecher's inequality.
2. Bilinear forms on operator spaces
Recall that an operator space E is a closed linear subspace of B(H) for some
Hilbert space H. For n ≥ 1, the embedding Mn(E) ⊂ Mn(B(H)) ∼= B(H n) gives
rise to a norm k.kn on Mn(E). In particular, C∗-algebras are operator spaces. A
linear map T : E → F between operator spaces induces a linear map Tn : Mn(E) →
Mn(F ) for each n ∈ N, defined by Tn([xij ]) = [T (xij)] for all x = [xij ] ∈ Mn(E).
The map T is called completely bounded if the completely bounded norm kTkcb :=
supn≥1 kTnk is finite.
There are two common ways to define a notion of complete boundedness for
bilinear forms on operator spaces. For the first one, we refer to [5]. Let E and
F be operator spaces contained in C∗-algebras A and B, respectively, and let u :
E × F → C be a bounded bilinear form. Let u(n) : Mn(E) × Mn(F ) → Mn(C) be
the map defined by ([aij], [bij]) 7→ [Pn
k=1 u(aik, bkj )].
Definition 2.1. The bilinear form u is called completely bounded if
kukcb := sup
n≥1ku(n)k
is finite. We put kukcb = ∞ if u is not completely bounded.
Equivalently (see Section 3 of [12] or the Introduction of [18]), u is completely
bounded if there exists a constant C ≥ 0 and states f on A and g on B such that
for all a ∈ E and b ∈ F ,
(1)
2 g(b∗b)
1
u(a, b) ≤ Cf (aa∗)
1
2 ,
and kukcb is the smallest constant C such that (1) holds.
For the second notion, we refer to [3], [9]. Let E and F be operator spaces
contained in C∗-algebras A and B, respectively, and let u : E×F → C be a bounded
ON THE JCB GROTHENDIECK THEOREM
3
bilinear form. Then there exists a unique bounded linear operator u : E → F ∗ such
that
u(a, b) = hu(a), bi
for all a ∈ E and b ∈ F , where h., .i denotes the pairing between F and its dual.
Definition 2.2. The bilinear form u is called jointly completely bounded if the map
u : E → F ∗ is completely bounded, and we set
We put kukjcb = ∞ if u is not jointly completely bounded.
kukjcb := kukcb.
Equivalently, if we define maps un : Mn(E) ⊗ Mn(F ) → Mn(C) ⊗ Mn(C) by
un
k
X
i=1
ai ⊗ ci,
l
X
j=1
bj ⊗ dj
=
k
l
X
i=1
X
j=1
u(ai, bj)ci ⊗ dj
for a1, . . . , ak ∈ A, b1, . . . , bl ∈ B, and c1, . . . , ck, d1, . . . , dl ∈ Mn(C), then we have
kukjcb = supn≥1 kunk.
3. KMS states on Cuntz algebras
For 2 ≤ n < ∞, let On denote the Cuntz algebra generated by n isometries, as
introduced by Cuntz in [6], in which one of the main results is that the algebras
On are simple. We now recall some results by Cuntz. If α = (α1, . . . , αk) denotes
a multi-index of length k = l(α), where αj ∈ {1, . . . , n} for all j, we write Sα =
It follows that for every nonzero word M in
Sα1 . . . Sαk , and we put S0 = 1.
i=1 S{S∗
{Si}n
For k ≥ 1, let F k
let F 0
n = C1. It follows that F k
n ⊂ F k+1
F k
n∞.
If we write Pn for the algebra generated algebraically by S1, . . . , Sn, S∗
ν l(µ) = l(ν) = k}, and
n is ∗-isomorphic to Mnk (C), and, as a consequence,
n is a UHF-algebra of type
i=1, there are unique multi-indices µ and ν such that M = SµS∗
ν .
. The C∗-algebra Fn generated by S∞
n be the C∗-algebra generated by {SµS∗
1 , . . . , S∗
n,
k=0 F k
i }n
n
each element A in Pn has a unique representation
1 )kA−k + A0 +
A =
(S∗
N
X
k=1
N
X
k=1
AkSk
1 ,
where N ∈ N and Ak ∈ Pn ∩ Fn. The maps Fn,k : Pn → Fn (k ∈ Z) defined by
Fn,k(A) = Ak extend to norm-decreasing maps Fn,k : On → Fn. It follows that
Fn,0 is a conditional expectation.
The existence of a unique KMS state on each Cuntz algebra was proved by Olesen
and Pedersen [15]. Firstly, we give some background on C∗-dynamical systems.
Definition 3.1. A C∗-dynamical system (A, R, ρ) consists of a C∗-algebra A and
a representation ρ : R → Aut(A), such that each map t 7→ ρt(a), a ∈ A, is norm
continuous.
C∗-dynamical systems can be defined in more general settings. In particular,
one can replace R with arbitrary locally compact groups.
Let Aa denote the dense ∗-subalgebra of A consisting of analytic elements, i.e.,
a ∈ Aa if the function t 7→ ρt(a) has a (necessarily unique) extension to an en-
tire operator-valued function. This extension is implicitly used in the following
definition.
4
TIM DE LAAT
Definition 3.2. Let (A, R, ρ) be a C∗-dynamical system. An invariant state φ on
A, i.e., a state for which φ ◦ ρt = φ for all t ∈ R, is a KMS state if
φ(ρt+i(a)b) = φ(bρt(a))
for all a ∈ Aa, b ∈ A and t ∈ R.
This definition is similar to the one introduced by Takesaki (see [20], Definition
13.1). It corresponds to φ being a β-KMS state for ρ−t with β = 1 according to
the conventions of [4] and [15]. In the latter, the following two results were proved
(see Lemma 1 and Theorem 2 therein). We restate these results slightly according
to the conventions of Definition 3.2.
t (Sk) = nitSk. Then ρn
Proposition 3.3. (Olesen-Pedersen) For all t ∈ R and the generators {Sk}n
k=1 of
On, define ρn
t extends uniquely to a ∗-automorphism of On
for every t ∈ R in such a way that (On, R, ρn) becomes a C∗-dymamical system.
Moreover, Fn is the fixed-point algebra of ρn in On, and Pn ⊂ (On)a.
Let τn = ⊗∞
k=1
1
n Tr denote the unique tracial state on Fn.
Proposition 3.4. (Olesen-Pedersen) For n ≥ 2, the C∗-dynamical system given
by (On, R, ρn) has exactly one KMS state, namely φn = τn ◦ Fn,0.
For a C∗-algebra A, let U(A) denote its unitary group. The following result was
proved by Archbold [1]. It implies the Dixmier property for On.
Proposition 3.5. (Archbold) For all x ∈ On,
φn(x)1On ∈ conv{uxu∗ u ∈ U(Fn)}
k.k
.
As a corollary, we obtain the following (well-known) fact (see also [7]).
Corollary 3.6. The relative commutant of Fn in On is trivial, i.e.,
(Fn)′ ∩ On = C1.
Proof. Let x ∈ (Fn)′ ∩On. By Proposition 3.5, we know that for every ε > 0, there
exists a finite convex combination Pm
i , where ui ∈ U(Fn), such that
kPm
i =
Pm
(cid:3)
i − φn(x)1Onk < ε. Since x ∈ (Fn)′ ∩ On, we have Pm
i = x. Hence, kx − φn(x)1Onk < ε. This implies that x ∈ C1.
i=1 λiuixu∗
i=1 λixuiu∗
i=1 λiuixu∗
i=1 λiuixu∗
Proposition 3.5 can be extended to finite sets in On, as described in the following
lemma, by similar methods as in [8], Part III, Chapter 5. For an invertible element
v in a C∗-algebra A, we define ad(v)(x) = vxv−1 for all x ∈ A.
Lemma 3.7. Let {x1, . . . , xk} be a subset of On, and let ε > 0. Then there exists
a convex combination α of elements in {ad(u) u ∈ U(Fn)} such that
kα(xi) − φn(xi)1Onk < ε
for all i = 1, . . . , k.
Moreover, there exists a net {αj}j∈J ⊂ conv{ad(u) u ∈ U(Fn)} such that
j kαj(x) − φn(x)1Onk = 0
lim
for all x ∈ On.
ON THE JCB GROTHENDIECK THEOREM
5
Proof. Suppose that kα′(xi) − φn(xi)1Onk < ε for i = 1, . . . , k − 1. By Proposition
3.5, we can find a convex combination α such that
k α(α′(xk)) − φn(α′(xk))1Onk < ε.
Note that φn(α′(xk)) = φn(xk) and 1On = α(1On). By the fact that k α(x)k ≤ kxk
for all x ∈ On, we conclude that α = α ◦ α′ satisfies kα(xi) − φn(xi)1Onk < ε for
i = 1, . . . , k.
Let J denote the directed set consisting of pairs (F, η), where F is a finite subset
of On and η ∈ (0, 1), with the ordering given by (F1, η1) (cid:22) (F2, η2) if F1 ⊂ F2 and
η1 ≥ η2. By the first assertion, this gives rise to a net {αj}j∈J with the desired
properties.
(cid:3)
4. Proof of the JCB Grothendieck Theorem
In this section, we explain the proof of the Grothendieck Theorem for jointly
completely bounded bilinear forms on C∗-algebras. As mentioned in Section 1, the
proof is along the same lines as the proof by Haagerup and Musat, but we tensor
with Cuntz algebras instead of type III factors.
Applying the GNS construction to the pair (On, φn), we obtain a ∗-representation
πn of On on the Hilbert space Hπn = L2(On, φn), with cyclic vector ξn, such that
φn(x) = hπn(x)ξn, ξniHπn . We identify On with its GNS representation. Note that
φn extends in a normal way to the von Neumann algebra O′′
n, which also acts on
Hπn . This normal extension is a KMS state for a W ∗-dynamical system with O′′
n as
the underlying von Neumann algebra (see Corollary 5.3.4 of [4]). The commutant
O′
n of On is also a von Neumann algebra, and using Tomita-Takesaki theory (see
[4], [20]), we obtain, via the polar decomposition of the closure of the operator
Sxξn = x∗ξn, a conjugate-linear involution J : Hπn → Hπn satisfying JOnJ ⊂ O′
n.
Lemma 4.1. For k ∈ Z, we have
n := {x ∈ On ρn
Ok
The proof of this lemma is analogous to Lemma 1.6 of [21]. Note that O0
t (x) = n−iktx∀t ∈ R} = {x ∈ On φn(xy) = n−kφn(yx)∀y ∈ On}.
n = Fn,
and that for all k ∈ Z, we have Ok
Lemma 4.2. For every k ∈ Z, there exists a ck ∈ On such that
n 6= {0}.
kck) = n
and, moreover, hckJckJξn, ξni = 1.
φn(c∗
k
2 ,
φn(ckc∗
k) = n− k
2 ,
The proof is similar to the proof of Lemma 2.1 of [12].
Proposition 4.3. Let A, B be C∗-algebras, and let u : A × B → C be a jointly
completely bounded bilinear form. There exists a bounded bilinear form u on
(A ⊗min On) × (B ⊗min JOnJ) given by
u(a ⊗ c, b ⊗ d) = u(a, b)hcdξn, ξni
for all a ∈ A, b ∈ B, c ∈ On and d ∈ JOnJ. Moreover, kuk ≤ kukjcb.
The C∗-algebra JOnJ is just a copy of On. This result is analogous to Pro-
position 2.3 of [12], and the proof is the same. Note that in our case, we use
kPk
i=1 cidikB(L2(On,φn)) = kPk
i=1 ci ⊗ dikOn⊗minJOnJ for all c1, . . . , ck ∈ On and
d1, . . . , dk ∈ JOnJ. This equality is elementary, since On is simple and nuclear. In
6
TIM DE LAAT
the proof of Haagerup and Musat, one takes the tensor product of A and a certain
type III factor M and the tensor product of B with the commutant M ′ of M ,
respectively. Note that JOnJ ⊂ O′
n.
One can formulate analogues of Lemma 2.4, Lemma 2.5 and Proposition 2.6
of [12]. They can be proved in the same way as there, and one explicitly needs
the existence and properties of KMS states on the Cuntz algebras (see Section 3).
The analogue of Proposition 2.6 gives the "transformation" of the JCB Grothen-
dieck Theorem to the noncommutative Grothendieck Theorem for bounded bilinear
forms.
Using Lemma 2.7 of [12], we arrive at the following conclusion, which is the
analogue of [12], Proposition 2.8.
Proposition 4.4. Let K(n) = q(n
2 )/2, and let u : A × B → C be a
jointly completely bounded bilinear form on C∗-algebras A, B. Then there exist
states f n
2 + n− 1
1
1 , gn
2 on A and gn
1 , f n
u(a, b) ≤ K(n)kukjcb (cid:16)f n
2 on B such that for all a ∈ A and b ∈ B,
2 (bb∗)
2 (a∗a)
1 (aa∗)
1 (b∗b)
2 + f n
2 gn
1
2(cid:17) .
1
2 gn
1
1
The above proposition is the JCB Grothendieck Theorem. However, the (uni-
versal) constant and states depend on n. This is because the noncommutative
Grothendieck Theorem gives states on A⊗min On and B ⊗min JOnJ, which clearly
depend on n, and these states are used to obtain the states on A and B. The
best constant we obtain in this way comes from the case n = 2, which yields the
constant K(2) = q(2
2 + 2− 1
1
2 )/2 ∼ 1.03.
5. The best constant
t ⊗ ρ3
In order to get the best constant K = 1, we consider the C∗-dynamical system
(A, R, ρ), with A = O2 ⊗ O3 and ρt = ρ2
t . It is straightforward to check that
it has a KMS state, namely φ = φ2 ⊗ φ3. It is easy to see that F = F2 ⊗ F3 is
contained in the fixed point algebra. (Actually, it is equal to the fixed point algebra,
but we do not need this.) These assertions follow by the fact that the algebraic
tensor product of O2 and O3 is dense in O2 ⊗ O3. Note that ρ is not periodic.
Applying the GNS construction to the pair (A, φ), we obtain a ∗-representation
π of A on the Hilbert space Hπ = L2(A, φ), with cyclic vector ξ, such that φ(x) =
hπ(x)ξ, ξiHπ . We identify A with its GNS representation. Using Tomita-Takesaki
theory, we obtain a conjugate-linear involution J : Hπ → Hπ satisfying JAJ ⊂ A′
(see also Section 4).
It follows directly from Proposition 3.5 that φ(x)1A ∈ conv{uxu∗ u ∈ U(F )}
for all x ∈ A. Also, the analogue of Lemma 3.7 follows in a similar way, as well as
the fact that F ′ ∩ A = C1.
It is elementary to check that
Aλ,k := {x ∈ A ρt(x) = λiktx∀t ∈ R} = {x ∈ A φ(xy) = λkφ(yx)∀y ∈ On}.
k.k
Let Λ := {2p3q p, q ∈ Z} ∩ (0, 1). For all λ ∈ Λ and k ∈ Z, we have Aλ,k 6= {0}.
This leads, analogous to Lemma 4.2, to the following result.
Lemma 5.1. Let λ ∈ Λ. For every k ∈ Z there exists a cλ,k ∈ A such that
φ(c∗
λ,kcλ,k) = λ− k
2 ,
φ(cλ,kc∗
λ,k) = λ
k
2
ON THE JCB GROTHENDIECK THEOREM
7
and
hcλ,kJcλ,kJξ, ξi = 1.
In this way, by the analogues of Lemma 2.4, Lemma 2.5 and Proposition 2.6 of
[12], we obtain the following result, which is the analogue of [12], Proposition 2.8.
be a jointly completely bounded bilinear form. Then there exist states f λ
and gλ
Proposition 5.2. Let λ ∈ Λ, and let C(λ) = q(λ
2 on B such that for all a ∈ A and b ∈ B,
u(a, b) ≤ C(λ)kukjcb (cid:16)f λ
1 (b∗b)
2 + λ− 1
1 (aa∗)
2 (a∗a)
2 + f λ
2 gλ
2 (bb∗)
1 , gλ
1
2(cid:17) .
1
2 gλ
2 )/2. Let u : A× B → C
2 on A
1 , f λ
1
1
1
Note that C(λ) > 1 for λ ∈ Λ. Let (λn)n∈N be a sequence in Λ converging to
+ and B∗
1. By the weak*-compactness of the unit balls (A∗
+,
respectively, the Grothendieck Theorem for jointly completely bounded bilinear
forms with K = 1 follows in the same way as in the "Proof of Theorem 1.1" in [12].
+)1 and (B∗
+)1 of A∗
Remark 5.3. By Kirchberg's second "Geneva Theorem" (see [14] for a proof), we
know that O2 ⊗ O3 ∼= O2. This implies that the best constant in Theorem 1.1 can
also be obtained by tensoring with the single Cuntz algebra O2, but considered
with a different action that defines the C∗-dynamical system. Since the explicit
form of the isomorphism is not known, we cannot adjust the action accordingly.
6. A remark on Blecher's inequality
In [2], Blecher stated a conjecture about the norm of elements in the algebraic
tensor product of two C∗-algebras. Equivalently, the conjecture can be formulated
as follows (see Conjecture 0.2′ of [18]). For a bilinear form u : A × B → C, put
ut(b, a) = u(a, b).
Theorem 6.1 (Blecher's inequality). There is a constant K such that any jointly
completely bounded bilinear form u : A × B → C on C∗-algebras A and B decom-
poses as a sum u = u1 + u2 of completely bounded bilinear forms on A × B, and
ku1kcb + kut
2kcb ≤ Kkukjcb.
A version of this conjecture for exact operator spaces and a version for pairs
of C∗-algebras, one of which is assumed to be exact, were proved by Pisier and
Shlyakhtenko [18]. They also showed that the best constant in Theorem 6.1 is
greater than or equal to 1. Haagerup and Musat proved that Theorem 6.1 holds
with K = 2 [12, Section 3]. We show that the best constant is actually strictly
greater than 1.
In the following, let OH(I) denote Pisier's operator Hilbert space based on ℓ2(I)
for some index set I. Recall the noncommutative little Grothendieck Theorem.
Theorem 6.2 (Noncommutative little Grothendieck Theorem). Let A be a C∗-
algebra, and let T : A → OH(I) be a completely bounded map. Then there exists
a universal constant C > 0 and states f1 and f2 on A such that for all a ∈ A,
kT ak ≤ CkTkcbf1(aa∗)
1
4 f2(a∗a)
1
4 .
For a completely bounded map T : A → OH(I), denote by C(T ) the smallest
constant C > 0 for which there exist states f1, f2 on A such that for all a ∈ A,
we have kT ak ≤ Cf1(aa)
In [12], Haagerup and Musat proved that
C(T ) ≤ √2kTkcb. Pisier and Shlyakhtenko proved in [18] that kTkcb ≤ C(T ) for
4 f2(a∗a)
4 .
1
1
8
TIM DE LAAT
all T : A → OH(I). Haagerup and Musat proved that for a certain T : M3(C) →
OH(3), the inequality is actually strict, i.e., kTkcb < C(T ) [13, Section 7]. We can
now apply this knowledge to improve the best constant in Theorem 6.1.
Theorem 6.3. The best constant K in Theorem 6.1 is strictly greater than 1.
Proof. Let A be a C∗-algebra, and let T : A → OH(I) be a completely bounded map
for which kTkcb < C(T ). Define the map V = T ∗JT from A to A∗ = A
, where
J : OH(I) → OH(I)∗ is the canonical complete isomorphism and T ∗ : OH(I)∗ → A∗
is the adjoint of T . Hence, V is completely bounded.
It follows that V = u
for some jointly completely bounded bilinear form u : A × A → C. Moreover,
kukjcb = kV kcb = kTk2
cb, where the last equality follows from the proof of Corollary
3.4 in [18]. By Blecher's inequality, i.e., Theorem 6.1, we have a decomposition
u = u1 + u2 such that ku1kcb + kut
By the second characterization of completely bounded bilinear forms (in the
2kcb ≤ Kkukjcb.
∗
Christensen-Sinclair sense) in Section 2, we obtain
u1(a, b) ≤ ku1kcbf1(aa∗)
1
2 g1(b∗b)
1
2 ,
u2(a, b) ≤ kut
2kcbf2(a∗a)
1
2 g2(bb∗)
1
2 .
It follows that
u(a, b) ≤ ku1kcbf1(aa∗)
1
2 g1(b∗b)
1
2 + kut
2kcbf2(a∗a)
1
2 g2(bb∗)
1
2 .
Let gi(a) = gi(a∗) for i = 1, 2, and define states
f = ku1kcbf1 + kut
ku1kcb + kut
2kcbg2
2kcb
and g = ku1kcbg1 + kut
ku1kcb + kut
2kcbf2
2kcb
.
We obtain
kT (a)k2 = u(a, a) ≤ ku1kcbf1(aa∗)
≤ (ku1kcbf1 + kut
≤ (ku1kcb + kut
Hence, ku1kcb + kut
2kcbg2)(aa∗)
2kcb) f (aa∗)
2 g(a∗a)
2kcb ≥ C(T )2 > kTk2
1
1
1
2 .
1
2 g1(a∗a)
2 (ku1kcbg1 + kut
1
2kcbf2(a∗a)
2 + kut
2kcbf2)(a∗a)
1
2
1
2 g2(aa∗)
1
2
cb = kukjcb. This proves the theorem.
(cid:3)
Acknowledgements
The question if a C∗-algebraic proof of Theorem 1.1 exists was suggested to me
by Uffe Haagerup and Magdalena Musat. I thank them for many useful comments.
References
[1] Archbold, R.J.: On the simple C ∗-algebras of J. Cuntz. J. London Math. Soc. (2) 21, 517 -- 526
(1980)
[2] Blecher, D.P.: Generalizing Grothendieck's program. In: Jaros, K. (ed.) Function spaces.
Lecture notes in pure and applied mathematics, vol. 136, pp. 45 -- 53. Marcel Dekker, Inc.,
New York (1992)
[3] Blecher, D.P., Paulsen, V.I.: Tensor products of operator spaces. J. Funct. Anal. 99, 262 -- 292
(1991)
[4] Bratteli, O., Robinson, D.W.: Operator Algebras and Quantum Statistical Mechanics 2.
Springer-Verlag, Berlin, (1997)
[5] Christensen, E., Sinclair, A.M.: Representations of completely bounded multilinear operators.
J. Funct. Anal. 72, 151 -- 181 (1987)
[6] Cuntz, J.: Simple C ∗-algebras generated by isometries. Comm. Math. Phys. 57, 173 -- 185
(1977)
ON THE JCB GROTHENDIECK THEOREM
9
[7] Cuntz, J.: Automorphisms of certain simple C ∗-algebras. In: Streit, L. (ed.) Quantum Fields
- Algebras, Processes, Proc. Sympos., Univ. Bielefeld, Bielefeld, 1978, pp. 187 -- 196. Springer,
Vienna, (1980)
[8] Dixmier, J.: Von Neumann Algebras. North-Holland Mathematical Library, vol. 27, North-
Holland, Amsterdam (1981)
[9] Effros, E., Ruan, Z.-J.: A new approach to operator spaces. Can. Math. Bull. 34, 329-337
(1991)
[10] Grothendieck, A.: R´esum´e de la th´eorie m´etrique des produits tensorielles topologiques.
Bol. Soc. Mat. Sao Paolo 8, 1 -- 79 (1953)
[11] Haagerup, U.: The Grothendieck inequality for bilinear forms on C ∗-algebras. Adv. Math. 56,
93-116 (1985)
[12] Haagerup, U., Musat, M.: The Effros-Ruan conjecture for bilinear forms on C ∗-algebras.
Invent. Math. 174, 139 -- 163 (2008)
[13] Haagerup, U., Musat, M.: Factorization and dilation problems for completely positive maps
on von Neumann algebras. Comm. Math. Phys. 303, 555 -- 594 (2011)
[14] Kirchberg, E., Phillips, N.C.: Embedding of exact C ∗-algebras in the Cuntz algebra O2.
J. Reine Angew. Math. 525, 17 -- 53 (2000)
[15] Olesen, D., Pedersen, G.K.: Some C ∗-dynamical systems with a single KMS state.
Math. Scand. 42, 111-118 (1978)
[16] Pisier, G.: Grothendieck's theorem for noncommutative C ∗-algebras, with an appendix on
Grothendieck's constants. J. Funct. Anal. 29, 397 -- 415 (1978)
[17] Pisier, G.: Grothendieck's theorem, past and present. Bull. Amer. Math. Soc. (N.S.) 49,
237 -- 323 (2012)
[18] Pisier, G., Shlyakhtenko, D.: Grothendieck's theorem for operator spaces, Invent. Math. 150,
185 -- 217 (2002)
[19] Regev, O, Vidick, T.: Elementary Proofs of Grothendieck Theorems for Completely Bounded
Norms. Preprint, arXiv:1206.4025 (2012)
[20] Takesaki, M.: Tomita's theory of modular Hilbert algebras and its applications. Lecture Notes
in Mathematics, Vol. 128 Springer-Verlag, Berlin, (1970)
[21] Takesaki, M.: The structure of a von Neumann algebra with a homogeneous periodic state.
Acta. Math. 131, 79 -- 121 (1973)
Department of Mathematical Sciences, University of Copenhagen,
Universitetsparken 5, DK-2100 Copenhagen Ø, Denmark
E-mail address: [email protected]
|
1905.07306 | 1 | 1905 | 2019-05-17T14:55:43 | Unbounded Derivations in Algebras Associated with Monothetic Groups | [
"math.OA"
] | Given an infinite, compact, monothetic group $G$ we study decompositions and structure of unbounded derivations in a crossed product C$^*$-algebra $C(G)\rtimes\Z$ obtained from a translation on $G$ by a generator of a dense cyclic subgroup. We also study derivations in a Toeplitz extension of the crossed product and the question whether unbounded derivations can be lifted from one algebra to the other. | math.OA | math | UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH
MONOTHETIC GROUPS
SLAWOMIR KLIMEK AND MATT MCBRIDE
Abstract. Given an infinite, compact, monothetic group G we study decompositions and
structure of unbounded derivations in a crossed product C∗-algebra C(G) ⋊ Z obtained from
a translation on G by a generator of a dense cyclic subgroup. We also study derivations in a
Toeplitz extension of the crossed product and the question whether unbounded derivations
can be lifted from one algebra to the other.
1. Introduction
Derivations naturally arise in studying differentiable manifolds, in representation theory
of Lie groups and in their noncommutative analogs. They also appear in mathematical
aspects of quantum mechanics, in particular in quantum statistical physics. Additionally,
derivations are important in analyzing amenability and other structures of operator algebras.
Good overviews are in
B
[1] and also in
S
[14].
9
1
0
2
y
a
M
7
1
]
.
A
O
h
t
a
m
[
1
v
6
0
3
7
0
.
5
0
9
1
:
v
i
X
r
a
In this paper we study classification and decompositions of unbounded derivations in C∗-
algebras associated to an infinite, compact, monothetic group G, which, by definition, is
a Hausdorff topological group with a dense cyclic subgroup. A group translation on G by
a generator of a cyclic subgroup is a minimal homeomorphism and one algebra associated
with G is the crossed product C∗-algebra B := C(G) ⋊ Z determined by the translation.
This algebra can be naturally represented in the ℓ2-Hilbert space of the full orbit.
If we
consider the analogous algebra on the forward orbit only, we obtain a Toeplitz extension
A of the algebra B. When the group is totally disconnected those algebras are precisely
Bunce-Deddens and Bunce-Deddens-Toeplitz algebras considered in
KMRSW2
[9].
The main objects of study in this paper are unbounded derivations d : A → A which are
defined on a subalgebra A of polynomials in generators of A. Similarly, we study derivations
δ : B → B, where B is the image of A under the quotient map A → A/K = B. The first
of the main results of this paper is that any derivation in those algebras can be uniquely
decomposed into a sum of a certain special derivation and an approximately inner derivation.
The special derivations are not approximately inner, and can be explicitly described.
It turns out that any derivation d : A → A preserves the ideal of compact operators K and
consequently defines a factor derivation [d] : B → B in B. It is an interesting and non-trivial
problem to describe properties of the map d 7→ [d]. For any C∗-algebra it is easy to see that
bounded derivations preserve closed ideals and so they define derivations on quotients. It
P
[12] that for bounded derivations and separable C∗-algebras the above map is
was proved in
onto, i.e. derivations can be lifted from quotients. In non-separable cases this is not true in
general. We prove here that lifting unbounded derivations from B to A is always possible
KMRSW2
[9]. However we give a
when G is totally disconnected, answering positively a conjecture in
Date: May 20, 2019.
1
2
KLIMEK AND MCBRIDE
simple counterexample of a special derivation in the algebra B for G = T1 that cannot be
lifted to a derivation in the algebra A. Instead, we conjecture that for any compact, infinite,
monothetic group approximately inner derivations in B can be lifted to approximately inner
derivation in A.
The paper is organized as follows. In section 2 we review monothetic groups and discuss
their properties. We also describe a crossed product C∗-algebra that is associated to a
monothetic group and that algebra Toeplitz extension, as well as discuss a Toeplitz map from
one algebra to another. In section 3 we classify all unbounded derivations on polynomial
domains in the C∗-algebras from section 2. Finally, in section 4 we consider lifting derivations
from a crossed product C∗-algebra to its Toeplitz extension. We prove that all derivations
can be lifted for totally disconnected, compact, infinite, monothetic groups and provide an
example that shows that not all derivations can be lifted in general.
2. Monothetic Groups and Associated C∗-algebras
2.1. Monothetic Groups. A topological (Hausdorff) group is called monothetic if it has
M
a dense cyclic subgroup. Andr´e Weil observed, Theorem 19 of
[11], that if G is a locally
compact monothetic group, then G ∼= Z or G is compact. In this paper we only consider the
case of compact G. It follows immediately that G is Abelian and separable. We first describe
HS
the structure of such groups following
[5]. The key tool is the character (dual) group and
Pontryagin duality, which translates properties of groups into properties of their duals.
Let S1 be the unit circle:
S1 = {z ∈ C : z = 1},
discrete.
and let bG denote the dual group G, the group of continuous homomorphisms from G to
S1 equipped with compact-open topology. It is well known that if G is compact then bG is
We typically use additive notation for an abelian group, however we use multiplicative
notation for the dual group. Given a monothetic group G, let x1 be a generator of a dense
cyclic subgroup, and we set xn = nx1 for n ∈ Z, so that x0 := 0 is the neutral element of
To better understand the structure of monothetic groups we look at the torsion subgroup
G. Then we can identify the dual group bG of G with a discrete subgroup of S1 via the map
given by:
Conversely, using Pontryagin duality, if H is a discrete subgroup of S1, then H is the dual
bG ∋ χ 7→ χ(x1) ∈ S1.
group of a compact monothetic group, namely bH, see
of its dual group. Given a monothetic G, the torsion subgroup of bGtor of bG is given by:
There are two extreme cases: we say bG is of pure torsion if bG = bGtor. We also say bG is torsion
free if bGtor = {0}. The following statements describe basic properties of monothetic groups.
First we look at the case of torsion free bG.
We provide short or outlined proofs with references. A good, concise book on Pontryagin
duality is
bGtor = {χ ∈ bG : χn = 1 for some n ∈ N}.
HS
[5].
M
[11].
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
3
Proposition 2.1. Let G be a compact monothetic group. G is connected if and only if bG is
torsion free.
M
[11], which only requires G to be compact,
(cid:3)
Proof. This is Corollary 4 of Theorem 30 of
Abelian.
We have the following remarkable result proved in
HS
[5].
con_com_sep_mono Theorem 2.2. Every connected compact separable Abelian topological group is monothetic.
The n-dimensional torus, Tn = Rn/Zn is an example of a compact, connected, separable,
Abelian group and thus by Theorem
con_com_sep_mono
2.2 is monothetic. Consider an element
of Tn. Then the cyclic subgroup generated by x1 is dense in Tn if and only if {1, θ1, . . . , θn}
are linearly independent over Z, see for example
KH
[6].
x1 = (θ1, . . . , θn)
Proposition 2.3. Let G be a compact monothetic group. G is totally disconnected if and
Next we consider the case when bG is of pure torsion.
only if bG is of pure torsion.
element of the discrete group bG is compact (i.e. the smallest closed subgroup containing it
Proof. This result follows for example from Corollary 1 of Theorem 30 of
is compact) if and only if it has finite order.
Before we state the next structural result we need to introduce odometers. Further details
D
[3]. The standard definition of an odometer (that inspired the
on odometers can be found in
name) uses a sequence of positive integers b := (bm)m∈N such that bm ≥ 2 for all m, called a
multibase. The odometer is then identified (as a set) with the direct product:
M
[11], since an
(cid:3)
G(b) :=Ym
Z/bmZ,
but addition is defined with the carry over rule. Equipped with the product topology G(b)
becomes a compact, totally disconnected topological group. It is easy to see that the cyclic
subgroup generated by x1 = (1, 0, 0, 0, . . .) is dense and so G(b) is a monothetic group.
An alternative representation of the odometer G(b) uses scales, and this is the description
lift_theo
that is used in the proof of Theorem
4.3. Let s = (sm)m∈N be a sequence of positive integers
such that sm divides sm+1 and sm < sm+1. There are natural homomorphisms between the
consecutive finite cyclic groups Z/sm+1Z → Z/smZ, namely congruence modulo sm. Thus
the inverse limit:
Gs = lim
←−
m∈N
Z/smZ
is well defined as the subset of the countable product Qm Z/smZ consisting of sequences
(y1, y2, y3, . . .) such that ym+1 ≡ ym (mod sm). Addition in this representation is coordinate-
wise, modulo sm in each coordinate m. Gs becomes a topological group when endowed
with the product topology over the discrete topologies in Z/smZ. Obviously, with our
assumptions, this group is infinite because s is unbounded.
4
KLIMEK AND MCBRIDE
The relation between the two definitions of an odometer is as follows. Given a multibase
b = (bm)m∈N define a scale s = (sm)m∈N by s1 = b1, s2 = b1g2, s3 = b1b2b3 and so on.
Equivalently, we have:
Then the map
s1 = b1, bn =
sn
sn−1
for n > 1.
G(b) ∋ (k1, k2, k3, . . .) 7→ (k1, k1 + k2b1, k1 + k2b1 + k3b1b2, . . . ) ∈ Gs
gives an isomorphism of the groups. In the scales representation of odometers the generator
x1 of a cyclic subgroup is given by x1 = (1, 1, 1, 1, . . .).
With the above definitions it is not transparent when two odometers are isomorphic, so
KMRSW2
[9]. A supernatural number
we describe yet another way to define odometers that we used in
N is defined as the formal product:
N = Yp−prime
pǫp,
ǫp ∈ {0, 1, · · · , ∞}.
IfP ǫp < ∞ then N is said to be a finite supernatural number (a regular natural number),
otherwise it is said to be infinite. If
is another supernatural number, then their product is given by:
pǫ′
p
N ′ = Yp−prime
NN ′ = Yp−prime
pǫp+ǫ′
p.
A supernatural number N is said to divide M if M = NN ′ for some supernatural number
N ′, or equivalently, if ǫp(N) ≤ ǫp(M) for every prime p.
Given a supernatural number N let JN be the set of finite divisors of N:
JN = {j : jN, j < ∞}.
Then (JN , ≤) is a directed set where j1 ≤ j2 if and only if j1j2N. Consider the collection
of cyclic groups {Z/jZ}j∈JN
and the family of group homomorphisms
satisfying
πij : Z/jZ → Z/iZ,
j ≥ i
πij(z) = z (mod i)
πik = πij ◦ πjk for all i ≤ j ≤ k.
Then the inverse limit of this system can be denoted as:
Z/NZ := lim
←−
j∈JN
Z/jZ =((zj) ∈ Yj∈JN
Z/jZ : πij(zj) = zi) .
In particular, if N is finite the above definition coincides with the usual meaning of the
symbol Z/NZ, while if N = p∞ for a prime p, then the above limit is equal to Zp, the ring
of p-adic integers, see for example
Robert
[13].
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
5
Given a scale s = (sm)m∈N we define the corresponding supernatural number N to be the
"limit" of sm:
N = lim
m→∞
sm,
(2.1)
N_limit
in the sense that each prime exponent ǫp(N) of N is defined to be the supremum of the prime
exponents ǫp(sm), m ∈ N. It follows that sm's are divisors of N and for every j ∈ JN there
Z/jZ is
is a natural number m(j) such that jsm(j). Consequently, a sequence (zj) ∈ lim
←−
j∈JN
completely determined by the subsequence (zsm) ∈ lim
←−
m∈N
Z/smZ, which gives an isomorphism
Z/NZ ∼= Gs. It turns out that odometers are classified by the supernatural number N, see
D
[3]. As before,
generates a dense cyclic subgroup.
x1 = (1, 1, 1, 1, . . .) ∈ Z/NZ
In general, we have the following simple consequence of the Chinese Reminder Theorem:
if N = Qp−prime
ǫp6=0
pǫp, then
Z/NZ ∼= Yp−prime
ǫp6=0
Z/pǫpZ.
Since the space Z/NZ is a compact, abelian topological group, it has a unique normalized
Haar measure µ. Also, if N is an infinite supernatural number then Z/NZ is a Cantor set
W
[15].
We are now ready to state the next structural result about compact monothetic groups.
tot_disc_mono Proposition 2.4. G is a compact, totally disconnected, monothetic group if and only if it
is an odometer. In particular, there exist a supernatural number N such that
G ∼= Z/NZ ∼=Yi
Z/pǫi
i Z,
where N =Q∞
i=1 pǫi
i .
HS
Proof. Let G be a compact totally disconnected monothetic group. In
[5], between Theorems
II ′ and III on pages 256-257, the authors show that G is isomorphic to a direct product of
groups Gpi where pi runs over all primes and where Gpi isomorphic to the zero group, the
cyclic group of order pǫi
for some ǫi or the group of p-adic integers, the last case corresponds
i
to ǫi = ∞.
(cid:3)
In general, for arbitrary bG we have the following structure for compact monothetic groups.
Proposition 2.5. Let G be a compact monothetic group. If G0 ≤ G is the connected com-
ponent of the neutral element 0, then G0 is a connected separable compact Abelian group and
G/G0 is a totally disconnected monothetic group. Moreover, there are natural isomorphisms:
\(G/G0) ∼= bGtor and cG0
∼= bG/bGtor
6
KLIMEK AND MCBRIDE
Proof. This proposition is not formally stated but appears as a note in
3 of Theorem 30 of
the annihilator A(G0) of G0 is given by:
HS
[5], see also Corollary
M
[11]. The first part follows from the previous propositions. Recall that
By Pontryagin duality, Theorem 27 of
A(G0) = {χ ∈ bG : χ(g) = 1, for all g ∈ G0}.
M
[11], we have:
A(G0) ∼= \(G/G0).
Notice the right-hand side of the equation is Abelian, discrete and of pure torsion. Thus
given χ ∈ A(G0), it defines a character class
Therefore, we have:
[χ] : G/G0 → S1.
[χ] ∈ \(G/G0) = \(G/G0)tor,
hence χ has finite order and thus A(G0) ≤ bGtor. Let χ ∈ bGtor, then
since G0 is connected and thus χ ∈ A(G0). Therefore we have A(G0) = bGtor and hence
The second isomorphism relation follows from Pontryagin duality:
χG0 ∈ (cG0)tor = {1}
\(G/G0) ∼= bGtor.
∼= bG/A(G0)
cG0
and the proof is complete.
(cid:3)
2.2. Minimal Systems. By a topological dynamical system (X, ϕ), we mean a topological
KH
space X and a continuous map ϕ : X → X, see
[6]. A topological dynamical system (X, ϕ)
is called topologically transitive if there exists a point x ∈ X such that its orbit {ϕn(x)}n∈Z
is dense in X. (X, ϕ) is called minimal if every orbit is dense in X. We say and write ϕ is
minimal for brevity.
Other equivalent characterization of minimal maps is as follows. A set A ⊆ X is called
ϕ-invariant if ϕ(A) ⊆ A. Then, ϕ is minimal if X does not contain any non-empty, proper,
closed ϕ-invariant subset. If in addition X is assumed to be Hausdorff and compact, then a
minimal map ϕ must be surjective. Moreover, if (X, ϕ) is topologically transitive then there
is no ϕ-invariant nonconstant continuous function on X.
Suppose that G is a compact monothetic group with x1 the generator of a dense cyclic
subgroup. Then we define the map ϕ : G → G by the formula:
ϕ(x) = x + x1.
It follows that (G, ϕ) is a minimal system. Let us remark that for metrizable spaces a
minimal, equicontinuous, dynamical systems coincide with translations by a generator of a
dense cyclic subgroup of a compact monothetic groups, see Theorem 2.4.2 in
K
[10].
We now turn our attention to the algebras that are present in this paper. Let G be a
compact infinite monothetic group, C(G) the complex-valued continuous functions on G
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
7
and µ a normalized Haar measure on G. Recall the notation for the elements of the cyclic
subgroup generated by x1:
xn = ϕn(0) = nx1,
for n ∈ Z. The set {xn}n∈Z is the full orbit of 0 under ϕ and {xn}n≥0 is the forward orbit. As
mentioned above, since ϕ is a minimal homeomorphism, the forward orbit {xn}n≥0 is dense
in G.
Consider the algebra of trigonometric polynomials on G:
F =(Xn
cnχn : χn ∈ bG, finite sums) .
We state below two simple but useful properties of F that we will need later in the paper.
First we have the following observation.
property_F Proposition 2.6. Let f ∈ F . Then RG f dµ = 0 if and only if there is a trigonometric
polynomial g ∈ F such that
f = g ◦ ϕ − g.
Proof. If f ∈ F then f has the following decomposition:
where χj are characters on G. Notice that we have
cjχj,
f =Xj
χj dµ =(cid:26) 1
ZG
if χj = 1
0 else,
which means thatRG f dµ = 0 if and only if χj 6= 1 for all j. Let χ be a nontrivial character,
then the goal is to find a g ∈ F such that
χ(x) = g(x + x1) − g(x).
(2.2)
cocyc
Notice that for a nontrivial character we must have χ(x1) 6= 1. Otherwise, if χ(x1) = 1, then
χ(xn) = χ(nx1) = 1 which in turn implies that χ = 1 on a dense set, and thus χ ≡ 1, which
is a contradiction. Therefore, we can choose
g(x) =
χ(x)
χ(x1) − 1
,
cocyc
which clearly satisfies (
2.2) for a
nontrivial character, we just take finite linear combinations of such functions for the general
case of a trigonometric polynomial, thus completing the proof.
(cid:3)
cocyc
2.2). Now that we can find a function g(x) that solves (
Next we describe another useful property of the space F .
property_G Proposition 2.7. For any nonzero n ∈ Z, there exists a trigonometric polynomial f ∈ F
such that
f − f ◦ ϕn 6= 0.
8
KLIMEK AND MCBRIDE
Proof. The key property of the characters is that they separate points of G, see Theorem 14
of
M
[11]. Therefore, if n 6= 0, we can pick χ such that:
χ(x1) 6= χ(xn+1) = χ(ϕn(x1)).
As in the previous proposition, the general case is handled by linearity and the proof is
complete.
(cid:3)
2.3. C∗-algebras. Let G be an infinite, compact, monothetic group. We will describe now
two types of C∗-algebras that can be naturally associated with such groups. They are defined
as concrete C∗-algebras of operators in the following Hilbert spaces. The first Hilbert space
is the ℓ2 space of the full orbit:
H = ℓ2({xl}l∈Z),
which is naturally isomorphic with ℓ2(Z). Let {El}l∈Z be the canonical basis in H. The
second Hilbert space is the ℓ2 space of the forward orbit:
H+ = ℓ2({xk}k∈Z≥0)
which is naturally isomorphic with the Hilbert space ℓ2(Z≥0). We also let {E+
canonical basis on H+.
k }k∈Z≥0 be the
The C∗-algebras associated to G are defined using the following operators. Let V : H → H
be the shift operator on H:
V El = El+1.
We also need the unilateral shift operator on H+:
k = E+
UE+
k+1.
Notice that V is a unitary while U is an isometry. We have:
[U ∗, U] = P0,
where P0 is the orthogonal projection onto the one-dimensional subspace spanned by E+
0 .
For a continuous function f ∈ C(G) we define two operators Mf : H → H and M +
f
H+ → H+ via formulas:
:
Mf El = f (xl)El and M +
f E+
k = f (xk)E+
k .
They are diagonal multiplication operators on H and H+ respectively. Due to the density
of the orbit {xk}k∈Z≥0, we immediately obtain:
kMf k = kM +
f k = sup
l∈Z
f (xl) = sup
k∈Z≥0
f (xk) = sup
x∈G
f (x) = kf k∞.
The algebras of operators generated by Mf 's or by M +
f 's are thus isomorphic to C(G) so they
carry all the information about the space G, while operators U and V reflect the dynamics
ϕ on G. The relation between those operators is:
Similarly we have:
V Mf V −1 = Mf ◦ϕ.
UM +
f = M +
f ◦ϕU.
There is also another, less obvious relation between U and M +
f 's, namely:
M +
f P0 = P0M +
f = f (x0)P0.
(2.3)
M_with_P_zero
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
9
We define the algebra B to be the C∗-algebra generated by operators V and Mf :
We claim that B is isomorphic with the crossed product algebra:
B = C ∗{V, Mf : f ∈ C(G)}.
B ∼= C(G) ⋊ϕ Z.
Indeed, observe that Z is amenable, the action of Z on G given by ϕ is a free action,
and ϕ is a minimal homoemorphism, thus the crossed product is simple and equal to the
F
[4]. Clearly, the operators V and Mf define a representation of
reduced crossed product, see
C(G) ⋊ϕ Z, which must be isomorphic to it, by simplicity of the crossed product.
The algebra B has a natural dense ∗-subalgebra B of polynomials in V , V −1, and the
Mχ's, where χ is a character of G. Equivalently, we have:
B =(Xn
V nMfn : fn ∈ F , finite sums) .
Next we define the other algebra that is of the main interest in this paper, a Toeplitz
extension of B. We define the algebra A to be the C∗-algebra generated by operators U and
M +
f :
A = C ∗{U, M +
f : f ∈ C(G)}.
To proceed further we need the following label operators on H and H+ respectively:
LEl = lEl and KE+
k = kE+
k .
The algebra A has a natural dense ∗-subalgebra A of polynomials in U, U ∗, M +
χ 's, where
χ is a character of G, which can be equivalently described as follows, using Proposition 3.1
from
KMRSW1
[8] and also Proposition
n_cov_set_eq
2.11 below:
A =(Xn≥0
) +Xn≥1
U n(a+
n (K) + M +
f +
n
(a−
n (K) + M +
f −
n
)(U ∗)n : f ±
n ∈ F , a±
n (k) ∈ c00(Z≥0)) ,
where the sums above are finite sums and c00(Z≥0) is the space of sequences that are eventu-
ally zero. Notice that if a ∈ A and x ∈ c00(Z≥0) ⊆ H+, then ax ∈ c00(Z≥0), an observation
that is often used below.
Next we establish the key relation between the two algebras A and B. Let P+ : H → H+
be the following map from H onto H+ given by
P+Ek =( E+
0
k
if k ≥ 0
if k < 0.
We also need another map s : H+ → H given by:
sE+
k = Ek.
Define the map T : B(H) → B(H+), between the spaces of bounded operators on H and
H+, in the following way: given b ∈ B(H)
T is known as a Toeplitz map. It has the following properties.
alg_props_T Proposition 2.8. Let T be the Toeplitz map defined above. Then:
T (b) = P+bs.
10
KLIMEK AND MCBRIDE
(1) T (IH) = IH+.
(2) T (bV n) = T (b)U n and T (V −nb) = (U ∗)nT (b) for n ≥ 0 and all b ∈ B(H).
(3) T (bMf ) = T (b)M +
f and T (Mf b) = M +
f T (b) for all f ∈ C(X) and all b ∈ B(H).
Consequently, it follows that T maps B to A and B to A.
Proof. For the first statement, if h ∈ H+ then we have the following calculation:
T (IH)h = (P+sIH)h = P+h = h = IH+h.
For the second statement we apply T (bV n) to the basis elements E+
k ) = (P+b)Ek+n = (P+bs)U nE+
k = (P+bV n)s(E+
T (bV n)E+
k of H+. We have
k = T (b)U nE+
k .
A similar calculation shows the other equality T (V −nb) = (U ∗)nT (b). Finally, for the last
statement, we apply T (bMf ) and T (Mf b) to the basis elements to get:
T (bMf )E+
k = (P+bMf )s(E+
k ) = (P+b)Mf Ek = (P+b)f (xk)Ek = (P+bs)M +
f E+
k
= T (b)M +
f E+
k .
This completes the proof.
(cid:3)
The next result describes the main relation between the two algebras A and B.
Proposition 2.9. The ideal of compact operators K in B(H+) is an ideal in A. Moreover,
B is the factor algebra:
B ∼= A/K
and
is an isomorphism.
Proof. Notice first that we have:
B ∋ b 7→ T (b) + K ∈ A/K
(2.4)
T_iso
P0 = I − UU ∗ ∈ A.
It follows that the operators Pk,l := U kP0(U ∗)l are also in A. Thus, all finite rank operators
with respect to the basis {E+
k } belong to A as they are finite linear combinations of Pk,l.
Moreover, since all compact operators in B(H+) are norm limits of these finite rank operators
and A is a C∗-algebra, it follows that K ⊆ A. It is clear that K is an ideal in A. Verifying that
T_iso
2.4) is an isomorphism, is analogous to the proof of Theorem
the map given by equation (
2.3 in
(cid:3)
KMRS
[7].
It follows from the two previous propositions that we have the following identification.
structure_cor Corollary 2.10. Under the isomorphism given by the equation (
T_iso
2.4), B is the factor algebra:
B ∼= [A].
For future reference we notice the following formulas:
[U] = V, [M +
f ] = Mf , [K] = L,
and also, for every b ∈ B:
[T (b)] = b.
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
11
Useful tools in classifying derivations on A and B are 1-parameter groups of automor-
phisms of A and B respectively that are given by the following equations:
θ (a) = eiθKae−iθK
ρK
ρL
θ (b) = eiθLbe−iθL
for a ∈ A
for b ∈ B,
where θ ∈ R/2πZ. We have the following formulas:
ρK
θ (U) = eiθU, ρK
θ (a(K)) = a(K),
and similarly for ρL
θ . It immediately follows that ρK
θ : A → A and that ρL
θ : B → B.
The automorphisms define natural Z-gradings on A and B given by the spectral subspaces:
An = {a ∈ A : ρK
Bn = {b ∈ B : ρL
θ (a) = einθa}
θ (a) = einθb}.
We call the elements of these sets the n-covariant elements of A and B respectively. When
n = 0 we call those elements invariant.
Let c0(Z≥0) be the space of sequences that converge to zero. The n-covariant elements of
A and B can described in detail.
n_cov_set_eq Proposition 2.11. We have the following set equalities:
An = {a ∈ A : a = U n(an(K) + M +
f ), a(k) ∈ c0(Z≥0), f ∈ C(G)}
for n ≥ 0 and
An = {a ∈ A : a = (a(K) + M +
f )(U ∗)−n, a(k) ∈ c0(Z≥0), f ∈ C(G)}
when n < 0. Similarly, we have:
if n ≥ 0, and
for n < 0.
Bn = {b ∈ B : b = V nMf , f ∈ C(G)},
Bn = {b ∈ B : b = Mf V n, f ∈ C(G)}
θ (a) = a. It follows from the definition
θ that these elements are precisely the diagonal operators in A. Moreover, we have the
Proof. Consider the invariant elements in A, that is ρK
of ρK
following unique decomposition, which is analogous to Proposition 2.4 in
KMRSW2
[9]:
a = a(K) + M +
f
where a(k) ∈ c0(Z≥0) and f ∈ C(G). Next we consider the n-covariant elements for n 6= 0.
Without loss of generality we only consider n > 0. Since we have:
ρK
θ (U n) = einθU n
ρK
θ (a(K) + M +
f ) = a(K) + M +
f
for a(k) ∈ c0(Z≥0) and f ∈ C(G), one containment follows immediately. On the other hand,
if a ∈ An then a(U ∗)n is an invariant element and thus by the above has the form
a(U ∗)n = a(K) + M +
f
for some a(k) ∈ c0(Z≥0) and f ∈ C(G). The other direction now follows. The same argument
also works for Bn, completing the proof.
(cid:3)
12
KLIMEK AND MCBRIDE
Similarly, we consider n-covariant elements from A and B:
An = {a ∈ A : ρK
n_cov_set_eq
2.11, a ∈ An if and only if a has the same element decomposition but with
θ (a) = einθa} and Bn = {b ∈ B : ρL
θ (b) = einθb}.
As in Proposition
a(k) ∈ c00(Z≥0) and f ∈ F . Again, there is an analogous result for b ∈ Bn.
3. Classification of Derivations
As in
KMRSW2
[9], one of the main goals in this paper is to classify unbounded derivations in A and
B. We begin with recalling the basic concepts.
Let M be a Banach algebra and let M be a dense subalgebra of M. A linear map
d : M → M is called a derivation if the Leibniz rule holds:
d(ab) = ad(b) + d(a)b
for all a, b ∈ M. We say a derivation d : M → M is inner if there is an element x ∈ M such
that
d(a) = [x, a]
for a ∈ M. We say a derivation d : M → M is approximately inner if there are xn ∈ M
such that
for a ∈ M.
Given n ∈ Z, a derivation d : A → A is said to be a n-covariant derivation if the relation
d(a) = lim
n→∞
[xn, a]
(ρK
θ )−1d(ρK
θ (a)) = e−inθd(a)
holds. We have a similar definition for a derivation δ : B → B. Like above, when n = 0 we
say the derivation is invariant.
3.1. Derivations in A. We first classify all invariant derivations d : A → A. An example
of an invariant derivation is given by
dK(a) = [K, a]
where a ∈ A. This derivation is well defined because A is the space of polynomials in U,
U ∗, and M +
f and we have [K, U] = U, [K, U ∗] = −U ∗ and [K, M +
f ] = 0.
der_alpha Lemma 3.1. For any {α(k)} ∈ c0(Z≥0) there exists a unique derivation dα : A → A such
that
dα(U) = Uα(K),
dα(U ∗) = −α(K)U ∗,
dα(a(K)) = 0
for every a(K) ∈ A0. Moreover this derivation is an approximately inner invariant deriva-
tion.
Proof. Define a sequence {αN (k)} as follows:
αN
0 (k) =(cid:26) α0(k)
0
for k < N
for k ≥ N.
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
13
Then αN (k) ∈ c00(Z≥0) and αN (K) converges to α(K) as N → ∞. Next, define a sequence
{βN (k)} by
so that βN (k) is eventually constant. Thus, dαN : A → A defined by
βN (k) =
αN (j),
k−1Xj=0
is an invariant inner derivation. We have
dαN (a) = [βN (K), a]
lim
N→∞
dαN (U) = Uα(K) ,
lim
N→∞
dαN (U ∗) = −α(K)U ∗ and dαN (a(K)) = 0
for all a(K) ∈ A0. Thus, by the Leibniz rule, the limit
lim
N→∞
dαN (a) = dα(a)
exists for all a ∈ A. Thus, this limit is a derivation from A to A.
approximately inner and invariant.
It follows that dα is
(cid:3)
der_f_zero Lemma 3.2. For any f ∈ C(G) such that
ZG
f dµ = 0,
there exists a unique derivation df : A → A such that
df (U ∗) = −M +
df (U) = UM +
f ,
f U ∗,
df (a(K)) = 0
where a(K) ∈ A0. Moreover df is an approximately inner invariant derivation.
Proof. By the density of F we can pick a sequence {f N } ⊆ F such that f N converges to f
and
By Proposition
We define
property_F
2.6, there exists a sequence {gN } ⊆ F such that
ZG
f N dµ = 0.
f N (x) = gN (ϕ(x)) − gN (x).
df N (a) = [M +
gN , a],
and notice that df N : A → A is an inner invariant derivation. By direct calculation we have
lim
N→∞
df N (U) = UM +
f ,
lim
N→∞
df N (U ∗) = −M +
f U ∗, and df N (a(K)) = 0,
for every a(K) ∈ A0. Thus, by the Leibniz rule, the limit
exists for all a ∈ A and is a derivation from A to A. It follows that df is approximately
inner and invariant.
(cid:3)
lim
N→∞
df N (a) = df (a)
14
KLIMEK AND MCBRIDE
invariant_der_A Proposition 3.3. Given any invariant derivation d : A → A there exists a number c0 ∈ C
such that d is of the unique form
where d : A → A is approximately inner.
d(a) = c0dK(a) + d(a)
Proof. Let a(K) ∈ A0 be a diagonal operator such that a(k) ∈ c00(Z≥0). Then, by invariance
of d, we have d(a(K)) ∈ A0. Notice that since A0 is precisely the algebra of diagonal operators
in A it is therefore a commutative algebra. Let P 2 = P be a projection in A0. Applying d
to both sides of the equation and using Leibniz's rule we have
2P d(P ) = d(P 2) = d(P )
which implies that (1 − 2P )d(P ) = 0 and hence d(P ) = 0. Since a(K) is a finite sum of
projections in A0, it follows that d(a(K)) = 0.
Let Pk be the one-dimensional orthogonal projection onto the span of Ek. Then Pk ∈ A0
and thus d(Pk) = 0. We have the following formula:
Therefore, applying d to both sides, we obtain:
M +
f Pk = f (xk)Pk.
d(M +
f )Pk + M +
f d(Pk) = f (xk)d(Pk).
It follows that d(M +
n_cov_set_eq
2.11 that d(a) = 0 for all a ∈ A0.
f )Pk = 0 for all k ∈ Z≥0 and so, d(M +
f ) = 0. It follows from Proposition
Notice that, by the invariance property of d, we have
d(U) = U(α0(K) + M +
f0)
d(U ∗) = −(α0(K) + M +
f0)U ∗
for some α0(k) ∈ c0(Z≥0) and f0 ∈ C(G).
(3.1)
d_inv_decomp
Let c0 be the following integral
and set f = f0 − c0 so thatRG
the decomposition
c0 =ZG
f0 dµ.
d = c0dK + dα0 + d f .
f dµ = 0. By Lemmas
der_alpha
3.1 and
der_f_zero
3.2 and equation (
d_inv_decomp
3.1), we have
Picking d = dα0 + d f completes the proof of existence of the decomposition.
Finally, to verify uniqueness of the decomposition, we only need to check that that dK
is not approximately inner. If dK is approximately inner then we can arrange that it can
be approximated by inner invariant derivations of the form dj(a) = [βj(K), a] with βj(k) ∈
c(Z≥0). Since {βj(k + 1) − βj(k)} ∈ c0(Z≥0) we would also get {(k + 1) − k} ∈ c0(Z), which
KMRSW2
is a contradiction. Full details of an analogous result are given in Theorem 3.10 of
[9]. (cid:3)
Next we classify n-covariant derivations in A.
property_G
2.7 choose f ∈ F such that f − f ◦ ϕn 6= 0. Since
Remark: Let n 6= 0 and by Proposition
f − f ◦ ϕn is continuous on a compact set, the minimum is achieved and is not equal to zero.
Therefore we have
f (x) − f (ϕn(x)) 6= 0
inf
x∈G
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
15
and hence M +
f − M +
the next proposition.
f ◦ϕn is an invertible operator. This remark is crucial for the proof of
covariant_der_A Proposition 3.4. Let d : A → A be an n-covariant derivation where n 6= 0. There exists
an β(K) ∈ A0 such that
d(a) = [U nβ(K), a]
d(a) = [β(K)(U ∗)−n, a]
for n > 0
for n < 0,
and hence d is an inner derivation.
Proof. We only discuss the case of n > 0 as the case of n < 0 is completely analogous. By
definition of n-covariance there exists an α(K) ∈ A0 such that
d(U) = U n+1α(K + I) and d(U ∗) = −U n−1α(K).
We define a "twisted" derivation d : A0 → A0 by d(a(K)) = U n d(a(K)) for a(K) ∈ A0. A
direct computation yields
d(a(K)b(K)) = d(a(K))b(K) + a(K + nI) d(b(K))
d(b(K)a(K)) = d(b(K))a(K) + b(K + nI) d(a(K))
for a(K), b(K) ∈ A0. Since A0 and A0 are commutative algebras we get
d(a(K)) (b(K) − b(K + nI)) = d(b(K)) (a(K) − a(K + nI)) .
Similarly to the proof in Theorem 3.4 in
KMRSW2
[9], there must exist a β(K) such that
d(a(K)) = β(K) (a(K) − a(K + nI)) .
Consequently, we have the following formula
d(a(K)) = U nβ(K)(a(K) − a(K + nI)).
Next we apply d to the commutation relation U ∗a(K) = a(K+I)U ∗ for a diagonal operator
a(K) ∈ A0, and obtain:
U n−1(−α(K)a(K) + β(K)a(K) − β(K)a(K + nI)) =
= U n−1(β(K − I)a(K) − a(K + nI)β(K − I) − a(K + nI)α(K)),
where we define β(−1) := 0. Rearranging these terms gives:
α(K)(a(K + nI) − a(K)) = (β(K) − β(K − I))(a(K + nI) − a(K))
for all a(K) ∈ A0. It therefore follows that β(K) − β(K − I) = α(K). Thus β(k) is uniquely
determined by
and it follows that
β(k) =
α(j),
kXj=0
d(a) = [U nβ(K), a]
16
KLIMEK AND MCBRIDE
for any a ∈ A, since both sides of the above equation are derivations, and they agree on
the generators of the polynomial algebra A. By the remark preceding the statement of the
proposition, if f ∈ F is such that M +
A0 ∋ d(M +
f ◦ϕn is invertible, we can apply d to M +
f − M +
f ) = U nβ(K)(M +
f − M +
f ◦ϕn).
f to get
Therefore, it follows that we must have β(K) ∈ A0, and the proof is complete.
(cid:3)
To classify all derivations d : A → A we need to define the Fourier coefficients of d following
the ideas of
BEJ
[2].
Fou_comp
Definition: If d is a derivation in A, the n-th Fourier component of d is defined as:
dn(a) =
einθ(ρK
θ )−1dρK
θ (a) dθ.
1
2πZ 2π
0
A direct calculation shows that if d : A → A is a derivation then dn : A → A is an
n-covariant derivation.
We have the following key Ces`aro mean convergence result for Fourier components of d,
which is more generally valid for unbounded derivations in any Banach algebra with the
continuous circle action preserving the domain of the derivation:
if d is a derivation in A
then
d(a) = lim
M →∞
1
M + 1
MXj=0 jXn=−j
dn(a)! ,
for every a ∈ A, see Lemma 4.2 in
KMRSW2
[9] for more details.
(3.2)
Ces_eq
The following theorem classifies all derivations d : A → A.
der_decomp_A Theorem 3.5. Let d : A → A be any derivation. Then there exists c0 ∈ C such that d has
the following decomposition:
d(a) = c0dK(a) + d(a)
where d is an approximately inner derivation.
Proof. Let d0 be the 0-th Fourier component of d.
Proposition
invariant_der_A
3.3 we have the unique decomposition:
It is an invariant derivation, so by
d0(a) = c0dK(a) + d0(a) = c0[K, a] + d0(a),
covariant_der_A
for every a ∈ A, where d0 is an approximately inner derivation. From Proposition
3.4 we
have that the Fourier components dn, n 6= 0 are inner derivations. It follows from equation
Ces_eq
3.2), by extracting d0, that we have:
(
d(a) = d0(a) + lim
M →∞
1
M + 1
MXj=1
Xn≤j, n6=0
dn(a) .
The terms under the limit sign are all finite linear combinations of n-covariant derivations
and so they are inner derivations themselves, meaning that the limit is approximately inner,
which ends the proof.
(cid:3)
We also have the following useful but weaker convergence result for the Fourier components
of derivations.
Fourier_comp_converge_A Proposition 3.6. Let d : A → A be any derivation. Then for every x ∈ c00(Z≥0) ⊆ ℓ2(Z≥0)
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
17
and a ∈ A,
dn(a)x = d(a)x.
Xn∈Z
We say thatPn∈Z dn(a) converges densely pointwise on the set c00(Z≥0).
Proof. By Leibniz rule we only need to verify the above formula on generators of A. More-
over, it is enough to consider only x = E+
k , since c00(Z≥0) consists of finite linear combinations
of such x's. Below we show the details for a = M +
f , as the calculations for a = U and a = U ∗
are very similar. We have the following basis decomposition:
d(a)E+
hE+
j , dn(a)E+
k iE+
j .
Using the definition of the n-th Fourier components dn of d and the fact that dn are n-
covariant, a direct calculation gives:
k =
∞Xj=0
k i =( hE+
∞Xn∈Z
dn(a)E+
0
hE+
j , dn(a)E+
It follows that
completing the proof.
*E+
j ,
j , d(a)E+
k i
if n + k = j
otherwise.
k+ = hE+
j , d(a)E+
k i,
(cid:3)
3.2. Derivations in B. Next we classify derivations in B starting with the invariant deriva-
tions. It turns out that there are new types of invariant derivations in B that were not present
in A. We describe these in the following lemma.
partial_der Lemma 3.7. Let ∂ : F → C(G) be any derivation such that
∂f ◦ ϕ = ∂(f ◦ ϕ)
for all f ∈ F , which we call a ϕ invariant derivation in C(G). Then there exists a unique
invariant derivation δ∂ : B → B such that
δ∂(V ) = 0 and δ∂(Mf ) = M∂f .
Proof. Since (V, Mf ) is a defining representation for C(G) ⋊ϕ Z, the only relation in the
polynomial algebra B is
V Mf V −1 = Mf ◦ϕ.
Define the δ∂ on the generators as above by δ∂(V ) = 0 and δ∂(Mf ) = M∂f . Using the Leibniz
rule we try to extend this definition to all B. To verify that δ∂ is a well-defined derivation
from B → B, we thus need to check that it preserves the relation. Applying δ∂ to both sides
of the relation yields M∂f ◦ϕ = M∂(f ◦ϕ), completing the proof.
(cid:3)
As with algebra A there is a simple example of an invariant derivation which is given by
where b ∈ B. This derivation is well defined because B is the space of polynomials in V ,
V −1, and Mf , and we have [L, V ] = V , [L, V −1] = −V −1 and [L, Mf ] = 0.
δL(b) = [L, a]
18
KLIMEK AND MCBRIDE
delta_f_zero Lemma 3.8. For any f ∈ C(G) such that
ZG
f dµ = 0,
there exists a unique derivation δf : B → B such that
δf (V ) = V Mf ,
δf (V −1) = −Mf V −1,
δf (Mg) = 0,
where Mg ∈ B0. Moreover df is an approximately inner invariant derivation.
The proof is identical to that of Lemma
der_f_zero
3.2.
invariant_der_B Proposition 3.9. Let δ : B → B be any invariant derivation, then there exist c0 ∈ C and a
ϕ invariant derivation in C(G), ∂ : F → C(G), such that δ is of the unique form
where δ∂ : B → B is the derivation defined in Lemma
partial_der
3.7 and δ is approximately inner.
δ(b) = c0δL(b) + δ∂(b) + δ(b)
Proof. Since δ is invariant, there exists f0 ∈ C(G) such that
δ(V ) = V Mf0.
Moreover, there exists a linear map ∂ : F → C(G) such that
Applying δ to the relation Mf g = Mf Mg gives
δ(Mf ) = M∂f .
M∂(f g) = M∂f Mg + Mf M∂g.
Hence ∂ satisfies the Leibniz rule and thus is a derivation. Applying δ to both sides of the
relation V Mf V −1 = Mf ◦ϕ yields:
∂f ◦ ϕ = ∂(f ◦ ϕ),
i.e. ∂ is ϕ invariant.
Now write f0 = c0 + f with c0 ∈ C and
ZG
f dµ = 0.
It follows that
δ(b) = c0δL(b) + δ∂(b) + δ f (b)
delta_f_zero
3.8.
where δ∂ : B → B is the derivation defined in Lemma
invariant_der_A
3.3 we obtain that δL is not approximately inner.
Arguing as in the proof of Proposition
To complete the proof we notice that a non-zero derivation δ∂ cannot be approximately
inner since F is commutative and hence has no non-zero inner and approximately inner
derivations. This proves the uniqueness of the decomposition and finishes the proof of the
proposition.
(cid:3)
partial_der
3.7 and δ f is defined in Lemma
Because the proof of classifying all n-covariant derivations in B is essentially the same as
in the case of A, we only state the result.
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
19
covariant_der_B Proposition 3.10. Let δ : B → B be an n-covariant derivation where n 6= 0. There exists
an η(L) ∈ B0 such that
δ(b) = [U nη(L), b] for n 6= 0
Moreover δ is an inner derivation.
Finally, putting Propositions
invariant_der_B
3.9 and
covariant_der_B
3.10 together along with the comment Ces`aro mean
convergence result for Fourier components of d we have the following result.
der_decomp_B Theorem 3.11. Let δ : B → B be any derivation. Then there exists c0 ∈ C and a ϕ
invariant derivation ∂ : F → C(G) such that δ has the following unique decomposition:
δ(b) = c0δL(b) + δ∂(b) + δ(b)
where δ∂ is the derivation defined in Lemma
partial_der
3.7 and δ is an approximately inner derivation.
We also state here a dense pointwise convergence result for Fourier components δn of a
Fourier_comp_converge_A
3.6 and has completely analogous
derivation δ : B → B, which is similar to Proposition
proof.
Fourier_comp_converge_B Proposition 3.12. Let δ : B → B be any derivation. Then for every x ∈ c00(Z) ⊆ ℓ2(Z)
and b ∈ B,
and we say thatPn∈Z δn(b) converges densely pointwise on the set c00(Z).
4. Lifting Derivations
δn(b)x = δ(b)x,
Xn∈Z
The first important observation is that any derivation in algebra A preserves compact
operators.
der_preserve_compact Proposition 4.1. If d : A → A is a derivation, then d : A ∩ K → K.
Proof. It is enough to prove that d(P0) is compact, where P0 is the orthogonal projection
onto the one-dimensional subspace spanned by E+
0 , because A ∩ K is comprised of linear
combinations of expressions of the form U lP0(U ∗)j and compactness would follow immedi-
ately from the Leibniz property. To see that d(P0) is compact, apply d to both sides of the
relation P0 = P 2
0 to obtain:
d(P0) = d(P0)P0 + P0d(P0) ∈ K,
which completes the proof.
(cid:3)
As a consequence of Proposition
der_preserve_compact
4.1, if d : A → A is a derivation in A, then [d] : B → B
defined by
gives a derivation in B, which, by Corollary
structure_cor
2.10, is defined on B ∼= [A].
As a consequence to Proposition
der_preserve_compact
4.1, we have:
[d](a + K) := [d(a)]
Clearly, if d is an approximately inner derivation, then so is [d]. In general, given a derivation
δ : B → B, if there exists a derivation d : A → A such that [d] = δ we call such a d a lift of
δ.
[dK] = δL.
20
KLIMEK AND MCBRIDE
der_decomp_A
3.5 and
A natural question is: which derivations δ : B → B can be lifted to a derivation d : A → A?
der_decomp_B
It follows from Theorems
3.11 that if there is a nonzero ϕ invariant derivation in
C(G), ∂ : F → C(G), then there is no d : A → A such that [d] = δ∂, because δ∂ is not
approximately inner. A natural example of this is G = T1 = R/Z with xk = θk (mod Z),
k ∈ Z and θ irrational, giving a dense subgroup of T1. In this case, F is the actual space
of trigonometric polynomials. Any derivation ∂ : F → C(T1) invariant with respect to
ϕ(x) = x + θ (mod Z) is of the form:
∂(f ) = const
d
dx
f (x).
In this case, the algebra B is generated by V and W = Me2πix satisfying the relation
V W = e2πiθW V.
Consequently, B is isomorphic with the irrational rotation algebra. B is the algebra of
polynomials in V and W and the derivation δd/dx : B → B is given on generators by
δd/dx(V ) = 0 and δd/dx(W ) = 2πiW
and it cannot be lifted to a derivation in A. The key reason is that there is an additional
M_with_P_zero
relation on A given by equation (
2.3) which prevents existence of such a lift. We conjecture
however, that for any compact infinite monothetic group, any approximately inner derivation
δ : B → B can be lifted to a derivation d : A → A.
For the remainder of the section we let G be totally disconnected, in other words G is an
tot_disc_mono
odometer, and thus by Proposition
2.4, there exists an infinite supernatural number N such
KMRSW2
that G ∼= Z/NZ. It was proved in
[9] that for such G's, the algebras A and B are precisely the
Bunce-Deddens-Toeplitz, A(N), and Bunce-Deddens algebras, B(N), respectively. It follows
KMRSW2
[9] that there are no nontrivial ϕ invariant derivations ∂ : F → C(G).
from Theorem 4.4 in
Below we prove one of the main results of this paper that for odometers, any unbounded
derivation in B(N) can be lifted to an unbounded derivation in A(N).
We will need the following useful result for computing Hilbert-Schmidt norms of opera-
tors in ℓ2(Z) and ℓ2(Z≥0). Since below we work mostly with algebra A, we only state the
corresponding version for brevity.
HSProp Proposition 4.2. Let a : ℓ2(Z≥0) → ℓ2(Z≥0) be defined by:
a =
U nan(K) +
an(K)(U ∗)−n,
∞Xn=0
−1Xn=−∞
where {an(k)}n∈Z,k∈Z≥0 ∈ ℓ2(Z×Z≥0). Then a is an integral operator with the Hilbert-Schmidt
norm given by:
Proof. Write f ∈ ℓ2(Z≥0) in the canonical basis:
kak2
HS =Xn∈Z
∞Xk=0
f =
an(k)2.
∞Xk=0
f (k)E+
k .
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
21
Applying the formula for a to f yields:
Resumming both terms gives:
an−k(n)f (k)E+
n
n .
an(k)f (k)E+
an(k + n)f (k)E+
k+n.
af =Xn≥0Xk≥0
af =Xn≥0Xn≥k
=Xn≥0 ∞Xk=0
k+n + X0≤k+nXn<0
n +Xn≥0Xn<k
an−k(min{n, k})f (k)! E+
an−k(k)f (k)E+
This shows that a is an integral operator with integral kernel
κ(k, n) = an−k(min{n, k}) ∈ ℓ2(Z × Z≥0).
Therefore, by writing a in the following way
(af )(n) =
∞Xk=0
κ(k, n)f (k) =
nXk=0
an−k(k)f (k) +Xk>n
an−k(n)f (k),
the Hilbert-Schmidt norm formula now follows, completing the proof.
(cid:3)
lift_theo Theorem 4.3. Let δ : B(N) → B(N) be any derivation. There exists a derivation d :
A(N) → A(N) such that [d] = δ.
Proof. Let δ : B(N) → B(N) be an approximately inner derivation in B(N), then by
Theorem
der_decomp_B
3.11 and by Propositions
invariant_der_B
3.9 and
covariant_der_B
3.10 we have
δ(b) ="Xn≥0
V nMfn +Xn<0
MfnV n, b#
for b ∈ B(N), where the convergence of infinite sums is understood as being densely pointwise
on c00(Z). In order to construct a lift of δ we need to consider derivations d : A(N) → A(N)
given by the following expression, densely pointwise convergent on c00(Z≥0):
d(a) ="Xn≥0
U nβn(K) +Xn<0
βn(K)(U ∗)−n, a#
for a ∈ A(N), where βn(k) has the following decomposition:
βn(K) = βn,0(K) + M +
fn
where βn,0(k) ∈ c0(Z≥0). We need to find conditions on βn,0(k) so that d is a well-defined
derivation in A(N) such that
[d](a) = δ([a])
for all a ∈ A(N). By the Leibniz rule we only need to check this equation on the generators
U, U ∗, and M +
χ , where χ is a character on Z/NZ.
22
KLIMEK AND MCBRIDE
A direct computation yields the following formula:
U n+1 (βn,0(K + I) − βn,0(K))
(βn,0(K) − βn,0(K − I)) (U ∗)−n−1 + P0 Xn<0
fn◦ϕ−1(U ∗)−n−1!
M +
Similarly, on U ∗ we have:
d(U) − T (δ(V )) =Xn≥0
+Xn<0
d(U ∗) − T (δ(V −1)) =Xn>0
+Xn≤0
Finally, we get the following expression for diagonal operators M +
χ :
U n−1 (βn,0(K − I) − βn,0(K))
(βn,0(K) − βn,0(K + 1)) (U ∗)−n+1 + Xn>0
χ ) − T (δ(Mχ)) =Xn≥0
+Xn<0
U nβn,0(K)(M +
βn,0(K)(M +
χ◦ϕn − M +
χ )(U ∗)−n.
d(M +
U n−1M +
fn◦ϕ−1! P0
χ − M +
χ◦ϕn)
The result follows provided we can choose βn,0(k) so that the right-hand sides of the above
equations are compact operators. We compute the Hilbert-Schmidt norm of the above opera-
HSProp
4.2 yields the following
tors to show the compactness. A direct calculation using Proposition
formulas:
I := kd(M +
χ ) − T (δ(Mχ))k2
βn,0(k)2χ(xk) − χ(xk+n)2
HS =Xn∈ZXk≥0
HS =Xn∈ZXk≥1
βn,0(k) − βn,0(k − 1)2 +Xn∈Z
= kd(U ∗) − T (δ(V −1))k2
HS.
II := kd(U) − T (δ(V ))k2
βn,0(0) − fn(x−1)2
We define βn,0(k) to have the following form:
− fn(x−1)(cid:18)Nn − k
Nn (cid:19) 0 ≤ k < Nn
k ≥ Nn,
βn,0(k) =
0
where the numbers Nn will be chosen later.
Notice that any character on Z/NZ is of the form:
M (cid:19) ,
χ(xk) = exp(cid:18) 2πijk
UNBOUNDED DERIVATIONS IN ALGEBRAS ASSOCIATED WITH MONOTHETIC GROUPS
23
where M N and j ∈ Z. Therefore I and II become
I =Xn∈Z
II =Xn∈Z
Nn (cid:19)2
fn(x−1)2(cid:18) Nn − k
NnXk=0
Nn−1Xk=0
=Xn∈Z
fn(x−1)2
N 2
n
fn(x−1)2
.
Nn
1 − e2πijn/M 2
The key observation used below is that the coefficients fn on the Fourier decomposition
of the derivation δ : B → B satisfly the following condition: for all M N:
fn(x−1)2 < ∞.
(4.1)
hs_condition
XM ∤n
fn(x−1)21 − e2πijn/M 2 = kP≥0δ(exp(2πijL/M)P−1k2
HS < ∞,
This follows from the formula:
and a similar formula for n < 0:
Xn≥0
Xn∈Z
fn(x−1)21 − e2πijn/M 2 = kP−1δ(exp(2πijL/M)P≥0k2
HS < ∞.
Here P−1 is the orthogonal projection in ℓ2(Z) onto the one-dimensional subspace spanned by
E−1, while P≥0 is the orthogonal projection onto the subspace spanned by {El}l≥0. Equations
above imply that we have:
fn(x−1)21 − e2πijn/M 2 =XM ∤n
fn(x−1)2,
∞ >Xn∈Z
> constXM ∤n
fn(x−1)21 − e2πijn/M 2
since the factor 1 − e2πijn/M has only finitely many values. This gives the following estimate:
Nn (cid:19)2
fn(x−1)2(cid:18)Nn − k
NnXk=0
I =Xn∈Z
fn(x−1)2(cid:18)Nn − k
Nn (cid:19)2
NnXk=0
≤ 4XM ∤n
1 − e2πijn/M 2
∼ constXM ∤n
Nnfn(x−1)2.
To proceed further we choose a scale s = (sm)m∈N for the supernatural number N, which
is a sequence of positive integers such that sm divides sm+1, sm < sm+1, and such that
N_limit
2.1). For every n ∈ Z there is an index m such that sm n but sm+1 ∤ n.
N = limm sm, see (
We then write
n = smn′,
where n′ is such that sm+1/sm ∤ n′. Using this decomposition we choose Nn = Cm to be
a constant depending on m only, to be determined later. Also, without loss of generality,
we can choose M, in the formula for the character χ, to be equal to one of the elements of
24
KLIMEK AND MCBRIDE
the scale: M = sq. It is then important to notice that sq ∤ n = smn′ if and only if m < q.
Consequently, we have the following expressions:
I ≤ const
≤ const
q−1Xm=1
q−1Xm=1
Cm Xsm+1/sm∤n′
Cm Xsm+1∤n
fn(x−1)2 < ∞
fsmn′(x−1)2 = const
q−1Xm=1
Cm Xsmn, sm+1∤n
fn(x−1)2
for any choice of Cm because the sum over sm+1 ∤ n is finite by equation (
hs_condition
4.1).
Next, for II we have an estimate:
II =
∞Xm=1
1
Cm Xsmn, sm+1∤n
fn(x−1)2 ≤
fn(x−1)2.
∞Xm=1
1
Cm Xsm+1∤n
By equation (
so that II < ∞. This completes the proof.
hs_condition
4.1) the interior sum is finite. Finally, we can always choose Cm large enough
(cid:3)
References
[1] Bratteli, O., Derivations, Dissipations and Group Actions on C∗ -algebras, Lecture Notes in Math. 1229,
Springer, 1986.
[2] Bratteli, O., Elliott, G. A., and Jorgensen, P. E. T., Decomposition of unbounded derivations into
invariant and approximately inner parts, Jour. Reine Ang. Math., 346, 166 - 193, 1984.
[3] Downarowicz, T., Survey of odometers and Toeplitz flows, Contemp. Math., 385, 7 - 37, 2005.
[4] Fillmore, P. A User's Guide to Operator Algebras, Wiley-Interscience Publication, 1996.
[5] Halmos, P. and Samelson, H., On Monothetic Groups, Proc. AMS, 28, 254 - 258, 1942.
[6] Katok, A. and Hasselblatt, B., Introduction to the modern theory of dynamical systems, Cambridge
University Press, 1996.
[7] Klimek, S., McBride, M., Rathnayake, S., and Sakai, K., The Quantum Pair of Pants, SIGMA, 11, 012,
1 - 22, 2015.
[8] Klimek, S., McBride, M., Rathnayake, S., Sakai, K., and Wang, H., Derivations and Spectral Triples on
Quantum Domains I: Quantum Disk, SIGMA, 13, 075, 1 - 26, 2017.
[9] Klimek, S., McBride, M., Rathnayake, S., Sakai, K., Wang, H., Unbounded Derivations in Bunce-
Deddens-Toeplitz Algebras, Jour. Math. Anal. Appl., 15, 988 - 1020, 2019.
[10] Kurka, P., Topological and Symbolic Dynamics, Soci´et´e math´ematique de France, 2003.
[11] Morris, S., Pontryagin Duality and the Structure of Locally Compact Abelian Groups, Cambridge Uni-
versity Press, 1977
[12] Pedersen, G. K., Lifting Derivations from Quotients of Separable C∗-algebras, Proc. Natl. Acad. Sci.,
73, 1414 - 1415, 1976
[13] Robert, A., A Course in p-adic Analysis, Springer, 2000.
[14] Sakai, S., Operator Algebras in Dynamical Systems, Cambridge University Press, 1991.
[15] Willard, S., General Topology, Addison-Wesley Publishing, 1970
Department of Mathematical Sciences, Indiana University-Purdue University Indianapo-
lis, 402 N. Blackford St., Indianapolis, IN 46202, U.S.A.
E-mail address: [email protected]
Department of Mathematics and Statistics, Mississippi State University, 175 President's
Cir., Mississippi State, MS 39762, U.S.A.
E-mail address: [email protected]
|
1608.00610 | 1 | 1608 | 2016-08-01T20:51:17 | Cohomology for spatial super-product systems | [
"math.OA"
] | We introduce a cohomology theory for spatial super- product systems and compute the $2-$cocycles for some basic examples called as Clifford super-product systems, thereby distinguish them up to isomorphism. This consequently proves that a family of $E_0-$semigroups on type III factors, which we call as CAR flows, are non-cocycle-conjugate for different ranks. Similar results follows for the even CAR flows as well. We also compute the automorphsim group of the Clifford super-product systems. | math.OA | math |
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT
SYSTEMS
OLIVER T. MARGETTS AND R. SRINIVASAN
Abstract. We introduce a cohomology theory for spatial super- prod-
uct systems and compute the 2−cocycles for some basic examples
called as Clifford super-product systems, thereby distinguish them
up to isomorphism. This consequently proves that a family of E0-
semigroups on type III factors, which we call as CAR flows, are non-
cocycle-conjugate for different ranks. Similar results follows for the
even CAR flows as well. We also compute the automorphsim group of
the Clifford super-product systems.
1. Introduction
The super-product system of Hilbert spaces is a generalisation of Arve-
son's product system of Hilbert spaces. These were originally defined in
[MaS1], but the idea was being discussed before that. Apart from being an
interesting mathematical object on its own, super-product systems arises
naturally as an invariant associated to E0-semigroups on factors, as shown
in [MaS2]. In a sense this generalizes Arveson's association of product sys-
tems to E0-semigroups on type I factors, but this association is not one-one
in the non-type I case. We recall the basic definitions and the association
of super-product system to an E0-semigroup in this introductory section.
Definition 1.1. A super-product system of Hilbert spaces is a one pa-
: t > 0}, together with
rameter family of separable Hilbert spaces {Ht
isometries
Us,t : Hs ⊗ Ht
7→ Hs+t for s, t ∈ (0,∞),
satisfying the following two axioms of associativity and measurability.
(i) (Associativity) For any s1, s2, s3 ∈ (0,∞)
Us1,s2+s3(1Hs1 ⊗ Us2,s3) = Us1+s2,s3(Us1,s2 ⊗ 1Hs3 ).
(ii) (Measurability) The space H = {(t, ξt) : t ∈ (0,∞), ξt ∈ Ht} is
equipped with a structure of standard Borel space that is compatible with
2010 Mathematics Subject Classification. Primary 46L55; Secondary 46L40, 46L53,
46C99.
Key words and phrases. *-endomorphisms, E0-semigroups, II1 factors, noncommuta-
tive probability, super-product systems.
1
2
O. MARGETTS AND R. SRINIVASAN
the projection p : H 7→ (0,∞) given by p((t, ξt) = t, tensor products and
the inner products (see [Arv, Remark 3.1.2]).
A super-product system is an Arveson product system if the isometries
Us,t are unitaries and further the condition of local triviality is satisfied,
that is there exists a single separable Hilbert space H satisfying H ∼=
(0,∞) × H as measure spaces (see [Arv, Remark 3.1.2]).
Definition 1.2. By an isomorphism between two super-product systems
(H 1
s,t) we mean an isomorphism of Borel spaces V :
H1 7→ H2 whose restriction to each fiber provides a unitary operator Vt :
H 1
t 7→ H 2
s,t) and (H 2
t satisfying
t , U 2
t , U 1
Vs+tU 1
s,t = U 2
s,t(Vs ⊗ Vt).
Basic examples of product systems are exponential product systems
and antisymmetric product systems, whose subsystems provides our basic
examples of proper super-product systems. These are discussed in the
next section.
The study of E0-semigroups was initiated by R. T. Powers (see [Pow1]),
and Arveson made many important contributions through a sequence of
papers in late 80s and 90s (see [Arv]). Arveson showed that E0-semigroups
on type I factors are completely classified by their associated product sys-
tems. This gives a rough division of E0-semigroups into three types, based
on the existence of units, namely I, II and III. The type I E0-semigroups
on type I factors are cocycle conjugate to the CCR flows ([Arv]), but there
are uncountably many E0-semigroups of types II and III ([Pow2], [Tsi],
[BhS], [IS1], [IS2] [VL]) on type I factors.
We say a von Neumann algebra M is in standard form if M ⊆ B(H)
has a cyclic and separating vector Ω ∈ H. By replacing a conjugate E0-
semigroup if needed, without loss of generality, we can always assume that
an E0-semigroup is acting on a von Neumann algebra in a standard form,
thanks to [MaS2, Lemma 2.4 ].
Definition 1.3. An E0-semigroup on a von Neumann algebra M is a
semigroup {αt : t ≥ 0} of normal, unital *-endomorphisms of M satisfying
(i) α0 = id,
(ii) αt(M) 6= M for all t > 0,
(iii) t 7→ ρ(αt(x)) is continuous for all x ∈ M, ρ ∈ M∗.
Definition 1.4. A cocycle for an E0-semigroup α on M is a strongly
continuous family of unitaries {Ut : t ≥ 0} ⊆ M satisfying Usαs(Ut) =
Us+t for all s, t ≥ 0.
For a cocycle {Ut
: t ≥ 0}, we automatically have U0 = 1. Fur-
t (x) := Utαt(x)U∗t defines an
thermore the family of endomorphisms αU
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
3
E0-semigroup. This leads to the equivalence of cocycle conjugacy on E0-
semigroups. The following definition is called as spatial conjugacy, but we
can assume any two conjugate E0-semigroups in standard form are spa-
tially conjugate, thanks to [MaS2, Lemma 2.4 ]. So we take the following
as the definition of conjugacy.
Definition 1.5. Two E0-semigroups α and β, acting standardly on M ⊆
B(H1) and N ⊆ B(H2) respectively, are said to be conjugate if there exists
a unitary U : H1 7→ H2 satisfying
(i) UMU∗ = N,
(ii) βt(x) = Uαt(U∗xU)U∗ for all t ≥ 0, x ∈ N,
Definition 1.6. Let α and β be E0-semigroups on von Neumann algebras
M and N. Then α and β are said to be cocycle conjugate if there exists a
cocycle {Ut : t ≥ 0} for α such that β is conjugate to αU .
Given an E0-semigroup θ on B(H), the associated Arveson product
system (Ht, Us,t) is defined by
Ht = {T ∈ B(H); αt(X)T = T X, ∀ X ∈ B(H)}
with inner product hT, Si1H = S∗T and Us,t(T ⊗ S) = T S for T ∈ Hs, S ∈
Ht.
Let α be an E0-semigroup on a factor M with cyclic and separating
vector Ω and let JΩ be the modular conjugation associated with Ω by the
Tomita-Takesaki theory. We can define a complementary E0-semigroup α′
on M′ by setting
α′t(x′) = JΩαt(JΩx′JΩ)JΩ
(x′ ∈ M′).
The complementary E0-semigroup is determined up to conjugacy, thanks
to the following Proposition (for proof see [MaS2, Proposition 3.1 ]).
Proposition 1.7. Let M and N be von Neumann algebras acting stan-
dardly with respective cyclic and separating vectors Ω1 ∈ H1, Ω2 ∈ H2. If
the E0-semigroups α on M, and β on N are cocycle conjugate, then α′ and
β′, defined with respect to Ω1 and Ω2 respectively, are cocycle conjugate.
Moreover, if α and β are conjugate, then α′ and β′ are spatially conjugate
and the implementing unitary can be chosen so that it also intertwines α
and β.
A super-product system is associated as an invariant to an E0-semigroup
on a factor through the following theorem (see [MaS2, Theorem 3.4]).
Theorem 1.8. Let M ⊆ B(H) be a factor in standard form and α an
E0-semigroup on M. For each t > 0, let
t = {X ∈ B(H) : ∀m∈M,m′∈M′ αt(m)X = Xm, α′t(m′)X = Xm′}.
H α
4
O. MARGETTS AND R. SRINIVASAN
Then H α = {H α
family of isometries Us,t (X ⊗ Y ) = XY .
t
: t > 0} is a super-product system with respect to the
Let α and β be E0-semigroups acting on respective factors M and N in
If α and β are cocycle conjugate then H α and H β are
standard form.
isomorphic.
In [MaS1], super-product systems of a family of E0-semigroups on type
II1 factors, called as Clifford flows and even Clifford flows, were com-
puted. Though the Clifford flows and even Clifford were shown to be
non-cocycle-conjugate for different ranks, the proof was indirect, using
boundary representations and a theory of C∗−semiflows. The associated
super-product systems were not shown to be non-isomorphic when their
ranks are different. In this paper we show those super-product systems
to be non-isomorphic for different ranks (see Section 5), which gives a di-
rect proof regarding the non-cocycle-conjugacy of Clifford flows and even
Clifford flows.
In [MaS2], using CCR representations, uncountable families consisting
of mutually non-cocycle-conjugate E0-semigroups were constructed on ev-
ery type IIIλ factors, for λ ∈ (0, 1]. These are called as CCR flows on
type III factors. Similar to CCR representations, it is also possible to
produce families of E0-semigroups using CAR representations. But useful
invariants are not yet found to distinguish them up to cocycle conjugacy.
This difficulty is mainly because the gauge group turns out to be trivial
for E0-semigroups constructed through CAR representations, unlike the
CCR representations.
In this paper we distinguish CAR flows on type III factors when their
ranks are different, by showing the associated super-product systems are
non-isomorphic (see Section 5). It remains open to classify among CAR
flows with same rank, similar to the classification done for CCR flows in
section 8 in [MaS2].
2. Preliminaries
N denotes the set of natural numbers, and we set N0 = N ∪ {0}, N =
N∪{∞}. We will only deal with complex Hilbert spaces in this paper. The
inner product is always conjugate linear in the first variable and linear in
the second variable. L2(S, k) denote the square integrable functions from
S taking values in a complex separable Hilbert space k. L2
loc(S, k) denotes
the functions which are square integrable on compact subsets. Throughout
this paper we denote by (Tt)t≥0 the right shift semigroup on L2((0,∞), k)
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
5
defined by
(Ttf )(s) = 0,
s < t,
= f (s − t),
s ≥ t,
for f ∈ L2((0,∞), k).
We will be dealing with antisymmetric tensors in all the examples con-
sidered in this paper, since they are the ones related to our examples of
E0-semigroups. But examples of super-product systems in this paper and
the facts proven about them can be extended to symmetric case as well,
by making analogous changes.
Let K be a complex separable Hilbert space. Let Γ(K) :=L∞n=0 (K)∧n
be the antisymmetric Fock space over K, where K∧0 is C. For any f ∈ K
the Fermionic creation operator a∗(f ) is the bounded operator defined by
the linear extension of
a∗(f )ξ =(cid:26) f
f ∧ ξ
if ξ = Ω,
if ξ ⊥ Ω,
where Ω is the vacuum vector. The annihilation operator is defined by
the adjoint a(f ) = a∗(f )∗. The creation and annihilation operators satisfy
the well-known anti-commutation relations (see Equations (1) in Section
4) and generate B(Γ(K)). For an isometry U : K1 7→ K2, the second quan-
tization Γ(U) : Γ(K1) 7→ Γ(K2) is the isometry defined by the extension
of
Γ(U)(x1 ∧ x2 ∧ · · · ∧ xn) = Ux1 ∧ Ux2 ∧ · · · ∧ Uxn.
For s ∈ (0,∞), t ∈ (0,∞], define
Us,t : Γ(L2((0, s), k) ⊗ Γ(L2((0, t), k) 7→ Γ(L2((0, s + t), k)
as the unitary the extension of
(ξ1∧ξ2∧· · ·∧ξm)⊗(η1∧η2∧· · ·∧ηn) 7→ Tsη1∧Tsη2∧· · ·∧Tsηn∧ξ1∧ξ2∧· · ·∧ξm.
To avoid messy notations, we will continue to use Us,t for isometric re-
strictions of Us,t to subspaces as well and for different k.
Example 2.1. The CAR flow of index n = dim k ∈ N is the E0-semigroup
: t ≥ 0} acting on B(Γ(L2((0,∞), k))) defined by the extension
θn = {θn
of
t
t (a(f )) := a(Ttf ), f ∈ L2((0,∞), k).
θn
Example 2.2. The Arveson product system associated with CAR flow of
index dim(k) is described by the CAR product system (H k(t), Us,t), where
H k(t) = Γ(L2((0, t), k). (see [MaS1, Remark 5.13].)
The following example gives several families of proper super-product
systems. The basic examples we are concerned with are Clifford super-
product systems and CAR super-product systems.
6
O. MARGETTS AND R. SRINIVASAN
Example 2.3. For an additive sub-semigroup G of N0, define
H k
G(t) :=Mn∈G
L2([0, t], k)∧n ⊆ H k(t);
Then (cid:0)H k
G(t), Us,t(cid:1) forms a super-product system.
When G = 2N0, we call them as the Clifford super-product systems, as
they are ones associated with Clifford flows (also with the even Clifford
flows) on hyperfinite II1 factor (see [MaS1, Corollary 8.13 ]).
Example 2.4. We can take tensor products of super-product systems,
where both the Hilbert spaces and isometries are tensored accordingly, to
produce new super-product systems. Define
Ek
2N0(t) := Mn1+n2∈2N0
L2([0, t], k)∧n1 ⊗ L2([0, t], k)∧n2 ⊆ H k(t) ⊗ H k(t).
Denote U 2s, t = Us,t ⊗ Us,t (and also its restrictions). Then (Ek
forms a super-product system.
2N0(t), U 2
s,t)
By restricting the natural isomorphism between the product systems
(cid:0)H k(t) ⊗ H k(t), U 2s, t(cid:1) and (cid:0)H k⊕k, Us,t(cid:1), it is easy to see that
2N0 (t), Us,t).
2N0(t), U 2
s,t) ∼= (H k⊕k
(Ek
3. Types of super-product systems
Imitating the definition of types for product systems, super-product
systems can also be broadly divided into three types. We further divide
type II super-product systems based on the existence of 2-addits, which
can be further refined by considering the n−th cohomology.
Definition 3.1. A unit for a super-product system (Ht, Us,t) is a measur-
able section u = {ut : ut ∈ Ht}t≥0 satisfying Us,t(us ⊗ ut) = us+t ∀ s, t ∈
(0,∞).
A super-product system is called spatial if it admits a unit. In a spatial
super-product system we usually fix a special unit called as the canonical
unit, denoted by Ω = {Ωt ∈ Ht}. An exponential unit is a unit u satisfying
hut, Ωti = 1. We denote the collection of all exponential units by UΩ(H).
For every unit u there exists a scalar λ ∈ C such that {e−λtut} ∈ UΩ(H).
Definition 3.2. An addit for a spatial super-product system (Ht, Us,t),
with respect to a canonical unit Ω, is a measurable family of vectors b =
{bt : bt ∈ Ht}t≥0 satisfying
Us,t(bs ⊗ Ωt) + Us,t(Ωs ⊗ bt) = bs+t ∀s, t ≥ 0.
We say an addit is centred if hΩt, bti = 0 for all t ≥ 0.
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
7
Denote the set of all centered addits by AΩ(H). Every addit b can be
written as bt = ct + λtΩt such that c is a centered addit and λ ∈ C.
The following theorem is proved in [MaS1, Theorem 5.11] (see also [MaS1,
Remark 5.12]). The second part follows from the construction of ExpΩ
and LogΩ.
Theorem 3.3. Let (Ht, Us,t) be a spatial super-product system with canon-
ical unit {Ωt}. There exists a bijection ExpΩ : AΩ(H) 7→ UΩ(H) with
inverse LogΩ satisfying
hExpΩ(b)t, ExpΩ(b′)ti = ehb1,b′
1it ∀ b, b′ ∈ AΩ(H).
Moreover ExpΩ(AΩ(H)) is contained in the product system generated
by the centered addits with respect to Ω, and LogΩ(UΩ(H)) is contained in
the product system generated by the units.
Let S be a collection of measurable sections of a super-product system
the closure of the
i=1 ti = t, si ∈ S}, and we continue
(Here the product is
t by Us,t.
tn in Ht, under the canonical unitary given
t , Us,t) is the super-product system
(Ht, Us,t). For any fixed t ∈ (0,∞), we denote by H S
linear span of the set {s1
to denote the restrictions of Us,t to H S
the image of s1
t1 ⊗ s2
by the associativity axiom.) Then (H S
generated by the S.
tn :Pn
t
t1s2
t2 · · · sn
t2 · · · ⊗ sn
A super-product system is said to be of type I if units exist and the
collection of units generate the super-product system. Thanks to Theorem
3.3, a spatial super-product system with canonical units Ω is type I if and
only if it is generated by {Ω} ∪ AΩ. It is easily verified that type I super-
product systems are indeed product systems.
It also follows that the
index, defined by Arveson through units, is nothing but the dimension of
AΩ, which is a Hilbert space with respect to ha, bi = ha1, b1i, satisfying
hat, bti = tha, bi (see [MaS1, Lemma 4.8]). Further the dimension of AΩ
does not depend on a particular unit Ω and this index is an invariant.
The centered addits of CAR product systems are given by one particle
vectors {ξ ⊗ 1[0,t] : ξ ∈ k}, which generate the product system ([MaS1,
Lemma 7.1]). Hence they are of type I and have index dim(k). This
shows that (H k(t), Us,t) are non-isomorphic if dim(k) varies. Since Arveson
systems form a complete invariant for E0-semigroups on B(H) with respect
to cocycle conjugacy, the following theorem, which was originally proved
by Arveson, implies that any E0-semigroup with a type I product system
of index n is cocycle conjugate to the CAR flow of index n. Approaching
with addits provides a much simpler proof.
Theorem 3.4. For n ∈ N any type I product system of index n is iso-
morphic to (H k(t), Us,t) for some k satisfying dim(k) = n.
8
O. MARGETTS AND R. SRINIVASAN
Proof. Let (Ht, Us,t) be a type I product system of index n ∈ N. Fix
a unit a0 and centered addits Aa0 with orthonormal basis {ai : i ∈ I}.
Let k be a separable Hilbert space of dimension n with orthonormal basis
s,t = 1(s,t) ⊗ ei for s, t ∈ (0,∞) and i ∈ I. Define
{ei : i ∈ I}. Set ei
Ut : Ht 7→ Γ(L2(0, t) ⊗ k) by Uta0 = Ω and
t1,t1+t2 ∧ · · · ∧ eik
tk(cid:1) = ei1
Ut(cid:0)ai1
t1+···+tk−1,t1+···+tk−1+tk ,
where t1 +t2 +· · ·+tk = t, {i1, i2 · · · ik} 6= {0} and e0
ti,ti+1 means there is no
vector in the antisymmetric tensor product. Us,t preserves inner products,
maps total set of vectors onto a total set, hence extends to a unitary
operator, and provides the required isomorphism of product systems. (cid:3)
t2 ⊗ · · · ⊗ aik
t1 ⊗ ai2
0,t1 ∧ ei2
A spatial super-product system with index n is said to be of type II (or
type IIn) if units do not generate the super-product system. It is type III
if units do not exist. Examples of product systems of type II and III are
complicated to construct, but constructing nontrivial examples of super-
product systems belonging to those types are readily given by Example
2.3. There are three possibilities for H k
G(t) defined in Example 2.3 namely
• G = N0: they are of type In.
• 0 ∈ G 6= N0: they are of type II0.
• 0 /∈ G: they are type III.
This follows immediately from Lemma 7.1, [MaS1].
If (Ht, Us,t) is a super-product system and (Et, Us,t) is a product sub-
system then (E⊥t ∩ Ht, Us,t) is a super-product system. So type III super-
product systems can be easily constructed from type II super-product
systems, by taking the orthogonal complement of the product system gen-
erated by units.
3.1. Cohomology for spatial super-product systems. Let (Ht, Us,t)
be a spatial super-product system with a distinguished unit Ω. The em-
beddings
ιs,t : Hs 7→ Hs+t
ξ 7→ Us,t(ξ ⊗ Ωt)
allow us to construct an inductive limit of the family of Hilbert spaces
(Hs)s>0, which we denote by H∞, together with embeddings ιs : Hs →
H∞. We will often abuse notation and identify Hs with its corresponding
image in H∞. Notice that each of the vector Ωs is mapped to the same
element, which we denote by Ω∞ ∈ H∞.
We can also define a second family of embeddings
κs,t : Ht 7→ Hs+t
ξ 7→ Us,t(Ωs ⊗ ξ).
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
9
Thanks to the associativity axiom, the squares
Hs
κr,s
ιs,t
Hs+t
κr,s+t
Hr+s ιr+s,t
/ Hr+s+t
commute for all r, s, t > 0. So there exist isometries (κt : H∞ 7→ H∞)t≥0,
which define an action of the semigroup R+ on H∞, satisfying κsιt =
ιs+tκs,t.
We say a function f : Rn
+ → H∞ ⊖ CΩ∞ is adapted if f (s1, . . . , sn) ∈
Hs1+···+sn for all s1, . . . , sn > 0. Let C n = C n(H, Ω) denote the space of
all adapted continuous maps f : Rn
+ → H∞ ⊖ CΩ∞, and dn : C n(H, Ω) →
C n+1(H, Ω) be defined by
dnf (s1, . . . , sn+1) := κs1f (s2, . . . , sn+1)
+
n
Xi=1
.
(−1)nf (s1, . . . , si + si+1, . . . , sn+1) + (−1)n+1f (s1, . . . , sn)
Lemma 3.5. (C, d) forms a cochain complex.
Proof. The map dn is the restriction of the usual coboundary map in the
group cohomology for the action R+ y H∞. We only need to check that
the image is adapted, but this follows since κs(Ht) ⊆ Hs+t.
Definition 3.6. For all n ≥ 0, the collection of n-cocycles for (H, Ω)
is the space Z n(H, Ω) := Ker(dn) and the collection of n-coboundaries is
defined by B1(H, Ω) = 0 and, for n ≥ 2, Bn(H, Ω) = Ran(dn−1). The
n-th cohomology group is the space Hn(E, Ω) := Z n(E, Ω)/Bn(E, Ω).
(cid:3)
The cohomology groups are invariant under isomorphisms of super-
product systems preserving the canonical unit. For all examples we con-
sider in this paper, the spaces of higher cocycles are infinite dimensional
and the cohomology groups are difficult to compute. We will instead con-
centrate on certain distinguished subspaces, which are more tractable. To
some extent they measure how far the super-product system is from being
a product system.
Definition 3.7. Let H = (Ht, Us,t) be a super-product system with canon-
ical unit Ω. A defective n-cochain for (H, Ω) is a member of C n(E, Ω)
satisfying
a(s1, . . . , sn) ⊥ ιs1+···+snUs1,...,sn(Hs1 ⊗ · · · ⊗ Hsn)
for all s1, . . . , sn > 0, where Us1,...,sn : Hs1 ⊗ · · · ⊗ Hsn 7→ Hs1,...,sn is
canonical unitary map determined uniquely by the associativity axiom.
/
/
/
10
O. MARGETTS AND R. SRINIVASAN
We denote the space of defective n-cochains by C n
similarly, the collection of defective n-cocycles Z n
aries Bn
def (H, Ω). The corresponding quotient Hn
fective cohomology group.
def (H, Ω) and define,
def (H, Ω) and cobound-
def (H, Ω) is the n-th de-
Clearly for a product system, one always has C n
following is immediate from definitions.
def (H, Ω) = {0}. The
def (K, Ω′) which preserves cocycles and coboundaries.
Proposition 3.8. Let H = (Ht, Us,t) and H = (Kt, U′s,t) be two spatial
super-product systems with canonical units Ω and Ω′ respectively. Let Vt :
Ht → Kt be an isomorphism of super-product systems taking the unit
(Ωt)t≥0 to (Ω′t)t≥0. Then, for each n ≥ 0, there is an isomorphism Φ :
C n
def (H, Ω) → C n
One can give an equivalent definition for the cohomology without refer-
ring to the inductive limit. For instance, there is a bijective correspondence
between the 1−cocycles and the addits we have defined earlier. Further
dim Z 1(H, Ω) is the index of H, which is independent of the choice of Ω. It
is also easy to see that there is a bijective correspondence with 2−cocycles
and the 2−addits defined below.
Definition 3.9. A 2−addit for a spatial super-product system (Ht, Us,t),
with respect to a canonical unit {Ωt}, is a measurable family of vectors
{as,t : s, t ≥ 0} satisfying
(i) as,t ∈ Hs+t ∀ s, t ≥ 0,
(ii) Ur+s,t(ar,s ⊗ Ωt) + ar+s,t = Ur,s+t(Ωr ⊗ as,t) + ar,s+t ∀ r, s, t ≥ 0.
A 2−addit is said to be defective if further as,t ∈ (Us,t(Hs ⊗ Ht))⊥
∀ s, t ≥ 0.
We denote the set of all defective 2−addits by A2
Ω(H). Note that a prod-
uct system, in particular a type I super-product system, does not admit
defective 2−addits. We can further divide type II0 super-product systems
depending upon the existence and abundance of defective 2−addits. We
say a type II0 super-product system, with a canonical unit Ω, is type II0-I
if it is generated by {Ω} ∪ A2
Ω in the sense, for each T > 0, HT is closure
of the linear span of the set
n
s1,t1a2
s2,t2 · · · an
sn,tn : ai ∈ {Ω} ∪ A2
{a1
Remark 3.10. For ξ, η ∈ k, it is easy to check that the family
Xi=1
aξ,η = {(10,s ⊗ ξ) ∧ (1s,s+t ⊗ η) : s, t ≥ 0}
Ω(H), i = 1,· · · , n
(si + ti) = T}.
is a 2−addit for the Clifford super-product system H k
direct verification to see that the subset {aξ,η : ξ, η ∈ k} of A2
2N0. Further it is a
Ω, together
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
11
with Ω, generate the Clifford super-product system. In Section 5 we get
the precise form of a 2−addit.
4. CAR flows and even CAR flows
In this section we recall the E0-semigroups associated with the canonical
anti-commutation relations, and compute their super-product systems.
CAR flows have been already discussed in [Am], [BISS] and [Bk]. But the
problem of showing non-cocycle-conjugacy for CAR flows with different
ranks has still remained open.
Let K be an infinite dimensional separable complex Hilbert space. We
denote by A(K) the CAR algebra over K, which is the universal C∗-
algebra generated by {a(x) : x ∈ K}, where x 7→ a(x) is an antilinear
map satisfying the CAR relations:
(1)
a(x)a(y) + a(y)a(x) = 0,
a(x)a(y)∗ + a(y)∗a(x) = hx, yi 1,
for all x, y ∈ K.
The quasi-free state ωA on A(K), associated with a positive contraction
A ∈ B(K), is the state determined by its 2n-point function as
ωA(a(xn)· · · a(x1)a(y1)∗ · · · a(ym)∗) = δn,m det(hxi, Ayji),
where det(·) denotes the determinant of a matrix. Given a positive con-
traction, it is a fact that such a state always exists and is uniquely deter-
mined by the above relation.
We denote by (HA, πA, ΩA) the GNS triple associated with a quasi-free
state ωA on A(K), and set MA := πA(A(K))′′. Every quasi-free state
ωA of the CAR algebra A(K) is a factor state, (see [PS, Theorem 5.1]).
We summarize standard results on the types of von Neumann algebras
obtained through the quasi-free representations, in the following theorem.
For a proof we refer to [PS, Lemma 5.3], and for (i) we refer to [Arv,
Chapter 13], [Pow2, Section II].
Theorem 4.1. Let K be a Hilbert space, let A ∈ B(K) be a positive
contraction.
(i) MA is of type I if and only if tr(A − A2) < ∞.
(ii) MA is of type II1 if and only if A − 1
(iii) MA is of type II∞ if and only if there exists a spectral projection P
of A, with both P and 1− P are of infinite dimensions, (P AP )2 −
P AP is trace-class and (1 − P )A(1 − P ) − (1 − P )/2 is Hilbert-
Schmidt.
2 is Hilbert-Schmidt.
(iv) MA is of type III otherwise.
12
O. MARGETTS AND R. SRINIVASAN
From here onwards we let K = L2((0,∞), k) and A ∈ B(K) be a
positive contraction satisfying Ker(A) = Ker(1 − A) = {0}. We further
assume that A is a Toeplitz operator, meaning T ∗t ATt = A for all t ≥ 0.
Let j be a conjugation on k, and we continue to denote the conjugation on
K also as j, obtained by (jf )(s) := jf (s) for f ∈ K for s ≥ 0. The GNS
representation associated with the quasi-free state ωA can be concretely
realized on the doubled anti-symmetric Fock space HA = Γ(K)⊗ Γ(K) by
πA(a(f )) = a(√1 − Af ) ⊗ Γ(−1) + 1 ⊗ a∗(j√Af ),
with cyclic and separating vector ΩA = Ω ⊗ Ω, where Ω is the vacuum
vector of Γ(K) and Γ(−1) is the second quantization (see [AWy] and [PS]).
Whenever it is convenient we will use this picture.
If X ∈ B(K) is a positive Toeplitz operator, then √XTt√X−1
extends
to an isometry on H (see [MaS2, Lemma 7.3]). We denote the extended
t )t≥0. Denote Yt = (cid:20)T 1−T
t j(cid:21) , for
semigroup of isometries by (T X
t ≥ 0. Now it follows from [Arv, Proposition 2.1.3], there exists a unique
E0-semigroup {θA
t (a(f )) = a(Ytf ) ∀f ∈
K ⊕ K.
Example 4.2. The restriction of θA to MA provides the unique E0-semigroup
αA = {αA
: t ≥ 0} on B(HA) satisfying θA
: t ≥ 0} on MA, determined by
0
jT T
t
t
0
t
αA
t (πA(a(f ))) = πA(a(Ttf )),
∀f ∈ K.
We call αA as the Toeplitz CAR flow on MA associated with A.
Let Me
A denotes the von Neumann subalgebra generated by the even prod-
ucts of πA(a(f )), πA(a∗(g)). This is the fixed point algebra of the action
given by πA(a(f )) 7→ −πA(a(f )). If we assume tr(A2 − A) = ∞, then this
action is outer and hence MA is a factor. The restriction of αA to Me
A is
called as the Toeplitz even CAR flow associated with A and we denote it
by βA.
By construction, all these E0-semigroups has a faithful normal invari-
ant state given by the normal extension of ωA, which implies the existence
of the canonical unit S = (St)t≥0 defined by the isometric extension of
StxΩ = αt(x)Ω for all x ∈ MA, t ≥ 0. The following Proposition charac-
terises when S is a multi-unit.
Proposition 4.3. For the E0-semigroups αA, βA, the canonical unit is a
multi-unit if and only if A = 1L2(R) ⊗ R for some R ∈ B(k).
Proof. One way is clear. For the other way, S is a multi-unit if and only
if the modular group (σΩ
for all t ≥ 0,
s ∈ R(see [MaS2, Proposition 3.4, (iii)]). The modular automorphism σΩ
s )s∈R satisfies αt = σΩ
−s ◦ αt ◦ σΩ
s
s
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
13
is the Bogoliubov automorphism associated with Ais(1 − A)−is. Hence
Ais(1 − A)−is commute with Tt for all t ≥ 0 and so it is of the form
1L2(R+) ⊗ Us. By considering the (analytic) generator, we infer that A(1−
A)−1 = 1L2(R) ⊗ X, for some densely defined self-adjoint operator X on k.
For f ∈ Dom((1 − A)−1) we have
(1 − A)−1f = (1 + A(1 − A)−1)f = (1L2(R+) ⊗ (1 + X))f.
This implies that (1 + X) has bounded inverse and
The proofs for βA is similar.
A = 1L2(R+) ⊗(cid:0)1 − (1 + X)−1(cid:1) .
(cid:3)
From here onwards, we assume that A is of the form 1L2(R+) ⊗ R for
some R ∈ B(k). Since we have assumed both A, 1 − A are injective (in
particular R2 − R 6= 0), tr(A2 − A) = tr(cid:0)1L2(R+) ⊗ (R2 − R)(cid:1) = ∞, and
hence MA is not of type I. In the same way A− 1
2 is Hilbert-Schmidt if and
only R = 1
2; in that case it is type II1. MA can not be of type II∞, since in
that case there will be a non-trivial subspace of k, which is an eigen space
for R with eigen value 1, which will contradict the assumption that 1 − A
is injective. (Even if we relax the condition 1 − A being injective, we will
only end up with a tensor product of Clifford flow (of even index) with
a CAR flow on type I∞ factor, which are completely classified by [MaS2,
Theorem 6.1].) So with our assumptions when R 6= 1
2, MA is of type III.
We may replace the suffixes A by R in all our notations and call αR, βR
as just CAR flows, even CAR flows respectively.
For m0 ∈ M, it is well-known that the map mΩ 7→ mm0Ω for all m ∈ M
is closable; we denote the closure by ρm0 and its domain by D(ρm0). The
proof of the first part of the following lemma is exactly same as [MaS1,
Lemma 8.3]. The only fact used there, by assuming M to be a II1 factor,
is St being a multi-unit (which was automatic).
Lemma 4.4. Let α be an E0-semigroup acting standardly on M with cyclic
and separating vector Ω satisfying hαt(m)Ω, Ωi = hmΩ, Ωi for all m ∈ M.
If further the canonical unit S = (St)t≥0 is a multi-unit, then
t = [MSt] ∩ αt(M)′St,
H α
where [MSt] denotes the weak operator closure of MSt.
For any A = T St ∈ H α
t , with T ∈ αt(M)′,
T Ω ∈ D(ραt(m)) and ραt(m)(T Ω) = T αt(m)Ω, ∀ m ∈ M.
Proof. Since T St ∈ [MSt], there exists a net {mλ}λ∈Λ ⊆ M such that mλSt
converges strongly to T St. This means {mλΩ} ⊆ D(ραt(m)) converges to
T Ω and mλαt(m)Ω converges to T αt(m)Ω for all m ∈ M, which means
the assertion of the lemma.
(cid:3)
14
O. MARGETTS AND R. SRINIVASAN
In [Bk], the super-product systems of CAR flows were computed, as-
suming few conditions. We provide here a direct proof. Let uR(f ) =
1√2
(πR(a(f )) + πR(a∗(f ))) for f ∈ L2((0,∞), k). Then
uR(f )uR(g) + uR(g)uR(f ) = Rehf, gi , ∀f, g ∈ L2((0,∞), k),
1
1
and MR = {uR(f ) : f ∈ L2((0,∞), k)}′′. Since the range of R
2 is dense,
we can choose a total set {ξk : k ∈ N} ⊆ k such that {R
2 ξk} forms
an orthonormal basis for k. (Choose any countable linearly independent
total subset of k, then its image under R
2 is also linearly independent and
total. Now use Gram-Schmidt orthogonalization.) If {en : n ∈ N} be any
orthonormal basis for L2(0,∞), then it is easy to check that the collection
{uR(ei1 ⊗ ξj1)uR(ei2 ⊗ ξj2)· · · uR(eim ⊗ ξjm)Ω} with 1 ≤ i1 < i2 < · · · < im,
1 ≤ j1 < j2 < · · · < jm, m ∈ N0 forms an orthonormal basis for the
GNS Hilbert space HR (when m = 0 we take Ω). Now it is clear that
the following choices can be made. Pick distinct posets Λ1, Λ2 order
isomorphic to N. Let
1
P ={I = (i1, i2 · · · im) ∈ Λm
F ={F = (j1, j2 · · · jm) ∈ Λm
1 : 1 ≤ i1 < i2 < · · · < im, m ∈ N0};
2 : 1 ≤ j1 < j2 < · · · < jm, m ∈ N0}.
Choose {fi}i∈Λ1 ⊆ L2((0, t), k) and {gj}j∈Λ2 ⊆ L2((t,∞), k), so that
{uR(I)uR(F )Ω : I ∈ P, F ∈ F} forms an orthonormal basis for the GNS
Hilbert space HR, where
u(I) = u(fi1)u(fi2)· · · u(fim); u(F ) = u(gj1)u(gj2)· · · u(gjm).
Proposition 4.5. Let αR and βR be respectively the CAR flow and even
CAR flow associated with A = 1 ⊗ R ∈ B(L2(R+, k). The super-product
system associated with αR and βR are both isomorphic to Ek
2N0.
Proof. Since A is of the form 1 ⊗ R, notice that
span{uR(f1)uR(f2)· · · uR(f2n)Ω : fi ∈ L2((0, t), k} = Ek
2N0(t).
2N0(t), define Tξt(mΩ) = αR
For ξt ∈ Ek
t (m)ξt. Then Tξt extends to a
bounded operator on HR = Γ(K) ⊗ Γ(K), and satisfies Tξtm = αR
t (m)Tξt
)∗ and Tξt(ξ) =
for all m ∈ M (indeed αR
U 2
s,∞(ξt ⊗ Stξ)). The modular conjugation JΩ on Γ(K)⊗ Γ(K) is the anti-
linear extension of ξ1 ∧ · · · ξn ⊗ η1 ∧ · · · ηm 7→ jηm ∧ · · · jη1 ⊗ jξn ∧ · · · jξ1.
It is easy to verify Tξt(m′Ω) = αR
t , and the map
t
ξt 7→ Tξt is isometric. We prove surjectivity as follows.
′(m′)ξt, hence Tξt ∈ H α
(1 ⊗ StXS∗t )(U 2
s,∞
t (X) = U 2
s,∞
Let T St ∈ H α
t , with T ∈ αt(M)′, and there exists a unique expansion
T Ω = XI∈P,J∈F
λ(I, F )uR(I)uR(F )Ω, λ(I, F ) ∈ C.
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
15
For any F ′ ∈ F , thanks to Lemma 4.4, T Ω ∈ D(ρuR(F ′)) , and
T Ω = uR(F ′)T uR(F ′)∗Ω
= uR(F ′)ρuR(F ′)∗(T Ω) (using Lemma 4.4)
λ(I, F )uR(F ′)uR(I)uR(F )uR(F ′)∗Ω
µF ′(I, F )λ(I, F )uR(I)uR(F )Ω,
= XI∈P,F∈F
= XI∈P,F∈F
2N0(t).
where µF ′(I, F ) = (−1)σF ′ (I,F ) with σF ′(I, F ) = IF ′+FF ′−F ∩F ′.
(In the third line we have used the fact that ρuR(F ′) is closed and the
orthogonal sum converges.) Since the expansion is unique we conclude
λ(I, F ) = 0 except for the terms indexed by (I, F ) satisfying I is even
and F is empty. Hence T Ω ∈ Ek
To compute the super product systems of the even CAR flows, observe
that the GNS Hilbert spaces, {Me
RΩ} can be identified with subspaces
2N0(∞). By restricting Tξt defined above, we conclude Ek
Ek
t Ω.
To prove the other inclusion, notice that the collection {uR(I)uR(F )Ω :
I ∈ P, F ∈ F}, with either both I and F have even lengths or both have
odd lengths, provides an ONB for Ek
2N0(∞). Similar arguments as for
CAR flows above, (modified as in the proof of [MaS1X, Proposition 8.16])
will complete the proof.
(cid:3)
2N0(t) ⊆ H β
5. 2-cocycles
In this section we compute the 2−addits for the Clifford super-product
systems, the family of super-product systems H k
2N0(t), discussed in Exam-
ple 2.3. From this computation, we get back the dimension of the Hilbert
space k ⊗ k, as the invariant defined as 2−index. This would prove that
two super-product systems H k
2N0(t) are isomorphic if and only
if k and k′ have same dimension. This also classifies CAR super-product
systems, the family of super-product systems discussed in Example 2.4,
since Ek
2N0(t) is isomorphic to H k⊕k
2N0(t) and H k′
2N0 (t).
We identify L2([0, t], k)⊗n with L2([0, t]n, k⊗n) by the natural isomor-
phism. Notice that the subspace L2([0, t], k)∧n is the collection of functions
f ∈ L2([0, t]n, k⊗n) satisfying
f (sσ(1), . . . sσ(n)) = ǫ(σ)Πσf (s1, . . . , sn),
for any permutation σ ∈ Sn, where Πσ is the corresponding tensor flip on
k⊗n. We use this identification and the following identification in all our
computations.
Notice that Rn ⊖ {(x1, x2, . . . , xn) : xi 6= xj ∀i 6= j} has full Lebesgue
measure, hence that in particular f is completely determined by its value
16
O. MARGETTS AND R. SRINIVASAN
on the simplex ∆n
t := {(s1, . . . , sn) ∈ [0, t]n : s1 > · · · > sn}, and
t
t
kf (s)k2 ds.
kf (σ(s))k2 ds = n!Z∆n
kfk2 =Z[0,t]n kf (s)k2 ds = Xσ∈SnZ∆n
Thus the map f 7→ n!f∆n
induces a Hilbert space isomorphism be-
tween L2([0, t], k)∧n → L2(∆n
t , k⊗n). This identification is also compat-
ible with right shifts, namely it intertwines the isometries L2([0, s], k)∧m⊗
L2([0, t], k)∧n → L2([0, s + t], k)∧n+m given by
f ⊗ g 7→(cid:0)(Tt)⊗mf(cid:1) ∧ g
s , k⊗m) ⊗ L2(∆n
t , k⊗n) → L2(∆m+n
, k⊗n+m)
t
t
with isometries Vs,t
given by
: L2(∆m
Vs,t(f ⊗ g)(s1, . . . , sn+m) = f (s1 − t, . . . , sm − t) ⊗ g(sm+1, . . . , sn+m)
when (s1, . . . , sn+m) ∈ [t, s + t]m × [0, s]n, and otherwise
Vs,t(f ⊗ g)(s1, . . . , sn+m) = 0.
Since the second quantization Γ(Tt) leaves n−particle spaces invariant,
projection of a 2−addit onto an n−particle space is again a 2−addit. So we
are basically looking for elements {a(s, t) : s, t ∈ (0, T )} ⊆ L2([0, T ], k)∧2n,
for arbitrarily fixed n, satisfying
a(r, s + t) = a(r + s, t) + a(r, s) − Sra(s, t), ∀ r, s, t ∈ (0, r + s + t),
loc(R+; k⊗2) such that
where Sr is the restriction of Γ(Tr). We continue to call them as 2−cocycles.
Further appropriate orthogonality conditions should be satisfied for the
them to be defective. Though we are interested only in the anti-symmetric
case, we prove it in a more general case, which may be useful later.
Lemma 5.1. For a defective 2−cocycle a(s, t) ∈ L2([0, s + t)], k)⊗2 there
exists f1, f2 ∈ L2
a(s, t)(x, y) = 1[s,s+t]×[0,s](x, y)f1(x−y)+1[0,s]×[s,s+t](x, y)f2(y−x), ∀ s, t, x, y ≥ 0.
Proof. The 2−cocycle {a(s, t) : s, t,≥ 0} is defective means
a(s, t) ⊥ Us,t(Hs⊗Ht) = L2((cid:16)([0, s] × [0, s])[([s, s + t] × [s, s + t])(cid:17) , k⊗2),
for all s, t ≥ 0. So the support of a(s, t) must be contained in ([s, s + t] ×
[0, s]) ∪ ([0, s] × [s, s + t]). Now, it follows from the relation
a(r, s + t) = a(r + s, t) + a(r, s) − Sra(s, t),
by looking at the various supports, that
a(r, s + t)[r,r+s]×[0,r] = a(r, s)[r,r+s]×[0,r]
a(r + s, t)[r+s,r+s+t]×[0,r] = a(r, s + t)[r+s,r+s+t]×[0,r],
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
17
for all r, s, t ≥ 0, and also that the same first and second relations hold true
as well in the region [0, r]×[r, r +s] and [0, r]×[r +s, r +s+t] respectively.
, k⊗2)
These equations separately imply the existence of g1, g2 ∈ L2
which together satisfies
loc(∆2
∞
a(s, t)(x, y) = 1[s,s+t]×[0,s](x, y)g1(x, y) + 1[0,s]×[s,s+t](x, y)g2(y, x),
for all s, t ≥ 0.
We claim that gi(x + s, y + s) = gi(x, y) for all x, y, s > 0, i = 1, 2; hence
g1(x, y) = f1(x−y) and g2(x, y) = f2(y−x) for some f1, f2 ∈ L2
loc(R+, k⊗2).
Indeed, by picking (x, y) ∈ [r + s, r + s + t] × [r, r + s], and evaluating the
functions in the following equality in this region
a(r + s, t) + a(r, s) = a(r, s + t) + Sra(s, t)
at (x, y), one finds that g1(x, y) = g1(x− r, y − r), as required. By picking
(x, y) ∈ [r, r + s] × [r + s, r + s + t], we get the relation for g2 and proof is
over.
(cid:3)
We write A ⋋ B, for the simplex {(s, t) :
s ∈ A, t ∈ B, s > t}.
Similarly, A⋋n will be used to denote the A ⋋ A· · · ⋋ A, where there are
n copies of A in the product
Proposition 5.2. Let {as,t : s, t ≥ 0} be a defective 2−addit for (H k
Then as,t ∈ L2([0, s + t)], k)∧2. Further there exists f ∈ L2
that
2N0(t), Us,t).
loc(R+, k⊗2) such
as,t(x, y) = 1[s,s+t]×[0,s](x, y)f (x − y) − 1[0,s]×[s,s+t](x, y)Πk⊗2f (y − x),
for all s, t, x, y ∈ (0,∞), where Πk⊗2 is the usual tensor-flip.
Proof. For S ⊆ [0, s + t], if PS is the projection in B(L2([0, s + t)], k)⊗2)
onto L2(S, k)⊗2, then PS (L2([0, s + t, k)∧2) = L2(S, k)∧2. Using this, it is
easy to verify that a defective 2−cocycle in L2([0, s + t)], k)∧2 continues
to be a defective cocycle in L2([0, s + t)], k)⊗2. By restricting the case of
Lemma 5.1 to the antisymmetric subspace, it is immediate from Equation
5, that any 2−addit as,t ∈ L2([0, s + t)], k)∧2 is of the form mentioned
in the proposition. So we only have to show that there are no defective
2−cocycles in the 2N-particle space when N > 1.
Pick an integer N > 1, and let a(s, t) be a defective 2-cocycle in the
2N-particle subspace L2([0, s + t)], k)∧2N . Since a(s, t) is defective, we
must have that a(s, t) is orthogonal to
N
Mn=0
L2([s, s + t]⋋2n, k⊗2n) ⊗ L2([0, s]⋋2(N−n), k⊗2(N−n))
=L2 N
[s, s + t]⋋2n × [0, s]⋋2(N−n), k⊗2N!
[n=0
18
O. MARGETTS AND R. SRINIVASAN
where for succinctness, we have abused notation in the cases n = 0, 2N.
Thus, up to a set of measure zero, the support Σ(s, t) of a(s, t) satisfies
N
Σ(s, t) ⊆
[s, s + t]⋋2n−1 × [0, s]⋋(2(N−n)+1)
[n=1
[s, s + t]⋋2(n−1) × [0, s]⋋2(N−n)! ⋋ [0, s].
= [s, s + t] ⋋ N
[n=1
Let Σ0(s, t) be the collection of points obtained by ignoring the first and
last coordinates of points in Σ(s, t), i.e.
Σ0(s, t) := {(s2, . . . , sN−1) : ∃s1,sN >0 (s1, s2, . . . , sN−1, sN ) ∈ Σ(s, t)}.
We will show that Σ0(s, t) has measure zero. We have Σ0(s, t) ⊆ A(s, t)
where we set
N
A(s, t) :=
[n=1
[s, s + t]⋋2(n−1) × [0, s]⋋2(N−n).
The cocycle identity, a(r, s+t) = a(r +s, t)+a(r, s)−Sra(s, t), ∀r, s, t > 0
asserts that
Σ0(r, s + t) ⊆ A(r, s + t) ∩(cid:0)(A(s, t) + r) ∪ A(r + s, t) ∪ A(r, s)(cid:1).
Now we set
B(r, s, t) := A(r, s + t) ∩ A(r + s, t)
N
=
=
N
[n=1
[r, r + s + t]⋋2(n−1) × [0, r]⋋2(N−n)
[n=1
[r + s, r + s + t]⋋2(n−1) × [0, r + s]⋋2(N−n)
∩
[k=1
[n=1
N
n
[r + s, r + s + t]⋋2(k−1) × [r, r + s]⋋2(n−k) × [0, r]⋋2(N−n).
Further notice that
N
(A(s, t) + r) =
[n=1
[r + s, r + s + t]⋋2(n−1) × [r, r + s]⋋2(N−n) ⊆ A(r + s, t),
A(r, s + t) ∩ A(r, s) = A(r, s) ⊆ B(r, s, t),
so we must have Σ0(r, s + t) ⊆ B(r, s, t). Since this holds for all r, s, t > 0,
we have
Σ0(r, u) ⊆ ∩{B(r, s, t) : s, t ∈ (0,∞), s + t = u}.
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
19
Now if we set
C(n, r, s, t) :=
n
[k=1
[r + s, r + s + t]⋋2(k−1) × [r, r + s]⋋2(n−k) × [0, r]⋋2(N−n)
then, by looking at the components of [0, r], note for any s1, s2, t1, t2 ∈
(0,∞) with s1 + t1 = u = s2 + t2, that
C(n, r, s1, t1) ∩ C(m, r, s2, t2) = ∅ if n 6= m,
so Σ0(r, u) is contained in
N
[n=1
∩{C(n, r, s, t) : s, t ∈ (0,∞), s + t = u}.
But for any n = 2, . . . , N and any s1 > s2 > . . . > s2N−2 such that
(s1, . . . , s2N−2) ∈ C(n, r, s, t), we have s1 /∈ [r, (s1+s2)
] and s2 /∈ [ (s1+s2)
s + t], so that
2
2
, r+
(s1, . . . , s2N−2) /∈ C(n, r,
(s1 + s2)
2
− r, r + s + t −
(s1 + s2)
2
);
so it follows that Σ0(r, u) ⊆ [0, r]⋋2(N−1).
We claim that Σ0(r, u) ⊆ S2n−1
]⋋2(N−1) for all n ∈ N0. We
prove the claim by induction. The claim is true when n = 0. Assuming
the claim for n and using
r
2
2n , (k+1)r
a(r, u) = a(
) − (T r
+ u) + a(
k=0 [ kr
)2N a(
, u),
r
2
r
2
2n
,
r
2
,
r
2
2n+1 , (k+1)r
[ kr
2
2n+1 ]⋋2(N−1). So Σ0(r, u) must be a
(cid:3)
we obtain Σ0(r, u) ⊆ S2n+1−1
set of measure zero.
k=0
The proof of the following Proposition follows form Proposition 5.2 and
2N0(t) and H k⊕k
s,t ⊗ Ω2) + (Ω1 ⊗ a2
Lemma 5.1, using the isomorphism between Ek
2N0 (t).
Proposition 5.3. Let {as,t : s, t ≥ 0} be a defective 2−addit for (Ek
Then as,t = (a1
s,t, with a1, a2 ∈ L2([0, s + t)], k)∧2
and a0
s,t ∈ L2([0, s + t)], k)⊗ L2([0, s + t)], k), where Ω1 and Ω2 are vacuum
vectors of the first and second Fock spaces respectively. Further there exist
f 1, f 2, f1, f2 ∈ L2
s,t(x, y) = 1[s,s+t]×[0,s](x, y)f i(x−y)−1[0,s]×[s,s+t](x, y)Πk⊗2f i(y−x), i = 1, 2
ai
loc(R+, k⊗2) such that
s,t) + a0
2N0(t), Us,t).
a12
s,t(x, y) = 1[s,s+t]×[0,s](x, y)f1(x − y) + 1[0,s]×[s,s+t](x, y)f2(y − x),
for all s, t, x, y ∈ (0,∞), i = 1, 2.
20
O. MARGETTS AND R. SRINIVASAN
Definition 5.4. Let (Ht, Us,t) be spatial super-product system with canon-
ical unit Ω. We say a 2−addit {as,t : s, t ≥ 0} is orthogonal to another
2−addit {bs,t : s, t ≥ 0} if as,t ⊥ bs,t for all s, t ≥ 0.
The 2−index with respect to Ω is defined as the supremum of the cardi-
nality of all sets containing mutually orthogonal 2−addits.
For f ∈ L2
s,t.
Clearly 2−index is an invariant under isomorphism preserving the canon-
ical unit. If the automorphism group of the super product system acts
transitively on the set of all units, then the 2−index do not depend on a
particular unit and it is an invariant for the super-product system. In par-
ticular when the unit is unique up to scalars, the 2−index is an invariant,
which is the case for Clifford and CAR super-product systems.
loc(R+, k⊗2) we denote the 2−addit described in Proposition
loc(R+, k⊗2) and T ∈ (0,∞], af
5.2, by af
Lemma 5.5. For f, g ∈ L2
s,t for all
s, t ∈ (0,∞) with s + t ≤ T if and only if f (r) ⊥ g(r) for almost all
r ∈ (0, T ).
Proof. For f, g ∈ L2
Z s+t
if and only if f (r) ⊥ g(r) for almost all r ∈ (0, T ). One way is clear. For
the other way, fix ε > 0. For any (s, t) ∈ (0, ε) × (0, T − ε), we have
Z s
0 hf (p − q), g(p − q)i dqdp = 0 ∀s, t ∈ (0,∞) with s + t ≤ T
loc(R+, k⊗2), we only have to prove that
s,t ⊥ ag
s
s
0 =Z s+t
Z s
0 hf (p − q), g(p − q)i dqdp
= −Z t
0 Z s
0 hf (u + v), g(u + v)i dvdu ,
so that hf (t + s), g(t + s)i = 0 for almost all (s, t) ∈ (0, ε)× (0, T − ε), for
all ǫ > 0. That is, hf (r), g(r)i = 0 for almost all r ∈ (0, T ).
Theorem 5.6. The 2−index of (H k
super-product systems (H k
only if dim(k) = dim(k′).
Also 2−index of (Ek
2N0(t), Us,t) is n2. Consequently the
2N0(t), Us,t) if and
2N0(t), Us,t) is 4n2. Consequently (Ek
2N0(t), Us,t) if and only if dim(k) = dim(k′).
2N0(t), Us,t) is isomorphic to (H k′
isomorphic to (Ek′
2N0(t), Us,t) is
(cid:3)
Proof. Pick an orthonormal basis (ei)n
i=1 for k (possibly with n = ∞) and
define functions fij ∈ L2
loc(R+, k⊗2) by fij(r) := ei ⊗ ej for all r ≥ 0. Let
aij denote the 2-cocycle with symbol fij, then it is clear that the aij are
pairwise locally orthogonal. Hence the 2−index is greater than or equal
to n2.
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
21
Now assume that we have a set of n2 orthogonal 2-cocycles a1, . . . , an2
with symbols f1, . . . , fn2. Then, by Lemma 5.5, f1, . . . , fn2 are orthogonal
a.e. Thus, for almost all r ≥ 0, (fi(r))n2
i=1 is an ortho-basis for k⊗2. If a is
a 2-cocycle with symbol f which is orthogonal to a1, . . . , an2 then, again
by Lemma 5.5, we must have f (r) ⊥ f1(r), . . . , fn2(r) for almost all r ≥ 0,
i.e. f = 0.
(cid:3)
The above Proposition provides a direct proof that Clifford flows on
type II1 factors are non-cocycle if they have different ranks. We further
have the following new result for the CAR flows on type III factors. Notice
under the assumption that R and 1 − R are injective Rank(R) = dim(k).
Corollary 5.7. Let αR1, αR2 and βR1, βR2 be respectively the CAR flows
and even CAR flows associated with 1 ⊗ R1, 1 ⊗ R2 ∈ B(L2(R+, k). Then
αR1 is not cocycle conjugate to αR′ if R1 and R2 have different ranks. Sim-
ilarly βR1 is also not cocycle conjugate to βR2 if R1 and R2 have different
ranks.
6. The automorphism group
To conclude the paper, we calculate the automorphism group of the
Clifford super-product system H k
2N0, which is clearly an invariant of the
super-product system. We show that it is much larger than the gauge
group of the corresponding Clifford flow, which was shown in [MaS1] to
be trivial.
Let M(R+;U(k⊗2)) denote the group of all measurable, unitary-valued
functions endowed with pointwise multiplication. For a given λ ∈ R and
F ∈ M(R+;U(k⊗2)) define
U(λ,F )(t)Ωt = eiλtΩt,
(U(λ,F )(t)f )(x, y) = eiλtF (x − y)f (x, y)
for all x, y ∈ [0, t], t > 0, f ∈ L2([0, t]; k)∧2, where Ωt is the vacuum vector
in H k
2N0(t). Now extend U(λ,F )(t) to L2([0, t]; k)∧2n by
U(λ,F )(t) (f1 ∧ · · · ∧ fn) = U(λ,F )(t)f1 ∧ · · · ∧ U(λ,F )(t)fn
for f1,· · · fn ∈ L2([0, t]; k)∧2. Clearly U(λ,F )(t) extends to a unitary oper-
ator on H k
2N0(t).
We denote the automorphism group of H k
2N0 by Aut(H k
2N0).
Theorem 6.1. Aut(H k
2N0) is isomorphic to (R, +) × M(R+;U(k⊗2)).
Proof. It is clear that the family {U(λ,F )(t) : t ≥ 0} provides an automor-
phism for any fixed (λ, F ), and that the map (λ, F ) 7→ U(λ,F ) induces a
homomorphism from (R, +) × M(R+;U(k⊗2)) into Aut(H k
2N0). To prove
injectivity, if the automorphism determined by (λ, F ) coincides with that
22
O. MARGETTS AND R. SRINIVASAN
of (µ, G), then the action on the vacuum ensures us that λ = µ and,
picking u, v ∈ k⊗2, T > 0 and setting
f = 1{0≤s<t≤T} ⊗ (u ⊗ v) − 1{0≤t<s≤T} ⊗ (v ⊗ u),
we obtain F (s − t)(u ⊗ v) = G(s − t)(u ⊗ v) for almost all 0 ≤ t < s ≤ T ,
but since u, v and T were arbitrary, F = G. Thus it remains only to show
that the given homomorphism is a surjection.
Since any automorphism θ ∈ Aut(H k
t we obtain an automorphism which preserves the unit Ω.
2N0) preserves units, it must satisfy
θt(Ωt) = eiλtΩt for some λ ∈ R. Thus, by setting θ′t(x) := e−iλtθt(x) for all
t > 0, x ∈ H α
Since θ′ preserves defective 2-cocycles, there exists a linear bijection
X on L2
loc(R+; k⊗2) such that θ′s+t(af (s, t)) = aXf (s, t). The equality
haf (s, t), ag(s, t)i = haXf (s, t), aXg(s, t)i is equivalent to
Z t
0 Z s
0 hf (p + q), g(p + q)i dqdp =Z t
0 Z s
0 h(Xf )(p + q), (Xg)(p + q)i dqdp
for all s, t > 0 (see the proof of Lemma 5.5). Thus
hf (r), g(r)i = h(Xf )(r), (Xg)(r)i , for almost all r > 0.
This implies that X restricts to an isometry X on L2(R+; k⊗2). On the
other hand (θ′)−1 implements the bijection X−1, which also restrict to an
isometry on L2(R+; k⊗2), and hence X is unitary.
We claim that X commutes with the orthogonal projections
Pt : L2(R+; k⊗2) → L2([0, t]; k⊗2),
f 7→ f[0,t]
for all t ≥ 0. To see this, note that if f ∈ Ker PT then af (s, t) ⊥ ag(s, t)
for all g ∈ L2
loc(R+; k⊗2) and s + t ≤ T . Hence aXf (s, t) ⊥ ag(s, t) for
loc(R+; k⊗2) and s + t ≤ T which implies, by Lemma 5.5, that
all g ∈ L2
Xf ∈ Ker PT , that is X(1 − PT )H ⊆ (1 − PT )H. Similarly, on the other
hand, X−1(1−PT )H ⊆ (1−PT )H and X−1 = X∗. Hence the claim. Thus,
we can identify X with an element of (L∞(R+) ⊗ 1)′ = L∞(R+; B(k⊗2)).
Since X is a unitary, it follows that X is given by an F ∈ M(R+;U(k⊗2)).
Now since the 2−addits generate the super-product system (see Remark
3.10), the automorphism θ is determined by its action on the 2−addits
and the automorphisms are as claimed.
(cid:3)
Remark 6.2. It is apparent that Aut(H k
2N0) consists of far more than the
restrictions of automorphisms of the Fock product system. The automor-
phisms of the Fock product system which leave the even subspaces invariant
are all of the form
θs(Ωt) = eiλsΩt, θs(f ) = eiλs(1L2([0,t]) ⊗ U)f (s, t > 0, f ∈ L2([0, t]; k)),
COHOMOLOGY FOR SPATIAL SUPER-PRODUCT SYSTEMS
23
for some λ ∈ R, U ∈ U(k. Thus they form a group isomorphic to (R, +)×
U(k).
References
[Ale] A. Alevras, One parameter semigroups of endomorphisms of factors of type II1,
J. Op. Thy., 51 (2004), 161-179.
[Am] Grigori G. Amosov, On cocycle conjugacy of quasi-free endomorphsms semi-
groups on the CAR algebra, Journal of Mathematical Sciences 07/2001;
105(6):2496-2503. DOI: 10.1023/A:1011304214659
[Ar1] H. Araki, A lattice of von Neumann algebras associated to the quantum theory
of the free Bose field, J. Math. Phys., 4 (1963), 1343–1362.
[Ar2] - - , Type of von Neumann algebra associated with free field, Progr. Theoret.
Phys. 32 (1964) 956-965.
[Ar3] - -, Some properties of modular conjugation operator of von Neumann alge-
bras and a non-commutative Radon-Nikodym theorem with a chain rule, Pacific
J. Math. 50 2 (1974), 309-354.
[AWd1] H. Araki, E. J. Woods, Complete boolean algebras of type I factors. Publ.
RIMS Kyoto University Ser. A 2 (1966), 157-242.
[AWd2] H. Araki, E.J. Woods, A classification of factors, Publ. RIMS Kyoto University
Ser. A Vol. 3 (1968), 51-130.
[AWy] H. Araki, W. Wyss, Representations of canonical anticommunication relations,
Helv. Phys. Acta 37 (1964) 139-159.
[ArY] H. Araki, S. Yamagami, On quasi-equivalence of quasifree states of the canonical
commutation relations, Publ. RIMS, Kyoto Univ. 18 (1982), 283-338.
[Arv] W. Arveson, "Noncommutative dynamics and E-semigroups", Springer mono-
graphs in mathematics, Springer, New York-Heidelberg 2003.
[Bak] B.M. Baker, Free states of the gauge invariant canonical anticommutations rela-
tions, Transactions of the American Mathematical Soceity, Volume 237, March
1978.
[BhS] B. V. Rajarama Bhat and R. Srinivasan, On product systems arising from sum
systems Infinite dimensional analysis and related topics, Vol. 8, Number 1,
March 2005.
[BISS] Panchugopal Bikram, Masaki Izumi, R. Srinivasan and V.S. Sunder, On Extend-
ability of Endomorphisms and E0-semigroups of factors, to appear in Kyushu
Journal of Mathematics.
Panchugopal Bikram, Non-extendable endomorphisms and E0-semigroups on
Type III factors,In?nite Dimensional Analysis, Quantum Probability and Re-
lated Topics Vol. 17,No. 4 (2014)
[Bk]
[BrR] O. Bratteli and D.W. Robinson, "Operator Algebras and Quantum Statisti-
cal Mechanics I, C ∗- and W ∗-algebras, Symmetry Groups, Decomposition
of States," Texts and Monographs in Physics, Springer-Verlag, New York-
Heidelberg, 1979.
[Con] A. Connes, Une classification des facteurs de type III, Annales scientifiques de
l ´E.N.S. 4e s´erie, tome 6, no 2 (1973), 133-252.
[Cnw] J.B. Conway, "A Course in Functional Analysis," Second Edition, Graduate
Texts in Mathematics, Springer, New York-Heidelberg, 1990.
[DAn] G.F. Dell'Antonio, Structure of the algebras of some free systems, Comm. Math.
Phys., 9 (1968), 81-117.
24
O. MARGETTS AND R. SRINIVASAN
[Hol] A.S. Holevo Quasifree states of the C ∗-algebra of CCR. II, Theoretical and
Mathematical Physics, Vol. 6, Issue 2 (1971), 103-107.
[IS1] M. Izumi, R. Srinivasan, Generalized CCR flows, Comm. Math. Phys., 281
(2008), 529-571.
[IS2] M. Izumi, R. Srinivasan, Toeplitz CAR flows and type I factorizations, Kyoto
J. Math., 50, no. 1 (2010), 1-32.
[Kat] T. Kato, "Perturbation theory for linear operators," Classics in Mathematics,
Springer-Verlag, Berlin-Heidelberg, 1995.
[VL] V. Liebscher: Random sets and invariants for (type II) continuous tensor prod-
uct systems of Hilbert spaces, Mem. Amer. Math. Soc., 199 (2009)
[MaS1] O. Margetts, R. Srinivasan, Invariants for E0-semigroups on II1 factors,
preprint, to appear in Comm. Math. Phys.
[MaS1X] O. Margetts, R. Srinivasan, arXiv:1209.1283 [math.OA]
[MaS2] O. Margetts, R. Srinivasan, Non-cocycle-conjugate E0-semigroups on factors,
preprint.
[Par] K. R. Parthasarathy, "An Introduction to Quantum Stochastic Calculus,"
Birkauser Basel, Boston, Berlin (1992).
[Pet] D. Petz, "An Invitation to the Algebra of Canonical Commutation Relations,"
Leuven Notes in Mathematical and Theoretical Physics. Series A: Mathematical
Physics 2 Leuven University Press, 1990.
Can. J. Math. 40, no. 1 (1988), pp. 86-114.
[Pow1] R. Powers, An index theory for ∗-endomorphisms of B(H) and type II1 factors,
[Pow2] R. T. Powers, A nonspatial continuous semigroup of ∗-endomorphisms of B(H).
[PS] R. T. Powers, and E. Størmer, Free States of the Canonical Anticommutation
Publ. Res. Inst. Math. Sci. 23 (1987), 1053-1069.
Relations. Comm. Math. Phys., 16 (1970), 1-33.
[Tsi] B. Tsirelson, Non-isomorphic product systems. Advances in Quantum Dynamics
(South Hadley, MA, 2002), 273–328, Contemp. Math., 335, Amer. Math. Soc.,
Providence, RI, 2003.
Department of Mathematics and Statistics, Fylde College, Lancaster
University, Lancaster LA1 4YF, U.K.
E-mail address: [email protected]
Chennai Mathematical Institute, H1, SIPCOT IT Park, Kelambakkam,
Siruseri 603103, India.
E-mail address: [email protected]
|
1711.04721 | 3 | 1711 | 2019-04-24T18:48:19 | Ranks of operators in simple C*-algebras with stable rank one | [
"math.OA"
] | Let $A$ be a separable, unital, simple C*-algebra with stable rank one. We show that every strictly positive, lower semicontinuous, affine function on the simplex of normalized quasitraces of $A$ is realized as the rank of an operator in the stabilization of $A$.
Assuming moreover that $A$ has locally finite nuclear dimension, we deduce that $A$ is $\mathcal{Z}$-stable if and only if it has strict comparison of positive elements. In particular, the Toms-Winter conjecture holds for separable, unital, simple, approximately subhomogeneous C*-algebras with stable rank one. | math.OA | math |
RANKS OF OPERATORS IN SIMPLE C∗-ALGEBRAS WITH
STABLE RANK ONE
HANNES THIEL
Abstract. Let A be a simple C ∗-algebra with stable rank one. We show that
every strictly positive, lower semicontinuous, affine function on the simplex
of normalized quasitraces of A is realized as the rank of an operator in the
stabilization of A.
Assuming moreover that A has locally finite nuclear dimension, we deduce
that A is Z-stable if and only if it has strict comparison of positive elements.
In particular, the Toms-Winter conjecture holds for simple, approximately
subhomogeneous C ∗-algebras with stable rank one.
1. Introduction
The rank of a matrix is one of the most fundamental notions in linear alge-
bra. In this context, the following two standard facts concerning matrix ranks are
repeatedly used:
• Comparison: We have rk(x) ≤ rk(y) if and only if x = rys for some r, s.
• Range: n × n-matrices realize the ranks 0, 1, 2, . . . , n.
The rank of a projection can be computed as its trace. Similarly, the rank of
a projection in the algebra B(H) of bounded operators on a separable, infinite-
dimensional Hilbert space H is computed by the canonical (unbounded) trace tr.
If two projections p and q satisfy p = rqs for some r, s, then one says that p is
Murray-von Neumann subequivalent to q, denoted p - q. The following facts hold
for projections in B(H):
• Comparison: We have tr(p) ≤ tr(q) if and only if p - q.
• Range: Projections in B(H) realize the ranks 0, 1, 2, . . . , ∞.
There is no bounded trace on B(H), but Murray and von Neumann discovered
the class of II1 factors, which are simple, infinite-dimensional von Neumann algebras
such that the unit is a finite projection. Every II1 factor has a unique bounded
trace (normalized at the unit). Murray and von Neumann proved the following
fundamental facts about projections in a II1 factor M :
(C) Comparison: We have tr(p) ≤ tr(q) if and only if p - q.
(R) Range: For every t ∈ [0, 1] there is a projection p ∈ M with tr(p) = t.
It is natural to ask whether these two properties have analogues for simple, tracial
C∗-algebras. The analogue of (C) is called strict comparison (see Remarks 9.2 for
the definition), and it is known not to hold automatically, unlike for II1 factors.
Date: April 26, 2019.
2010 Mathematics Subject Classification. Primary 46L05; Secondary 06B35, 06F05, 19K14,
46L35, 46L80.
Key words and phrases. simple C*-algebra, rank of operator, stable rank one, Cuntz semigroup,
Toms-Winter conjecture, quasitrace, dimension function.
The author was partially supported by the Deutsche Forschungsgemeinschaft (SFB 878 Groups,
Geometry & Actions).
1
2
HANNES THIEL
On the other hand, it is not known if the analogue of (R) is automatic for
simple C∗-algebras. The main result of this paper is that it is automatic for simple
C∗-algebras with stable rank one.
Recall that a unital C∗-algebra is said to have stable rank one if its invertible
elements form a dense subset. The stable rank is a noncommutative dimension
theory that was introduced to study nonstable K-theory. Just as topological spaces
are more tractable if they have low dimension, C∗-algebras with minimal stable rank
(that is, stable rank one) are accessible to techniques that do not apply in general.
Let us describe the C∗-algebraic analogue of (R). While II1 factors have a unique
normalized trace, this is no longer the case for simple C∗-algebras. This means that
there is more than one way to 'measure' the rank of a projection. Given a general
unital, simple C∗-algebra A, the normalized traces on A form a Choquet simplex
T1(A). The rank of a projection p in A is defined as the function rk(p) : T1(A) →
[0, 1], given by rk(p)(τ ) := τ (p). The question is then which functions T1(A) → [0, 1]
are realized as the rank of a projection in A. Given a II1 factor M , the space T1(M )
is a singleton. Therefore, property (R) says that every function T1(M ) → [0, 1] is
realized by a projection in M .
To formulate the accurate C∗-algebraic analogue of (R), we have to apply two
changes. First, we have to replace traces on A by 2-quasitraces (see Paragraph 2.14
for definitions). This is a minor change that can be ignored in many important
cases. For example, for exact C∗-algebras there is no distinction between traces
and 2-quasitraces. Second, we need to replace projections in A by positive elements
in the stabilization A ⊗ K. This change is more fundamental and cannot be ignored
in applications. The reason is that many interesting C∗-algebras have no nontrivial
projections. In this case it is necessary to consider the rank of more general elements
to obtain the correct analogue of (R).
As for traces, the normalized 2-quasitraces form a Choquet simplex QT1(A).
Given a positive element x in A ⊗ K, the rank of x at τ ∈ QT1(A) is defined as
dτ (x) = lim
n→∞
τ (x1/n).
We call the resulting map rk(x) : QT1(A) → [0, ∞], given by rk(x)(τ ) := dτ (x),
the rank of x. The function rk(x) is lower semicontinuous and affine. If x 6= 0,
then rk(x) is also strictly positive, and we write rk(x) ∈ LAff(QT1(A))++; see
Paragraph 3.1. Thus, the precise question is:
Question 1.1. Let A be a separable, unital, simple, non-elementary, stably finite
C∗-algebra, and let f ∈ LAff(QT1(A))++. Is there x ∈ (A ⊗ K)+ with rk(x) = f ?
This question was first explicitly posed by N. Brown. Affirmative answers have
been obtained in the following cases:
(1) under the additional assumption that A tensorially absorbs the Jiang-Su al-
gebra Z, by Elliott-Robert-Santiago [ERS11, Corollary 6.8], extending ear-
lier work by Brown-Perera-Toms that covered the tensorially Z-absorbing,
exact case [BPT08, Theorem 5.5];
(2) under the assumption that A is exact, has strict comparison of positive ele-
ments, and such that QT1(A) is a Bauer simplex whose extreme boundary
has finite covering dimension by Dadarlat-Toms [DT10, Theorem 1.1].
In general, Question 1.1 is still open. The main result of this paper provides an
affirmative answer under the assumption that A has stable rank one:
Theorem (8.11). Let A be a separable, unital, simple, non-elementary C∗-algebra
with stable rank one. Then for every f ∈ LAff(QT1(A))++ there exists x ∈ (A⊗K)+
with rk(x) = f .
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
3
An important regularity property in the theory of simple C∗-algebras is tensorial
absorption of the Jiang-Su algebra Z: A C∗-algebra A is said to be Z-stable if
A ∼= A ⊗ Z. This is the C∗-algebraic analogue of being a McDuff factor. The
Jiang-Su algebra is a separable, unital, simple, non-elementary C∗-algebra that is
KK-equivalent to C.
By [Rør04, Theorem 6.7], every unital, simple, stably finite, Z-stable C∗-algebra
has stable rank one. Thus, Theorem 8.11 generalizes the results in [ERS11] and
[BPT08] mentioned above. However, the assumption of stable rank one is much less
restrictive than that of Z-stability, and it can in fact be used to prove Z-stability;
see for example Theorem 9.5.
Moreover, vast classes of naturally occurring simple, non-Z-stable C∗-algebras
have stable rank one, including a wealth of non-classifiable, nuclear C∗-algebras,
[EHT09, Vil98, Tom08], reduced group C∗-algebras of free products, [DHR97], and
crossed products of dynamical systems with a Cantor factor, [AP15, GK10].
Theorem 8.11 has important consequences for the structure of simple, nuclear
C∗-algebras with stable rank one. In particular, it provides the first verification of
the Toms-Winter conjecture for a large class of C∗-algebras without restriction on
the geometry of the simplex of traces. The Toms-Winter conjecture predicts that
three regularity properties of very different natures are equivalent for a separable,
unital, simple, non-elementary, nuclear C∗-algebra A:
(1) A has finite nuclear dimension.
(2) A is Z-stable.
(3) A has strict comparison of positive elements.
We discuss these conditions and the previously known implications in Section 9.
The condition of finite nuclear dimension is of great importance since it leads to
classification by K-theoretic and tracial data. The completion of the classification
program is one of the great achievements in operator algebras, obtained in a series of
remarkable breakthroughs in the last three years, building on an extensive body of
work over decades by numerous people; see [EGLN15] and [TWW17, Corollary D].
However, in applications it is sometimes difficult to verify finite nuclear dimension
directly. A positive solution of the Toms-Winter conjecture allows finite nuclear
dimension to be deduced from relatively simple, verifiable conditions such as strict
comparison of positive elements. We obtain the following partial verifications of
the Toms-Winter conjecture:
Theorem (9.5). Let A be a separable, unital, simple, non-elementary C∗-algebra
with stable rank one and locally finite nuclear dimension. Then A is Z-stable if and
only if A has strict comparison of positive elements.
Theorem (9.6). The Toms-Winter conjecture holds for approximately subhomoge-
neous C∗-algebras with stable rank one.
In particular, we obtain that separable, unital, simple, approximately subhomo-
geneous C∗-algebras with stable rank one and strict comparison of positive elements
are classified by K-theoretic and tracial data.
Theorems 9.5 and 9.6 apply to large and natural classes of C∗-algebras. For
instance, no nuclear C∗-algebra is known which does not have locally finite nuclear
dimension. Further, all previous partial verifications of the Toms-Winter conjecture
required restrictions on the geometry of the simplex of traces. The most general
results so far assumed that the traces form a Bauer simplex with finite-dimensional
extreme boundary. The novelty of our result is that they have no restrictions on
the geometry of the trace simplex.
Methods. The main tool to obtain the results of this paper is the Cuntz semigroup.
It provides a convenient way to organize the comparison theory of positive elements
4
HANNES THIEL
in a C∗-algebra into a positively ordered monoid. Coward-Elliott-Ivanescu, [CEI08],
initiated a systematic study of Cuntz semigroups by introducing the category Cu of
abstract Cuntz semigroups, also called Cu-semigroups. We refer to Subsections 2.1
and 2.2 for details.
Let us explain how to translate Question 1.1 into the setting of Cu-semigroups.
Given a separable, unital, simple, non-elementary, stably finite C∗-algebra A, its
Cuntz semigroup S := Cu(A) is a countably based, simple, non-elementary, stably
finite Cu-semigroup satisfying certain axioms (O5) and (O6), and the class u :=
[1] is a compact, full element in S. The simplex K of normalized 2-quasitraces
on A can be identified with the space of functionals λ : S → [0, ∞] that satisfy
λ(u) = 1; see Paragraph 2.14. Given a ∈ S, we obtain rk(a) ∈ LAff(K) defined by
rk(a)(λ) := λ(a). For x ∈ (A ⊗ K)+, we have rk(x) = rk([x]). Thus, an affirmative
answer to Question 1.1 follows directly from one to the following:
Question 1.2. Let S be a countably based, simple, non-elementary, stably finite
Cu-semigroup satisfying (O5) and (O6), let u ∈ S be a compact, full element,
let K denote the simplex of functionals on S that are normalized at u, and let
f ∈ LAff(K)++. Does there exist a ∈ S with rk(a) = f ?
Let us outline our approach to answer Question 1.2. Consider the set R :=
{rk(a) : a ∈ S, a 6= 0} of ranks that are realized by nonzero elements in S. We want
to verify R = LAff(K)++. For this, we use the basic order theoretic properties of
LAff(K), which we study in Section 3. We proceed in four steps:
First, we show that S 'realizes chisels': Given λ in the extreme boundary ∂eK,
we let σλ ∈ LAff(K) be the function that takes value 0 at λ and value ∞ elsewhere.
Given t > 0, we call t + σλ the chisel at λ with value t; see Definition 5.3. Under
certain conditions on S, we have t + σλ ∈ R, for every λ ∈ ∂eK and t > 0; see
Proposition 5.4. This step does not require stable rank one; see Theorem 5.5.
Second, we show that S 'realizes functional infima': If f, g ∈ R, then f ∧ g ∈ R;
see Theorem 7.5. It is at this step that the assumption of stable rank one is needed.
The Cuntz semigroups of C∗-algebras with stable rank one satisfy a certain axiom
(O6+) (explained below). We use (O6+) to realize functional infima. (In [APRT18]
we improve this result by showing that Cuntz semigroups of separable C∗-algebras
with stable rank one admit general infima, that is, they are inf-semilattices.)
Third, we show that S 'realizes ranks approximately': Under certain conditions
on S, for all g ∈ Aff(K)++ and ε > 0 there exists f ∈ R with g ≤ f ≤ g + ε; see
Lemma 8.1. The basic idea to obtain this, is to approximate g by infima of chisels.
Fourth, we show that R is closed under passing to suprema of increasing se-
quences. Combined with the approximate realization of ranks from step three, we
obtain R = LAff(K)++; see Theorem 8.7.
To summarize, our approach to realizing a given rank is to approximate it by
infima of chisels. This should be contrasted with the approach in [DT10], where
the basic idea is to approximate a given rank by step functions, that is, by sums of
scalar multiples of characteristic functions.
The main novel techniques of this paper are two new properties for Cu-semi-
groups. In Definition 4.1, and inspired by [Edw69], we introduce Edwards' con-
dition, which roughly says that the functional infimum of two element a and b
can be pointwise approximated by elements dominated by a and b. Using AW ∗-
completions, we verify Edwards' condition for Cuntz semigroups of all unital C∗-
algebras; see Theorem 4.7.
In Definition 6.1, we introduce axiom (O6+), a strengthened version of axiom
(O6) of almost Riesz decomposition. We show that axiom (O6+) is satisfied by
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
5
Cuntz semigroups of C∗-algebras with stable rank one (Theorem 6.4), but not for
all C∗-algebras (Example 6.7).
In Section 5, we show that S realizes chisels if it satisfies Edwards' condition.
This does not require the assumption of stable rank one, and in fact chisels are
realized as the ranks of operators in all simple C∗-algebras; see Theorem 5.5. In
Section 7, we show that S realizes functional infima if it satisfies (O6+).
Acknowledgements
I thank Nate Brown, Tim de Laat, Leonel Robert, Stuart White and Wilhelm
Winter for valuable comments. I am grateful to Eusebio Gardella for his feedback
on the first draft of this paper. Further, I want to thank the anonymous referee for
many helpful suggestions.
2.1. The category Cu of abstract Cuntz semigroups.
2. Preliminaries
2.1. Let S be a partially ordered set. Recall that S is said to satisfy Riesz in-
terpolation if for all x, y, a, b ∈ S with x, y ≤ a, b, there exists z ∈ S such that
x, y ≤ z ≤ a, b. If for all a, b ∈ S the infimum a ∧ b = sup{x : x ≤ a, b} exists,
then S is called an inf-semilattice. A subset D ⊆ S is said to be upward directed
(downward directed ) if for every a, b ∈ D there exists c ∈ D with a, b ≤ c (c ≤ a, b).
Recall that S is said to be directed complete, or a dcpo for short, if every directed
subset D of S has a supremum sup D; see [GHK+03, Definition O-2.1, p.9].
If S is an inf-semilattice then S satisfies Riesz interpolation.
Indeed, given
x, y ≤ a, b, we have x, y ≤ a ∧ b ≤ a, b. The converse holds if S is directed complete:
Given a, b ∈ S, it follows from Riesz interpolation that the set {x : x ≤ a, b} is
directed, whence directed completeness implies that a ∧ b exists.
Given a dcpo S and a, b ∈ S, recall that a is said to be way-below b, or that a is
compactly contained in b, denoted a ≪ b, if for every increasing net (bj)j in S with
b ≤ supj bj there exists j such that a ≤ bj; see [GHK+03, Definition I-1.1, p.49]. A
domain is a dcpo S such that for every a ∈ S the set {x ∈ S : x ≪ a} is upward
directed and has supremum a; see [GHK+03, Definition I.1.6, p.54].
In the context of a Cu-semigroups, the symbol '≪' is used to denote the se-
quential way-below relation; see Paragraph 2.3. In general, the way-below relation
in a dcpo is stronger than its sequential version, and it may be strictly stronger.
Nevertheless, it will always be clear from context to which relation the symbol '≪'
refers. Moreover, we will now see that under suitable 'separability' assumptions,
the two notions agree.
Let S be a domain. A subset B ⊆ S is called a basis if for every a′, a ∈ S with
a′ ≪ a there exists b ∈ B with a′ ≤ b ≤ a. Equivalently, every element in S is the
supremum of a directed net in B; see [GHK+03, Definition III-4.1, Proposition III-
4.2, p.240f]. In particular, S is said to be countably based if it contains a countable
basis. In this case, the way-below relation agrees with its sequential version and
every element in S is the supremum of a ≪-increasing sequence.
2.2. A partially ordered monoid is a commutative monoid M together with a partial
order ≤ such that a ≤ b implies a + c ≤ b + c, for all a, b, c ∈ M . The partial order
on M is said to be positive if we have 0 ≤ a for every a ∈ M . The partial order
is said to be algebraic if for all a, b ∈ M we have a ≤ b if and only if there exists
c ∈ M with a + c = b. Accordingly, we speak of positively ordered monoids and
algebraically ordered monoids.
A cone is a commutative monoid C together with a scalar multiplication by
(0, ∞), that is, with a map (0, ∞) × C → C, denoted (t, λ) 7→ tλ, that is additive
6
HANNES THIEL
in each variable, and such that (st)λ = s(tλ) and 1λ = λ, for all s, t ∈ (0, ∞) and
λ ∈ C. We do not define scalar multiplication with 0 in cones. An ordered cone is
a cone together with a partial order ≤ such that λ1 ≤ λ2 implies λ1 + µ ≤ λ2 + µ
and tλ1 ≤ tλ2, for all λ1, λ2, µ ∈ C and t ∈ (0, ∞). If the partial order is positive
(algebraic), then we speak of a positively (algebraically) ordered cone.
Let M be a partially ordered monoid. Then M is said to satisfy Riesz decom-
position if for all a, b, c ∈ M with a ≤ b + c there exist a1, a2 ∈ M such that
a = a1 + a2, a1 ≤ b and a2 ≤ c. Further, M is said to satisfy Riesz refinement if
for all a1, a2, b1, b2 ∈ M with a1 + a2 = b1 + b2 there exist xi,j ∈ M , for i, j = 1, 2,
such that ai = xi,1 + xi,2 for i = 1, 2, and bj = x1,j + x2,j for j = 1, 2. Among
cancellative and algebraically ordered monoids, the properties Riesz decomposition,
Riesz refinement, and Riesz interpolation are equivalent.
Further, M is said to be inf-semilattice-ordered if M is an inf-semilattice where
addition is distributive over ∧, that is,
a + (b ∧ c) = (a + b) ∧ (a + c),
for all a, b, c ∈ M .
2.3. In [CEI08], Coward, Elliott and Ivanescu introduced the category Cu of ab-
stract Cuntz semigroups. Recall that for two elements a and b in a partially ordered
set S, one says that a is way-below b, or that a is compactly contained in b, denoted
a ≪ b, if for every increasing sequence (bn)n in S for which supn bn exists and
satisfies b ≤ supn bn, there exists N such that a ≤ bN . This is a sequential version
of the usual way-below relation used in lattice theory; see Paragraph 2.1.
An abstract Cuntz semigroup, also called a Cu-semigroup, is a positively ordered
monoid S (see Paragraph 2.2) satisfying the following axioms:
(O1) Every increasing sequence in S has a supremum.
(O2) Every element in S is the supremum of a ≪-increasing sequence.
(O3) Given a′, a, b′, b ∈ S with a′ ≪ a and b′ ≪ b, we have a′ + b′ ≪ a + b.
(O4) Given increasing sequences (an)n and (bn)n in S, we have supn(an + bn) =
supn an + supn bn.
Given Cu-semigroups S and T , a map ϕ : S → T is a Cu-morphism if it preserves
addition, order, the zero element, the way below relation, and suprema of increasing
sequences. If ϕ is not required to preserve the way-below relation then it is called
a generalized Cu-morphism.
We often use the following additional axioms:
(O5) Given a′, a, b′, b, c ∈ S satisfying a + b ≤ c, a′ ≪ a and b′ ≪ b, there exists
x ∈ S (the 'almost complement') such that a′ + x ≤ c ≤ a + x and b′ ≪ x.
(O6) Given a′, a, b, c ∈ S satisfying a′ ≪ a ≤ b + c, there exist e, f ∈ S such that
a′ ≤ e + f , e ≤ a, b and f ≤ a, c.
Axiom (O5) means that S has 'almost algebraic order', and axiom (O6) means that
S has 'almost Riesz decomposition'.
Recall that a Cu-semigroup S is said to have weak cancellation, or to be weakly
cancellative, if for all a, b, x ∈ S, if a + x ≪ b + x then a ≪ b. This is equivalent to
requiring that for all a, b, x ∈ S, if a + x ≪ b + x then a ≤ b. It is also equivalent
to requiring that for all a, b, x′, x ∈ S satisfying a + x ≤ b + x′ and x′ ≪ x we have
a ≤ b. We refer to [APT18, Section 4] for details.
A Cu-semigroup S is said to be countably based if there exists a countable subset
B ⊆ S such that every element in S is the supremum of an increasing sequence
with elements in B. For a discussion and proof of the following basic result we refer
to [APT18, Remarks 3.1.3, p.21f].
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
7
Proposition 2.4. Let S be a countably based Cu-semigroup. Then S a domain
(see Paragraph 2.1) where the way-below relation agrees with its sequential version.
In particular, every upward directed subset D ⊆ S has a supremum.
2.5. Let S be a Cu-semigroup. A sub-Cu-semigroup of S is a submonoid T ⊆ S
such that: T is closed under passing to suprema of increasing sequences; T is a
Cu-semigroup for the inherited partial order; and the inclusion map T → S is a
Cu-morphism. Given a sub-Cu-semigroup T ⊆ S, the way-below relation in T
agrees with the restriction of the way-below relation in S to T .
An ideal in S is a submonoid J ⊆ S that is closed under passing to suprema of
increasing sequences and that is downward hereditary, that is, if a ∈ S and b ∈ J
satisfy a ≤ b, then a ∈ J. Every ideal is a sub-Cu-semigroup, and a sub-Cu-sem-
igroup is an ideal if and only if it is downward hereditary. We call S simple if it
only contains the ideals {0} and S. Note that S is simple if and only if for every
nonzero a ∈ S the element ∞a := supn na is the largest element of S.
An element a ∈ S is said to be finite if a 6= a + b for every b 6= 0. Further, S is
stably finite if a ∈ S is finite whenever there exists a ∈ S with a ≪ a; see [APT18,
Paragraph 5.2.2]. Every weakly cancellative, simple Cu-semigroup is stably finite.
A functional on S is a generalized Cu-morphism S → [0, ∞]. We use F (S) to
denote the set of functionals on S. Equipped with pointwise addition and order,
F (S) has the structure of a positively ordered monoid. A scalar multiple of a
functional is again a functional, which gives F (S) the structure of a positively
ordered cone.
If S satisfies (O5) and (O6), then F (S) is an algebraically inf-
semilattice-ordered cone; see Proposition 2.2.3 and Theorem 4.1.2 in [Rob13a].
It was shown in [ERS11, Theorem 4.8] that F (S) has a natural compact, Haus-
dorff topology such that a net (λj )j in F (S) converges to λ ∈ F (S) if and only if
for all a′, a ∈ S with a′ ≪ a we have
lim sup
j
λj (a′) ≤ λ(a) ≤ lim inf
j
λj(a).
Given a positively ordered monoid M , we let M ∗ denote the order-preserving
monoid morphisms M → [0, ∞]. Equipped with pointwise order and addition and
the obvious scalar multiplication, M ∗ has the structure of a positively ordered cone.
If C is a positively ordered cone, then every f ∈ C∗ is automatically homogeneous
(that is f (tλ) = tf (λ) for t ∈ (0, ∞) and λ ∈ C). We also say that f is a linear
functional on C. We set
lsc :=(cid:8)f ∈ F (S)∗ : f lower semicontinuous(cid:9).
Given a ∈ S, we let ba : F (S) → [0, ∞] be given by ba(λ) := λ(a), for λ ∈ F (S).
Thenba belongs to F (S)∗
lsc. We callba the rank of a. If a ≪ a, thenba is continuous.
Fu7→1(S) :=(cid:8)λ ∈ F (S) : λ(u) = 1(cid:9).
only if bu is continuous (for example, if u is compact, that is, u ≪ u). We refer to
[APT18, Theorem 5.2.6, p.50] for a characterization of when Fu7→1(S) is nonempty.
If S is simple, stably finite, satisfies (O5) and (O6) and u is nonzero and compact,
then Fu7→1(S) is a Choquet simplex; see Lemma 3.8.
Then Fu7→1(S) is a convex subset of F (S). Moreover, Fu7→1(S) is closed if and
F (S)∗
Given u ∈ S, set
2.6. Let S be a Cu-semigroup, and let a, b ∈ S. The element a is said to be
compact if a ≪ a. We use Sc to denote the set of compact elements in S. The
element a is said to be stably below b, denoted a <s b, if there exists n ∈ N such
that (n + 1)a ≤ nb. Further, a is said to be soft if a′ <s a for every a′ ∈ S
satisfying a′ ≪ a; see [APT18, Definition 5.3.1]. We use Ssoft to denote the set of
soft elements in S. Set S×
soft := Ssoft \ {0}.
8
HANNES THIEL
By [APT18, Theorem 5.3.11], Ssoft is a submonoid of S that is closed under
passing to suprema of increasing sequences. Moreover, Ssoft is absorbing in the
sense that a + b is soft as soon as a is soft and b ≤ ∞a. In particular, if S is simple,
then a + b is soft whenever a is soft and nonzero.
A simple Cu-semigroup is said to be elementary if it contains a minimal nonzero
element. The typical elementary Cu-semigroup is N := {0, 1, 2, . . . , ∞}.
Next, we summarize basic results about the structure of simple Cu-semigroups.
Part (1) is [APT18, Proposition 5.3.16], parts (2) and (3) follow from (the proof
of) [APT18, Proposition 5.3.18], and part (4) is [Rob13a, Proposition 5.2.1].
Proposition 2.7. Let S be a simple, non-elementary, stably finite Cu-semigroup
satisfying (O5) and (O6). Then:
(1) Every nonzero element in S is either soft or compact. Thus, S = Sc ⊔ S×
soft.
(2) Ssoft is a sub-Cu-semigroup of S that is itself a simple, stably finite Cu-
semigroup satisfying (O5) and (O6).
(3) For every nonzero a ∈ S there exists x ∈ Ssoft with 0 6= x ≤ a.
(4) For every nonzero a ∈ S and n ∈ N there exists x ∈ S with 0 6= nx ≤ a.
Let S be a partially ordered semigroup. Recall that S is said to be unperforated
if for all a, b ∈ S we have a ≤ b whenever na ≤ nb for some n ≥ 1. Further, S is
said to be almost unperforated if for all a, b ∈ S we have a ≤ b whenever a <s b
(that is, whenever (n + 1)a ≤ nb for some n ≥ 1).
Proposition 2.8. Let S be a simple, stably finite, non-elementary Cu-semigroup
satisfying (O5) and (O6). Then the following are equivalent:
(1) S is almost unperforated.
(2) Ssoft is almost unperforated.
(3) The map κ : Ssoft → F (S)∗
lsc is an order-embedding, that is, for all a, b ∈
Ssoft we have a ≤ b wheneverba ≤bb.
Proof. By Proposition 2.7, Ssoft is a Cu-semigroup. It is clear that (1) implies that
Ssoft is almost unperforated, and that (3) implies that Ssoft is unperforated. If Ssoft
is almost unperforated, then (3) follows from [APT18, Theorem 5.3.12]. To show
that (3) implies (1), assume that the order on Ssoft is determined by the functionals.
To show that S is almost unperforated, let a, d ∈ S and n ∈ N satisfy (n+1)a ≤ nd.
We need to verify a ≤ d. We may assume that there exists d with d ≪ d. (For
every a′ ≪ a there exists d′ ≪ d with (n + 1)a′ ≤ nd′. If for all such a′, d′ we can
deduce a′ ≤ d′, then we obtain a ≤ d.) We will repeatedly use the results from
Proposition 2.7. We approximate a and d by suitable soft elements.
We have (2n + 2)a ≤ (2n)d. Choose x ∈ S×
soft with (2n + 1)x ≤ a. Set b := a + x.
Then a ≤ b, and b is soft since S×
soft is absorbing; see Paragraph 2.6. Then
(2n + 1)b = (2n + 1)a + (2n + 1)x ≤ (2n + 2)a.
Choose y′, y ∈ S×
soft with y′ ≪ y and (2n + 1)y ≤ d. Apply (O5) for y′ ≪ y ≤ d to
obtain z ∈ S with y′ + z ≤ d ≤ y + z. Set c := y′ + z. Then c ≤ d and c is soft. We
have
(2n + 1)d ≤ (2n + 1)y + (2n + 1)z ≤ d + (2n + 1)z ≤ d + (2n + 1)c.
Let λ∞ ∈ F (S) be the largest functional, which satisfies λ∞(s) = ∞ for every
nonzero s ∈ S. Since there exists d with d ≪ d and since S is simple, we have
λ(d) < ∞ for every λ ∈ F (S) with λ 6= λ∞. We deduce (2n)bd ≤ (2n + 1)bc. Thus,
(2n + 1)bb ≤ (2n + 2)ba ≤ 2nbd ≤ (2n + 1)bc.
By assumption, we obtain b ≤ c, and hence a ≤ b ≤ c ≤ d.
(cid:3)
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
9
Proposition 2.9 ([APT18, Lemma 5.4.3, Proposition 5.4.4]). Let S be a countably
based, simple, stably finite, non-elementary Cu-semigroup satisfying (O5) and (O6).
Then for every a ∈ Sc there exists x ∈ Ssoft with x < a and bx =ba.
If S is weakly cancellative or almost unperforated, then for each a ∈ S the set
{x ∈ Ssoft : x ≤ a} contains a largest element. Moreover, the map : S → Ssoft
given by (a) := max{x ∈ Ssoft : x ≤ a} is a generalized Cu-morphism.
there exists x ∈ Pa with ba = bx.
Proof. Given a ∈ S set Pa := {x ∈ Ssoft : a ≤ x}. By [APT18, Lemma 5.4.3],
If S is almost unperforated, then such x is
If S weakly cancellative, then
the largest element in Pa, by Proposition 2.8.
it follows from [APT18, Proposition 5.4.4] that Pa contains a largest element.
Hence, is well-defined and order-preserving. It is straightforward to verify that
preserves suprema of increasing sequences.
(This also follows as in the proof
of Proposition 8.9, using that is the upper adjoint of a Galois connection; see
Remarks 2.10.) To show that is additive, let a, b ∈ S. By Proposition 2.7, each
of a and b is either soft or compact. If a and b are soft, there is nothing to show.
If a ∈ Sc and b ∈ S×
soft, then it follows from [APT18, Proposition 5.4.4] that
a + b = (a) + b, and consequently
(a + b) = ((a) + b) = (a) + b = (a) + (b).
Lastly, it is enough to consider the case that a and b are compact. We have
(a) + (b) ≤ (a + b). Further,
a + (b) ≤ (a + b) ≤ a + b.
Applying (O5) for a ≪ a ≤ (a + b), we obtain x ∈ S such that a + x = (a + b).
Then x ∈ Ssoft and a + x ≤ a + b. Using that S is weakly cancellative or almost
unperforated (and simple), we obtain x ≤ b. Then x ≤ (b) and thus
(a + b) = a + x ≤ a + (b) = (a) + (b),
as desired.
(cid:3)
Remarks 2.10. (1) By Proposition 2.9, a function in F (S)∗
some a ∈ S if and only if it is realized as the rank of some soft element.
lsc is realized asba for
(2) Given a ∈ Sc nonzero, the element (a) is a 'predecessor' of a. We therefore
call the map : S → Ssoft from Proposition 2.9 the predecessor map. Let ι : Ssoft →
S denote the inclusion map. Given a ∈ Ssoft and b ∈ S, we have a ≤ (b) if and
only if ι(a) ≤ b. Thus, and ι form a (order theoretic) Galois connection between
S and Ssoft; see [GHK+03, Definition O-3.1, p.22]. In Paragraph 8.8 we show that
there is also a Galois connection between LAff(K)++ and S×
soft.
Let S be a partially ordered semigroup. Recall that S is said to be divisible if
for all a ∈ S and k ≥ 1 there exists x ∈ S such that kx = a. If S is a Cu-semigroup,
then is said to be almost divisible if for all a′, a ∈ S with a′ ≪ a and all k ∈ N there
exists x ∈ S such that kx ≤ a and a′ ≤ (k + 1)x; see [APT18, Definition 7.3.4].
Given a compact, convex set K, we use LAff(K)++ to denote the strictly posi-
tive, lower semicontinuous, affine functions K → [0, ∞]; see Paragraph 3.1. In the
following result, K is a Choquet simplex by Lemma 3.8.
Proposition 2.11. Let S be a countably based, simple, stably finite, non-elementary
Cu-semigroup satisfying (O5) and (O6), let u ∈ S be a compact, full element, and
set K := Fu7→1(S). Assume that S is almost unperforated. Then the following are
equivalent:
(1) S is almost divisible.
(2) Ssoft is (almost) divisible.
(3) The map κ : Ssoft → F (S)∗
lsc is an order-isomorphism.
10
HANNES THIEL
(4) For every f ∈ LAff(K)++, there exists a ∈ Ssoft withbaK = f .
(5) For every g ∈ Aff(K)++ and ε > 0 there exists a ∈ S with g ≤baK ≤ g + ε.
Proof. By Proposition 2.7, Ssoft is a simple Cu-semigroup satisfying (O5) and (O6).
We will use the predecessor map : S → Ssoft from Proposition 2.9. Since S
is almost unperforated, so is Ssoft, by Proposition 2.8. We will frequently use
that the map κ : Ssoft → F (S)∗
lsc is an order-embedding. Since S is simple, we
obtain that K is a compact base for the cone F (S) \ {λ∞}.
It follows that
the restriction map F (S)∗
lsc → LAff(K), f 7→ fK, defines an order-isomorphism
∼= LAff(K)++ ∪ {0}. This shows that (3) and (4) are equivalent. It is clear
F (S)∗
lsc
that (4) implies (5).
To show that (1) implies (2), assume that S is almost divisible. We first show
that Ssoft is almost divisible. Let a′, a ∈ Ssoft satisfy a′ ≪ a and let k ∈ N. Since S
is almost divisible, we obtain x ∈ S with kx ≤ a and a′ ≤ (k + 1)x. Set y := (x).
Then ky ≤ a and a′ ≤ (k + 1)y, which shows that Ssoft is almost divisible. It follows
from [APT18, Theorem 7.5.4] that Ssoft is divisible.
To show that (2) implies (3), assume that Ssoft is divisible. Applying [APT18,
Theorem 7.5.4], it follows that Ssoft has 'real multiplication' in the sense of Robert,
[Rob13a, Definition 3.1.2]. Let L(F (S)) ⊆ F (S)∗
lsc be defined as in [Rob13a, Sec-
tion 3.2]. Using that S is simple and hence the cone F (S) \ {λ∞} has a com-
pact base (namely K), we deduce L(F (S)) = F (S)∗
lsc. The inclusion Ssoft → S
induces a natural identification F (S) ∼= F (Ssoft). We obtain a natural order-
isomorphism L(F (Ssoft)) ∼= F (S)∗
lsc. It follows from [Rob13a, Theorem 3.2.1] that
Ssoft → F (Ssoft)∗
implies (3).
lsc, a 7→ba, defines an order-isomorphism Ssoft → L(F (Ssoft)), which
To show that (5) implies (4), one proceeds as in the proof of Theorem 8.7:
Given f ∈ LAff(K)++, choose an increasing sequence (gn)n in Aff(K)++ and a
decreasing sequence (εn)n of positive numbers such that g0 − ε0 ∈ Aff(K)++ and
f = supn(gn − εn). By assumption, for reach n we obtain an ∈ S with gn − εn ≤
for each n. Since bn and bn+1 are soft, we deduce bn ≤ bn+1. Thus, the sequence
To show that (3) implies (2), let a ∈ Ssoft and let k ∈ N. We have 1
bbnK =canK ≤ gn − εn+1 ≤ gn+1 − εn+1 ≤ [an+1K = dbn+1K,
canK ≤ gn − εn+1. Set bn := (an). Then
(bn)n is increasing. Set b := supn bn. ThenbbK = f .
By assumption, there exists x ∈ Ssoft withbx = 1
To show that (2) implies (1), assume that Ssoft is divisible. Let a′, a ∈ S satisfy
a′ ≪ a and let k ∈ N. By assumption, we obtain x ∈ Ssoft with kx = (a). Then
kx = (a) ≤ a. By [APT18, Proposition 5.4.4], we have a + x = (a) + x. Using
this at the third step, we obtain
kba ∈ F (S)∗
kba. Then kbx =ba and thus kx = a.
lsc.
a′ ≤ a ≤ a + x = (a) + x = (k + 1)x,
as desired.
(cid:3)
2.2. The Cuntz semigroup of a C*-algebra.
2.12. Let A be a C∗-algebra. We use A+ to denote the positive elements in A.
Given a ∈ A+ and ε > 0, we write aε for the 'ε-cut-down' (a − ε)+, obtained by
applying continuous functional calculus for the function f (t) = max{0, t − ε} to a.
Let a, b ∈ A+. We write a - b if a is Cuntz subequivalent to b, that is, if there
exists a sequence (rn)n in A+ such that limn ka − rnbr∗
nk = 0. We write a ∼ b if a
is Cuntz equivalent to b, that is, if a - b and b - a. We write a ⊆ b if a belongs to
bAb, the hereditary sub-C∗-algebra of A generated by b.
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
11
Rørdams's lemma states that a - b if and only if for every ε > 0 there exist
δ > 0 and x ∈ A such that
aε = xx∗, and x∗x ⊆ bδ.
It follows that a - b if and only if aε - b for every ε > 0. If A has stable rank one,
unitization of A, such that uaεu∗ ⊆ b; see [Rør92, Proposition 2.4].
then a - b if and only if for every ε > 0 there exists a unitary u in eA, the minimal
The Cuntz semigroup of A, denoted Cu(A), is the set of Cuntz equivalence classes
of positive elements in the stabilization A⊗ K. The class of a ∈ (A⊗ K)+ in Cu(A) is
denoted by [a]. The relation - induces a partial order on Cu(A). Using a suitable
∼=−→ A ⊗ K, one introduces an addition on Cu(A) by
isomorphism ϕ : A ⊗ K ⊗ M2
setting [a] + [b] := [ϕ( a 0
0 b )]. This gives Cu(A) the structure of a positively ordered
monoid. It was shown in [CEI08] that Cu(A) is always a Cu-semigroup.
The next result summarizes the basic properties of Cuntz semigroups of C∗-alge-
bras. We refer to [APT18, Section 4] for details. In Section 6, we show that Cu(A)
satisfies a strengthened version of (O6) if A has stable rank one.
Proposition 2.13. Let A be a C∗-algebra. Then Cu(A) is a Cu-semigroup satis-
fying (O5) and (O6). If A is separable, then Cu(A) is countably-based. If A has
stable rank one, then Cu(A) is weakly cancellative. If A is simple, then Cu(A) is
simple.
2.14. The terminology used in connection with quasitraces on C∗-algebras is non-
uniform across the literature. We follow [BK04, Subsection 2.9].
Let A be a C∗-algebra. A quasitrace on A is a map τ : A+ → [0, ∞] with τ (0) = 0,
satisfying τ (xx∗) = τ (x∗x) for every x ∈ A, and such that τ (a + b) = τ (a) + τ (b) for
all a, b ∈ A+ that commute. A quasitrace τ is bounded if τ (a) < ∞ for all a ∈ A+,
in which case τ extends canonically to a (non-linear) functional A → C satisfying
τ (a) = τ (a+)−τ (a−) for all self-adjoint a ∈ A, and satisfying τ (a+ib) = τ (a)+iτ (b)
for all self-adjoint a, b ∈ A.
A 2-quasitrace on A is a quasitrace that extends to a quasitrace τ2 on A ⊗ M2
with τ2(a ⊗ e1,1) = τ (a) for every a ∈ A+, where e1,1 is the upper-left rank-one
projection. A quasitrace τ is said to be lower semicontinuous if τ (a) = supε>0 τ (aε)
for every a ∈ A+. A bounded quasitrace is automatically lower semicontinuous.
We let QT(A) denote the set of lower semicontinuous, 2-quasitraces on A. Every
τ ∈ QT(A) is order-preserving on A+; see [BK04, Remarks 2.27]. The set QT(A)
has the structure of an algebraically ordered cone and it can be equipped with a
natural compact, Hausdorff topology; see [ERS11, Section 4.1] We warn the reader
that in [BH82] quasitraces are assumed to be bounded, and in [Haa14] they are also
assumed to be 2-quasitraces.
Let τ ∈ QT(A). There is a unique extension of τ to a lower semicontinuous
quasitrace τ∞ : (A ⊗ K)+ → [0, ∞] satisfying τ∞(a ⊗ e) = τ (a) for a ∈ A+ and for
every rank-one projection e ∈ K. Abusing notation, we usually denote τ∞ by τ .
The induced dimension function dτ : (A ⊗ K)+ → [0, ∞] is given by
dτ (a) := lim
n
τ (a1/n),
for a ∈ (A ⊗ K)+.
If a, b ∈ (A ⊗ K)+ satisfy a - b, then dτ (a) ≤ dτ (b). Therefore, dτ induces a
well-defined, order-preserving map Cu(A) → [0, ∞], which we also denote by dτ .
One can show that this map is a functional on Cu(A). Moreover, every functional
on Cu(A) arises in this way; see [ERS11, Proposition 4.2].
It follows that the
assignment τ 7→ dτ is an additive bijection between QT(A) and F (Cu(A)), and it
is even an isomorphism of ordered topological cones; see [ERS11, Theorem 4.4].
12
HANNES THIEL
If A is unital, we let QT17→1(A) denote the set of 2-quasitraces τ satisfying
τ (1) = 1. If A is stably finite, then QT17→1(A) is a nonempty, compact, convex
subset of QT(A) that has the structure of a Choquet simplex; see [BH82, Theo-
reom II.4.4]. Under the identification QT(A) ∼= F (Cu(A)), the Choquet simplex
QT17→1(A) corresponds to F[1]7→1(Cu(A)).
3. Lower semicontinuous affine functions on Choquet simplices
3.1. Let K be a compact, convex set, by which we always mean a compact, convex
subset of a locally convex, Hausdorff, real topological vector space. The set of
extreme points in K is denoted by ∂eK. We use Aff(K) to denote the space
of continuous, affine functions K → R. We let LAff(K) denote the set of lower
semicontinuous, affine functions K → (−∞, ∞]. We equip Aff(K) and LAff(K)
with pointwise order and addition. We let Aff(K)++ and LAff(K)++ denote the
subsets of strictly positive elements in Aff(K) and LAff(K), respectively.
Recall that a partially ordered set S is called directed complete, or a dcpo for
short, if every (upward) direct subset of S has a supremum; equivalently, every
increasing net in S has a supremum; see [GHK+03, Definition O-2.1, p.9]. Given
an increasing net in LAff(K), the pointwise supremum is a function in LAff(K).
This shows that LAff(K) is a dcpo.
Proposition 3.2 ([Alf71, Corollary I.1.4, p.3]). Let K be a compact, convex set.
Then every element of LAff(K) is the supremum of an increasing net in Aff(K).
Proposition 3.3 (Bauer's minimum principle, [Bau58]). Let K be a compact,
convex set, and let f : K → (−∞, ∞] be a lower semicontinuous, concave function.
Then inf f (K) = min f (∂eK). This holds in particular for all f ∈ LAff(K).
Proof. The result is formulated in German as the Corollary to Satz 2 in [Bau58].
A simplified proof of the dual result (Bauer's maximum pricniple) can be found
in [Cho69, Theorem 25.9, p.102]: Every upper semicontinuous, convex function
K → R attains its maximum on ∂eK. This implies Bauer's minimum principle for
lower semicontinuous, concave functions K → R. The general case (allowing the
value ∞) is clear if inf f (X) = ∞. Otherwise, the result follows by considering the
function given as the pointwise infimum of f and a fixed value > inf f (K).
(cid:3)
The following result is not as well known. Since the only reference I could find
is in German, I include a short argument for the convenience of the reader.
Proposition 3.4 (Bauer, [Bau63]). Let K be a compact, convex set, and let f, g ∈
LAff(K). Then f ≤ g if and only if f∂eK ≤ g∂eK.
Proof. Apply Proposition 3.2 to obtain an increasing net (fj)j∈J in Aff(K) with
supremum f . For each j ∈ J, set gj := g − fj. Since fj is continuous and affine,
gj belongs to LAff(K). Moreover, we have 0 ≤ gj(λ) for every λ ∈ ∂eK. Applying
Bauer's minimum principle (Proposition 3.3), we obtain 0 ≤ gj, and thus fj ≤ g.
Passing to the supremum over j, we deduce f ≤ g.
(cid:3)
Remark 3.5. Let K be a metrizable, compact, convex set, and let f, g : K → R be
affine, lower semicontinuous functions satisfying f∂eK ≤ g∂eK. Then the conclusion
of Proposition 3.4 may also be obtained as follows: By Choquet's theorem, [Alf71,
Corollary 1.4.9, p.36], for every λ ∈ K there exists a boundary measure µ (a Borel
probability measure µ with µ(∂eK) = 1) with barycenter λ, that is, such that
h(λ) =ZK
h(λ)dµ(λ) =Z∂eK
h(λ)dµ(λ),
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
13
for all h ∈ Aff(K). The above formula holds also for affine functions K → R of first
Baire class, in particular for f and g; see [Phe01, p.88]. Given λ ∈ K, we choose a
boundary measure µ with barycenter λ, and obtain
f (λ) =Z∂eK
f (λ)dµ(λ) ≤Z∂eK
g(λ)dµ(λ) = g(λ).
Recall the definition of a domain and the way-below relation from Paragraph 2.1.
Lemma 3.6. Let K be a compact, convex set. Then LAff(K) is a domain. Given
f, g ∈ LAff(K), we have f ≪ g if and only if there exist h ∈ Aff(K) and ε > 0 with
f + ε ≤ h ≤ g. Further, K is metrizable if and only if LAff(K) is countably based.
Proof. It is straightforward to verify that LAff(K) is a dcpo. To verify the char-
acterization of the way-below relation, let f, g ∈ LAff(K). We first assume that
f ≪ g. Apply Proposition 3.2 to obtain an upward directed subset D ⊆ Aff(K)
such that g = sup D. Set
D′ :=(cid:8)h − ε : h ∈ D, ε > 0(cid:9).
It is straightforward to show that D′ is upward directed and that g = sup D′. By
definition of the way-below relation, we can choose h′ ∈ D′ with f ≤ h′. Choose
h ∈ D and ε > 0 such that h′ = h − ε. Then f + ε ≤ h ≤ g, as desired.
Conversely, let h ∈ Aff(K) and ε > 0 with f + ε ≤ h ≤ g. To verify f ≪ g, let D
be an upward directed subset of LAff(K) with g ≤ sup D. Given x ∈ K, we have
h(x) − ε < h(x) ≤ g(x) = sup(cid:8)g′(x) : g′ ∈ D(cid:9),
which allows us to choose gx ∈ D with h(x) − ε < gx(x). Using that h is continuous
and that gx is lower-semicontinuous, we can choose a neighborhood Vx of x such
that h − ε < gx on Vx. Use that K is compact to choose x1, . . . , xn such that K =
k=1 Vxk . Since D is upward directed, we can pick g′ ∈ D with gx1, . . . , gxn ≤ g′.
Sn
Then f ≤ g′, as desired.
Using that every element in LAff(K) is the supremum of an upward directed
subset of Aff(K), see Proposition 3.2, and using the above characterization of the
way-below relation, it follows that LAff(K) is a domain.
Using the description of the way-below relation, it follows that LAff(K) is count-
ably based if and only if Aff(K) is separable (for the topology induced by the
supremum-norm). Assume that K is metrizable. Then the Banach algebra C(K, R)
of continuous functions K → R is separable. Since Aff(K) is a closed subspace of
C(K, R), it is separable as well; see [Goo86, Proposition 14.10, p.221].
Conversely, assume that Aff(K) is separable. Since the elements of Aff(K)
separate the points of K, the family of sets of the form f −1((0, ∞]), with f ranging
over a countable dense subset of Aff(K), forms a countable basis for the topology
of K. Hence, K is second-countable and therefore metrizable.
(cid:3)
Lemma 3.7. Let K be a compact, convex set. Then the following are equivalent:
(1) K is a Choquet simplex.
(2) LAff(K) is an inf-semilattice.
(3) LAff(K) satisfies Riesz interpolation.
Moreover, if the above hold, then LAff(K) is inf-semilattice-ordered and we have
(3.7.1)
(f ∧ g)(λ) = f (λ) ∧ g(λ),
for all f, g ∈ LAff(K) and λ ∈ ∂eK.
Proof. By [Alf71, Theorem II.3.8, p.89], K is a Choquet simplex if and only if the
upper semicontinuous, affine functions K → [−∞, ∞) form a sup-semilattice. The
latter is equivalent to LAff(K) being an inf-semilattice, which shows the equivalence
14
HANNES THIEL
of (1) and (2). A dcpo is an inf-semilattice if and only if it satisfies Riesz refinement;
see Paragraph 2.1. Using that LAff(K) is a dcpo, we obtain that (2) and (3) are
equivalent.
Assume that statements (1)-(3) are satisfied. To show (3.7.1), let f, g ∈ LAff(K).
Consider the function h : K → (−∞, ∞] given by h(λ) := f (λ) ∧ g(λ), for λ ∈ K.
Then h is lower semicontinuous and concave. Let h be the lower envelope of h; see
[Alf71, p.4]. It follows from [Alf71, Theorem II.3.8] that h is lower semicontinuous
and affine. This shows that h is the infimum of f and g in LAff(K).
By [Alf71, Proposition I.4.1], we have k(λ) = k(λ) for every lower semicontinuous
function k : K → (−∞, ∞] and every λ ∈ ∂eK. Applied for h, we obtain
(f ∧ g)(λ) = h(λ) = h(λ) = f (λ) ∧ g(λ),
for all λ ∈ ∂eK.
To show that LAff(K) is inf-semilattice-ordered, let f, g, h ∈ LAff(K). Using
(3.7.1) at the first and last step, we obtain
(f + (g ∧ h))(λ) = f (λ) + g(λ) ∧ h(λ)
= (f (λ) + g(λ)) ∧ (f (λ) + h(λ)) = ((f + g) ∧ (f + h))(λ),
for all λ ∈ ∂eK. Using Proposition 3.4, we deduce f +(g ∧h) = (f +g)∧(f +h). (cid:3)
Lemma 3.8. Let S be a simple, stably finite Cu-semigroup satisfying (O5) and
(O6), and let u ∈ S be nonzero such that u ≪ a for some a ∈ S, and such that bu
is continuous (for example, u is nonzero and compact). Then K := Fu→1(S) is a
nonempty Choquet simplex. If S is countably based, then K is metrizable.
Proof. By Proposition 2.2.3 and Theorem 4.1.2 in [Rob13a], F (S) is an algebraically
ordered, complete lattice. Let λ∞ ∈ F (S) be the largest functional, which satisfies
λ∞(a) = ∞ for all nonzero a ∈ S. Set C := F (S)\{λ∞}. Then C is an algebraically
ordered, locally compact cone. Moreover, C is cancellative by the remarks above
Proposition 3.2.3 in [Rob13a]. (In the notation used there, we have C = FS(S).)
It follows that C satisfies Riesz refinement.
Using thatbu is continuous, we deduce that K is a closed, convex subset of F (S).
Let λ0 be the zero functional. Given λ ∈ F (S) with λ 6= λ0, λ∞, we have λ(u) 6= 0
(using that u is full) and λ(u) 6= ∞ (using that there exists a ∈ S with u ≪ a).
1
Hence, the functional
λ(u) λ. It follows that K
is a compact base of C. Since C satisfies Riesz refinement, it follows from [Alf71,
Proposition II.3.3, p.85] that K is a Choquet simplex.
1
λ(u) λ belongs to K and λ = λ(u) ·
If S is countably based, it follows as in the proof of Lemma 3.6 that K is metriz-
(cid:3)
able.
Proposition 3.9. Let K be a metrizable, compact, convex set. Consider S :=
LAff(K)++ ∪ {0}. Then S is a weakly cancellative, countably based, simple, stably
finite Cu-semigroup satisfying (O5). Further, the following are equivalent:
(1) K is a Choquet simplex.
(2) S is semilattice-ordered.
(3) S satisfies (O6+). (See Definition 6.1.)
(4) S satisfies (O6).
Proof. It follows from Lemma 3.6 that S is a countably based Cu-semigroup. Every
nonzero element in S is a strictly positive function on K and therefore full in S,
which shows that S is simple. For a, b ∈ S we have a ≪ b in S if and only if a ≪ b
in LAff(K). Using that finite-valued functions in LAff(K) can be canceled, one
shows that S is weakly cancellative and stably finite.
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
15
To verify that S satisfies (O5), let a′, a, b′, b, c ∈ S satisfy
a + b ≤ c, and a′ ≪ a, and b′ ≪ b.
We need to find x ∈ S satisfying a′ +x ≤ c ≤ a+x and b′ ≪ x. If a′ = 0, then x := c
has the desired properties. If a′ 6= 0, apply Lemma 3.6 to obtain f ∈ Aff(K)++
and ε > 0 such that
a′ ≤ f, and f + ε ≤ a.
Define x : K → [0, ∞] by x(λ) := c(λ) − f (λ), for λ ∈ K. Since f ∈ Aff(K), we
have x ∈ LAff(K). It follows that x has the desired properties.
If K is a Choquet simplex, then LAff(K) is semilattice-ordered by Lemma 3.6,
which implies that S is semilattice-ordered as well. Assume that S is semilattice-
ordered. To verify (O6+), let a, b, c, x′, x, y′, y ∈ S satisfy
a ≤ b + c, and x′ ≪ x ≤ a, b, and y′ ≪ y ≤ a, c.
Set e := a ∧ b and f := a ∧ c. We have x′ ≪ e ≤ a, b and y′ ≪ f ≤ a, c. Moreover,
e + f = (a ∧ b) + (a ∧ c) = ((a ∧ b) + a) ∧ ((a ∧ b) + c)
= (2a) ∧ (a + b) ∧ (a + c) ∧ (b + c) ≥ a,
which shows that e and f have the desired properties.
It is clear that (O6+) implies (O6). Lastly, let us assume that S satisfies (O6).
Set u := 1, the constant function of value 1 on K. Then bu is continuous and
Fu7→1(S) is affinely homeomorphic to K. It follows from Lemma 3.8 that K is a
Choquet simplex.
(cid:3)
4. Edwards' condition
Throughout this section, we let S be a simple, stably finite Cu-semigroup sat-
isfying (O5) and (O6), we let u ∈ S be a compact, full element, and we set
K := Fu→1(S), the compact, convex set of normalized functionals. It follows from
Lemma 3.8 that K is a nonempty Choquet simplex.
Definition 4.1. Given λ ∈ ∂eK, we say that S satisfies Edwards' condition for λ
if
(4.1.1)
min{λ(a), λ(b)} = sup(cid:8)λ(c) : c ≤ a, b(cid:9),
for all a, b ∈ S. If this holds for all λ ∈ ∂e(K), then we say that S satisfies Edward's
condition for ∂e(K).
Remarks 4.2. (1) There is natural way to formulate Edwards' condition for ar-
bitrary functionals. This is especially relevant for studying ranks of operators in
non-simple or non-unital C∗-algebras. We will pursue this in forthcoming work.
(2) Since K is a Choquet simplex, we may apply (3.7.1) to identify the left hand
side in (4.1.1) with (baK ∧bbK)(λ).
(3) Let λ ∈ ∂eK. Then S satisfies Edwards' condition for λ if for all a, b ∈ S
and t ∈ [0, ∞) satisfying t < λ(a), λ(b), there exists c ∈ S such that
t < λ(c),
and c ≤ a, b.
Similarly, by Proposition 4.3, if S satisfies Edwards' condition for λ, then for all
a, b ∈ S and t ∈ (0, ∞] satisfying λ(a), λ(b) < t, there exists d ∈ S such that
a, b ≤ d, and λ(d) < t.
These formulations are similar to condition (2) considered by Edwards in [Edw69],
and this is the reason for the terminology in Definition 4.1.
The statement of the following result can be considered as a dual version of
Edwards' condition. It will be used in Proposition 5.4.
16
HANNES THIEL
Proposition 4.3. Let λ ∈ ∂eK be such that S satisfies Edwards' condition for λ.
Then
for all a, b ∈ S.
max{λ(a), λ(b)} = inf(cid:8)λ(d) : a, b ≤ d(cid:9),
Proof. The inequality '≤' is clear. To show the converse, let a, b ∈ S. Without
loss of generality, we may assume that λ(a) ≤ λ(b) < ∞. Let t ∈ (0, ∞) satisfy
λ(b) < t. We need to find d ∈ S such that
a, b ≤ d, and λ(d) < t.
If λ(a) = 0, then a = 0 by simplicity, and thus d := b has the desired properties.
Assume that 0 < λ(a). Choose ε > 0 such that
and choose s ∈ [0, ∞) such that
λ(b) < t − 2ε,
s < λ(a) ≤ s + ε.
Applying Edwards' condition, we obtain c ∈ S such that
s < λ(c),
and c ≤ a, b.
Choose c′ ≪ c such that
s < λ(c′), and λ(c) ≤ λ(c′) + ε.
Applying (O5) for c′ ≪ c ≤ a and c′ ≪ c ≤ b, we obtain x, y ∈ S such that
c′ + x ≤ a ≤ c + x, and c′ + y ≤ b ≤ c + y.
Set d := c + x + y. We have
λ(c′) + λ(x) ≤ λ(a) ≤ s + ε ≤ λ(c′) + ε.
Since c′ ≪ c ≤ ∞u, and λ(u) = 1, we have λ(c′) < ∞, which allows us to cancel it
from the above inequality to obtain λ(x) ≤ ε. We deduce
λ(d) = λ(c) + λ(x) + λ(y) ≤ λ(c′) + 2ε + λ(y) ≤ λ(b) + 2ε < t.
Further, we have
a ≤ c + x ≤ c + x + y = d, and b ≤ c + y ≤ c + y + x = d,
which shows that d has the desired properties.
(cid:3)
Our next goal is to verify that Cuntz semigroups of C∗-algebras satisfy Edwards'
condition; see Theorem 4.7. A main ingredient in the proof are AW ∗-completions
as developed in [Haa14], which in turn is based on [BH82]. We recall some details.
4.4. Let A be a unital C∗-algebra. Recall that QT17→1(A) denotes the normalized
2-quasitraces; see Paragraph 2.14. Let τ ∈ QT17→1(A). Then τ extends canonically
to a (possibly non-linear) functional A → C, which we also simply denote by τ .
Given x ∈ A, set
By [Haa14, Lemma 3.5(2)], we have
kxkτ := τ (x∗x)1/2.
(4.4.1)
for all x, y ∈ A.
kx + yk2/3
τ ≤ kxk2/3
τ + kyk2/3
,
τ
If x ∈ A is normal, then the restriction of τ to the sub-C∗-algebra generated by
x is linear, whence we can apply the Cauchy-Schwarz inequality to obtain
(4.4.2)
τ (x) ≤ kxkτ .
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
17
For general x, let a and b be the real and imaginary parts of x. Applying (4.4.1)
at the third step, we obtain
kak2
2
1
1
≤
and similarly kbk2
(x + x∗)(cid:13)(cid:13)2
τ ≤ 2kxk2
τ =
τ (cid:1)3
4(cid:0)kx + x∗k2/3
τ . We deduce that
τ =(cid:13)(cid:13) 1
τ (x) = τ (a) + iτ (b) =(cid:0)τ (a)2 + τ (b)2(cid:1) 1
Using (4.4.1), one can show that Kτ := (cid:8)x : kxkτ = 0(cid:9) is a closed, two-sided
ideal. If Kτ = {0}, then τ is said to be faithful. If A is simple, then τ is automati-
cally faithful. Define distτ : A × A → [0, ∞) by setting
4(cid:0)kxk2/3
2 ≤(cid:0)kak2
τ (cid:1)3
τ(cid:1) 1
τ + kx∗k2/3
= 2kxk2
τ ,
τ + kbk2
2 ≤ 2kxkτ .
distτ (x, y) := kx − yk2/3
τ = τ(cid:0)(x − y)∗(x − y)(cid:1)1/3
,
for x, y ∈ A. Then distτ is a pseudometric on A. If τ is faithful, then distτ is a
metric; see [Haa14, Lemma 3.5, Definition 3.6]. (In [Haa14], distτ is denoted by
dτ . We introduce different notation since we reserve dτ to denote the dimension
function induced by τ .)
Let ℓ∞(A) denote the C∗-algebra of bounded sequences in A. Set
(4.4.3)
(4.4.4)
A :=(cid:8)(xn)n ∈ ℓ∞(A) : (xn)n is a distτ -Cauchy sequence(cid:9), and
J :=(cid:8)(xn)n ∈ ℓ∞(A) : xn → 0 in the distτ -pseudometric(cid:9),
Set M := A/ J, and let π : A → M denote the quotient ∗-homomorphism. It
follows that τ induces a map A → C by (xn)n 7→ limn→∞ τ (xn). One verifies that
this is a normalized 2-quasitrace on A that vanishes on J. We let ¯τ : M → C be
the induced map satisfying
¯τ (π(x)) := lim
n→∞
τ (xn),
for x = (xn)n ∈ A. Given x ∈ A, the constant sequence (x)n belongs to A. We
obtain a ∗-homomorphism ι : A → M by ι(x) := π((x)n). We have ¯τ ◦ ι = τ . If τ
is faithful, then ι is injective and we can consider A as a sub-C∗-algebra of M .
The algebra M is an AW ∗-algebra and ¯τ is a faithful, normal, normalized 2-
quasitrace on M ; see [Haa14, Section 4]. If τ is an extreme point in QT1→1(A),
then M is a finite AW ∗-factor; see [Haa14, Proposition 4.6].
4.5. Let A be a unital C∗-algebra, and let τ ∈ QT17→1(A).
Let e ∈ A+. Then e∗e ≤ keke. Using at the second step that τ is order-
preserving, we obtain
(4.5.1)
kekτ = τ (e∗e)1/2 ≤ τ (keke)1/2 = kek1/2τ (e)1/2.
Let a ∈ A+ be a contraction. Then the sequence ¯a := (a1/n)n is bounded. Let us
verify that ¯a belongs to A, as defined in (4.4.3). Given m ≤ n, we have a1/m ≤ a1/n.
Using (4.5.1) and ka1/n − a1/mk ≤ 1 at the second step, and using at the last step
that a1/n and a1/m commute, we deduce that
distτ (a1/n, a1/m)1/3 = ka1/n − a1/mk2
τ ≤ τ (a1/n − a1/m) = τ (a1/n) − τ (a1/m).
We have limn τ (a1/n) = dτ (a), which implies that (a1/n)n is a distτ -Cauchy se-
quence. We have verified that ¯a ∈ A, which allows us to set
pa := π(¯a).
Then
(4.5.2)
¯τ (pa) = lim
n
τ (a1/n) = dτ (a),
and ¯τ (p2
a) = lim
n
τ (a2/n) = dτ (a).
18
HANNES THIEL
For each n, we have a2/n ≤ a1/n, and therefore p2
and since ¯τ is faithful on M , it follows that p2
projection of ι(a) in M , that is, the smallest projection acting as a unit on ι(a).
a commute,
a = pa. Note that pa is the support
a ≤ pa. Since pa and p2
Lemma 4.6. Let A be a unital C∗-algebra, let τ ∈ ∂e QT17→1(A), and let a, b ∈ A+.
Then
min{dτ (a), dτ (b)} = sup(cid:8)dτ (c) : c ∈ A+, c - a, b(cid:9).
Proof. The inequality '≥' is clear. Let us show the converse. Without loss of
generality, we may assume that a and b are contractions, and that dτ (a) ≤ dτ (b).
The statement is clear if dτ (a) = 0. Thus, we may assume that 0 < dτ (a). Let
t ∈ [0, ∞) satisfy t < dτ (a). We need to find c ∈ A+ such that
t < dτ (c),
and c - a, b.
Let (M, ¯τ ) be AW ∗-completion of (A, τ ) as described in Paragraph 4.4. Since τ
is extremal, M is a factor; see [Haa14, Proposition 4.6]. Consider A as in (4.4.3).
As explained in Paragraph 4.5, we have ¯a, ¯b ∈ A. Set
Then pa and pb are projections satisfying
pa := π(¯a), and pb := π(¯b).
¯τ (pa)
(4.5.2)
= dτ (a) ≤ dτ (b)
(4.5.2)
= ¯τ (pb).
The order of projections in an AW ∗-factor is determined by its normalized qua-
sitrace; see [Ber72, Proposition 27.1, p.160]. We may therefore choose v ∈ M such
that pa = vv∗ and v∗v ≤ pb. Lift v to a contractive element ¯v = (vn)n in A. Set
wn := a1/nvnb1/n,
for n ∈ N, and ¯w := (wn)n. Then ¯w = ¯a¯v¯b ∈ A, and
π(¯a¯v¯b) = pavpb = v.
Thus, ¯w is also a contractive lift of v. For each n, we have
na1/n ≤ a2/n,
n = a1/nvnb2/nv∗
wnw∗
and thus ¯a2 − ¯w ¯w∗ ≥ 0. We deduce that
τ(cid:0)a2/n − wnw∗
for each n. We have
and thus ¯a2 − ¯w ¯w∗ ∈ J. It follows that
(4.6.1)
lim
n
Using (4.5.1) at the second step, we deduce that
(4.4.2)
a − vv∗ = 0,
π(¯a2 − ¯w ¯w∗) = p2
n(cid:1) =(cid:12)(cid:12)τ (a2/n − wnw∗
n)(cid:12)(cid:12)
τ(cid:0)a2/n − wnw∗
n(cid:1) ≤ lim
n(cid:1)1/2
≤ (cid:13)(cid:13)a2/n − wnw∗
n(cid:13)(cid:13)τ ,
n(cid:13)(cid:13)τ = 0.
n(cid:1)1/2
≤ τ(cid:0)a2/n − wnw∗
ka2/n − wnw∗
n) + wnw∗
n
n(cid:1)1/2
+ τ(cid:0)wnw∗
.
τ (a2/n)1/2 = τ(cid:0)(a2/n − wnw∗
Using (4.6.1) and using that limn τ (a2/n) = dτ (a) > t, we can choose n ∈ N such
that τ (wnw∗
n) > t.
Moreover,
n is contractive, we have dτ (wnw∗
n) > t. Since wnw∗
n) ≥ τ (wnw∗
Hence, c := wnw∗
wnw∗
n = a1/nvnb2/nv∗
n has the desired properties.
na1/n - a, b.
(cid:3)
Theorem 4.7. Let A be a unital C∗-algebra. Then Cu(A) satisfies Edwards' con-
dition for ∂eF[1]7→1(Cu(A)).
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
19
Proof. We identify QT1→1(A) with F[1]→1(Cu(A)), as in Paragraph 2.14. Let
τ ∈ ∂e QT1→1(A). We let τ∞ denote the (unique) extension of τ to a lower-
semicontinuous 2-quasitrace on A ⊗ K. Further, dτ : (A ⊗ K)+ → [0, ∞] denotes
the induced dimension function. Let a, b ∈ (A ⊗ K)+, and let t ∈ R satisfy
t < dτ (a), dτ (b). To verify Edwards' condition, we need to find c ∈ (A ⊗ K)+
such that t < dτ (c) and c - a, b.
We have
dτ (a) = sup
ε>0
dτ (aε),
and dτ (b) = sup
ε>0
dτ (bε).
Thus, we can choose ε > 0 such that
t < dτ (aε), dτ (bε).
We have [aε/2] ≪ [a] and [bε/2] ≪ [b] in Cu(A), which allows us to choose n such
that [aε/2], [bε/2] ≤ n[1]. Note that n[1] is the class of 1 ⊗ 1n, the unit of the
hereditary subalgebra A ⊗ Mn in A ⊗ K. We have
aε = (aε/2 − ε
2 )+, and aε/2 - 1 ⊗ 1n.
Applying Rørdam's Lemma, we can choose a positive element a′ in the hereditary
sub-C∗-algebra generated by 1 ⊗ 1n (that is, in (A ⊗ Mn)+) such that aε ∼ a′.
Similarly, we obtain b′ ∈ (A ⊗ Mn)+ such that bε ∼ b′.
The restriction of 1
n τ∞ to A ⊗ Mn is an extreme, normalized 2-quasitrace of
A ⊗ Mn. Applying Lemma 4.6, we obtain c ∈ (A ⊗ Mn)+ such that t < dτ (c) and
c - a′, b′. Considering c as an element in A ⊗ K, we have c - a, b. This shows that
c has the desired properties.
(cid:3)
5. Realizing chisels
Let K be a compact, convex set. In this section, we introduce chisels, which are
defined as functions in LAff(K) that take the value ∞ except at one extreme point
of K; see Definition 5.3. We show that Edwards' condition implies that strictly
positive chisels are realized; see Proposition 5.4. We start with two preparatory
lemmas.
Lemma 5.1. Let S be a simple, non-elementary Cu-semigroup satisfying (O5) and
(O6), and let λ ∈ F (S). Then λ(S) = {0, ∞} or λ(S) = [0, ∞].
Proof. Let a ∈ S such that λ(a) 6= 0, ∞. We need to verify that λ(S) = [0, ∞]. It
follows from Robert's Glimm Halving for simple Cu-semigroups, [Rob13a, Propo-
sition 5.2.1], that for every n ∈ N there exists x ∈ S with x 6= 0 and nx ≤ a. It
follows that for every n ≥ 1 there exists yn ∈ S with 0 < λ(yn) < 1
n . Let t ∈ (0, ∞).
n=1 knyn)l is
We may choose kn ∈ N such that t =Pn knλ(yn). The sequence (Pl
increasing. Its supremum y satisfies
λ(y) = λ sup
l
knyn! = sup
l
λ lXn=1
lXn=1
knyn! = sup
l
lXn=1
knλ(yn) = t,
(cid:3)
which shows that t ∈ λ(S).
Lemma 5.2. Let S be a simple, non-elementary Cu-semigroup satisfying (O5) and
(O6), let u ∈ S be a compact, full element, set K := Fu7→1(S), and let λ ∈ ∂eK
and µ ∈ K with λ 6= µ. Assume that S satisfies Edwards' condition for λ. Then
for every n ≥ 1 there exists a ∈ S with λ(a) < 1
n and µ(a) ≥ n.
Proof. We first show that λ (cid:2) µ. By [Rob13a, Proposition 2.2.3], the cone F (S)
is algebraically ordered. Thus, assuming λ ≤ µ, there exists ϕ ∈ F (S) such that
λ + ϕ = µ. Since λ(u) = µ(u) = 1, we obtain ϕ(u) = 0 and consequently ϕ = 0.
But then λ = ν, a contradiction.
20
HANNES THIEL
Thus, we have λ (cid:2) µ, which allows us to choose x ∈ S with λ(x) > µ(x). Choose
x′ ∈ S with x′ ≪ x and λ(x′) > µ(x′). Let n ≥ 1. Choose k, m ∈ N such that
Then
k ≤ mλ(x′),
and mµ(x′) ≤ (k − n).
k − 1
n < λ(ku), λ(mx′).
Using that S satisfies Edwards' condition for λ, we obtain y ∈ S such that
k − 1
n < λ(y),
and y ≤ ku, mx′.
Choose y′ ∈ S with
Applying (O5) for y′ ≪ y ≤ ku, we obtain a ∈ S such that
y′ ≪ y, and k − 1
n < λ(y′).
y′ + a ≤ ku ≤ y + a.
Then
k − 1
n + λ(a) < λ(y′) + λ(a) ≤ λ(ku) = k,
which implies λ(a) < 1
n . Using that y ≤ mx′, we deduce
k = µ(ku) ≤ µ(y) + µ(a) ≤ µ(mx′) + µ(a) ≤ k − n + µ(a),
which implies that n ≤ µ(a).
(cid:3)
Let K be a compact, convex set, and let λ ∈ ∂eK. Set U := K \ {λ}, which is an
open, convex set. Therefore the characteristic function of U is lower semicontinuous
and concave. Multiplying by ∞, the function becomes affine, which justifies the
following:
Definition 5.3. Let K be a compact, convex set, and let λ ∈ ∂eK. We let
σλ ∈ LAff(K) be given by σλ(λ) = 0 and σλ(µ) = ∞ for µ 6= λ. Given t ∈ R, the
chisel at λ with value t is the function t + σλ ∈ LAff(K).
Proposition 5.4. Let S be a countably based, simple, non-elementary Cu-sem-
igroup satisfying (O5) and (O6), let u ∈ S be a compact, full element, and set
K := Fu7→1(S). Assume that S satisfies Edwards' condition for ∂eK.
Then for every λ ∈ ∂eK and t > 0 there exists a ∈ S withbaK = t + σλ.
Proof. Let λ ∈ ∂eK and t > 0. Consider the set
L :=(cid:8)a : λ(a) < t(cid:9).
It follows from Proposition 4.3 that L is upward directed. Since S is countably
based, every upward directed subset of S has a supremum; see Proposition 2.4.
Thus, we may define
a := sup L.
By Lemma 5.1, we have λ(S) = [0, ∞], which implies λ(a) = t. Given µ ∈ K with
(cid:3)
µ 6= λ, it follows from Lemma 5.2 that µ(a) = ∞. HencebaK = t + σλ.
Theorem 5.5. Let A be a separable, unital, simple, non-elementary, stably finite
C∗-algebra. Set K := QT17→1(A). Then for every τ ∈ ∂eK and t > 0 there exists
a ∈ (A ⊗ K)+ such that dτ (a) = t and dµ(a) = ∞ for every µ ∈ K \ {τ }.
Proof. Let τ ∈ ∂eK and t > 0. Set S := Cu(A). It follows from the properties of
A that S is a countably based, simple, non-elementary, stably finite Cu-semigroup
satisfying (O5) and (O6). The class u := [1A] is a compact, full element in S. Under
the identification of QT(A) with F (S), see Paragraph 2.14, the set K corresponds
to Fu7→1(S), and τ corresponds to dτ .
It follows from Theorem 4.7 that S satisfies Edwards' condition for ∂eK. We
Any a ∈ (A ⊗ K)+ with s = [a] has the desired properties.
may therefore apply Proposition 5.4 for S to obtain s ∈ S such thatbsK = t + σdτ .
(cid:3)
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
21
6. A new Riesz decomposition axiom for Cuntz semigroups
In this section, we introduce a new axiom (O6+) for Cu-semigroups. We show
that this axiom is satisfied by Cuntz semigroups of C∗-algebras with stable rank
one (Theorem 6.4), although it is not satisfied by the Cuntz semigroups of all C∗-
algebras; see Example 6.7. We also show that (O6+) does not simply follow from
weak cancellation and (O6) (which are both known to hold for Cuntz semigroups
of C∗-algebras with stable rank one); see Example 6.8.
Definition 6.1. Let S be a Cu-semigroup. We say that S satisfies strengthened
almost Riesz decomposition, or that S satisfies (O6+), if for every a, b, c, x′, x, y′, y ∈
S satisfying
a ≤ b + c,
x′ ≪ x ≤ a, b, and y′ ≪ y ≤ a, c,
there exist e, f ∈ S such that
a ≤ e + f,
x′ ≪ e ≤ a, b, and y′ ≪ f ≤ a, c.
Remark 6.2. Axiom (O6+) is a strengthening of the axiom of 'almost Riesz de-
composition', also called (O6), as introduced by Robert in [Rob13a]. It is known
that (O6) holds for the Cuntz semigroup of every C∗-algebra; see [Rob13a, Propo-
sition 5.1.1]. The strengthening of (O6) to obtain (O6+) is similar to the step from
the original formulation of the axiom of 'almost algebraic order' to its strengthened
version as introduced in [APT18, Definition 4.1]. All these axioms state that a cer-
tain problem (to find a complement, or to decompose an element below a sum) has
an approximate solution, but the strengthened versions state that the element(s)
can be found to improve a previously known (partial) solution of the problem.
Note that the strengthened version of (O5) holds for the Cuntz semigroup of
every C∗-algebra; see [APT18, Theorem 4.7]. However, in Example 6.7 we show
that (O6+) does not hold for Cuntz semigroups of all C∗-algebras.
Lemma 6.3. Let S be a Cu-semigroup. Assume that for every a′, a, b, c, x′, x ∈ S
satisfying
a′ ≪ a ≤ b + c,
and x′ ≪ x ≤ a, b,
there exists e ∈ S such that
a′ ≤ e + c,
and x′ ≪ e ≤ a, b.
Then S satisfies (O6+).
Proof. Claim 1: Let a, b, c, x′, x ∈ S satisfy
a ≤ b + c, and x′ ≪ x ≤ a, b.
Then there exists e ∈ S such that
a ≤ e + c, and x′ ≪ e ≤ a, b.
To verify the claim, set e′
0 ≪ e0 ≪ x. Then
choose a ≪-increasing sequence (an)n≥0 with supremum a and such that a0 = e0.
For each n ≥ 1, choose a′
n ≪ an. We will inductively find
elements e′
0 := x′, and choose e0 such that e′
n satisfying an−1 ≪ a′
n, en, for n ≥ 1, such that
n ≤ e′
a′
n + c, and e′
n−1 ≪ e′
n ≪ en ≤ an+1, b,
for n ≥ 1. For the base case of the induction, we have e′
0 = x′ and e0 = a0.
For the induction step, let n ≥ 1, and assume that we have chosen e′
k and ek for
k ≤ n − 1. To obtain e′
n and en, we use that
an ≪ an+1 ≤ b + c, and e′
By assumption, we obtain en ∈ S such that
an ≤ en + c, and e′
n−1 ≪ en−1 ≤ an+1, b.
n−1 ≪ en ≤ an+1, b.
22
HANNES THIEL
Using that a′
n ≪ an and e′
n + c, and e′
n and en have the desired properties.
a′
n ≤ e′
Then e′
n−1 ≪ en, we can choose e′
n−1 ≪ e′
n such that
n ≪ en.
Using that (e′
n)n is an increasing sequence, we may set e := supn e′
n. For each
n + c, and therefore a ≤ e + c. Further, we have
n ≤ b, and
n ≤ an+2 ≤ a and e′
1 ≤ e. Moreover, for each n we have e′
n we have an−1 ≤ a′
x′ = e′
thus e ≤ a, b. This proves the claim.
n ≤ e′
0 ≪ e′
To verify that S satisfies (O6+), let a, b, c, x′, x, y′, y ∈ S satisfy
a ≤ b + c, and x′ ≪ x ≤ a, b, and y′ ≪ y ≤ a, c.
Applying claim 1 for a ≤ b + c and x′ ≪ x ≤ a, b, we obtain e ∈ S such that
a ≤ e + c, and x′ ≪ e ≤ a, b.
Applying claim 1 for a ≤ e + c and y′ ≪ y ≤ a, c, we obtain f ∈ S with
a ≤ e + f, and y′ ≪ f ≤ a, c.
It follows that e and f have the desired properties.
(cid:3)
Recall that a unital C∗-algebra is said to have stable rank one if its invertible
elements are dense. A nonunital C∗-algebra is defined to have stable rank one
if its minimal unitization does. Stable rank one passes to quotients, hereditary
sub-C∗-algebras, matrix algebras and stabilizations.
Theorem 6.4. Let A be a C∗-algebra that has stable rank one. Then Cu(A) sat-
isfies (O6+).
Proof. We use the notation for Cuntz (sub)equivalence and ε-cut-downs as in
Paragraph 2.12. Given δ > 0, let fδ : [0, ∞) → [0, 1] be the continuous function
that takes value 0 at 0, that takes value 1 on [δ, ∞), and that is linear on the inter-
val [0, δ]. Given z ∈ A+, the element fδ(z) acts as a unit on zδ, and consequently
fδ(z)d = d = dfδ(z) for every d ∈ zδAzδ.
Since stable rank one passes to stabilizations, we may assume that A is stable.
Set S := Cu(A). To verify the assumption of Lemma 6.3 for S, let a′, a, b, c, x′, x ∈ S
satisfy
a′ ≪ a ≤ b + c, and x′ ≪ x ≤ a, b.
Choose x ∈ S such that x′ ≪ x ≪ x. Choose r, s, t ∈ A+ representing a, b and c,
respectively. We may assume that s and t are orthogonal. Then b + c = [s + t].
Choose ε > 0 such that x, a′ ≤ [rε]. Since x ≪ x, we can choose δ′ > 0 such that
x ≤ [sδ′ ].
Since r - s + t, there exists δ > 0 and r ∈ A+ such that rε ∼ r ⊆ (s + t)δ. We
may assume that δ ≤ δ′, so that also x ≤ [sδ]. Choose g ∈ A+ representing x. Take
α > 0 such that x′ ≪ [gα]. We have
Thus, there exist β > 0 and g ∈ A+ such that gα ∼ g ⊆ rβ. Then fβ(r) acts as a
unit on g and therefore
[g] = x ≤ [rε] = [r].
g = fβ(r)gfβ(r) ≤ kgkfβ(r)2 ≤ kgkβ−1r.
We have x′ ≪ [gα] = [g]. Choose η > 0 such that x′ ≪ [gη].
Set B := (s + t)δA(s + t)δ. We have
g ∼ gα - g - sδ,
in A. Since both g and sδ belong to B, we deduce that g - sδ in B. Since stable
rank one passes to hereditary sub-C∗-algebras, B has stable rank one. Thus, there
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
23
exists a unitary u in the minimal unitization of B such that ugηu∗ ⊆ sδ; see [Rør92,
Proposition 2.4]. Set
g := ugηu∗, and r := uru∗.
Then r ∼ r ⊆ (s + t)δ. Set
e := [rfδ(s)r], and f := [rfδ(t)r].
We have
rfδ(s)r ∼ fδ(s)1/2r2fδ(s)1/2 - r2 ∼ r, and rfδ(s)r - fδ(s) - s,
and therefore e ≤ a and e ≤ b. Similarly, we obtain f ≤ a, c.
Since s and t are orthogonal, we have fδ(s + t) = fδ(s) + fδ(t). Using this at the
fifth step, and using at the fourth step that fδ(s + t) acts as a unit on r, we deduce
a′ ≤ [rε] = [r] = [r2] = [rfδ(s + t)r]
= [rfδ(s)r + rfδ(t)r]
≤ [rfδ(s)r] + [rfδ(t)r] = e + f ≤ e + c.
For κ := kgkβ−1 we have g ≤ κr and consequently
g = ugηu∗ ≤ ugu∗ ≤ κuru∗ = κr.
Then
which implies
as desired.
g = fδ(s)1/2gfδ(s)1/2 ≤ κfδ(s)1/2rfδ(s)1/2 ∼ rfδ(s)r,
x′ ≪ [gη] = [ugηu∗] = [g] ≤ [rfδ(s)r] = e,
(cid:3)
Remark 6.5. Theorem 6.4 also holds if A ⊗ K is only assumed to 'have almost
stable rank one'. Recall that a C∗-algebra A is said to have almost stable rank one
if every hereditary sub-C∗-algebra of A is contained in the closure of the invertible
elements of its minimal unitization; see [Rob16, Definition 3.1].
In the proof of
Theorem 6.4, the assumption of stable rank one was only used to obtain the unitary
u in the minimal unitization of B. The existence of such a unitary follows also if
B is contained in the closure of the invertible elements of its minimal unitization.
If A has stable rank one, then it has almost stable rank one. The converse holds
if A is unital. It was shown by Robert, [Rob16, Corollary 3.2], that every (even not
necessarily simple) Z-stable, projectionless C∗-algebra has almost stable rank one.
It is not known if such algebras actually have stable rank one.
While stable rank one passes from a C∗-algebra to its stabilization, the same
does not hold for almost stable rank one; see Example 6.6. Thus, to obtain (O6+)
for the Cuntz semigroup of a C∗-algebra A, we need to require almost stable rank
for A ⊗ K and not just for A.
We thank Leonel Robert for providing the following example.
Example 6.6 (Robert). Almost stable rank one does not pass to matrix algebras.
The Cuntz semigroup of the Jiang-Su algebra Z is isomorphic to N⊔(0, ∞], with the
elements in N being compact (given by classes of projections), and with elements in
(0, ∞] being soft. Choose a positive element h in Z representing the soft 1, and set
A := hZh, the hereditary sub-C∗-algebra of Z generated by h. Then A is Z-stable,
non-unital, and projectionless. However, A ⊗ M2 contains the class of the unit of
Z. Hence, Z is isomorphic to a hereditary sub-C∗-algebra of A. In particular, A is
projectionless, but not stably projectionless.
Let D := {z ∈ C : z ≤ 1} be the 2-disc, and consider B := C(D, A). This
C∗-algebra is again Z-stable and projectionless. Therefore, it has almost stable
rank one, by [Rob16, Corollary 3.2]. However, B ⊗ M2 contains C(D, Z) as a
24
HANNES THIEL
hereditary sub-C∗-algebra. Let f : D → Z be given by f (z) = z · 1Z. We claim
that f can not be approximated by invertible elements in C(D, Z). Identify the
boundary of D with the 1-sphere S1. Then fS 1 defines a nontrivial element in
K1(C(S1) ⊗ Z) ∼= K1(C(S1)) ∼= Z. Let g ∈ C(D, Z) be invertible. Then gS 1
defines the trivial element in K1(C(S1) ⊗ Z). However, if kf − gk∞ < 1, then fS 1
and gS 1 define the same element in K1(C(S1) ⊗ Z). Hence, kf − gk∞ ≥ 1.
Therefore C(D, Z) does not have stable rank one. Since C(D, Z) is a unital,
hereditary sub-C∗-algebra of B ⊗ M2, it follows that B ⊗ M2 does not have almost
stable rank one.
In Proposition 3.9, we have observed that for certain Cu-semigroups, (O6) is
equivalent to (O6+). However, while the Cuntz semigroup of every C∗-algebra
satisfies (O6), the next example shows that the same does not hold for (O6+).
Example 6.7. Axiom (O6+) does not hold for the Cuntz semigroups of all C∗-
algebras. Consider C(S2), the C∗-algebra of continuous functions on the 2-sphere.
Nonzero vector bundles over S2 are classified by their ranks (taking values in N>0)
together with their first Chern classes (taking values in Z).
It follows that the
Murray-von Neumann semigroup of C(S2) is
V (C(S2)) ∼=(cid:8)(0, 0)(cid:9) ∪(cid:8)(n, m) : n > 0(cid:9) ⊆ Z × Z,
with the algebraic order.
Let Lsc(S2, N) denote the set of lower semicontinuous functions S2 → N with
pointwise order and addition. Given an open subset U ⊆ S2, we let 1U ∈ Lsc(S2, N)
denote the characteristic function of U . We set
Lsc(S2, N)nc := Lsc(S2, N) \ {n1S 2 : n ≥ 1},
the set of nonconstant functions in Lsc(S2, N) together with zero. By [Rob13b,
Theorem 1.2], we have
Cu(C(S2)) ∼= V (C(S2)) ⊔(cid:0) Lsc(S2, N) \ {n1S 2 : n ≥ 1}(cid:1)
∼= (N>0 × Z) ⊔ Lsc(S2, N)nc.
Let us describe the order and addition. We retain the usual addition and order (alge-
braic and pointwise, respectively) in the two components N>0 × Z and Lsc(S2, N)nc.
Let (n, m) ∈ N>0 × Z and let f ∈ Lsc(S2, N)nc. We set (n, m) + f := n1S 2 + f in
Lsc(S2, N)nc. We have (n, m) ≤ f if and only if n1S 2 ≤ f in Lsc(S2, N). Similarly,
f ≤ (n, m) if and only if f ≤ n1S 2 .
Let U, V ⊆ S2 be proper, nonempty, open subsets such that the closure of U is
contained in V . Then 1U ≪ 1V in Cu(C(S2)), and 1V , 1U ∈ Lsc(S2, N)nc. We
have
(1, 0) ≤ (1, 1) + 1U , and 1U ≪ 1V ≤ (1, 0), (1, 1).
If Cu(C(S2)) satisfied (O6+), then there would be e ∈ Cu(C(S2)) such that
(1, 0) ≤ e + 1U , and 1U ≤ e ≤ (1, 0), (1, 1).
The condition e ≤ (1, 0), (1, 1) implies that e belongs to Lsc(S2, N)nc. Thus, there
exists t ∈ S2 with e(t) = 0. Since 1U ≤ e, we have t /∈ U . But then (e + 1U )(t) = 0,
which implies that (1, 0) is not dominated by e + 1U , a contradiction.
Example 6.8. Axiom (O6+) does not follow from weak cancellation and (O6).
For example, consider again S := Cu(C(S2)). Since S is the Cuntz semigroup of
a C∗-algebra, it satisfies (O6) (and also (O5)). Using the description of S from
Example 6.7, it is straightforward to check that S is weakly cancellative. However,
by Example 6.7, S does not satisfy (O6+).
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
25
Remark 6.9. The Cuntz semigroup of a C∗-algebra with stable rank one need
not satisfy Riesz decomposition. Consider for example the Jiang-Su algebra Z. We
have Cu(Z) ∼= N ⊔ (0, ∞]. The elements in N are compact, while the elements in
(0, ∞] are soft.
3 . We have a = 1 ≤ 4
Let a = 1, the compact element of value one, and let b = c = 2
3 , the soft element
of value 2
3 = b + c. The only compact elements below 1 are 0
and 1. Moreover, the sum of any element with a nonzero soft element is again soft.
Therefore, if a = a1 + a2, then either a1 = 0 and a2 = 1, or reversed. In either case,
we do not have a1, a2 ≤ 2
3 .
On the positive side, Perera has shown that the Cuntz semigroup of a σ-unital
C∗-algebra with stable rank one and real rank zero satisfies Riesz decomposition;
see [Per97, Theorem 2.13].
7. Realizing functional infima
Let S be a simple, stably finite Cu-semigroup satisfying (O5) and (O6), let
u ∈ S be a compact, full element, and set K := Fu7→1(S). Then LAff(K) is an
inf-semilattice by Lemma 3.8 and Lemma 3.7. We will consider the property that
Definition 7.1. Let S be a simple, stably finite Cu-semigroup satisfying (O5) and
(O6), let u ∈ S be a compact, full element, and set K := Fu7→1(S). We say that
S realizes functional infima over K if for all a, b ∈ S there exists c ∈ S satisfying
the set {baK : a ∈ S} is closed under infima:
bcK =baK ∧bbK.
In Theorem 7.5, we show that (O6+) together with Edwards' condition ensures
that S realizes functional infima. We prepare the proof in a sequence of lemmas.
In the following result, note that the backward implications in (1) and (2) follow
immediately from c ≤ a, b.
Lemma 7.2. Let S be a Cu-semigroup satisfying (O6+), let B ⊆ S be a countable
set, and let z′, z, a, b ∈ S satisfy z′ ≪ z ≤ a, b. Then there exists c ∈ S such that
z′ ≪ c ≤ a, b and:
(1) For every t ∈ B, we have a ≤ b + t if and only if a ≤ c + t.
(2) For every t ∈ B, we have b ≤ a + t if and only if b ≤ c + t.
Proof. Let Ba be the set of elements t ∈ B such that a ≤ b + t, and similarly for Bb.
Choose a sequence (tn)n of elements in Ba such that every element of Ba appears
cofinally often. Similarly, choose the sequence (rn)n of elements in Bb.
Choose ≪-increasing sequences (an)n and (bn)n with supremum a and b, respec-
tively. We will inductively choose elements x′
n, xn, y′
n, yn ∈ S satisfying
an ≤ x′
n + tn, and y′
n−1 ≪ x′
n ≪ xn ≤ a, b,
and
Set x′
0, x0 := 0, y′
for k ≤ n − 1. We have a ≤ b + tn and y′
obtain xn such that
n + rn, and x′
bn ≤ y′
0 := z′ and y0 := z. Assume that we have chose x′
n ≪ yn ≤ a, b.
n ≪ y′
k, yk
n−1 ≪ yn−1 ≤ a, b. Applying (O6+), we
k, xk, y′
a ≤ xn + tn, and y′
n−1 ≪ xn ≤ a, b.
Using that an ≪ a, we can choose x′
n such that
an ≤ x′
n + tn, and y′
n−1 ≪ x′
n ≪ xn.
We have b ≤ a + rn and x′
n ≪ xn ≤ a, b. Applying (O6+), we obtain yn such that
b ≤ yn + rn, and x′
n ≪ yn ≤ a, b.
26
HANNES THIEL
Using that bn ≪ b, we can choose y′
n such that
n + rn, and x′
bn ≤ y′
n ≪ y′
n ≪ yn.
Set c := supn x′
n. Note that c = supn y′
therefore c ≤ a, b. Moreover, we have z′ = y′
n. For each n, we have x′
0 ≪ c.
n ≤ a, b, and
Let t ∈ Ba. By the choice of the sequence (tn)n, there exists a strictly increasing
sequence (nk)k of natural numbers such that t = tnk for each k. We have
ank ≤ x′
nk + tnk = x′
nk + t ≤ c + t,
for each k ∈ N, and therefore a ≤ c + t.
Similarly, we obtain b ≤ c + t for every t ∈ Bb.
(cid:3)
Lemma 7.3. Let S be countably based Cu-semigroup satisfying (O5) and (O6+),
and let z′, z, a, b ∈ S satisfy z′ ≪ z ≤ a, b. Then there exists c ∈ S satisfying
z′ ≪ c ≤ a, b and:
(1) Given y′ ≪ y ≤ a, b, there exists r ∈ S such that y′ + r ≤ a ≤ c + r.
(2) Given y′ ≪ y ≤ a, b, there exists s ∈ S such that y′ + s ≤ b ≤ c + s.
Proof. Using that S is countably based we can choose x′
that x′
that y′ ≤ x′
that
n, xn ∈ S, for n ∈ N, such
n ≪ xn for each n, and such that for any y′ ≪ y ≤ a, b there exists n such
n ≪ xn ≤ a to obtain rn such
n ≪ xn ≤ y. Given n, apply (O5) for x′
Similarly, we obtain sn such that
x′
n + rn ≤ a ≤ xn + rn.
x′
n + sn ≤ b ≤ xn + sn.
Let B be the set consisting of the elements rn and sn.
Apply Lemma 7.2 for z′, z, a, b and B to obtain c ∈ S satisfying z′ ≪ c ≤ a, b and
the statements of Lemma 7.2. Let y′, y ∈ S satisfy y′ ≪ y ≤ a, b. By construction,
we can choose n such that y′ ≤ x′
n ≪ xn ≤ y. Then
a ≤ xn + rn ≤ b + rn.
Using Lemma 7.2(1), we obtain a ≤ c + rn. Then
y′ + rn ≤ x′
n + rn ≤ a ≤ c + rn.
Similarly, one deduces the second statement from Lemma 7.2(2).
(cid:3)
Lemma 7.4. Let S be countably based, simple, stably finite Cu-semigroup satisfying
(O5) and (O6+), let u ∈ S be a compact, full element, let z′, z, a, b ∈ S satisfy z′ ≪
z ≤ a, b, and set K := Fu7→1(S). Then there exists c ∈ S satisfying z′ ≪ c ≤ a, b
and such that
for every λ ∈ K with λ(a) < ∞.
sup(cid:8)λ(x) : x ≤ a, b(cid:9) ≤ λ(c),
Proof. Apply Lemma 7.3 for z′, z, a, b to obtain c ∈ S satisfying z′ ≪ c ≤ a, b and
the statements of Lemma 7.3. To show that c has the desired property, let λ ∈ K
satisfy λ(a) < ∞. Let x ∈ S satisfy x ≤ a, b. Let x′ ≪ c. Apply Lemma 7.3(1) to
obtain r ∈ S such that
Then
x′ + r ≤ a ≤ c + r.
λ(x′) + λ(r) ≤ λ(c) + λ(r).
Since λ(a) < ∞, we have λ(r) < ∞, which allows us to cancel it from the above
inequalities. We deduce that
λ(x′) ≤ λ(c).
Since λ(x) = sup{λ(x′) : x′ ≪ x}, we obtain λ(x) ≤ λ(c), as desired.
(cid:3)
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
27
Theorem 7.5. Let S be countably based, simple, stably finite Cu-semigroup satis-
fying (O5) and (O6+), let u ∈ S be a compact, full element, and set K := Fu7→1(S).
Assume that S satisfies Edwards' condition for ∂eK. Then for every a, b ∈ S there
exists c ∈ S satisfying c ≤ a, b andbcK =baK ∧bbK.
Proof. By Lemma 3.8, K is a metrizable Choquet simplex. Let a, b ∈ S. Choose a
≪-increasing sequence (an)n with supremum a. Set
for each n. Then (fn)n is an increasing sequence in LAff(K) with pointwise supre-
mum f .
f :=baK ∧bbK, and fn :=canK ∧bbK,
By Lemma 3.6, LAff(K) is a countably based domain. Therefore, the way-
below relation in LAff(K) agrees with its sequential version, and every element in
LAff(K) is the supremum of a ≪-increasing sequence. We can therefore choose an
increasing sequence (f ′
n ≪ fn for each n, and such that
supn f ′
n)n in LAff(K) such that f ′
n = f . We may assume f ′
0 ≤ 0.
We will inductively construct c′
n, cn ∈ S for n ≥ 0 such that
c′
n−1 ≪ c′
n ≪ cn ≤ an, b, and f ′
nK,
n ≤ bc′
for n ≥ 1. Set c′
Lemma 7.4 for c′
0 = c0 = 0. Assume that we have chosen c′
n−1 ≪ cn−1 ≤ an, b to obtain cn ∈ S satisfying
k, ck for k ≤ n − 1. Apply
c′
n−n ≪ cn ≤ an, b, and sup(cid:8)λ(x) : x ≤ an, b(cid:9) ≤ λ(cn),
for every λ ∈ K with λ(an) < ∞. Since an ≪ a ≤ ∞u, there exists N ∈ N such
that an ≤ N u. Every λ ∈ K satisfies λ(u) = 1 and hence λ(an) < ∞. Given
λ ∈ ∂eK, we apply Edwards' condition at the first step to obtain
min{λ(an), λ(b)} = sup(cid:8)λ(x) : x ≤ an, b(cid:9) ≤ λ(cn) ≤ min{λ(an), λ(b)}.
It follows that bcn(λ) = min{λ(an), λ(b)} = fn(λ), for all λ ∈ ∂eK. Thus, bcnK = fn,
n−1 ≪ cn to choose c′
by Proposition 3.4.
n ∈ S such that
n ≪ fn and c′
Use that f ′
c′
n−n ≪ c′
n ≪ cn, and f ′
nK.
n ≤ bc′
n bc′
nK =bcK,
f = sup
n
f ′
n ≤ sup
The sequence (c′
n)n is increasing, which allows us to set c := supn c′
n. We have
c′
n ≤ a, b for each n, and therefore c ≤ a, b. Further, we have
which shows that c has the desired properties.
(cid:3)
Remark 7.6. In [APRT18] we show that (O6+) together with a new technique
of adjoining compact elements can be used to show the existence of infima (not
just functional infima). In particular, it follows that Cuntz semigroups of separable
C∗-algebras with stable rank one are inf-semilattices.
8. Realizing ranks
Throughout this section, we fix a weakly cancellative, countably based, simple,
stably finite, non-elementary Cu-semigroup S satisfying (O5) and (O6+), together
with a compact, full element u ∈ S, and we set K := Fu7→1(S). We also assume
that S satisfies Edwards' condition for ∂eK. Note that K is a Choquet simplex.
g + ε.
Lemma 8.1. For every g ∈ Aff(K)++ and ε > 0 there exists a ∈ S with g ≤baK ≤
28
HANNES THIEL
Proof. Let g ∈ Aff(K)++ and ε > 0. For each λ ∈ ∂eK, apply Proposition 5.4 to
obtain aλ ∈ S with
By Lemma 8.5, we have ca′
gλ ∈ Aff(K) satisfying
We may assume that aλ is soft; see Remarks 2.10. By Lemma 3.6, we have
g ≪ g + ε
2 ≤caλK in LAff(K), which allows us to choose a′
λK.
a′
2 + σλ.
λ ∈ S such that
caλK = g(λ) + ε
λ ≪ aλ, and g ≤ca′
λK ≪ caλK in LAff(K). Use Lemma 3.6 to choose
ca′
λK ≤ gλ ≤caλK.
hM := ^λ∈M
gλ.
Consider the family F of finite subsets of ∂eK, which is an upward directed set
ordered by inclusion. For each M ∈ F set
Then hM ≤ hN if M, N ∈ F satisfy M ⊇ N . Thus, (hM )M∈F is a downward
directed net in Aff(K). Set
h := lim
M∈F
hM = inf
M∈F
hM ,
the pointwise infimum of the family (hM )M∈F . Then h is an upper semicontinuous,
affine function on K. For each λ ∈ ∂eK, we have
h(λ) ≤ h{λ}(λ) = gλ(λ) ≤caλ(λ) = g(λ) + ε
2 − h(λ) for all λ ∈ ∂eK. Since the function g + ε
This implies that 0 ≤ g(λ) + ε
belongs to LAff(K), we may apply Proposition 3.4 to deduce that 0 ≤ g + ε
Thus, h ≤ g + ε
2 .
2 .
By Lemma 3.6, we have −g − ε ≪ −g − ε
2 . Hence, the function g + ε is way-
above g + ε
2 in the sense that for every downward directed subset D of upper
semicontinuous functions with inf D ≤ g + ε
2 , there exists d ∈ D with d ≤ g + ε.
Thus, we obtain M ∈ F such that hM ≤ g + ε. Applying Theorem 7.5, we obtain
a ∈ S such that
2 − h
2 − h.
For each λ ∈ M , we have g ≤ca′
g ≤baK = ^λ∈Mca′
which shows that a has the desired properties.
λK.
baK = ^λ∈Mca′
λK and therefore g ≤baK. Then
λK ≤ ^λ∈M
gλ = hM ≤ g + ε,
(cid:3)
Remark 8.2. In the proof of Lemma 8.1, the assumption (O6+) is only used to
realize functional infima over K.
Lemma 8.3. For every f ∈ LAff(K)++, the set
L′
is upward directed.
Proof. Let f ∈ LAff(K)++. To show that L′
satisfy
f :=(cid:8)a′ ∈ S : ∃ a ∈ S with a′ ≪ a, andbaK ≪ f in LAff(K)(cid:9),
a′ ≪ a, and baK ≪ f, and b′ ≪ b, and bbK ≪ f.
By Lemma 3.6, LAff(K) is a domain, which allows us to choose f ′ such that
We need to find c′, c ∈ S such that a′, b′ ≤ c′ ≪ c andbcK ≪ f .
baK,bbK ≤ f ′ ≪ f . Apply Lemma 3.6 to choose g ∈ Aff(K) and ε > 0 such that
f is upward directed, let a′, a, b′, b ∈ S
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
29
f ′ ≤ g and g + 2ε ≤ f . Note thatbaK,bbK ≤ g and g + ε ≪ f . We may also assume
that a, b ≪ ∞. This allows us to choose n ∈ N such that a, b ≤ nu and g ≤ cnuK.
Then there exists h ∈ Aff(K) such that g + h =cnuK.
a′ + r ≤ nu ≤ a + r, and b′ + s ≤ nu ≤ b + s.
Apply (O5) to obtain r, s ∈ S such that
Then
g + h =cnuK ≤baK +brK ≤ g +brK,
we obtain z ∈ S satisfying
and therefore h ≤brK. Similarly, we deduce that h ≤bsK. Applying Theorem 7.5,
Choose z′ ∈ S with z′ ≪ z and h − ε ≤ bz′
h ≤bzK, and z ≤ r, s.
K. Apply (O5) to obtain c ∈ S such that
z′ + c ≤ nu ≤ z + c.
Then
Since S is weakly cancellative, we deduce that a′ ≪ c. Similarly, b′ ≪ c. Then
a′ + r ≤ nu ≪ nu ≤ z + c ≤ r + c.
h − ε +bcK ≤ bz′
K +bcK ≤cnuK = g + h,
the desired properties, which shows that L′
and thusbcK ≤ g + ε ≪ f . Choose c′ ≪ c such that a′, b′ ≤ c′. Then c′ and c have
By Proposition 2.4, every upward directed set in a countably based Cu-semi-
f is upward directed.
(cid:3)
group has a supremum. This justifies the following:
Definition 8.4. We define α : LAff(K)++ → S by
α(f ) := sup L′
f ,
for f ∈ LAff(K)++.
in LAff(K).
Proof. Choose c ∈ S with a ≪ c ≪ b. Since b is soft, we have c <s b. This implies
Lemma 8.5. Let a, b ∈ S satisfy a ≪ b, and assume that b is soft. ThenbaK ≪bbK
that we can choose ε > 0 such that (1 + ε)bc ≤bb. By [Rob13a, Lemma 2.2.5],
to K, is an order-isomorphism. It follows thatbaK ≪bbK in LAff(K).
Lf :=(cid:8)a ∈ S :baK ≪ f in LAff(K)(cid:9),
Since S is simple, the map LAff(F (S)) → LAff(K), given by restricting a function
(cid:3)
and Sf =(cid:8)a ∈ Ssoft :baK ≤ f(cid:9).
ba ≪ (1 + ε)bc ≤bb.
Then α(f ) = sup Lf = max Sf . In particular, α(f ) is soft.
Proposition 8.6. Let f ∈ LAff(K)++. Consider the sets
Proof. To verify that α(f ) is an upper bound for Lf , let a ∈ Lf . For every a′ ∈ S
with a′ ≪ a we have a′ ∈ L′
f = α(f ). Using a = sup{a′ :
a′ ≪ a}, we deduce that a ≤ α(f ). Since L′
f ⊆ Lf , we deduce that sup Lf exists
and agrees with sup L′
f .
f and therefore a′ ≤ sup L′
To verify that α(f ) is an upper bound for Sf , let a ∈ Sf . For every a′ ∈ S with
a′ ≤ α(f ). Hence, a ≤ α(f ).
Next, we verify that every element in Lf is below an element in Sf . Let a ∈ Lf .
a′ ≪ a we have ba′
K ≪baK by Lemma 8.5, which implies that a′ ∈ Lf and thus
ThenbaK ≪ f . Applying Lemma 3.6, we obtain ε > 0 such that ba + ε ≤ f . By
Proposition 2.7 and Proposition 2.9, we can choose x ∈ Ssoft nonzero withbxK ≤ ε.
30
HANNES THIEL
Since S×
[a + xK ≤ f . Thus, a ≤ a + x ∈ Sf . It follows that sup Sf = sup Lf = α(f ).
soft is absorbing, it follows that a + x is soft; see Paragraph 2.6. Further,
Let B be a countable basis for S. To show that B ∩ L′
f . Use that L′
b1, b2 ∈ B ∩ L′
By definition of L′
a′ ≪ b ≪ a. Then b belongs to B ∩ L′
f , there is a ∈ S with a′ ≪ a andbaK ≪ f . Choose b ∈ B with
f and satisfies a1, a2 ≤ b, as desired.
f with α(f ) = supn an.
We can therefore choose an increasing sequence (an)n in L′
f is upward directed to obtain a′ ∈ L′
f is upward directed, let
f with b1, b2 ≤ a′.
Since every an satisfiescanK ≤ f , we have
[α(f )K = sup
n canK ≤ f.
2 ≪ f.
canK ≤ gn − εn
It follows from Proposition 2.7 that L′
f 6= {0}, and thus α(f ) 6= 0. Hence, α(f ) is
either compact or soft. But if α(f ) were compact, then there would be n such that
α(f ) ≤ an, and therefore [α(f )K ≪ f . Arguing as above, we would obtain x ∈ Ssoft
nonzero such that α(f ) + x ∈ Sf , and hence α(f ) + x ≤ α(f ), contradicting that
It follows that α(f ) is soft. Thus, α(f ) belongs to Sf and is
S is stably finite.
therefore the maximal element in Sf .
(cid:3)
Theorem 8.7. We have dα(f ) = f , for every f ∈ LAff(K)++. In particular, for
every f ∈ LAff(K)++, there exists a ∈ S withbaK = f .
Proof. Let f ∈ LAff(K)++. As observed in the proof of Proposition 8.6, we have
[α(f ) ≤ f . To show the converse inequality, choose an increasing sequence (gn)n in
Aff(K)++ with supremum f . Choose a decreasing sequence (εn)n that converges to
zero. We may assume that g0 − ε0 is strictly positive. We have supn(gn − εn) = f .
2 .
For each n, apply Lemma 8.1 to obtain an satisfying gn − εn ≤canK ≤ gn − εn
Using Lemma 3.6 at the second step, we have
Thus, an belongs to Lf , as defined in Proposition 8.6, which implies an ≤ sup Lf =
α(f ). It follows that
as desired.
n
f = sup
(gn − εn) ≤ sup
n canK ≤ [α(f )K,
(cid:3)
8.8. Consider the map α : LAff(K)++ → S as defined in Definition 8.4. By
Proposition 8.6, the image of α is contained in Ssoft. Set S×
soft := Ssoft \ {0} and de-
fine κ : S×
it follows that κ ◦ α = id and α ◦ κ ≤ id. Given a ∈ S×
soft and f ∈ LAff(K)++,
we have a ≤ α(f ) if and only if κ(a) ≤ f . Thus, α and κ form a (order theoretic)
Galois connection between LAff(K)++ and S×
soft; see [GHK+03, Definition O-3.1,
p.22]. The map α is the upper adjoint, and the map κ is the lower adjoint.
soft → LAff(K)++ by κ(a) :=baK. Using Theorem 8.7 and Proposition 8.6
Proposition 8.9. The map α : LAff(K)++ → Ssoft preserves order, infima and
suprema of directed sequences.
Proof. It follows from Proposition 8.6 that α is order-preserving. By [GHK+03,
Theorem O-3.3, p.24], the upper adjoint of a Galois connection preserves arbitrary
existing infima. Since K is a Choquet simplex, LAff(K)++ is an inf-semilattice;
see Lemma 3.7. Thus, given f, g ∈ LAff(K)++, we have
α(f ∧ g) = α(f ) ∧ α(g).
By [GHK+03, Theorem IV-1.4, p.268], if the lower adjoint of a Galois connection
preserves the way-below relation, and if the source of the lower adjoint is a domain,
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
31
then the upper adjoint preserves suprema of increasing nets. By Lemma 8.5, the
map κ preserves the way-below relation, whence α preserves suprema of increasing
sequences.
(cid:3)
Question 8.10. Is the map α : LAff(F (S))++ → Ssoft additive? Does it preserve
the way-below relation?
Theorem 8.11. Let A be a separable, unital, simple, non-elementary C∗-algebra
with stable rank one. Set K := QT17→1(A). Then for every f ∈ LAff(K)++ there
exists a ∈ (A ⊗ K)+ such that dτ (a) = f (τ ) for every τ ∈ K.
Proof. Let f ∈ LAff(K)++. Set S := Cu(A). It follows from the properties of A
that S is a countably based, simple, non-elementary Cu-semigroup satisfying (O5).
By [RW10, Theorem 4.3], the Cuntz semigroup of a C∗-algebra with stable rank one
is weakly cancellative; see also [APT18, Chapter 4]. It follows from Theorem 6.4
that S satisfies (O6+).
The class u := [1A] is a compact, full element in S. Under the identification
of QT(A) with F (S), see Paragraph 2.14, the set K corresponds to Fu7→1(S). It
follows from Theorem 4.7 that S satisfies Edwards' condition for ∂eK.
We may therefore apply Theorem 8.7 for S and f to obtain s ∈ S such that
(cid:3)
bsK = f . Then any a ∈ (A ⊗ K)+ with s = [a] has the desired property.
Corollary 8.12. Let A be a separable, unital, simple, non-elementary C∗-algebra
with stable rank one. Assume that Cu(A) is almost unperforated, that is, that A has
strict comparison of positive elements. Then Cu(A) is almost divisible. Moreover,
there are order-isomorphisms
Cu(A) ∼= V (A) ⊔ LAff(QT17→1(A))++ ∼= Cu(A ⊗ Z).
The map A → A ⊗ Z, a 7→ a ⊗ 1, induces the isomorphism Cu(A) ∼= Cu(A ⊗ Z).
Proof. Set K := QT17→1(A).
It follows from the properties of A that Cu(A) is
a countably based, simple, stably finite, non-elementary Cu-semigroup satisfying
(O5) and (O6). By Theorem 8.11, Cu(A) satisfies statement (4) of Proposition 2.11.
We deduce that Cu(A) is almost divisible and that κ : Cu(A)soft → LAff(K)++
is an order-isomorphism. There is a natural affine homeomorphism between the
simplex of normalized 2-quasitraces on A and A ⊗ Z. We therefore have an order-
isomorphisms
Cu(A)soft ∼= LAff(K)++ ∼= Cu(A ⊗ Z)soft.
We have an order-isomorphism between Cu(A)c and the Muray-von Neumann semi-
group V (A). Since A is unital and has stable rank one, V (A) is order-isomorphic
to K0(A)+, the positive part of the partially ordered group K0(A). Similarly,
V (A ⊗ Z) ∼= K0(A ⊗ Z)+.
By [GJS00, Theorem 1], the map ι induces an order-isomorphism K0(A) →
K0(A ⊗ Z) if and only if K0(A) is weakly unperforated. This condition is verified
using [Rør04, Section 3]. Thus, we have order-isomorphisms
Cu(A)c ∼= V (A) ∼= K0(A)+ ∼= K0(A ⊗ Z)+ ∼= V (A ⊗ Z) ∼= Cu(A ⊗ Z)c.
The result follows using the decomposition of a simple Cu-semigroup into its com-
pact and soft part; see Proposition 2.7(1).
(cid:3)
9. The Toms-Winter conjecture
Conjecture 9.1 (Toms-Winter). Let A be a separable, unital, simple, non-ele-
mentary, nuclear C∗-algebra. Then the following are equivalent:
(1) A has finite nuclear dimension.
(2) A is Z-stable, that is, A ∼= A ⊗ Z.
32
HANNES THIEL
(3) Cu(A) is almost unperforated.
Remarks 9.2. (1) The nuclear dimension is a non-commutative analogue of topo-
logical covering dimension, introduced by Winter-Zacharias [WZ10, Definition 2.1].
(2) The Jiang-Su algebra Z, introduced in [JS99], is a separable, unital, sim-
ple, non-elementary, approximately subhomogeneous C∗-algebra with unique tra-
cial state and K0(Z) ∼= Z and K1(Z) = 0. Being Z-stable is considered as the
C∗-algebraic analogue of being a McDuff von Neumann factor.
(3) Let A be a unital, simple C∗-algebra. Recall that A is said to have strict
comparison of positive elements if for all a, b ∈ (A ⊗ K)+ we have a - b whenever
dτ (a) < dτ (b) for all τ ∈ QT17→1(A). Let S be a simple Cu-semigroup with a
compact, full element u ∈ S. Then S is almost unperforated if and only if for all
a, b ∈ S we have a ≤ b whenever λ(a) < λ(b) for all λ ∈ Fu7→1(S); see [APT18,
Proposition 5.2.14]. For S = Cu(A) we have a natural identification of QT17→1(A)
with F[1]7→1(Cu(A)), that maps τ to dτ . It follows that A has strict comparison of
positive elements if and only if Cu(A) is almost unperforated.
(4) The regularity conjecture of Toms-Winter is intimately connected to the
Elliott classification program, which seeks to classify simple, nuclear C∗-algebras
by K-theoretical and tracial data. Examples of Toms, [Tom08], showed that an
additional regularity assumption is necessary to obtain such a classification.
Due to a series of remarkable breakthroughs in the last three years, building on
an extensive body of work over decades by numerous people, the classification of
separable, unital, simple, non-elementary C∗-algebras with finite nuclear dimen-
sion that satisfy the universal coefficient theorem (UCT) has been completed; see
[EGLN15] and [TWW17, Corollary D].
In particular, it follows from Theorem 9.6 that separable, unital, simple, approx-
imately subhomogeneous C∗-algebras with stable rank one and strict comparison
of positive elements are classified by K-theoretic and tracial data.
9.3. The implications '(1)⇒(2)' and '(2)⇒(3)' of the Toms-Winter conjecture have
been confirmed in general. The first implication is due to Winter, [Win12, Corol-
lary 7.3]. The second was shown by Rørdam, [Rør04, Theorem 4.5].
The implication '(2)⇒(1)' was shown to hold under the additional assumption
that ∂eT (A) is compact, see [BBS+19, Theorem B], generalizing the earlier solution
of the monotracial case, [MS14], [SWW15]. Very recently, it was shown that the
implication '(2)⇒(1)' holds in general; see Remark 9.11.
The implication '(3)⇒(2)' was shown to hold under the additional assumption
that ∂eT (A) is compact and finite dimensional by independent works of Kirchberg-
Rørdam, [KR14], Sato, [Sat12], and Toms-White-Winter, [TWW15], extending the
work of Matuia-Sato, [MS12], that covered the case of finitely many extreme traces.
This was further extended by Zhang, [Zha14], who showed that the implication
'(3)⇒(2)' also holds in certain cases where ∂eT (A) is finite dimensional but not
necessarily compact.
9.4. Let A be a separable, unital, simple, non-elementary, nuclear C∗-algebra. The
Cuntz semigroup of every Z-stable C∗-algebra is almost unperforated and almost
divisible. Thus, the Toms-Winter conjecture predicts in particular that Cu(A)
is almost divisible whenever it is almost unperforated. Moreover, to prove the
implication '(3)⇒(2)' of the Toms-Winter conjecture, the verification of almost
divisibility of Cu(A) is crucial. Winter showed that it is even enough under the
additional assumption that A has locally finite nuclear dimension; see [Win12,
Corollary 7.4]. By Corollary 8.12, if A has stable rank one, then Cu(A) is almost
divisible whenever it is almost unperforated. Combined with the result of Winter,
we obtain:
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
33
Theorem 9.5. Let A be a separable, unital, simple, non-elementary C∗-algebra
with stable rank one and locally finite nuclear dimension. Then A is Z-stable if
and only if A has strict comparison of positive elements. (See also Remark 9.11.)
It was shown by Elliott-Niu-Santiago-Tikuisis, [ENST15, Theorem 1.1], that the
nuclear dimension of every (not necessarily simple or unital) separable, approxi-
mately subhomogeneous, Z-stable C∗-algebra is finite. Ng-Winter, [NW06], proved
that every separable, approximately subhomogeneous C∗-algebra has locally finite
nuclear dimension. Combining with the above result, we deduce:
Theorem 9.6. The Toms-Winter conjecture holds for approximately subhomoge-
neous C∗-algebras with stable rank one. (See also Remark 9.11.)
Remark 9.7. The class of algebras covered by Theorem 9.6 is very rich and it
includes both Z-stable and non-Z-stable algebras. For example, by [EHT09, The-
orem 4.1], all diagonal AH-algebras have stable rank one, whence they are covered
by Theorem 9.6. This includes in particular the Villadsen algebras of first type,
[Vil98], and Toms' celebrated examples, [Tom08].
9.8. Let X be an infinite, compact, metrizable space, and let α : X → X be a
minimal homeomorphism. Then α induces an automorphism of C(X). The crossed
product A := C(X) ⋊ Z is a separable, unital, simple, non-elementary, nuclear C∗-
algebra. Let u ∈ A be the canonical unitary implementing α. Choose x ∈ X and
set
Ax := C∗(cid:0)C(X), C0(X \ {x})u(cid:1) ⊆ A,
the 'orbit breaking subalgebra' of A corresponding to x. Then Ax is a separable,
unital, simple, non-elementary, approximately subhomogeneous C∗-algebra. By
combining [Phi14, Theorem 7.10] and [AP15, Theorem 4.6], we obtain that Ax
is a 'centrally large subalgebra' of A, in the sense of [AP15, Definition 3.2]. By
[Phi14, Theorem 6.14], Cu(A) is almost unperforated if and only if Cu(Ax) is. By
Theorem 3.3 and Corollary 3.5 in [ABP18], A is Z-stable if and only if Ax is.
Since Ax is approximately subhomogeneous, it follows from [ENST15, Theo-
rem A] that Ax has finite nuclear dimension (denoted dimnuc(Ax) < ∞) whenever
Ax is Z-stable. By [ENST15, Corollary 4.8], the analogous statement holds also
for A. We show the various implications in the following diagram. The state-
ments in the three columns correspond to the three statements of the Toms-Winter
conjecture. The horizontal implications towards the right hold in general; see
Paragraph 9.3.
ENST
dimnuc(A) < ∞
/ A ∼= A ⊗ Z
Cu(A) almost unperforated
ENST
ABP
P
dimnuc(Ax) < ∞
/ Ax ∼= Ax ⊗ Z
Cu(Ax) almost unperforated
Thus, to verify the Toms-Winter conjecture for A and Ax it remains only to verify
that almost unperforation of their Cuntz semigroups implies Z-stability. Moreover,
A satisfies the Toms-Winter conjecture if and only if Ax does.
The large subalgebra Ax is approximately subhomogeneous (and separable, uni-
tal, simple, non-elementary, nuclear). Therefore, we may apply Theorem 9.6 to
verify the Toms-Winter conjecture for Ax if we know that Ax has stable rank one.
It is conjectured by Archey-Niu-Phillips, [AP15, Conjecture 7.2], that A always has
stable rank one. By [AP15, Theorem 6.3], A has stable rank one whenever Ax does
and they use this to verify their conjecture in the case that the dynamical system
has a Cantor factor. We obtain:
/
/
/
t
t
O
O
O
O
/
/
/
t
t
34
HANNES THIEL
Theorem 9.9. Let X be an infinite, compact, metrizable space, together with a
minimal homeomorphism on X. Then the Toms-Winter conjecture holds for the
crossed product A = C(X) ⋊ Z if it has large subalgebra with stable rank one. In
particular, the Toms-Winter conjecture holds for the crossed product A = C(X)⋊ Z
if the dynamical system has a Cantor factor.
It was also shown by Rørdam that every unital, simple, stably finite, Z-stable
C∗-algebra has stable rank one; see [Rør04, Theorem 6.7]. Therefore, if a stably
finite C∗-algebra satisfies either condition (1) or (2) of the Toms-Winter conjecture,
then it has stable rank one. We are led to ask the following:
Question 9.10. Let A be a separable, unital, simple, stably finite, nuclear C∗-al-
gebra such that Cu(A) is almost unperforated. Does A have stable rank one?
Remark 9.11. After this paper had been submitted, it was shown that the im-
plication '(2)⇒(1)' of the Toms-Winter conjecture holds in general; see [CET+19,
Theorem A]. It follows that Theorems 9.5 and 9.6 can be generalized as follows: The
Toms-Winter conjecture holds for C∗-algebras with locally finite nuclear dimension
and stable rank one.
References
[Alf71]
E. M. Alfsen, Compact convex sets and boundary integrals, Springer-Verlag, New
York-Heidelberg, 1971, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 57.
MR 0445271. Zbl 0209.42601.
[APRT18] R. Antoine, F. Perera, L. Robert, and H. Thiel, C ∗-algebras of stable rank one
and their Cuntz semigroups, preprint (arXiv:1809.03984 [math.OA]), 2018.
[APT18] R. Antoine, F. Perera, and H. Thiel, Tensor products and regularity properties of
Cuntz semigroups, Mem. Amer. Math. Soc. 251 (2018), viii+191. MR 3756921.
[ABP18] D. Archey, J. Buck, and N. C. Phillips, Centrally large subalgebras and tracial
[AP15]
[Bau58]
[Bau63]
[Ber72]
[BH82]
[BK04]
Z-absorption, Int. Math. Res. Not. IMRN (2018), 1857 -- 1877. MR 3801476.
D. Archey and N. C. Phillips, Permanence of stable rank one for centrally
large subalgebras and crossed products by minimal homeomorphisms, preprint
(arXiv:1505.00725 [math.OA]), 2015.
H. Bauer, Minimalstellen von Funktionen und Extremalpunkte, Arch. Math. 9 (1958),
389 -- 393. MR 0100774. Zbl 0082.32601.
H. Bauer, Kennzeichnung kompakter Simplexe mit abgeschlossener Extremalpunkt-
menge, Arch. Math. 14 (1963), 415 -- 421. MR 0164053. Zbl 0196.42202.
S. K. Berberian, Baer *-rings, Springer-Verlag, New York-Berlin,
1972,
Die Grundlehren der mathematischen Wissenschaften, Band 195. MR 0429975.
Zbl 0242.16008.
B. Blackadar and D. Handelman, Dimension functions and traces on C ∗-algebras,
J. Funct. Anal. 45 (1982), 297 -- 340. MR 650185. Zbl 0513.46047.
E. Blanchard and E. Kirchberg, Non-simple purely infinite C ∗-algebras: the Haus-
dorff case, J. Funct. Anal. 207 (2004), 461 -- 513. MR 2032998. Zbl 1048.46049.
[BBS+19] J. Bosa, N. P. Brown, Y. Sato, A. Tikuisis, S. White, and W. Winter, Covering
dimension of C ∗-algebras and 2-coloured classification, Mem. Amer. Math. Soc. 257
(2019), vii+97. MR 3908669.
[BPT08] N. P. Brown, F. Perera, and A. S. Toms, The Cuntz semigroup, the Elliott con-
jecture, and dimension functions on C ∗-algebras, J. Reine Angew. Math. 621 (2008),
191 -- 211. MR 2431254. Zbl 1158.46040.
[CET+19] J. Castillejos, S. Evington, A. Tikuisis, S. White, and W. Winter, Nuclear
[Cho69]
[CEI08]
[DT10]
dimension of simple C ∗-algebras, preprint (arXiv:1901.05853 [math.OA]), 2019.
G. Choquet, Lectures on analysis. Vol. II: Representation theory, Edited by J. Mars-
den, T. Lance and S. Gelbart, W. A. Benjamin, Inc., New York-Amsterdam, 1969.
MR 0250012. Zbl 0331.46003.
K. T. Coward, G. A. Elliott, and C. Ivanescu, The Cuntz semigroup as an in-
variant for C ∗-algebras, J. Reine Angew. Math. 623 (2008), 161 -- 193. MR 2458043.
Zbl 1161.46029.
M. Dadarlat and A. S. Toms, Ranks of operators in simple C ∗-algebras, J. Funct.
Anal. 259 (2010), 1209 -- 1229. MR 2652186. Zbl 1202.46061.
RANKS OF OPERATORS IN SIMPLE C ∗-ALGEBRAS WITH STABLE RANK ONE
35
[DHR97] K. Dykema, U. Haagerup, and M. Rørdam, The stable rank of some free product
[Edw69]
C ∗-algebras, Duke Math. J. 90 (1997), 95 -- 121. MR 1478545. Zbl 0905.46036.
D. A. Edwards, On uniform approximation of affine functions on a compact convex
set, Quart. J. Math. Oxford Ser. (2) 20 (1969), 139 -- 142. MR 0250044. Zbl 0177.16303.
[EGLN15] G. A. Elliott, G. Gong, H. Lin, and Z. Niu, On the classification of simple
amenable C ∗-algebras with finite decomposition rank, II, preprint (arXiv:1507.03437
[math.OA]), 2015.
[EHT09] G. A. Elliott, T. M. Ho, and A. S. Toms, A class of simple C ∗-algebras with stable
rank one, J. Funct. Anal. 256 (2009), 307 -- 322. MR 2476944. Zbl 1184.46059.
[GK10]
[Haa14]
[JS99]
[KR14]
[Goo86]
[GJS00]
[ENST15] G. A. Elliott, Z. Niu, L. Santiago, and A. Tikuisis, Decomposition rank of approx-
imately subhomogeneous C ∗-algebras, preprint (arXiv:1505.06100 [math.OA]), 2015.
[ERS11] G. A. Elliott, L. Robert, and L. Santiago, The cone of lower semicontinuous traces
on a C ∗-algebra, Amer. J. Math. 133 (2011), 969 -- 1005. MR 2823868. Zbl 1236.46052.
[GHK+03] G. Gierz, K. H. Hofmann, K. Keimel, J. D. Lawson, M. Mislove, and D. S. Scott,
Continuous lattices and domains, Encyclopedia of Mathematics and its Applications
93, Cambridge University Press, Cambridge, 2003. MR 1975381. Zbl 1088.06001.
J. Giol and D. Kerr, Subshifts and perforation, J. Reine Angew. Math. 639 (2010),
107 -- 119. MR 2608192. Zbl 1201.46055.
G. Gong, X. Jiang, and H. Su, Obstructions to Z-stability for unital simple C ∗-al-
gebras, Canad. Math. Bull. 43 (2000), 418 -- 426. MR 1793944. Zbl 0978.46037.
K. R. Goodearl, Partially ordered abelian groups with interpolation, Mathematical
Surveys and Monographs 20, American Mathematical Society, Providence, RI, 1986.
MR 845783. Zbl 0589.06008.
U. Haagerup, Quasitraces on exact C ∗-algebras are traces, C. R. Math. Acad. Sci.
Soc. R. Can. 36 (2014), 67 -- 92. MR 3241179. Zbl 1325.46055.
X. Jiang and H. Su, On a simple unital projectionless C ∗-algebra, Amer. J. Math.
121 (1999), 359 -- 413. MR 1680321. Zbl 0923.46069.
E. Kirchberg and M. Rørdam, Central sequence C ∗-algebras and tensorial ab-
sorption of the Jiang-Su algebra, J. Reine Angew. Math. 695 (2014), 175 -- 214.
MR 3276157. Zbl 1307.46046.
H. Matui and Y. Sato, Strict comparison and Z-absorption of nuclear C ∗-algebras,
Acta Math. 209 (2012), 179 -- 196. MR 2979512. Zbl 1277.46028.
H. Matui and Y. Sato, Decomposition rank of UHF-absorbing C ∗-algebras, Duke
Math. J. 163 (2014), 2687 -- 2708. MR 3273581. Zbl 1317.46041.
P. W. Ng and W. Winter, A note on subhomogeneous C ∗-algebras, C. R. Math.
Acad. Sci. Soc. R. Can. 28 (2006), 91 -- 96. MR 2310490. Zbl 1138.46037.
F. Perera, The structure of positive elements for C ∗-algebras with real rank zero,
Internat. J. Math. 8 (1997), 383 -- 405. MR 1454480. Zbl 0881.46044.
R. R. Phelps, Lectures on Choquet's theorem, second ed., Lecture Notes in Mathe-
matics 1757, Springer-Verlag, Berlin, 2001. MR 1835574. Zbl 0997.46005.
N. C. Phillips, Large subalgebras, preprint (arXiv:1408.5546 [math.OA]), 2014.
[Phi14]
[Rob13a] L. Robert, The cone of functionals on the Cuntz semigroup, Math. Scand. 113 (2013),
[MS12]
[MS14]
[Per97]
[Phe01]
[NW06]
161 -- 186. MR 3145179. Zbl 1286.46061.
[Rob13b] L. Robert, The Cuntz semigroup of some spaces of dimension at most two, C. R.
[Rob16]
[Rør92]
[Rør04]
[RW10]
[Sat12]
Math. Acad. Sci. Soc. R. Can. 35 (2013), 22 -- 32. MR 3098039. Zbl 1295.46045.
L. Robert, Remarks on Z-stable projectionless C ∗-algebras, Glasg. Math. J. 58
(2016), 273 -- 277. MR 3483583. Zbl 1356.46046.
M. Rørdam, On the structure of simple C ∗-algebras tensored with a UHF-algebra. II,
J. Funct. Anal. 107 (1992), 255 -- 269. MR 1172023. Zbl 0810.46067.
M. Rørdam, The stable and the real rank of Z-absorbing C ∗-algebras, Internat. J.
Math. 15 (2004), 1065 -- 1084. MR 2106263. Zbl 1077.46054.
M. Rørdam and W. Winter, The Jiang-Su algebra revisited, J. Reine Angew. Math.
642 (2010), 129 -- 155. MR 2658184. Zbl 1209.46031.
Y. Sato, Trace spaces of simple nuclear C ∗-algebras with finite-dimensional extreme
boundary, preprint (arXiv:1209.3000 [math.OA]), 2012.
[SWW15] Y. Sato, S. White, and W. Winter, Nuclear dimension and Z-stability, Invent.
Math. 202 (2015), 893 -- 921. MR 3418247. Zbl 1350.46040.
[TWW17] A. Tikuisis, S. White, and W. Winter, Quasidiagonality of nuclear C ∗-algebras,
[Tom08]
Ann. of Math. (2) 185 (2017), 229 -- 284. MR 3583354. Zbl 1367.46044.
A. S. Toms, On the classification problem for nuclear C ∗-algebras, Ann. of Math. (2)
167 (2008), 1029 -- 1044. MR 2415391. Zbl 1181.46047.
36
HANNES THIEL
[TWW15] A. S. Toms, S. White, and W. Winter, Z-stability and finite-dimensional tra-
cial boundaries, Int. Math. Res. Not. IMRN (2015), 2702 -- 2727. MR 3352253.
Zbl 1335.46054.
J. Villadsen, Simple C ∗-algebras with perforation, J. Funct. Anal. 154 (1998), 110 --
116. MR 1616504. Zbl 0915.46047.
[Vil98]
[Win12] W. Winter, Nuclear dimension and Z-stability of pure C ∗-algebras, Invent. Math.
187 (2012), 259 -- 342. MR 2885621. Zbl 1280.46041.
[WZ10] W. Winter and J. Zacharias, The nuclear dimension of C ∗-algebras, Adv. Math.
224 (2010), 461 -- 498. MR 2609012. Zbl 1201.46056.
[Zha14] W. Zhang, Tracial state space with non-compact extreme boundary, J. Funct. Anal.
267 (2014), 2884 -- 2906. MR 3255477. Zbl 1311.46057.
Hannes Thiel Mathematisches Institut, Fachbereich Mathematik und Informatik der
Universitat Munster, Einsteinstrasse 62, 48149 Munster, Germany.
E-mail address: [email protected]
URL: www.math.uni-muenster.de/u/hannes.thiel/
|
1809.10053 | 1 | 1809 | 2018-09-26T15:11:26 | The $\kappa$-Poincar\'e Group on a $C^*$-level | [
"math.OA",
"hep-th",
"math-ph",
"math-ph",
"math.QA"
] | The $C^*$-algebraic $\kappa$-Poincar\'{e} Group is constructed. The construction uses groupoid algebras of differential groupoids associated to Lie group decomposition. It turns out the underlying $C^*$-algebra is the same as for "$\kappa$-Euclidean Group" but a comultiplication is twisted by some unitary multiplier. Generators and commutation relations among them are presented. | math.OA | math |
THE κ-POINCAR´E GROUP ON A C∗-LEVEL.
PIOTR STACHURA
Abstract. The C ∗-algebraic κ-Poincar´e Group is constructed. The construction uses groupoid algebras of
differential groupoids associated to Lie group decomposition. It turns out the underlying C ∗-algebra is the
same as for "κ-Euclidean Group" but a comultiplication is twisted by some unitary multiplier. Generators
and commutation relations among them are presented.
1. Introduction
The history of κ-deformation has begun in 1992 with the work of J. Lukierski, A. Nowicki and H. Ruegg
[5], where this deformation of the enveloping algebra of Poincar´e Group appeared for the first time. Next
important step was the paper by S. Majid and H. Ruegg in 1994 [7] identifying bicrossedproduct structure
of κ-Poincar´e algebra. Since then, a vast literature on the subject has been produced with many attempts
to apply this deformation to physical problems, but because (almost) all work was done on the level of pure
algebra (or formal power series) it was hard to get more then rather formal conclusions. This is not a review
paper and we refer to [6] (also an older one [2]) for a discussion and an extensive bibliography of the subject.
This article, however, is not about the κ-Poincar´e algebra but about the κ-Poincar´e Group - deformation
of an algebra of functions on Poincar´e Group. It appeared in 1994, on a Hopf ∗-algebra level, in the work of
S. Zakrzewski [22], where it was also shown it is a quantization of a certain Poisson-Lie structure. Soon, it
became clear that this particular Poisson structure is not special for dimension 4 but has analogues in any
dimension and is related to certain decompositions of orthogonal Lie algebras [10] and is dual to a certain
Lie algebroid structure [14].
The main result of this work is a topological version of κ-Poincar´e Group. Since [7] it has been clear
that had it existed it should have been given by some bicrossedproduct construction. The main problem is
that the decomposition of a Lie algebra g = a ⊕ c doesn't lift to a decomposition of a Lie Group G = AC
(of course it lifts to a local decomposition, but the complement of the set of decomposable elements i.e.
G \ (AC ∩ CA) has a non empty interior) therefore the construction of S. Vaes and L. Vainerman presented
in [15] can't be directly applied. As it has been shown already in [10] there is a non connected extension
of a group A, let's denote it by A, such that AC ∩ C A is open and dense in G. Therefore it fits into the
framework of [15] and the κ-deformation of Poincar´e Group, or rather its non connected extension, exists
as a locally compact quantum group.
Here we use approach different then used in [15]. Although less general, it has some advantages -- it is
more geometric and it is easier to see that what we get is really a quantization of a Poisson-Lie structure.
It is based on the use of groupoid algebras for differential groupoids naturally related to decompositions of
Lie groups. For a global decomposition this construction was described in [12]; the result is that given a Lie
group G with two closed subgroups B, C ⊂ G satisfying G = BC one can define two differential groupoid
structures on G (over B and C, this is described briefly in the second part of this introduction). It turns out
that C∗-algebras of these groupoids carry quantum group structures, in fact, sweeping under the rug some
1
2
PIOTR STACHURA
universal/reduced algebras problems, one may say that all main ingredients of quantum group structure are
just C∗-lifting of natural groupoid objects. As said above the situation with κ-Poincar´e is not so nice, but
there is a global decomposition "nearby"one can try to use; this framework was described in [13].
Let us explain briefly the construction; geometric details were presented in [14]. By the (restricted)
Poincar´e Group it's meant here the semidirect product of the (restricted) Lorentz Group A := SO0(1, n)
and n + 1-dimensional vector Minkowski space. It turns out, that it can be realized as a subgroup (T A)0 of
T ∗G for G := SO0(1, n + 1), where A is embedded naturally into G (as the stabilizer of a spacelike vector).
On (T A)0 there is a Poisson structure dual to a Lie algebroid structure related to the decomposition of g
-- the Lie algebra of G into two subalgebras g = a ⊕ c, where a is the Lie algebra of A. This Lie algebroid
is the algebroid of the Lie groupoid AC ∩ CA ⊂ G, here C is the Lie subgroup with algebra c. As said
above, the set AC ∩ CA is too small, but one can find A ⊂ G -- a non-connected extension of A, in fact this
is the normalizer of A in G, such that Γ := AC ∩ C A is open and dense in G. The set Γ is a differential
groupoid (over A) and C∗-algebra of this groupoid is the C∗-algebra of κ-Poincar´e Group. The method
used in [13] relies essentially on the fact that the algebra c has a second complementary algebra b and the
decomposition g = c ⊕ b lifts to a global decomposition G = BC and this is just the Iwasawa decomposition
(i.e. B = SO(n + 1)). This global decomposition defines a groupoid GB : G ⇒ B and its C∗-algebra is
the underlying algebra of the quantum group which may be called "Quantum κ-Euclidean Group". Our
groupoid Γ embeds into GB in such a way that it is possible to prove that their C∗-algebras are the same
but the comultiplication of κ-Poincar´e is comultiplication of κ-Euclidean twisted by a unitary multiplier.
So one may say that quantum spaces underlying κ-Poincar´e and κ-Euclidean groups are the same and only
group structures are different.
The embedding Γ ֒→ GB essentially is given by embedding of A ֒→ SO(n + 1) as a dense open subset,
so it is a kind of compactification of A (which consists of two copies of the (restricted) Lorentz Group
SO0(1, n)). This compactification solves the problem, that some natural operators, that "should be" self-
adjoint elements affiliated with C∗(Γ) are defined by non complete vector fields, so they are not essentially
self-adjoint on their "natural" domains and this embedding just defines "correct" domains.
In this work we consider only C∗-algebra with comultiplication, and not discuss other ingredients like
antipode, Haar weight, etc.; they can be constructed using methods presented in [12].
In the remaining part of the Introduction we recall basics of groupoid algebras, groupoids related to
decomposition of groups and results of [13] essential in the following. The short description of the content
of each section is given at the end of the Introduction.
1.1. Groupoid algebras. We will use groupoid algebras, so now we recall basic facts and establish the
relevant notation. We refer to [12, 11] for a detailed exposition and to [13] for basics of formalism.
All manifolds are smooth, Hausdorff, second countable and submanifolds are embedded. For a manifold
M by Ω1/2
(M ) we denote the vector space of smooth, compactly supported, complex half densities on M ;
the associated norm. Clearly, if we choose some ψ0 -- non vanishing, real half density on M , there is the
equality Ω1/2
(M ) = {f ψ0 , f ∈ D(M )}, where D(M ) stands for smooth, complex and compactly supported
functions on M .
c
Let Γ ⇒ E be a differential groupoid. By eR (eL) we denote the source (target) projection and called it
R denote bundles of complex half
right (left) projection; a groupoid inverse is denoted by s. Let Ω1/2
L , Ω1/2
c
it is equipped with the scalar product (ψ1 ψ2) := ZM
ψ1ψ2 and L2(M ) is the completion of Ω1/2
c
(M ) in
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
3
L ⊗Ω1/2
densities along left and right fibers. A groupoid *-algebra A(Γ) is a vector space of compactly supported,
smooth sections of Ω1/2
R together with a convolution and ∗-operation. To write explicit formulae let us
choose λ0 - a real, non vanishing, left invariant half density along left fibers (in fact that means we choose a
Haar system on Γ, however nothing depends on this choice, details are given in [11].) Let ρ0 := s(λ0) be the
corresponding right invariant half density and ω0 := λ0 ⊗ ρ0. Then any ω ∈ A(Γ) can be written as ω = f ω0
for a unique function f ∈ D(Γ). With such a choice we write (f1ω0) (f2ω0) =: (f1 ∗ f2)ω0, (f ω0)∗ =: (f∗)ω0
and:
(1)
(f1 ∗ f2)(γ) :=ZFl(γ)
λ2
0(γ′)f1(γ′)f2(s(γ′)γ) =ZFr(γ)
0(γ′)f1(γs(γ′)f2(γ′) , f∗(γ) := f (s(γ))
ρ2
Fl(γ) and Fr(γ) are left and right fibers passing through γ e.g. Fl(γ) := e−1
The choice of ω0 defines a norm that makes A(Γ) a normed ∗-algebra:
L (eL(γ)).
f ω00 =: f0 = max(sup
e∈EZe−1
L (e)
λ2
0(γ)f (γ), sup
e∈EZe−1
R (e)
0(γ)f (γ))
ρ2
There is a faithful representation πid of A(Γ) on L2(Γ) described as follows: choose ν0 - a real, non
vanishing half density on E; since eR is a surjective submersion one can define ψ0 := ρ0 ⊗ ν0 - this is a real,
non vanishing, half density on Γ. For ψ = f2ψ0, f2 ∈ D(Γ) the representation is given by πid(f1ω0)(f2ψ0) =:
(πid(f1)f2)ψ0 and πid(f1)f2 = f1∗f2 is as in (1). The estimate πid(ω) ≤ ω0 makes possible the definition:
The reduced C∗-algebra of a groupoid -- C∗r (Γ) is the completion of A(Γ) in the norm ω := πid(ω). We
will also use the following fact which is a direct consequence of the definition of the norm f0:
Lemma 1.1. Let U ⊂ Γ be an open set with compact closure. There exists M such that f0 ≤ M supf (γ)
for any f ∈ D(Γ) with support in U . If fn ∈ D(Γ) have supports in a fixed compact set and fn converges to
f ∈ D(Γ) uniformly then fnω0 converges to f ω0 in C∗r (Γ).
We will use Zakrzewski category of groupoids [20, 21]. Morphisms are not mappings (functors) but relations
⊲ Γ′ of differential groupoids defines a mapping
satisfying certain natural properties. A morphism h : Γ
h : A(Γ) → L(A(Γ′)) (linear mappings of A(Γ′)), this mapping commutes with (right) multiplication in
A(Γ′) i.e.
h(ω)(ω′)ω′′ = h(ω)(ω′ω′′) , ω ∈ A(Γ) , ω′, ω′′ ∈ A(Γ′)
and we use notation h(ω)ω′ (see formula (107) in the Appendix for an example of such a mapping); there
is also a representation πh of A(Γ) on L2(Γ′); these objects satisfy some obvious compatibility conditions
with respect to multiplication and ∗-operation (see [11] for details).
A bisection B of Γ ⇒ E is a submanifold such that eLB, eRB : B → E are diffeomorphisms. Bisections
⊲ Γ′
act on C∗r (Γ) as unitary multipliers and are transported by morphisms: if B ⊂ Γ is a bisection and h : Γ
is a morphism then the set h(B) is a bisection of Γ′.
1.2. Group decompositions, related groupoids and quantum groups. Now we briefly recall some
facts about double groups. Let G be a group and A, B ⊂ G subgroups such that A∩ B = {e}. Every element
g in the set Γ := AB ∩ BA can be written uniquely as
g = aL(g)bR(g) = bL(g)aR(g) , aL(g), aR(g) ∈ A , bL(g), bR(g) ∈ B.
4
PIOTR STACHURA
These decompostions define surjections: aL, aR : Γ → A and bL, bR : Γ → B (in fact aL, bR are defined on
AB and bL, aR on BA, we will denote these extensions by the same symbols). The formulae:
E := A ,
s(g) := bL(g)−1aL(g) = aR(g)bR(g)−1 ,
Gr(m) := {(b1ab2; b1a, ab2) : b1a, ab2 ∈ Γ}
define the structure of the groupoid ΓA : Γ ⇒ A; the analogous formulae define the groupoid ΓB : Γ ⇒ B.
On the other hand for a subgroup B ⊂ G there is a (right) transformation groupoid (B \ G) ⋊ B. The
following lemma [13] explains relation between these groupoids.
Lemma 1.2. The map:
ΓA ∋ g 7→ ([aL(g)], bR(g)) ∈ (B\G) ⋊ B
is an isomorphism of the groupoid ΓA with the restriction of a (right) transformation groupoid (B\G) ⋊ B
to the set {[a] : a ∈ A} ⊂ B\G.
If AB = G (i.e. Γ = G) the triple (G; A, B) is called a double group and in this situation we will
denote groupoids ΓA, ΓB by GA, GB. It turns out that the transposition of multiplication relation mB i.e.
δ0 := mT
⊲ GA × GA is a coassociative morphism of groupoids. Applying the lemma 1.2 to the
groupoid GA we can identify it with the transformation groupoid (B\G) ⋊ B. So GA = A ⋊ B is a right
transformation groupoid for the action (a, b) 7→ aR(ab) i.e the structure is given by:
B : GA
E := {(a, e) : a ∈ A} , s(a, b) := (aR(ab), b−1),
m := {(a1, b1b2; a1, b1, aR(a1b1), b2) : a1 ∈ A , b1, b2 ∈ B}
⊲ Γ with its graph, i.e. subset of Γ× Γ× Γ. We will
In the formula above, we identified a relation m : Γ× Γ
use such notation throughout the paper. If G is a Lie group, A, B are closed subgroups, A∩B = {e} , AB = G
then (G; A, B) is called a double Lie group, abbreviated in the following as DLG. It turns out that the
mapping bδ0, defined by the morphism δ0 (compare (107) in Appendix), extends to the coassociative morphism
∆ of C∗r (GA) and C∗r (GA × GA) = C∗r (GA) ⊗ C∗r (GA) which satisfies density conditions:
cls{∆(a)(I ⊗ b) : a, b ∈ C∗r (GA)} = cls{∆(a)(b ⊗ I) : a, b ∈ C∗r (GA)} = C∗r (GA) ⊗ C∗r (GA),
where cls denotes the closed linear span. There are other objects that make the pair (C∗r (GA), ∆) a locally
compact quantum group; we refer to [12] for details.
1.3. Framework for κ-Poincar´e. The framework for κ-Poincare group we are going to use was presented
in [13]. Let us now recall basic facts established there. Let G be a group and A, B, C ⊂ G subgroups
satisfying conditions:
B ∩ C = {e} = A ∩ C , BC = G.
i.e. (G; B, C) is a double group. As described above, in this situation, there is the groupoid GB, and the
(coassociative) morphism δ0 : GB
⊲ GB × GB; explicitly the graph of δ0 is equal to:
(2)
δ0 = {(b1c, cb2; b1cb2) : b1, b2 ∈ B , c ∈ C}
Using the lemma 1.2 we see that this is a transformation groupoid (C\G) ⋊ C and the isomorphism is
(C\G) × C ∋ ([g], c) 7→ bR(g)c ∈ G
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
5
Let Γ := AC ∩ CA and consider on Γ the groupoid structure ΓA : Γ ⇒ A described above, together with a
relation (the transposition of the multiplication in ΓC : Γ ⇒ C):
mT
C := {(a1c1, c1a2; a1c1a2) : a1c1, c1a2 ∈ Γ} ⊂ Γ × Γ × Γ.
The corresponding projections will be denoted by cL, cR and aR, aL. Again, by the lemma 1.2 we identify the
groupoid ΓA with the restriction of (C\G) ⋊ C and then with the restriction of GB to the set B′ := B ∩ CA,
i.e. with b−1
R (B′). This restriction will be denoted by ΓB′ (instead of more adequate but rather
inconvenient GBB′). This isomorphism and its inverse are given by:
(3)
L (B′) ∩ b−1
ΓA ∋ac 7→ bR(a)c ∈ ΓB′
, ΓB′ ∋ bc 7→ aR(b)c ∈ ΓA
The image of mT
C inside ΓB′ × ΓB′ × ΓB′ is equal to:
{(bR(a1)c1, bR(a2)c2; bR(a1a2)c2) : a1c1 = c1a1, c1a2 = a2c2}
The following object plays the major role in what follows:
(4)
T := {(g, b) : cR(g)b ∈ A} = {(b1cL(b2)−1, b2) : b1 ∈ B, b2 ∈ B′} ⊂ GB × GB.
Using the definition (2) of δ0 one easily computes images of T by relations id × δ0 and δ0 × id:
(id × δ0)T = {(g1, b2, b3) : b2b3 ∈ B′ , cR(g1) = cL(b2b3)−1}
(δ0 × id)T = {(g1, g2, b3) : cR(g1) = cL(g2) , b3 ∈ B′ , cR(g2) = cL(b3)−1}
Let us also denote T12 := T × B ⊂ GB × GB × GB and T23 := B × T ⊂ GB × GB × GB.
Main properties of T are listed in the following lemma (proven in [13]):
Proposition 1.3.
(1) T is a section of left and right projections (in GB × GB) over the set B × B′ and
(2) (id × δ0)T is a section of left and right projections (in GB × GB × GB) over the set B × δ0(B′) =
a bisection of GB × ΓB′;
{(b1, b2, b3) : b2b3 ∈ B′};
(3) (δ0 × id)T is a section of left and right projections over the set B × B × B′;
(4) T23(id × δ0)T = T12(δ0 × id)T (equality of sets in GB × GB × GB), moreover this set is a section of
the right projection over B × (δ0(B′) ∩ (B × B′)) and the left projection over B × B′ × B′.
Due to this proposition the left multiplication by T , which we denote by the same symbol, is a bijection of
GB × b−1
(5)
⊲ GB × GB be a relation defined by:
L (B′). Let AdT : GB × GB
(g1, g2; g3, g4) ∈ AdT ⇐⇒ ∃t1, t2 ∈ T : (g1, g2) = t1(g3, g4)(sB × sB)(t2).
and let us define the relation δ := AdT · δ0 : GB
⊲ GB × GB
(6)
δ = {(bR(b3b−1
2 cL(b2))cL(b2)−1cL(b2c2)cL(bR(b2c2)), b2c2; b3c2) :
b3 ∈ B, c2 ∈ C, b2, bR(b2c2) ∈ B′}.
The relation between δ and mT
C is explained in the lemma: [13]
Lemma 1.4. δ is an extension of mT
C i.e. mT
C ⊂ δ
Addition of some differential conditions to this situation makes possible to use T to twist the comultiplication
on C∗r (GB):
6
PIOTR STACHURA
Assumptions 1.5.
(1) G is a Lie group and A, B, C are closed Lie subgroups such that
B ∩ C = {e} = A ∩ C , BC = G (i.e. (G; B, C) is a DLG).
(2) The set Γ := CA ∩ AC is open and dense in G.
(3) Let U := b−1
L (B′) and A(U ) be the linear space of elements from A(GB) supported in U . We assume
that A(U ) is dense in C∗r (GB).
(4) For a compact set KC ⊂ C, open V ⊂ B and (b1, b2) ∈ B × B′ let us define a set Z(b1, b2, KC ; V ) :=
KC ∩ {c ∈ C : bR(b1c)b2 ∈ V } and a function:
B × B′ ∋ (b1, b2) 7→ µ(b1, b2, KC ; V ) :=ZZ(b1,b2,KC ;V )
dlc.
For compact sets K1 ⊂ B and K2 ⊂ B′ let µ(K1, K2, KC; V ) := sup{µ(b1, b2, KC; V ) : b1 ∈ K1 , b2 ∈
K2} We assume that
Remark 1.6.
∀ ǫ > 0∃ V − a neighborhood of B \ B′ in B : µ(K1, K2, KC; V ) ≤ ǫ
• It follows from the first and the second assumptions that B′ is open and dense in B.
• The second assumption can be replaced by the following two conditions:
a) g = a ⊕ c, where g, a, c are lie algebras of G, A, C, respectively. (then AC and CA are open);
b) AC ∩ CA is dense in G.
• These assuumptions are not very pleasent and probably, at least some of them, redundant; but they are
sufficient to get results in case of quantum "ax+b" group and κ-Poincar´e. This framework, however,
doesn't work for dual to κ-Poncar´e and at the moment it is unclear if and how that example may be
handled (with this approach).
The following proposition was proven in [13]:
Proposition 1.7. Assume the conditions listed in (1.5) are satisfied. Then
a) C∗r (ΓB′) = C∗r (GB) -- for this equality, it is sufficient to satisfy (1),(2),(3) from 1.5;
satisfies:
b) The mapping T : A(GB × U ) → A(GB × U ) extends to the unitary bT ∈ M (C∗r (GB)⊗ C∗r (GB)) which
c) Because of b), the formula ∆(a) := bT ∆0(a)bT −1
(bT ⊗ I)(∆0 ⊗ id)bT = (I ⊗bT )(id ⊗ ∆0)bT
defines a coassociative morphism. For this morphism
("cls" stands for "closed linear span"):
(7)
cls{∆(a)(I ⊗ c) , a, c ∈ C∗r (GB)} = cls{∆(a)(c ⊗ I) , a, c ∈ C∗r (GB)} = C∗r (GB) ⊗ C∗r (GB).
The second section describes the situation for κ-Poincar´e,
i.e. we define groups G, A, B, C, compute
explicit formulae for decompositions and describe structure of the groupoid GB. In the third one, we verify
Assumptions 1.5 and, by the Prop. 1.7 get the C∗-algebra and comultiplication for κ-Poincar´e Group. The
fourth section describes generators of this C∗-algebra, computes commutation relations and comultiplication
on generators; also, the twist is described in more details.
In the last but one section we compare our
formulae to the ones in [22] and in the last one we discuss "quantum κ-Minkowski Space". Finally there is
an Appendix with some formulae needed here and proven elsewhere.
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
7
2. Decompositions defining κ-Poincar´e Group
Let (V, η) be n + 2, n ≥ 1 dimensional vector Minkowski space (signature is (+,−, . . . ,−)). Let us
choose an orthonormal basis (e0, e1, . . . , en+1) and identify the (special) orthogonal group SO(η) with the
corresponding group of matricies SO(1, n + 1).
Let G := SO0(1, n+1) ⊂ SO(1, n+1) be the connected component of identity and g := span{Mαβ , α, β =
0, . . . , n + 1} be its Lie algebra (see Appendix for notation). Consider three closed subgroups A, B, C ⊂ G:
The group A is a non-connected extension of SO0(1, n) inside SO0(1, n + 1) defined by:
(8)
A :=( u 0
0 1 ! In
0 h ! : u ∈ SO0(1, n), h := d 0
0 d ! , d = ±1) .
0
In fact, it is not hard to see that A is the normalizer of SO0(1, n) (embedded into upper left corner) inside
SO0(1, n + 1). The Lie algebra of A is a := span{M0m , m = 1, . . . , n}.
We parameterize SO0(1, n) by:
{z ∈ Rn : z < 1} × SO(n) ∋ (z, U ) 7→ 1+z2
2
1−z2
1−z2 z
2
1−z2 ztU
1−z2 zzt)U ! ∈ SO0(1, n)
(I + 2
Remark 2.1. This is the standard boost × rotation parametrization of the Lorentz Group; the parameter z
is related to a velocity v by:
v
2z
(9)
z =
, v =
1 + z2
In this way we obtain the parametrization of A:
1 +p1 − v2
{z ∈ Rn : z < 1} × SO(n) × {−1, 1} ∋ (z, U, d) 7→
where D = In−1 0
The group B is SO(n + 1) embedded into G by: SO(n + 1) ∋ g 7→ 1 0
1−z2 ztU D
0
1−z2 zzt)U D 0
d
d !; we will also denote by (z, U, d) the corresponding element of G.
1+z2
1−z2
2
1−z2 z
(I + 2
0
0
0
2
,
0 g ! ∈ G; its Lie algebra is
b := {Mkl , k, l = 1, . . . , n + 1}. Elements of SO(n + 1) will be written as:
(Λ, u, w, α) := Λ u
wt α ! , Λ ∈ Mn(R), u, w ∈ Rn , α ∈ [−1, 1],
(10)
and Λ, u, w, α satisfy:
(11)
ΛΛt + uut = I , Λw + αu = 0 , ΛtΛ + wwt = I , Λtu + αw = 0 , u2 + α2 = w2 + α2 = 1;
these equations imply that α = det(Λ). Again we will denote by (Λ, u, w, α) the corresponding element of
G.
The group C is:
(12)
C :=
s2+1+y2
2s
−y
s2−1−y2
2s
− 1
s yt
I
1
s yt
s2−1+y2
2s
−y
s2+1−y2
2s
s ∈ R+, y ∈ Rn
⊂ G ;
8
PIOTR STACHURA
it is isomorphic to the semidirect product of R+ and Rn {(s, y) ∈ R+×Rn} with multiplication (s1, y1)(s2, y2) :=
(s1s2, s2y1 + y2). As before we will use (s, y) to denote the corresponding element of G. The Lie algebra of
C is c := span{Mβ0 − Mβ(n+1) , β = 0, . . . , n} = span{Mk0 − Mk(n+1) , k = 1, . . . , n + 1}. The coordinates
(s, y) are related to basis in c as:
(13)
(s, y) = exp(−(log s)M0(n+1)) exp(M(y)) ,
M(y) :=
nXk=1
yk(Mk(n+1) − Mk0)
Remark 2.2. More geometric description of data defining groups A, B, C was given in [14]; essentially we
have to choose two orthogonal vectors in n + 2 dimensional Minkowski(vector) space, one spacelike and one
timelike.
The Iwasawa decomposition for G is G = BC = CB.
In the following we will need explicit relation
between two forms of this decomposition i.e. solutions of the equation
(14)
(Λ, u, w, α)(s, y) = (s, y)(Λ, u, w, α)
Lemma 2.3. (a) Let (s, y) ∈ C and (Λ, u, w, α) ∈ B. The equation (14) has the (unique) solution (s, y) ∈ C
and (Λ, u, w, α) ∈ B given by formulae:
(15)
y = Λy −
u
s2 − 1 − y2
y(cid:19)
2s
1 − α
s
1
M s(cid:18)w −
yt)(cid:21)
1 − α
s
w =
(w −
1 − α
s
y)(wt −
1
s = M s = −wty +
M s(cid:18)(1 −
wty
u =
s
Λ =(cid:20)Λ −
α = 1 −
s2 − 1 − y2
2s
s
1
1
+ α
)u −
2s
1 − α
s2 + 1 + y2
Λy(cid:19)
uwt(cid:21)(cid:20)I −
2(1 − α) (cid:18)1 − α
s (cid:19)2
s2 + 1(cid:19) −
2(cid:18) 1
s2 + y2
α − 1
1 − α
M s2 , where
=
1
1
M (1 − α)
M :=
1 − α
s
y2! =
s2 − 1(cid:19) −
+ w −
α
2(cid:18) 1
s2 + y2
wty
s
(b) Let (s, y) ∈ C and (Λ, u, w, α) ∈ B. The equation (14) has the (unique) solution (s, y) ∈ C and
(Λ, u, w, α) ∈ B given by formulae:
s =
u =
s
M
s
M
=
2s(1 − α)
s2(1 − α)2 + u + (1 − α)y2
(u + (1 − α)y)
(16)
y =
w =
1
s2 + y2 − 1
M (cid:18)Λt y −
w(cid:19)
M (cid:16)(1 − α)Λt y + (1 + ut y) w(cid:17)
s
2
Λ = Λ −
=(cid:20)I −
α = 1 −
1
M (cid:20)(1 + ut y)y wt −
M (1 − α)
s2(1 − α)
, where
1
M
s2 + y2 − 1
2
(u + (1 − α)y)(u + (1 − α)y)t(cid:21)(cid:20)Λ +
u wt + (u + (1 − α)y)yt Λ(cid:21) =
u wt(cid:21)
1 − α
1
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
9
M :=
=
1
2(1 − α)(cid:0)s2(1 − α)2 + (1 − α)y + u2(cid:1) =
s2 + y2 + 1
s2 + y2 − 1
+ ut y
2
− α
2
Proof: Direct computation using (10) and (12).
By direct computations one also verifies that (14) implies equality
s(α − 1) =
α − 1
s
Clearly, it follows that α = 1 ⇐⇒ α = 1 and then:
u = w = u = w = 0 , Λ = Λ , s = s , y = Λy
For α 6= 1 (or, equivalently, α 6= 1) we have:
(17)
1
Λ −
α − 1
uwt = Λ −
u wt
1
α − 1
Moreover, for two given elements of B: (Λ, u, w, α) and (Λ, u, w, α) with α 6= 1, α 6= 1 that satisfy (17),
there exist (not unique) (s, y) ∈ C and (s, y) ∈ C such that equation (14) is fulfilled; they are given by:
ss =
,
y =
,
y =
u
u
s(1 − α) −
1 − α
w
α − 1 −
sw
α − 1
α − 1
α − 1
The decomposition (14) defines the groupoid GB : G ⇒ B. We will write (Λ, u, w, α; s, y) for the product
(Λ, u, w, α)(s, y) and use (Λ, u, w, α; s, y) to denote elements of GB.
The multiplication relation is given by:
mB :=n(cid:16)Λ, u, w, α; ss1, s1y + y1; ; Λ, u, w, α; s, y ; ; Λ, u, w, α; s1, y1(cid:17) :
(Λ, u, w, α; s, y) = (s, y)(Λ, u, w, α)o ⊂ GB × GB × GB,
We will use a detailed structure of this groupoid to verify (3) of Assumptions 1.5; the structure is summarized
in the following lemma.
Lemma 2.4.
(1) The isotropy group of (Λ, 0, 0, 1) ∈ B (i.e. of elements of SO(n) embedded in SO(n+1)
in the upper left corner) is equal to {(Λ, 0, 0, 1; s, y) : (s, y) ∈ C)} ≃ C and for (Λ, u, w, α) ∈ B , α 6= 1
is one dimensional {(Λ, u, w, α; s, s−1
1−α w) : s ∈ R+} ≃ R+
(2) GB is a disjoint union of an open groupoid Γ0 over SO(n + 1) \ SO(n) and a group bundle Γ1 :=
{(Λ, 0, 0, 1; s, y) : Λ ∈ SO(n), (s, y) ∈ C} ≃ SO(n) × C;
(3) The open groupoid Γ0 is a product Γ0 = O(n)− × Γ00 of a manifold-groupoid O(n)− and a transi-
tive groupoid Γ00, where O(n)− stands for the component of the orthogonal group with a negative
determinant;
(4) The transitive groupoid Γ00 is a product Γ00 = R+ × (Rn × Rn) of a group and a pair groupoid.
Proof: 1) and 2) are direct consequences of formulae (15) and (16);
3) For v ∈ Rn let R(v) be the orthogonal reflection in Rn+1 along the direction of (v,−1)t i.e.
R(v) = I − 2vvt
1+v2
2vt
1+v2
2v
1+v2 !
1+v2
v2−1
10
PIOTR STACHURA
Let us consider the map:
Φ : O(n)− × Rn ∋ (K, v) 7→ K 0
1 ! R(v) ∈ SO(n + 1)
0
Clearly Φ(K, v) ∈ SO(n + 1) \ SO(n). Moreover, using (11), it is easy to see that for α 6= 1 the matrix
Λ − 1
α−1 uwt is orthogonal and
Λ u
wt α ! I
α−1 uwt) = −1 and we can define
Ψ : SO(n + 1) \ SO(n) ∋ Λ u
wt
1−α
0
therefore det(Λ − 1
0
1 ! I −w
1 ! = Λ − 1
α−1 uwt
wt
1−α
0
−1 ! ,
wt α ! 7→(cid:18)Λ −
uwt,
1
α − 1
w
1 − α(cid:19) ∈ O(n)− × Rn.
By direct computation one verifies that Ψ = Φ−1. Clearly, both mappings are smooth, so both are diffeo-
morphisms. By (15) we obtain:
Φ(K, v) (s, y) = (s, y) Φ(K, sv − y)
1 + v2 (vty +
, y = K(cid:18)y −
2v
1 + sv − y2
s(1 + v2)
s =
s2 − y2 − 1
2s
)(cid:19)
These formulae show that Γ00 := Rn × C is a (right) transformation groupoid with the action: Rn × C ∋
(v; s, y) 7→ sv − y ∈ Rn. It is clear that this action is transitive.
4) We will use the following general but simple:
Lemma 2.5. Let Γ ⇒ E be a transitive groupoid. For e0 ∈ E let p : E ∋ e 7→ p(e) ∈ e−1
L (e0) be a section
of the right projection, such that p(e0) = e0. Let G be an isotropy group of e0. The map G × E × E ∋
(g, e1, e2) 7→ s(p(e1))gp(e2) ∈ Γ is an isomorphism of groupoids. (G × E × E is a product of a group and
pair groupoid).
The application of the lemma (choose v0 = 0) gives us a groupoid isomorphism:
R+ × Rn × Rn ∋ (s; x1, x2) 7→ (x1; s, sx1 − x2) ∈ Rn × C,
which in our situation clearly is a diffeomorphism.
To find the set B′ i.e. B ∩ CA we have to solve the equation:
(18)
(z, U, d) = (s, y)(Λ, u, w, α) , (z, U, d) ∈ A , (s, y) ∈ C , (Λ, u, w, α) ∈ B
Using (9, 10) and (12), by direct computation, one verifies:
Lemma 2.6. (a) For (z, U, d) ∈ A solutions (s, y) ∈ C , (Λ, u, w, α) ∈ B of (18) are given by:
s =
1 + z2
1 − z2
,
2zzt
(19)
Λ = (I −
1 + z2 )U D ,
y = −2z
1 − z2
u = −2dz
1 + z2
,
,
w =
2DU tz
1 + z2
,
α =
d(1 − z2)
1 + z2
.
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
11
(b) For (Λ, u, w, α) ∈ B with α 6= 0, solutions (s, y) ∈ C , (z, U, d) ∈ A of (18) are given by:
,
s =
1
α
sgn(α)u
z = −
1 + α
y =
u
α
,
U = (Λ −
sgn(α)
1 + α
,
uwt)D ,
d = sgn(α).
(20)
This way we obtain B′ and projections cL and aR defined by decomposition CA ∋ g = cL(g)aR(g) (restricted
to B′).
(21)
(22)
(23)
B′ = {(Λ, u, w, α) ∈ B : α 6= 0}
cL(Λ, u, w, α) = (α,−sgn(α)u)
aR(Λ, u, w, α) =(cid:18)−
sgn(α)u
1 + α
, (Λ −
sgn(α)
1 + α
uwt)D, sgn(α)(cid:19)
Finally, using the definition (4) and the formula above we obtain the twist T :
T = {(Λ1, u1, w1, α1;
,
1
α2
u2
α2
; ; Λ2, u2, w2, α2; 1, 0) ∈ GB × GB : α2 6= 0}
3. The C∗-algebra
In this section we will use Prop. 1.7 to get the C∗-algebra and the comultiplication. To this end, conditions
listed in assumptions 1.5 have to be verified. The first one is clear and the second was proven in [10], it
L (B′) and A(U ) be the linear space of elements from A(GB) supported in
remains to check (3) and (4).
Condition 1.5 (3) Let U := b−1
U . We will show that A(U ) is dense in C∗r (GB).
Let B′ be as in (21) and define B′′ := {(Λ, u, w, α) ∈ B : α 6= 1}, then B = B′ ∪ B′′ i.e. {B′, B′′} is an
open cover of B and b−1
L (B′′) = Γ0. Let χ′, χ′′ be a partition of unity subordinated to the cover {B′, B′′},
χ′(γ) := χ′(bL(γ)) and χ′′(γ) := χ′′(bL(γ)). For ω ∈ A(GB) : ω = χ′ω + χ′′ω and χ′ω ∈ A(U ) , χ′′ω ∈ A(Γ0);
so to prove that A(U ) is dense w C∗r (GB) is sufficient to show that any element in A(Γ0) can be approximated
by elements from A(U ), in particular from A(U∩Γ0). In this way we can transfer the whole problem to Γ0 and
use its structure described in lemma 2.4; in this presentation U∩Γ0 = {(K, t, x1, x2) ∈ O(n)−×R+×Rn×Rn :
x1 6= 1}. Let dK denotes a measure on O(n)− defined by (the square of) some smooth, positive, non
vanishing half-density; dx1 be the Lebesgue measure on Rn and ds
s be the Haar measure on R+. Then
L2(Γ0) can be identified with L2(cid:0)O(n)− × R+ × Rn × Rn; dK ds
(π(f )Ψ)(K, t, x1, x2) := (f ∗ Ψ)(K, t, x1, x2) :=Z ds
s dx1dx2(cid:1) and A(Γ0) acts on L2(Γ0) by:
dy f (K, s, x1, y)Ψ(K, t/s, y, x2);
(24)
s
note that it is sufficient to consider Ψ ∈ D(Γ0). We will prove the following estimate:
Lemma 3.1. Let f be a continuous function supported in a product of compact sets L × M × N × R ⊂
O(n)− × R+ × Rn × Rn. Then the norm of the operator π(f ) given by (24) satisfies:
(25)
π(f ) ≤ supf ν(M )pµ(N )µ(R),
where ν denotes the Haar measure on R+ : ν(M ) :=RM
ds
s and µ the Lebesgue measure on Rn.
12
PIOTR STACHURA
For ǫ > 0 let Oǫ be an open neighborhood of the unit sphere in Rn with µ(Oǫ) ≤ ǫ; let χǫ be a smooth
function supported in Oǫ such that 0 ≤ χǫ ≤ 1, χǫ = 1 on the unit sphere and the function χǫ on Γ0 be
defined by χǫ(K, t, x1, x2) := χǫ(x1).
Let f ∈ D(Γ0) be supported in L× M × N × R. We have f = (f − χǫf ) + χǫf and (f − χǫf ) ∈ D(U ∩ Γ0).
By the lemma 3.1
π(χǫf ) ≤ supfν(M )pµ(R)pµ(Oǫ ∩ N ) ≤ supfν(M )pµ(R)√ǫ
So really f can be approximated by elements from D(U ∩ Γ0). It remains to prove lemma 3.1.
Proof of lemma 3.1: We just apply the Schwartz inequality several times. Let Ψ be smooth and compactly
supported; by (24):
(Ψf ∗ Ψ) =Z dK
(Ψf ∗ Ψ) ≤Z dK
dt
t
dt
t
dx1dx2 Ψ(K, t, x1, x2)Z ds
dx1dx2 Ψ(K, t, x1, x2)Z ds
s
s
dy f (K, s, x1, y)Ψ(K, t/s, y, x2)
dy f(K, s, x1, y)Ψ(K, t/s, y, x2)
We write the integral as iterated integral: R dKdx2R dx1R dy R dt
For fixed (K, y, x1, x2) let us define functions: Ψ1 : R+ ∋ t 7→ Ψ1(t) := Ψ(K, t, x1, x2)
t R ds
s
f1 : R+ ∋ t 7→ f1(t) := f (K, t, x1, y) ; Ψ2 : R+ ∋ t 7→ Ψ2(t) := Ψ(K, t, y, x2)
and
(26)
and
(27)
With these definitions we have the estimate:
t Ψ(K, t, x1, x2)Z ds
Z dt
s f(K, s, x1, y)Ψ(K, t/s, y, x2) = (Ψ1f1 ∗ Ψ2) ≤ Ψ12Ψ22f11,
where the scalar product is in L2(R+, ds
and f1 is L1 norm; these norms are continuous functions of remaining variables.
Let us now define (continuous, compactly supported) functions Ψ1, Ψ2, f1 : O(n)− × Rn × Rn → R:
s ) and · 2 norms refer to this space, ∗ is the convolution in R+
Ψ1(K, x1, x2) := Ψ12 =(cid:20)Z dt
t Ψ(K, t, x1, x2)2(cid:21)1/2
, Ψ2(K, x2, y) := Ψ22
The right hand side of (26) is estimated by:
f1(K, x1, y) := f11 =Z t
dt
t f (K, t, x1, y).
Z dKdx2Z dx1 Ψ1(K, x1, x2)Z dy f1(K, x1, y) Ψ2(K, x2, y)
By the Schwartz inequality for y-integration we get an estimate for the integral above by:
Z dKdx2Z dx1 Ψ1(K, x1, x2)(cid:20)Z dy ( f1(K, x1, y))2(cid:21)1/2 (cid:20)Z dy ( Ψ2(K, x2, y))2(cid:21)1/2
Again let us denote f2(K, x1) :=hR dy ( f1(K, x1, y))2i1/2
The integral (27) reads
and Ψ3(K, x2) :=hR dy ( Ψ2(K, x2, y))2i1/2
Schwartz inequality again and we can estimate it by:
Z dKdx2 Ψ3(K, x2)Z dx1 f2(K, x1) Ψ1(K, x1, x2).
Z dKdx2 Ψ3(K, x2) Ψ4(K, x2) f3(K),
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
13
where Ψ4(K, x2) :=hR dx1 ( Ψ1(K, x1, x2))2i1/2
And, finally, this integral we estimate by
and f3(K) :=hR dx1 ( f2(K, x1))2i1/2
sup f3(cid:20)Z dKdx2( Ψ4(K, x2))2(cid:21)1/2 (cid:20)Z dKdx2( Ψ3(K, x2))2(cid:21)1/2
.
(again the Schwartz inequality was used).
But
and, in a similar way,
Z dKdx2( Ψ4(K, x2))2 =Z dKdx2Z dx1 ( Ψ1(K, x1, x2))2 =
=Z dKdx2Z dx1 Z dt
t Ψ2(K, t, x1, x2) = kΨk2
Z dKdx2( Ψ3(K, x2))2 = kΨk2
Thus we get an inequality (Ψf ∗ Ψ) ≤ sup f3kΨk2, therefore π(f ) ≤ sup f3.
s f (K, s, x1, y)(cid:21)2
dy (supf)2ν2(M ) = (sup f)2ν2(M )µ(N )µ(R)
f3(K)2 =Z dx1( f2(K, x1))2 =Z dx1Z dy(cid:20)Z ds
≤ZN
dx1ZR
≤
This is the estimate (25) and the lemma is proven.
This way by Prop. 1.7 (a) we obtain the equality of C∗-algebras:
(28)
C∗r (ΓA) = C∗r (ΓB′) = C∗r (GB).
To get the twist and the comultiplication we have to verify:
Condition 1.5 (4) For M > 1 and 1 > δ > ǫ > 0 let us define compact sets KM ⊂ C, Kδ ⊂ B′ and an open
neighborhood Vǫ of B \ B′ in B:
KM := {(s, y) ∈ C :
1
M ≤ s ≤ M , y ≤ M} , Kδ := {(Λ, u, w, α) ∈ B : α ≥ δ} ,
Vǫ := {(Λ, u, w, α) ∈ B : α < ǫ}.
We will show that for any M > 1 any 0 < δ < 1 and any 0 < ǫ < δ there exists ǫ′ such that
µ(B, Kδ, KM ; Vǫ′) < ǫ (notation as in Assumptions 1.5.) Since any compact in B′ is contained in some
Kδ and any compact in C is contained in some KM , this is sufficient.
Let b = (Λ, u, w, α) and b1 = (Λ1, u1, w1, α1) ∈ Kδ; we want to find the set Z(b, b1, KM ; Vǫ) := KM ∩ {c ∈
C : bR(bc)b1 ∈ Vǫ}. Let c = (s, y) , bR(bc) =: (Λ, u, w, α) and bR(bc)b1 =: (Λ2, u2, w2, α2).
Then α2 = wtu1 + αα1 and using solutions of eq. (14) we get:
α =( r2−1
, α2 =( 2rtu1+α1(r2−1)
, w =( 2r
r2+1 α 6= 1
α = 1
1
1+r2 α 6= 1
α = 1
0
1+r2
α1
α 6= 1
α = 1
,
1−α − y.
where r := sw
Now we solve for (s, y) the inequality α2 < ǫ with the additional assumption 0 < ǫ < δ;
There is no solution for α = 1 and for α 6= 1 we have:
(cid:12)(cid:12)(cid:12)(cid:12)r +
or in terms of (s, y):
sgn(α1)
α1 − ǫ
sw
1 − α
+
sgn(α1)
α1 − ǫ
(cid:12)(cid:12)(cid:12)(cid:12)
<
<
2
2
u1(cid:12)(cid:12)(cid:12)(cid:12)
u1 − y(cid:12)(cid:12)(cid:12)(cid:12)
1 − ǫ2
(α1 − ǫ)2 and (cid:12)(cid:12)(cid:12)(cid:12)r +
(α1 − ǫ)2 and (cid:12)(cid:12)(cid:12)(cid:12)
1 − ǫ2
sw
1 − α
+
sgn(α1)
α1 + ǫ
u1 − y(cid:12)(cid:12)(cid:12)(cid:12)
2
>
1 − ǫ2
(α1 + ǫ)2
α1−ǫ u1 with a radius
√1−ǫ2
α1+ǫ .
For fixed s this is the intersection of the (larger) ball centered at y1 := sw
1−α + sgn(α1)
r1 :=
Because of the inequality
α1 + ǫ(cid:19)p1 − ǫ2 = r1 − r2,
√1−ǫ2
α1−ǫ with the exterior of the (smaller) ball centered at y2 := sw
y1 − y2 =(cid:18)
α1 − ǫ −
α1+ǫ u1 with a radius r2 :=
the smaller ball is contained in the larger one, and the volume of this intersection is equal to:
1−α + sgn(α1)
1
1
F (n)(rn
1 − rn
2 ) = F (n)
(1 − ǫ2)n/2
(α1 − ǫ)n
2nǫ
α1 + ǫ ≤
1
1
α1 + ǫ(cid:19)q1 − α2
1 <(cid:18)
α1 − ǫ −
(α1 − ǫ)n(cid:18)1 −(cid:18)1 −
α1 + ǫ(cid:19)n(cid:19) ≤ F (n)
δ(δ − ǫ)n ,
2nǫ
δ ≤ 2nF (n)
(1 − ǫ2)n/2
(δ − ǫ)n
≤ F (n)
2ǫ
1
ǫ
14
PIOTR STACHURA
after some manipulation we get for α1 > 0:
−ǫ(1 + r2) < 2rtu1 + α1(r2 − 1) < ǫ(1 + r2)
and for α1 < 0:
r +
r +
u1
α1 − ǫ2 <
u1
α1 − ǫ2 >
1 − ǫ2
(α1 − ǫ)2 and r +
1 − ǫ2
(α1 − ǫ)2 and r +
Both situations can be described uniformly as:
u1
α1 + ǫ2 >
1 − ǫ2
(α1 + ǫ)2
u1
α1 + ǫ2 <
1 − ǫ2
(α1 + ǫ)2 .
sgn(α1)
α1 + ǫ
2
u1(cid:12)(cid:12)(cid:12)(cid:12)
>
1 − ǫ2
(α1 + ǫ)2
where F (n)rn is a volume of n-dimensional ball. In this way we obtain
µ(b, b1, KM ; Vǫ) =ZZ(b,b1,KM ;Vǫ)
ds
s
dy ≤ ǫ
log M
δ
4nF (n)
(δ − ǫ)n
i.e.
µ(B, Kδ, KM ; Vǫ) ≤ ǫ
log M
δ
4nF (n)
(δ − ǫ)n
The right hand side goes to 0 as ǫ → 0, and the fourth condition of (1.5) is satisfied.
Now, by the Prop. 1.7 (b), (c), we get the comultiplication ∆ on C∗r (ΓB′ ) = C∗r (GB) satisfying the density
condition (7).
4. Generators and relations
In the previous section the C∗-algebra of the quantum κ-Poincar´e Group together with comultiplication
was defined. In this section we describe its generators, commutation relations among them and look closer
at the twist.
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
15
4.1. General formulae. Let u be a bisection of a differential groupoid Γ ⇒ E and f a smooth, bounded
function on E. They define bounded operators on L2(Γ), denoted by u and f , which are multipliers of C∗r (Γ):
u acts by a push-forward of half-densities and the action of f is defined by ( f ψ)(γ) := f (eL(γ))ψ(γ) , ψ ∈
Ω1/2
(Γ) [11]. These operators satisfy:
c
(29)
(30)
In particular, for a one-parameter group of bisections ut, we have
u f = bfu u , where fu(e) := f (eL(u−1e)) ,
but fdu−t = bft , where ft(e) := f (eL(u−te))
e ∈ E
For a DLG (G; B, C) and c0 ∈ C, by Bc0 we denote the bisection of GB defined as Bc0 := {bc0 : b ∈ B}. It
acts on GB by
(Bc0)(bc) = bR(bc−1
(31)
GB is a (right) transformation groupoid B ⋊ C for the action B × C ∋ (b, c) 7→ bR(bc) ∈ B. The C∗-algebra
C∗r (GB) is the reduced crossed product C0(B) ⋊r C (see e.g [13], prop. 5.2, where this identification is
described in details). If C is an amenable group, and this is our case (or more generally, this is the case
0 )bc = cR(c0b−1)bc
0 )c0c = c−1
L (bc−1
of C = AN coming from the Iwasawa decomposition G = K(AN )) reduced and universal crossed products
coincide ([16], Thm 7.13, p. 199). Moreover, since the universal C∗-algebra of a differential groupoid
C∗(GB) as defined in [11] is, for transformation groupoids, "something between" universal and reduced
crossed products, for amenable groups C we have C∗(GB) = C∗r (GB); thus any morphism of differential
groupoids h : GB
⊲ Γ defines a C∗-morphism of corresponding reduced and universal algebras.
Remark 4.1. Thus our C∗(GB) has SO(n) family of "classical points" i.e. characters given by 0-dimensional
orbits of GB described in lemma 2.4.
The canonical morphisms iC ∈ M or(C∗(C), C∗(GB)) and iB ∈ M or(C0(B), C∗(GB)) are given by (ex-
tensions of) the following actions on A(GB):
iC (c0)ω := (Bc0)ω ,
(iB(f )ω)(g) := f (bL(g))ω(g) ,
c0 ∈ C, g ∈ G, f ∈ C∞0 (B), ω ∈ A(GB)
If X1, . . . , Xm are generators of C∗(C), f1, . . . , fk are generators of C0(B) (in the sense of [18]) and C
is amenable then (by the universality of crossed product) C∗(GB) is generated by iC(X1), . . . , iC(Xm),
iB(f1), . . . , iB(fk).
In our situation, it is clear that C(B) is generated by matrix elements (Λ, u, w, α) in (10) and, by the
results of [18], C∗(C) is generated by any basis in c. Thus we have
Proposition 4.2. Let groups B and C be defined by (10, 12);
let (X0, . . . , Xn) be a basis in c and
(Λ, u, w, α) be matrix elements of B. Elements (iC(X0), . . . , iC (Xn), iB(Λ), iB(u), iB(w), iB (α)) are gen-
erators of C∗(GB).
For c ∈ c, let u c and X r
vector field on GB defined by
c be, respectively, the one-parameter group of bisections and the right invariant
(32)
u c(t) := B exp(t c)
, X r
c (bc) :=
u c(t)(bc) = (Adc(b)( c))bc,
where Adc(g) is defined in (101) (the last equality follows easily from (31)).
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
16
PIOTR STACHURA
Let Π(X r
c ) be the projection of X r
c onto B by bL i.e.
(33)
Π(X r
c )(b) :=
by the straightforward computation, we get:
bL(u c(t)(b))
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
(34)
Π(X r
c )(b) := −(PbAd(b) c)b
Proposition 4.3. Let (G; B, C) be a DLG and c ∈ c. Let u c, X r
(33).
(1) The one-parameter group bu c is strongly continuous and strictly continuous (as a group of multipliers
of C∗r (GB)). Let A c be its generator. The action of A c on Ω1/2
A c = ιLX r
affiliated to C∗r (GB).
(GB) and on A(GB) is given by
(GB) is an essential domain for A c and A c is
(the Lie derivative). The linear space Ω1/2
c ) be objects defined in (32) and
c and Π(X r
c
c
c
(2) For c, e ∈ c, generators A c, A e, as operators on Ω1/2
c
(GB) (or A(GB)), satisfy:
(35)
[A c, A e] = −ιA[ c, e]
(3) Let c ∈ c and f be a smooth function on B. As operators on Ω1/2
c
(GB) (or A(GB)), A c and f satisfy:
(36)
[A c, f ] = ι( Π(X r
c )f )b
Proof: Since bu c is a one-parameter group of multipliers in C∗r (GB) for its strict continuity it is sufficient
to check continuity at t = 0 of the mapping R ∋ t 7→ bu c(t)(ω) ∈ C∗r (GB) for ω ∈ A(GB). Let us choose
ω0 = λ0 ⊗ ρ0 as in (103, 104), then ω = f ω0 for f ∈ D(GB) and we can write bu c(t)(f ω0) =: (bu c(t)f )ω0,
where the function (bu c(t)f ) is given by (compare (108)):
(37)
(bu c(t)f )(bc) = f (u c(−t)bc) jC (exp(t c))−1/2
Since the mapping R × GB ∋ (t, g) 7→ u c(t)g ∈ GB is continuous and f ∈ D(GB), for δ > 0 and t < δ
supports of all functions (bu c(t)f ) are contained in a fixed compact set, so by the lemma 1.1, it is sufficient
to prove that (bu c(t)f ) converges, as t → 0, uniformly to f . But this is clear, since everything happens in a
fixed compact set and all functions and mappings appearing in (37) are smooth.
Recall that the domain of a generator A of a (strongly continuous) one-parameter group of unitaries Ut
on a Hilbert space is defined as set of those vectors ψ for which the limit lim
t→0
(−ι)
the value of this limit. Moreover if a dense linear subspace is contained in the domain of A and is invariant
exists and Aψ is
Utψ − ψ
t
for all Ut's then it is a core for A.
Let us choose ν0 -- real non vanishing half density on B and let Ψ0 = ρ0 ⊗ ν0. This is (real, non vanishing)
half density on GB and any ψ ∈ Ω1/2
(GB) can be written as ψ = f Ψ0 for f ∈ D(GB). The action of bu c(t)
can be written as bu c(t)(f Ψ0) =: (bu c(t)f )Ψ0 and (bu c(t)f ) is given by the formula (108). Using this formula
and formulae (32, 102) one verifies that:
c
(38)
lim
t→0
1
t
((bu c(t)f )(bc) − f (bc)) = −(cid:18)(X r
c f )(bc) +
1
2
T r(ad( c)c)f (bc)(cid:19) ,
where ad( c)( e) := [ c, e] , c, e ∈ c. Again, since in the formula above, we stay, for a given f , in a fixed
compact subset and everything is smooth, the limit is, in fact, uniform and therefore also in L2(GB). Thus
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
17
ψ
(GB) ⊂ Dom(A c) and
Ω1/2
c
(39)
Clearly Ω1/2
c
A c(f Ψ0) =: A c(f ) Ψ0 , A c(f ) := ι(cid:18)X r
T r(ad( c)c)f(cid:19)
(GB) is dense in L2(GB) and bu c(t) invariant, so it is a core for A c.
c , the limit lim
t→0
c f +
1
2
1
t
ψ.
Since bu c(t)(ψ) is a push-forward by the flow of X r
and, consequently, A cψ = ιLX r
fundamental vector field for this action is defined as X c(g) :=
Let us prove (35). The mapping C × GB ∋ (c0, g) 7→ (Bc0)g ∈ GB is a left action of C. For c ∈ c, the
u c(−t)(g) and the map c 7→ X c is a Lie
the
algebra homomorphism (see e.g [4]) Comparing with (32) we see that X r
formula (35) follows.
((bu c(t)ψ)(g) − ψ(g)) is equal to −LX r
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
c = (−1)X c. Since A c = ιLX r
d
c
c
c
The formula (36) is a direct consequence of (30) and (33).
4.2. Commutation relations for κ-Poincar´e. Let us consider the one-parameter group c(t) := exp(tM0(n+1));
in (s, y) coordinates c(t) = (s(t), y(t)) := (e−t, 0). Let S(t) = Bc(t) be the corresponding group of bisections
i.e.
S(t) := {(Λ, u, w, α; e−t, 0) , (Λ, u, w, α) ∈ B , t ∈ R}.
Define (Λ(t), u(t), w(t), α(t)) := bR(Λ, u, w, α; et, 0) and let S be the generator of dS(t). By (31,33) and (36)
we get:
By (15) we have
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
[ S, Q](Λ, u, w, α) = ι(cid:18) d
Q(t)(cid:19)b
Λ(t) = Λ −
u(t) =
sinh t
uwt
cosh t + α sinh t
u
cosh t + α sinh t
, Q = Λ, u, w, α
α(t) =
w(t) =
α cosh t + sinh t
cosh t + α sinh t
w
cosh t + α sinh t
,
and by differentiation we obtain commutation relations (as operators on A(GB) or Ω1/2
c
(GB)):
[ S, w] = −ια w
[ S, u] = −ιαu
[ S, α] = −ι(α2 − 1)
or with indices put explicitly:
(40)
(41)
[ S, Λ] = −ιduwt
[ S,cΛkl] = −ιcukcwl
[ S,cwk] = −ιαcwk
[ S,cuk] = −ιαcuk
Now, for y0 ∈ Rn, let c(t) := exp(tM(y0)), where, as in (13), M(y0) := Pn
i=1(y0)i(Mi(n+1) − Mi0)); or
in (s, y) coordinates: c(t) = (s(t), y(t)) = (1, ty0). Let Y0(t) be the corresponding one-parameter group of
bisections Bc(t); as before we put (Λ(t), u(t), w(t), α(t)) := bR(Λ, u, w, α; 1,−ty0). By (15) we get:
t2(1 − α)
M (t)
t2y02
2M (t)
t(1 + twty0)
Λ(t) = Λ +
Λy0wt −
uwt −
Λy0yt
0
0 −
M (t)
M (t)
uyt
t
[ S, α] = −ι(α2 − 1)
α(t) = 1 −
1 − α
M (t)
u(t) =
u + t(wty0u + (1 − α)Λy0)
M (t)
w(t) =
w + t(1 − α)y0
M (t)
,
where M (t) := y02
2
(1 − α)t2 + twty0 + 1
18
PIOTR STACHURA
Denoting by cY0 the generator of [Y0(t) we obtain commutation relations (again as operators on A(GB) or
(GB)):
Ω1/2
c
[cY0, Λ] = −ι(cid:0)uyt
0 + Λy0wt(cid:1)b
[cY0, w] = ι(cid:0)(1 − α)y0 − wty0w(cid:1)b
[cYm,cΛkl] = −ι(cid:16)cukδml + dΛkmcwl(cid:17)
[cYm,cwk] = ι ((1 − α)δmk −dwmcwk)
[cY0, α] = ι(cid:0)(1 − α)wty0(cid:1)b
[cY0, u] = ι ((1 − α)Λy0)b
[cYm, α] = ι(1 − α)dwm
[cYm,cuk] = ι(1 − α)dΛkm
(42)
(43)
(44)
or for y0 := em ∈ Rn with corresponding cYm and with indices of matrix elements:
For completeness let us write relations between S andcY0. Since S (cY0) is the generator of the group bu c
for c = M0(n+1) ( c = M(y0)) by the formula (35) and (110) we obtain:
[ S,cY0] = −icY0 ,
[cY1,cY0] = 0,
wherecY1 is defined in the same way ascY0 for a vector y1 ∈ Rn.
Our generators S and Yk are related, via the mapping c ∋ c 7→ A c used in the previous subsection, to the
following basis in c:
(45)
( S, Yk) ! ( c0, ck) := (M0(n+1), Mk(n+1) − Mk0) ,
k = 1, . . . , n
⊲ GB ×GB defined in (2) i.e. δ0 = {(b−1
4.3. Formulae for comultiplication. Now we consider again a general DLG (G; B, C) together with the
1 cL(b1bc), b1bc; bc) : b, b1 ∈ B, c ∈ C}. It defines the
relation δ0 : GB
mapping δ0 given by the formula (107), which extends to the coassociative ∆0 ∈ M or(C∗r (GB), C∗r (GB) ⊗
C∗r (GB)) [12]. For a smooth, bounded function f on B, let ∆B(f ) be the value of the comultiplication of
the group B on f i.e. ∆B(f )(b1, b2) := f (b1b2). Let f be the multiplier of C∗r (GB) defined by f ; the formula
(107) implies
(46)
∆0( f ) = \∆B(f )
One easily computes the image of a bisection Bc0 by δ0:
and its action on GB × GB:
δ0(Bc0) = {(b1cL(bc0), bc0) : b, b1 ∈ B}
(47)
δ0(Bc0)(b1c1, b2c2) = (c−1
L (b1cL(b2c−1
= (bR(b1cL(b2c−1
= (cR(cR(c0b−1
0 ))b1c1, c−1
L (b2c−1
0 ))c−1
L (b2c−1
0 )b2c2) =
0 )c1, bR(b2c−1
0 )c0c2) =
2 )b−1
1 )b1c1, cR(c0b−1
2 )b2c2)
For c ∈ c, let u c be the one-parameter group of bisections defined in (32) and define u c(t) := δ0(u c(t)); this
is a one-parameter group of bisections of GB × GB. Let X r
c be the corresponding right invariant vector field
on GB × GB (compare (32)). By (47):
X r
c (b1c1, b2c2) =
(c1(t)b1c1, c2(t)b2c2), where
c1(t) := cR(b1cR(b2 exp(t c)b−1
2 )b−1
1 ) , c2(t) = cR(b2 exp(t c)b−1
2 ),
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
19
and
(48)
X r
c (b1c1, b2c2) = (Adc(b1)Adc(b2)( c)b1c1, Adc(b2)( c)b2c2)
Let (Xα) be a basis in c , uα, Aα, X r
invariant vector fields respectively, and let uα := δ0(uα) with corresponding Aα and X r
reads (after identification T(x,y)(X × Y ) = TxX ⊕ TyY ) :
(49)
α be corresponding one-parameter groups, their generators and right-
α. The formula (48)
Adc
X r
βα(b2)X r
β(b1c1) + X r
α(b2c2),
α(b1c1, b2c2) =Xβ
where functions Adc
in Prop.4.3 the action of Aα on Ω1/2
Since Aα = ∆0(Aα) we obtain the following equality on Ω1/2
βα : B → R are matrix elements of AdcB (101), i.e. Adc(b)(Xα) =:Pβ Adc
βα(b)Xβ. As
(GB × GB) is given by (ι times) the Lie derivative with respect to X r
α.
∆0(Aα) = I ⊗ Aα +Xβ
(GB × GB)
Aβ ⊗ Adc
(50)
βα
c
c
c
c
c
(GB) ⊗ Ω1/2
(GB) and is symmetric there. But since elements of Ω1/2
Remark 4.4. Let us comment on the meaning of the equality above. The operator Aα = ∆0(Aα) is es-
sentially self-adjoint on Ω1/2
(GB × GB). The operator on the right hand side has immediate meaning on
Ω1/2
(GB × GB) can be approximated
by elements of Ω1/2
(GB × GB) (i.e. uniformly with all derivatives on
compact sets) and operators Aα are differential operators, the space Ω1/2
(GB × GB) is in the domain of the
(GB) ⊗ Ω1/2
closure of the right hand side (treated as operator on Ω1/2
(GB)). Therefore this closure is the
self-adjoint operator ∆0(Aα).
(GB) in topology of Ω1/2
(GB) ⊗ Ω1/2
c
c
c
c
c
c
c
(51)
4.4. Comultiplication for κ-Poincar´e. For a bisection Bc0 let us consider the map:
(b1c1, b2c2) 7→
δ(Bc0)(b1c1, b2c2), where the relation δ is given by (6). By the use of that formula one finds the domain of
this map -- {(b1c1, b2c2) : b2, bR(b2c−1
0 ) ∈ B′} and
=(cid:0)c−1
δ(Bc0)(b1c1, b2c2) =(cid:0)bR[b1c−1
0 )cL(b2)c1, bR(b2c−1
L (b2c−1
0 )c0c2 = (Bc0)(b2c2) i.e. the action on the right "leg" is the action of the bisection Bc0.
We will use this expression to get the comultiplication for generators. Let c0 := c(t) := exp(t c) for c ∈ c.
Since B′ is open in B, for fixed b2 ∈ B′ the right hand side of (51) is well defined for t sufficiently close to
0; therefore we can define a vector field on G × B′C which we denote by δ(X r
L (b2)cL(b2c−1
L (b2)cL(b2c−1
L (b2c−1
0 )]c−1
0 ))b1c1, c−1
0 )c0c2(cid:1) =
0 )b2c2(cid:1) .
Note that bR(b2c−1
c ). It is given by:
L (b1c−1
δ(X r
c )(b1c1, b2c2) :=
(c1(t)b1c1, c2(t)b2c2),
c1(t) := c−1
L (b1c−1
L (b2)cL(b2 exp(−t c))) ,
c2(t) := c−1
L (b2 exp(−t c))
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
Let ePc be the projection onto c corresponding to the decomposition g = c ⊕ a and
(52)
gAdc(g) :=ePcAd(g)c.
Computing derivatives we obtain
(53)
δ(X r
c )(b1c1, b2c2) =(cid:16)(Adc(b1)gAdc(aR(b2) c)b1c1 , (Adc(b2) c)b2c2(cid:17) .
20
PIOTR STACHURA
For a basis (Xα) in c we obtain (compare 49):
βα(aR(b2))X r
β(b1c1) + X r
α(b2c2),
δ(X r
α)(b1c1, b2c2) =Xβ gAdc
∆(Aα) = I ⊗ Aα +Xβ
Aβ ⊗ (gAdc
βα · aR)
We may try to go one step further and write this equation in a form similar to (50) as:
But, contrary to (50), where the right hand side has a well defined meaning, here we have rather formal
expression. Certainly we have equality as operators on Ω1/2
(B′C) but this space is not a core
for the left hand side so it is not true that the closure of the right hand side is equal to the self-adjoint
(GB) ⊗ Ω1/2
c
c
operator on the left hand side. The precise formula for comultiplication is given in Prop.1.7.
Let us now, compute the comultiplication for our generators (bS,bYk) (formally, in the sense of (55)). Recall
that they are related (45) to the following basis in c:
( cβ) := ( c0, ck) := (M0(n+1), Mk(n+1) − Mk0) ,
k = 1, . . . , n.
We need to find matrix elements of representation gAdc(a), a ∈ A in the basis ( cβ). Let us denote this matrix
by W(a):
gAdc(a) cβ =Xα
Wαβ(a) cα
We will use lemma 7.1. It is easy to see that the orthogonal complement of a, with respect to the form k
defined in (111), is a⊥ = span{Mβ(n+1) : β = 0, . . . , n} and bases (eβ) and (ρβ) defined as:
(56)
β = 0, . . . , n
ρβ := k(eβ)
eβ := Mβ(n+1)
,
,
are orthonormal basis in a⊥ and a0, respectively. The projection ePc : a⊥ → c acts as ePc(eβ) = Mβ(n+1) −
Mβ0 = cβ. By the lemma 7.1, matrix of gAdc(a) (i.e. the matrix W(a)) is equal to the matrix of Ad#(a)a0
in basis (ρα). For a = (z, U, d) ∈ A, by direct computations (e.g. using formulae in the Appendix) one gets:
W(z, U, d) = d 1+z2
1−z2
d 2
1−z2 z
2
1−z2 ztU D1
1−z2 zzt)U D1 ! , where D1 = dIn−1 0
1 ! .
0
(I + 2
Finally, using (23) one gets:
(58)
W(aR(Λ, u, w, α)) = 1
− u
α
α
wt
α
α ) !
sgn(α)(Λ − uwt
And formulae for comultiplication on generators are:
(54)
(55)
(57)
(59)
∆(bS) = I ⊗bS +bS ⊗
∆(bYi) = I ⊗bYi +bS ⊗
1
α
wi
α
α
+Xk bYk ⊗ −uk
+Xk bYk ⊗ sgn(α)(Λki −
ukwi
α
)
It remains to compute ∆ for generators (Λ, u, w, α). Let f be a smooth and bounded function on B. By (6)
∆(f ), as a function on B × B′, is given by the formula:
(60)
(∆f )(b1, b2) = f (bR(b1aR(b2))) = f (bR(b1(cL(b2))−1)b2) , (b1, b2) ∈ B × B′
Using formulae (22) and (15), after some computations, one gets:
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
21
αΛkl ⊗ ul , ∆(wk) = I ⊗ wk + P −1Xl
wl ⊗ αΛlk
∆(Λkl) =Xj
Λkj ⊗ Λjl + P −1Xmj
Λkmwj ⊗ sgn(α)umΛjl ,
wk ⊗ sgn(α)uk. Notice that P is invertble on B× B′ and right hand sides are well defined
∆(uk) = uk ⊗ sgn(α) + P −1Xl
∆(α) = P −1(α ⊗ α) ,
(61)
where P := 1−Xk
(as smooth functions on B × B′).
4.5. A closer look at the twist.
In this subsection groups G, A, B, C fulfill assumptions (1.5) and C is an exponential Lie group i.e. the
Let logC := exp−1
C and for b ∈ B′ let
exponential mapping expC : c → C is a diffeomorphism.
(62)
ct(b) := expC (t logC(cL(b)−1)) = expC(−t logC(cL(b))) ,
t ∈ R
be the one-parameter group defined by − logC(cL(b)) ∈ c and
(63)
Tt := {(b1ct(b2), b2) : b1 ∈ B, b2 ∈ B′} ⊂ GB × ΓB′.
Note that T1 is the twist (4).
Lemma 4.5. Tt is a one-parameter group of bisections GB × ΓB′.
Proof: It is easy to verify that Tt is a bisection and a submanifold for any t ∈ R. For s, t ∈ R:
(b1c1, b2c2) ∈ TtTs ⇐⇒ ∃ b3, b5 ∈ B , b4, b6 ∈ B′ : (b1c1, b2c2) = (b3ct(b4), b4)(b5cs(b6), b6),
(on the right hand side there is the multiplication in GB × ΓB′) so c2 = e and b2 = b4 = b6 ∈ B′,
and b1c1 = mB(b3ct(b6), b5cs(b6)). Therefore b5 = bR(b3ct(b6)) and b1c1 = b5ct(b6)cs(b6) = b5ct+s(b6), i.e.
TtTs = {(b1ct+s(b2), b2) : b1 ∈ B, b2 ∈ B′} = Tt+s.
Let X r
T be the right invariant vector field on GB × ΓB′ defined by Tt (compare (32)) i.e.
X r
T (b1c1, b2c2) :=
Tt(b1c1, b2c2)
Using the definition (63) we get for b2 ∈ B′:
Tt(b1c1, b2c2) = (bR(b1ct(b2)−1)ct(b2)c1, b2c2) = (cL(b1ct(b2)−1)−1b1c1 , b2c2) =
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
and
= (cR(ct(b2)b−1
1 )b1c1 , b2c2),
(64)
For a basis (Xα) of c let Dα : B′ → R be coordinates of logC(cL(b)):
Dα(b)Xα.
(65)
T (b1c1, b2c2) =(cid:0)(−Adc(b1) logC(cL(b2)))b1c1, 0b2c2(cid:1)
X r
Using the definition (32) of X r
α we can write (64) as
(66)
Dα(b2)X r
α(b1c1)
logC(cL(b)) =:Xα
T (b1c1, b2c2) = −Xα
X r
22
PIOTR STACHURA
Proposition 4.6. Let bTt be the one-parameter group of unitaries in L2(GB × ΓB′) = L2(GB) ⊗ L2(ΓB′) =
L2(GB) ⊗ L2(GB) defined by Tt. The action of bTt on A(GB × ΓB′) is given by
(67)
bTt(F (ω0 ⊗ ω0)) =: (bTtF )(ω0 ⊗ ω0) ,
(bTtF )(g1, g2) := F (T−t(g1, g2)) jC (c−t(bL(g2)))−1/2,
where ω0 = λ0 ⊗ ρ0 for λ0 and ρ0 defined in (103,104) and F ∈ D(GB × ΓB′).
Let T be the generator of bTt . T ie essentially self-adjoint on Ω1/2
(GB × ΓB′):
Ω1/2
c
c
(GB × ΓB′) and for F (ψ0 ⊗ ψ0) ∈
T (F (ψ0 ⊗ ψ0)) =: (T F )(ψ0 ⊗ ψ0) ,
(T F )(b1c1, b2c2) = ι(cid:18)(X r
T F )(b1c1, b2c2) −
1
2
T r(ad(logC(cL(b2)))c)F (b1c1, b2c2)(cid:19)
where ψ0 = ρ0 ⊗ ν0 for some real, non vanishing half density ν0 on B. Moreover on Ω1/2
c
(GB × ΓB′)
T = −Xα
Aα ⊗ Dα
(68)
(69)
Proof:
As for any bisection bTtF is given by [11] :
(bTtF )(Tt(g1, g2)) = F (g1, g2)
(ρ0 ⊗ ρ0)(vg1 ⊗ wg2)
(ρ0 ⊗ ρ0)(Tt(vg1 ⊗ wg2))
, v, w ∈ Λmax(TeC)
Let c(s) ⊂ C be a curve with c(0) = e. For (g1, g2) ∈ GB × ΓB′ let (g1, g2) := Tt(g1, g2). By (63) we get
Tt(c(s)g1, g2) = (c1(s)g1, g2)
Tt(g1, c(s)g2) = (c2(s)g1, c(s)g2),
for some curves c1(s), c2(s). Identifying tangent spaces to right fibers with TeC in corresponding points, one
sees that the map we have to consider has the form M1 M2
I !, so its action on densities is determined by
0
M1 i.e. (derivative of) c(s) 7→ c1(s). But this is exactly as the action of the bisection Bct(bL(g2)) (compare
(31) and by (108) we obtain:
(bTtF )(b1c1, b2c2) = F (T−t(b1c1, b2c2)) jC (ct(b2))−1/2
Since ct(b2) = expC(−t logC(cL(b2))) we can proceed exactly as in the proof of Prop.4.3 and by putting
c = − logC(cL(b2)) in formula (39) we obtain (68). As before, since Ω1/2
a core for T . Formula (69) is a direct consequence of (66) and (68).
(GB × ΓB′) is invariant for bTt it is
4.6. Twist for κ-Poincar´e. Let ( s, y) ∈ R× Rn denotes element of c. The exponential mapping expC and
its inverse logC are given by:
c
expC : c ∋ ( s, y) 7→ (exp( s), y
) ∈ C ,
logC : C ∋ (s, y) 7→ (log(s), y
exp( s) − 1
s
log(s)
s − 1
) ∈ g
With respect to the basis (45) the decomposition is: ( s, y) = − s c0 +Xk
cL : B′ ∋ (Λ, u, w, α) 7→ (α,−sgn(α)u) ∈ C. Thus
yk ck. Recall from (22) the mapping
logC (cL(Λ, u, w, α)) = (log(α),−sgn(α)
u) = − log(α) c0 − sgn(α)
uk ck
log(α)
α − 1
(70)
i.e.
D0(Λ, u, w, α) = − log(α) , Dk(Λ, u, w, α) = −sgn(α)
log(α)
α − 1 Xk
log(α)
α − 1
uk
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
23
The bisection Tt is equal:
Tt := {(b1,α2−t,
sgn(α2)(α2−t − 1)
1 − α2
And from (69) and (70) we get the formula for the twist:
u2; Λ2, u2, w2, α2, 1, 0) : α2 6= 0 , b1 ∈ B} ⊂ GB × ΓB′
T = exp(ιT ) ,
T = bS ⊗ log α +Xk bYk ⊗
sgn(α) log α
α − 1
uk
5. Comparison with relations coming from Poisson-Poincar´e Group
The classical Poincar´e Group, our (C∗(GB), ∆) is a quantization of, is (T A)0 ⊂ T ∗G. After identifying
T ∗G with the semidirect product g∗ ⋊ G (113), (T A)0 is identified with a0 ⋊ A. Invariant, bilinear form
k (111) on g induces the form on a0 which makes it a vector Minkowski space and the action of A is by
orthogonal transformations. The bundle (T A)0 is dual to the Lie algebroid of groupoid ΓA, so objects related
to this groupoid have more direct relation to functions on a0 ⋊ A (see [14] for a detailed description). The
groupoid GB was used to overcome some functional analytic problems, let us now transport our expressions
to the groupoid ΓA to relate them to formulae in [22, 8].
5.1. Back to ΓA picture. Right invariant fields X r
d
c . By (31) and (32) X r
L (bc−1
t )bc, where ct := exp(t c). If bc ∈ ΓB′ i.e. b, bR(bc) ∈ B′ then it is easy to see that bR(bc−1
c (bc) = d
dt(cid:12)(cid:12)t=0 bR(bc−1
t )ctc =
t ) ∈ B′
c (bc) to ΓA by the mapping
t )ctc ∈ ΓB′. So we can push X r
dt(cid:12)(cid:12)t=0 c−1
for t sufficiently close to 0 and, consequently, bR(bc−1
(3) bc 7→ aR(b)c:
bR(bc−1
t ))ctc = aR(aR(b)c−1
t
t )ctc 7→ aR(bR(bc−1
t )) = aR(bc−1
t ) = aR(aR(b)c−1
since aR(bR(bc−1
ΓA by X A,r
:
c
)(ctaR(b)−1)aR(b)c = c−1
L (aR(b)c−1
t
)aR(b)c
t ). Let us denote the resulting (right-invariant) vector field on
(71)
X A,r
c
(ac) :=
where gAdc was defined in (52).
Anchor. Let us transfer formula (33):
c−1
L (a exp(−t c))ac = (gAdc(a) c)ac,
aR(bR(bc−1
t )) =
aR(aR(b)c−1
t ) =
R (b))aR(b)
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
d
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
aR(aR(b)c−1
t a−1
= (−Ada(aR(b)) c)aR(b)
Thus, denoting the anchor of ΓA by ΠA, we obtain:
(72)
ΠA(X A,r
c
)(a) := (−Ada(a) c)a.
Comparing this formula and (71) with (34) and (32) we see that we can immediately write commutation
relations analogous to relations (35) and (36):
(73)
[AA
c , AA
e ] = −ιAA
[ c, e], c, e ∈ c
[AA
c , f ] = ι( ΠA(X A,r
c
c are, as before, Lie derivatives along X A,r
Operators AA
(ΓA)
(and A(ΓA)), however, their status is "weaker" then one of relations (35) and (36. Firstly, operators AA
c
are not essentially self-adjoint on Ω1/2
(ΓA); secondly, even though each smooth function on A defines an
element affiliated to C∗r (ΓA) = C∗r (GB) (or even a multiplier, if f is bounded), not every such a function
c multiplied by i. These relations hold on Ω1/2
c
c
)f )b, f ∈ C∞(A).
24
PIOTR STACHURA
defines smooth (or even continuous) function on B and only for such functions formula (36) has immediate
meaning.
Comultiplication. By computations similar to ones before (71) we obtain from (53):
and, for a basis (Xα) in c:
(74)
δ(X A,r
δ(X A,r
c
)(a1c1, a2c2) =(cid:16)(gAdc(a1)gAdc(a2) c)a1c1 , (gAdc(a2) c)a2c2(cid:17) .
α )(a1c1, a2c2) =Xβ gAdc
β (a1c1) + X A,r
βα(a2)X A,r
α (a2c2),
or, as operators on Ω1/2
c
(75)
(ΓA × ΓA) we can write (compare to (50):
∆(AA
α ) = I ⊗ AA
α +Xβ
βα
AA
β ⊗gAdc
But again, because of problems with domains, this expression is rather formal.
Finally, let us check that if we transfer formula (60) we get what we expect. The formula (3) gives
diffeomorphism φ : A → B′:
(76)
φ(a) := bR(a) , φ−1(b′) = aR(b′) , a ∈ A, b′ ∈ B′
For f ∈ C∞(A) using (60) let us compute:
[(φ∗ ⊗ φ∗)−1∆(φ∗f )](a1, a2) = ∆(φ∗f )](bR(a1), bR(a2)) = (φ∗f )(bR[bR(a1)aR(bR(a2))]) =
= (φ∗f )(bR(bR(a1)a2)) = (φ∗f )(bR(a1a2)) = f (aR(bR(a1a2))) =
= f (a1a2)
5.2. Hopf ∗-algebra of κ-Poincar´e from quantization of Poisson-Lie structure. On a Hopf *-
algebra level, relations for κ-Poincare Group were given in [22] (and [8]). The ∗-algebra is generated by
self-adjoint elements {aα, Lαβ , α, β = 0, 1 . . . , n} that satisfy the following commutation relations for ηαβ :=
diag(1,−1, . . . ,−1) and some h ∈ R (to compare with [8] substitute κ := h−1):
(77)
η = LtηL
[a0, ak] = ihak ,
[Lαβ, Lγδ] = 0 ,
[ak, al] = 0 ,
[a0, L00] = ih((L00)2 − 1) ,
[a0, L0m] = ihL00L0m ,
(78)
[a0, Lm0] = ihL00Lm0 ,
[a0, Lmn] = ihLm0L0n ,
together with coproduct ∆:
[ak, L00] = ih(L00 − 1)Lk0
[ak, L0m] = ih(L00 − 1)Lkm
[ak, Lm0] = ih(Lk0Lm0 − δkm(L00 − 1))
[ak, Lmn] = ih(Lm0Lkn − δkmL0n)
antypode S and counit ǫ:
∆(aα) = aα ⊗ I +Xβ
S(aα) = −Xβ
S(Lαβ)aβ ,
(79)
(80)
(81)
ǫ(aα) = 0 ,
ǫ(Lαβ) = δαβ
Lαβ ⊗ aβ ,
∆(Lαβ) =Xγ
Lαγ ⊗ Lγβ
S(Lαβ) = (L−1)αβ = (ηLtη)αβ
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
25
5.3. Comparison of formulae. Now we want to compare our generators and relations to formulae (77 - 79).
We restrict computations only to commutation relations and comultiplication (and partially to antipode)
and are not going to investigate in details remaining parts of quantum group structure on (C∗r (ΓA), ∆) (i.e.
counit, antipode and Haar weight; this can be done with the use of expressions given in [12]).
Comparing (116) and (75) we see that, with respect to the comultiplication, our generators (bYα) (where
we put bY0 := bS) behave like S(aα). Such an identification would not be consistent with self-adjointness,
however. Instead we use the unitary part of antipode (see e.g. [19]), which, for a quantum group defined by
DLG, is implemented by the group inverse [12]. As said above, we are not going to present all functional
analytic details but summarize computations in the following:
Lemma 5.1. Let (M, ∆) be a ∗-bialgebra generated by self-adjoint elements (ak, Vkl, V c
satisfying (1)Xm
Xl
mkVml = δklI , (2) ∆(Vkl) = Xm
lm = Xm
Vkl ⊗ al. Then
VkmV c
kl) , k, l = 1, . . . , n
Vkm ⊗ Vml and (3) ∆(ak) = ak ⊗ I +
V c
a) Xm
V c
lmVkm =Xm
b) Elements Ak := −
1
2Xm
VmlV c
mk = δklI and ∆(V c
kl) =Xm
V c
km ⊗ V c
ml.
(amV c
mk + V c
mkam) are self-adjoint and satisfy
c) Moreover, if Xmk
V c
mk[am, Vlk] =Xmk
ak = −
and M is generated by (Ak, Vkl, V c
kl).
∆(Ak) = I ⊗ Ak +Xm
[am, Vlk]V c
mk then
Am ⊗ V c
mk
1
2Xm
(VkmAm + AmVkm)
(82)
(83)
as:
(84)
Proof: a) The first equality is just ∗ applied to (1); for the second one, apply ∆ to (1), use (2) and then (1);
b) self-adjointness is evident and the formula for ∆(Ak) is a simple computation;
c) Clearly, equality (83) is sufficient for statement about generation. Let us rewrite the assumption in (c)
by (1) and (a) expressions in brackets are δlmI and we obtain:
V c
mkamVlk −Xm Xk
Xmk
am Xk
VlkV c
mk! −Xmk
VlkamV c
mk
V c
mkVlk! am =Xm
2al =Xmk
(V c
mkamVlk + VlkamV c
mk)
Let us compute using definition of Ak and (1):
Xk
VskAk = −
VskamV c
mk −
1
2Xmk
VskV c
mkam = −
1
2Xmk
VskamV c
mk −
1
2
as
In the similar way, using (a):
as −
Adding these two equalities and using (84) we get (83).
AkVsk = −
1
2
mkamV c
V c
sk
1
2Xmk
1
2Xmk
Xk
The version of the lemma above, with ak and Ak interchanged, is proven in the same way:
kl, Vkl) , k, l = 1, . . . , n
ml and (3) ∆(Ak) = I ⊗ Ak +
26
PIOTR STACHURA
V c
VmlV c
Lemma 5.2. Let (M, ∆) be a ∗-bialgebra generated by self-adjoint elements (Ak, V c
satisfying (1)Xm
Xl
Al ⊗ V c
a) Xm
lmVkm = Xm
lm =Xm
kl) = Xm
mk = δklI , (2) ∆(V c
Vkm ⊗ Vml ;
km ⊗ V c
V c
lk. Then
VkmV c
(VkmAm + AmVkm) are self-adjoint and satisfy
b) Elements ak = −
1
V c
mkVml = δklI and ∆(Vkl) =Xm
2Xm
∆(ak) = ak ⊗ I +Xl
Vkl ⊗ al.
c) Moreover, if Xmk
[V c
ks, Am]Vkm =Xmk
Ak = −
and M is generated by (ak, V c
kl, Vkl).
Vkm[V c
ks, Am] then
(amV c
mk + V c
mkam)
1
2Xm
(85)
(86)
Looking at the formulae (59) and using the fact that W given by (57) is a representation of the group A
we use the lemma 5.2 with V c = W , where matrix elements of W are expressed by (Λ, w, u, α) as in (58)
(using identification of A with B′ as in (76)). So we define:
− wt
(87)
L := 1
α
u
α
α
α ) !
sgn(α)(Λ − uwt
LαβbYβ −
1
2Xβ
[bYβ, Lαβ] =
and
(88)
aα := −
1
2Xβ (cid:16)LαβbYβ +bYβLαβ(cid:17) = −Xβ
2Xβ
[bYβ, Lαβ]
1
=: aα −
In the formula above and in what follows we treat bYα (recall that bY0 := bS) and matrix elements of L as
(ΓA) (or A(ΓA)).
operators on Ω1/2
c
Remark 5.3. Notice that matrix elements of L, despite being non-continuous functions on B, are affiliated
to C∗(GB) (wich is equal to C∗r (GB)) because they are smooth functions on A and C∗(GB) = C∗r (ΓA).
[aµ, Lβγ] = [aµ, Lβγ]
operators, it is clear that
α , sgn(α) and 1
α
Since for f ∈ C∞(A) commutators [bYµ, f ] ∈ C∞(A) and functions on A commute, we have
We will need commutators of bYβ with 1
α(cid:21) =(cid:20)bYβ,
By this equality we have (cid:20)bYβ,
α(cid:21) =(cid:20)bYβ,
α(cid:21) α +
0 =(cid:20)bYβ,
[bYβ, sgn(α)] = 0.
(cid:21) = sgn(α)(cid:20)bYβ,
αhbYβ, αi ⇒ (cid:20)bYβ,
αhbYβ, αi 1
α(cid:21) and, for the last commutator,
α(cid:21) = −
. Since α 6= 0 on A and bYβ are differential
sgn(α)
(89)
1
α
α
α
1
1
1
1
1
1
.
Now, using (41), (43) we obtain:
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
27
1
α
] = ι(1 −
1
α2 ) ,
[bS,
1
α
] = ι
α − 1
α2 wm .
[bYm,
(90)
matrix elements of L:
Having these relations together with (41), (43) by direct computation on obtains commutators of bYγ with
where we use sgn(γ) :=( 1
hbYγ, Lβµi = ι (δγµ(δβ0 − Lβ0) − sgn(γ)Lβγ(L0µ − δ0µ)) ,
LργhbYγ, Lβµi =
[aρ, Lβµ] = [aρ, Lβµ] = −Xγ hLργbYγ, Lβµi = −Xγ
. Now we get
γ = 0
γ > 0
−1
for
for
(91)
= ι (Lρµ(Lβ0 − δβ0) + sgn(ρ)δρβ(L0µ − δ0µ)) .
and these are relations (78) (for h = 1).
Finally, lets us verify (77). By (90) and (44):
Xγ
[aµ, aν ] =Xγδ hLµγbYγ, LνδbYδi =Xγδ
[aµ, aν] =haµ −
= ι(δ0µaν − δ0ν aµ).
ιn
2
(δµ0 − Lµ0), aν −
ιn
2
[bYγ, Lβγ] = ιn(δβ0 − Lβ0)
LµγLνδhbYγ,bYδi + LµγhbYγ, LνδibYδ − LνδhbYδ, LµγibYγ =
(δν0 − Lν0)i =
Lµ0)(cid:17) =
ιn
2
ιn
2
ιn
2
ι (δµ0Lν0 − δν0Lµ0) =
= [aµ, aν] +
= ι(cid:16)δµ0(aν +
= ι(cid:16)δµ0(aν +
([aµ, Lν0)] − [aν, Lµ0)]) = ι(δ0µaν − δ0ν aµ) +
ιn
Lν0) − δν0(aµ +
2
ιn
(Lν0 − δν0)) − δν0(aµ +
2
(Lµ0 − δµ0)(cid:17) =
ιn
2
= ι (δµ0aν − δν0aµ)
and
(92)
as in (77).
6. (Some remarks on) Quantum κ-Minkowski space.
In this last section we are going to make few remarks on "quantum κ-Minkowski space". Under this name
is usually understood "∗-algebra generated by self-adjoint elements (x0, xk) satisfying relations" (77):
[x0, xk] = ihxk ,
[xk, xl] = 0 ,
k = 1, . . . , n
As a Hopf algebra it was introduced already in [7]. This is considered either on a pure algebra level or as
operators on Hilbert space [1] or within a star-product formulation of simplified two-dimensional version in
[3]. But if one wants to consider it on a C∗-algebra level, it is clear (in fact since [10], [23] and certainly
since [15]) that the "unique candidate" for this name is C∗(C) with the group C defined in (12) i.e. C is
the AN group from the Iwasawa decomposition SO0(1, n) = SO(n)AN (this group appears in [1] under the
name "κ-Minkowski Group"). But to call a space the "Minkowski space" it should carry an action of the
Poincar´e Group, so in our case we have to show the action of our quantum group (C∗(ΓA), ∆) on C∗(C).
28
PIOTR STACHURA
Let us now show, how the groupoid framework gives natural candidate for such an action and postpone
analytic details to future publication.
Remark 6.1. A "quantum space" C∗(C) has a family of classical points {pλ : λ ∈ R} i.e. characters. They
are given by 1-dimensional unitary representations of C pλ(s, y) := sιλ. These are representations induced
from whole line of 0-dimensional symplectic leaves on which the Poisson bivector for the related Poisson
Minkowski space (affine) described e.g. in [14] vanishes.
Let us begin with simpler situation of global decomposition and let (G; B, C) be a double group with
⊲ GB × GB defined by (2). Consider relations δL : C ⊲ GB × C and δR : C ⊲ C × GB defined
δ0 : GB
by:
(93)
δL := {(g, cR(g); cL(g)) : g ∈ G}
,
δR := {(cL(g), g; cR(g)) : g ∈ G}
The following lemma can be proven by direct computation:
Lemma 6.2. δL and δR are morphisms of groupoids that satisfy:
(94)
(δ0 × id)δL = (id × δL)δL
,
(id × δ0)δR = (δR × id)δR
Let σ : C × GB ∋ (c, g) 7→ (g, c) ∈ GB × C be the flip and RB, RC denote the inverse in a group G and its
restriction to C respectively. Then the following equality holds:
(95)
δL = σ(RC × RB)δRRC
If (G; B, C) is a DLG, relations δL, δR are morphisms of differential groupoids and lifting them one obtains
∆L, ∆R -- morphisms of corresponding C∗-algebras, e.g. ∆L ∈ M or(C∗(C), C∗(GB × C)). One may expect
that in "nice" cases C∗(GB × C) = C∗(GB) ⊗ C∗(C) and reduced/universal algebras problem can be also
handled. This way one gets actions (left or right) of (C∗(GB), ∆0) on C∗(C). Let us show that ∆L should
be considered as "quantization" of the canonical affine action of the semidirect product b0 ⋊ B on b0:
(96)
(b0 ⋊ B) × b0 ∋ (ϕ, b; ψ) 7→ ϕ + Ad#(b)ψ ∈ b0
Let Γ1 ⇒ E1 and Γ2 ⇒ E2 be differential groupoids. For a (differentiable) relation h : Γ1
⊲ Γ2 by
T ∗h : T ∗Γ1
⊲ T ∗Γ2 we denote the relation (cotangent lift of h) defined by
(97)
(ϕ2, ϕ1) ∈ T ∗h ⇐⇒ ∀(v2, v1) ∈ T h < ϕ2, v2 >=< ϕ1, v1 > .
If h is a morphism of differential groupoids then T ∗h is a morphism of symplectic groupoids and its base
map is a Poisson map (T E2)0 → (T E1)0 [21].
By (93), (97) and (114) for Φ -- the base map of T ∗δL, we have Φ : (T B)0 × c∗ → c∗ and for (φ, ψ) ∈
(TbB)0 × c∗, c ∈ c ⊂ g and identifying c∗ with b0 ⊂ g∗:
< Φ(φ, ψ), c > =< (φ, ψ), ( cb, cR( cb) >=< φ, cb > + < ψ, Adc(b−1)( c) >=
=< φ, cb > + < Ad#(b)(ψ), c) >=< ϕ + Ad#(b)(ψ), c >,
where, using right trivialization, we represent φ ∈ (TbB)0 by (ϕ, b) ∈ b0× B. This way we get Φ(ϕ, b; ψ) =
ϕ + Ad#(b)(ψ)
exactly as in (96).
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
29
In the situation of κ-Poincar´e, relations defined as δL, δR in (93) e.g.
δL : C ⊲ ΓA × C ,
δL := {(g, cR(g); cL(g)) : g ∈ ΓA}
are not morphisms of differential groupoids. It can be directly verified that m′(δL × δL) * δLm , where
m and m′ denote multiplication relations in C and ΓA × C, respectively. Or one may observe that T ∗δL
restricted to sets of units gives the action of Poisson-Poincar´e group a0 ⋊ A on c∗ ≃ a0 as in (96). It is
known that this is a Poisson action which is non complete, so it cannot be the base map of a morphism of
symplectic groupoids [21]. Essentially we face the similar problem as in the very beginning: some operators
that "should be" self-adjoint are not essentially self-adjoint on their "natural" domains and we can try to
overcome it in the similar way -- passing from ΓA to GB.
It is easier to work with δR : C ⊲ C × ΓA given by δR := {(cL(g), g; cR(g)) : g ∈ ΓA} , or after passing
to ΓB′ (and using the same symbol δR):
(98)
Let us define
δR = {(cL(bL(g))−1cL(g), g; cR(g)) : g ∈ ΓB′} ⊂ (C × ΓB′) × C
(99)
T C := {(cL(b)−1, b) : b ∈ B′} = (cR × id)T ⊂ C × GB , T C
12 := T C × B ⊂ C × GB × GB
The following lemma may be compared to Prop. 1.3.
Lemma 6.3.
(1) T C is a section of left and right projections over {e} × B′ ⊂ C × GB (e is the neutral
element in C) and a bisection of C × ΓB′.
(2) (id × δ0)T C is a section of left and right projections (in C × GB × GB) over the set {e} × δ0(B′) =
{(e, b2, b3) : b2b3 ∈ B′};
(3) (δR × id)T C is a section of left and right projections over the set {e} × B × B′;
(4) T C
12(δR × id)T C = T23(id × δ0)T C (equality of sets in C × GB × GB), moreover this set is a section
of the right projection over {e} × (δ0(B′) ∩ (B × B′)) and the left projection over {e} × B′ × B′.
Proof: The first statement is clear from the definition of T C. By a straightforward computation:
(100)
(id × δ0)T C = {(cL(b1b2)−1, b1, b2) : b1b2 ∈ B′} ⊂ C × GB × GB
(δR × id)T C = {(cL(g), g, b) : b ∈ B′, cR(g) = cL(b)−1} =
= {(cL(b2cL(b3)−1), b2cL(b3)−1, b3) : b2 ∈ B, b3 ∈ B′} ⊂ C × GB × GB,
the second and third statement follow easily from these expressions. Now, using these expressions, it is easy
to compute T C
12(δR × id)T C and get:
12(δR × id)T C = {cL(b2)−1cL(b2cL(b3)−1), b2cL(b3)−1, b3) : b2, b3 ∈ B′},
T C
and the same result for T23(id × δ0)T C . The statement about left projection is clear, the only relation we
need to check is that for b2, b3 ∈ B′ the product bR(b2cL(b3)−1)b3 is again in B′. Let b2 = c2a2 and b3 = c3a3.
Then
bR(b2cL(b3)−1)b3 = bR(c2a2c−1
3 )b3 = bR(c2a2c−1
3 b3) = bR(c2a2a3) = bR(a2a3) ∈ B′
T C is used to twist δR to δR:
30
PIOTR STACHURA
Lemma 6.4. Let δR : C ⊲ C × GB be the morphism defined in (93) and δR be as in (98). They are related
by: δR = AdT C · δR.
Proof: Recall that AdT C : C × GB
⊲ C × GB is defined by (compare (5)):
(c1, g2; c3, g4) ∈ AdT C ⇐⇒ ∃t1, t2 ∈ T C : (c1, g2) = t1(c3, g4)(sC × sB)t2
The multiplication above is in the groupoid C × GB and sC, sB are groupoid inverses (i.e. sC stands for the
inverse in the group C). Let us compute:
(c1, g2; c3, g4) ∈ AdT C ⇐⇒ ∃b5, b6 ∈ B′ : (c1, g2) = (cL(b5)−1, b5)(c3, g4)(cL(b6), b6),
i.e. b5 = bL(g4) ∈ B′ , b6 = bR(g4) ∈ B′ therefore g2 = g4 ∈ ΓB′ and c1 = cL(b5)−1c3cL(b6); so this relation
is in fact bijection
AdT C : C × ΓB′ ∋ (c, g) 7→ (cL(bL(g))−1c3cL(bR(g)), g) ∈ C × ΓB′
defined by the bisection T C of C × ΓB′.
Now we have:
(c1, g2, c3) ∈ AdT C δR ⇐⇒ ∃g ∈ G : (c1, g2; cL(g), g) ∈ AdT C , cR(g) = c3
therefore g ∈ ΓB′ and c1 = cL(bL(g))−1cL(g)cL(bR(g)) = cL(bL(g))−1cL(g) and
AdT C δR = {(cL(bL(g))−1cL(g), g, cR(g)) : g ∈ ΓB′}
exactly as δR in (98).
T C, as a bisection of C × ΓB′, defines unitary multiplier cT C of C∗r (C × ΓB′) = C∗r (C) ⊗ C∗r (ΓB′) =
C∗(C)⊗C∗r (GB) = C∗(C)⊗C∗(GB), so having ∆R -- action of the quantum group (C∗(GB), ∆0) on "quantum
space" represented by C∗(C) (lifted from δR) we can, due to properties of T C described in Lemma 6.3, define
the action of our quantum group (C∗(ΓA), ∆) on the same "space" C∗(C) by e∆R(a) := cT C∆R(a)cT C−1
.
Whether this construction gives continuous action of κ-Poincar´e on C∗(C) still needs to be verified.
7. Appendix
Here we collect some formulae proven in [12] and used in this paper. (G; B, C) is a double Lie group,
g, b, c are corresponding Lie algebras and g = b ⊕ c (direct sum of vector spaces). Let Pb, Pc be projections
in g corresponding to the decomposition g = b ⊕ c. Let us define:
(101)
Adb(g) := PbAd(g)b , Adc(g) := PcAd(g)c
Clearly Adb and Adc are representations when restricted to B or C.
Modular functions. Let us define:
(102)
jB(g) := det(Adb(g)) , jC (g) := det(Adc(g))
The choice of ω0. Choose a real half-density µ0 6= 0 on TeC and define left-invariant half-density on GB by
(103)
λ0(g)(v) := µ0(g−1v) , v ∈ ΛmaxT l
gGB.
The corresponding right-invariant half-density is given by:
(104)
ρ0(g)(w) := jC(bL(g))−1/2µ0(wg−1) , w ∈ ΛmaxT r
g GB.
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
31
Multiplication and comultiplication in A(GB) After the choice of ω0 as above, the multiplication in A(GB)
reads: (f1ω0)(f2ω0) =: (f1 ∗ f2)ω0 and
(105)
(f1 ∗ f2)(g) =ZC
=ZC
dlc f1(bL(g)c)f2(cL(bL(g)c)−1g) =
drc jC (bL(cbR(g)))−1f1(gcR(cbR(g))−1)f2(cbR(g)),
where dlc and drc are left and right Haar measures on C defined by µ0.
The · l defined by this ω0 is given by:
(106)
Let δ0 := mT
C : GB
b∈BZC
⊲ GB × GB. The formula for δ0 reads
fl = sup
dlcf (bc)
(107)
(δ0(f )F )(b1c1, b2c2) =ZC
δ0(f ω0)(F (ω0 ⊗ ω0)) =: (δ0(f )F )(ω0 ⊗ ω0)
dlc jC (cL(b2c))−1/2f (b1b2c)F (cL(b1b2c)−1b1c1, bR(b2c)c−1c2)
Action of bisections on bidensities. Let us write ω = f ω0 , f ∈ D(GB). The action of a bisection Bc0 on
A(GB) is given by:
(108)
(Bc0f )(g) = f (B(c0)−1g)jC (c0)−1/2
(Bc0)(f ω0) =: (Bc0f ) ω0 ,
Notation for orthogonal Lie algebras. Let (V, η) be a real, finite dimensional vector space with a
bilinear, symmetric and non degenerate form η; by η we denote also the isomorphism V → V ∗ defined by
< η(x), y >:= η(x, y). A basis (vα) of V is called orthonormal if η(vα, vβ) = η(vα, vα)δαβ ,
η(vα, vα) = 1.
For a subset S ⊂ V the symbol S⊥ is used for the orthogonal complement of S, the symbol S0 ⊂ V ∗ stands
for the annihilator of S. Let us define operators in End(V ):
(109)
Mxy := x ⊗ η(y) − y ⊗ η(x)
,
x, y ∈ V.
For a basis (vα) in V we write Mαβ instead of Mvα,vβ . Operators Mxy satisfy:
(110)
[Mxy, Mzt] = η(x, t)Myz + η(y, z)Mxt − η(x, z)Myt − η(y, t)Mxz
and so(η) = span{Mxy : x, y ∈ V }. Notice that for g ∈ O(η) we have Ad(g)(Mxy) = Mgx,gy. We will use a
bilinear, non degenerate form k : so(η) × so(η) → R defined by:
(111)
k(Mxy, Mzt) := η(x, t)η(y, z) − η(x, z)η(y, t)
It is easy to see that for g ∈ O(η): Ad(g) ∈ O(k) i.e.
k(gMxyg−1, gMztg−1) = k(Mxy, Mzt) , g ∈ O(η)
(of course k is proportional to the Killing form on so(η)). By Ad# we denote the coadjoint representation
of O(η) on so(η)∗: Ad#(g) := Ad(g−1)∗. If k is the isomorphism so(η) → so(η)∗ defined by the form k then
Ad#(g)k(X) = k(Ad(g)X) , X ∈ so(η), g ∈ O(η)
Let us also define a bilinear form k on so(η)∗ by:
(112)
k(ϕ, ψ) := k(k−1(ϕ), k−1(ψ)) , ϕ, ψ ∈ so(η)∗
32
PIOTR STACHURA
so k(ϕ, ψ) =< ϕ, k−1(ψ) >; again it is clear that if g ∈ O(η) then Ad#(g) ∈ O(k), and
k(Ad#(g)k(X), k(Y )) = k(Ad(g)X, Y ) , X, Y ∈ so(η).
Adjoint, coadjoint representations and Hopf algebra structure. For a Lie group G we identify the
group T ∗G, via right translations, with the semidirect product g∗ ⋊ G (with coadjoint representation):
(113)
(ϕ, g)(ψ, h) := (ϕ + Ad#(g)ψ, gh) , ϕ, ψ ∈ g∗ , g, h ∈ G,
If B ⊂ G is a subgroup with a Lie algebra b ⊂ g then b0 × B is a subgroup of g∗ ⋊ G.
If c ⊂ g is any complementary subspace to b i.e. g = b⊕ c, then Adc(b) := PcAd(b)c is a representation of
B on c. The spaces c and b0 are dual to each other and the representation Adc is contragradient to Ad#B,
i.e. for ϕ ∈ b0, c ∈ c and b ∈ B:
< Ad#(b)ϕ, c > =< ϕ, Ad(b−1) c >=< ϕ, PcAd(b−1) c >=< ϕ, Adc(b−1) c >=
(114)
=< (Adc(b−1))∗ϕ, c >
Let (ρk) be a basis in b0, ( ck) dual basis in c and Ad#
in corresponding bases:
lk, Adc
lk : B → R matrix elements of Ad#(b) and Adc(b)
< ρl, cm >= δlm ,
Ad#(b)ρk =Xl
Ad#
lk(b)ρl
,
Adc(b) ck =Xl
Adc
lk(b) cl
Clearly, the equality (114) implies Ad#
lk(b−1) = Adc
kl(b), i.e.
Let us use the same symbols for extensions of functions Ad#
lk, ck to b0 ⋊ B i.e.
X Adc
km(b)Ad#
kl(b) =X Ad#
lk(b)Adc
lk, Adc
mk(b) = δlm.
Ad#
kl(ϕ, b) := Ad#
kl(b) , Adc
kl(ϕ, b) := Adc
kl(b) ,
ck(ϕ, b) :=< ϕ, ck >
It is straightforward to compute action of comultiplication, counit and antipode, defined by the group b0 ⋊B,
on functions ck, Ad#
kl, Adc
kl:
S( Ak) = −Xl
Al Ad#
kl ,
ck = −Xl
Ad#
kl
Al
Let us assume additionally, that g is equipped with invariant, non degenerate, symmetric bilinear form k
and that kb is non degenerate. Thus we have two decompositions
g = b ⊕ b⊥ = b ⊕ c
Let Pb be the projection on b defined by the first decomposition and Pc projection on c defined by the
second one. The next lemma is straightforward.
ǫ( ck) = 0 ,
S( ck) = −Xm
S(Ad#
km) cm = −Xm
Adc
mk cm ;
ǫ(Ad#
kl) = δkl ,
ǫ(Adc
kl) = δkl ,
S(Ad#
kl) = Adc
lk ;
S(Adc
kl) = Ad#
lk .
(115)
∆( ck) = ck ⊗ I +Xm
kl) =Xm
kl) =Xm
∆(Ad#
∆(Adc
Ad#
km ⊗ cm ,
Ad#
Adc
km ⊗ Ad#
ml ,
km ⊗ Adc
ml ,
Let us define Ak := S( ck) = −Xl
∆( Ak) = I ⊗ Ak +Xm
(116)
Adc
lk cl , then:
Am ⊗ Adc
mk ,
THE κ-POINCAR´E GROUP ON A C ∗-LEVEL.
33
Lemma 7.1. (a) The restriction of k to b⊥ is an isomorphism of b⊥ and b0. For any c ∈ c and any f ∈ b⊥:
Pc(I − Pb) c = c , (I − Pb)Pc f = f , in other words the restriction of Pc to b⊥ is an isomorphism of b⊥ and
c, and the inverse mapping is the restriction of I − Pb to c.
(b) The mapping φ := k·(I − Pb)c : c → b0 is an isomorphism with the inverse φ−1 = Pc·(k−1b0), moreover
for any b ∈ B
φ · Adc(b) · φ−1 = Ad#(b)b0
(c) Let (ek) be o.n. basis in b⊥; it defines bases (k(ek)) and (Pc(ek)) in b0 and c, respectively. Then
φ(Pc(ek)) = k(ek) and, consequently, matrix elements of Ad(b)b⊥, Ad#(b) and Adc(b) in bases (ek), (k(ek))
and (Pc(ek)), respectively, are equal.
References
[1] A. Agostini, κ-Minkowski representations on Hilbert spaces, J. Math. Phys. 48 (2007), 052305;
[2] A. Borowiec, A. Pachol, κ-Minkowski Spacetimes and DSR Algebras: Fresh Look and Old Problems, SIGMA 6 (2010) 086;
[3] B. Durhuus, A. Sitarz, Star product realizations of κ-Minkowski space, J. Noncommut. Geom. 7 (2013), 605645;
[4] P. Liebermann, C-M. Marle, Symplectic Geometry and Analytical Mechanics, D. Reidel 1987 p. 421;
[5] J. Lukierski, A Nowicki and H. Ruegg, New quantum Poincar´e algebra and κ-deformed field theory, Phys. Lett. B 293 (1992)
344-352;
[6] J. Lukierski, Kappa-deformations: historical developments and recent results, J. Phys.: Conf. Ser. 804 (2017) 012028; also
arXiv:1611.10213;
[7] S. Majid, H. Ruegg , Bicrossproduct structure of κ-Poincar´e group and non-commutative geometry, Phys. Lett. B 334 (1994),
348-354;
[8] P. Kosinski, P. Maslanka , The κ-Weyl group and its algebra, Proc. 21 Intern. Coll. On Group Theor. Methods in Phys.,
Goslar 1996, Heron, Sophia 1997, also arXiv: q-alg/9512018;
[9] G. Skandalis, Duality for locally compact 'quantum groups (joint work with S. Baaj), Mathematisches Forschungsinstitut
Oberwolfach, Taungsbericht 46/1991, C ∗-algebren, 20,10-26.10.1991,p. 20;
[10] P. Stachura , Double Lie Algebras and Manin triples, arXiv:q-alg/9712040;
[11] P. Stachura , Differential groupoids and C ∗-algebras, math.QA/9905097, for a shorter exposition see:, C ∗-algebra of a
differential groupoid, Banach Center Publ 51, Inst. Math. Polish Acad. Sci., 2000, 263-281;
[12] P. Stachura , From double Lie groups to quantum groups, Fund. Math. 188 (2005), 195-240;
[13] P. Stachura , On the quantum 'ax+b' group, J. Geom. and Phys. 73 (2013) 125-149;
[14] P. Stachura , On Poisson structures related to κ-Poincar´e group, Int. J. Geom. Methods M. 14 (2017) 1750133;
[15] S. Vaes, L. Vainerman , Extensions of locally compact quantum groups and the bicrossed product constraction, Adv. in Math
175 (1) (2003), 1-101;
[16] D. Williams , Crossed Products of C ∗-Algebras, Math. Surveys and Monographs, vol 134. AMS 2007;
[17] S. L. Woronowicz, K. Napi´orkowski , Operator theory in the C*-algebra framework, Rep. Math. Phys. 31 (3) (1992), 353-371;
[18] S. L. Woronowicz, C ∗-algebras generated by unbounded elements, Rev.Math. Phys. 7 (1995), no. 3, 481-521;
[19] S. L. Woronowicz, From multiplicative unitaries to quantum groups, Int. J. Math. 7 no. 1 (1996), 127-149;
[20] S. Zakrzewski, Quantum and Classical pseudogroups I, Comm. Math. Phys. 134 (1990), 347-370;
[21] S. Zakrzewski, Quantum and classical pseudogroups. II. Differential and symplectic pseudogroups, Comm. Math. Phys.,
134 (2) (1990), 371-395;
[22] S. Zakrzewski, Quantum Poincare group related to the κ -Poincar´e algebra, J. Phys. A Math. Gen. 27 (1994) 2075-2082;
[23] S. Zakrzewski, Poisson structures on the Poincar´e group , Comm. Math. Phys. 185 (1997) 285 -311.
Faculty of Applied Informatics and Mathematics, Warsaw University of Life Sciences-SGGW, ul Nowoursyn-
owska 166, 02-787 Warszawa, Poland, e-mail: piotr [email protected]
|
1011.6034 | 2 | 1011 | 2013-07-24T00:09:45 | Non-cocommutative C$^{*}$-bialgebra defined as the direct sum of free group C$^{*}$-algebras | [
"math.OA",
"math.QA"
] | Let ${\Bbb F}_{n}$ be the free group of rank $n$ and let $\bigoplus C^{*}({\Bbb F}_{n})$ denote the direct sum of full group C$^{*}$-algebras $C^{*}({\Bbb F}_{n})$ of ${\Bbb F}_{n}$ $(1\leq n<\infty$). We introduce a new comultiplication $\Delta_{\varphi}$ on $\bigoplus C^{*}({\Bbb F}_{n})$ such that $(\bigoplus C^{*}({\Bbb F}_{n}),\,\Delta_{\varphi})$ is a non-cocommutative C$^{*}$-bialgebra. With respect to $\Delta_{\varphi}$, the tensor product $\pi\otimes_{\varphi}\pi'$ of any two representations $\pi$ and $\pi'$ of free groups is defined. The operation $\ptimes$ is associative and non-commutative. We compute its tensor product formulas of several representations. | math.OA | math |
Non-cocommutative C∗-bialgebra defined as the
direct sum of free group C∗-algebras ∗
Katsunori Kawamura†
College of Science and Engineering, Ritsumeikan University,
1-1-1 Noji Higashi, Kusatsu, Shiga 525-8577, Japan
Abstract
Let F
n be the free group of rank n and let L C ∗(F
n) denote
n (1 ≤ n <
the direct sum of full group C∗-algebras C ∗(F
n) of F
∞). We introduce a new comultiplication ∆ϕ on L C ∗(F
n) such that
(L C ∗(F
n), ∆ϕ) is a non-cocommutative C∗-bialgebra. With respect
to ∆ϕ, the tensor product π ⊗ϕ π′ of any two representations π and
π′ of free groups is defined. The operation ⊗ϕ is associative and non-
commutative. We compute its tensor product formulas of several rep-
resentations.
Mathematics Subject Classifications (2010). 46K10, 16T10.
Key words.
regular representation
free group C∗-algebra; C∗-bialgebra; tensor product; quasi-
1
Introduction
A C∗-bialgebra is a generalization of bialgebra in the theory of C∗-algebras,
which was introduced in C∗-algebraic framework for quantum groups [15,
18]. For example, if G is a locally compact group, then the full group C∗-
algebra C∗(G) of G is a cocommutative C∗-bialgebra with respect to the
standard (diagonal) comultiplication.
In this paper, a C∗-bialgebra arising from certain group homomor-
phisms among free groups is given as follows: Let Fn denote the free group
of rank n with free generators g(n)
n . For n, m ≥ 1, define the group
1 , . . . , g(n)
∗This is the revision of the previous version.
†e-mail: [email protected].
1
homomorphism φn,m from Fnm to Fn × Fm by
φn,m(g(nm)
m(i−1)+j ) := (g(n)
i
, g(m)
j
)
(i = 1, . . . , n, j = 1, . . . , m).
(1.1)
The map φn,m is well-defined on the whole of Fnm by the universality of
Fnm. Then the following diagram is commutative for each n, m, l ≥ 1:
Figure 1.1
φn,ml
✶
Fn × Fml
idn × φm,l
Fnml
q
✶
Fn × Fm × Fl.
φnm,l
q
Fnm × Fl
φn,m × idl
Group homomorphisms in (1.1) can be lifted as ∗-homomorphisms ϕn,m
among full group C∗-algebras and their minimal tensors. For {ϕn,m}, the
following diagram is also commutative for each n, m, l ≥ 1:
Figure 1.2
C∗(Fn) ⊗ C∗(Fml)
ϕn,ml
✯
idn ⊗ ϕm,l
C∗(Fnml)
❥
✯
C∗(Fn) ⊗ C∗(Fm) ⊗ C∗(Fl).
ϕnm,l
❥
ϕn,m ⊗ idl
C∗(Fnm) ⊗ C∗(Fl)
By using {ϕn,m}, we can construct a new comultiplication ∆ϕ on the
direct sum
M C∗(Fn) = C∗(F1) ⊕ C∗(F2) ⊕ C∗(F3) ⊕ · · ·
(1.2)
for all finite-rank free groups {Fn : 1 ≤ n < ∞} such that (L C∗(Fn), ∆ϕ)
is a non-cocommutative C∗-bialgebra without antipode (Theorem 1.6).
For any two unitary representations of free groups, we can define the
tensor product ⊗ϕ by using the comultiplication ∆ϕ which is not commuta-
tive (Fact 3.1). Especially, the ⊗ϕ-tensor product of any two quasi-regular
2
representations is a direct sum of quasi-regular representations (Theorem
3.2):
λFn/H ′ ⊗ϕ λFm/H ′′ ∼= M
λFnm/Hµ.
(1.3)
µ
In this section, we show our motivation, definitions and the main the-
orem.
1.1 Motivation
According to [9], given two representations of a group G, their tensor prod-
uct is a new representation of G, which decomposes into a direct sum of
indecomposable representations. The problem of finding this decomposition
is called the Clebsch-Gordan problem and the resulting formula for the de-
composition is called the tensor product formula (or Clebsch-Gordan formula
[9]). A generalization of the Clebsch-Gordan problem for groups is to con-
sider modules over associative algebras instead of group algebras. However,
there lies an obvious obstruction in that there is no known way to define the
tensor product of two left modules over an arbitrary associative algebra. For
group algebras, the extra structure coming from the group yields the tensor
product. For a bialgebra A, the associative tensor product of representa-
tions (=special modules) of A can be defined by using the comultiplicatoin.
Hence one of most important motivations of the study of bialgebras is the
tensor product of their representations.
We have studied a new kind of C∗-bialgebras which are defined as
direct sums of well-known C∗-algebras, for example, Cuntz algebras, UHF
algebras, matrix algebras [12] and Cuntz-Krieger algebras [13]. They are
non-commutative and non-cocommutative, and there never exist antipodes
on them. Such bialgebra structures do not appear before one takes direct
sums. With respect to their comultiplications, new tensor products among
representations of these C∗-algebras and their tensor product formulas were
In [12], we gave a general method to construct a C∗-
obtained [11, 14].
bialgebra from a given system of C∗-algebras and special ∗-homomorphisms
among them. The essential part of this construction is how to construct
such ∗-homomorphisms for each concrete example. One of our interests is
to construct new examples of C∗-bialgebra from various C∗-algebras.
On the other hand, group C∗-algebras are important examples of C∗-
algebras [4, 6, 19]. Furthermore, quantum groups in the C∗-algebra approach
are founded on the study of group C∗-algebras [15, 18].
Hence we consider to construct a new C∗-bialgebra associated with
group C∗-algebras by using a new comultiplication instead of their standard
3
comultiplications. In this paper, we choose free group C∗-algebras for this
purpose, and try to construct a new comultiplication on them according to
our method [12].
1.2 C∗-bialgebra
In this subsection, we review terminology about C∗-bialgebra according to
[7, 15, 18]. For two C∗-algebras A and B, we write Hom(A, B) as the set of
all ∗-homomorphisms from A to B. We assume that every tensor product
⊗ as below means the minimal C∗-tensor product.
Definition 1.3 A pair (A, ∆) is a C∗-bialgebra if A is a C∗-algebra and
∆ ∈ Hom(A, M (A ⊗ A)), where M (A ⊗ A) denotes the multiplier algebra of
A ⊗ A, such that the linear span of {∆(a)(b ⊗ c) : a, b, c ∈ A} is norm dense
in A ⊗ A and the following holds:
(∆ ⊗ id) ◦ ∆ = (id ⊗ ∆) ◦ ∆.
(1.4)
We call ∆ the comultiplication of A.
We say that a C∗-bialgebra (A, ∆) is strictly proper if ∆(a) ∈ A ⊗ A for any
a ∈ A; (A, ∆) is unital if A is unital and ∆ is unital; (A, ∆) is counital if
there exists ε ∈ Hom(A, C) such that
(ε ⊗ id) ◦ ∆ = id = (id ⊗ ε) ◦ ∆.
(1.5)
We call ε the counit of A and write (A, ∆, ε) as the counital C∗-bialgebra
(A, ∆) with the counit ε. Remark that Definition 1.3 does not mean ∆(A) ⊂
A⊗A. If A is unital, then (A, ∆) is strictly proper. A bialgebra in the purely
algebraic theory [1, 10] means a unital counital strictly proper bialgebra with
the unital counit with respect to the algebraic tensor product, which does
not need to have an involution. Hence a C∗-bialgebra is not a bialgebra in
general. In Definition 1.3, if A is unital and ∆ is unital, then the condition
of the dense subspace in A ⊗ A can be omitted.
According to [12], we recall several notions of C∗-bialgebra.
Definition 1.4
(i) For two C∗-bialgebras (A1, ∆1) and (A2, ∆2), f is
a C∗-bialgebra morphism from (A1, ∆1) to (A2, ∆2) if f is a non-
degenerate ∗-homomorphism from A1 to M (A2) such that (f ⊗f )◦∆1 =
∆2 ◦ f . In addition, if f (A1) ⊂ A2, then f is called strictly proper.
(ii) A map f is a C∗-bialgebra endomorphism of a C∗-bialgebra (A, ∆) if f
is a C∗-bialgebra morphism from A to A. In addition, if f (A) ⊂ A and
f is bijective, then f is called a C∗-bialgebra automorphism of (A, ∆).
4
(iii) A pair (B, Γ) is a right comodule-C∗-algebra of a C∗-bialgebra (A, ∆)
if B is a C∗-algebra and Γ is a non-degenerate ∗-homomorphism from
B to M (B ⊗ A) such that the following holds:
(Γ ⊗ id) ◦ Γ = (id ⊗ ∆) ◦ Γ
(1.6)
where both Γ ⊗ id and id ⊗ ∆ are extended to unital ∗-homomorphisms
from M (B ⊗ A) to M (B ⊗ A ⊗ A). The map Γ is called the right
coaction of A on B.
(iv) A proper C∗-bialgebra (A, ∆) satisfies the cancellation law if ∆(A)(I ⊗
A) and ∆(A)(A ⊗ I) are dense in A ⊗ A where ∆(A)(I ⊗ A) and
∆(A)(A ⊗ I) denote the linear spans of sets {∆(a)(I ⊗ b) : a, b ∈ A}
and {∆(a)(b ⊗ I) : a, b ∈ A}, respectively.
Let (B, m, η, ∆, ε) be a bialgebra in the purely algebraic theory, where
m is a multiplication and η is a unit of the algebra B. An endomorphism S
of B is called an antipode for (B, m, η, ∆, ε) if S satisfies m ◦ (id ⊗ S) ◦ ∆ =
η ◦ ε = m ◦ (S ⊗ id) ◦ ∆ [1, 10].
1.3 Free group algebras and homomorphisms among them
In this subsection, we briefly review free group C∗-algebras [4, 6], and intro-
duce new homomorphisms among them in order to define a comultiplication.
For n = ∞, 1, 2, 3, . . ., let Fn denote the free group of rank n where we
use the symbol "∞" as the countable infinity for convenience in this paper.
Let (Kn, ηn) denote a direct sum of all irreducible representations (up to
unitary equivalence) of the Banach algebra ℓ1(Fn). Let C∗(Fn) denote the
full group C∗-algebra of Fn, which is defined as the C∗-algebra generated
by the image of ℓ1(Fn) by ηn. Remark that C∗(F1) is ∗-isomorphic to the
C∗-algebra C(T) of all complex-valued continuous functions on the torus T.
With respect to the natural identification of the group algebra CFn over the
coefficient field C with a subalgebra of C∗(Fn), CFn is dense in C∗(Fn). For
n = ∞, 1, 2, 3, . . ., let {g(n)
i } be the free generators of Fn. We also identify
g(n)
i with the unitary ηn(g(n)
) in C∗(Fn).
i
We introduce ∗-homomorphisms among C∗(Fn)'s as follows.
Lemma 1.5
(i) For 1 ≤ n, m < ∞, define the map ϕn,m from C∗(Fnm)
to the minimal tensor product C∗(Fn) ⊗ C∗(Fm) by
ϕn,m(g(nm)
m(i−1)+j ) := g(n)
i ⊗ g(m)
j
(i = 1, . . . , n, j = 1, . . . , m).
(1.7)
Then it is well-defined on the whole of C∗(Fnm) as a unital ∗-homomorphism.
5
(ii) For 1 ≤ n < ∞, define the map ϕ∞,n from C∗(F∞) to the minimal
tensor product C∗(F∞) ⊗ C∗(Fn) by
i ⊗ g(n)
n(i−1)+j ) := g(∞)
ϕ∞,n(g(∞)
(i ≥ 1, j = 1, . . . , n).
(1.8)
j
Then it is well-defined on the whole of C∗(F∞) as a unital ∗-homomorphism.
(iii) If n, m ≥ 2, then ϕn,m is not injective.
(iv) Let C∗r (Fn) denote the reduced group C∗-algebra of Fn, which is defined
as the C∗-algebra generated by the image of the left regular represen-
tation of Fn. Then the map ϕn,m in (1.7) can not be extended as a
∗-homomorphism from C∗r (Fnm) to C∗r (Fn) ⊗ C∗r (Fm).
Especially, ϕ1,1 equals the standard comultiplication of C∗(F1). The proof
of Lemma 1.5 will be given in § 2.2.
1.4 Main theorem
In this subsection, we show our main theorem. Let C∗(Fn), {g(n)
{ϕn,m}n,m≥1 and {ϕ∞,n}n≥1 be as in § 1.3.
Theorem 1.6 Define the C∗-algebra A as the direct sum
i }n
i=1, CFn,
A := M
1≤n<∞
C∗(Fn)
(1.9)
and define ∆ϕ ∈ Hom(A, A ⊗ A) and ε ∈ Hom(A, C) by
∆ϕ(x) := X
ϕm,l(x) when x ∈ C∗(Fn),
(1.10)
m,l; ml=n
(1.11)
where ε1 ∈ Hom(C∗(F1), C) is defined as ε1F1 = 1, and E1 is the projection
from A onto C∗(F1). Then the following holds:
ε := ε1 ◦ E1
(i) The C∗-algebra A is a strictly proper counital C∗-bialgebra with the
comultiplication ∆ϕ and the counit ε.
(ii) The C∗-bialgebra (A, ∆ϕ) satisfies the cancellation law.
(iii) By the smallest unitization, (A, ∆ϕ, ε) can be extended to the unital
counital C∗-bialgebra ( A, ∆ϕ, ε).
6
(iv) There never exists any antipode for any dense unital counital subbial-
gebra of ( A, ∆ϕ, ε) in (iii).
(v) Define the algebraic direct sum CF∗ := ⊕alg{CFn : 1 ≤ n < ∞}.
Then ∆ϕ(CF∗) ⊂ CF∗ ⊙ CF∗ where ⊙ means the algebraic tensor
product, and CF∗ is identified with a ∗-subalgebra of A with respect
to the canonical embedding.
(vi) Define Γϕ ∈ Hom(C∗(F∞), M (C∗(F∞) ⊗ A)) by
Γϕ(x) := Y
1≤n<∞
ϕ∞,n(x)
(x ∈ C∗(F∞)),
(1.12)
where we identify the multiplier M (C∗(F∞)⊗A) with the direct product
Qn≥1 C∗(F∞) ⊗ C∗(Fn). Then C∗(F∞) is a right comodule-C∗-algebra
of (A, ∆ϕ) with respect to the coaction Γϕ.
Remark 1.7
C∗(Fn).
(i) The R.H.S. in (1.10) is always a finite sum when x ∈
(ii) The C∗-bialgebra (A, ∆ϕ) is non-cocommutative. In fact, the following
holds:
∆ϕ(g(6)
2 ) = g(1)
1 ⊗ g(6)
1 . (1.13)
2 + g(2)
1 ⊗ g(3)
2 + g(3)
1 ⊗ g(2)
2 + g(6)
2 ⊗ g(1)
(iii) In (1.9), every free group C∗-algebras C∗(Fn) (1 ≤ n < ∞) appear at
once. This is an essentially new structure of the class of free group
C∗-algebras. On the other hand, C∗(F∞) appears as a comodule-C∗-
algebra of (A, ∆ϕ). This shows a certain naturality of this bialgebra
structure.
In § 2, we prove Theorem 1.6. In § 3, we show tensor product formulas
of representations of Fn's with respect to ∆ϕ, and show some C∗-bialgebra
automorphisms.
2 Proofs of theorems
In this section, we prove Lemma 1.5 and Theorem 1.6.
7
2.1 C∗-weakly coassociative system
According to § 3 in [12], we recall a general method to construct a C∗-
bialgebra from a set of C∗-algebras and ∗-homomorphisms among them. A
monoid is a set M equipped with a binary associative operation M × M ∋
(a, b) 7→ ab ∈ M, and a unit with respect to the operation. For example, N =
{1, 2, 3, . . .} is an abelian monoid with respect to the multiplication. In order
to show Theorem 1.6, we give a new definition of C∗-weakly coassociative
system which is a generalization of the older in Definition 3.1 of [12].
Definition 2.1 Let M be a monoid with the unit e. A data {(Aa, ϕa,b) :
a, b ∈ M} is a C∗-weakly coassociative system (= C∗-WCS) over M if Aa is
a unital C∗-algebra for a ∈ M and ϕa,b is a unital ∗-homomorphism from
Aab to Aa ⊗ Ab for a, b ∈ M such that
(i) for all a, b, c ∈ M, the following holds:
(ida ⊗ ϕb,c) ◦ ϕa,bc = (ϕa,b ⊗ idc) ◦ ϕab,c
(2.1)
where idx denotes the identity map on Ax for x = a, c,
(ii) there exists a counit εe of Ae such that (Ae, ϕe,e, εe) is a counital C∗-
bialgebra,
(iii) for each a ∈ M, the following holds:
(εe ⊗ ida) ◦ ϕe,a = ida = (ida ⊗ εe) ◦ ϕa,e.
(2.2)
The condition (2.2) is weaker than the older, "ϕe,a(x) = Ie ⊗x and ϕa,e(x) =
x ⊗ Ie for x ∈ Aa and a ∈ M" ([12], Definition 3.1) .
In fact, the older
definition satisfies (2.2). From the new definition, the same result holds as
follows.
Theorem 2.2 ([12], Theorem 3.1). Let {(Aa, ϕa,b) : a, b ∈ M} be a C∗-
WCS over a monoid M. Assume that M satisfies that
#Na < ∞ for each a ∈ M
(2.3)
where Na := {(b, c) ∈ M × M : bc = a}. Define C∗-algebras
A∗ := ⊕{Aa : a ∈ M}, Ca := ⊕{Ab ⊗ Ac : (b, c) ∈ Na}
(a ∈ M).
8
Define ∆(a)
by
ϕ ∈ Hom(Aa, Ca), ∆ϕ ∈ Hom(A∗, A∗ ⊗ A∗) and ε ∈ Hom(A∗, C)
∆(a)
ϕ (x) := X
ϕb,c(x)
(x ∈ Aa), ∆ϕ := ⊕{∆(a)
ϕ : a ∈ M},
(b,c)∈Na
ε := εe ◦ Ee
(2.4)
where Ee denotes the projection from A∗ onto Ae. Then (A∗, ∆ϕ, ε) is a
strictly proper counital C∗-bialgebra.
Proof. By (2.3), ∆(a)
ϕ is
unital for each a. Since M × M = `a∈M Na, A∗ ⊗ A∗ = ⊕{Af ⊗ Ag : f, g ∈
M} = ⊕{Ca : a ∈ M}. Since ∆(a)
ϕ is unital for each a, ∆ϕ is non-degenerate.
From (2.1), the following holds for x ∈ Aa:
ϕ is well-defined. Furthermore, Ca is unital and ∆(a)
{(∆ϕ ⊗ id) ◦ ∆ϕ}(x) = Pb,c,d∈M, bcd=a(ϕb,c ⊗ idd)(ϕbc,d(x))
= Pb,c,d∈M, bcd=a(idb ⊗ ϕc,d)(ϕb,cd(x))
= {(id ⊗ ∆ϕ) ◦ ∆ϕ}(x).
(2.5)
Hence (∆ϕ ⊗ id) ◦ ∆ϕ = (id ⊗ ∆ϕ) ◦ ∆ϕ on A∗. Therefore ∆ϕ is a comulti-
plication of A∗. On the other hand, for x ∈ Aa, we see that
{(ε ⊗ id) ◦ ∆ϕ}(x) = (ε ⊗ id)(∆(a)
ϕ (x))
(ε ⊗ id)(ϕb,c(x))
= P(b,c)∈Na
= (εe ⊗ ida)(ϕe,a(x))
= x
(from (2.2)).
(2.6)
In like wise, we see that (id ⊗ ε) ◦ ∆ϕ = id.
Hence (ε ⊗ id) ◦ ∆ϕ = id.
Therefore ε is a counit of (A∗, ∆ϕ). In consequence, we see that (A∗, ∆ϕ, ε)
is a counital C∗-bialgebra. By definition, (A∗, ∆ϕ) is strictly proper.
We call (A∗, ∆ϕ, ε) in Theorem 2.2 by a (counital) C∗-bialgebra associated
with {(Aa, ϕa,b) : a, b ∈ M}.
The following lemma holds independently of the generalization in Def-
inition 2.1(iii).
Lemma 2.3 For the following C∗-WCS {(Aa, ϕa,b) : a, b ∈ M}, we assume
the condition (2.3).
(i) ([12], Lemma 2.2) For a given strictly proper non-unital counital C∗-
bialgebra (A, ∆, ε), let A := A ⊕ C denote the smallest unitization of
A. Then there exist a unique extension ( ∆, ε) of (∆, ε) on A such that
( A, ∆, ε) is a strictly proper unital counital C∗-bialgebra.
9
(ii) ([12], Lemma 3.2) For a C∗-WCS {(Aa, ϕa,b) : a, b ∈ M} over M, let
(A∗, ∆ϕ, ε) be as in Theorem 2.2 and let ( A∗, ∆ϕ, ε) be the smallest
unitization of (A∗, ∆ϕ, ε) in (i). Assume that any element in M has
no left inverse except the unit e. Then the antipode for any dense
unital counital subbialgebra of ( A∗, ∆ϕ, ε) never exists.
(iii) ([12], Lemma 3.1) Let {(Aa, ϕa,b) : a, b ∈ M} be a C∗-WCS over
a monoid M and let (A∗, ∆ϕ) be as in Theorem 2.2 associated with
{(Aa, ϕa,b) : a, b ∈ M}. Define
Ya,b := ϕa,b(Aab)(Ia ⊗ Ab)
Xa,b := ϕa,b(Aab)(Aa ⊗ Ib),
(a, b ∈ M)
(2.7)
where ϕa,b(Aab)(Aa ⊗ Ib) and ϕa,b(Aab)(Ia ⊗ Ab) mean the linear spans
of {ϕa,b(x)(y ⊗ Ib) : x ∈ Aab, y ∈ Aa} and {ϕa,b(x)(Ia ⊗ y) : x ∈
Aab, y ∈ Ab}, respectively. If both Xa,b and Ya,b are dense in Aa ⊗ Ab
for each a, b ∈ M, then (A∗, ∆ϕ) satisfies the cancellation law.
(iv) ([12], Theorem 3.2) For a C∗-WCS {(Aa, ϕa,b) : a, b ∈ M} over a
monoid M, assume that B is a unital C∗-algebra and a set {ϕB,a : a ∈
M} of unital ∗-homomorphisms such that ϕB,a ∈ Hom(B, B ⊗ Aa) for
each a ∈ M and the following holds:
(ϕB,a ⊗ idb) ◦ ϕB,b = (idB ⊗ ϕa,b) ◦ ϕB,ab
(a, b ∈ M).
(2.8)
Then B is a right comodule-C∗-algebra of the C∗-bialgebra (A∗, ∆ϕ)
with the unital coaction Γϕ := Qa∈M ϕB,a.
2.2 Homomorphisms among free groups
In this subsection, we show properties of φn,m in (1.1) and prove Lemma
1.5.
Lemma 2.4 For each n ≥ 1, we write 1 as the unit of Fn.
(i) For any x ∈ Fn, there exists (y, z) ∈ Fm × Fnm such that φn,m(z) =
(x, y).
(ii) For any y ∈ Fm, there exists (x, z) ∈ Fn × Fnm such that φn,m(z) =
(x, y).
(iii) For any (x, y) ∈ Fn × Fm, there exists (x′, z) ∈ Fn × Fnm such that
φn,m(z)(x′, 1) = (x, y).
10
(iv) For any (x, y) ∈ Fn × Fm, there exists (y′, z) ∈ Fm × Fnm such that
φn,m(z)(1, y′) = (x, y).
(v) When n, m ≥ 2, φn,m is not injective.
· · · aεl
il
1 · · · bεl
1 and z := cε1
m(i1−1)+1 · · · cεl
Proof.
(i) Let a1, . . . , an, b1, . . . , bm, c1, . . . , cnm be the free generators of
Fn, Fm, Fnm, respectively. Assume that x ∈ Fn is written as a reduced word
x = aε1
where εi = 1 or −1 for i = 1, . . . , l. For example, define
i1
(y, z) ∈ Fm × Fnm by y := bε1
m(il−1)+1. Then
y belongs to the abelian subgroup generated by the single element b1, and
it is not always a reduced word in Fm. Then the statement holds for (y, z).
(ii) As the proof of (i), this is proved.
(iii) From (ii), we can find (x′′, z) ∈ Fn × Fnm such that φn,m(z) = (x′′, y).
Define x′ := (x′′)−1x, then the statement holds.
(iv) As the proof of (iii), this is proved from (i).
(v) Let c1, . . . , cnm be as in the proof of (i). For i, l ∈ {1, . . . , n}, k, j ∈
{1, . . . , m}, define x(i, l; j, k) ∈ Fnm by
x(i, l; j, k) := cm(i−1)+j c−1
m(i−1)+k cm(l−1)+k c−1
m(l−1)+j .
(2.9)
Then x(i, l; j, k) 6= 1 when k 6= j, i 6= l, but x(i, l; j, k) ∈ ker φn,m for any
i, l, j, k.
In the proof of Lemma 2.4(v),
c1c−1
in F4 satisfies φ2,2(c1c−1
2 c4c−1
3
if n = m = 2, then the reduced word
2 c4c−1
3 ) = (1, 1).
n,m of Fnm on Kn ⊗ Km by ϕ0
n,m is included in C∗(Fn) ⊗ C∗(Fm), ϕ0
Proof of Lemma 1.5 (i) Let φn,m be as in (1.1) and let (Kn, ηn) be as in §
1.3. Define the unitary representation ϕ0
n,m :=
(ηn ⊗ ηm) ◦ φn,m. The representation ϕ0
n,m is well-defined by the universality
of Fnm. Since the image of ϕ0
n,m is
uniquely extended to ϕn,m in (1.7) such that ϕn,m(ηnm(x)) = ϕ0
n,m(x) for
each x ∈ Fnm ([4], Proposition 2.5.2). Hence the statement holds.
(ii) In analogy with (i), the statement holds.
(iii) For x(i, l; j, k) in (2.9), we see that ϕn,m(x(i, l; j, k) − 1) = 0 for each
i, l, j, k. Hence the statement holds.
(iv) If such an extension ϕn,m of ϕn,m exists, then ϕn,m must be injective
because C∗r (Fnm) is simple when nm ≥ 2 ([6], Corollary VII.7.5 and its
proof). On the other hand, ϕn,m never be injective from (iii).
11
2.3 Proof of Theorem 1.6
We prove Theorem 1.6 in this subsection. Let N := {1, 2, 3, . . .}. Remark
that (2.3) holds for any element in the multiplicative monoid (N, ·).
(i) From Theorem 2.2, it is sufficient to show that {(C∗(Fn), ϕn,m) : n, m ∈
N} is a C∗-WCS over the monoid N. By the definition of ϕn,m in (1.7), we
can verify that (ϕn,m ⊗ idl) ◦ ϕnm,l = (idn ⊗ ϕm,l) ◦ ϕn,ml for n, m, l ∈ N
where ida denotes the identity map on C∗(Fa) for a = n, l. Hence (2.1) is
satisfied. On the other hand, since ε1F1 = 1, {(ε1 ⊗ idn) ◦ ϕ1,n}(g(n)
) =
(ε1 ⊗ idn)(g(1)
for each j = 1, . . . , n and n ∈ N.
By the same token, we obtain (idn ⊗ ε1) ◦ ϕn,1 = idn. Hence (2.2) is verified.
Therefore {(C∗(Fn), ϕn,m) : n, m ∈ N} is a C∗-WCS over the monoid N.
(ii) For n, m ∈ N, define three subsets Pn,m, Qn,m, Rn,m of C∗(Fn) ⊗ C∗(Fm)
by
) = ε1(g(1)
1 ⊗ g(n)
j
j
1 ) g(n)
j = g(n)
j
Pn,m := {ϕn,m(z)(x ⊗ Im) : x ∈ Fn, z ∈ Fnm},
(2.10)
Qn,m := {ϕn,m(z)(In ⊗ y) : y ∈ Fm, z ∈ Fnm},
(2.11)
Rn,m :=
{x ⊗ y : x ∈ Fn, y ∈ Fm}.
(2.12)
Then their linear spans are dense subspaces of ϕn,m(C∗(Fnm))(C∗(Fn)⊗Im),
ϕn,m(C∗(Fnm))(In ⊗ C∗(Fm)) and C∗(Fn) ⊗ C∗(Fm), respectively. From
Lemma 2.4(iii), it is sufficient to show that Rn,m ⊂ Pn,m and Rn,m ⊂ Qn,m.
We prove Rn,m ⊂ Pn,m as follows: For (x, y) ∈ Fn × Fm, there exists
(x′, z) ∈ Fn × Fnm such that φn,m(z)(x′, 1) = (x, y) from Lemma 2.4(iii).
By definitions of φn,m and ϕn,m, this implies ϕn,m(z)(x′ ⊗ Im) = x ⊗ y.
Therefore Rn,m ⊂ Pn,m.
In a similar fashion, we obtain Rn,m ⊂ Qn,m from Lemma 2.4(iv).
Hence the statement holds.
(iii) From (i) and Lemma 2.3(i), the statement holds.
(iv) Remark that the monoid N has no left invertible element except the
unit 1. From the proof of (i) and Lemma 2.3(ii), the statement holds.
(v) From the proof of (i) and the definition of ϕn,m in (1.7), we see that
ϕn,m(CFnm) ⊂ CFn ⊙ CFm. This implies the statement.
(vi) By definition, we see that (ϕ∞,n ⊗ idm) ◦ ϕ∞,m = (id∞ ⊗ ϕn,m) ◦ ϕ∞,nm
for n, m ∈ N. From Lemma 2.3(iv) for ϕC ∗(F∞),n := ϕ∞,n, the statement
holds.
12
3 Tensor product formulas of representations, and
automorphisms
In this section, we show tensor product formulas of unitary representations
of Fn's with respect to the comultiplication ∆ϕ in Theorem 1.6, and C∗-
bialgebra automorphisms.
3.1 General facts about representations
We introduce a new tensor product of representations of Fn's and show
tensor product formulas of quasi-regular representations.
3.1.1 Representations of Fn
We identify Fn with the unitary subgroup of C∗(Fn) with respect to the
canonical embedding. Let Repu
Fn denote the class of all unitary represen-
tations of Fn. For (π, π′) ∈ Repu
Fn ×Repu
Fm, define the new representation
π ⊗ϕ π′ ∈ Repu
Fnm by
π ⊗ϕ π′ := (π ⊗ π′) ◦ ϕn,m,
(3.1)
where ϕn,m is as in (1.7). Then we see that the new operation ⊗ϕ is asso-
ciative, and it is distributive with respect to the direct sum. Furthermore,
⊗ϕ is well-defined on the unitary equivalence classes of representations. It
will be shown that ⊗ϕ is non-commutative in Fact 3.1.
3.1.2 Tensor product formulas of 1-dimensional unitary repre-
sentations of free groups
In this subsection, we show tensor product formulas of 1-dimensional unitary
representations of free groups with respect to ⊗ϕ in (3.1) as a basic example
of tensor product formula. For groups G and H, let Hom(G, H) denote the
set of all homomorphisms from G to H. Define
Ch(Fn) := Hom(Fn, U (1)).
(3.2)
Since an element in Ch(Fn) can be regarded as a one-dimensional unitary
representation of Fn, we can regard π1 ⊗ϕ π2 as an element in Ch(Fnm) for
π1 ∈ Ch(Fn) and π2 ∈ Ch(Fm).
Let g1, . . . , gn be free generators of Fn. For z = (z1, . . . , zn) ∈ Tn :=
U (1)n, define χ(n)
z ∈ Ch(Fn) by
χ(n)
z (gi) := zi
(i = 1, . . . , n).
(3.3)
13
Clearly, χ(n)
unitarily equivalent if and only if z = w. By the correspondence
is irreducible for any z ∈ Tn. For z, w ∈ Tn, χ(n)
z
and χ(n)
w are
z
Tn ∋ z 7→ χ(n)
z ∈ Ch(Fn),
(3.4)
we see that Ch(Fn) is equivalent to Tn as a set.
For z ∈ Tn and w ∈ Tm, define the Kronecker product z ⊠ w ∈ Tnm as
(z ⊠ w)m(i−1)+j := ziwj for (i, j) ∈ {1, . . . , n} × {1, . . . , m}. By definition,
the following holds:
χ(n)
z ⊗ϕ χ(m)
w = χ(nm)
z⊠w
(z ∈ Tn, w ∈ Tm).
(3.5)
This implies that the correspondence in (3.4) gives a semigroup isomorphism
Tn, ⊠). This shows a naturality of the
from (Sn≥1 Ch(Fn), ⊗ϕ) to (Sn≥1
operation ⊗ϕ.
Fact 3.1 The operation ⊗ϕ in (3.1) is non-commutative as the following
sense: There exist n, m ≥ 2 and representations π1, π2 of Fn and Fm, re-
spectively such that π1 ⊗ϕ π2 and π2 ⊗ϕ π1 are not unitarily equivalent.
In (3.5), let (n, m) = (2, 3) and z = (1, −1) ∈ T2 and w = (1, 1, 1) ∈
because z ⊠ w =
Proof.
T3. Then χ(2)
(1, 1, 1, −1, −1, −1) and w ⊠ z = (1, −1, 1, −1, 1, −1).
w⊠z = χ(3)
z⊠w 6∼= χ(6)
z ⊗ϕ χ(3)
w = χ(6)
w ⊗ϕ χ(2)
z
3.1.3 Quasi-regular representations
In this subsection, we review quasi-regular representations of discrete groups,
and show the general formula of the ⊗ϕ-tensor product of quasi-regular
representations of free groups.
For a discrete group Γ and a subgroup Γ0, let Γ/Γ0 denote the left
coset space, that is, Γ/Γ0 := {xΓ0 : x ∈ Γ}. Define the quasi-regular
representation [8] (or permutation representation [16]) (ℓ2(Γ/Γ0), λΓ/Γ0 ) of
Γ associated with Γ0 as the natural (unitary) left action of Γ on the standard
basis of ℓ2(Γ/Γ0):
λΓ/Γ0 : Γ y ℓ2(Γ/Γ0).
(3.6)
Especially, the regular representation λ and the trivial representation 1 of Γ
are quasi-regular representations associated with subgroups {e} and Γ of Γ,
respectively. For the trivial representation 1Γ0 of Γ0, λΓ/Γ0 coincides with
the induced representation IndΓ
Γ0(1Γ0 ).
14
Since Γ/Γ0 is a Γ-homogeneous space, λΓ/Γ0 is a cyclic representation.
When Γ acts on a set X, the permutation representation of Γ on ℓ2(X) are
decomposed into the direct sum of quasi-regular representations as follows:
ℓ2(X) ∼= M
ℓ2(Γ/Hµ)
µ
(3.7)
where Hµ is a subgroup of Γ such that Xµ ∼= Γ/Hµ for the orbit decomposi-
tion X = `µ Xµ with respect to the Γ-action. About the irreducibility and
unitary equivalence of quasi-regular representations, see Appendix A.
Next, we consider the tensor product ⊗ϕ in (3.1) for quasi-regular
representations of free groups.
Theorem 3.2 Let H′, H′′ be subgroups of Fn and Fm, respectively. For
φn,m in (1.1), define the left action φn,m of Fnm on the direct product set
X := Fn/H′ × Fm/H′′ by
φn,m(g)(xH′, yH′′) := (g′xH′, g′′yH′′)
( (xH′, yH′′) ∈ X, g ∈ Fnm ) (3.8)
where (g′, g′′) := φn,m(g). With respect to the action φn,m, let X = `µ Xµ
be the orbit decomposition and choose Hµ as a stabilizer subgroup of Fnm
associated with Xµ. Then the following holds:
λFn/H ′ ⊗ϕ λFm/H ′′ ∼= M
λFnm/Hµ.
(3.9)
µ
Proof. Let {ξ′a : a ∈ Fn/H′}, {ξ′′b : b ∈ Fnm/H′′} denote standard bases of
ℓ2(Fn/H′) and ℓ2(Fnm/H′′), respectively. By definition, (λFn/H ′⊗ϕλFm/H ′′)(g)(ξ′a⊗
ξ′′b ) = ξ′g′a ⊗ ξ′′g′′b for g ∈ Fnm) where (g′, g′′) := φn,m(g). On the other hand,
the permutation representation (ℓ2(X), L) of Fnm by φn,m satisfies Lgξ(a,b) =
ξ φn,m(g)(a,b) = ξ(g′a, g′′b) where {ξ(a,b) : (a, b) ∈ X} denotes the standard
basis of ℓ2(X). By the natural identification of ℓ2(X) with ℓ2(Fn/H′) ⊗
ℓ2(Fm/H′′), we see that L and λFn/H ′ ⊗ϕ λFm/H ′′ are unitarily equivalent.
By definition, Xµ ∼= Fnm/Hµ as Fnm-homogeneous spaces, and the
statement holds from the decomposition in (3.7).
Theorem 3.2 states that the ⊗ϕ-tensor product of any two quasi-regular
representations is decomposed into the direct sum of quasi-regular represen-
tations. That is, the category of direct sums of quasi-regular representations
of free groups is closed with respect to the ⊗ϕ-tensor product. This shows a
naturality of the ⊗ϕ-tensor product. In § 3.2 and 3.3, we will show concrete
examples of the formula (3.9).
15
3.2 Tensor product of some irreducible representations
In this subsection, we show tensor product formulas of some irreducible
quasi-regular representations of Fn's as examples of Theorem 3.2.
We review some irreducible quasi-regular representations of Fn [2, 20].
Let g1, . . . , gn be the free generators of Fn. Fix i ∈ {1, . . . , n} and let H (n)
denote the abelian subgroup of Fn generated by the single element gi:
i
H (n)
i
:= {gl
i : l ∈ Z} ⊂ Fn.
(3.10)
Proposition 3.3
(i) For any i = 1, . . . , n, λFn/H (n)
i
is irreducible.
(ii) λFn/H (n)
i
∼= λFn/H (n)
j
if and only if i = j.
Proof. See Appendix A.
We show the tensor product formula of λFn/H (n)
i
's in Proposition 3.3.
For the map φn,m in (1.1), define the subgroup Gn,m of Fnm by
Gn,m := ker φn,m.
(3.11)
From Lemma 2.4(v), Gn,m 6= {1} when n, m ≥ 2 and Gn,1 = G1,n = {1}
for any n ≥ 1. Since Gn,m is a normal subgroup of Fnm, Gn,mH := {gh :
(g, h) ∈ Gn,m × H} is also a subgroup of Fnm and Gn,mH = HGn,m for any
subgroup H of Fnm.
Theorem 3.4 Let H (n)
Define Kn,m,l := Gn,mH (nm)
i
l
for l = 1, . . . , nm.
and Gn,m be as in (3.10) and (3.11), respectively.
(i) For n, m ≥ 1 and (i, j) ∈ {1, . . . , n} × {1, . . . , m},
λFn/H (n)
i
⊗ϕ λFm/H (m)
j
∼= λFnm/Kn,m,m(i−1)+j .
(3.12)
(ii) For any l = 1, . . . , nm, λFnm/Kn,m,l is irreducible.
(iii) λFnm/Kn,m,l and λFnm/Kn,m,l′ are unitarily equivalent if and only if l =
l′.
Proof. See Appendix B.
16
3.3 Tensor product of regular representations
In this subsection, we consider regular representations of Fn's as examples
of Theorem 3.2.
We recall characterization of representations by using von Neumann
algebras. A nondegenerate representation π of a C∗-algebra A is said to be
(pure) type X if π(A)′′ is of type X for X=I, II, III, II1, II∞ [3]. For a group
G and unitary representation U of G, U is said to be type X if U (G)′′ is of
type X for X=I, II, III, II1, II∞. U is factor if the center of U (G)′′ is trivial.
In the proof of Theorem 3.2, λFn = λFn/H ′ and λFm = λFm/H ′′ when
H′ = {1} ⊂ Fn and H′′ = {1} ⊂ Fm.
Proposition 3.5 Let n, m ≥ 1.
(i) λFn ⊗ϕ λFm
∼= (λFnm/Gn,m)⊕∞ where Gn,m is as in (3.11).
(ii) If n, m ≥ 2, then λFnm/Gn,m is a type II1 factor representation.
(i) Let X := Fn × Fm and let φn,m be the action of Fnm on X in
Proof.
(3.8). From Lemma 2.4(iii), we see that the orbit decomposition is given as
follows:
Ox, Ox := { φn,m(y)(x, 1) : y ∈ Fnm}.
(3.13)
X = a
x∈Fn
Furthermore, the equivalence (Ox, φn,mOx) ∼= (Fnm/Gn,m, L) holds as Fnm-
homogeneous spaces for all x ∈ Fn where L denotes the natural left action of
Fnm on Fnm/Gn,m. Therefore the equivalence (X, φn,m) ∼= (Fnm/Gn,m, L)#Fn
of Fnm-homogeneous spaces holds. From this and the proof of Theorem 3.2,
the statement holds.
(ii) See Appendix C.
3.4 Automorphisms
In this subsection, we show examples of some C∗-bialgebra automorphism
of (A, ∆ϕ). For t ∈ R, define α(n)
t ∈ AutC∗(Fn) by
√−1t log n g(n)
i
α(n)
t
(g(n)
i
) := e
(i = 1, . . . , n).
(3.14)
Then α(∗)
t ◦ α(∗)
α(∗)
t
s = α(∗)
:= ⊕n≥1α(n)
t
t+s for s, t ∈ R.
is a C∗-bialgebra automorphism of (A, ∆ϕ) such that
17
Define β(n) ∈ AutC∗(Fn) by
β(n)(g(n)
i
) := g(n)
n−i+1
(i = 1, . . . , n).
(3.15)
Then β(∗) := ⊕n≥1β(n) is a C∗-bialgebra automorphism of (A, ∆ϕ) such that
β(∗) ◦ β(∗) = id.
The automorphism β(∗) commutes α(∗)
t
for each t. Hence these give the
action of the group R × (Z/2Z) on the C∗-bialgebra (A, ∆ϕ).
Acknowledgment: The author would like to express his sincere thanks to
reviewers for correcting errors in the previous version.
Appendix
A Applications of commensurator to quasi-regular
representations of discrete groups
Recall that two subgroups Γ0 and Γ1 of a group Γ are commensurable if
Γ0∩Γ1 is of finite index in both Γ0 and Γ1, that is, [Γ0, Γ0∩Γ1]·[Γ1, Γ0∩Γ1] <
∞, in such case, we write Γ0 ≈ Γ1. Define the commensurator ComΓ(Γ0) of
Γ0 in Γ as
ComΓ(Γ0) := {g ∈ Γ : Γ0 ≈ gΓ0g−1}.
(A.1)
Then the inclusions Γ0 ⊂ ComΓ(Γ0) ⊂ Γ of subgroups hold. According to [5],
we review applications of commensurator to quasi-regular representations of
discrete groups by Mackey [16].
Theorem A.1 ([5], Theorem 2.1) Let Γ be a discrete group and let Γ0, Γ1
be subgroups of Γ.
(i) The representation (ℓ2(Γ/Γ0), λΓ/Γ0 ) of Γ is irreducible if and only if
ComΓ(Γ0) = Γ0.
(ii) Assume ComΓ(Γi) = Γi for i = 0, 1. Then λΓ/Γ0 and λΓ/Γ1 are uni-
tarily equivalent if and only if Γ0 and Γ1 are quasiconjugate in Γ, that
is, there exists g ∈ Γ such that Γ0 ≈ gΓ1g−1.
Lemma A.2 Let H (n)
i
be as in (3.10).
(i) If a, b ∈ Fn satisfy ab = ba, then there exists w ∈ Fn such that a = wl
and b = wl′ for some l, l′ ∈ Z.
18
(ii) If a, b ∈ Fn satisfy albk = bkal for some l, k ∈ Z, then ab = ba.
(iii) If a ∈ Fn satisfies ab = ba for some b ∈ H (n)
i
and b 6= 1, then a ∈ H (n)
i
.
(iv) If i, j ∈ {1, . . . , n} and g ∈ Fn satisfy H (n)
i ≈ gH (n)
j g−1, then i = j
and g ∈ H (n)
i
.
.
i
i = wl′. Therefore ck
i wl′ = wl′ck
Proof. Let c1, . . . , cn be free generators of Fn and let Hi := H (n)
(i) See [17], p42, 6.
(ii) See [17], p41, 4.
(iii) By (i), both a and b can be written as a = wl and b = wl′
w ∈ Fn and l, l′ ∈ Z. By the choice of b, b = ck
Hence ck
we see w ∈ Hi. Therefore a = wl ∈ Hi.
(iv) If i, j and g satisfy the assumption, then [Hi, Hi ∩ gHjg−1] < ∞ by
definition. Since #Hi = ∞, there exists x ∈ Hi ∩ gHjg−1 such that
x 6= 1. Then cl
in (3.3),
i = χ(n)
j ) = zl′
z (cl′
for any z ∈ Tn. This im-
zl
j
ig−1 for some l ∈ Z. By the
plies i = j and l = l′. In consequence, cl
i = gcl
choice of x and l, l 6= 0. From (iii), g ∈ Hi.
for some
i for some k ∈ Z and k 6= 0.
i . From (ii), ciw = wci. From (i),
j g−1 for some l, l′ ∈ Z. For χ(n)
i = x = gcl′
z (gcl′
j g−1) = χ(n)
i) = χ(n)
z (cl
z
Proof of Proposition 3.3. (i) From Lemma A.2(iii), ComFn(H (n)
By Theorem A.1(i), the statement holds.
(ii) From Lemma A.2(iii), H (n)
and H (n)
i = j. From this and Theorem A.1(ii), the statement holds.
j
i
i
are quasiconjugate if and only if
) = H (n)
i
.
B Proof of Theorem 3.4
In this section, we prove Theorem 3.4. Let φn,m, H (n)
, Gn,m be as in (1.1),
(3.10) and (3.11), respectively. Since Gn,m is a normal subgroup of Fnm, we
can define the quotient group
i
Qn,m := Fnm/Gn,m.
(B.1)
Then Q1,n ∼= Qn,1 ∼= Fn for any n ≥ 1.
Let g1, . . . , gn be free generators of Fn. Define p(n) ∈ Hom(Fn, Z) by
p(n)(g) := ε1 + · · · + εk when g = gε1
i1
· · · gεk
ik
∈ Fn
(B.2)
where εi ∈ {1, −1} for i = 1, . . . , k. By definition, the following holds.
19
Fact B.1
(i) For g ∈ Fnm, if φn,m(g) = (g′, g′′), then p(nm)(g) = p(n)(g′) =
p(m)(g′′).
(ii) If g ∈ Gn,m, then p(nm)(g) = 0.
(iii) For any i = 1, . . . , n, the restriction p(n)
morphism.
: H (n)
i → Z is an iso-
H (n)
i
(iv) Let p(nm) : Qn,m → Z by p(nm)(g) := p(nm)(g) for g = gGn,m ∈ Qn,m
and define the subgroup H (nm)
l
of Qn,m by
H (nm)
l
:= {hGn,m ∈ Qn,m : h ∈ H (nm)
l
}.
(B.3)
Then p is well-defined and is a group homomorphism. Furthermore,
the restriction p(nm) H (nm)
→ Z is an isomorphism for any
l = 1, . . . , nm.
: H (nm)
l
l
Remark that H (1·n)
l
∼= H (n·1)
l
∼= H (n)
l
for any n ≥ 1.
Lemma B.2 Let n, m ≥ 1,(i, j) ∈ {1, . . . , n} × {1, . . . , m} and g ∈ Fnm.
(i) If φn,m(g) ∈ H (n)
φn,m(h) = φn,m(g).
i × H (m)
j
, then there exists h ∈ H (nm)
m(i−1)+j such that
(ii) If φn,m(g) ∈ H (n)
i × H (m)
j
, then g ∈ Gn,mH (nm)
m(i−1)+j.
(i) Let (g′, g′′) := φn,m(g). From Fact B.1(i), p(n)(g′) = p(nm)(g) =
)l by Fact B.1(iii).
Proof.
)l and g′′ = (g(m)
p(m)(g′′). When l := p(nm)(g), g′ = (g(n)
Hence h := (g(nm)
m(i−1)+j satisfies the relation.
(ii) From (i), φn,m(h−1g) = (1, 1) for some h ∈ H (nm)
m(i−1)+j Gn,m = Gn,mH (nm)
h−1g ∈ Gn,m and g ∈ hGn,m ⊂ H (nm)
m(i−1)+j . Therefore
m(i−1)+j.
m(i−1)+j )l ∈ H (nm)
j
i
Lemma B.3 For n, m ≥ 1 and (i, j) ∈ {1, . . . , n} × {1, . . . , m}, two Fnm-
homogeneous spaces (Fnm/(Gn,mH (nm)
are equivalent where L denotes the natural left action of Fnm on Fnm/K and
φn,m is as in (3.8).
m(i−1)+j ), L) and (Fn/H (n)
i ×Fm/H (m)
j
, φn,m)
20
Proof. Rewrite K := Gn,mH (nm)
H (m)
here. Define the map
j
m(i−1)+j , φ := φn,m, H′ := H (n)
i
and H′′ :=
θ : Fnm/K → Fn/H′ × Fm/H′′;
θ([x]) := ([x′], [x′′])
(B.4)
where (x′, x′′) := φ(x) and [a] denotes the coset with the representative a.
By definition, we see that the map θ is well-defined and satisfies θ ◦ Lg =
φn,m(g) ◦ θ for all g ∈ Fnm. It is sufficient to show that θ is injective and
surjective.
(i) Injectivity: For [x], [y] ∈ Fnm/K, assume θ([x]) = θ([y]). Then [x′] = [y′]
and [x′′] = [y′′]. Hence x′ = y′h′ and x′′ = y′′h′′ for some (h′, h′′) ∈ H′ × H′′.
Therefore φ(x) = (x′, x′′) = (y′h′, y′′h′′) = φ(y)(h′, h′′). Hence φ(y−1x) =
(h′, h′′). From Lemma B.2(ii), y−1x ∈ K. Therefore [x] = [y]. Hence θ is
injective.
(ii) Surjectivity: For ([x′], [x′′]) ∈ Fn/H′×Fm/H′′, there exists (w′, z) ∈ Fn×
Fnm such that φ(z)(w′, 1) = (x′, x′′) from Lemma 2.4(iii). Let g(n)
, . . . , g(n)
n
be free generators of Fn. Assume w′ = (g(n)
)εl for εi ∈ {1, −1}.
j1
Let h′′ := (g(m)
φ(z)φ(w) = φ(zw) for some w ∈ Fnm. Therefore ([x′], [x′′]) = ([x′], [x′′h′′]) =
θ([zw]). Hence θ is surjective.
)ε1 · · · (g(n)
jl
)ε1 · · · (g(m)
)εl = (g(m)
1
j
j
j
)ε1+···+εl. Then (x′, x′′h′′) = φ(z)(w′, h′′) =
Let χ(n)
z be as in (3.3) and let Tn ⊠ Tm := {z ⊠ w : (z, w) ∈ Tn × Tm}.
(gh) =
by (3.5). From this, χ(nm)
u
For u ∈ Tn ⊠ Tm, we see Gn,m ⊂ ker χ(nm)
χ(nm)
u
(g) for any g ∈ Fnm and h ∈ Gn,m. Hence
u
χ(nm)
u
: Qn,m → U (1);
χ(nm)
u
(gGn,m) := χ(nm)
u
(g)
(B.5)
is well-defined for any u ∈ Tn ⊠ Tm as a homomorphism.
Lemma B.4 Let n, m ≥ 1 and let H (nm)
and g ∈ Qn,m satisfy H (nm)
l
≈ g H (nm)
j
i
be as in (B.3). If i, j ∈ {1, . . . , nm}
g−1, then i = j and g ∈ H (nm)
.
i
i
Proof. Let Hi := H (nm)
, G := Gn,m and let c1, . . . , cnm be free generators of
Fnm. For g ∈ Fnm, let g := gG ∈ Qn,m. If i, j and g satisfy the assumption,
then [ Hi, Hi ∩ g Hj g−1] < ∞ by definition. Since # Hi = ∞, there exists
x ∈ Hi ∩ g Hj g−1 such that x 6= 1. Then cl
j g−1G for some
l, l′ ∈ Z. For χ(nm)
j g−1G) = ul′
(cl
j for
any u = (u1, . . . , unm) ∈ Tn ⊠ Tm. This implies i = j and l = l′. From
ig−1G for some
this, the first statement is verified. In consequence, cl
iG = xG = gcl′
iG) = χ(nm)
(gcl′
i = χ(nm)
in (B.5), ul
iG = gcl
u
u
u
21
l ∈ Z. By the choice of x and l, l 6= 0. Since cl
some w ∈ G. For (g′, g′′) := φn,m(g),
igG = gcl
iG, cl
ig = gcl
iw for
(al
i′g′, bl
i′′g′′) = φn,m(cl
ig) = φn,m(gcl
iw) = φn,m(gcl
i) = (g′al
i′, g′′bl
i′′)
(B.6)
where a1, . . . , an and b1, . . . , bm are free generators of Fn and Fm, respec-
tively, and (i′, i′′) ∈ {1, . . . , n} × {1, . . . , m} is defined as i = m(i′ − 1) + i′′.
Hence al
i′′. From these and Lemma A.2(iii),
g′ ∈ H (n)
. There-
i′
fore g ∈ HiG by Lemma B.2(ii). Hence g ∈ Hi.
i′′g′′ = g′′bl
. Hence φn,m(g) = (g′, g′′) ∈ H (n)
and g′′ ∈ H (m)
i′′
i′ × H (m)
i′g′ = g′al
i′ and bl
i′′
Proposition B.5 Let Qn,m and H (nm)
tively.
l
be as in (B.1) and (B.3), respec-
(i) For any l ∈ {1, . . . , nm}, λ
Qn,m/ H (nm)
l
is irreducible.
(ii) For l, l′ ∈ {1, . . . , nm}, λ
Qn,m/ H (nm)
l
∼= λ
Qn,m/ H (nm)
l′
if and only if l = l′.
(i) From Lemma B.4, ComQn,m( H (nm)
Proof.
Theorem A.1(i), the statement holds.
(ii) From Lemma B.4, H (nm)
l = l′. From this and Theorem A.1(ii), the statement holds.
and H (nm)
) = H (nm)
l′
l
l
l
are quasiconjugate if and only if
. From this and
Proof of Theorem 3.4. (i) From Lemma B.3, the statement holds.
(ii) Let Kl := Gn,mH (nm)
Fnm-homogeneous space, we see that
. By identifying Fnm/Kl with Qn,m/ H (nm)
l
l
as a
λFnm/Kl(g(nm)
l
) = λ
Qn,m/ H (nm)
l
(g(nm)
l
)
(l = 1, . . . , nm).
(B.7)
Hence λFnm/Kl(Fnm) = λ
is also irreducible.
(iii) From Proposition B.5(ii) and (B.7), the statement holds.
Qn,m/ H (nm)
l
(Qn,m). From Proposition B.5(i), λFnm/Kl
C Proof of Proposition 3.5(ii)
In this section, we prove Proposition 3.5(ii). For this purpose, we prove the
following proposition.
22
Proposition C.1 Let Qn,m be as in (B.1). If n, m ≥ 2, then the group
Qn,m is ICC, that is, every conjugacy class in Qn,m, other than its unit is
infinite [3].
l
, blg′′b−1
l
l
6= al′g′a−1
Proof. Rewrite φ := φn,m and G := Gn,m, and let g := gG ∈ Qn,m and
(g′, g′′) := φ(g) for g ∈ Fnm. Fix g ∈ Qn,m \ {1}. By the choice of g,
(g′, g′′) 6= (1, 1).
Assume g′
6= 1. Choose an infinite sequence {al : l ≥ 1} ⊂ Fn such
that alg′a−1
l′ when l 6= l′. Since Fn is ICC, such a sequence
always exists. By the choice of {al}, al 6= al′ when l 6= l′. From Lemma
2.4(i), there exists {(bl, cl) : l ≥ 1} ⊂ Fm × Fnm such that φ(cl) = (al, bl)
for any l ≥ 1. By the choice of {cl}, φ(clgc−1
) where
φ : Qn,m → Fn × Fm denotes the natural homomorphism induced by φ.
From this and the choice of {al}, φ(clgc−1
l′ G) when l 6= l′.
From this, clgc−1
l′ when l 6= l′. Therefore
{clgc−1
l G) 6= φ(cl′gc−1
: l ≥ 1} is an infinite subset of the conjugacy class of g.
l
If g′ = 1, then g′′ 6= 1. In a similar way, we can construct an infinite
subset of the conjugacy class of g from g′′ by using Lemma 2.4(ii). Hence
the statement is verified.
l G) = (alg′a−1
l = clgc−1
l G 6= cl′gc−1
l′ G = cl′ gc−1
Proof of Proposition 3.5(ii). By definition, the group Qn,m in (B.1) acts
on ℓ2(Fnm/Gn,m) = ℓ2(Qn,m) by its left regular representation λQn,m. For
the natural left action L′ of Qn,m on Qn,m, L(g)(hGn,m) = ghGn,m =
L′(gGn,m)(hGn,m) for any g, h ∈ Fnm where L denotes the natural left action
of Fnm on Fnm/Gn,m. Hence λFnm/Gn,m (Fnm) = λQn,m(Qn,m). By Proposi-
tion C.1, λQn,m is a type II1 factor representation ([3], III.3.3.7 Proposition).
Hence the statement holds.
References
[1] E. Abe, Hopf algebras. Cambridge University Press, 1977.
[2] C. Akemann, S. Wassermann, N. Weaver, Pure states on free group
C∗-algebras. Glasgow Math. J. 52 (2010) 151 -- 154.
[3] B. Blackadar, Operator algebras. Theory of C∗-algebras and von Neu-
mann algebras. Springer-Verlag Berlin Heidelberg New York, 2006.
[4] N.P. Brown, N. Ozawa, C∗-algebras and finite-dimensional approxima-
tions. American Mathematical Society, 2008.
23
[5] M. Burger, P. de la Harpe, Constructing irreducible representations of
discrete groups. Proc. Indian Acad. Sci. Math. Sci. 107 (1997) 223 -- 235.
[6] K.R. Davidson, C∗-algebras by example. American Mathematical Soci-
ety, 1996.
[7] M. Enock, J.M. Schwartz, Kac algebras and duality of locally compact
groups. Springer-Verlag, 1992.
[8] H. Hatem, Decomposition of quasi-regular representations induced from
discrete subgroups of nilpotent Lie groups. Lett. Math. Phys. 81 (2007)
135 -- 150.
[9] M. Herschend, On the representation ring of a quiver. Algebr. Repre-
sent. Theor. 12 (2009) 513 -- 541.
[10] C. Kassel, Quantum groups. Springer-Verlag, 1995.
[11] K. Kawamura, A tensor product of representations of Cuntz algebras.
Lett. Math. Phys. 82 (2007) 91 -- 104.
[12] K. Kawamura, C∗-bialgebra defined by the direct sum of Cuntz alge-
bras. J. Algebra 319 (2008) 3935 -- 3959.
[13] K. Kawamura, C∗-bialgebra defined as the direct sum of Cuntz-Krieger
algebras. Comm. Algebra 37 (2009) 4065 -- 4078.
[14] K. Kawamura, Tensor products of
representa-
tions of Cuntz-Krieger algebras. Algebr. Represent. Theor. DOI
10.1007/s10468-012-9362-2, math.OA/0805.0667v1.
type III
factor
[15] J. Kustermans, S. Vaes, The operator algebra approach to quantum
groups. Proc. Natl. Acad. Sci. USA 97(2) (2000) 547 -- 552.
[16] G.W. Mackey, the theory of unitary group representations. The Uni-
versity of Chicago Press Chicago and London, 1976.
[17] W. Magnus, A. Karrass, D. Solitar, Combinatorial group theory, Inter-
science Publishers, 1966.
[18] T. Masuda, Y. Nakagami, S.L. Woronowicz, A C∗-algebraic framework
for quantum groups. Int. J. Math. 14 (2003) 903 -- 1001.
[19] G.K. Pedersen, C∗-algebras and their automorphism groups. Academic
Press, 1979.
24
[20] H. Yoshizawa, Some remarks on unitary representations of the free
group. Osaka Math. J. 3(1) (1951) 55 -- 63.
25
|
1908.07440 | 1 | 1908 | 2019-08-17T14:41:46 | Classification of tensor decompositions for II$_1$ factors | [
"math.OA"
] | In the mid thirties Murray and von Neumann found a natural way to associate a von Neumann algebra $L(\Gamma)$ to any countable discrete group $\Gamma$. Classifying $L(\Gamma)$ in term of $\Gamma$ is a notoriously complex problem as in general the initial data tends to be lost in the von Neumann algebraic regime. An important problem in the theory of von Neumann algebras is to completely describe all possible tensor decompositions of a given group von Neumann algebra $L(\Gamma)$. In this direction the main goal is to investigate how exactly a tensor decomposition of $L(\Gamma)$ relates to the underlying group $\Gamma$.
In this dissertation we introduce several new classes of groups $\Gamma$ for which all tensor decompositions of $L(\Gamma)$ are parametrized by the canonical direct product decompositions of $\Gamma$. Specifically, we show that whenever $L(\Gamma)\cong M_1\bar\otimes M_2$ where $M_i$ are any diffuse von Neumann algebras then there exists a non-canonical direct product decomposition $\Gamma=\Gamma_1\times\Gamma_2$ such that up to amplifications we have that $M_1\cong L(\Gamma_1)$ and $M_2\cong L(\Gamma_2)$. Our class include large classes of icc (infinite conjugacy class) amalgamated free products and wreath product groups. In addition we obtain similar classifications of tensor decompositions for the von Neumann algebras associated with the $T_0$ and $T_1$ group functors introduced by McDuff in $1969$. | math.OA | math | CLASSIFICATION OF TENSOR DECOMPOSITIONS FOR II1 FACTORS
9
1
0
2
g
u
A
7
1
]
.
A
O
h
t
a
m
[
1
v
0
4
4
7
0
.
8
0
9
1
:
v
i
X
r
a
by
Wanchalerm Sucpikarnon
A thesis submitted in partial fulfillment of the
requirements for the Doctor of Philosophy
degree in Mathematics
in the Graduate College of
The University of Iowa
December 2019
Thesis Supervisor: Ionut Chifan, Associate Professor
Copyright by
WANCHALERM SUCPIKARNON
2019
All Rights Reserved
Graduate College
The University of Iowa
Iowa City, Iowa
CERTIFICATE OF APPROVAL
PH.D. THESIS
This is to certify that the Ph.D. thesis of
Wanchalerm Sucpikarnon
has been approved by the Examining Committee for the
thesis requirement for the Doctor of Philosophy degree
in Mathematics at the December 2019 graduation.
Thesis committee:
Ionut Chifan, Thesis Supervisor
Raul Curto
Palle Jorgensen
Surjit Khurana
Victor Camillo
ACKNOWLEDGEMENTS
First of all I am thankful to the Development and Promotion of Science and
Technology Talents Project (DPST) in Thailand for the financial support and to my
family for their love and encouragement.
I also thank Raul Curto, Victor Camillo, Palle Jorgensen, and Surjit Khurana
for their time and servingonmy dissertation committee
And last but not least, I would like to expressmysincerestgratitudeto my advi-
sor Ionut Chifan for introducing me the research area and providing me opportunities
with his kindness and support during my academic journey. In addition as an advisee,
I want to thank him for being a forgiver when I made mistakes, being a navigator
when I lost my ways, being an excellent role model and showing me what a great
mathematician should be with his passion in mathematics, and finally thank for be-
ing my advisor.
ii
ABSTRACT
In the mid thirties Murray and von Neumann found a natural way to associate
a von Neumann algebra L(Γ) to any countable discrete group Γ. Classifying L(Γ) in
term of Γ is a notoriously complex problem as in general the initial data tends to be
lost in the von Neumann algebraic regime. An important problem in the theory of
von Neumann algebras is to completely describe all possible tensor decompositions
of a given group von Neumann algebra L(Γ). In this direction the main goal is to
investigate how exactly a tensor decomposition of L(Γ) relates to the underlying group
Γ.
In this dissertation we introduce several new classes of groups Γ for which
all tensor decompositions of L(Γ) are parametrized by the canonical direct product
decompositions of Γ. Specifically, we show that whenever L(Γ) ∼= M1 ¯⊗M2 where Mi
are any diffuse von Neumann algebras then there exists a non-canonical direct product
decomposition Γ = Γ1 × Γ2 such that up to amplifications we have that M1
∼= L(Γ1)
and M2
∼= L(Γ2). Our class include large classes of icc (infinite conjugacy class)
amalgamated free products and wreath product groups. In addition we obtain similar
classifications of tensor decompositions for the von Neumann algebras associated with
the T0 and T1 group functors introduced by McDuff in 1969.
iii
PUBLIC ABSTRACT
In the study of tensor decomposition of von Neumann algebra, Popa introduced
the notion of primeness which is analogous to prime numbers. However, the unique
prime factorization of von Neumann algebras are much more complicated.
In our
work we consider von Neumann algebra arising from a group and we obtain many
new classes of groups Γ that satisfy this classification result. This includes large
families of amalgamated free products, wreath products, McDuffs groups.
iv
TABLE OF CONTENTS
CHAPTER
1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1 Statements of main results
. . . . . . . . . . . . . . . . . . . . .
2 VON NEUMANN ALGEBRAS . . . . . . . . . . . . . . . . . . . . .
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Group von Neumann algebra . . . . . . . . . . . . . . . . . . . .
2.3 Group measure space . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Conditional expectation . . . . . . . . . . . . . . . . . . . . . . .
2.6 Amplification . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.7 Ultrapower von Neumann algebras and property gamma . . . . .
3 INTERTWINING RESULTS IN AMALGAMATED FREE PRODUCT
VON NEUMANN ALGEBRAS . . . . . . . . . . . . . . . . . . . . .
3.1 Popa's intertwining techniques
. . . . . . . . . . . . . . . . . . .
4 FINITE INDEX INCLUSIONS OF VON NEUMAN ALGEBRAS . .
5 AMENABILITY AND RELATIVE AMENABILITY . . . . . . . . . .
5.1 Amenable groups with their von Neumann algebras . . . . . . . .
5.2 Amenable von Neumann algebras . . . . . . . . . . . . . . . . . .
5.3 Relative amenability for von Neumann algebras . . . . . . . . . .
6 MAIN RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
von Neumann algebras
6.1 Tensor product decompositions of amalgamated free products of
. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . .
6.2 Spatially commensurable von Neumann algebras
6.3 Classification of tensor product decompositions of II1 factors aris-
ing from groups
. . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.1 Amalgamated free product groups . . . . . . . . . . . . .
6.3.2 Direct product of wreath product groups
. . . . . . . . .
6.3.3 McDuff's group functors T0 and T1 . . . . . . . . . . . . .
1
2
7
7
10
12
14
15
18
19
23
23
26
32
32
32
33
37
37
42
58
59
62
68
v
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
vi
CHAPTER 1
INTRODUCTION
1
An important step towards understanding the structure of II1 factors is the
study of their tensor product decompositions. A factor is called prime if it cannot be
decomposed as a tensor product of diffuse factors. Using this notion of ∗-orthogonal
von Neumann algebras, S. Popa was able to show in [Po83] that the (non-separable)
II1 factor L(FS) arising from the free group FS with uncountably many generators
S is prime. More than a decade later, using Voiculescu's influential free probability
theory, Ge managed to prove the same result about the free group factors L(Fn)
with countably many generators, n ≥ 2 [Ge98]. Using a completely different perspec-
tive based on C ∗-techniques, Ozawa obtained a far-reaching generalization of this by
showing that for every icc hyperbolic group Γ the corresponding factor L(Γ) is in fact
solid (for every diffuse A ⊂ L(Γ) von Neumann subalgebra, its relative commutant
A′ ∩L(Γ) is amenable) [Oz03]. Developing a new approach rooted in the study of clos-
able derivations, Peterson showed primeness of L(Γ), whenever Γ is any nonamenable
icc group with positive first Betti number [Pe06]. Within the powerful framework of
his deformation/rigidity theory Popa discovered a new proof of solidity of the free
group factors [Po06]. These methods laid out the foundations of a rich subsequent
activity regarding the study of primeness and other structural aspects of II1 factors
[Oz04, CH08, CI08, Si10, Fi10, CS11, CSU11, SW11, HV12, Bo12, BHR12, DI12,
CKP14, Is14, HI15, Ho15, DHI16, Is16].
2
1.1 Statements of main results
The techniques introduced in the deformation/rigidity framework also opened
up a whole array of new possibilities towards understanding novel aspects in the
classification of tensor product decompositions of factors. For example, motivated in
part by the results in [CdSS15], Drimbe, Hoff and Ioana have discovered in [DHI16]
a new classification result regarding the study of tensor product decompositions of
II1 factors. Precisely, whenever Γ is an icc group that is measure equivalent to a
direct product of non-elementary hyperbolic groups then all possible tensor product
decompositions of the corresponding II1 factor L(Γ) can arise only from the canonical
direct product decompositions of the underlying group Γ. Pant and de Santiago
showed the same result holds when Γ is a poly-hyperbolic group with non-amenable
factors in its composition series [dSP17]. In this dissertation we make new progress in
this direction by introducing several new and fairly large classes of groups for which
this tensor product rigidity phenomenon still holds. This include many new families of
groups that were not previously investigated in this framework such as amalgamated
free products and McDuff groups. Our results also improve significantly upon a series
of previous results on primeness and unique prime factorisations including [CH08,
SW11]. Below we briefly describe these results also placing them in a context and
explaining their importance and the methods involved.
Basic properties in Bass-Serre theory of groups show that the only way an
amalgam Γ1 ∗Σ Γ2 could decompose as a direct product is through its core Σ. Pre-
cisely, if Γ1 ∗Σ Γ2 = Λ1 × Λ2 then there is a permutation s of {1, 2} so that Λs(1) < Σ.
3
This further gives Σ = Λs(1) × Σ0, Γ1 = Λs(1) × Γ0
1, Γ2 = Λs(1) × Γ0
2 for some groups
Σ0 < Γ0
1, Γ0
2 and hence Λs(2) = Γ0
1 ∗Σ0 Γ0
2. An interesting question is to investigate
situations when this basic group theoretic aspect could be upgraded to the von Neu-
mann algebraic setting. It is known this fails in general since there are examples of
product indecomposable icc amalgams whose corresponding factors are McDuff and
hence decomposable as tensor products. However, under certain indecomposability
assumptions on the core algebra, we are able to provide a positive answer to our
question.
Theorem 1.1. Let Γ = Γ1 ∗Σ Γ2 be an icc group such that [Γ1 : Σ] ≥ 2 and [Γ2 : Σ] ≥
3. Assume that Σ is finite-by-icc and any corner of L(Σ) is virtually prime. Suppose
that L(Γ) = M1 ¯⊗M2, for diffuse Mi's. Then there exist decompositions Σ = Ω × Σ0
with Σ0 finite, Γ1 = Ω × Γ0
1, Γ2 = Ω × Γ0
2, for some groups Σ0 < Γ0
1, Γ0
2, and hence
Γ = Ω × (Γ0
1 ∗Σ0 Γ0
2). Moreover, there is a unitary u ∈ L(Γ), t > 0, and a permutation
s of {1, 2} such that
Ms(1) = uL(Ω)tu∗
and Ms(2) = uL(Γ0
1 ∗Σ0 Γ0
2)1/tu∗.
In particular this result provides many new examples of prime group factors
and factors that satisfies Ozawa-Popa's unique prime decomposition property. This
includes factors associated with simple groups such as Burger-Mozes groups which is
a premiere in the subject.
In [Po07] Popa was able to establish primeness for all factors L(Γ) associ-
ated with non-canonical wreath product groups Γ = A ≀ G where A is amenable and
4
Γ is non-amenable. Using the deformation techniques from [CPS12] Sizemore and
Winchester were able to extend this result by establishing various unique tensor de-
composition properties from von Neumann algebras arising from direct products of
such groups.
In this dissertation we extend this even further by showing that all
tensor decompositions of such factors are in fact parametrized by the canonical direct
product decompositions of the underlying group. Specifically, for product of groups
in the class WR (see section 6.3.2 for the definition) we have the following result
Theorem 1.2. Let Γ1, Γ2, . . . , Γn ∈ WR and let Γ = Γ1 × Γ2 × · · · × Γn. Consider
the corresponding von Neumann algebra M = L(Γ) and let P1, P2 be non-amenable
II1 factors such that M = P1 ¯⊗P2. Then there exist a scalar t > 0 and a partition
I1 ⊔ I2 = {1, 2, . . . , n} such that
L(ΓI1) ∼= P t
1
and L(ΓI2) ∼= P 1/t
2
.
In the celebrated work [Mc69] McDuff introduced an (uncountable) family
of groups that give rise to non-isomorphic II1 factors, thus solving a long standing
open problem at the time. Her construction of these groups was quite involved being
essentially based on the iteration of the so-called T0 and T1 group functors. These
functors are in part inspired by the earlier work of Dixmier and Lance [DL69] which
in turn go back to the pioneering work of Murray and von Neumann [MvN43].
Let Γ be a group. For i ≥ 1, let Γi be isomorphic copies of Γ and Λi be
isomorphic to Z. Define Γ =Li≥1 Γi and let S∞ be the group of finite permutations
of the positive integers N. Consider the semidirect product Γ ⋊ S∞ associated to
5
the natural action of S∞ on Γ which permutes the copies of Γ. Following [Mc69] we
define
• T0(Γ) = the group generated by Γ and Λi, i ≥ 1 with the only relation that Γi
and Λj commutes for i ≥ j ≥ 1.
• T1(Γ) = the group generated by Γ ⋊ S∞ and Λi, i ≥ 1 with the only relation
that Γi and Λj commute for i ≥ j ≥ 1.
From definitions it is evident that Ti(Γ) give rise to II1 factors L(Ti(Γ)) that
have an abundence of assymptoticaly central sequences and hence by [Mc69] they
admit many tensor product decompositions by the hyperfinite factor, i.e. L(Ti(Γ)) ∼=
L(Ti(Γ)) ¯⊗R. However, besides this classic result, virtually noting is known towards
describing the other possible tensor decompositions of these factors. In this thesis
we completely answer this question by showing that in fact these are all the possible
tensor decompositions of these factors.
Theorem 1.3. Fix Γ a non-amenable group and let α ∈ {0, 1}. If L(Tα(Γ)) = P1 ¯⊗P2
then either P1 or P2 is isomorphic to the hyperfinite II1 factor.
All the aforementioned results are obtained through the developments of sev-
eral new technical innovations in the deformation/rigidity technology. These new
methods are highlighted in the chapter 6 of this thesis which also contains the bulk
of the results. Particularly important in most of the proofs is the notion of spatial
commensurability for von Neumann subalgebras introduced in the section 6.2 as well
as the assymptotic analysis on bimodules and clustering von Neumann sugalgebras
presented in the proof of theorem 6.14. These new methods shed new light in the
6
study of tensor decomposition aspects present excellent potential to tackle more dif-
ficult groups that will be investigated in the future. In this dissertation, Theorem 1.1
is from the previous work with R. de Santiago while Theorem 1.2 and Theorem 1.3
are the results of a collaboration with I. Chifan.
7
CHAPTER 2
VON NEUMANN ALGEBRAS
2.1
Introduction
Let H be a Hilbert space and B(H) be the space of bounded linear operators
on H. Recall that B(H) is a Banach space with the operator norm k · k∞.
We define the convergences on H as the following:
• The uniform topology is a topology defined by the operator norm, i.e,
xn → x uniformly
if and only if
kxn − xk∞ → 0.
• The strong operator topology (SOT) is a topology generated by the family of
semi-norm kxξk for all x ∈ B(H) and ξ ∈ H, i.e.
xn → x SOT if and only if
k(xn − x)ξk → 0 for all ξ ∈ H.
• The weak operator topology (WOT) is a topology generated by the family of
semi-norm hxξ, ζi for all x ∈ B(H) and ξ, ζ ∈ H, i.e.
xn → x WOT if and only if
h(xn − x)ξ, ζi → 0 for all ξ, ζ ∈ H.
Note that the topologies on B(H) can be compared as the following relation:
WOT ≺ SOT ≺ uniform.
Definition 2.1. A von Neumann algebra is a ∗-subalgebra of B(H) containing
the unit 1 and being closed in the weak topology.
8
Definition 2.2. Let B ⊂ B(H), the commutant of B is defined by
B′ = {x ∈ B(H) xy = yx for all y ∈ B}.
Theorem 2.1 (Double commutant Theorem). Let A be a ∗-subalgebra of B(H) con-
taining the unit 1.
A′′ = A
W OT
SOT
= A
.
In particular, A is a von Neumann algebra if and only if A = A′′.
In general, by Double commutant Theorem, for any subset S ⊂ B(H) we call
(S∪S∗)′′ a von Neumann algebra generated by S. Moreover for subsets S1, S2 ⊂ B(H),
we write S1 ∨ S2 as the von Neumann algebra generated by S1 and S2.
If S ⊂ M are von Neumann algebras, then
P ′ ∩ M = {x ∈ M xp = px for p ∈ P }
is called the relative commutant of P in M.
Theorem 2.2 (Kaplansky Density Theorem). Let N ⊂ B(H) be a von Neumann
algebra and A be a strongly ∗-subalgebra, not assumed to be unital.
(i) If x ∈ N, then there exists a net (xα) from A converging ∗-strongly to x and
satisfying kxαk ≤ kxk for all α.
(ii) If x ∈ N is a self-adjoint then the net in (i) may be chosen with the additional
property that each xα is self-adjoint.
(iii) If u ∈ N is a unitary and A is a unital C∗-algebra, then there is a net (uα) of
unitaries from A converging ∗-strongly to u.
Definition 2.3. Let A be a subset of B(H). Then the center of A is defined by
9
Z(A) = A ∩ A′.
Definition 2.4. A von Neumann algebra M is called factor if it has the trivial
center, i.e. Z(M) = C1.
Definition 2.5. A von Neumann algebra M is finite if it has a faithful normal tracial
state τ : M → C satisfying:
• τ is a positive linear functional with τ (1) = 1;
• τ is faithful, i.e. if τ (x∗x) = 0 then x = 0;
• τ is normal, i.e. weakly continuous on (M)1, the unit ball of M with respect to
the uniform norm k · k∞;
• τ is trace, i.e. τ (xy) = τ (yx) for all x, y ∈ M.
If M is an infinite dimensional finite von Neumann algebra, then M is called an von
Neumann algebra of type II1.
Let M ∈ B(H) be a von Neumamm algebra on a Hilbert space H and p a
projection in H. Then
pMp = {pxp x ∈ M}
is a von Neumann algebra in B(pH). One says that pMp is a corner of M.
Definition 2.6. A von Neumann algebra M is diffuse if there are no nonzero minimal
projecton or an atom in M. Recall that a nonzero projection p ∈ M is said to be
minimal if the corner pMp = Cp.
10
Definition 2.7. We say that two von Neumann algebra M1 and M2 are isomorphic
if there exists a bijection ∗-homomorphism (called an isomorphism) between M1 and
M2, and denoted by M1
∼= M2.
Remark. If a finite factor (M, τ ) has a minimal projection, then M ∼= Mn(C) for some
n. A finite factor M is diffuse if and only if it is infinite dimensional. M is then a
type II1 factor.
2.2 Group von Neumann algebra
Let Γ be a discrete group and ℓ2(Γ) be the space of square summable se-
quences over Γ which has a natural orthonormal basis {δh h ∈ Γ}, where δh is a
Kronecker delta. Consider the left regular representation of Γ on ℓ2(Γ) defined by
λg(δh) = δgh
for g, h ∈ Γ. The von Neumann algebra L(Γ) is a von Neumann algebra
generated by the set {λg g ∈ Γ}. By the bicommutant theorem
L(Γ) = span{λg g ∈ Γ}′′ ∼= C[Γ]
SOT
(2.1)
In addition, recall that for f, f0 ∈ ℓ2(Γ), the convolution product
defined by
Lf (f0) = f ∗ f1
(f ∗ f0)(t) =Xs∈Γ
f (s)f0(s−1t).
By Cauchy-Schwarz inequality, we have kf ∗ f0k∞ ≤ kf k2kf0k2 and it follows that
f ∗ f0 ∈ ℓ∞(Γ). We say that f is a left convolver for Γ if f ∗ f0 ∈ ℓ2(Γ) for every
f0 ∈ ℓ2(Γ). We denote LC(Γ) the space of all left convolvers for Γ. Since LC(Γ)
11
commutes with ρ(Γ), so we can view LC(Γ) as a subspace of ρ(Γ)′ in B(ℓ2(Γ)). Plus,
it is easy to check that λ(Γ) is contained in LC(Γ). Therefore, L(Γ) ⊂ LC(Γ).
Similarly, we can introduce RC(Γ) as the space of all right convolvers for Γ. As a
consequence, one can show that
LC(Γ) = L(Γ) = RC(Γ)′ and RC(Γ) = R(Γ) = LC(Γ)′
The von Neumann algebra L(Γ) is the (left) group von Neumann algebra
of Γ and R(Γ) is the right group von Neumann algebra of Γ. Note that since the left
and right-regular representations are equivalent it follows that L(Γ) ∼= R(Γ).
Moreover for x ∈ L(Γ),
τ (x) = hxδe, δei
defines a normal faithful trace on L(Γ). In particular, L(Γ) is a finite von Neumann
algebra.
If x =Pg∈Γ αgδg ∈ ℓ2(Γ) is a left-convolver, then we will often also write x or
Pg∈Γ αgug to denote the operator Lx ∈ LC(Γ). (Instead of δg we use ug to emphasize
that ug is a unitary operator.) And we call the set {αg}g∈Γ the Fourier coefficients
of x. Thus writing x =Pg∈Γ αgug should be considered as an abbreviation for writing
Lx = LPg∈Γ αg δg.
Theorem 2.3. [MvN43] Let Γ be a discrete group A group von Neumann algebra
L(Γ) is a factor if and only if Γ is an infinite conjugacy class (icc) group, i.e. each
conjugacy class of non-trivial elements in Γ is an infinite set.
Proof. Suppose h ∈ Γ \ {e} and the conjugacy class hΓ = {ghg g ∈ Γ} is finite. Then
12
x =Pk∈hΓ uk 6∈ C and x ∈ {ug}′
g∈Γ ∩ L(Γ) = Z(L(Γ)).
Conversely, suppose Γ is icc and x = Pg∈Γ αgug ∈ Z(L(Γ)) \ C, then for all
h ∈ Γ we have
x = uhxu∗
h =Xg∈Γ
αhuhgh−1 =Xg∈Γ
αh−1ghug.
Thus the Fourier coefficients for x are constant on conjugacy classes and sincePg∈Γ αg <
∞. Then we have αg = 0 for all g 6= e and hence x = αe ∈ C.
The followings are examples of countable icc groups
• Fn, n ≥ 2 the free group on n generators.
• S∞ = ∪∞
n=1
Sn the group of finite permutations on N
• wreath products H ≀I Γ := (⊕IH) ⋊ Γ where H, Γ are countably infinite and
Γ y I with infinite orbits.
• icc property is closed under products.
• amalgamated free products Γ = Γ1 ∗Σ Γ2 where [Γ1 : Σ] ≥ 2, [Γ2 : Σ] ≥ 3 and
Σ ∩ gΣg−1 < ∞.
2.3 Group measure space
Let Γ y (X, µ) be a probability measure preserving (p.m.p.) action of Γ on
a probability space (X, µ). Recall that L∞(X, µ) acts naturally by multiplication on
L2(X, µ). Let σ : Γ y L2(X, µ) be an action of Γ on L2(X, µ) defined by
σg(f )(x) = f (g−1x)
for all
g ∈ Γ, x ∈ (X, µ).
13
Define the space
A[Γ] :=(cid:26)Xg∈Γ
agg(cid:12)(cid:12) ag ∈ L∞(X, µ) and ag = 0 for all g ∈ Γ but finite(cid:27).
The product is defined by (a1g)(a2h) = a1σg(a2)gh and the involution by (ag)∗ =
σg−1(a∗)g−1 where a∗ = ¯a. To avoid any confusion, we write ug instead of 1L∞(X,µ)g
According to the first step, it follows that A[Γ] is a ∗-algebra of operators acting on
the Hilbert space H = L2(X, µ) ⊗ ℓ2(Γ) by sending
a 7→ L(a) := a ⊗ 1 , ug 7→ L(ug) := σg ⊗ λg
and
and L(ug)L(a)L(ug)∗ = L(σg(a)).
The group measure space von Neumann algebra L∞(X, µ) ⋊ Γ associated
with Γ y (X, µ) or crossed product is the von Neumann algebra generated by
L(L∞) ∪ {L(ug) g ∈ Γ}.
In particular, the elements in L∞(X, µ)⋊Γ may be identified to elements of L2(X, µ)⊗
ℓ2(Γ) by x 7→ xUe and hence are written as
xgug,
x =Xg∈Γ
L2(X,µ) < ∞. The coefficient xg ∈ L∞(X, µ) are called Fourier coeffi-
withPg∈Γ kxgk2
cients of x and the ug are called the canonical unitaries of the crossed product.
With the trace defined by
τ (x) = hxue, uei =ZX
xedµ where x =Xg∈Γ
xgug.
In particular, the group von Neumann algebra is a specific case when X is just a
singleton.
14
2.4 Tensor product
Definition 2.8. Let M1 ∈ B(H1) and M2 ∈ B(H2) be von Neumann algebras. The
algebraic tensor product M1 ⊙ M2 of M1 and M2 is defined by
(x1 ⊗ x2)(ξ1 ⊗ ξ2) = (x1ξ1 ⊗ x2ξ2)
for any xi ∈ Mi, ξi ∈ Hi and i = 1, 2. Obviously, M1 ⊙ M2 is a ∗-algebra and its SOT-
closure gives a von Neumann algebra acting on H1 ⊙ H2. We call it von Neumann
tensor product denoted by
M1 ¯⊗M2.
There is a celebrating theorem established by Tomita in 1960.
Theorem 2.4 (Tomita's Commutant theorem). Let H1, H2 be Hilbert spaces. Let
M1 ⊂ B(H1) and M2 ⊂ B(H2) be von Neumann algebras. Then
(M1 ¯⊗M2)′ = M ′
1 ¯⊗M ′
2
According to Tomita's Theorem, we have that M1 ¯⊗M2 is a factor if each
component Mi needs to be a factor for i = 1, 2. We, furthermore, have the following
basic proposition.
Proposition 2.5. Given any von Neumann algebras M1 and M2. Then
1. If M1 and M2 are tracial factors, then so is M1 ¯⊗M2;
2. If M1 and M2 are II1 factors, then so is M1 ¯⊗M2.
Definition 2.9. Let M be a II1 factor. We says M is prime provided that if M is
isomorphic to a tensor product M1 ¯⊗M2 of von Neumann algebras M1, M2 then either
M1 or M2 is finite dimensional.
15
In the same spirit with Choda's Galois correspondence theorem [Ch78]. Ge
obtained a splitting theorem for tensors that we recall below. This is instrumental in
deriving some of main results in this thesis.
Theorem 2.6 (Theorem A in [Ge96]). If M is a finite factor, N is a finite von
Neumann algebra, and B is a von Neumann sub algebra of M ¯⊗N, there exists a von
Neumann sub algebra N0 of N such that
B = M ¯⊗N0
2.5 Conditional expectation
Theorem 2.7 (GNS-Construction). Let A∗ be a C∗-algebra and ϕ a positive linear
functional on A. Then there exists a Hilbert space L2(A, ϕ) and a unique (up to
equivalence) representation
π : A → B(L2(A, ϕ))
with a unit cyclic vector 1ϕ ∈ L2(A, ϕ) such that
ϕ(x) = hπ(x)1ϕ, 1ϕi
for all x ∈ A.
Throughout the section N denote a finite von Neumann algebra withe a fixed
faithful normal trace τ and B denote a von Neumann subalgebra of N. Using GNS
construction, we can define the Hilbert space L2(N) which is defined over the dense
linear subspace N by
hx, yi = τ (xy∗)
for all x, y ∈ N.
This L2(B) is a Hilbert subspace of L2(N) with the restricted inner product on L2(N).
Denote by eB : L2(N) → L2(B) be the canonical orthogonal projection. We define
16
EB = eBN .
For the further use, we recall the following basic properties of this projection
Theorem 2.8. Let B ⊂ N be von Neumann subalgebras.
1. EB = eBN is a norm reducing map from N onto B with EB(1) = 1;
2. EB(bxc) = bEB(x)c for all x ∈ N and b, c ∈ B;
3. τ (xEB(y)) = τ (EB(x)EB(y)) = τ (EB(x)y) for all x ∈ N;
4. {eB}′ ∩ N = B and B′ = (N ′ ∪ {eB})′′;
5. EB is normal complete positive map;
6. eBJ = JeB and EBJ = JEB
7. For the uniqueness, if φ : N → B with
φ(b1xb2) = b1φ(x)b2
and τ (φ(x)) = τ (x)
for all x ∈ N and b1, b2 ∈ B, then φ = EB.
Definition 2.10. Let B ⊂ N be finite von Neumann algebras. From Theorem 2.8,
the conditional expectation EB : N → B is defined by EB = eBN .
Below we record some conditional expectation that will be useful subsequently.
(1) Let Λ < Γ be groups. Consider L(Λ) ⊂ L(Γ).We have EL(Λ)(x) =
Pg∈Λ xgug where x =Pg∈Γ xgug ∈ L(Γ).
17
(2) Let L∞(X) ⋊ Γ be a crossed product. The conditional expectation EL(Γ) :
L∞(X)⋊Γ → L(Γ) is defined by EL(Γ)(P xgug) =Pg∈Γ τ (xg)ug where x =Pg∈Γ xgug ∈
L∞(X) ⋊ Γ, EL∞(X)(P xgug) = xe.
(3) Let N ⊂ M be finite von Neumann algebras and p ∈ N be a projection.
Define EpN p : pMp → pNp by EpN p(x) = pEN (x)EN (p)−1p for all x ∈ pMp.
Note that kxk2
2,p = τ (p)−1kxk2
2 where k · k2, k · k2,p are the norms on L2(M)
and L2(pMp) respectively.
(4) Let N ⊂ M be finite von Neumann algebras and p be a projection in
N ′ ∩ M. Define EN p : pMp → Np by EN p(x) = EN (x)EN (p)−1p. for all x ∈ pMp.
In the cases (3) and (4) if N is a factor, then EN (p) = τ (p)1.
To study structural property of inclusions of von Neumann algebras, an im-
portant tool is the associated basic construction. This algebra was introduced by E.
Christensen in order to study perturbations of algebras and later was used to great
extended theory of finite index subfactors by V.F.R. Jones. The basic construction
plays a key roles in Popa's deformation/rigidity theory especially in the intertwining
technique that we will see use in this dissertation.
Definition 2.11. If B is a von Neumann subalgebra of a finite von Neumann algebra
N with faithful normal trace τ, the basic construction from the inclusion B ⊂ N
is defined to be the von Neumann algebra hN, eBi := (N ∪ {eB})′′.
Theorem 2.9. Let B be a von Neumann subalgebra of finite von Neumann algebra
N with a fixed faithful normal trace τ. Then hN, eBi is a semifinite von Neumann
algebra with a faithful semifinite normal trace T r satisfying the following properties:
1. hN, eBi = JB′J, hN, eBi′ = JBJ, and the ∗-subalgebra NeBN = span{xeBy x, y ∈
18
N} is weakly dense in hN, eBi;
2. the central support of eB in hN, eBi is 1;
3. eBhN, eBieB = BeB;
4. eBN and NeB are weakly and strongly dense in respectively ebhN, eBi and
hN, eBieB;
5. the map x 7→ xeB : N → NeB ⊂ hN, eBieB is injective;
6. T r(xeBy) = τ (xy) for all x, y ∈ N;
7. NeBN is dense in L2(hN, eBi, T r) in k · k2,T r-norm.
2.6 Amplification
Let M ⊂ B(H) be a von Neumann algebra For every n ≥ 1, let Mn(M) be a
space of n × n matrices with entries in M. Clearly, Mn(M) ⊂ B(H ⊕n). Moreover, it
is a straightforward proof to show that Mn(M) is also a von Neumann algebra. If M
is a type II1 factor then Mn(M) is also a type II1 factor.
Denote Trn ⊗τ its trace defined by
(Trn ⊗τ )([xij]) =Xi
τ (xii).
Moreover, we embed Mn(M) into Mn+1(M) by putting the zero entries in the last
row and the last column, we obtain the increasing algebras
M(M) = [n≥1
Mn(M)
19
For any two projections p, q ∈ M(M), there is an n ≥ 0 such that both p and
q must belong to Mn(M). Since Mn(M) is a factor and the trichotomy property for
projections on factors, we have
p and q are equivalent
if and only if
(Trn ⊗τ )(p) = (Trn ⊗τ )(q)
This follows that
Define
p M(M) p = p Mn(M) p ≃ q Mn(M) q.
M t = p Mn(M) p, where
t = (Tr ⊗τ )(p).
It is not too hard to check that M t is well-defined for every t > 0 and unique up to
isomorphism. We call M t an amplification of M by t.
Theorem 2.10. Let M, M0 be II1 factors and s, t > 0. Then the following hold:
(a) (M ¯⊗M0)t = M ¯⊗M t
0 = M t ¯⊗M0.
(b) (M s)t = M st.
(c) M ¯⊗M0 = M t ⊗ M 1/t
0
.
Corollary 2.11. Given two groups Γ1, Γ2 and t > 0, we have the relation
L(Γ1 × Γ2) = L(Γ1) ¯⊗L(Γ2) = L(Γ1)t ¯⊗L(Γ2)1/t.
2.7 Ultrapower von Neumann algebras and property gamma
In this section we introduce the ultrapower von Neumann algebra N ω asso-
ciated to a given von Neumann algebra N. This is an important tool that provides
20
algebraic framework to understand various asymptotic properties such as central se-
quence. We fix a free ultrafilter ω on N. Recall that ω is an element of βN \ N where
βN is the Stone-Cech compatification of N. For any bounded sequence (cn) of complex
numbers, limω cn is defined as the value at ω of this sequence, viewed as a continuous
function on βN.
Let (Mn, τn) is a sequence of tracial von Neumann algebras. The product alge-
bra Πn≥1Mn is the C∗-algebra of bounded sequences x = (xn)n with xn ∈ Mn for every
n, endowed with the norm kxk = supn kxnk. The (tracial) ultraproduct ΠωMn is the
quotient of Πn≥Mn by the ideal Iω of all sequences (xn)n such that limω τn(x∗
nxn) = 0.
It is easily seen that Iω is a normed closed two-sided ideal, so that ΠωMn is a C∗-
algebra. If xω denotes the class of x ∈ Πn≥1Mn, then τω(x) := limω τn(xn) defines
without ambiguity a faithful tracial state on ΠωMn. We set kyk2,ω = τω(y∗y)1/2
whenever y ∈ ΠωMn.
When (Mn, τn) = (M, τ ) for all n, we set M ω = ΠωM and we says that
(M ω, τω) is the (tracial) ultrapower of (M, τ ) along ω.
Proposition 2.12. We have the followings.
1. (ΠωMn, τω) is a tracial von Neumann algebra.
2. If Mn are finite factors such that limn dim Mn = +∞, then ΠωMn is a II1 factor.
Next we recall Murray-von Neumann property Gamma associates with a von
Neumann algebra. This was the first invariant introduced to distinguish the hyper-
fintie II1 factor R from the free group factor L(F2). This showed the existence of
non-hyperfinite II1 factors.
Definition 2.12. A II1 factor M is said to have Property Gamma if given ε > 0
and x1, . . . , xk ∈ N, there exists a trace zero unitary u ∈ M such that
21
kux1 − xiuk2 < ε,
1 ≤ i ≤ k.
An alternative formulation is the existence, for a fixed but arbitrary finite set F ⊂ M,
of a sequence {un}∞
n=1 of trace zero unitaries in N satisfying
lim
n→∞
kunx − xunk = 0,
x ∈ F.
Theorem 2.13 ([Mc69]). Let M be a separable II1 factor and let ω be free ultrafilter
on N. The following conditions are equivalent:
1. M has Property Gamma;
2. M ′ ∩ M ω 6= C1;
3. M ′ ∩ M ω is diffuse.
Definition 2.13 ([Mc69]). Let M be a separable II1 factor. For ω be free ultrafilter
on N, if the central sequence algebra M ′ ∩ M ω is non-abelian then M ∼= M ¯⊗R and
M is said to be McDuff.
We finish this section by recording the important result for our development.
Theorem 2.14 (Theorem 3.1in [CSU13]). Let Γ ba a countable discrete group together
with a family of subgroups G such that satisfies condition NC(G). Let (A, τ ) be any
amenable von Neumann algebra equipped that ω is a free ultrafilter on the positive
integers N.
Then for any asymptotically central sequence (xn)n ∈ M ′ ∩ M ω, there exists
a finite subset F ⊂ G such that (xn)n ∈ ∨Σ∈F (A ⋊ Σ)ω ∨ M (i.e. the von Neumann
subalgebra of M ω generated by M and (A ⋊ Σ)ω for Σ ∈ F ).
22
INTERTWINING RESULTS IN AMALGAMATED FREE PRODUCT
VON NEUMANN ALGEBRAS
CHAPTER 3
23
3.1 Popa's intertwining techniques
Over a decade, S. Popa has developed the following powerful method in [Po03,
Theorem 2.1 and Corollary 2.3] to identify intertwines between arbitrary subalgebras
of tracial von Neumann algebras.
In order to study the structural theory of von Neumann algebras, S. Popa
introduced the following concept of the intertwining subalgebras which has been very
instrumental in the recent development in the classification of von Neumann algebra.
Theorem 3.1 (Popa's intertwining by bimodule technique). Let (M, τ ) be a finite
von Neumann algebra. Suppose P, Q be von Neumann subalgebras of M. Then the
following are equivalent:
1. There exist projections p ∈ P , p ∈ Q, a nonzero partially isometry v ∈ pP q and
a ∗-homomorphism ψ : pP p → qQq such that
ψ(x)v = vx for all x ∈ pP p.
and such that v∗v ∈ ψ(pP p)′ ∩ qMq and vv∗ ∈ (pP p)′ ∩ pMp.
2. For any group G ⊂ U(P ) such that G′′ = P , there is no sequence (un)n ⊂ G
satisfying for all x, y ∈ M
kEQ(xuny)k2 → 0.
3. There exists a Q-P -submodule H of L2(M) with dimQ H < ∞.
24
4. There exists a positive element a ∈ hM, eQi; the basic construction with Tr(a) <
∞ such that the ultraweakly closed convex hull of {w∗aw w ∈ P unitary} does
not contain 0.
If one of the conditions in Theorem 3.1 above holds, we say Q embeds in P
inside M and denoted by P ≺M Q. Otherwise, we write P 6≺M Q. In the condition
(1) the partial isometry v is also called an intertwiner between P and Q.
Moreover, if we have P p′ ≺M Q for any nonzero projection p′ ∈ P ′ ∩ 1P M1P ,
then we write P ≺s
M Q.
Next we record several well-known important results that will be used in the
subsequent sections.
Theorem 3.2 (Corollary F.14 in [BO08]). Let M be a finite von Neumann algebra
with separable predual. Suppose (An) ⊂ M is a sequence of von Neumann subalgebras
and N ⊂ pMp be a von Neumann subalgebra such that N 6≺M An for any n. Then
there exists a diffuse abelian von Neumann subalgebra B ⊂ N such that N 6≺M An
for any n.
Proposition 3.3. Let M = M1 ∗P M2 be an amalgamated free product von Neumann
algebra. If for each i there is a unitary ui ∈ U(Mi) such that EP (ui) = 0 then
M ⊀M Mk
for all k = 1, 2.
Proof. Let u = u1u2 ∈ U(M). Using freeness and basic approximation properties one
can see that limn→∞ kEMk(xuny)k2 = 0 for all x, y ∈ M. Then Theorem 3.1 (b) gives
the conclusion.
Theorem 3.4 (Lemma 2.2 in [CI17]). Let Γ1, Γ2 ≤ Γ be countable groups such that
25
L(Γ1) ≺L(Γ) L(Γ2).
Then there exists g ∈ Γ such that [Γ1 : Γ1 ∩ gΓ2g−1] < ∞.
FINITE INDEX INCLUSIONS OF VON NEUMAN ALGEBRAS
CHAPTER 4
26
In this section we recall several basic facts from the pioneering work of V.F.R
Jones [Jo81] on the theory of finite index inclusion of factors.
Definition 4.1. Let B ⊂ M be an inclusion of finite von Neumann algebras. The
a set (mi)1≤i≤n ∈ M is called a (left) Pimsner-Popa basis if m ∈ M has a unique
expression form
where bi ∈ piB.
m =
mibi
n
Xi=1
Theorem 4.1. Let M be a II1 factor and B ⊂ M a von Neumann subalgebra. Then
L2(M)B is finite generated if and only if m1, . . . , mn ∈ M such that
(i) EB(m∗
i mj) = δi,jpj is a projection in B for all i, j;
i = 1.
(ii) P1≤i≤n mieBm∗
If these conditions hold, we haveP1≤i≤n mimi∗ = dim(L2(M)B)1 and x =P1≤i≤n miEB(m∗
i x) =
1 for every x ∈ M.
Definition 4.2. Let B be a subfactor of a II1 factor M. The Jones' index of B in
M is defined as the dimension of L2(M) as a left B-module, i.e.,
[M : B] = dimC(L2(M)B).
By the definition, we have [M : B] is finite if and only if hM, eBi is a type II1
factor if and only if L2(M)B is finitely generated.
Theorem 4.2 (Downward basic construction, Lemma 3.1.8 in [Jo81]). Let N ⊂ M
be II1 factors such that [M : N] < ∞. Then there exists a subfactor P ⊂ N and a
27
projection eP ∈ M such that
• EP (eP ) = τ (eP )1,
• epxeP = Ep(x)eP for all x ∈ N, and
• M = hN, eP i.
While V.F.R. Jones defined the notation of finite index on factors, Pimsner
- Popa found a more probabilistic general notion of finite index that works for all
inclusions of finite von Neumann algebras.
Definition 4.3 ([PP86]). If B ⊂ M is a subfactor of the type II1 factor, then
[M : B]−1 = inf(cid:8)kEB(x)k2
2/kxk2
2 x ∈ M+, x 6= 0(cid:9)
with the convention ∞−1 = 0. If [M : B] 6= 0 then we says that B ⊂ M has textbffinite
index or is an finite index inclusion. In the case that B ⊂ M are II1 factors then it
coincides with the notion of indexes by Jones.
For the following proposition, we record some basic properties of finite index
inclusions of von Neumann algebras that will be needed throughout our work. Even
if they are well known, we also include their proofs for the sake of completeness.
Proposition 4.3. Let N ⊂ M be von Neumann algebras with [M : N] < ∞. Then
the following hold:
1. If N is a factor, then
dimC(N ′ ∩ M) ≤ [M : N] + 1.
2. [Po95, 1.1.2(iv)] If Z(M) is purely atomic1 then Z(N) is also purely atomic, .
3. [Po95, 1.1.2(ii)] If N is a factor and r ∈ N ′ ∩ M then
28
[rMr : Nr] ≤ τ (r)[M : N] < ∞.
Proof. (1) Fix 0 6= p ∈ N ′ ∩ M a nonzero projection. Since N is a factor then
EN (p) = τ (p)1. As [M : N] < ∞, we have
τ (p)2 = kEN (p)k2
2 ≥ [M : N]−1kpk2
2 = [M : N]−1τ (p).
Since p is an arbitrary projection in N ′ ∩ M, we obtain τ (p) ≥ [M : N]−1 for all
projections p ∈ N ′ ∩ M. Hence,
dimC(N ′ ∩ M) ≤ [M : N] + 1.
(2) Let p ∈ Z(N) be a maximal projection such that Z(N)p is purely atomic
and Z(N)(1 − p) is diffuse. To prove the conclusion it suffices to show that q = 1 − p
vanishes. Since the inclustion N ⊂ M is finite index, we have qNq ⊂ qMq is finite
index. This implies that qMq ≺qM q qNq. Hence,
qNq′ ∩ qMq ≺qM q qMq′ ∩ qMq = Z(M)q.
Therefore, Z(N)q ≺ Z(M)q. Since Z(M) is purely atomic, it follows that there
exists a minimal projection of Z(N) under q. This forces q = 0, as desired.
(3) Since r ∈ N ′ ∩ M and N is a factor, we have EN (r) = τ (r)1. Thus,
EN r(rxr) = τ (r)−1EN (rxr)r
for all x ∈ M.
1The unit 1 can be expressed as a sum of minimal projection
Hence, we have
kEN r(rxr)k2
2,r = τ (r)−1kEN r(rxr)k2
2
29
= τ (r)−1(cid:0)τ (r)−1kEN (rxr)rk2
2(cid:1)
= τ (r)−2kEN (rxr)rk2
2
≥ τ (r)−2[M : N]−1krxrk2
2
= τ (r)−1[M : N]−1krxrk2
2,r
which shows [rMr : Nr] ≤ τ (r)[M : N].
Definition 4.4. Let M be a factor. We say M is virtually prime if A, B ⊂ M are
commuting diffuse subfactors of M, then [M : A ∨ B] = ∞.
Lemma 4.4. Let N ⊂ M be a finite index inclusion of II1 factors. Then one can
find projections p ∈ M, q ∈ N, a partial isometry v ∈ M, and a unital injective
∗-homomorphism φ : pMp → qNq such that
1. φ(x)v = vx for all x ∈ pMp, and
2. [qNq : φ(pMp)] < ∞.
Proof. Since [M : N] < ∞ then M ≺M N. Thus there exist projections p ∈ M,
q ∈ N, a partial isometry v ∈ M, and a unital injective ∗-homomorphism φ : pMp →
qNq so that
φ(x)v = vx
for all
x ∈ pMp.
(4.1)
Denoting by Q = φ(pMp) ⊂ qNq, we notice that vv∗ ∈ Q′ ∩ qMq and v∗v = p.
Moreover by restricting vv∗ if necessary we can assume wlog the support projection
30
of EN (vv∗) equals q. Also from the condition (4.1), we have that
Qvv∗ = vMv∗ = vv∗Mvv∗
Since M is a factor, passing to relative commutants we have
vv∗(Q′ ∩ qMq)vv∗ = (Qvv∗)′ ∩ vv∗Mvv∗
= Z(vv∗Mvv∗)
= Cvv∗.
Since Q′ ∩ qNq ⊂ Q′ ∩ qMq, there is a projection r ∈ Q′ ∩ qNq such that
r(Q′ ∩ qNq)r = Qr′ ∩ rNr = Cr.
Since q = s(EN (vv∗)) one can check that rv 6= 0. Thus replacing Q by Qr, φ(·)
by φ(·)r, q by r, and v by the partial isometry from the polar decomposition of rv
then the intertwining relation (4.1) still holds with the additional assumption that
Q′ ∩ qMq = Cq. In particular, EqN q(vv∗) = cq where c is a positive scalar.
To finish the proof we only need to argue that [qNq : Q] < ∞. Consider the
von Neumann algebra hqNq, vv∗i generated by qNq and vv∗ inside qMq. Therefore
we have the following inclusions
Q ⊂ qNq ⊂ hqNq, vv∗i ⊂ qMq.
Since vv∗Mvv∗ = Qvv∗ then
vv∗qNq
and vv∗ = Qvv∗.
31
Moreover, since vv∗ ∈ Q′ ∩ qMq and EqN q(vv∗) = c1, one can check that hqNq, vv∗i
is isomorphic to the basic construction of Q ⊂ qNq. Therefore, Q ⊂ qNq has index
c, hence finite.
Lemma 4.5 (Lemma 3.9 in [Va08]). Let (M, τ ) be a tracial von Neumann algebra
and A, B, N von Neumann subalgebras. Let A ⊂ N be a finite index inclusion. Then
the followings hold
(1) If A ≺M B, then N ≺M B.
(2) If B ≺M A, then B ≺M A.
AMENABILITY AND RELATIVE AMENABILITY
CHAPTER 5
32
Amenability is one of the important standard term in studying von Neumann
algebra which was first introduced by Connes in 1976. In this chapter, we discuss
about the amenabilities on groups and on von Neumann algebra. Finally we provide
the concept of relative amenability for von Neumann algebras.
5.1 Amenable groups with their von Neumann algebras
Definition 5.1. A group Γ is said to be amenable if one of the following conditions
holds:
a) there exists a left Γ- invariant mean on ℓ∞(Γ)
b) there exists a sequence of unit vectors (ξn) in ℓ2(Γ) such that for every g ∈ Γ,
lim
n
kλG(g)ξn − ξnk2 = 0
c) there exists a sequence of finitely supported positive definite functions on Γ
which converges pointwise to 1
d) Følner ; For any finite subset E ⊂ Γ and ε > 0 there is a finite subset F ⊂ Γ
such that
max
s∈E
sF ∆ F
F
< ε.
5.2 Amenable von Neumann algebras
Definition 5.2. A von Neumann algebra M is said to be amenable or injective
33
a) if it has a concrete representation as a von Neumann subalgebra of some B(H)
such that there exists c conditional expectation E : B(H) → M.
b) for every inclustion A ⊂ B of unital C ∗-algebra, every unital completely positive
map φ : A → M extends to a completely positive map from B to M.
c) for any B(H) which contains M as a von Neumann subalgebra, there is a con-
ditional expectional expectation from B(H) onto M.
We use the word "amenable" to emphasis the analogy of the amenability for
groups. By previous section we can show that a countable group Γ is amenable if and
only if group von Neumann algebra L(Γ) is amenable.
Theorem 5.1 ([Co76]). The hyperfinite factor R is amenable.
5.3 Relative amenability for von Neumann algebras
In practice we will use the following characterization, which comes from [OP07]
wihch was introduced by Ozawa-Popa.
Definition 5.3. Let P ⊂ M be an inclusion of a von Neumann algebras. A state
ψ : M → C is P -central if
ψ(mx) = ψ(xm)
for every x ∈ P and every m ∈ M.
Following Section 2.2 in [OP07], we have the following definition
Definition 5.4. Let P, Q be von Neumann subagebras of a tracial von Neumann
algebra (M, τ ) Then P is amenable relative to Q inside M and denoted by
P ⋖M Q if one of the following conditions holds:
34
a) there exists a conditional expectation from hM, eQi onto P whose restriction to
M is EM
P
b) there is a P -central state ψ on hM, eQi such that ψM = τ
c) there is a P -central state ψ on hM, eQi such that ψ is normal on M and faithful
on Z(P ′ ∩ M)
d) there is a net (ξi) of norm-one vector in L2(hM, eQi) such that
lim
i
kxξi − ξxk = 0 for every x ∈ P
and
lim
hξixξii = τ (x)
i
for every x ∈ M.
e) M L2(M)P is weakly contained in M L2(M) ⊗ L2(M)P .
Moreover, if M is amenable relative to Q inside M, one simply says that M
is amenable relative to Q or that Q is co-amenable in M. In particular, M is
amenable if and only if M is amenable relative to C1.
Proposition 5.2 (Ioana). Let P, Q be von Neumann subalgebras of a finite von Neu-
mann algebra (M, τ ). If P ≺s
M Q, then P ⋖M Q.
Proposition 5.3 (Transitive property , Proposition 2.4 (3) in [OP07]). Let P, Q, N ⊂
M be finite von Neumann algebras. If N ⋖M P and P ⋖M Q, then N ⋖M Q.
Next we record several important results that will be used in our subsequent
development.
35
Theorem 5.4 (Theorem A in [Va13]). Let M = M1 ∗P M2 be the amalgamated free
product of the tracial von Neumann algebra (Mi, τ ) with the common von Neumann
subalgebra P ⊂ Mi with respect to the unique trace preserving conditional expectations.
Let p be a nonzero projection, A ⊂ pMp a von Neumann subalgebra that is amenable
relative to one of the Mi inside M. Then at least one of the following statement holds.
• A ≺M P .
• There is an i ∈ {1, 2} such that NpM p(A)′′ ≺M Mi
• NpM p(A)′′ is amenable relative to P inside M.
Proposition 5.5 (Proposition 2.7 in [PV11]). Let (M, τ ) be a tracial von Neumann
algebra with von Neumann subalgebras Q1, Q2 ⊂ M. Assume that Q1 and Q2 form a
commuting square and that Q1 is regular in M. If a von Neumann algebra P ⊂ pMp
is amenable relative to both Q1 and Q2, then P is amenable relative to Q1 ∩ Q2.
Lemma 5.6 (Lemma 2.6 in [DHI16]). Let (M, τ ) be a tracial von Neumann algebra,
and P ⊂ pMp, Q ⊂ M be von Neumann subalgebras.
(1) Assume that P is amenable relative tot Q. Then P p′ is amenable relative to Q
for every projection p′ ∈ P ′ ∩ pMp.
(2) Assume that p0P p0p′ is amenable relative to Q for some projection p0 ∈ P,
p′ ∈ P ′ ∩ pMp. Let z be the smallest projection belonging to NpM p(P )′ ∩ pMp
such that p0p′ ≤ z. Then P z is amenable relative to Q.
(3) Assume that P ≺s
M Q. Then P is amenable relative to Q.
Lemma 5.7 (Lemma 2.6 in [IS19]). Let (M, τ ) be a tracial von Neumann algebra and
36
Q ⊂ M a von Neumann subalgebra. Assume that there exists nets of von Neumann
algebras Qn, Mn ⊂ M such that
(1) Q ⊂ Mn ∩ Qn and QnL2(M)Mn ⊂weak QnL2(Qn) ⊗Q L2(Mn)Mn for every n,
(2) limn kx − EMn(x)k2 = 0 for every x ∈ M.
If P ⊂ M is a von Neumann subalgebra which is amenable relative to Qn inside M,
for every n then P is amenable relative to Q inside M.
37
CHAPTER 6
MAIN RESULTS
6.1 Tensor product decompositions of amalgamated free products of
von Neumann algebras
In this section we preset a general result that completely describe all the
tensor product decompositions for a large class of amalgamated free product von
Neumann algebras M1 ∗P M2. Specifically, we will show that every tensor product
product decomposition essentially splits the core P . This is a phenomenon that
parallels results in Bass-Serre theory for groups. The precise statement is Theorem
6.3. However in order to prove our result we first need the following result which
essentially relies on the usage of [Va13, Theorem A] (see also [Io12, Theorem 7.1]).
Theorem 6.1. Let M1, M2 be tracial von Neumanna algebras with the common von
Neumann subalgebra P ⊂ Mi such that for each i = 1, 2 there is a unitary ui ∈ U(Mi)
so that EP (ui) = 0. Let M = M1∗P M2 be the corresponding amalgamated free product
von Neumann algebra and assume in addition that M is not amenable relative to P
inside M. Let p ∈ M be a nonzero projection and assume A1, A2 ⊂ pMp are two
commuting diffuse subalgebras that A1 ∨ A2 ⊂ pMp has finite index. Then
Ai ≺M P for some
i = 1, 2.
Proof. Fix A ⊂ A1 an arbitrary diffuse amenable subalgebra of A1. Using Theorem
5.4, one of the following holds:
(1) A ≺M P ;
38
(2) A2 ≺M Mi for some i = 1, 2; or
(3) A2 is amenable relative to P inside M.
If (6.1) holds then either
(4) A2 ≺M P ; or
(5) A1 ∨ A2 ≺M Mi.
If (6.1) holds, since [pMp : A1 ∨ A2] < ∞, then we must have M ≺M Mi. Then
Proposition 3.3 will lead to a contradiction. If case (6.1) holds, then applying Theorem
5.4 again we get one of the following
(6) A2 ≺M P ;
(7) A1 ∨ A2 is a amenable relative to P inside M; or
(8) A1 ∨ A2 ≺M Mi for some i.
If (6.1) holds, since [pMp : A1 ∨ A2] < ∞, it follows that pMp is a amenable relative
to P inside M, contradicting the initial assumption. Notice that the condition (6.1)
is similar to the condition (6.1) which was already eliminated before. To summary,
we have obtained that for any subalgebra A ⊂ A1 amenable we have either
A ≺M P or A2 ≺M P.
(6.1)
Here, suppose A1 6≺M P By using Theorem 3.2 and setting An = P and
N = A1 We obtain that there exist a diffuse von Neumann subalgebra B ⊂ A1 such
that B 6≺M P . From above since A is any arbitrary diffuse subalgebra, it is forced
that A2 ≺M P . So we can conclude that A1 ≺M P or A2 ≺M P.
Corollary 6.2. Let Γ = Γ1 ∗Σ Γ2 such that [Γ1 : Σ] ≥ 2 and [Γ2 : Σ] ≥ 3. Denote by
39
M = L(Γ) let p be a projection in M and assume A1, A2 ⊂ pMp are two commuting
diffuse subalgebras such that A1 ∨ A2 ⊂ pMp has finite index. Then
Ai ≺M L(Σ)
for some
i = 1, 2.
(6.2)
Proof. Since [Γ1 : Σ] ≥ 2 and [Γ2 : Σ] ≥ 3 then by the proof of Theorem 7.1 in [Io12]
it follows that L(Γ) is not amenable relative to L(Σ). The conclusion follows then
from Theorem 6.1.
With these preparations at hand we are ready to prove the main theorem of
this section.
Theorem 6.3. Let M = M1∗P M2 be an amalgamated free product such that M, M1, M2, P
are II1 factors and [Mk : P ] = ∞ for all k = 1, 2. Assume A1, A2 ⊂ M are diffuse
factors such that M = A1 ¯⊗A2. Then there exist tensor product decompositions
P = C ¯⊗P0, M1 = C ¯⊗M 0
1 , and M2 = C ¯⊗M 0
2
and hence M = C ¯⊗(M 0
1 ∗P0 M 0
2 ). Moreover, there exist t > 0 and a permutation
σ ∈ S2 such that
At
σ(1)
∼= C
and A1/t
σ(2)
∼= M 0
1 ∗P0 M 0
2 .
Proof. By Theorem 6.1 we have that Ai ≺M P for some i ∈ {1, 2}. Since M = A1 ¯⊗A2,
by symmetry it suffices to assume A1 ≺M P . It follows directly from the the definition
that there exist nonzero projections a ∈ A1, p ∈ P , a nonzero partial isometry v ∈ M,
and a unital injective ∗-homomorphism
Φ : aA1a → pP p
such that
Φ(x)v = vx for all x ∈ aA1a.
40
(6.3)
Shrinking a if necessary we can assume there is an integer m such that τ (p) =
m−1. Letting B = φ(aA1a), it is easy to chech that vv∗ ∈ B′ ∩ pMp. Also we can
assume wlog that s(EP (vv∗)) = p and using factoriality of Ai that v∗v = r1 ⊗ r2.
Thus by (6.3) there is a unitary u ∈ M which is extended from v so that
Bvv∗ = vA1v∗ = u(r1A1r1 ⊗ r2)u∗.
(6.4)
Passing to relative commutants we also have
vv∗(B′ ∩ pMp)vv∗ = vv∗B′vv∗ ∩ vv∗pMpvv∗
= (Bvv∗)′ ∩ vv∗Mvv∗
= (vA1v∗)′ ∩ vMv∗
= u((r1A1r1 ⊗ r2)′ ∩ (r1A1r1 ⊗ r2A2r2))u∗
= u(r1 ⊗ r2A2r2)u∗.
(6.5)
Combing (6.4) and (6.5) together, we have
vv∗(B ∨ B′ ∩ pMp)vv∗ = u(r1A1r1) ¯⊗(r2A2r2)u∗
= vv∗Mvv∗.
Letting z be the central support of vv∗ in B ∨ B′ ∩ pMp we conclude that
(B ∨ B′ ∩ pMp)z = zMz.
(6.6)
Note by construction we actually have z ∈ Z(B′ ∩ pMp).
In addition, we have
p ≥ z ≥ vv∗ and hence
41
p ≥ s(EP (z)) ≥ s(EP (vv∗)) = p.
Thus s(EP (z)) = p. Also notice that p ≥ s(EMk(z)) ≥ z ≥ vv∗. For every t > 0,
denote ek
t = χ[t,∞)(EMk (z)). Using relation (6.6) and [CIK13, Lemma 2.3] it follows
that the inclusion (B ∨B′ ∩pMkp)ek
t ⊂ ek
t Mkek
t is finite index. This, together with the
assumptions and [Va08, Lemma 3.7] further imply that (B ∨ B′ ∩ pMkp)ek
t ⊀Mk P .
But ek
t z commutes with (B ∨ B′ ∩ pMkp)ek
t and hence by [IPP05, Theorem 1.2.1]
we have ek
t z ∈ Mk. Since ek
t z → z in W OT , as t → 0, we obtain that z ∈ pMkp,
for all k = 1, 2.
In conclusion z ∈ pM1p ∩ pM2p = pP p and hence z = p. Thus
using factoriality and (6.6) we get that pMp = B ¯⊗(B′ ∩ pMp). Moreover, we have
B ⊂ pP p ⊂ pMp = B ¯⊗(B′ ∩ pMp) and since B is a factor it follows from Theorem
2.6 that pP p = B ¯⊗(B′ ∩ pP p). Similarly one can show that pMkp = B ¯⊗(B′ ∩ pMkp)
for all k = 1, 2. Thus,
B′ ∩ pMp = (B′ ∩ pM1p) ∨ (B′ ∩ pM2p)
= (B′ ∩ pM1p) ∗(B′∩pP p) (B′ ∩ pM2p).
Combining these observations, we now have
pMp = B ¯⊗(B′ ∩ pMp)
= B ¯⊗((B′ ∩ pM1p) ∗(B′∩pP p) (B′ ∩ pM2p))
= (B ¯⊗(B′ ∩ pM1p)) ∗B ¯⊗(B′∩pP p) (B ¯⊗(B′ ∩ pM2p)).
42
Tensoring by Mm(C) this further gives
M = Mm(C) ¯⊗pMp
= Mm(C) ¯⊗B ¯⊗((B′ ∩ pM1p) ∗(B′∩pP p) (B′ ∩ pM2p))
= (Mm(C) ¯⊗B ¯⊗(B′ ∩ pM1p)) ∗Mm(C) ¯⊗B ¯⊗(B′∩pP p) (Mm(C) ¯⊗B ¯⊗(B′ ∩ pM2p))
Letting
C := Mm(C) ¯⊗B, P0 := B′ ∩ pP p,
and M 0
k := B′ ∩ pMkp,
altogether, the previous relations show that
P = C ¯⊗P0, M1 = C ¯⊗M 0
1 , M2 = C ¯⊗M 0
2 , and M = C ¯⊗(M 0
1 ∗P0 M 0
2 ).
For the remaining part of the conclusion, notice that relations (6.4), (6.5) and p =
z(vv∗) show that
Aτ (r1)
i
∼= B, Aτ (r2)
i+1
∼= (B′ ∩ pMp)τ (vv∗).
Using amplifications these further imply that
Amτ (r1)
i
∼= C, Aτ (r2)/(mτ (vv∗ ))
i+1
∼= M 0
1 ∗P0 M 0
2 .
Letting t = mτ (r1) we get the desired conclusion.
6.2 Spatially commensurable von Neumann algebras
In the context of Popa's concept of weak intertwining of von Neumann algebras
we introduce a notion of commensurable von Neumann algebras up to corners. This
notion is essential to this work as it can be used very effectively to detect tensor
product decompositions of II1 factors (see Theorems 6.6 and 6.8 below). It is also the
43
correct notion which translate in the von Neumann algebraic language to the notion
of commensurability for groups.
In the first part of section we build the necessary technical tools to prove
these two results. Several of the arguments developed here are inspired by ideas from
[CdSS15] and [DHI16].
Definition 6.1. Let P, Q ⊂ M (not necessarily unital) be inclusions of von Neumann
algebras. We write P ∼=com
M Q (and we say a corner of P is spatially commensurable
to a corner of Q) if there exist nonzero projections p ∈ P , q ∈ Q, a nonzero partial
isometry v ∈ M and a ∗-homomorphism φ : pP p → qQq such that
φ(x)v = vx
for all
x ∈ pP p
[qQq : φ(pP p)] < ∞
s(EQ(vv∗)) = q.
(6.7)
(6.8)
(6.9)
When just the condition (6.7) is satisfied together with φ(pP p) = qQq.
In other
words, φ is a ∗-isomorphism. We write pP p ∼=φ,v
M qQq.
Remark. When pP p is a II1 factor then so is φ(pP p). By Proposition 4.3 (1), φ(pP p)′∩
qQq) is finite dimensional, so there exists r ∈ φ(pP p)′ ∩ qQq) such that rv 6= 0. Thus
replacing φ(·) by φ(·)r and v by the isometry in the polar decomposition of rv one can
check (6.7) still holds. Also from Proposition 4.3 (3) it follows that φ(pP p)r ⊂ rQr
is an finite index inclusion of II1 factors. Hence throughout this article, whenever
P ∼=com
M Q and P is a factor, we will always assume the algebras in (6.8) are II1
factors.
44
For further use we recall the following result from [CKP14, Lemma 2.6].
Proposition 6.4 (Proposition 2.4). [CKP14] Let (M, τ ) be a tracial von Neumann
algebra and let z ∈ M be a nonzero projection. Suppose that P ⊂ zMz and N ⊂ M
are von Neumann subalgebras such that P ∨ (P ′ ∩ zMz) ⊂ zMz has finite index and
that P ≺M N. Then there exist a scalar s > 0, nonzero projections r ∈ N, p ∈ P ,
a subalgebra P0 ⊂ rNr, and a ∗-isomorphism θ : pP p → P0 such that the following
properties are satisfied:
1. P0 ∨ (P ′
0 ∩ rNr) ⊂ rNr has finite index;
2. there exist a nonzero partial isometry v ∈ M such that
r EN (vv∗) = EN (vv∗)r ≥ sr
and
θ(pP p)v = P0v = rvpP p;
3. EN (v(pP ′p ∩ pMp)v∗)′′ ⊂ P ′
0 ∩ rNr.
We record next a technical variation of [CKP14, Proposition 2.4] in the context
of commensurable von Neumann algebras that will be essential to deriving the main
results of this section.
Lemma 6.5. Let Σ < Γ be groups where Γ is icc. Assume Z(L(Σ)) is purely atomic1,
r ∈ L(Γ) is a projection, and there exist commuting II1 subfactors P, Q ⊂ rL(Γ)r such
that P ∨ Q ⊂ rL(Γ)r has finite index. If P ≺M L(Σ) then one of the following holds:
1. There exist projections p ∈ P, e ∈ L(Σ), a partial isometry w ∈ M, and a unital
injective ∗-homomorphism Φ : pP p → eL(Σ)e such that
1The unit 1 can be expressed as a sum of minimal projection
45
(a) Φ(x)w = wx for all x ∈ pP p;
(b) s(EL(Σ)(ww∗)) = e;
(c) If B := Φ(pP p) then B ∨(B′ ∩eL(Σ)e) ⊂ eL(Σ)e is a finite index inclusion
of II1 factors.
2. P ∼=com
L(Γ) L(Σ).
Proof. From the assumption, P ≺L(Γ) L(Σ) so there exist projections p ∈ P , q ∈
L(Σ), a nonzero partial isometry v ∈ L(Γ), and a unital injective ∗-homomorphism
φ : pP p → qL(Σ)q such that
φ(x)v = vx
for all x ∈ pP p.
(6.10)
Let C := φ(pP p). Note v∗v ∈ pP p′ ∩ pL(Γ)p, vv∗ ∈ C ′ ∩ qL(Γ)q and we can also
assume that
s(EL(Σ)(vv∗)) = q
Clearly Q ⊂ P ′. Since P ∨ Q ⊂ rL(Γ)r has finite index, we have P ∨ (P ′ ∩ rL(Γ)r)
also has finite index in rL(Γ)r. By Proposition 6.4, it implies that
C ∨ (C ′ ∩ qL(Σ)q) ⊂ qL(Σ)q
(6.11)
is also a finite index inclusion of algebras. By Proposition 4.3(2), Z(C ′ ∩ qL(Σ)q) is
purely atomic and there is a nonzero projection e ∈ Z(C ′ ∩ qL(Σ)q) so that ev 6= 0
and we have either
i) (C ′ ∩ qL(Σ)q)e is a II1 factor, or
ii) (C ′ ∩ qL(Σ)q)e = Mn(C)e for some n ∈ N.
46
Consider Φ : pP p → Ce =: B given by
Φ(x) = φ(x)e
for all x ∈ pP p
and let w be the partial isometry in the polar decomposition of ev. Then condition
(6.10) implies that
Φ(x)w = wx for all x ∈ pP p.
Moreover, we have evv∗e ≤ ww∗. Applying conditional expectation to the
relation, with its properties we have
EqL(Σ)q(vv∗)e = EqL(Σ)q(evv∗e) ≤ EqL(Σ)q(ww∗).
By considering support vectors, we obtain that
e = s(EqL(Σ)q(vv∗))e = s(EqL(Σ)q(vv∗)e) ≤ s(EqL(Σ)q(ww∗)).
Since w is the partial isometry in the polar decomposition of ev, by its unique-
ness, we have eEqL(Σ)q(ww∗) = EqL(Σ)q(eww∗) = EqL(Σ)q(ww∗).
it follows that
s(EqL(Σ)q(ww∗)) ≤ e and therefore s(EL(Σ)(ww∗)) = e.
Assume case i) above. Using (6.11), we have
B ∨ (B′ ∩ eL(Σ)e) = Ce ∨ (C ′ ∩ qL(Σ)q)e ⊂ eL(Σ)e)
is a finite index inclusion of II1 factors. Altogether, these lead to possibility (1) in
the statement.
Assume case ii) above. Then relation (6.11) implies that
C = Be ⊂ eL(Σ)e
is a finite index inclusion which gives possibility (2) in the statement.
47
Theorem 6.6 (Claims 4.7-4.12 in [CdSS15]). Let Σ < Λ be finite-by-icc groups2.
Also assume there exists 0 6= p ∈ Z(L(Σ)′ ∩ L(Λ)) such that L(Σ) ∨ (L(Σ)′ ∩ L(Λ))p ⊂
pL(Λ)p admits a finite Pimnser-Popa basis. Then there exists Ω < Λ such that
[Σ, Ω] = 1 and [Λ : ΣΩ] < ∞.
The next result is a basic von Neumann's projections equivalence property for
inclusions of von Neumann algebras. Its proof is standard and we include it only for
reader's convenience.
Lemma 6.7. Let N ⊂ (M, τ ) be finite von Neumann algebras, where N is a II1
factor. Then for every projection 0 6= e ∈ M there exists a projection f ∈ N and a
partial isometry w ∈ M such that e = w∗w and ww∗ = f .
Theorem 6.8. Let Σ < Γ be countable groups, where Γ is icc and Σ is finite-by-icc.
Let r ∈ L(Γ) be a projection and let P, Q ⊂ rL(Γ)r be commuting II1 factors such
that P ∨ Q ⊂ rL(Γ)r has finite index. If P ∼=com
L(Γ) L(Σ) then there exist a subgroup
Ω < CΓ(Σ) satisfying the following properties:
(a) [Γ : ΣΩ] < ∞;
(b) Q ∼=com
M L(Ω).
Proof. Since P ∼=com
L(Γ) L(Σ), by the definition there exist nonzero projections p ∈
P, q ∈ L(Σ), a nonzero partial isometry v ∈ L(Γ), and an injective, unital ∗-
homomorphism Φ : pP p → eL(Σ)e so that
2A group G is called finite-by-icc if it has a normal subgroup N that is finite and the
quotient G/N is icc.
48
(1) Φ(x)v = vx for all x ∈ pP p, and
(2) Φ(pP p) ⊂ qL(Σ)q is a finite index inclusion of II1 factors.
Now we denote by R := Φ(pP p) ⊂ qL(Σ)q. Let T ⊂ R ⊂ qL(Σ)q be the downward
basic construction for inclusion R ⊂ qL(Σ)q. Since [qL(Σ)q : R] < ∞, according to
Theorem 4.2 let a ∈ T ′ ∩ qL(Σ)q be the Jones' projection satisfying
qL(Σ)q = hR, ai and aL(Σ)a = T a.
(6.12)
Also note that [qL(Σ)q : R] = [R : T ]. As the ∗-homomorphism Φ : pP p → qL(Σ)q
is injective, the restriction Φ−1 : T → pP p is an injective ∗-homomorphism such that
U := Φ−1(T ) ⊂ pP p is a finite Jones index subfactor and
Φ−1(x)v∗ = v∗x for all x ∈ T.
(6.13)
Notice that T ⊂ qL(Σ)q and the projection a ∈ T ′ ∩ qL(Σ)q. Let θ′ : T a → T
be the ∗-isomorphism given by θ′(xa) = x for all x ∈ T.
We can check that v∗a 6= 0 and from the polar decomposition of v∗a, let w0
be a nonzero partial isometry so that v∗a = w∗
0v∗a. Since from above we know
T a = aL(Σ)a, combining together with (6.13) we have that the compostion map
θ = Φ−1 ◦ θ′ : aL(Σ)a → pP p
is an injective ∗-homomorphism such that its image
θ(aL(Σ)a) = Φ−1 ◦ θ′(aL(Σ)a) = Φ−1(T ) = U ⊂ pP p and
θ(y)w∗
0 = w∗
0y
for all
y ∈ aL(Σ)a.
(6.14)
49
By the assumption P ∨Q ⊂ rL(Γ)r has finite index. It follows that pP p∨Qp ⊂
pL(Γ)p also has finite index as well. From (6.12) we have U ⊂ pP p has finite index
so it follows that U ∨ Qp ⊂ pL(Γ)p has finite index. Since these all are factors, it
follows that U ∨ Qp ⊂ pL(Γ)p admits a finite Pimsner-Popa basis. From construction
we have
U ∨ Qp ⊂ U ∨ (U ′ ∩ pL(Γ)p) ⊂ pL(Γ)p
and hence U ∨ Qp ⊂ U ∨ (U ′ ∩ pL(Γ)p) admits a finite Pimsner-Popa basis. Also since
U ∨ Qp is a factor, we have by Proposition 4.3(1) that
dimC(cid:18)(cid:2)U ∨ (U ′ ∩ pL(Γ)p(cid:3) ∩ (U ∨ Qp)′(cid:19) < ∞.
Since [U ∨ (U ′ ∩ pL(Γ)p(cid:3) ∩ (U ∨ Qp)′ = [U ′ ∩ pL(Γ)p] ∩ (U ∨ Qp)′, we conclude that
dimC(cid:0)[U ′ ∩ pL(Γ)p] ∩ (U ∨ Qp)′(cid:1) < ∞.
Using Proposition 4.3(3) for every minimal projection b ∈ [U ′ ∩ pL(Γ)p] ∩ (U ∨ Qp)′,
then we have
(U ∨ Qp)b ⊂(cid:0)U ∨ (U ′ ∩ pL(Γ)p)(cid:1)b
is a finite inclusion of II1 factors.
Claim: Qb ⊂ (U ′ ∩ pL(Γ)p)b has finite index.
Now we have known from above that (U ∨ Qp)b ⊂ (cid:0)U ∨ (U ′ ∩ pL(Γ)p)(cid:1)b is a finite
inclusion. Thus, by 4.3 there exists Cb > 0 such that for all x ∈ U+ and y ∈
(U ′ ∩ pL(Γ)p)+ we have
kEU ∨Qb(xyb)k2
2,b ≥ Cbkxybk2
2,b,
(6.15)
50
where k · k2,b is the norm on L2(bL2(Γ)b). Since EU ∨Qp(b) = τp(b)p we have
EU ∨Qb(zb) = EU ∨Qp(zb)bτ −1
p (b)
for all
z ∈ U ∨ (U ′ ∩ pL(Γ)p).
Thus for every x ∈ U and y ∈ (U ′ ∩ pL(Γ)p) we have
EU ∨Qb(xyb) = EU ∨Qp(xyb)bτ −1
p (b)
= xEU ∨Qp(yb)bτ −1
p (b)
= xEQp(yb)bτ −1
p (b)
= xEQb(yb).
Also since U is a factor, we can check that we have
kxybk2
2 = kxk2
2kybk2
2
for all x ∈ U and y ∈ (U ′ ∩ pL(Γ)p). This further implies that
kxybk2
2,b = kxk2
2kybk2
2,b.
Using these formulas together with (6.15) we see that
kxk2
2kEQb(yb)k2
2,b = kxEQb(yb)k2
2,b
= kEU ∨Qb(xyb)k2
2,b
≥ Cbkxybk2
2,b
= Cbkxk2
2kybk2
2,b
and hence
kEQb(yb)k2
2,b ≥ Cbkybk2
2,b
51
for all y ∈ (U ′ ∩ pL(Γ)p)+. Hence Qb ⊂ U ′ ∩ pL(Γ)pb is a finite index inclusion of II1
factors for every minimal projection b ∈ [U ′ ∩ pL(Γ)p] ∩ (U ∨ Qp)′.
Choose a minimal projection b ∈ [U ′∩pL(Γ)p]∩(U ∨Qp)′ so that w∗ = bw∗
0 6= 0.
Thus (6.14) gives
θ(y)w∗ = w∗y, for all y ∈ aL(Σ)a.
(6.16)
Notice that w∗w ∈ (U ′ ∩ pL(Γ)p)b and ww∗ ∈ aL(Σ)a′ ∩ aL(Γ)a. Let u ∈ pL(Γ)p be
a unitary so that uw∗w = w, then relation (6.16) entails
uUw∗wu∗ = ww∗aL(Σ)a.
(6.17)
Passing through relative commutants we also have
uw∗w(U ′ ∩ pL(Γ)p)w∗wu∗ = ww∗(aL(Σ)a′ ∩ aL(Γ)a)ww∗
= ww∗(L(Σ)′ ∩ L(Γ))ww∗
Altogether, (6.17) and (6.19) imply that
uw∗w(U ∨ (U ′ ∩ pL(Γ)p))w∗wu∗ = ww∗(aL(Σ)a ∨ (aL(Σ)a′ ∩ aL(Γ)a))ww∗
= ww∗(L(Σ) ∨ (L(Σ)′ ∩ L(Γ)))ww∗.
(6.18)
(6.19)
(6.20)
Since from assumptions pP p∨Qp = p(P ∨Q)p ⊂ pL(Γ)p is a finite index and U ⊂ pP p
is finite index it follows that U ∨ Qp ⊂ pL(Γ)p has finite index as well. Also notice
U ∨ Qp ⊂ U ∨ (P ′ ∩ rL(Γ)r)p
= U ∨ (pP p′ ∩ pL(Γ)p)
⊂ U ∨ (U ′ ∩ pL(Γ)p)
52
Thus U ∨ (U ′ ∩ pL(Γ)p) ⊂ pL(Γ)p is finite index. Combining with (6.20) we obtain
ww∗(L(Σ) ∨ (L(Σ)′ ∩ L(Γ)))ww∗ ⊂ ww∗L(Γ)ww∗
is a finite index inclusion of II1 factors. By using Theorem 6.6, there exists a subgroup
Ω < Λ such that
[Σ, Ω] = 1
and [Γ : ΣΩ] < ∞.
(6.21)
Since Γ is an icc group, it follows that Σ, Ω also are icc groups as well; in particular,
both L(Σ) and L(Ω) are II1 factors. By Lemma 6.7, there exist unitaries u1 ∈
U ′ ∩ pL(Γ)p and u2 ∈ L(Σ)′ ∩ L(Γ) such that
u1w∗wu∗
1 = q1 ∈ Qb
and u∗
2ww∗u2 = q2 ∈ L(Ω).
We denote t := u∗
2uu∗
1 then the relation (6.19) can be rewritten as
tq1(U ′ ∩ pL(Γ)p)q1t∗ = q2(L(Σ)′ ∩ L(Γ))q2.
(6.22)
Since by (6.21) [Γ : ΣΩ] < ∞, we obtain that
q2L(ΣΩ)q2 ⊂ q2L(Γ)q2
has finite index. Since L(Ω) ⊂ L(Σ)′ ∩ L(Γ), it follows that
q2L(ΣΩ)q2 ⊂ q2(cid:0)L(Σ) ∨ (L(Σ)′ ∩ L(Γ))(cid:1)q2
is a finite index inclusion. Therefore following the same argument as the previous
claim, we obtain that
q2L(Ω)q2 ⊂ q2L(Σ)′ ∩ L(Γ)q2
53
is a finite index inclusion of II1 factors. By Lemma 4.4, there exist projections r1, r2 ≤
q2, a partial isometry w1 ∈ q2L(Σ)′ ∩ L(Γ)q2, and a ∗-isomorphism
φ′ : r1L(Σ)′ ∩ L(Γ)r1 → B ⊂ r2L(Ω)r2
such that
(3) φ′(x)w1 = w1x for all x ∈ r1L(Σ)′ ∩ L(Γ)r1;
(4) [r2L(Ω)r2 : B] < ∞.
Using Lemma 6.7, relation (6.22), and perturbing more the unitary t, we can assume
there exists a projection q3 ∈ Q such that q3b ≤ q1 and
tq3(U ′ ∩ pL(Γ)p)q3bt∗ = r1(L(Σ)′ ∩ L(Γ))r1.
(6.23)
Consider the ∗-isomorphism Ψ′ : q3Qq3 → tq3Qq3bt∗ given by
Ψ′(x) = txbt∗
for x ∈ q3Qq3
and we set Ψ = φ′ ◦ Ψ′ : q3Qq3 → r2L(Ω)r2. Clearly Ψ is a ∗-homomorphis. Using (3)
above for every x ∈ q3Qq3 we have
Ψ(x)w1 = φ′(Ψ′(x))w1t = w1Ψ′(x)t = w1txbt∗t
= w1txb = w1tbx.
Next we will show that w1tb 6= 0.
Indeed, suppose by contradiction that
w1tb = 0 then w1tbq1t∗ = 0. This implies that w1q2 = 0. Thus
w1 = w1r1 = w1r1q2 = 0,
a contradiction. So letting w to be the partial isometry in the polar decomposition
of w1tb = ww1tb, simply denoting q := q3 and f := r2, we get that
54
Ψ : qQq → f L(Ω)f
is an injective, unital ∗-homomorphism so that
Ψ(x) w = wx for all x ∈ qQq.
Moreover since Qb ⊂ q1(U ′ ∩ pL(Γ)p)q1 is finite index, using (4) above and (6.23) one
gets that
Ψ(qQq) ⊂ r2L(Ω)r2
has finite index. Altogether these show that Q ∼=com
L(Γ) L(Ω) as desired.
We end this section presenting the second main result. This roughly asserts
that tensor product decompositions of group von Neumann algebras whose factors
are commensurable with subalgebras arising commuting subgroups can be "slightly
perturbed" to tensor product decompositions arising from the actual direct product
decompositions of the underlying group. The proof uses the factor framework in an
essential way and it is based on arguments from [OP03, Proposition 12] and [CdSS15,
Theorem 4.14] (see also [DHI16, Theorem 6.1]).
Theorem 6.9. Let Γ be an icc group and assume that M = L(Γ) = M1 ¯⊗M2, where
Mi are diffuse factors. Also assume there exist commuting, non-amenable, icc sub-
groups Σ1, Σ2 < Γ such that
[Γ : Σ1Σ2] < ∞, M1
∼=com
M L(Σ1),
and M2
∼=com
M L(Σ2).
55
Then there exist a group decomposition Γ = Γ1 × Γ2, a unitary u ∈ M and t > 0 such
that
M1 = uL(Γ1)tu∗
and M2 = uL(Γ2)1/tu∗.
Proof. Since M1
∼=com
M L(Σ1), in particular we have L(Σ1) ≺M M1. Since M =
M1 ¯⊗M2 then proceeding as in the proof of [OP03, Proposition 12] there exist a scalar
µ > 0 and a partial isometry v ∈ M satisfying
p := vv∗ ∈ M 1/µ
2
,
q := v∗v ∈ L(Σ1)′ ∩ M and
vL(Σ1)v∗ ⊂ M µ
1 p.
(6.24)
Let Ω2 = {γ ∈ Γ OΣ1(γ) < ∞} where OΣ1(γ) = {ηγη−1 η ∈ Σ1} is the
orbit of γ ∈ Γ under the conjugate action of Σ1. Notice that for γ1, γ2 ∈ Γ, it is easy
to show that OΣ1(γ1γ2) ⊂ OΣ1(γ1)OΣ1(γ2). This implies that
OΣ1(γ1γ2) ≤ OΣ1(γ1)OΣ1(γ2)
and hence Ω2 is a subgroup of Γ. Clearly Σ2 < Ω2 because from the assumption
Σ1, Σ2 commute. Since [Γ : Σ1Σ2] < ∞, it follows that [Γ : Ω2Σ1] < ∞.
Now setting Ω1 = CΣ1(Ω2), the centerizer of Ω2 in Σ1, we can easily see that
Ω1, Ω2 < Γ are commuting, non-amenable, icc subgroups.
Claim: [Σ1 : Ω1] < ∞ and [Γ : Ω1Ω2] < ∞.
First, we will show that [Σ1 : Ω1] < ∞. Assume by a contradiction that Ω1 has infinite
index in Σ1 and {hk} ⊂ Σ1 is an infinite sequence of representatives of distinct right
cosets of Ω1 in Σ1. Since [Γ : Ω2Σ1] < ∞, there is a right coset Ω2Σ1γ such that
Ω2Σ1γ ∩ Ω1hi
6= ∅. Then consider the subsequence {hi} of {hk}, for each i ≥ 1,
hi = xiγ for some xi ∈ Ω2Σ1. Then
56
hih−1
1 ∈ Ω2Σ1
1 = (xiγ)(x1γ)−1 = xiγγ−1x−1
1 = xix−1
for all i ≥ 2. Then for each i ≥ 2, It follows that hih−1
1 = ωiσi for some ωi ∈ Ω2
and σi ∈ Σ1. Notice that ωjσj = hjh−1
1
6= hih−1
1 = ωiσi. From the construction,
we have ωiΣ1 6= ωjΣ1 for all i 6= j It is easy to check that OΩ2Σ1(ωi) < ∞ for all
i ≥ 2. Also since [Γ : Ω2Σ1] < ∞, it implies further that OΓ(ωi) < ∞. However,
it contradicts the assumption Γ is icc. Hence [Σ1 : Ω1] < ∞. Furthermore, as a
consequence [Γ : Ω1Ω2] < ∞ as well.
Also notice that since CΓ(Σ1) ⊂ Ω2, we have L(Σ1)′ ∩ M ⊂ L(Ω2) and by
relation (6.24) we have
vL(Ω1)v∗ ⊂ M µ
1 p.
Since L(Ω2) and M 1/µ
2
are factors then as in the proof of [OP03, Proposition 12], we
can find partial isomoetries
w1, . . . , wm ∈ L(Ω2) and
u1, . . . , um ∈ M 1/µ
2
satisfying
wiwi
∗ = q′ ≤ q, ui
∗ui = p′ = uq′u∗ ≤ p for any i
and
Pj wj
∗wj = 1L(Ω2), Pj ujuj
∗ = 1M 1/µ
2
.
Combining with the above, we can check that u = Pj ujvwj ∈ M is a unitary
satisfying uL(Ω1)u∗ ⊂ M µ
1 . Since we know that M = M µ
1 ¯⊗M 1/µ
2
57
, it is forced that
M 1/µ
2 ⊂ u(L(Ω1)′ ∩ M)u∗.
(6.25)
Similarly, let Θ2 = {λ ∈ Γ OΩ1(λ) < ∞} and Θ1 = CΩ1(Θ2). As before it
follows that Θ1, Θ2 < Λ are commuting, non-amenable and icc subgroups such that
[Γ : Θ1Θ2] < ∞ and [Σ1 : Θ1] < ∞.
Moreover, Since CΓ(Ω1) ⊂ Θ2, by (6.25) we have
Since M = M µ
1 ¯⊗M 1/µ
2
M 1/µ
2 ⊂ uL(Θ2)u∗.
, by Theorem 2.6, there exists a subfactor B ⊂ M µ
1 such that
uL(Θ2)u∗ = B ¯⊗M 1/µ
2
.
Since M2
∼=com
M L(Σ2), we have uL(Σ2)u∗ ≺M M 1/µ
2
. Since [Ω2 : Σ2] < ∞, it follows
that uL(Ω2)u∗ ≺M M 1/µ
2
as well. Since B ⊂ uL(Ω2)u∗ we have that B ≺M M 1/µ
2
.
However since B ⊂ M µ
1 and M = M µ
1 ¯⊗M 1/µ
2
, these force that B has an atomic corner.
As B is a factor, then we get
B = Mk(C),
for some k ∈ N.
Altogether, we have
uL(Θ2)u∗ = B ¯⊗M 1/µ
2 = Mk(C) ¯⊗M 1/µ
2 = M t
2,
where t = k/µ. Since M = M 1/t
1 ¯⊗M t
2, we also get
u(L(Θ2)′ ∩ M)u∗ = M 1/t
1
.
(6.26)
(6.27)
58
Let Γ1 = {λ ∈ Γ OΘ2(λ) < ∞} and since Θ2 is an icc group, it follows that
Γ1 ∩ Θ2 = {1}. By construction as CΓ(Θ2) ⊂ Γ1, we obtain uL(Γ1)u∗ ⊇ u(L(Θ2)′ ∩
M)u∗ = M 1/t
1
. Therefore, agian applying Theorem 2.6, we have that
uL(Γ1)u∗ = A ¯⊗M 1/t
1 = A ¯⊗u(L(Θ2)′ ∩ M)u∗,
for some subfactor A ⊂ uL(Θ2)u∗.
In particular, we have A = uL(Γ1)u∗∩uL(Θ2)u∗ = C1 since Γ1∩Θ2 = {1} and,
hence uL(Γ1)u∗ = u(L(Θ2)′ ∩ M)u∗. Letting Γ2 = Θ2, it follows that the subgroups
Γ1 and Γ2 are commuting, non-amenable subgroups of Γ such that Γ1 ∩ Γ2 = {1},
Γ1Γ2 = Γ. And from equation (6.26) and (6.27) above, uL(Γ1)u∗ = M 1/t
1
, and
uL(Γ2)u∗ = M t
2.
6.3 Classification of tensor product decompositions of II1 factors arising
from groups
Motivated by the prior work [CdSS15], Drimbe, Hoff and Ioana have discovered
in [DHI16] a new classification result in the study of tensor product decompositions of
II1 factors. Specifically they unveiled the first examples of icc groups Γ for which all
diffuse tensor product decompositions of L(Γ) are "paramatrized" by the canonical
direct product decompositions of the underlying group Γ. Their examples include
remarkable groups such as the class of all icc groups Γ that are measure equivalent
to products of non-elementary hyperbolic groups. Similar results where obtained
subsequently in [CdSS17, dSP17]. In this dissertation we obtained similar results for
new classes of groups including amalgamated free products, direct products of wreath
59
product groups and MsDuff's groups. For the ease of presentation the results will be
presented in independent subsections.
6.3.1 Amalgamated free product groups
In Section 6.1 we have seen that for a large class of AFP von Neumann alge-
bras M = M1 ∗P M2 all their tensor factorizations essentially split P and the entire
inclusions P ⊂ Mi. However in the particular case when M arises for amalgam groups
Γ = Γ1 ∗Σ Γ2 this is insufficient to determine whether this further splits the group Σ as
well. In fact it is well known this does not happen all the time (see the Remark after
the Theorem 6.11) and hence a separate analysis is required to understand this aspect.
In this direction we isolate several situations when indeed the tensor decompositions
arise from the direct product splittings of Γ. One instance is when the algebra L(Σ)
is virtually prime 3.
Before stating our result we need a group theoretic preliminary.
Lemma 6.10. Let Γ = Γ1 ∗Σ Γ2 be an amalgamated free product. Suppose Γ = Λ1 ×Λ2
for some subgroups Λ1, Λ2. Then we can find a permutation σ ∈ S2 satisfying
• Σ = Λσ(1) × Σ0,
• Γ1 = Λσ(1) × Γ0
1,
• Γ2 = Λσ(1) × Γ0
2,
• Λσ(2) = Γ0
1 ∗Σ0 Γ0
2.
Proof. Since Γ = Γ1 ∗Σ Γ2, considering the von Neumann algebra of Γ, we have
3See Definition 4.4.
60
L(Γ) = L(Γ1) ∗L(Σ) L(Γ2). By using Theorem 6.1, we have
L(Γσ(1)) ≺L(Γ) L(Σ)
for some σ ∈ S2. Since Γ = Λ1 × Λ2, by applying Theorem 3.4, there is an element
h ∈ Γ so that [Λσ(1) : hΣh−1 ∩ Λσ(1)] < ∞. Since Λσ(1) is normal in Γ, conjugating by
h we can assume that [Λσ(1) : Σ ∩ Λσ(1)] < ∞. Also passing through a finite index
subgroup, we can also assume Σ ∩ Λσ(1) is normal in Γ. Therefore, we have
Γ(cid:14)(Σ ∩ Λσ(1)) = (Γ1(cid:14)(Σ ∩ Λσ(1))) ∗
Σ(cid:14)(Σ∩Λσ(1))
= Λσ(1)(cid:14)(Σ ∩ Λσ(1)) × Λσ(2).
(Γ2(cid:14)(Σ ∩ Λσ(1)))
Since Λσ(1)(cid:14)(Σ ∩ Λσ(1)) is finite, [KS70, Theorem 10] implies that
Λσ(1)(cid:14)(Σ ∩ Λσ(1)) < Σ(cid:14)(Σ ∩ Λσ(1))
and thus Λσ(1) < Σ. Since Λσ(1) < Σ ⊂ Λ1 × Λ2 and clearly Λσ(1) is normal in Λ1 × Λ2,
there is a subgroup Σ0 of Σ such that Σ = Λσ(1) × Σ0. With the same argument,
since Σ < Γ1, Γ2 < Λ1 × Λ2, for i = 1, 2 there are subgroups Γ0
i < Γi such that
Γi = Λσ(1) × Γ0
i . Moreover,
Λσ(1) × Λσ(2) = Γ = Γ1 ∗Σ Γ2
= (Λσ(1) × Γ0
1) ∗(Λσ(1)×(Σ0) Λσ(1) × Γ0
2)
= Λσ(1) × (Γ0
1 ∗Σ0 Γ0
2).
Hence, we can conclude that Λσ(2) = Γ0
1 ∗Σ0 Γ0
2.
61
Theorem 6.11. Let Γ = Γ1 ∗Σ Γ2 be an icc group with [Γ1 : Σ] ≥ 2 and [Γ2 : Σ] ≥ 3.
Assume that Σ is finite-by-icc and any corner of L(Σ) is virtually prime. Suppose
that L(Γ) = M1 ¯⊗M2 for diffuse Mi's. Then there exist direct product decompositions
Σ = Ω × Σ0, Γ1 = Ω × Γ0
1, and Γ2 = Ω × Γ0
2
with Σ0 finite, for some groups Σ0 < Γ0
1, Γ0
2, and hence Γ = Σ×(Γ0
1 ∗Σ0 Γ0
2). Moreover,
there exist a unitary u ∈ L(Γ), a scalar t > 0 and σ ∈ S2 such that
Mσ(1) = uL(Ω)tu∗
and Mσ(2) = uL(Γ0
1 ∗Σ0 Γ0
2)1/tu∗.
Proof. Since M1 ¯⊗M2 = L(Γ), by Corollary 6.2 we can assume Mσ(1) ≺ L(Σ). Since
any corner of L(Σ) is virtually prime then by Lemma 6.5 we must have
Mσ(1)
∼=com
M L(Σ),
and further applying Theorems 6.8 and 6.9 there exist infinite groups Λi so that
Γ = Λ1 × Λ2. Thus the desired conclusion follows by using Lemma 6.10.
Remark. The previous theorem illustrates a situation when a true von Neumann
algebraic counterpart of Lemma 6.10 could be successfully obtained. However, if
one drops the primeness assumption on L(Σ), the conclusion of the theorem is no
longer true. Precisely, there are icc amalgams Γ = Γ1 ∗Σ Γ2 whose group factors L(Γ)
admit non-canonical tensor product decompositions while Γ is indecomposable as a
nontrivial direct product. For instance, consider a group inclusion Σ < Ω satisfying
the following conditions:
i) [Jo98] for each finite E ⊂ Ω there are γ, λ ∈ Σ so that
[γ, E] = [λ, E] = 1
and
[γ, λ] 6= 1;
62
ii) for each γ ∈ Σ there is λ ∈ Ω so that [γ, λ] 6= 1.
Concrete such examples are Σ = ⊕S∞H < Ω = ∪n∈N(H ≀ Sn), where H is any icc
group and S∞ is the group of finite permutations of N.
Then the inclusion Σ < Γ = Ω ∗Σ Ω still satisfies i) and by [Jo98, Proposition
2.4] L(Γ) is McDuff so L(Γ) = L(Γ) ¯⊗R, where R is the hyperfinite factor. On the
other hand, combining Lemma 6.10 with ii) one can see that Γ cannot be written as
a nontrivial direct product.
6.3.2 Direct product of wreath product groups
Throughout this section, we denote by WR, the class of generalized wreath
product groups in the form Γ = A ≀I G, where G is a group acting on a set I, A is an
amenable group whose stabilizers StabΓ(i) are finite for all i ∈ I.
For further use we recall the following result, which is a particular case of
[IPV10, Corollary 4.3].
Theorem 6.12 ([IPV10]). Let Γ = A ≀I Γ0 ∈ WR and let B be a finite von Neumann
algebra B. Denote by M = B ¯⊗L(Γ) the corresponding tensor product algebra. Let
P1, P2 ∈ pMp be two commuting von Neumann subalgebras such that P1 ∨ P2 ⊂ pMp
is a finite index inclusion, Then either
i) there exists a nonzero p0 ∈ P ′
1 ∩ pMp such that P1p0 is amenable relative to B
or
ii) P2 ≺M B
Proof. Apply [IPV10, Corollary 4.3], one of the following must hold:
(1) There exists p1 ∈ (P1)′ ∩ M such that (P1)p1 is amenable relative to B inside
63
M;
(2) P2 ≺M B;
(3) P1 ∨ (P ′
1 ∩ pMp) ≺M M ¯⊗L(AI).
To finish the proof we only need to show that (3) does not hold. Assuming
by contradiction it holds then, since P2 ⊂ P ′
1 ∩ pMp, we have P1 ∨ P2 ≺M B ¯⊗L(AI).
Together with the assumption that P1 ∨ P2 ⊂ pMp has finite index, these imply that
pMp ≺M B ¯⊗L(AI). This further implies that B ≺B L(AI) which is a contradiction.
Notation. Let Γ1, Γ2, . . . , Γn be groups and let Γ = Γ1 ×Γ2 ×· · ·×Γn the correspond-
ing n-folded direct product. For every subset I ⊂ {1, 2, . . . , n} we will be denoting
by ΓI < Γ the subproduct groups supported on I, i.e. ΓI = Πi∈IΓi.
Next we present the main result of the section which classify all tensor product
decompositions of II1 factors associated with n-folded products of wreath product
groups. In particular our result generalizes the unique prime decompositions results
for such factors obtained by Sizemore and Winchester [SW11].
Theorem 6.13. Let Γ1, Γ2, . . . , Γn ∈ WR and let Γ = Γ1 × Γ2 × · · · × Γn. Consider
the corresponding von Neumann algebra M = L(Γ) and let P1, P2 be non-amenable
II1 factors such that M = P1 ¯⊗P2. Then there exist a scalar t > 0 and a partition
64
I1 ⊔ I2 = {1, 2, . . . , n} such that
L(ΓI1) ∼= P t
1
and L(ΓI2) ∼= P 1/t
2
.
Proof. Pick I1, I2 ⊂ {1, 2, . . . , n} be minimal (nonempty) subsets so that P1 ≺M
L(ΓI1) and P2 ≺M L(ΓI2). Next we argue that I1 $ {1, 2, . . . , n} and I2 $ {1, 2, . . . , n}.
We will only show the first statement as the second will follow similarly. Fix i ∈
{1, 2, . . . , n}. Write M = L(Γi) ⊗ L(Γi) where Γi := Γ{1,...,n}\{i} and using Theorem
6.12 for Bi = L(Γi) we have that either
(a) P1 ≺M L(Γi) or
(b) P2 ⊗ pi is amenable relative to L(Γi) inside M for some nonzero projection
pi ∈ P1.
Notice that using Lemma 5.6 (2), since P2 is a factor, case (b) above is equivalent to
(b') P2 is amenable relative to L(Γi) inside M.
Assume by contradiction that for all i ∈ {1, . . . , n} we have only case (b'). Since
EL(Γi) ◦ EL(Γj ) = EL(Γj ) ◦ EL(Γi) for all i, j and L(Γj) ⊂ M is regular, by using
Proposition 5.5 inductively we have that P2 is amenable relative to Tn
inside M. In particular, this implies that P2 is amenable which contradicts the initial
i=1 L(Γi) = C1
assumption. Therefore, there exists an io ∈ {1, . . . , n} such that P1 ≺M L(Γi0). In
particular this show that I1 ⊂ {1, . . . , n}\{i0}. Similarly we have that I2 $ {1, . . . , n}.
Next we prove the following
P1
∼=com
M L(ΓI1).
(6.28)
To see this recall that P1 ≺M L(ΓI1). Since P1 ∨ P2 = M and ΓI1 is icc, by
65
using Lemma 6.5 one of the followings must hold:
(a) P1
∼=com
M L(ΓI1), or
(b) there exist nonzero projections p1 ∈ P1, q1 ∈ L(ΓI1), a nonzero partial isometry
v ∈ q1Mp1, and a ∗-isomorphism ψ : p1P1p1 → Q ⊂ q1L(ΓI1)q1 such that
(i) ψ(x)v = vx for x ∈ p1P1p1;
(ii) Q and Q′ ∩ q1L(ΓI1)q1 are II1 factors so that Q ∨ (Q′ ∩ q1L(ΓI1)q1) ⊂
q1L(ΓI1)q1 has finite index;
(iii) s(EL(ΓI1 )(vv∗)) = q1.
So to show (6.28) we only need to argue that the case (b) above does not hold. Assume
by contradiction it does. As it is well-known that the algebras L(Γi) are prime for all
i ∈ {1, . . . , n} (see for instance [Po07, 6.4]), the part (ii) above implies that I1 ≥ 2.
Fix j ∈ I1. From (ii) we have that Q ∨ (Q′ ∩ q1L(ΓI1)q1) ⊂ q1L(ΓI1)q1 has finite index,
and hence using Theorem 6.12 we have that either
(c) Q ≺q1L(ΓI1 )q1 L(ΓI1\{j}), or
(d) there exists a nonzero projection p0 ∈ (Q′ ∩ q1L(ΓI1)q1)′ ∩ q1L(ΓI1)q1 such that
(Q′ ∩ q1L(ΓI1)q1)p0 is amenable relative to L(ΓI1\{j}) inside L(ΓI1).
Since Q ∨ (Q′ ∩ q1L(ΓI1)q1) is a factor, one can easily see that the inclusion Q ∨ Q′ ∩
q1L(ΓI1)q1 ⊂ q1L(ΓI1)q1 is irreducible 4; in particular the normalizer satisfies that
(cid:0)Nq1L(ΓI1 )q1(Q′ ∩ q1L(ΓI1)q1)(cid:1)′ ∩ q1L(ΓI1)q1 = C1.
4A subfactor of finite index N ⊂ M is said to be irreducible if the relative commutant
N ′ ∩ M = C.
66
Hence, using Lemma 5.6 we see that the condition (d) is equivalent to
(d') Q′ ∩ q1L(ΓI1)q1 is amenable relative to L(ΓI1\{j}) inside L(ΓI1).
Assume that for every j ∈ I1 only the possibility (d') holds. Since EL(ΓI1\{j1}) ◦
EL(ΓI1 \{j2}) = EL(ΓI1\{j2}) ◦ EL(ΓI1\{j1}) for all j1, j2 ∈ I1 and L(ΓI1\{j}) are regular
in L(ΓI1) then applying Proposition 5.5 inductively we get that Q′ ∩ q1L(ΓI1)q1 is
amenable relative to Tj∈I1
L(ΓI1\{j}) = C1. It follows that Q′ ∩ q1L(ΓI1)q1 is isomor-
phic to the hyperfinite II1 factor. In particular, Q ∨ (Q′ ∩ q1L(ΓI1)q1) is a factors with
McDuff's property. In particular, it has property Gamma of Murray-von Neumann.
Since Q ∨ (Q′ ∩ q1L(ΓI1)q1) ⊂ q1L(ΓI1)q1 has finite index, it follows from [PP86,
Proposition 1.11] that q1L(ΓI1)q1 has property Gamma as well. Therefore, for every
ω non-principal ultrafilter on N we have that
L(ΓI1)′ ∩ L(ΓI1)ω 6= C1.
(6.29)
Thus L(ΓI1) has property Gamma. Notice that L(ΓI1) = L(ΓI1\{j}) ¯⊗L(Γj) and using
both Example 1.4 c 5and Theorem 3.16 in [CSU13] we have that
L(ΓI1)′ ∩ L(ΓI1)ω ⊂ L(ΓI1\{j})ω ∨ L(Γj).
Since this holds for all j ∈ I1 then we have that
L(ΓI1)′ ∩ L(ΓI1)ω ⊂ \j∈I1(cid:0)L(ΓI1\{j})ω ∨ L(Γj)(cid:1).
5 Let H, Γ be countably infinite discrete group, let G y I , and consider the generalized
wreath product group H ≀I Γ := (⊕I H) ⋊ Γ.. Let G := {StabΓi i ∈ I}. We have this group
statisfies condition NC with respect to G
6 See Theorem 2.14
67
But by using the same argument from [CP10, Corollary 1.2] one can check that
Tj∈I1(cid:0)L(ΓI1\{j})ω ∨ L(Γj)(cid:1) = L(ΓI1) and hence
L(ΓI1)′ ∩ L(ΓI1)ω ⊂ L(ΓI1) ∩ L(ΓI1)′ = C1
which is a contradiction to (6.29). Thus there must exist j0 ∈ I1 such that Q ≺L(ΓI1 )
L(ΓI1\{j0}). It follows that there exists nonzero projections r ∈ Q, t ∈ L(ΓI1\{j0})
and a nonzero partial isometry w ∈ tL(ΓI1)r and an injective ∗-homomorphism Φ :
rQr → tL(ΓI1\{j0})t such that
Φ(y)w = wy
for
y ∈ rQr.
(6.30)
Since ψ is an isomorphism, there is a nonzero projection p0 ∈ P1 such that ψ(p0) = r.
Thus the relation (i) implies that
ψ(x)v = vx for x ∈ p0P1p0.
(6.31)
Applying (6.31) in (6.30), we see that for all x ∈ P1 we have that
Φ(ψ(x))wv = wψ(x)v = wvx.
(6.32)
Next we argue that
wv 6= 0.
(6.33)
Assume by contradiction that wv = 0. Thus wvv∗ = 0 end hence
0 = EL(ΓI1 )(wvv∗) = wEL(Γ1)(vv∗).
But this implies that 0 = w s(EL(ΓI1 )(vv∗)), where s(EL(ΓI1 )(vv∗)) is the support
projection of EL(ΓI1 )(vv∗). By (iii) we get 0 = wq1 and since by construction r ≤ q1
68
and w ∈ tL(ΓI1)r then we get wq1 = w, hence w = 0 which is a contradiction. This
proves (6.33).
Therefore wv 6= 0 and by taking the polar decomposition of wv = w0wv, we
see that (6.32) implies
Φ ◦ ψ(x)w0 = w0x for all x ∈ p0P1p0.
(6.34)
Since Φ ◦ ψ : p0P1p0 → tL(ΓI1\{j0})t is a ∗-homomorphism, it follows that P1 ≺L(ΓI1 )
L(ΓI1\{j0}) but this contradicts the minimality of I1 and therefore we have reached a
contradiction. As a consequence, case (b) does not hold altogether.
Using relation (6.28) and Theorem 6.8 there exist a subgroup Ω 6 CΓ(ΓI1) =
ΓI\I1 such that Ω × ΓI1 6 Γ is finite index and P ∼=com
M L(Ω). Hence by Theorem 6.9
we conclude that there exist Γ1 × Γ2 = Γ a product decomposition and a scalar t > 0
and a unitary u ∈ U(M) such that
L(Γ1) = uP t
1u∗
and L(Γ2) = uP 1/t
2 u∗.
Moreover, it is implicit in the proof of Theorem 6.9 that ΓI1 is commensurable to Γ1
and Γ{1,...,n}\I1 = ΓI2 is commensurable to Γ2. It only remains to argue that ΓI1 = Γ1
and ΓI2 = Γ2 which follows from basic group theoretic considerations.
6.3.3 McDuff's group functors T0 and T1
In this subsection we establish tensor product decomposition results for II1
factors associated with groups that arise via T0, T1-group functorial constructions
introduced by D. McDuff in [Mc69]. Before doing so we recall those notations from
69
[Mc69]. These constructions are inspired by the earlier work of Dixmier and Lance
[DL69] which in turn go back to the pioneering work of Murray and von Neumann
[MvN43].
Let Γ be a group. For i ≥ 1, let Γi be isomorphic copies of Γ and Λi be
isomorphic to Z. Define Γ =Li≥1 Γi and let S∞ be the group of finite permutations
of the positive integers N. Consider the semidirect product Γ ⋊ S∞ associated to
the natural action of S∞ on Γ which permutes the copies of Γ. Following [Mc69] we
define
• T0(Γ) = the group generated by Γ and Λi, i ≥ 1 with the only relation that Γi
and Λj commutes for i ≥ j ≥ 1.
• T1(Γ) = the group generated by Γ ⋊ S∞ and Λi, i ≥ 1 with the only relation
that Γi and Λj commute for i ≥ j ≥ 1.
Using a basic iterative procedure, these famous functorial group constructions
were used to provide the first infinite family of non-isomorphic II1 factors, the so
called L(Kα(Γ))'s where α ∈ {0, 1}N. One key feature, which also played a crucial
role in McDuff's work, is that the corresponding group factors L(Tα(Γ)) possess lots of
central sequences. In particular these algebras have McDuff property, i.e. L(Tα(Γ)) ∼=
L(Tα(Γ)) ¯⊗R, where R is the hyperfinite II1 factor. However we will prove below that
these are the only possible tensor decompositions. Specifically we have the following
type of unique prime factorization result
Theorem 6.14. Fix Γ a non-amenable group and let α ∈ {0, 1}.
If L(Tα(Γ)) =
P1 ¯⊗P2 then either P1 or P2 is isomorphic to the hyperfinite II1 factor.
70
Proof. First denote by Γn := ⊕i≥nΓi. Let α = 0 and define
• Σn 6 T0(Γ) be the subgroup generated by Γ, Λ1, Λ2, . . . , Λn;
• ∆n 6 T0(Γ) be the subgroup generated by Γn, Λn+1, Λn+2, . . ..
Similarly in the case of α = 1, we define
• Σn 6 T1(Γ) is the subgroup generated by Γ ⋊ S∞, Λ1, Λ2, . . . , Λn;
• ∆n 6 T1(Γ) is the subgroup generated by Γn, Λn+1, Λn+2.
In both cases, one can check that
Tα(Γ) = Σn ∗Γn ∆n. Thus, L(Tα(Γ)) = L(Σn) ∗L(Γn) L(∆n).
And we denote by Σ′
i=1 Γi) ∨ Λ1 ∨ Λ2 ∨ · · · ∨ Λn < Σn.
n := (Ln−1
Now let M = L(T0(Γ)) = P1 ¯⊗P2. Then by Theorem 6.1 there exist i ∈ {1, 2}
such that Pi ≺M L(Γn). Since Pi are factor, we have
Pi ≺s
M L(Γn).
(6.35)
Next denote by Qn := L(Γn) and Mn := L(Σ′
n). With these notations at hand we
show the followings hold.
lim
n→∞
kx − EMn(x)k2 = 0 for all x ∈ M.
(6.36)
as Qn − Mn bimodules we have QnL2(M)Mn ≺ QnL2(Qn) ¯⊗L2(M)Mn.(6.37)
To justify these statements notice first, since Σ′
i=1 Γi) ∨ Λ1 ∨ Λ2 ∨ · · · ∨
n րSn≥1 Σ′
n = T0(Γ) and hence M = L(T0(Γ)) =Sn L(Σ′
. This clearly shows (6.36).
SOT
=
n)
Λn, then clearly Σ′
SOT
Sn Mn
n := (Ln−1
Now we show (6.37). As before we have that T0(Γ) = Σn ∗Γn ∆n. Notice that
71
Σn = Σ′
n × Γn. Fix F a set of left coset representatives for Σ′
n in Γ and we isolate the
following subsets of F :
F1 = {w w = a1b1a2b2 . . . akbk or b1a2b2 . . . akbk where ai ∈ Σn \ Γn, bi ∈ ∆n \ Γn};
F0 = {w w ∈ Γn}.
We can check that F1 ⊔ F0 = F .
Next we prove that if Γnw1Σ′
n = Γnw2Σ′
n for w1, w2 ∈ F1, then w−1
2 w1 ∈ Γn.
Indeed, let m1, m2 ∈ Γ′
n, k1, k2 ∈ Σ′
n such that
m1w1k1 = m2w2k2. Thus m1w1k1k−1
2 w−1
2 m−1
2 = 1.
As wi = . . . a(i)
k b(i)
k where a(i)
k ∈ Σn \ Γn and b(i)
k ∈ ∆n \ Γn, we see that the previous
equation implies that
m1 . . . b(1)
k−1a(1)
k b(1)
k k1k−1
2 (b(2)
k )−1(a(2)
k )−1(b(2)
k−1)−1 . . . m−1
2 = 1
(6.38)
Consider the part b(1)
k k1k−1
2 (b(2)
k )−1 and notice that if k1k−1
2
6= 1 then k1k−1
2 ∈ Σ′
n \
{1} ⊂ Σ′
n \ Γn because Σ′
n ∩ Γn = {1}. Therefore, the left-hand side in (6.38)
is already in its reduced form so it cannot be trivial since it has alternating word
length at least 2. Thus k1k−1
2 = 1 which means k1 = k2 and m1w1 = m2w2 so
that w−1
2 w1 = m−1
1 m2 ∈ Γn. Moreover, observe that if w1, w2 ∈ F0, then clearly
2 w1 ∈ Γn
w−1
From above, on the set F we can introduce the following equivalence relation:
w1 ∼ w2
if there exists an m ∈ Γn such that mw1 = w2.
72
Next let G be a transversal set for F(cid:14) ∼ , i.e., pick an element w in each equivalence
class of F / ∼. Note that T0(Γ) = ⊔w∈G ΓnwΣ′
n is the double coset decomposition.
Thus as Qn-Mn bimodules we have the following decomposition:
QnL2(M)Mn
∼=Mw∈G
QmuwMn
k·k2.
(6.39)
Next let K be a right cosets representatives for the inclusion Γn < T0(Γ). Thus as
Qn-Mn bimodules we have that
QnL2(M) ¯⊗L2(M)Mn
Qm(uk ⊗ uw)Mn
k·k2
∼= Mk∈K,w∈F
∼= Mk∈K,w∈G Mδ∼w
Qm(uk ⊗ uδ)Mn
k·k2!
(6.40)
Next we argue that argue that for all w ∈ G, δ ∈ F and k ∈ K we have that
QnQnuwMn
k·k2
Mn
∼= QnL2(Qn) ⊗ L2(Mn)Mn
∼= QnQm(uk ⊗ uδ)Mn
k·k2
Mn
(6.41)
as Qn-Mn-bimodules.
To see the first part of (6.41) fix q1, q2 ∈ Qn and n1, n2 ∈ Mn and notice that
hq1uwn1, q2uwn2i = τ (q1uwn1n∗
2uw−1q∗
2)
= τ (q1uwEQn(n1n∗
2)uw−1q∗
2)
= τ (n1n∗
2)τ (q1uwuw−1q∗
2)
= τ (n1n∗
2)τ (q1q∗
2)
= hq1 ⊗ n1, q2 ⊗ n2i
This computation shows that the map quwn 7→ q ⊗ n induces an Qn-Mn-bimodules
isomorphism between QmuwMn
k·k2 and L2(Qn) ¯⊗L2(Mn).
73
The second part of (6.41) follows in a similar manner as the map quk ⊗ uδn 7→
q ⊗ n does the job. Indeed fixing q1, q2 ∈ Qn and n1, n2 ∈ Mn we see that
hq1(uk ⊗ uδ)n1, q2(uk ⊗ uδ)n2i = hq1uk, q2ukihuδn1, uδn2i
= hq1, q2iL2(Qn)hn1, n2iL2(Mn)
= hq1 ⊗ n1, q2 ⊗ n2iL2(Qn)⊗L2(Mn).
Now combining relations (6.39), (6.41) and (6.40) we see that, as Qn-Mn bimodules
we have the following
QnL2(M)Mn
QnQnuwMn
k·k2
Mn
∼=Mw∈G
∼=Mw∈G
≺ Mk∈K,w∈G(cid:0)Mδ∼w
∼= Mk∈K,w∈G(cid:0)Mδ∼w
QnL2(Qn) ¯⊗L2(Mn)Mn
QnL2(Qn) ¯⊗L2(Mn)Mn(cid:1)
Mn(cid:1)
QnQn(uk ⊗ uδ)Mn
k·k2
∼= QnL2(Qn) ¯⊗L2(Mn)Mn.
This concludes the proof of (6.37).
Notice that relations (6.36) and (6.37) show that the conditions in Lemma 5.7
are satisfied. Since Pi ≺s
M L(Γn) by (6.35) then we have that Pi is amenable relative
to ∩nQn = C1. Thus, Pi is amenable and we are done. In the case α = 1 and can let
i=1)⋊Sn ∨Λ1 ∨Λ2 · · ·∨Λn and the same method above applies verbatim.
Σ′
n = (Ln
Notice that the previous theorem can be generalized by to the case of products
Ω = Ω1 × ... × Ωn of McDuff's groups Ωi = Tαi(Γ). Specifically it asserts that all
74
possible tensor splittings L(Ω) = P1 ¯⊗P2 occurs only in the "amenable rooms" around
the subproducts Γ of Γ. The proof follows essentially the same arguments as in the
proof Theorem 6.14 and is left to the reader.
Theorem 6.15. For n ≥ 2 and i ∈ {1, . . . , n}, fix Γi non-amenable groups. Let
αi ∈ {0, 1} and let Ωi = Tαi(Γi). Denote by Ω = Ω1 × Ω2 × · · · × Ωn and assume that
M = L(Ω) = P1 ¯⊗P2 where Pi are non-amenable factors. Then there exist i ∈ {1, 2}
and a subset I ( {1, 2, . . . , n} such that Pi is amenable relative to L(ΩI ) inside M.
REFERENCES
75
[AP] C. Anantharaman and S. Popa, An introduction to II1 factors, University of
California, Los Angeles.
[BO08] N. P. Brown, N. Ozawa, C∗-algebras and finite-dimensional approximations,
Graduate Studies in Mathematics, vol. 88, AMS, Providence, RI.
[Bo12] R. Boutonnet, On solid ergodicity for Gaussian action, J. Funct. Anal. 263
(2012), 1040 -- 1063.
[BHR12] R. Boutonnet, C. Houdayer, and S. Raum, Amalgamated free product type
III factors with at most one Cartan subalgebra, Compos. Math. 150 (2014), 143 --
174.
[BHV05] B. Bekka, P. de la Harpe, and A. Valette, Kazhdan property (T), New
Mathematical Monographs, vol. 11, CUP, Cambridge 2008.
[B00] B. Blackdar, Operator Algebras, Theory of C∗-algebras and von Neumann Al-
gebras
[CdSS15] I. Chifan, R. de Santiago, and T. Sinclair, W∗-rigidity for the von Neumann
algebras of products of hyperbolic groups, Geom. Funct. Anal. 26 (2016), 136 -- 159.
[CdSS17] I. Chifan, R. de Santiago, and W. Sucpikarnon, Tensor product decompo-
sitions ofII1 factors arising from extensions of amalgamated free product groups,
Comm. Math. Phys. 364 (2018), 1163 -- 1194.
[CH08] I. Chifan, C. Houdayer, Bass-Serre rigidity results in von Neumann algebras,
Duke Math. J. 153 (2010), 23 -- 54.
[CI17] I. Chifan, A. Ioana, Amalgamated free product rigidity for group von Neumann
algebras, Preprint. arXiv:1705.07350.
[Ch78] H. Choda, A Galois correspondence in a von Neumann algebra, Tohoku Math.
J. 30(1978), 491 -- 504.
[CI08] I. Chifan, A. Ioana, Ergodic subequivalence relations induced by a Bernoulli
action, Geom. Funct. Anal. 20 (2010), 53 -- 67.
[CIK13] I. Chifan, A. Ioana and Y. Kida, W ∗-superrigidity for arbitrary actions of
central quotients of braid groups, Math. Ann. 361 (2015), 563 -- 582.
76
[CKP14] I. Chifan, Y. Kida, S. Pant, Primeness Results for von Neumann Algebras
Associated with Surface Braid Groups, Int. Math. Res. Not. 16 (2016), 4807 -- 4848.
[Co76] A. Connes, Classification of injective factor, Ann. of Math. 101 (1976), 73 --
115.
[CP10] I. Chifan and J. Peterson, Some unique group measure space decomposition
results, Duke Math. J. 162 (2013), no. 11, 1923 -- 1966.
[CPS12] I. Chifan, S. Popa and O. Sizemore, Some OE and W ∗-rigidity results for
actions by wreath product groups, J. Funct. Anal. 263 (2012), 3422 -- 3448.
[CS11] I. Chifan, T. Sinclair, On the structural theory of II1factors of negatively
curved groups, Ann. Sci. ´Ec. Norm. Sup. 46 (2013), no. 1, 1 -- 34.
[CSU11] I. Chifan, T. Sinclair, and B. Udrea, On the structural theory of II1 factors
of negatively curved groups, II. Actions by product groups, Adv. Math. 245 (2013),
208 -- 236.
[CSU13] I. Chifan, T. Sinclair, B. Udrea, Inner amenability for groups and central
sequences in factors, Ergodic Theory Dynam. Systems 36 (2016), no. 4, 1106 --
1029.
[DHI16] D. Drimbe, D. Hoff, A. Ioana, Prime II1 factors arising from irreducible
lattices in products of rank one simple Lie groups, J. Reine Angew. Math. to
appear, arXiv:1611.02209.
[DI12] Y. Dabraowski and A. Ioana, Unbounded derivations, free dilations and inde-
composability results for II1 factors, Trans. Amer. Math. Soc. 368 (2016), no. 7,
4525 -- 4560.
[DL69] J. Dixmier and E.C. Lance, Deux nouveaux facteurs de type II1, Invent. Math.
7 (1969), 226 -- 234.
[dSP17] R. de Santiago, S. Pant, Classification of Tensor Decompositions of II1 Fac-
tors Associated With Poly-Hyperbolic Groups, arXiv:1802.09083.
[Fi10] P. Fima, A note on the von Neumann algebra of a Baumslag-Solitar group, C.
R. Acad. Sci. Paris, Ser. I 349 (2011), 25 -- 27.
[Ge96] L. Ge, On maximal injective subalgebras of factors, Adv. Math. 118 (1996),
no. 1, 34 -- 70.
[Ge98] L. Ge, Applications of Free Entropy to Finite von Neumann Algebras, II Ann,.
of Math. Second Series, 147, No. 1 (Jan., 1998), pp. 143 -- 157.
77
[HI15] C. Houdayer, Y. Isono, Unique prime factorization and the bicentralizer prob-
lem for a class of type III factors, Adv. Math. 305 (2017), 402 -- 455.
[Ho15] D. Hoff, von Neumann Algebras of Equivalence Relations with Nontrivial One-
Cohomology. J. Funct. Anal. 270 (2016), no. 4, 1501 -- 1536.
[HV12] C. Houdayer, S. Vaes. Type III factors with unique Cartan decomposition, J.
Math´ematiques Pures et Appliqu'ees 100 (2013), 564-590.
[Io10] A. Ioana, W ∗-superrigidity for Bernoulli actions of property (T) groups, J.
Amer. Math. Soc. 24 (2011), 1175 -- 1226.
[Io12] A. Ioana, Cartan subalgebras of amalgamated free product II1 factors, Ann. Sci.
´Ec. Norm. Sup.(4) 48 (2015), no. 1, 71 -- 130.
[IPP05] A. Ioana, J. Peterson, S. Popa, Amalgamated free products of w-rigid factors
and calculation of their symmetry groups. Acta Math. 200 (2008), 85 -- 153.
[IPV10] A. Ioana, S. Popa and S. Vaes, A Class of superrigid group von Neumann
algebras, Ann. of Math. (2) 178 (2013), 231 -- 286.
[Is14] Y. Isono, Some prime factorization results for free quantum group factors, J.
Reine Angew. Math. 722 (2017), 215 -- 250.
[Is16] Y. Isono, On fundamental groups of tensor product II1 factors, preprint
arXiv:1608.06426.
[IS19] A. Ioana and P. Spaas, II1 factors with exotic central sequence algebras
preprint, arXiv:1904.06816.
[Je13] J. Peterson, Notes on von Neumann algebras, Vanderbilt University, 2013.
[Jo81] V.F.R. Jones, Index for subfactors, Invent. Math. 72 (1983), 1 -- 25.
[Jo98] P. Jolissaint, Central Sequences in the Factor Associated with the Thompson
Group F, Annales de l'Institut Fourier 48 (1998), 1093 -- 1106.thm
[KS70] A. Karrass, D. Solitar, The subgroups of a free product of two groups with an
amalgamated subgroup, Trans. Amer. Math. Soc. 150 (1970), 227 -- 255.
[Mc69] D. McDuff, Central sequences and the hyperfinite factor, Proc. London Math.
Soc. 21 (1970), 443 -- 461.
[MvN36] F.J. Murray, J. von Neumann, On rings of operators, Ann. Math. 37 (1936),
116 -- 229.
78
[MvN43] F.J. Murray, J. von Neumann, Rings of operators IV, Ann. Math. 44 (1943),
716 -- 808.
[OP03] N. Ozawa, S. Popa, Some prime factorization results for type II1 factors,
Invent. Math. 156 (2004), 223 -- 234.
[OP07] N. Ozawa, S. Popa, On a class of II1 factors with at most one Cartan subal-
gebra, Ann. Math. 172 (2010), 713 -- 749.
[Oz03] N. Ozawa, Solid von Neumann algebras, Acta Math. 192 (2004), 111 -- 117.
[Oz04] N. Ozawa, A Kurosh-type theorem for type II1 factors, Int. Math. Res. Not.
(2006), Art. ID 97560, 21.
[Pe06] J. Peterson, L2-rigidity in von Neumann algebras, Invent. Math. 175 (2009)
no. 2, 417 -- 433.
[Po03] S. Popa, Strong rigidity of II1 factors arising from malleable actions of w-rigid
groups I, Invent. Math. 165 (2006), 369 -- 408.
[Po06] S. Popa, On Ozawa's property for free group factors, Int. Math. Res. Not.
(2007), no. 11, 10pp.
[Po07] S. Popa, On the Superrigidity of Malleable Actions with Spectral Gap, J. Am.
Math. Soc. 21(2008), no. 4, 981 -- 1000
[Po95] S. Popa, Classification of subfactors and their endomorphisms, CBMS Re-
gional Conference Series in Mathematics, vol. 86, Published for the Conference
Board of the Mathematical Sciences, Washington, DC; by the American Mathe-
matical Society, Providence, RI, 1995.
[Po83] S. Popa, Orthogonal pairs of ∗-subalgebras in finite von Neumann algebras, J.
Operator Th. 9 (1983), no. 2, 253 -- 268.
[PP86] M. Pimsner, S. Popa, Entropy and index for subfactors, Ann. Sci. ´Ecole Norm.
Sup. 19 (1986), 57 -- 106.
[PV11] S. Popa, S. Vaes, Unique Cartan decomposition for II1 factors arising from
arbitrary actions of free groups, Acta Math. 212 (2014), 141 -- 198.
[Si10] T. Sinclair, Strong solidity of group factors from lattices in SO(n, 1) and
SU(n, 1), J. Funct. Anal. 260 (2011), no. 11, 3209 -- 3221.
[SS08] A. Sinclair and R. Smith, Finite von Neumann Algebra and Masas
79
[SW11] J. O. Sizemore and A. Winchester, Unique prime decomposition results for
factors coming from wreath products, Pacific J. Math. 265 (2013), no. 1, 221 -- 232.
[Va08] S. Vaes, Explicit computations of all finite index bimodules for a family of II1
factors, Ann. Sci. ´Ec. Norm. Sup. 41 (2008), 743 -- 788.
[Va13] S. Vaes, Normalizers inside amalgamated free products von Neumann algebras,
Publ. Res. Inst. Math. Sci. 50 (2014), 695 -- 721.
|
1512.00346 | 2 | 1512 | 2017-01-03T01:00:15 | Quotients of Ultragraph C*-Algebras | [
"math.OA"
] | Let $\mathcal{G}$ be an ultragraph and let $C^*(\mathcal{G})$ be the associated $C^*$-algebra introduced by Mark Tomforde. For any gauge invariant ideal $I_{(H,B)}$ of $C^*(\mathcal{G})$, we analyze the structure of the quotient $C^*$-algebra $C^*(\mathcal{G})/I_{(H,B)}$. For simplicity's sake, we first introduce the notion of quotient ultragraph $\mathcal{G}/(H,B)$ and an associated $C^*$-algebra $C^*(\mathcal{G}/(H,B))$ such that $C^*(\mathcal{G}/(H,B))\cong C^*(\mathcal{G})/I_{(H,B)}$. We then prove the gauge invariant and the Cuntz-Krieger uniqueness theorems for $C^*(\mathcal{G}/(H,B))$ and describe primitive gauge invariant ideals of $C^*(\mathcal{G}/(H,B))$. | math.OA | math |
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
HOSSEIN LARKI
Abstract. Let G be an ultragraph and let C ∗(G) be the associated C ∗-
algebra introduced by Mark Tomforde. For any gauge invariant ideal
I(H,B) of C ∗(G), we analyze the structure of the quotient C ∗-algebra
C ∗(G)/I(H,B). For simplicity's sake, we first introduce the notion of quo-
tient ultragraph G/(H, B) and an associated C ∗-algebra C ∗(G/(H, B))
such that C ∗(G/(H, B)) ∼= C ∗(G)/I(H,B). We then prove the gauge in-
variant and the Cuntz-Krieger uniqueness theorems for C ∗(G/(H, B))
and describe primitive gauge invariant ideals of C ∗(G/(H, B)).
1. Introduction
In order to bring graph C ∗-algebras and Exel-Laca algebras together un-
der one theory, Tomforde introduced in [17] the notion of ultragraphs and
associated C ∗-algebras. An ultragraph is basically a directed graph in which
the range of each edge is allowed to be a nonempty set of vertices rather than
a single vertex. However, the class of ultragraph C ∗-algebras are strictly
lager than the graph C ∗-algebras as well as the Exel-Laca algebras (see [18,
Section 5]). Due to some similarities, parts of fundamental results for graph
C ∗-algebras, such as the Cuntz-Krieger and the gauge invariant uniqueness
theorems, simplicity, and K-theory computation have been extended to the
setting of ultragraphs [17, 18].
Furthermore, by constructing a specific topological quiver Q(G) from an
ultragraph G, Katsura et al. described some properties of the ultragraph
C ∗-algebra C ∗(G) using those of topological quivers [13]. In particular, they
showed that every gauge invariant ideal of C ∗(G) is of the form I(H,B) cor-
responding to an admissible pair (H, B) in G. Recall that the graph C ∗-
algebras (and also the ultragraph C ∗-algebras) are simple examples of the
topological quiver algebras [14] in which the sets of vertices and edges are
considered with discrete topology and Radon measures are special counting
measures. So, it seems that working with topological quiver algebras is more
complicated than with the ultragraph C ∗-algebras.
Recall that for any gauge invariant ideal I(H,B) of a graph C ∗-algebra
∼=
C ∗(E), there is a (quotient) graph E/(H, B) such that C ∗(E)/I(H,B)
C ∗(E/(H, B)) (see [2, 1]). So, the class of graph C ∗-algebras contains such
Date: September 1, 2018.
2010 Mathematics Subject Classification. 46L55.
Key words and phrases. Quotient ultragraph, ultragraph C ∗-algebra, graph C ∗-algebra,
ideal, primitive ideal.
1
2
HOSSEIN LARKI
quotients, and we may apply the results and properties of graph C ∗-algebras
for their quotients. For examples, some contexts such as simplicity, K-
theory, primitivity, and topological stable rank are directly related to the
structure of ideals and quotients.
Unlike the C ∗-algebras of graphs and topological quivers, the class of
ultragraph C ∗-algebras is not closed under quotients. This causes some ob-
stacles in studying the structure of ultragraph C ∗-algebras. The aims of
this article are to prove two kinds of uniqueness theorems, known as the
gauge invariant and the Cuntz-Krieger uniqueness theorems, for quotients
of ultragraph C ∗-algebras and apply them for analyzing the ideal struc-
ture of an ultragraph C ∗-algebra. Suppose that I(H,B) is a gauge invari-
ant ideal of an ultragraph C ∗-algebra C ∗(G). For the sake of convenience,
we first introduce the notion of quotient ultragraph G/(H, B) and a rela-
∼= C ∗(G/(H, B)) and
tive C ∗-algebra C ∗(G/(H, B)) such that C ∗(G)/I(H,B)
then prove the gauge invariant and Cuntz-Krieger uniqueness theorems for
C ∗(G/(H, B)). The C ∗-algebra C ∗(G/(H, B)) is called a quotient ultragraph
C ∗-algebra. The uniqueness theorems help us to show when a representation
of C ∗(G/(H, B)) is injective. We see that the structure of C ∗(G/(H, B)) is
close to that of graph and ultragraph C ∗-algebras. Hence, many traditional
graph C ∗-algebraic techniques may be extended to the ultragraph setting
by applying quotient ultragraphs (see Sections 6 and 7). We note that the
initial idea for defining quotient ultragraphs has been inspired from quo-
tient labelled graphs in [9]. However, many main results of this paper are
not known for labelled graphs. Moreover, any restrictive condition such as
set-finiteness or receiver set-finiteness does not be assumed here for ultra-
graphs.
This article is organized as follows. We begin in Section 2 by giving some
definitions and preliminaries about the ultragraphs and their C ∗-algebras
which will be used in the next sections. In Section 3, for any admissible pair
(H, B) in an ultragraph G, we introduce the quotient ultragraph G/(H, B)
and an associated C ∗-algebra C ∗(G/(H, B)). For this, the ultragraph G is
modified by an extended ultragraph G and we define an equivalent relation
∼ on G. Then the quotient ultragraph G/(H, B) is ultragraph G with the
equivalent classes {[A] : A ∈ G
}. In Section 4, by approaching with graph
C ∗-algebras, the gauge invariant and the Cuntz-Krieger uniqueness theorems
will be proved for the quotient ultragraphs C ∗-algebras.
0
In Section 5, we describe the gauge invariant ideals of C ∗(G/(H, B)). We
first prove that C ∗(G/(H, B)) is isometrically isomorphic to the quotient C ∗-
algebra C ∗(G)/I(H,B). We then see that every gauge invariant ideal I(K,S) in
C ∗(G) with K ⊇ H and K ∪ S ⊇ B induces an ideal J[K,S] in C ∗(G/(H, B))
and all gauge invariant ideals of C ∗(G/(H, B)) are of this form. In Sections
6 and 7, using quotient ultragraphs, we can apply some graph C ∗-algebra
methods for the ultragraph C ∗-algebras. In Section 6, we define Condition
(K) for G/(H, B) under which all ideals of C ∗(G/(H, B)) are gauge invariant
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
3
and the real rank of C ∗(G/(H, B)) is zero. As a corollary, we can recover
both [13, Proposition 7.3] and [12, Proposition 5.26] by a different approach.
Finally, in Section 7, all primitive gauge invariant ideals of ultragraph C ∗-
algebras will be characterized.
2. preliminaries
In this section, we review basic definitions and properties of ultragraph
C ∗-algebras which will be needed through the paper. For more details, we
refer the reader to [17] and [13].
Definition 2.1 ([17]). An ultragraph is a quadruple G = (G0, G1, rG, sG) con-
sisting of a countable vertex set G0, a countable edge set G1, the source map
sG : G1 → G0, and the range map rG : G1 → P(G0) \ {∅}, where P(G0) is
the collection of all subsets of G0. If rG(e) is a singleton vertex for each edge
e ∈ G1, then G is an ordinary (directed) graph.
For our convenience, we use the notation G0 of [13] rather than [17, 18].
For any set X, a nonempty subcollection of the power set P(X) is said to
be an algebra if it is closed under the set operations ∩, ∪, and \. If G is
an ultragraph, the smallest algebra in P(G0) containing {{v} : v ∈ G0} and
{rG(e) : e ∈ G1} is denoted by G0. We simply denote every singleton set {v}
by v. So, G0 may be considered as a subset of G0.
Definition 2.2. For each n ≥ 1, a path α of length α = n in G is a sequence
α = e1 . . . en of edges such that s(ei+1) ∈ r(ei) for 1 ≤ i ≤ n − 1. If also
s(e1) ∈ r(en), α is called a loop or a closed path. We write α0 for the set
{sG(ei) : 1 ≤ i ≤ n}. The elements of G0 are considered as the paths of
length zero. The set of all paths in G is denoted by G∗. We may naturally
extend the maps sG, rG on G∗ by defining sG(A) = rG(A) = A for A ∈ G0,
and rG(α) = rG(en), sG(α) = sG(e1) for each path α = e1 . . . en.
Definition 2.3 ([17]). Let G be an ultragraph. A Cuntz-Krieger G-family is
a set of partial isometries {se : e ∈ G1} with mutually orthogonal ranges
and a set of projections {pA : A ∈ G0} satisfying the following relations:
(1) p∅ = 0, pApB = pA∩B, and pA∪B = pA + pB − pA∩B for all A, B ∈ G0,
(2) s∗
(3) ses∗
ese = prG (e) for e ∈ G1,
e ≤ psG (e) for e ∈ G1, and
(4) pv =PsG (e)=v ses∗
e whenever 0 < s−1
G (v) < ∞.
The C ∗-algebra C ∗(G) of G is the (unique) C ∗-algebra generated by a uni-
versal Cuntz-Krieger G-family.
By [17, Remark 2.13], we have
C ∗(G) = span(cid:8)sαpAs∗
β : α, β ∈ G∗, A ∈ G0, and rG(α) ∩ rG(β) ∩ A 6= ∅(cid:9) ,
where sα := se1 . . . sen if α = e1 . . . en, and sα := pA if α = A.
4
HOSSEIN LARKI
Remark 2.4. As noted in [17, Section 3], every graph C ∗-algebra is an ultra-
graph C ∗-algebra. Recall that if E = (E0, E1, rE, sE) is a directed graph, a
collection {se, pv : v ∈ E0, e ∈ E1} containing mutually orthogonal projec-
tions pv and partial isometries se is called a Cuntz-Krieger E-family if
(1) s∗
(2) ses∗
ese = prE(e) for all e ∈ E1,
e ≤ psE(e) for all e ∈ E1, and
(3) pv =PsE(e)=v ses∗
E (v) < ∞.
We denote by C ∗(E) the universal C ∗-algebra generated by a Cuntz-Krieger
E-family.
e for every vertex v ∈ E0 with 0 < s−1
By the universal property, C ∗(G) admits the gauge action of the unit cir-
cle T. By an ideal, we mean a closed two-sided ideal. Using the properties
of quiver C ∗-algebras [13], the gauge invariant ideals of C ∗(G) were char-
acterized in [13, Theorem 6.12] via a one-to-one correspondence with the
admissible pairs of G.
Definition 2.5. A subset H ⊆ G0 is said to be hereditary if the following
properties holds:
(1) sG(e) ∈ H implies rG(e) ∈ H for all e ∈ G1.
(2) A ∪ B ∈ H for all A, B ∈ H.
(3) If A ∈ H, B ∈ G0, and B ⊆ A, then B ∈ H.
G (v) < ∞, then rG(s−1
Moreover, a subset H ⊆ G0 is called saturated if for any v ∈ G0 with 0 <
s−1
G (v)) ⊆ H implies v ∈ H. The saturated hereditary
closure of a subset H ⊆ G0 is the smallest hereditary and saturated subset
H of G0 containing H.
Let H be a saturated hereditary subset of G0. The set of breaking vertices
of H is denoted by
BH :=(cid:8)w ∈ G0 : s−1
G (w) = ∞ but 0 < rG(s−1
G (w)) ∩ (G0 \ H) < ∞(cid:9) .
Following [12], an admissible pair (H, B) in G is a saturated hereditary set
H ⊆ G0 together with a subset B ⊆ BH. For any admissible pair (H, B) in
G, we define the ideal I(H,B) of C ∗(G) generated by
{pA : A ∈ G0} ∪(cid:8)pH
w := pw −PsG(e)=w, rG (e) /∈H ses∗
where pH
e. Note that the ideal I(H,B) is gauge
invariant and [11, Theoerm 6.12] implies that every gauge invariant ideal I
of C ∗(G) is of the form I(H,B) for
w : w ∈ B(cid:9) ,
H := {A : pA ∈ I} and B :=(cid:8)w ∈ BH : pH
3. Quotient Ultragraphs and their C ∗-algebras
w ∈ I(cid:9) .
In this section, for any admissible pair (H, B) in an ultragraph G, we intro-
duce the quotient ultragraph G/(H, B) and its relative C ∗-algebra C ∗(G/(H, B)).
We will show in Proposition 5.1 that C ∗(G/(H, B)) is isometrically isomor-
phic to the quotient C ∗-algebra C ∗(G)/I(H,B).
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
5
Let us fix an ultragraph G = (G0, G0, rG, sG) and an admissible pair (H, B)
in G. For defining our quotient ultragraph G/(H, B), we first modify G by
an extended ultragraph G as follows. Add the vertices {w′ : w ∈ BH \ B} to
G0 and denote A := A ∪ {w′ : w ∈ A ∩ (BH \ B)} for each A ∈ G0. We now
define the new ultragraph G = (G0, G
, r′, s′) by
1
0
G
1
G
:= G0 ∪ {w′ : w ∈ BH \ B},
:= G1,
the source map
s′(e) :=(cid:26) (sG(e))′
sG(e)
if sG(e) ∈ BH \ B and rG(e) ∈ H
otherwise,
and the rang map r′(e) := rG(e) for every e ∈ G1. In Proposition 3.3 below,
we will see that the C ∗-algebras of G and G coincide.
Example 3.1. Suppose G is the ultragraph
e
e
g
u
w
(∞)
f
v
A
H
where (∞) indicates infinitely many edges. If H is the saturated hereditary
subset of G0 containing {v} and A, then we have BH = {w}. For B := ∅,
consider the admissible pair (H, ∅) in G. Then the ultragraph G associated
to (H, ∅) would be
e
u
e
g
e
w
w′
(∞)
f
v
A
H
Indeed, since BH \ B = {w}, for constructing G we first add a vertex w′ to
G. We then define
r′(f ) := A = A,
r′(e) := {v, w} = {v, w, w′}, and
r′(g) := {u} = {u}.
For the source map s′, for example, since sG(f ) ∈ BH \ B and rG(f ) ∈ H,
we may define s′(f ) := w′. Note that the range of each edge emitted by w′
belongs to H.
6
HOSSEIN LARKI
0
1
0
As usual, we write G
0
for the algebra generated by the elements of G
∪
}. Note that A = A for every A ∈ H, and hence, H would be
as well. Moreover, the set of breaking
H = B).
{r′(e) : e ∈ G
a saturated hereditary subset of G
vertices of H in G is B (meaning BG
Remark 3.2. Suppose that C ∗(G) is generated by a Cuntz-Krieger G-family
{se, pA : A ∈ G0, e ∈ G1}. If a family M = {Se, Pv, PA : v ∈ G0, A ∈ G0, e ∈
1
} in a C ∗-algebra X satisfies relations (1)-(4) in Definition 2.3, we may
G
generate a Cuntz-Krieger G-family N = {Se, PA : A ∈ G
} in X.
is the algebra generated by {v, w′, r′(e) : v ∈ G0, w ∈
For this, since G
BH \ B, e ∈ G
}, it suffices to define
, e ∈ G
0
0
1
1
PA∩B := PAPB
PA∪B := PA + PB − PAPB
PA\B := PA − PAPB
0
and generate projections PA for all A ∈ G
G-family in X, and the C ∗-subalgebras generated by M and N coincide.
. Then N is a Cuntz-Krieger
Proposition 3.3. Let G be an ultragraph, and let (H, B) be an admissible
pair in G. If G is the extended ultragraph as above, then C ∗(G) ∼= C ∗(G).
Proof. Suppose that C ∗(G) = C ∗(te, qA) and C ∗(G) = C ∗(se, pC).
define
If we
Pv := qv
tet∗
e
Pw :=PsG (e)=w
Pw′ := qw −PsG(e)=w
rG (e) /∈H
rG (e) /∈H
PA := qA
Se := te
for v ∈ G0 \ (BH \ B)
for w ∈ BH \ B
tet∗
e
for w ∈ BH \ B
0
for A ∈ G
1
for e ∈ G
then, by Remark 3.2, the family {Pv, Pw, Pw′, PA, Se} induces a Cuntz-
Krieger G-family in C ∗(G). Since all vertex projections of this family are
nonzero (which follows all set projections PA are nonzero for ∅ 6= A ∈ G
),
the gauge-invariant uniqueness theorem [17, Theorem 6.8] implies that the
∗-homomorphism φ : C ∗(G) → C ∗(G) with φ(p∗) = P∗ and φ(s∗) = S∗ is
injective. On the other hand, the family generates all C ∗(G), and hence, φ
is an isomorphism.
(cid:3)
0
To define a quotient ultragraph G/(H, B), we use the following equivalent
relation on G.
Definition 3.4. Suppose that (H, B) is an admissible pair in G, and that G
is the extended ultragraph as above. We define the relation ∼ on G
by
0
A ∼ B ⇐⇒ ∃V ∈ H such that A ∪ V = B ∪ V.
Note that A ∼ B if and only if both sets A \ B and B \ A belong to H.
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
7
Lemma 3.5. The relation ∼ is an equivalent relation on G
the operations
0
. Furthermore,
[A] ∪ [B] := [A ∪ B], [A] ∩ [B] := [A ∩ B], and [A] \ [B] := [A \ B]
are well-defined on the equivalent classes {[A] : A ∈ G
0
}.
Proof. It is straightforward to verify that the relation ∼ is reflexive, symmet-
ric, and associative. To see the second assertion, suppose that [A1] = [A2],
[B1] = [B2], and A1 ∪ V = A2 ∪ V , B1 ∪ W = B2 ∪ W for some V, W ∈ H.
The hereditary property of H yields V ∪ W ∈ H and we get
(A1 ∪ B1) ∪ (V ∪ W ) = (A2 ∪ B2) ∪ (V ∪ W )
=⇒
[A1 ∪ B1] = [A2 ∪ B2].
Also, we have
(A1 ∩ B1) ∪ (V ∪ W ) = (A1 ∪ (V ∪ W )) ∩ (B1 ∪ (V ∪ W ))
= (A2 ∪ (V ∪ W )) ∩ (B2 ∪ (V ∪ W ))
= (A2 ∩ B2) ∪ (V ∪ W )
=⇒
[A1 ∩ B1] = [A2 ∩ B2].
For the third operation, first note that B1 \ B2, B2 \ B1 ⊆ W because
B1 ∪ W = B2 ∪ W . Then we have
(A1 \ B1) ∪ W = (A1 \ ((B1 ∩ B2) ∪ (B1 \ B2))) ∪ W
= (A1 \ (B1 ∩ B2)) ∪ W
and similarly,
(A2 \ B2) ∪ W = (A2 \ (B1 ∩ B2)) ∪ W.
Now the fact A1 ∪ V = A2 ∪ V implies that
(A1 \ B1) ∪ (V ∪ W ) = (A1 \ (B1 ∩ B2)) ∪ (V ∪ W )
= ((A1 ∪ V ) \ (B1 ∩ B2)) ∪ (V ∪ W )
= ((A2 ∪ V ) \ (B1 ∩ B2)) ∪ (V ∪ W )
= (A2 \ B2) ∪ (V ∪ W ).
Therefore, we obtain [A1 \ B1] = [A2 \ B2], as desired.
(cid:3)
Definition 3.6. Let G be an ultragraph, let (H, B) be an admissible pair in
G, and consider the equivalent relation of Definition 3.4 on the extended
, r′, s′). The quotient ultragraph of G by (H, B) is the
ultragraph G = (G
quintuple G/(H, B) = (Φ(G0), Φ(G0), Φ(G1), r, s), where
, G
0
1
Φ(G0) :=(cid:8)[v] : v ∈ G0 \ H(cid:9) ∪(cid:8)[w′] : w ∈ BH \ B(cid:9) ,
Φ(G0) :=n[A] : A ∈ G
Φ(G1) :=ne ∈ G
0o ,
: r′(e) /∈ Ho ,
1
8
HOSSEIN LARKI
and r : Φ(G1) → Φ(G0), s : Φ(G1) → Φ(G0) are the range and source maps
defined by
r(e) = [r′(e)]
and
s(e) := [s′(e)].
We refer to Φ(G0) as the vertices of G/(H, B).
Remark 3.7. Lemma 3.5 implies that Φ(G0) is the smallest algebra contain-
ing
(cid:8)[v], [w′] : v ∈ G0 \ H, w ∈ BH \ B(cid:9) ∪n[r′(e)] : e ∈ G
1o .
Notation.
(1) For every vertex v ∈ G
(2) For A, B ∈ G
(3) Through the paper, we will denote the range and the source maps
, we write [A] ⊆ [B] whenever [A] ∩ [B] = [A].
\ H, we usually denote [v] instead of [{v}].
0
0
of G by rG, sG, those of G by r′, s′, and those of G/(H, B) by r, s.
Now we introduce representations of quotient ultragraphs and their rela-
tive C ∗-algebras.
Definition 3.8. Let G/(H, B) be a quotient ultragraph. A representation
of G/(H, B) is a set of partial isometries {Te : e ∈ Φ(G1)} and a set of
projections {Q[A] : [A] ∈ Φ(G0)} which satisfy the following relations:
(1) Q[∅] = 0, Q[A∩B] = Q[A]Q[B], and Q[A∪B] = Q[A] + Q[B] − Q[A∩B].
(2) T ∗
(3) TeT ∗
e Te = Qr(e) and T ∗
e Tf = 0 when e 6= f .
e ≤ Qs(e).
e , whenever 0 < s−1([v]) < ∞.
(4) Q[v] =Ps(e)=[v] TeT ∗
The C ∗-algebra C ∗(G/(H, B)) of G/(H, B) is the universal C ∗-algebra gen-
erated by a representation {te, q[A] : [A] ∈ Φ(G0), e ∈ Φ(G1)} which exists
by Theorem 3.11 below.
Remark 3.9. Recall that the universality of C ∗(G/(H, B)) means that if
{Te, Q[A]} is a representation of G/(H, B) in a C ∗-algebra X, then there
exists a ∗-homomorphism φ : C ∗(G/(H, B)) → X such that φ(q[A]) = Q[A]
and φ(te) = Te for all [A] ∈ Φ(G0) and e ∈ Φ(G1).
Note that if α = e1 . . . en is a path in G and r′(α) /∈ H, then the hereditary
property of H yields r′(ei) /∈ H, and so ei ∈ Φ(G1) for all 1 ≤ i ≤ n. In this
case, we denote tα := te1 . . . ten. Moreover, we define
∗
(G/(H, B))∗ := {[A] : [A] 6= [∅]} ∪nα ∈ G
: r(α) 6= [∅]o
as the set of finite paths in G/(H, B) and we can extend the maps s, r on
(G/(H, B))∗ by setting
s([A]) := r([A]) := [A] and s(α) := s(e1), r(α) := r(en).
The proof of next lemma is similar to the arguments of [17, Lemmas 2.8
and 2.9].
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
9
Lemma 3.10. Let G/(H, B) be a quotient ultragraph and let {Te, Q[A]} be
a representation of G/(H, B). Then any nonzero word in Te, Q[A], and T ∗
f
may be written as a finite linear combination of the forms TαQ[A]T ∗
β for
α, β ∈ (G/(H, B))∗ and [A] ∈ Φ(G0) with [A] ∩ r(α) ∩ r(β) 6= [∅].
Theorem 3.11. Let G/(H, B) be a quotient ultragraph. Then there ex-
ists a (unique up to isomorphism) C ∗-algebra C ∗(G/(H, B)) generated by a
universal representation {te, q[A] : [A] ∈ Φ(G0), e ∈ Φ(G1)} for G/(H, B).
Furthermore, all the te's and q[A]'s are nonzero for [∅] 6= [A] ∈ Φ(G0) and
e ∈ Φ(G1).
Proof. By a standard argument similar to the proof of [17, Theorem 2.11],
we may construct such universal C ∗-algebra C ∗(G/(H, B)). Note that the
universality implies that C ∗(G/(H, B)) is unique up to isomorphism. To
show the last statement, we generate a representation for G/(H, B) as fol-
lows. Suppose C ∗(G) = C ∗(se, pA) and consider I(H,B) as an ideal of C ∗(G)
by the isomorphism of Proposition 3.3. If we define
(cid:26) Q[A] := pA + I(H,B)
Te := se + I(H,B)
for [A] ∈ Φ(G0)
for e ∈ Φ(G1),
then the family {Te, Q[A] : [A] ∈ Φ(G0), e ∈ Φ(G1)} is a representation for
G/(H, B) in the quotient C ∗-algebra C ∗(G)/I(H,B). Note that the definition
of Q[A]'s is well-defined. Indeed, if A1 ∪ V = A2 ∪ V for some V ∈ H, then
pA1 + pV \A1 = pA2 + pV \A2 and hence pA1 + I(H,B) = pA2 + I(H,B) by the
facts V \ A1, V \ A2 ∈ H.
Moreover, all elements Q[A] and Te are nonzero for [∅] 6= [A] ∈ Φ(G0),
e ∈ Φ(G1). In fact, if Q[A] = 0, then pA ∈ I(H,B) and we get A ∈ H by [13,
Theorem 6.12]. Also, since T ∗
e Te = Qr(e) 6= 0, all partial isometries Te are
nonzero.
Now suppose that C ∗(G/(H, B)) is generated by the family {te, q[A]
:
[A] ∈ Φ(G0), e ∈ Φ(G1)}. By the universality of C ∗(G/(H, B)), there is a
∗-homomorphism φ : C ∗(G/(H, B)) → C ∗(G)/I(H,B) such that φ(te) = Te
and φ(q[A]) = Q[A], and thus, all elements {te, q[A] : [∅] 6= [A] ∈ Φ(G0), e ∈
Φ(G1)} are nonzero.
(cid:3)
Using Lemma 3.10, one may easily show that
C ∗(G/(H, B)) = span(cid:8)tαq[A]t∗
β : α, β ∈ (G/(H, B))∗, r(α)∩[A]∩r(β) 6= [∅](cid:9).
4. Uniqueness Theorems
After defining the C ∗-algebras of quotient ultragraphs, in this section,
we prove the gauge invariant and the Cuntz-Krieger uniqueness theorems
for them. To do this, we approach to a quotient ultragraph C ∗-algebra by
graph C ∗-algebras and then apply the corresponding uniqueness theorems of
graph C ∗-algebras. This approach is a developed version of the dual graph
method of [15, Section 2] and [17, Section 5] with more complications.
10
HOSSEIN LARKI
We fix again an ultragraph G, an admissible pair (H, B) in G, and the
quotient ultragraph G/(H, B) = (Φ(G0), Φ(G0), Φ(G1), r, s).
Definition 4.1. We say that a vertex [v] ∈ Φ(G0) is a sink if s−1([v]) = ∅. If
[v] emits finitely many edges of Φ(G1), [v] is called a regular vertex. We say
[v] is a singular vertex if it is not regular. We denote the set of all singular
vertices of Φ(G0) by
Φsg(G0) :=(cid:8)[v] ∈ Φ(G0) : s−1([v]) = 0 or ∞(cid:9).
Let F be a finite subset of Φsg(G0)∪Φ(G1), and denote F 0 := F ∩Φsg(G0)
and F 1 := F ∩ Φ(G1) = {e1, . . . , en}. We want to construct a special graph
GF such that C ∗(GF ) is isomorphic to C ∗(te, q[v] : [v] ∈ F 0, e ∈ F 1). For
each ω = (ω1, . . . , ωn) ∈ {0, 1}n \ {0n}, we write
r(ei) \ [ωj=0
r(ω) := \ωi=1
Note that r(ω) ∩ r(ν) = [∅] for distinct ω, ν ∈ {0, 1} \ {0n}. If
r(ej) and R(ω) := r(ω) \[ F 0.
Γ0 :=(cid:8)ω ∈ {0, 1}n \ {0n} : ∃[v1], . . . , [vm] ∈ Φ(G0) such that
R(ω) =
m[i=1
[vi] and ∅ 6= s−1([vi]) ⊆ F 1 for 1 ≤ i ≤ m(cid:9),
we consider the finite set
Γ := {ω ∈ {0, 1}n \ {0n} : R(ω) 6= [∅] and ω /∈ Γ0} .
Now we define the finite graph GF = (G0
F , G1
F , rF , sF ) containing the
vertices G0
F := F 0 ∪ F 1 ∪ Γ and the edges
G1
F :=(cid:8)(e, f ) ∈ F 1 × F 1 : s(f ) ⊆ r(e)(cid:9)
∪(cid:8)(e, [v]) ∈ F 1 × F 0 : [v] ⊆ r(e)(cid:9)
∪(cid:8)(e, ω) ∈ F 1 × Γ : ωi = 1 when e = ei(cid:9)
with the source map sF (e, f ) = sF (e, [v]) = sF (e, ω) = e, and the range map
rF (e, f ) = f , rF (e, [v]) = [v], rF (e, ω) = ω.
Proposition 4.2. Let G/(H, B) be a quotient ultragraph and let F be a
finite subset of Φsg(G0) ∪ Φ(G1). If C ∗(G/(H, B)) = C ∗(te, q[A]), then the
elements
Qe := tet∗
e,
Q[v] := q[v](1 −Pe∈F 1 tet∗
T(e,[v]) := teQ[v],
e), Qω := qR(ω)(1 −Pe∈F 1 tet∗
T(e,ω) := teQω
T(e,f ) := teQf ,
form a Cuntz-Krieger GF -family generating the C ∗-subalgebra C ∗(te, q[v] :
[v] ∈ F 0, e ∈ F 1) of C ∗(G/(H, B)). Moreover, all projections Q∗ are
nonzero.
e)
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
11
Proof. We first note that all the projections Qe, Q[v], and Qω are nonzero.
Indeed, each [v] ∈ F 0 is a singular vertex in G/(H, B), so Q[v] is nonzero.
Also, by definition, for every ω ∈ Γ we have ω /∈ Γ0 and R(ω) 6= [∅]. Hence,
for any ω ∈ Γ, if there is an edge f ∈ Φ(G1) \ F 1 with s(f ) ⊆ R(ω), then
0 6= tf t∗
f ≤ Qω. If there is a sink [w] such that [w] ⊆ R(ω) = r(ω) \S F 0,
e) = Qω. Thus Qω is nonzero in either
case. In addition, the projections Qe, Q[v], and Qω are mutually orthogonal
then 0 6= q[w] ≤ qR(ω)(1 −Pe∈F 1 tet∗
because of the factor 1 −Pe∈F 1 tet∗
Now we show the collection {Tx, Qa : a ∈ G0
F , x ∈ G1
F } is a Cuntz-Krieger
e and the definition of R(ω).
GF -family by checking the relations (1)-(3) in Remark 2.4.
(1): Since Q[v], Qω ≤ qr(e) for (e, [v]), (e, ω) ∈ G1
F , we have
T ∗
(e,f )T(e,f ) = Qf t∗
eteQf = tf t∗
f qr(e)tf t∗
f = tf qr(f )t∗
f = Qf ,
and
T ∗
(e,[v])T(e,[v]) = Q[v]t∗
eteQ[v] = Q[v]qr(e)Q[v] = Q[v],
T ∗
(e,ω)T(e,ω) = Qωt∗
eteQω = Qωqr(e)Qω = Qω.
(2): This relation may be checked similarly.
(3): Note that any element of F 0 ∪ Γ is a sink in GF . So, fix some ei ∈ F 1
as a vertex of G0
F . Write qF 0 :=P[v]∈F 0 q[v]. We compute
f = qr(ei) Xf ∈F 1
Qf = qr(ei) Xf ∈F 1
tf t∗
qr(ei) Xf ∈F 1
s(f )⊆r(ei)
s(f )⊆r(ei)
tf t∗
f ;
(i)
(ii)
(iii)
qr(ei) X[v]∈F 0,
[v]⊆r(ei)
Q[v] = qr(ei) X[v]∈F 0
q[v](1 − Xe∈F 1
tet∗
e)
tet∗
e);
= qr(ei)qF 0(1 − Xe∈F 1
e) = Xωi=1
tet∗
qR(ω)(1 − Xe∈F 1
tet∗
e),
We can use these relations to get
Qω = Xω∈Γ,ωi=1
Xω∈Γ,ωi=1
qR(ω)(1 − Xe∈F 1
becausePωi=1 qR(ω) = qr(ei)(1 − qF 0).
T(ei,f ) + X[v]∈F 0, [v]⊆r(ei)
= teiqr(ei) Xe∈F 1
T(ei,[v]) + Xω∈Γ, ωi=1
e + qr(ei)qF 0(Xe∈F 1
tet∗
tet∗
Xs(f )⊆r(ei)
T(ei,ω)
e) + qr(ei)(1 − qF 0)(Xe∈F 1
tet∗
e)
12
HOSSEIN LARKI
= teiqr(ei)Xe∈F 1
= tei.
tet∗
e + (qF 0 + 1 − qF 0)(1 − Xe∈F 1
tet∗
e)
(4.1)
Now if ei is not a sink as a vertex in GF (i.e. {x ∈ G1
we conclude that
F : sF (x) = ei} > 0),
T(ei,ω)T ∗
(ei,ω)
Xf ∈F 1, s(f )⊆r(ei)
T(ei,f )T ∗
T(ei,[v])T ∗
(ei,f ) + X[v]∈F 0, [v]⊆r(ei)
ei +X teiQ[v]t∗
=X teiQf t∗
ei +X teiQωt∗
= teiqr(ei)(X Qf +X Q[v] +X Qω)t∗
= teit∗
ei = Qei.
ei
ei
(ei,[v]) + Xω∈Γ, ωi=1
which establishes the relation (3).
Furthermore, equation (4.1) says that tei ∈ C ∗(T∗, Q∗) for every ei ∈ F 1.
Also, for each [v] ∈ F 0, we have
Q[v] + Xe∈F 1,s(e)=[v]
tet∗
Qe = t[v](1 − Xe∈F 1
= t[v] − t[v] Xe∈F 1
e) + Xe∈F 1,s(e)=[v]
e + t[v] Xe∈F 1
tet∗
e
tet∗
tet∗
e
= t[v].
Therefore, the family {Tx, Qa : a ∈ G0
C ∗({te, q[v]
plete.
F } generates the C ∗-subalgebra
: e ∈ F 1, [v] ∈ F 0}) of C ∗(G/(H, B)) and the proof is com-
(cid:3)
F , x ∈ G1
Corollary 4.3. If F is a finite subset of Φsg(G0) ∪ Φ(G1), then C ∗(GF ) is
isometrically isomorphic to the C ∗-subalgebra of C ∗(G/(H, B)) generated by
{te, q[v] : [v] ∈ F 0, e ∈ F 1}.
Proof. Suppose that X is the C ∗-subalgebra generated by {te, q[v] : [v] ∈
F 0, e ∈ F 1} and let {Tx, Qa : a ∈ G0
F } be the Cuntz-Krieger
GF -family in Proposition 4.2. If C ∗(GF ) = C ∗(sx, pa), then there exists a
∗-homomorphism φ : C ∗(GF ) → X with φ(pa) = Qa and φ(sx) = Tx for
every a ∈ G0
F . Since each Qa is nonzero by Proposition 4.2, the
gauge invariant uniqueness theorem implies that φ is injective. Moreover,
the family {Tx, Qa} generates X, so φ is an isomorphism.
(cid:3)
F , x ∈ G1
F , x ∈ G1
Note that if F1 ⊆ F2 are two finite subsets of Φsg(G0) ∪ Φ(G1) and X1, X2
are the C ∗-subalgebras of C ∗(G/(H, B)) corresponding to GF1 and GF2,
respectively, we then have X1 ⊆ X2 by Proposition 4.2.
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
13
Remark 4.4. Using the relations in Definition 3.8, each q[A] for [A] ∈ Φ(G0),
can be produced by the elements of
{q[v] : [v] ∈ Φsg(G0)} ∪ {te : e ∈ Φ(G1)}
with finitely many operations. So, the ∗-subalgebra of C ∗(G/(H, B)) gener-
ated by
{q[v] : [v] ∈ Φsg(G0)} ∪ {te : e ∈ Φ(G1)}
is dense in C ∗(G/(H, B)).
As for graph C ∗-algebras, we can apply the universal property to have a
strongly continuous gauge action γ : T → Aut(C ∗(G/(H, B))) such that
γz(te) = zte and γz(q[A]) = q[A]
for every [A] ∈ Φ(G0), e ∈ Φ(G1), and z ∈ T. Now we are ready to prove
the uniqueness theorems.
Theorem 4.5 (The Gauge Invariant Uniqueness Theorem). Let G/(H, B)
be a quotient ultragraph and let {Te, Q[A]} be a representation for G/(H, B)
such that Q[A] 6= 0 for [A] 6= [∅]. If πT,Q : C ∗(G/(H, B)) → C ∗(Te, Q[A]) is
the ∗-homomorphism satisfying πT,Q(te) = Te, πT,Q(q[A]) = Q[A], and there
is a strongly continuous action β of T on C ∗(Te, Q[A]) such that βz ◦ πT,Q =
πT,Q ◦ γz for every z ∈ T, then πT,Q is faithful.
Proof. Select an increasing sequence {Fn} of finite subsets of Φsg(G0)∪Φ(G1)
such that ∪∞
n=1Fn = Φsg(G0) ∪ Φ(G1). For each n, Corollary 4.3 gives an
isomorphism
πn : C ∗(GFn) → C ∗({te, q[v] : [v] ∈ F 0, e ∈ F 1})
that respects the generators. We can apply the gauge invariant uniqueness
theorem for graph C ∗-algebras to see that the homomorphism
πT,Q ◦ πn : C ∗(GFn) → C ∗(Te, Q[A])
is faithful. Hence, for every Fn, the restriction of πT,Q on the ∗-subalgebra of
C ∗(G/(H, B)) generated by {te, q[v] : [v] ∈ F 0
n} is faithful. This turns
out that πT,Q is injective on the ∗-subalgebra C ∗(te, q[v] : [v] ∈ Φsg(G0), e ∈
Φ(G1)). Since, this subalgebra is dense in C ∗(G/(H, B)), we conclude that
πT,Q is faithful.
(cid:3)
n , e ∈ F 1
To prove a verson of Cuntz-Krieger uniqueness theorem, we extend Con-
dition (L) for quotient ultragraphs.
Definition 4.6. We say that G/(H, B) satisfies Condition (L) if for every loop
α = e1 . . . en in G/(H, B), at least one of the following conditions holds:
(i) r(ei) 6= s(ei+1) for some 1 ≤ i ≤ n, where ei+1 := e1 (or equivalently,
r(ei) \ s(ei+1) 6= [∅]).
(ii) α has an exit; that means, there exists f ∈ Φ(G1) such that s(f ) ⊆ r(ei)
and f 6= ei+1 for some 1 ≤ i ≤ n.
14
HOSSEIN LARKI
Lemma 4.7. Let F be a finite subset of Φsg(G0) ∪ Φ(G1).
satisfies Condition (L), so does the graph GF .
If G/(H, B)
Proof. Suppose that G/(H, B) satisfies Condition (L). As the elements of
F 0 ∪Γ are sinks in GF , every loop in GF is of the formeα = (e1, e2) . . . (en, e1)
corresponding with a loop α = e1 . . . en in G/(H, B). So, fix a loop eα =
(e1, e2) . . . (en, e1) in GF . Then α = e1 . . . en is a loop in G/(H, B) and by
Condition (L), one of the following holds:
(i) r(ei) 6= s(ei+1) for some 1 ≤ i ≤ n, where ei+1 := e1, or
(ii) there exists f ∈ Φ(G1) such that s(f ) ⊆ r(ei) and f 6= ei+1 for some
1 ≤ i ≤ n.
We can suppose in the case (i) that s(ei+1) ( r(ei) and r(ei) emits only
the edge ei+1 in G/(H, B). Then, by the definition of Γ, there exists either
[v] ∈ F 0 with [v] ⊆ r(ei)\s(ei+1), or ω ∈ Γ with ωi = 1. Thus, either (ei, [v])
∃ω ∈ Γ with ωi = 1 such that [v] ⊆ R(ω).
f /∈ F 1, for [v] := s(f ) we have either [v] /∈ F 0 or
or (ei, ω) is an exit for the loop eα in GF , respectively.
Now assume case (ii) holds. If f ∈ F 1, then (ei, f ) is an exit for eα. If
Hence, (ei, [v]) or (ei, ω) is an exit for eα, respectively. Consequently, in any
case, eα has an exit.
Theorem 4.8 (The Cuntz-Krieger Uniqueness Theorem). Suppose that
G/(H, B) is a quotient ultragraph satisfying Condition (L). If {Te, QA} is a
Cuntz-Krieger representation for G/(H, B) in which all the projection Q[A]
are nonzero for [A] 6= [∅], then the ∗-homomorphism πT,Q : C ∗(G/(H, B)) →
C ∗(Te, Q[A]) with πT,Q(te) = Te and πT,Q(q[A]) = Q[A] is an isometrically
isomorphism.
(cid:3)
Proof. It suffices to show that πT,Q is faithful. Similar to the proof of
4.5, choose an increasing sequence {Fn} of finite sets such that ∪∞
n=1Fn =
Φsg(G0) ∪ Φ(G1). By Corollary 4.3, there are isomorphisms πn : C ∗(GFn) →
C ∗(cid:0){te, q[v] : [v] ∈ F 0
graphs GFn satisfy Condition (L) by Lemma 4.7, the Cuntz-Krieger unique-
ness theorem for graph C ∗-algebras implies that the ∗-homomorphisms
n}(cid:1) that respect the generators. Since all the
n , e ∈ F 1
πT,Q ◦ πn : C ∗(GFn) → C ∗(Te, Q[A])
are faithful. Therefore, πT,Q is faithful on the subalgebra C ∗(te, q[v] : [v] ∈
Φsg(G0), e ∈ Φ(G1)) of C ∗(G/(H, B)). Since this subalgebra is dense in
C ∗(G/(H, B)), we conclude that πT,Q is a faithful homomorphism.
(cid:3)
5. Gauge Invariant Ideals
Here we want to describe the gauge invariant ideals of a quotient ultra-
graph C ∗-algebra C ∗(G/(H, B)) to apply in Sections 6 and 7. For this,
we first prove that C ∗(G/(H, B)) is isomorphic to the quotient C ∗-algebra
C ∗(G)/I(H,B), and then use the ideal structure of C ∗(G) [13, Section 6].
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
15
Proposition 5.1. Let G be an ultragraph. If (H, B) is an admissible pair
in G, then C ∗(G/(H, B)) ∼= C ∗(G)/I(H,B).
Proof. Using Proposition 3.3, we can consider I(H,B) as an ideal of C ∗(G).
Suppose that C ∗(G) = C ∗(se, pA) and C ∗(G/(H, B)) = C ∗(te, q[A]). If we
define
Te := se + I(H,B) and Q[A] := pA + I(H,B)
for every [A] ∈ Φ(G0) and e ∈ Φ(G1), then the family {Te, Q[A]} is a rep-
resentation for G/(H, B) in C ∗(G)/I(H,B). So, there is a ∗-homomorphism
φ : C ∗(G/(H, B)) → C ∗(G)/I(H,B) such that φ(te) = Te and φ(q[A]) = Q[A].
Moreover, all Q[A] with [A] 6= [∅] are nonzero because pA + I(H,B) = I(H,B)
implies A ∈ H. Then, an application of Theorem 4.5 yields that φ is faithful.
On the other hand, the family {Te, Q[A] : [A] ∈ Φ(G0), e ∈ Φ(G1)} generates
the quotient C ∗(G)/I(H,B), and hence, φ is surjective as well. Therefore, φ
is an isomorphism and the result follows.
(cid:3)
In the rest of section, we fix a quotient ultragraph G/(H, B) and assume
that C ∗(G) = C ∗(se, pA) and C ∗(G/(H, B)) = C ∗(te, q[A]). If we order the
admissible pairs in G by
(K1, S1) (cid:22) (K2, S2) ⇐⇒ K1 ⊆ K2 and S1 ⊆ K2 ∪ S2,
then [13, Theorem 6.12] shows that the map (K, S) 7→ I(K,S) is a one-to-one
order preserving correspondence between admissible pairs in G and gauge
invariant ideals of C ∗(G). Moreover, for every gauge invariant ideal I(K,S) in
C ∗(G), we have {A ∈ G0 : pA ∈ I(K,S)} = K and {w ∈ BK : pK
w ∈ I(K,S)} =
S.
Lemma 5.2. Let φ : C ∗(G) → C ∗(G/(H, B)) be the canonical surjection
described in Proposition 5.1, and let (K, S) be an admissible pair in G. Then
the image of ideal I(K,S) of C ∗(G) under φ is the ideal J[K,S] in C ∗(G/(H, B))
generated by
nq[A], q[w′] : A ∈ K, w ∈ S ∩ (BH \ B)o
q[w] − XsG (e)=w,rG (e) /∈K
tet∗
e : w ∈ S \ (BH \ B)
.
Proof. Using the isomorphisms in Propositions 3.3 and 5.1, we have φ(pA) =
q[A] and φ(pH
w ) = q[w′] for every A ∈ K and w ∈ S ∩ (BH \ B). Also, if
w ∈ S \ (BH \ B), then
∪
φ(pw − XsG (e)=w,rG(e) /∈K
ses∗
e) = q[w] − XsG(e)=w,rG (e) /∈K
tet∗
e.
(cid:3)
These follow the result.
Theorem 5.3. Let G/(H, B) be a quotient ultragraph of G. Then
16
HOSSEIN LARKI
(1) the map (K, S) 7→ J[K,S] is a one-to-one order preserving corre-
spondence between the admissible pairs (K, S) in G with H ⊆ K,
B ⊆ K ∪ S and the gauge invariant ideals of C ∗(G/(H, B));
(2) if J[K,S] is a gauge invariant ideal in C ∗(G/(H, B)), then
C ∗(G/(H, B))
J[K,S]
∼= C ∗(G/(K, S)).
Proof. (1): Suppose that J is a gauge invariant ideal in C ∗(G/(H, B)). If
φ : C ∗(G) → C ∗(G/(H, B)) is the canonical map of Lemma 5.2 and we define
I := φ−1(J), then I is a gauge invariant ideal of C ∗(G) with I(H,B) ⊆ I. By
[13, Theorem 6.12], there exists an admissible pair (K, S) in G such that
I = I(K,S). In particular, we have H ⊆ K and B ⊆ K ∪ S. Now Lemma 5.2
implies that J = φ(I(K,S)) = J[K,S], as demanded.
(2): Let J[K,S] be a gauge invariant ideal in C ∗(G/(H, B)). Since φ(I(K,S)) =
J[K,S] by Lemma 5.2, Proposition 5.1 yields that
C ∗(G)
I(K,S)
C ∗(G)/I(H,B)
I(K,S)/I(H,B)
J[K,S]
C ∗(G/(H, B))
∼=
∼=
∼= C ∗(G/(K, S)).
(cid:3)
6. Condition (K)
It is known that a graph E satisfies Condition (K) if and only if every
ideal in C ∗(E) is gauge invariant (see [1, 5] among others).
In [13, Sec-
tion 7], the authors proved a similar result for ultragraph C ∗-algebras by
applying the C ∗-algebras of topological graphs. In this section, we extend
Condition (K) for a quotient ultragraph G/(H, B) under which all ideals of
C ∗(G/(H, B)) are gauge invariant. In particular, we can alternatively obtain
[13, Proposition 7.3] by applying the quotient ultragraphs. To do this, we
first see that a quotient ultragraph G/(H, B) satisfies Condition (K) if and
only if every quotient ultragraph G/(K, S) with H ⊆ K satisfies Condition
(L). Then the main results will be shown.
Let α = e1 . . . en be a path in an ultragraph G. For 1 ≤ k ≤ l ≤ n, the
path β = ekek+1 . . . el is called a subpath of α. We simply write β ⊆ α when
β is a subpath of α; otherwise, we write β * α.
Definition 6.1. We say that a quotient ultragraph G/(H, B) of G satisfies
Condition (K) if every vertex v ∈ G0 \ H either is the base of no loops, or
there are at least two loops α, β in G based at v such that neither α nor β
is a subpath of the other.
Remark 6.2. If G/(∅, ∅) is an ultragraph, the above definition for Condition
(K) coincide with that of [13, Definition 7.1].
In the following, we show in the absence of Condition (K) for G/(H, B)
that there is a quotient ultragraph G/(K, S) with K ⊇ H such that it does
not satisfy Condition (L). To d this, let G/(H, B) be a quotient ultragraph
and set
X :=(cid:8)rG(α) \ γ0 : α ∈ G∗, α ≥ 1, sG (α) ∈ γ0(cid:9) ,
Ai : A1, . . . , An ∈ X ∪ H, n ∈ N) ,
Y :=( n[i=1
K0 :=(cid:8)B ∈ G0 : B ⊆ A for some A ∈ Y(cid:9) .
G (w)) ⊆ Kn−1(cid:9)
∪(cid:8)w ∈ B : w /∈ Kn−1 and rG(s−1
G (w) < ∞ and rG(s−1
Sn :=(cid:8)w ∈ G0 : 0 < s−1
and
We construct a specific saturated hereditary subset of G0 containing H as
follows: for any n ∈ N inductively define
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
17
not satisfying Condition (K). Then G/(H, B) contains a loop γ = e1 . . . en
such that there are no loops α with s(α) = s(γ), α * γ, and γ * α.
If
γ0 := {sG(e1), . . . , sG(en)}, define
G (w)) ⊆ Kn−1(cid:9)
Kn := {A ∪ F : A ∈ Kn−1 and F ⊆ Sn is a finite subset} .
Then we can easily show that the set
K =
∞[n=0
Kn =(A ∪ F : A ∈ K0 and F ⊆
Sn is a finite subset)
∞[n=1
is hereditary and saturated in G.
Lemma 6.3. Suppose that G/(H, B) is a quotient ultragraph of G, and
γ = e1 . . . en is a loop in G/(H, B) such that there are no loops α with
s(α) = s(γ) and α * γ, γ * α. If we construct the set K as above, then K
is a saturated hereditary subset of G0 such that H ⊆ K and B ⊆ K ∪ BK .
Moreover, we have A ∩ γ0 = ∅ for every A ∈ K.
Proof. First, we show inductively that every Kn is a hereditary subset of G0
by checking the conditions of Definition 2.5. To verify condition (1) for K0,
let us take e ∈ G1 with sG(e) ∈ K0. Then sG(e) ∈ H ∪ X. If sG(e) ∈ H,
we have rG(e) ∈ H ⊆ K0 by the hereditary property of H. If sG(e) ∈ X,
there is α ∈ G∗ such that sG(α) ∈ γ0 and sG(e) ∈ rG(α) \ γ0. Hence,
sG(αe) = sG(α) ∈ γ0. Also, rG(αe) ∩ γ0 = ∅ because the otherwise implies
the existence of a path β ∈ G∗ with sG(β) = sG(γ) and β * γ, γ * β that
contradicts the hypothesis. It turns out
rG(e) = rG(αe) = rG(αe) \ γ0 ∈ X ⊆ K0.
Hence, K0 satisfies condition (1) in Definition 2.5. We may easily verify
conditions (2) and (3) for K0, so K0 is hereditary. Moreover, for every
w ∈ Sn, the range of each edge emitted by w belongs to Kn−1 by definition.
Thus, we can inductively check that each Kn is hereditary, and so is K =
∪∞
n=1Kn. The saturation property of K is similar to the proof of [18, Lemma
3.12], and it is omitted.
18
HOSSEIN LARKI
Furthermore, it is clear that H ⊆ K by the fact H ⊆ Y . To see B ⊆
K ∪ BK, take an arbitrary vertex w ∈ B \ K. Then w is an infinite emitter
and since w /∈ ∪∞
n=1Kn = K. Thus w emits
some edges into G0 \ K which implies w ∈ BK as desired.
n=1Sn, we have rG(s−1
G (w)) * ∪∞
It remains to show A ∩ γ0 = ∅ for every A ∈ K. To do this, note that
A ∩ γ0 = ∅ for every A ∈ K0 because this property holds for all A ∈ H and
A ∈ X. We claim that (∪∞
n=1Sn) ∩ γ0 = ∅. Indeed, if v = sG(ei) ∈ γ0 for
some ei ∈ γ, then rG(ei) ∩ γ0 6= ∅ and rG(ei) /∈ K0. Hence, {rG(e) : e ∈
G1, sG(e) = v} * K0 that turns out v /∈ S1. So, we have S1 ∩ γ0 = ∅. An
inductive argument shows Sn ∩ γ0 = ∅ for n ≥ 1, and the claim holds. Now
since
K = ∪∞
n=1Kn = {A ∪ F : A ∈ K0 and F ⊆ ∪∞
n=1Sn is a finite subset} ,
we conclude that A ∩ γ0 = ∅ for all A ∈ K.
(cid:3)
Proposition 6.4. A quotient ultragraph G/(H, B) satisfies Condition (K) if
and only if for every admissible pair (K, S) in G with H ⊆ K and B ⊆ K ∪S,
the quotient ultragraph G/(K, S) satisfies Condition (L).
Proof. Suppose that G/(H, B) satisfies Condition (K) and (K, S) is an ad-
missible pair in G with H ⊆ K. Let α = e1 . . . en be a loop in G/(K, S).
Since α is also a loop in G/(H, B) and G/(H, B) satisfies Condition (K),
there is a loop β = f1 . . . fm in G with sG(α) = sG(β), and neither α ⊆ β
nor β ⊆ α. Without loos of generality, assume e1 6= f1. By the fact
sG(α) = sG(β) ∈ rG(β), we have rG(β) /∈ K, and so rG(f1) /∈ K by the
hereditary property of K. Therefore, f1 is an exit for α in G/(K, S) and we
conclude that G/(K, S) satisfies Condition (L).
For the converse, suppose on the contrary that G/(H, B) does not satisfy
Condition (K). Then there exists a loop γ = e1 . . . en in G/(H, B) such that
there are no loops α with s(α) = s(γ), α * γ, and γ * α. As Lemma 6.3,
construct a saturated hereditary subset K of G0 and consider the quotient
ultragraph G/(K, BK ) = (Ψ(G0), Ψ(G0), Ψ(G1), r′, s′). We denote by ⌊.⌋ the
equivalent classes in Ψ(G0). We show that γ as a loop in G/(K, BK ) has no
exits and r′(ei) = s′(ei+1) for 1 ≤ i ≤ n. If f is an exit for γ in G/(K, BK )
such that s′(f ) = s′(ej) and f 6= ej, then rG(f ) /∈ K and rG(f ) ∩ γ0 6= ∅ (if
rG(f ) ∩ γ0 = ∅, then rG(f ) = rG(f ) \ γ0 ∈ X ⊆ K, a contradiction). So,
there is el ∈ γ such that sG(el) ∈ rG(f ). If we set α := e1 . . . ej−1f el . . . en,
then α is a loop in G with sG(α) = sG(γ), and α * γ, γ * α, that contradicts
the hypothesis. Therefore, γ has no exits in G/(K, BK ). Moreover, we have
r′(ei) ∩ ⌊γ0⌋ = s′(ei+1) for each 1 ≤ i ≤ n, because the otherwise gives an
exit for γ in G/(K, BK ) by the construction of K. Hence,
r′(ei) \ s′(ei+1) = r′(ei) \ ⌊γ0⌋ = ⌊∅⌋
and we get r′(ei) = s′(ei+1) (note that the fact rG(ei) \ γ0 ∈ K implies
r′(ei) \ ⌊γ0⌋ = ⌊rG(ei) \ γ0⌋ = ⌊∅⌋). Therefore, the quotient ultragraph
G/(K, BK ) does not satisfy Condition (L) as desired.
(cid:3)
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
19
To prove the main result of this section, Theorem 6.6, we need also the
following lemma.
Lemma 6.5. Let G/(H, B) = (Φ(G0), Φ(G0), Φ(G1), r, s) be a quotient ultra-
graph of G. If G/(H, B) does not satisfy Condition (L), then C ∗(G/(H, B))
contains an ideal Morita-equivalent to C(T). In particular, C ∗(G/(H, B))
contains non-gauge invariant ideals.
Proof. Suppose that γ = e1 . . . en is a loop in G/(H, B) without exits and
r(ei) = s(ei+1) for 1 ≤ i ≤ n. If C ∗(G/(H, B)) = C ∗(te, q[A]), for each i we
have
t∗
eitei = qr(ei) = qs(ei+1) = tei+1t∗
ei+1.
Write [v] := s(γ) and let Iγ be the ideal of C ∗(G/(H, B)) generated by q[v].
Since γ has no exits in G/(H, B) and we have
qs(ei) = (tei . . . ten)q[v](t∗
en . . . t∗
ei)
for every 1 ≤ i ≤ n, an easy argument shows that
Iγ = span(cid:8)tαq[v]t∗
β : α, β ∈ (G/(H, B))∗, [v] ⊆ r(α) ∩ r(β)(cid:9) .
So, we have
q[v]Iγq[v] = span(cid:8)(tγ)nq[v](t∗
γ)m : m, n ≥ 0(cid:9) ,
where (tγ)0 = (t∗
γ)0 := q[v]. We show that q[v]Iγq[v] is a full corner in Iγ
which is isometrically isomorphic to C(T). For this, associated to the graph
E
f
w
define Qw := q[v] and Tf := tγ (= tγq[v]). Then {Tf , Qw} is a Cuntz-Krieger
E-family in q[v]Iγq[v]. Assume C ∗(E) = C ∗(sf , pw). Since Qw 6= 0, the
gauge-invariant uniqueness theorem for graph C ∗-algebras implies that the
∗-homomorphism φ : C ∗(E) → q[v]Iγq[v] with pw 7→ Qw and sf 7→ Tf is
faithful. Moreover, the C ∗-algebra q[v]Iγq[v] is generated by {Tf , Qw}, and
hence, φ is an isomorphism. We know that C ∗(E) ∼= C(T), so q[v]Iγq[v] is
isomorphic to C(T). The fullness of q[v]Iγq[v] in Iγ may be easily shown.
Indeed, if J is an ideal in Iγ with q[v]Iγq[v] ⊆ J, we then have q[v] ∈ J and
J = Iγ. Therefore, Iγ is Morita-equivalent to q[v]Iγq[v]
∼= C(T) as desired.
On the other hand, C(T) contains infinitely many non-gauge invariant
ideals (corresponding with closed subsets ∅ 6= U ( T). Therefore, the
corner q[v]Iγq[v] contains non-gauge invariant ideals, and so does Iγ by the
Morita-equivalence. This follows the second statement of lemma because
each ideal of Iγ is an ideal of C ∗(G/(H, B)).
(cid:3)
20
HOSSEIN LARKI
We now ready to prove the main result of this section as a generalization
of [1, Corollary 3.8] and [13, Proposition 7.3].
Theorem 6.6. A quotient ultragraph G/(H, B) satisfies Condition (K) if
and only if all ideals of C ∗(G/(H, B)) are gauge invariant.
Proof. Suppose that G/(H, B) satisfies Condition (K). Take an arbitrary
ideal J in C ∗(G/(H, B)) and let I be its corresponding ideal in C ∗(G) with
I(H,B) ⊆ I (by Proposition 5.1). If C ∗(G) = C ∗(se, pA) and set
K :=(cid:8)A ∈ G0 : pA ∈ I(cid:9) , S :=
w ∈ BK : pw − Xs(e)=w,r(e) /∈K
ses∗
,
e ∈ I
then [18, Lemma 3.4] implies that (K, S) is an admissible pair in G. Since
I(K,S) ⊆ I, the map
(cid:26) φ : C ∗(G)/I(K,S) −→ C ∗(G)/I
a + I(K,S) 7−→ a + I
(for a ∈ C ∗(G))
is a well-defined ∗-homomorphism. Let us denote π : C ∗(G/(K, S)) →
C ∗(G)/I(K,S) the isomorphism of Proposition 5.1. Since the quotient ul-
tragraph G/(K, S) satisfies Condition (L) by Proposition 6.4, the Cuntz-
Krieger uniqueness theorem, Theorem 4.8, implies that φ ◦ π is injective.
So, φ is injective that follows I = I(K,S) and I is gauge invariant. Now as
∼= C ∗(G/(H, B)), we
J is the image of I(K,S) into the quotient C ∗(G)/I(H,B)
conclude that J = J[K,S] is a gauge invariant ideal in C ∗(G/(H, B)).
Conversely, assume G/(H, B) does not satisfy Condition (K). By Propo-
sition 6.4, there exists an admissible pair (K, S) in G with H ⊆ K and
B ⊆ K ∪ S such that the quotient ultragraph G/(K, S) does not satisfy Con-
dition (L). Note that in this case, I(H,B) ⊆ I(K,S) and if J[K,S] is the image
∼=
of I(K,S) into C ∗(G)/I(H,B)
C ∗(G/(K, S)). But Lemma 6.5 says that C ∗(G/(K, S)) contains infinitely
many non-gauge invariant ideals. On the other hand, for every gauge in-
variant ideal J of C ∗(G/(H, B)), the ideal J + J[K,S] is also gauge invari-
∼= C ∗(G/(K, S)). This concludes that the C ∗-
ant in C ∗(G/(H, B))/J[K,S]
algebra C ∗(G/(H, B)) contains non-gauge invariant ideals and completes the
proof.
(cid:3)
∼= C ∗(G/(H, B)), then C ∗(G/(H, B))/J[K,S]
We gather the results of this section in the following corollary. Recall
from [3] that the real rank of a unital C ∗-algebra A is the smallest integer
RR(A) such that for every ε > 0, positive integer n ≤ RR(A)+1 and n-tuple
(x1, . . . , xn) of self-adjoint elements in A, there is an n-tuple (y1, . . . , yn) of
self-adjoint elements of A such thatPn
(xi − yi)2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
nXi=1
(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
< ε.
If A is non-unital, RR(A) is the real rank of its unitization.
i=1 y2
i is invertible and
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
21
Corollary 6.7. Let H be a saturated hereditary subset of G0. Then the
following conditions are equivalent:
(1) The quotient ultragraph G/(H, BH ) satisfies Condition (K).
(2) The quotient ultragraph G/(H, B) satisfies Condition (K) for some
B ⊆ BH.
(3) For every admissible pair (K, S) with H ⊆ K, G/(K, S) satisfies
Condition (L).
(4) If B ⊆ BH , all ideals of C ∗(G/(H, B)) ∼= C ∗(G)/I(H,B) are gauge
invariant.
(5) If B ⊆ BH, the real rank of C ∗(G/(H, B)) ∼= C ∗(G)/I(H,B) is zero.
Proof. We have shown the equivalence of conditions (1)-(4). So, it suffices
to show the implications (2) ⇒ (5) and (5) ⇒ (3). For (2) ⇒ (5), suppose
B ⊆ BH and the quotient ultragraph G/(H, B) satisfies Condition (K).
Select an increasing sequence {Fn}∞
n=1 of finite subsets of Φsg(G0) ∪ Φ(G1)
such that ∪∞
n=1Fn = Φsg(G0)∪Φ(G1). Then, similar to the proof of Theorem
4.5, C ∗(G/(H, B)) is isomorphic to the inductive limit lim−→C ∗(GFn). Since
loops in each GFn come from those of G/(H, B), we may select Fn's such
that each finite graph GFn satisfies Condition (K), and so, the real rank of
C ∗(GFn) is zero [10, Theorem 4.1]. Thus the real rank of C ∗(G/(H, B)) is
zero by [3, Proposition 3.1].
For (5) ⇒ (3), suppose that there is an admissible pair (K, S) with
H ⊆ K such that the quotient ultragraph G/(K, S) does not satisfy Con-
dition (L). We can also assume that B ⊆ K ∪ S by Lemma 6.3. Then, by
Lemma 6.5, C ∗(G/(K, S)) contains an ideal Morita-equivalent to C(T), and
so RR(C ∗(G/(K, S))) 6= 0 by [3, Corolary 2.8]. As C ∗(G/(K, S)) is a quo-
tient of C ∗(G/(H, B)), Theorem 3.14 of [3] implies that RR(C ∗(G/(H, B))) 6=
0.
(cid:3)
For ultragraph C ∗-algebras, Corollary 6.7 shows both [13, Proposition 7.3]
and [12, Proposition 5.26] because every ultragraph can be considered as a
quotient ultragraph with the trivial admissible pair ({∅}, {∅}). However, our
proof is quite different from those of [13] and [12].
Corollary 6.8. An ultragraph G satisfies Condition (K) if and only if all
ideals of C ∗(G) are gauge invariant if and only if the real rank of C ∗(G) is
zero.
7. Primitive ideals in C ∗(G)
In this section, we apply quotient ultragraphs to describe primitive gauge
invariant ideals of an ultragraph C ∗-algebra. Recall that since every ultra-
graph C ∗-algebra C ∗(G) is separable (as assumed G0 to be countable), a
prime ideal of C ∗(G) is primitive and vice versa [4, Corollaire 1].
Definition 7.1. Let G be an ultragraph. For two sets A, B ∈ G0, we write
A ≥ B if either B ⊆ A, or there exists α ∈ G∗ with α ≥ 1 such that
22
HOSSEIN LARKI
s(α) ∈ A and B ⊆ r(α). We simply write A ≥ v, v ≥ B, and v ≥ w if
A ≥ {v}, {v} ≥ B, and {v} ≥ {w}, respectively. A subset M ⊆ G0 is said to
be downward directed whenever for every A, B ∈ M , there exists ∅ 6= C ∈ M
such that A, B ≥ C.
To prove Proposition 7.3 below, we need a simple lemma.
Lemma 7.2. If G/(H, B) satisfies Condition (L), then every nonzero ideal
of C ∗(G/(H, B)) contains some projection q[A] with [A] 6= [∅].
Proof. Take an arbitrary ideal J in C ∗(G/(H, B)). If there are no q[A] ∈ J
with [A] 6= [∅], then the Cuntz-Krieger uniqueness theorem implies that the
quotient homomorphism φ : C ∗(G/(H, B)) → C ∗(G/(H, B))/J is injective.
Hence, we have J = ker φ = (0).
(cid:3)
Proposition 7.3. Let H be a saturated hereditary subset of G0. Then the
ideal I(H,BH ) in C ∗(G) is primitive if and only if the quotient ultragraph
G/(H, BH ) satisfies Condition (L) and the collection G0 \ H is downward
directed.
∼=
Proof. Let I(H,BH ) be a primitive ideal of C ∗(G). Since C ∗(G)/I(H,BH )
C ∗(G/(H, BH )), the zero ideal in C ∗(G/(H, BH )) is primitive. If G/(H, BH )
does not satisfy Condition (L), then C ∗(G/(H, BH )) contains an ideal J
Morita-equivalent to C(T) by Lemma 6.5. Select two ideals I1, I2 in C(T)
with I1 ∩ I2 = (0), and let J1, J2 be their corresponding ideals in J. Then
J1 and J2 are two nonzero ideals of C ∗(G/(H, BH )) with J1 ∩ J2 = (0),
contradicting the primness of C ∗(G/(H, BH )). Therefore, G/(H, B) satisfies
Condition (L).
Now we show that M := G0 \ H is downward directed. For this, we take
arbitrary sets A, B ∈ M and consider the ideals
J1 := C ∗(G/(H, BH ))q[A]C ∗(G/(H, BH ))
and
J2 := C ∗(G/(H, BH ))q[B]C ∗(G/(H, BH ))
in C ∗(G/(H, BH )) generated by q[A] and q[B], respectively. Since A, B /∈ H,
the projections q[A], q[B] are nonzero by Theorem 3.11, and so are the ideals
J1, J2. The primness of C ∗(G/(H, BH )) implies that the ideal
J1J2 = C ∗ (G/(H, BH )) q[A]C ∗ (G/(H, BH )) q[B]C ∗ (G/(H, BH ))
is nonzero, and hence q[A]C ∗(G/(H, BH ))q[B] 6= {0}. As the set
span(cid:8)tαq[D]t∗
β : α, β ∈ (G/(H, B))∗, r(α) ∩ [D] ∩ r(β) 6= [∅](cid:9)
is dense in C ∗(G/(H, BH )), there exist α, β ∈ (G/(H, BH ))∗ and [D] ∈ Φ(G0)
such that q[A](tαq[D]t∗
β)q[B] 6= 0. In this case, we must have s(α) ⊆ [A] and
s(β) ⊆ [B] and thus, A, B ≥ C for C := rG(α) ∩ D ∩ rG(β).
In order to prove the converse, assume that G/(H, BH ) satisfies Condition
(L) and the collection M = G0 \ H is downward directed. Fix two nonzero
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
23
ideals J1, J2 of C ∗(G/(H, BH )). By Lemma 7.2, there are nonzero projec-
tions q[A] ∈ J1 and q[B] ∈ J2. Then A, B /∈ H and, since M is downward
directed, there exists C ∈ M such that A, B ≥ C. Hence, the ideal J1 ∩ J2
contains the nonzero projection q[C]. Since J1 and J2 were arbitrary, this
implies that the C ∗-algebra C ∗(G/(H, BH )) is primitive and consequently,
I(H,BH ) is a primitive ideal in C ∗(G) by Proposition 5.1.
(cid:3)
The next proposition describes another kind of primitive ideals in C ∗(G).
Proposition 7.4. Let (H, B) be an admissible pair in G and let B = BH \
{w}. Then the ideal I(H,B) in C ∗(G) is primitive if and only if A ≥ w for
all A ∈ G0 \ H.
Proof. Suppose that I(H,B) is a primitive ideal and take an arbitrary A ∈
G0 \ H. If A := A ∪ {v′ : v ∈ A ∩ (BH \ B)}, then q[A] and q[w′] are two
nonzero projections in C ∗(G/(H, B)). Consider two ideals
J[A] := C ∗(G/(H, B))q[A]C ∗(G/(H, B))
and
J[w′] := C ∗(G/(H, B))q[w′]C ∗(G/(H, B))
of C ∗(G/(H, B)) generated by q[A] and q[w′], respectively. The primness of
C ∗(G/(H, B)) ∼= C ∗(G)/IH,B implies that the ideal
J[A]J[w′] = C ∗(G/(H, B))q[A]C ∗(G/(H, B))q[w′]C ∗(G/(H, B))
is nonzero, and hence q[A]C ∗(G/(H, B))q[w′] 6= {0}. Then there exist α, β ∈
(G/(H, B))∗ such that q[A]tαt∗
βq[w′] 6= 0. Since [w′] is a sink in G/(H, B), we
must have q[A]tαq[w′] 6= 0. If α = 0, then [w′] ⊆ [A], w′ ∈ A and w ∈ A.
If α ≥ 1, then s(α) ⊆ [A] and [w′] ⊆ r(α), which yield sG(α) ∈ A and
w ∈ rG(α). Therefore, we obtain A ≥ w in either case.
Conversely, assume A ≥ w for every A ∈ G0 \ H. Then the collection
G0 \ H is downward directed. Moreover, this hypothesis implies that, for
every [∅] 6= [A] ∈ Φ(G0), there exists α ∈ (G/(H, B))∗ such that s(α) ⊆ [A]
and [w′] ⊆ r(α). Since [w′] is a sink in G/(H, B), we see that the quotient
ultragraph G/(H, B) satisfies Condition (L). Now similar to the last part of
the proof of Proposition 7.3, we can show that I(H,B) is a primitive ideal. (cid:3)
Recall that the loops in a quotient ultragraph G/(H, B) come from those
in the initial ultragraph G. So, to check Condition (L) for a quotient ultra-
graph G/(H, B), we can use the following definition.
Definition 7.5. Let H be a saturated hereditary subset of G0 and denote
M := G0 \ H. A loop α = e1 . . . en is said to be in G \ H if rG(α) ∈ M . In
this case, we say that α has an exit in G \ H if either rG(ei) \ sG(ei+1) ∈ M
for some i, or there is an edge f with rG(f ) ∈ M such that sG(f ) = sG(ei)
and f 6= ei, for some 1 ≤ i ≤ n.
24
HOSSEIN LARKI
It is easy to verify that a quotient ultragraph G/(H, B) satisfies Condition
(L) if and only if every loop in G\H has an exit in G\H. Now we characterize
all primitive gauge invariant ideals of an ultragraph C ∗-algebra C ∗(G) that
is a generalization of [1, Theorem 4.7] and [5, Theorem 4.5].
Theorem 7.6. Let G be an ultragraph. A gauge invariant ideal I(H,B) of
C ∗(G) is primitive if and only if one of the following holds:
(1) B = BH , G0 \ H is downward directed, and every loop in G \ H has
exits in G \ H.
(2) B = BH \ {w} for some w ∈ BH, and A ≥ w for all A ∈ G0 \ H.
Proof. Let I(H,B) be a primitive ideal in C ∗(G). Then C ∗(G/(H, B)) ∼=
C ∗(G)/I(H,B) is a primitive C ∗-algebra. We claim that BH \ B ≤ 1. In-
deed, if w1, w2 are two distinct vertices in BH \ B, similar to the proof of
Propositions 7.3 and 7.4, the primitivity of C ∗(G/(H, B)) implies that the
2] is nonzero. So, there exist α, β ∈ (G/(H, B))∗
corner q[w′
6= 0. But we must have α = β = 0 because
such that q[w′
[w′
2] 6= 0 which is impossi-
ble because q[w′
2}] = q[∅] = 0. Thus, the claim holds. Now
we may apply Propositions 7.3 and 7.4 to obtain the result.
(cid:3)
2] are two sinks in G/(H, B). Hence, q[w′
1]C ∗(G/(H, B))q[w′
1]q[w′
2] = q[{w′
1]q[w′
1]tαt∗
βq[w′
2]
1], [w′
1}∩{w′
Corollary 6.8 says that if G satisfies Condition (K), then all ideals of C ∗(G)
are of the form I(H,B). So, we have the following.
Corollary 7.7. If an ultragraph G satisfies Condition (K), then Theorem
7.6 describes all primitive ideals of C ∗(G).
In the end of paper, we establish primitive gauge invariant ideals of a quo-
tient ultragraph C ∗-algebra C ∗(G/(H, B)) by applying Theorem 7.6. Recall
from Theorem 5.3 that every gauge invariant ideal of C ∗(G/(H, B)) is of the
form J[K,S] with H ⊆ K and B ⊆ K ∪ S. Since
C ∗(G/(H, B))
J[K,S]
∼= C ∗(G/(K, S)) ∼=
C ∗(G)
I(K,S)
by Theorem 5.3, a gauge invariant ideal J[K,S] of C ∗(G/(H, B)) is primitive
if and only if I(K,S) is a primitive ideal of C ∗(G). Therefore, Theorem 7.6
conclude that:
Theorem 7.8. Let G/(H, B) be a quotient ultragraph of G. A gauge in-
variant ideal J[K,S] of C ∗(G/(H, B)) is primitive if and only if one of the
following conditions holds:
(1) S = BK, G0 \ K is downward directed, and every loop in G \ K has
an exit in G \ K.
(2) S = BK \ {w} for some w ∈ BK, and A ≥ w for all A ∈ G0 \ K.
In particular, if G/(H, B) satisfies Condition (K), these conditions charac-
terize all primitive ideals of C ∗(G/(H, B)).
QUOTIENTS OF ULTRAGRAPH C ∗-ALGEBRAS
25
References
[1] T. Bates, J.H. Hong, I. Raeburn and W. Szyma´nski, The ideal structure of the C ∗-
algebras of infinite graphs, Ilinois J. Math. 46 (2002), 1159-1176.
[2] T. Bates, D. Pask, I. Raeburn and W. Szyma´nski, The C ∗-algebras of row-finite
graphs, New York J. Math. 6 (2000), 307-324.
[3] L.G. Brown and G.K. Pedersen, C*-algebras of real rank zero, J. Funct. Anal. 99
(1991), 131-149.
[4] J. Dixmier, Sur les C ∗-algebres, Bull. Soc. Math. France 88 (1960), 95-112.
[5] D. Drinen and M. Tomforde, The C ∗-algebras of arbitrary graphs, Rocky Mt. J. Math.
35 (2005), 105-135.
[6] R. Exel and M. Laca, Cuntz-Krieger algebras for infinite matrices, J. Reine Angew.
Math. 512 (1999), 119-172.
[7] N. Fowler, M. Laca and I. Raeburn, The C ∗-algebras of infinite graphs, Proc. Amer.
Math. Soc. 8 (2000), 2319-2327.
[8] J.H. Hong and W. Szyma´nski, Purely infinite Cuntz-Krieger algebras of directed
graphs, Bull. London Math. Soc. 35 (2003), 689-696.
[9] J.A. Jeong, S.H. Him and G.H. Park, The structure of gauge-invariant ideals of la-
belled graph C ∗-algebras, J. Funct. Anal. 262 (2012), 1759-1780.
[10] J.A. Jeong, G.H. Park, Graph C ∗-algebras with real rank zero, J. Funct. Anal. 188
(2002), 216-226.
[11] T. Katsura, A class of C ∗-algebras generalizing both graph algebras and homeomor-
phism C ∗-algebras, I. Fundamental results, Trans. Amer. Math. Soc. 356 (2004),
4287-4322.
[12] T. Katsura, P.S. Muhly, A. Sims and M. Tomforde, Graph algebras, Exel-Laca alge-
bras, and ultragraph C ∗-algebras coincide up to Morita equivalence, J. Reine Angew.
Math. 640 (2010), 135-165.
[13] T. Katsura, P.S. Muhly, A. Sims and M. Tomforde, Utragraph algebras via topological
quivers, Studia Math. 187 (2008), 137-155.
[14] P.S. Muhly and M. Tomforde, Topological Quivers, Internat. J. Math. 16 (2005),
693-755.
[15] I. Raeburn and W. Szymanski, Cuntz-Krieger algebras of infinite graphs and matrices,
Trans. Amer. Math. Soc. 356 (2004), 39-59.
[16] W. Szyma´nski, Simpliciy of Cuntz-Krieger algebras of infinite matrices, Pacific J.
Math. 199 (2001), 249-256.
[17] M. Tomforde, A unified approach to Exel-Laca algebras and C ∗-algebras associated to
graphs, J. Oper. Theory 50 (2003), 345-368.
[18] M. Tomforde, Simplicity of ultragraph algebras, Indiana Univ. Math. J. 52 (2003)
901-925.
Department of Mathematics, Faculty of Mathematical Sciences and Com-
puter, Shahid Chamran University of Ahvaz, Iran
E-mail address: [email protected]
|
1108.5219 | 1 | 1108 | 2011-08-26T01:02:39 | The Correlation Numerical Range of a Matrix and Connes' Embedding Problem | [
"math.OA",
"math.FA"
] | We define a new numerical range of an n\timesn complex matrix in terms of correlation matrices and develop some of its properties. We also define a related numerical range that arises from Alain Connes' famous embedding problem. | math.OA | math |
The Correlation Numerical Range of a Matrix
and Connes' Embedding Problem∗
Don Hadwin
Department of Mathematics and Statistics
University of New Hampahire
Durham, NH 03824, USA
Deguang Han
Department of Mathematics
University of Central Florida
Orlando, FL 32816, USA
November 7, 2018
Abstract
We define a new numerical range of an n × n complex matrix in terms
of correlation matrices and develop some of its properties. We also define a
related numerical range that arises from Alain Connes' famous embedding
problem.
2010 Mathematics Subject Classification. Primary 46L10, 15A48.
Key words and phrases. Numerical Range, Correlation Numerical Range,
Connes' Embedding Problem
1 Introduction and Preliminaries
In this paper we define and study a new numerical range for n × n complex
matrices. This numerical range is loosely related to new reformulations [5], [7] of
Connes' Embedding problem (CEP) [1]. We derive some of the basic properties
of this new range, pose some questions. We answer all of the questions in the
2 × 2 case. We introduce another related numerical range with many of the
same properties that is directly related to the reformulation of CEP in [5] and
further developed in [2].
If T is an operator on a Hilbert space H, the numerical range W (T ) of T is
the set
W (T ) = {hT x, xi : x ∈ H,kxk = 1} .
∗Email: [email protected] and [email protected]. This Research is supported in part
by an NSF grant
1
Suppose A = (αij) is an n × n complex matrix. We define A = (αijIn ) acting
on H = Cn⊕ ··· ⊕ Cn (n copies), and let
1
√n
E =
e1
e2
...
en
∈ H : ke1k = ··· = kenk = 1
Wc (A) =nD Ae, eE : e ∈ Eo .
. We define the correlation numerical range, or C-numerical range, of A as
The term correlation comes from the fact that an n × n correlation matrix
[9] is a matrix B = (bij) ∈ Mn (C) such that B ≥ 0 and bii = 1 for 1 ≤
i ≤ n. Equivalently, B is an n × n correlation matrix if there are unit vectors
e1, . . . , en ∈ Cn such that
B = (hei, eji) .
Let En denote the set of n × n correlation matrices. If A ∈ Mn (C), then AT
denotes the transpose of T . We also use Dn to denote the set of all the n × n
diagonal matrices, and Dn,0 the set of trace zero n × n diagonal matrices.
2 Basic Results
We first prove some of the basic properties of Wc (T ).
Theorem 1 Suppose A ∈ Mn (C). Then
1. Wc (A) = {τn(AB} : B = {(bij) ∈ Mn (C) , B ≥ 0, bii = 1 for 1 ≤ i ≤ n}.
2. Wc (A) ⊆ W (A)
3. Wc (A) is convex
4. τn (A) ∈ Wc (A)
5. If D ∈ Dn, then
Wc (A + D) = Wc (A) + τn (D)
6. Wc (A) = {λ} if and only if A is diagonal and τn (A) = λ
7. Wc (A) ⊆ R if and only if Im A is diagonal and τn (Im A) = 0.
8. If {Ak} is a sequence in Mn (C) and kAk − Ak → 0, then
λk : λk ∈ Wc (Ak) for k ≥ 1 and lim
k→∞
λk exists(cid:27) .
k→∞
Wc (A) =(cid:26) lim
9. Wc(cid:0)AT(cid:1) = Wc (A) .
2
Proof. (1). This is a direct calculation,
(2). This follows from the definition of Wc (A) and the fact that W (A) =
W (cid:16) A(cid:17)(3) . This follows from (1).
(4). Choose a vector in E with {e1, . . . , en} an orthonormal basis for Cn.
(5) This is a direct computation.
(6). The "if" part follows from (5). For the other direction, suppose
Wc (A) = {λ}. We know from (4) that λ = τn (A). Suppose i 6= j. Choose unit
vectors {e1, . . . , en} in Cn so that {ek : i 6= k 6= j} is orthonormal and orthogo-
nal to {ei, ej} and such that ej = β. We have that
λ =D Ae, eE = τn (A) + αij ¯β + αjiβ.
Since β ∈ C with β = 1 it follows that αij = 0.
(7). This follows from (6) and the obvious fact that
Re Wc (A) = Wc (Re A) and Im Wc (A) = Wc (Im A) .
correlation matrices.
(8). This is an easy consequence of the compactness of the set En of n × n
(9). This follows from (1), the fact that En = (cid:8)BT : B ∈ En(cid:9) , and the fact
that τn(cid:0)ST(cid:1) = τn (S) for every S ∈ Mn (C).
Here is a fundamental problem in this paper.
Problem 1: What is a necessary and sufficient condition for Wc (A) ⊆
[0,∞) or (0,∞)? In particular, is it true that Wc (A) ⊆ [0,∞) if an only if A is
the sum of a trace-zero diagonal operator and a positive semidefinite operator?
Note that if A is a limit of matrices of the form "positive semidefinite +
zero-trace diagonal", then Wc (A) ⊆ [0.∞). However, this set of matrices is
norm closed.
Lemma 2 Suppose A ∈ Mn (C). The following are equivalent.
1. A is the sum of a positive semidefinite matrix and a trace-zero diagonal
matrix
2. There is a sequence {Ak} of positive semidefinite operators and a sequence
{Dk} of trace-zero diagonal operators such that kAk + Dk − Ak → 0.
Proof. The implication (1) =⇒ (2) is obvious. Suppose (2) is true. Then
T r (Ak) = T r (Ak + Dk) → T r(A).
3
Since 0 ≤ Ak, we have kAkk ≤ T r(Ak). Hence there is a subsequence (cid:8)Akj(cid:9)
that converges to some P ≥ 0. Also
Dkj =(cid:0)Akj + Dkj(cid:1) − Akj → A − P = D
for some zero-trace diagonal operator D. Hence A = P + D shows that (1) is
true.
It is easily shown that W (S ⊕ T ) is the convex hull of W (S)∪ W (T ). Here
is the analogue for Wc.
Lemma 3 Suppose A ∈ Mn (C) and A = (cid:18) S1
0
for k = 1, 2. Then
0
S2 (cid:19) where Sj ∈ Mkj (C)
Wc (A) =
k1
n
Wc (S1) +
k2
n
Wc (S2) .
It follows from Theorem 1 that in Wc is actually a function on Mn (C) /Dn,0,
i.e., in comparing Wc (T ) with W (T ) we see that Dn plays the role of CIn and
Dn,0 plays the role of 0. Here are some more examples.
It is true that W (U ∗T U ) = W (T ) for every operator T and every unitary
operator U . It is known that, for every ∗-automorphism α of Mn (C) , there is
a unitary matrix U such that α = adU , i.e.,
α (T ) = U ∗T U
for every T ∈ Mn (C). Let Gn be the group of unitary matrices generated by
the diagonal unitaries and the permutation matrices.
Proposition 4 Suppose U ∈ Mn (C) unitary. The following are equivalent.
1. adU (Dn) ⊆ Dn
2. adU (Dn,0) ⊆ Dn,0
3. Wc (U ∗AU ) = Wc (A) for every A ∈ Mn (C)
4. U ∈ Gn
Proof. (1) ⇔ (2). This is obvious since adU always preserve the trace and
adU (I) = I
(3) =⇒ (1). This follows from the fact that Dn is the set of all T ∈ Mn such
that Wc (T ) is a singleton.
4
(1) =⇒ (4). This is well-known. Since adU is a automorphism of Dn
and Dn is ∗-isomorphic to C ({1, 2, . . . , n}), and since every automorphism on
C ({1, 2, . . . , n}) is composition with a homeomorphism on {1, 2, . . . , n}, there
is a unitary permutation matrix W ∈ Mn (C) such that
adU (D) = adW (D)
for every D ∈ Dn. Hence U W ∗ commutes with every diagonal matrix, i.e.,
V = U W ∗ is a diagonal unitary matrix, so U = V W ∈ Gn.
(4) =⇒ (3) . It is easily seen that if U ∈ Gn and B ∈ Mn (C), then B is a
correlation matrix if and only if U BU ∗ is a correlation matrix. Moreover,
τn (adU (A) B) = τn (AU BU ∗) ,
so it follows from part (1) of theorem 1 that Wc (adU (A)) = Wc (A) whenever
U ∈ Gn.
3 The Correlation Numerical Radius
The numerical radius w (T ) is defined by
w (T ) = sup{λ : λ ∈ W (T )} .
We define the C-numerical radius as
wc (A) = max{λ : λ ∈ Wc (A)} .
A classical result is that
Define
kTk /2 ≤ w (T ) ≤ kTk .
kTkc = inf
D∈Dn,0 kT − Dk ,
which is the norm of the image of T in Mn (C) /Dn,0.
Proposition 5 Suppose n ∈ N. There is a number κn > 0 such that
κn kTkc ≤ wc (T ) ≤ kTkc
for every T ∈ Mn (C). Moreover, when n ≥ 2,
2
4n + 2 ≤ κn ≤
n
1
.
5
Proof. We know that kkc and wc are seminorms on Mn (C) that are 0 exactly
on Dn,0. Since Mn (C) is finite-dimension, these seminorms are equivalent and
the existence of κn is proved. Moreover, for any T ∈ Mn (C) and any D ∈ Dn,0,
we have
wc (T ) = wc (T − D) ≤ kT − Dk .
Hence wc (T ) ≤ kTkc.
Suppose n ≥ 2 and let A ∈ Mn (C) be the direct sum of (cid:18) 0
(n − 2) × (n − 2) zero matrix. It follows from Lemma 3 that wc (A) = 2
kTkc = 1. Thus κn ≤ 2/n.
from [12] that there is a diagonal projection matrix P such that
Suppose T = T ∗ is a matrix all of whose diagonal entries are 0. It follows
n and
1
0 (cid:19) with an
0
dist (T,Dn) ≤ 2 kT P − P Tk .
Let kk1 denote the trace-class norm on Mn (C), i.e., kSk1 = T r(cid:16)(S∗S)
2(cid:17).
Suppose X ∈ Mn (C) and X = (1 − P ) XP and kXk ≤ 1. Then X + X ∗ is a
selfadjoint zero-diagonal contraction, so P + X + X ∗ = B = (bij) is a correlation
matrix. Moreover, if we write T = (tij) and (1 − P ) T P = (sij)
1
T r(cid:0)T BT(cid:1) =Xi6=j
so we get
Taking the supremum over all X we obtain
wc (T ) ≥(cid:12)(cid:12)τn(cid:0)T BT(cid:1)(cid:12)(cid:12) ≥
wc (T ) ≥
2
n k(1 − P ) T Pk1 ≥
tij bis = 2 ReXsijxij ,
Xsij xij(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
n(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
2
.
2
n k(1 − P ) T Pk ≥
2
n
dist (T,Dn) .
n
2
wc (T ) .
But if we choose D ∈ Dn such that kT − Dk = dist (T,Dn) , we have that
dist (D,Dn,0) = τn (D) = τn (T − D) ≤
Hence,
kTkc ≤ kT − Dk + dist (D,Dn,0) ≤ nwc (T ) .
For an arbitrary T = T ∗, we can write T = T1 + D1 with T1 = T ∗
1 a
zero-diagonal matrix and D1 ∈ Dn. We have
kTkc ≤ kT1kc + kD1kc = kT1kc + τn (T ) ≤
nwc (T1) + wc (T ) ≤ n [wc (D1) + wc (T )] + wc (T ) ≤ (2n + 1) wc (T ) .
For the general case, we know from the fact that wc (T ∗) = wc (T ) and kT ∗kc =
kTkc that, for an arbitrary T ∈ Mn (C) we have
kTkc ≤ kRe Tkc + kIm Tkc ≤ (2n + 1) [wc (Re T ) + wc (Im T )] ≤
6
(2n + 1) [wc (T ) + wc (T )] ≤ (4n + 2) wn (T ) .
Hence κn ≥ 1
4n+2 .
In the preceding proposition, there was no attempt to get the best estimates
of κn. This leads to a natural question.
Problem 2. What is the exact value of κn?
The following result relates w and wc.
Proposition 6 Suppose A ∈ Mn (C). Then
wc (A) ≤ inf
D∈Dn,0
w (A + D) .
Proof. We know that wc (A) = wc (A + D) ≤ w (A + D) for every D ∈ Dn,0,
and the result follows.
4 Connes' Embedding Problem; Correlation Ma-
trices From Unitaries
In this section we consider a special class of correlation matrices. Suppose k is a
positive integer. We can make Mk (C) into a Hilbert space with inner product
hS, Ti = τk (T ∗S) .
In this case we have that each unitary matrix U in Mk (C) is a unit vector. We
say that an n × n correlation matrix is unitarily induced if there is a positive
integer k and unitary k × k matrices U1, . . . , Uk such that
A =(cid:0)τn(cid:0)U ∗
j Ui(cid:1)(cid:1) .
The set Fn of n × n unitarily induced correlation matrices is generally not
convex. This is because if U and V are unitary and s, t are nonzero numbers,
then sU + tV is unitary if and only if
σ (s + tU ∗V ) ⊂ T,
where T = {λ ∈ C : λ = 1}; equivalently if
1
t
σ (U ∗T ) ⊂
T −
s
t
.
Details of this are worked out in [2]. If T ∈ Mn (C), we define
Wuc (T ) = co ({τn (T A) : A ∈ Fn}) =
{τn (T B) : B ∈ coFn} .
Since Fn ⊂ En, it is clear that Wuc (T ) ⊂ Wc (T ) . It was proved in [9] (later in
[2]) that En is the convex hull of Fn if and only if n ≤ 3.
7
Lemma 7 Suppose n is a positive integer. Then
1. If n ≤ 3 and T ∈ Mn (C), then
Wc (T ) = Wuc (T ) .
2. If n > 3, then there is a T = T ∗ in Mn (C) such that
Wuc (T ) 6= Wc (T ) .
Proof. (1) . This follows from the fact [9] that En = coFn when n ≤ 3.
(2). Suppose n > 3. It follows from [9] that there is a B ∈ En such that
B /∈ coFn. It follows from the Hahn Banach theorem that there is a continuous
linear functional ϕ on Mn (C) and a real number t such that
Re ϕ (A) < t < ϕ (B)
for every A ∈ coFn. We know that there is a K ∈ Mn (C) such that
ϕ (S) = τn (SK)
for every S ∈ Mn (C). If S ≥ 0, then
Re ϕ (S) = Re τn(cid:16)S
Hence, if T = Re K, we have
1
2 KS
1
2(cid:17) = τn (S Re K) .
Wuc (T ) ⊂ (−∞.t) and t < τn(T B) ∈ Wc (T ) .
Hence Wuc (T ) 6= Wc (T ).
We now relate Wuc (T ) to Connes' famous embedding problem [1], which
asks if every finite von Neumann algebra can be tracially embedded in an ultra-
product of an ultrapower of the hyperfinite II1 factor. Fortunately, the reader
does not need to know the meaning of any of the terms in the preceding sen-
tence, because of a lovely reformulation [5] of Connes' embedding problem in
terms of matrices. This reformulation, which is an extension of results in [3]
and [7] , was further studied in [2].
Let Fn denote the free group on n generators {u1, . . . , un}, and let An denote
the group algebra of Fn. The definition u∗ = u−1 on Fn extends to an involution
∗ on An.
If T = (αij ) ∈ Mn (C), we define an element pT (u1, . . . , un) ∈ An by
pT (u1, . . . , un) =
αij u∗
i uj.
n
Xi,j=1
Let Pn be the set of elements of An that can be written in the form
m
m
q∗
j qj +
Xj=1
Xj=1
(fjgj − gjfj) .
8
elements q1, . . . , qm, f1, g1, . . . fm, gm ∈ An.
following:
The revised version of Connes' embedding problem in [5] is equivalent to the
For every positive integer n, and every selfadjoint A in Mn (C), if Wuc (A) ⊂
(0,∞), then pA ∈ Pn.
This question was answered affirmatively by Popovych [5] when n = 3.
Note that Pn is closed under addition and multiplication by nonnegative
scalars and that the set of A = A∗ in Mn (C) such that Wuc ((0,∞)) is also
closed under addition and multiplication by nonnegative scalars. Note that the
map T → pT from Mn (C) to An is linear. Here is one simple observation.
Lemma 8 Suppose A ∈ Mn (C) and A ≥ 0. Then pA ∈ Pn.
Proof. Every nonnegative A is a sum of rank-one nonnegative matrices. It fol-
lows from the remarks preceding this lemma that we can assume that rankA =
1, which means that A can be written as
Then
pA =
n
βi ¯βju∗
A = (β1, . . . , βn) (β1, . . . , βn)∗ =(cid:0)βi ¯βj(cid:1) .
Xi,j=1
¯βiui!∗ n
Xi=1
i uj = n
Xi=1
¯βiui! ∈ Pn.
Remark 9 One more observation is that pS = pT if and only if S − T ∈ Dn,0.
The the map T 7→ pT is really a function on Mn (C) /Dn,0. Thus if A is the
sum of a positive semidefinite matrix and a zero-trace diagonal matrix, then
pA ∈ Pn. This means that Problem 1 is related to this scenario.
An affirmative answer to Problem 1 yields an affirmative answer to the fol-
lowing problem. If Connes' embedding problem has an affirmative answer, then
so must the following problem. Since the set Fn of correlation matrices has such
a simple definition, this question should be easier to resolve.
Problem 3. If Wc (A) ⊂ (0,∞), the must pA ∈ Pn?
5 The Case n = 2.
All of the questions can be answered when n = 2. A complete description of
W (T ) when T is a 2 × 2 matrix is given in [4].
9
Lemma 10 If T =(cid:18) a b
d (cid:19) , then
c
Wc (T ) = Wuc (T ) =
a + d
2
+ W (cid:18)(cid:18) 0
c
and W (cid:18)(cid:18) 0
c
b
0 (cid:19)(cid:19) is
b
0 (cid:19)(cid:19) = W (cid:18)(cid:18) a+d
2
c
a+d
b
2 (cid:19)(cid:19) ,
1. the disk centered at 0 with radius 1
2 max (b ,c) if bc = 0.
2. the segment from −
3. the elliptical disk with foci ±√bc if bc 6= 0 and b 6= c .
√bc to √bc if b = c , and
The following result give the answers to most of the questions in the preced-
ing question when n = 2.
Corollary 11 Suppose T = (cid:18) a b
[0,∞), then T is the sum of a trace-zero diagonal matrix (cid:18) a 0
and a positive semidefinite matrix (cid:18) a+d
2 (cid:19). So pT ∈ P2.
d (cid:19) ∈ M2 (C).
If Wuc (T ) = Wc (T ) ⊆
0 d (cid:19) − τ2 (T ) I2
a+d
2
c
c
b
A well known result of V. Pelligrini [10] and a result of R. Kadison [6] implies
that a linear numerical range-preserving map ϕ : Mn (C) → Mn (C) has the
form
or
ϕ (S) = U ∗SU
ϕ (S) = U ∗ST U
for some unitary matrix U . See [9] for more general results.
Theorem 12 Suppose ϕ : M2 (C) → M2 (C) is linear. The following are
equivalent.
1. For every T ∈ M2 (C) we have Wc (ϕ (T )) = Wc (T ) .
2. There is a linear map α : M2 (C) → D2,0 and a unitary U ∈ D2 such that
ϕ (T ) = U ∗T U + α (T )
10
Proof. We know that if T = T ∗, then Wc (ϕ (T )) ⊂ R, which implies Im ϕ (T ) ∈
D2,0. More generally, this implies that, for every T ∈ M2 (C), we have
ϕ (T ∗) − ϕ (T )∗ ∈ D2,0.
Let {eij : 1 ≤ i, j ≤ 2} be the standard matrix units for M2 (C). We know that
Wc (ϕ (e12)) = W (e12) is the disk centered at 0 with radius 1/2. Hence, by
Lemma 10, there is a λ ∈ C with λ = 1 and a D12 ∈ Dn,0 such that either
ϕ (e12) = λe12 + D12 or ϕ (e12) = λe21 + D12.
Case 1: ϕ (e12) = λe12. It follows from the fact that e11 ∈ D2,0 that
Wc (ϕ (e11) − e11) = Wc (ϕ (e11)) − τ2 (e11) = {0} .
Whence, D11 = ϕ (e11)− e11 ∈ D2,0. Similarly, D22 = ϕ (e22)− e22 ∈ D2,0. Also
D21 = ϕ (e21) − ¯λe21 = ϕ (e∗
12) − ϕ (e12)∗ ∈ D2,0.
Define α : M2 (C) → D2,0 by
Hence
c
α(cid:18)(cid:18) a b
ϕ(cid:18)(cid:18) a b
(cid:18) λ 0
d (cid:19)(cid:19) = aD11 + bD12 + cD21 + dD22.
d (cid:19)(cid:19) =(cid:18) a
d (cid:19)(cid:19) =
1 (cid:19)(cid:18) a b
d (cid:19)(cid:19) .
d (cid:19) + α(cid:18)(cid:18) a b
1 (cid:19) + α(cid:18)(cid:18) a b
d (cid:19)(cid:18) λ 0
¯λc
λb
0
0
c
c
c
c
Case 2: ϕ (e12) = λe21 +D12. If we define ψ (A) = ϕ (A)T , then Wc (ψ (T )) =
12. Hence, by Case 1, there is a linear
Wc (T ) always holds and ψ (e12) = λe12 +DT
function α : M2 (C) → D2,0 such that
ψ(cid:18)(cid:18) a b
c d (cid:19)(cid:19) =(cid:18) λ 0
1 (cid:19)(cid:18) a b
0
c
Hence,
c
ϕ(cid:18)(cid:18) a b
(cid:18) 0
λ 0 (cid:19)(cid:18) a b
1
c
d (cid:19)(cid:19) .
c
0
d (cid:19)(cid:18) λ 0
1 (cid:19) + α(cid:18)(cid:18) a b
d (cid:19)(cid:19)T
+ α(cid:18)(cid:18) a b
d (cid:19)(cid:19)T
d (cid:19)(cid:19) = ψ(cid:18)(cid:18) a b
d (cid:19)(cid:18) 0
λ 0 (cid:19)∗
=
1
c
c
.
Acknowledgement. The second author is supported by a research grant from
the National Science Foundation.
11
References
[1] A. Connes, Classification of injective factors, Cases II1, II∞, IIIλ, λ 6= 1,
Ann. of Math. (2) 104 (1976), no. 1, 73 -- 115.
[2] K. Dykema, K. Juschenko, Matrices of unitary moments, arXiv:0901.0288,
to appear in Math. Scand.,
[3] D. Hadwin, A noncommutative moment problem, Proc. Amer. Math. Soc.
129 (2001), no. 6, 1785 -- 1791.
[4] P. Halmos, A Hilbert space problem book. Second edition. Graduate Texts
in Mathematics, 19. Encyclopedia of Mathematics and its Applications, 17.
Springer-Verlag, New York-Berlin, 1982.
[5] K. Juschenko, S. Popovych, Algebraic reformulation of Connes' embedding
problem and the free group algebra, Israel J. Math, 181(2011), 305 -- 315.
[6] R. V. Kadison, Isometries of operator algebras, Ann. of Math 54 (1951)
325-338.
[7] I. Klep, M. Schweighofer, Connes' embedding conjecture and sums of her-
mitian squares (2008) Advances in Mathematics, 217 (4), pp. 1816-1837.
[8] Chi-Kwong Li and Ahmed Ramzi Sourour, Linear operators on matrix
algebras that preserve the numerical range, numerical radius or the states,
Canad. J. Math 56 (2004) 134-167.
[9] Chi-Kwong Li, Bit Shun Tam, A note on extreme correlation matrices,
SIAM J. Matrix Anal. Appl. 15 (1994), no. 3, 903 -- 908.
[10] V. J. Pelligrini, Numerical range preserving operators on a Banach Algebra,
Studia Math. 54 (1975) 143-147.
[11] S. Popovych, Trace-positive complex polynomials in three unitaries Proc.
Amer. Math. Soc. 138 (2010), 3541-3550.
[12] S. Rosenoer, Distance estimates for von Neumann algebras. Proc. Amer.
Math. Soc. 86 (1982) 248 -- 252.
12
|
1207.6741 | 1 | 1207 | 2012-07-29T03:07:55 | Amenability and Uniqueness | [
"math.OA"
] | The main result of this paper is a characterization of properly infinite injective von Neumann algebras and of nuclear C*-algebras by using a uniqueness theorem, based on generalizations of Voiculescu's famous Weyl-von Neumann theorem. | math.OA | math |
AMENABILITY AND UNIQUENESS
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
1. Introduction
One of the most important achievements up to now in the theory of opera-
tor algebras is the complete classification up to isomorphism of injective von
Neumann factors with separable preduals. This remarkable classification
built on important earlier work by several mathematician was mostly ac-
complished by A. Connes with the final case settled by U. Haagerup. Part
of Connes and Haagerup's classification is the proof that several natural
classes of factors (the injective, semi-discrete, approximately finite dimen-
sional) coincide. This classification has very close connections and analogies
with the theory of nuclear C*-algebras. Nuclear C*-algebras form an im-
portant class of C*-algebras since the important work of Choi and Effros,
Connes, Effros and Lance, and Haagerup (see for example [5], [6].
[7], [8],
[15] ) and are the main object of study in the Elliott classification program.
This class includes important examples of C*-algebras coming from group
representation theory, dynamical systems and mathematical physics.
The main result of this paper is a characterization of properly infinite in-
jective von Neumann algebras and of nuclear C*-algebras by using a unique-
ness theorem, based on generalizations of Voiculescu's famous Weyl-von
Neumann theorem.
Before stating Hadwin's generalization of Voiculescu's theorem, proved in
[18], let us recall that if H is a Hilbert space and T ∈ B(H), then the rank
of T , denoted by rank(T ) is the dimension of the closure of the range of T .
Theorem 1.1 ([18]). Let A be a C*-algebra and H be a Hilbert space. Let
φ, ψ be two *-homomorphisms from A to B(H).
Then a necessary and sufficient condition for φ and ψ to be approximately
unitarily equivalent is that
rank (φ(a)) = rank (ψ(a)),
for all a ∈ A .
Our first goal in this paper is to present versions of Theorem 1.1, where
either B(H) is replaced by a semidiscrete von Neumann algebra or the C*-
algebra A is nuclear and weaker notions of approximate unitary equivalence
are used.
Date: September 9, 2018.
2010 Mathematics Subject Classification. 46L05, 46L10.
A.C. thanks AARMS for postdoctoral support. T.G. and Z.N. were partially supported
by a grant from NSERC Canada.
1
2
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
Let us introduce these notions.
Definition 1.1. Let A, B be C*-algebras, with B unital, and let φ and ψ be
two *-homomorphisms from A to B. Then
(a) φ and ψ are said to be approximately unitarily equivalent if there exists
a net {uα} of unitaries in B such that for all a ∈ A
uαφ(a)u∗
α → ψ(a)
in the norm topology on B .
(b) φ and ψ are said to be weakly approximately unitarily equivalent if there
exist two nets {uα} and {vβ} of unitaries in B such that for all a ∈ A,
and
uαφ(a)u∗
α → ψ(a)
vβψ(a)v∗
β → φ(a)
in the relative weak topology on B (ie., the σ(B, B∗)-topology).
(c) If B is a von Neumann algebra M , then φ and ψ are weak* approxi-
mately unitarily equivalent if there exist two nets {uα} and {vβ} of unitaries
in M such that for all a ∈ A,
and
uαφ(a)u∗
α → ψ(a)
vβψ(a)v∗
β → φ(a)
in the weak* topology on M (i.e., the σ(M, M∗)-topology).
These notions of approximate unitarily equivalence have been previously
defined and studied, with B = B(H) (see [3], [13] and [17]) and in [19] for
a von Neumann algebra B. (The relationship between our work and [19] is
for a nuclear C*-algebra.)
Before stating our first result, let us recall the following notion of rank
introduced by Hadwin in [18]: two *-homomorphisms φ and ψ from a C*-
algebra A to a von Neumann algebra M have the same W*-rank if for
each positive element a ∈ A, the support projections of φ(a) and ψ(a) are
Murray-von Neumann equivalent in M.
Our first result is then:
Theorem 1.2. Let A be a C*-algebra and M be a von Neumann algebra,
and φ and ψ be two *-homomorphisms from A to M.
If either A is nuclear or M is semidiscrete, then for φ and ψ to be weak*
approximately unitarily equivalent, it is necessary and sufficient that they
have the same W*-rank.
The proof of Theorem 1.2 is obtained by first considering the case of
factors with separable preduals, then of general von Neumann algebras also
with separable preduals, using reduction theory, before proving the general
case.
Theorem 1.2 and its converse for a properly infinite von Neumann algebra
give a new characterization of semidiscreteness. Indeed we have:
AMENABILITY AND UNIQUENESS
3
Theorem 1.3. Let M be a properly infinite von Neumann algebra. Then
the following statements are equivalent:
(1) M is semidiscrete.
(2) Let A be a C*-algebra and φ, ψ : A → M two *-homomorphisms.
Then W*-rank(φ) = W*-rank(ψ) if and only if φ and ψ are weak*
approximately unitarily equivalent.
As in Theorem 1.3, our new characterization for nuclear C*-algebras uses
two uniqueness properties for C*-algebras. To state it we need first (see Def-
inition 5.1) to introduce a notion of C*-rank for pairs of *-homomorphisms
from a C*-algebra A to a unital one B, based on studies of the Cuntz semi-
group associated to a C*-algebra.
Theorem 1.4. Let A be a separable C*-algebra. Then the following are
equivalent:
(1) A is nuclear.
(2) A has the weak* uniqueness property (ie.
for every von Neumann
algebra M, for all pairs of *-homomorphisms φ and ψ from A to
M, φ and ψ have the same W*-rank if and only if they are weak*
approximatively unitarily equivalent).
(3) A has the weak uniqueness property (ie. for every unital C*-algebra
B, for all pairs of *-homomorphisms φ and ψ from A to B, φ and ψ
have the same C*-rank if and only if they are weakly approximatively
unitarily equivalent).
In C*-algebra theory, uniqueness theorems play an important role both
in extension theory (see for example [3], [23], [11] ) and in Elliott classifica-
tion program for nuclear C*-algebras, but there the approximately unitarily
equivalence in norm is used.
Notations: Let M be a von Neumann algebra and M∗ be its predual.
- For all ρ ∈ M∗, let · w∗,ρ denote the semi-norm defined for x ∈ M by
xw∗,ρ = ρ(x) .
The weak* topology (or σ(M, M∗)-topology), also called the σ-weak topol-
ogy, is the topology induced by the semi-norms · w∗,ρ , ρ ∈ M∗.
- For all φ ∈ M+
x ∈ M by
φ) denote the semi-norm defined for
∗ , let · φ (resp. · ♯
xφ =pφ(x∗x)
(resp. x♯
φ =pφ(x∗x + xx∗) )
The strong topology (or s(M, M∗)-topology) is the topology induced by
∗ and the strong* topology (or s∗(M, M∗)-
the semi-norms · φ , φ ∈ M+
topology) the topology induced by the semi-norms · ♯
φ , φ ∈ M+
∗ .
It is well-known (see for example [25], II. 4. 10) that on the unitary group
U (M) all the above topologies coincide.
4
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
2. Uniqueness for finite von Neumann algebra codomains
Proposition 2.1. Let M be a countably decomposable finite von Neumann
algebra. Let A be a C*-algebra.
Suppose that either A is nuclear or M is injective.
Let φ, ψ : A → M be two *-homomorphisms. Then
τ ◦ φ = τ ◦ ψ
for every normal tracial state τ on M if and only if there exist two nets
{uα} and {vβ} of unitaries in M such that for all a ∈ A,
and
uαφ(a)u∗
α → ψ(a)
vβψ(a)v∗
β → φ(a)
where the convergence is in the σ-strong* topology on M.
Proof. The if direction is clear. So it suffices to prove the only if direction.
Note that the condition τ ◦φ = τ ◦ψ for every τ implies that φ and ψ have
the same kernel. Hence, we may assume that both φ and ψ are injective,
A ⊂ M, and φ is the inclusion of A into M.
Let ǫ > 0 and let F = {a1, a2, ..., an} ⊂ A be a finite subset. Let G
be a finite collection of normal tracial states on M. As A′′ is hyperfinite,
there exists a finite dimensional von Neumann subalgebra F ⊂ A′′ and
x1, x2, ..., xn ∈ F such that kxi − aikτ < ǫ for 1 ≤ i ≤ n, and for every τ ∈ G.
(See, for example, [6], [7], [10].)
Let ψ be the extension of ψ to A′′. We have τ ◦ ψ = τ for every normal
tracial state τ on M. Hence,
kψ(ai) − ψ(xi)kτ = kψ(ai − xi)kτ = kxi − aikτ
for every τ ∈ G and for 1 ≤ i ≤ n.
Since F and ψ(F ) are finite dimensional von Neumann subalgebras of M
and τ ◦ ψ = τ for every normal tracial state τ on M, there exists a unitary
u ∈ M such that
Hence, for 1 ≤ i ≤ n, and for every τ ∈ G,
ψF = Ad(u)
kψ(ai) − uaiu∗kτ
≤ kψ(ai) − ψ(xi)kτ + kψ(xi) − uxiu∗kτ + ku(xi − ai)u∗kτ
< ǫ + 0 + ǫ
= 2ǫ
as required.
(cid:3)
AMENABILITY AND UNIQUENESS
5
3. Uniqueness for properly infinite von Neumann factor
codomains
We will need the following excision of pure states result (a generalization
of Glimm's Lemma), due to Akemann, Anderson and Pedersen, whose proof
can be found in [1] Proposition 2.2. (See also [11] Lemma 8 or [23] Lemma
5.3.2 for a short proof of special case.):
Lemma 3.1. Let A be a C*-algebra. Let ρ be a pure state on A. Then
there exists a net {aα} of positive elements in A with kaαk = ρ(aα) = 1 for
all α, such that
limαkaα(a − ρ(a))aαk = 0.
for all a ∈ A.
Lemma 3.2. Let M be a properly infinite von Neumann algebra and let A be
a C*-algebra with a faithful state. Let σ : A → Mn(C) and η : Mn(C) → M
be completely positive contractive maps. Let ψ : A → M be the completely
positive contractive map given by
ψ =df η ◦ σ
Then ψ can be approximated in the pointwise-norm operator topology by
finite sums of maps of the form
a 7→ mρn(a∗
0aa0)m∗
where ρ is a pure state on A, ρn =df ρ ⊗ idMn(C) : Mn(A) → Mn(C) is the
natural map induced by ρ, m is a row matrix in Mn and a0 is a row matrix
in An.
Proof. Firstly, by [2] Lemma 4.4 and the Krein -- Milman Theorem, the map
σ can be approximated on finite sets (i.e., in the pointwise norm topology)
by finite sums of maps of the form
A → Mn(C) : a 7→ ρn(a∗
0aa0)
where a0 is a row matrix over A with length n and where ρ is a pure state
on A.
Hence, to complete the proof, it suffices to prove that there exists a row
matrix m over M, with length n, such that ν can be expressed in the form
But this follows from [16] Proposition 2.1.
Mn → M : x 7→ mxm∗.
(cid:3)
The above lemma also works with the von Neumann algebra M replaced
with a unital C*-algebra which contains a unital copy of the Cuntz algebra
O2. The proof involves a variation on the argument of [16] Proposition 2.1.
In order to make our arguments go through, we need to restrict ourselves
to the factor case. We will eventually remove this condition.
6
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
For a properly infinite von Neumann factor M, we let KM denote the
Breuer ideal of M, i.e., the C*-ideal of M generated by the finite projections.
(Hence, if M is type III then KM = 0.)
Lemma 3.3. Let M be a countably decomposable properly infinite von Neu-
mann factor. Let bi ∈ M for 1 ≤ i ≤ n, and let p′, q′ ∈ M be infinite (i.e.,
p′, q′ /∈ KM) projections.
Then there exist infinite projections p, q ∈ M such that p ≤ p′, q ≤ q′ and
pbiq = 0 for 1 ≤ i ≤ n.
Sketch of proof: By induction, it suffices to prove this for n = 1.
Let p′′ ∈ M be the left support projection of p′b1q′. Then p′′ ≤ p′.
If p′ − p′′ is an infinite projection, then take p =df p′ − p′′ and q =df q′.
Suppose that p′ − p′′ is not an infinite projection. Then p′′ is infinite and
hence, a properly infinite projection. Hence, let p1, p2 ∈ M be pairwise
orthogonal projections such that p1 ∼ p2 ∼ p′′ and p′′ = p1 + p2. Let
q1 ∈ M be the right support projection of p1b1q′. Then q2 =df q′ − q1 is
an infinite projection such that p1b1q2 = 0. So take p =df p1 and q =df q2.
(Clearly, p1b1q2 = 0. So it suffices to show that q2 is an infinite projection.
2 be the left support projection of p′b1q2. Then since q1 ⊕ q2 = q′,
Let p′
p1 ∨ p′
2 = p′′ = p1 ⊕ p2. Hence, p1 ∨ p′
2 − p1 = p2 which is an infinite
projection. But, by [25] Proposition V.1.6, (p1 ∨ p′
2).
Hence, p′
2 is an infinite projection. Hence, q2 is an infinite projection.) (cid:3)
2) − p1 ∼ p′
2 − (p1 ∧ p′
Before continuing, we recall some notation used, for example, in the study
of the Cuntz semigroup (e.g., [24]). For ǫ > 0, let (t − ǫ)+ : (−∞, ∞) →
[0, ∞) be the function which is given by
(t − ǫ)+ =(t − ǫ
0
if t ≥ ǫ
otherwise
Let A be a C*-algebra. For a self-adjoint element a ∈ A and for ǫ > 0,
let (a − ǫ)+ ∈ A be the positive element gotten by applying (t − ǫ)+ and
the continuous functional calculus to a. For a positive element c ∈ A, let
Her(c) ⊆ A denote the hereditary C*-subalgebra generated by c (i.e., the
smallest hereditary C*-subalgebra of A that contains c.)
Lemma 3.4. Let M be a countably decomposable properly infinite von Neu-
mann factor and let A be a C ∗-subalgebra of M with a unit. Let p ∈ KM be
a finite projection. Suppose that σ : A → Mn(C) and η : Mn(C) → M are
two completely positive maps with the following properties:
(a) If ψ =df η ◦ σ then ψ(1A) is a projection in M
(b) σA∩KM = ψA∩KM = 0
Then ψ can be approximated in the pointwise-norm topology by maps of
the form
a 7→ v∗av
AMENABILITY AND UNIQUENESS
7
where v is a partial isometry in M such that
pv = 0
Proof. If A ⊂ KM then we can take v = 0. Hence, we may assume that A
is not a subset of KM.
First, let ǫ > 0 be given and let F be a finite subset of A. We may assume
that 1A ∈ F. Also, we may assume that p ∈ A and ap, pa, pap ∈ F for all
a ∈ F. We will approximate ψ on F in the norm topology. For simplicity,
we may assume that the elements of F all have norm less than or equal to
one. Let δ > 0 be arbitrary.
By Lemma 3.2, let ρ1, ρ2, ..., ρk be a finite set of pure states on A with
ρiA∩KM = 0 for 1 ≤ i ≤ k, let m1, m2, ..., mk be a set of row matrices in
Mn and let a1, ..., ak be a set of row matrices in An such that on F, ψ is
within δ of the map
k
a 7→
miρi
n(a∗
i aai)m∗
i .
Xi=1
By Lemma 3.1, for each i, let ci be a positive element of A with kcik = 1
and ρi(ci) = 1 such that
diag(ci, ci, ..., ci)ρi
n(a∗
i aai)diag(ci, ci, ..., ci)
is within
δ
2k(kmik2+1) of
diag(ci, ci, ..., ci)a∗
i aaidiag(ci, ci, ..., ci)
for every a ∈ F. Here, diag(ci, ci, ..., ci) is the element of Mn(A) with ci's
in the diagonal and zeroes everywhere else.
Note that for 1 ≤ i ≤ k, since ρiA∩KM = 0 and ρi(ci) = 1, ci is a full
element of M. Hence, since M is a properly infinite factor, for each i, let
xi be an element of M with norm less than 5/4 such that for every a ∈ F,
i x∗
(a) xic2
(b) xicia∗
i = 1M for each i, and
i aajcjx∗
j = 0 for i 6= j.
(Indeed, since kcik = ρi(ci) = 1, for small enough ǫ > 0, the element
(ci − 1 + ǫ)+ is not compact and hence is full. Therefore, Her((ci − 1 + ǫ)+)
contains an infinite projection, say, p′
i ∈ M for each 1 ≤ i ≤ n. Then, by
repeatedly applying Lemma 3.3, one gets infinite subprojections pi ≤ p′
i in
M such that picia∗
j = 0 for i 6= j. Note that pi ∼ 1M (1 ≤ i ≤ n).
Then take xi =df yipi for appropriate yi ∈ M. Since pi ∈ Her((ci − 1 + ǫ)+),
yi can be chosen so that kxik < 5/4 (1 ≤ i ≤ n).)
i aajcjp∗
i=1 mixicia∗
Take x∗ =df Pk
i . Then on F, ψ is within 2δ of the map
a 7→ x∗ax. Since ap, pa, pap ∈ F for all a ∈ F, and since ψ(ap) = ψ(pa) =
ψ(pap) = 0 (since p ∈ KM) for all a ∈ F, we have that
kx∗paxk, kx∗apxk, kx∗papxk < 2δ
8
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
for all a ∈ F. Now take y =df (1 − p)x ∈ M. Then for all a ∈ F,
ky∗ay − ψ(a)k ≤ kx∗ax − ψ(a)k + kx∗apxk + kx∗paxk + kx∗papxk < 8δ.
Also, by hypothesis, ψ(1A) is a projection and ky∗1Ay − ψ(1A)k < δ.
Hence, since δ was arbitrary, if we chose δ to be small enough then we
can find a partial isometry v ∈ M such that pv = 0 and on F, ψ is within
ǫ of the map
a 7→ v∗av,
as required.
(cid:3)
Corollary 3.5. Let M be a countably decomposable properly infinite von
Neumann factor and let A be a C*-subalgebra of M. Let p ∈ KM be a finite
projection, and suppose that ψ : A → M is a *-homomorphism such that
ψA∩KM = 0
Then if A is nuclear (resp. M is semidiscrete) then ψ can be approxi-
mated in the pointwise-norm (resp. pointwise-σ-strong*) topology by maps
of the form a 7→ v∗av where v is a partial isometry in M such that pv = 0.
strong. One can check that ψ((A + C1A
strong and extend ψ to ψ where ψA = ψ and ψ(1A
Proof. We may assume that A is unital. For otherwise, we can replace A
with A + C1A
strong ) =
strong) ∩ KM) = 0. (Indeed, say
1
ψ(A)
that a ∈ A is a self-adjoint element with kak ≤ 1 such that 1A
strong −a ∈ KM.
Choose an approximate unit {bα} for A such that a ≤ bα for all α. Then
strong −ψ(a) in the strong topology.
bα−a → 1A
But 0 ≤ bα − a ≤ 1A
strong − a ∈ KM for all α. Hence, ψ(bα − a) = 0 for all
α. Hence, ψ(1A
strong − a) = 1
strong − ψ(a) = 0.)
strong −a and ψ(bα−a) → 1
ψ(A)
ψ(A)
The corollary then follows from Lemma 3.4 and the definitions of nucle-
(cid:3)
arity and semidiscreteness.
The following notion of rank was introduced by Hadwin ([18]).
Definition 3.1. Let A be a C*-algebra and let M be a von Neumann alge-
bra. Let φ, ψ : A → M be two *-homomorphisms.
Then we say that φ and ψ have the same W*-rank (and write "W*-rank(φ) =
W*-rank(ψ)") if for every positive element a ∈ A, the support projections
of φ(a) and ψ(a) are Murray-von Neumann equivalent in M.
With notation as in Definition 3.1, by [4] Theorem III.2.5.7, we have that
in the case of a finite von Neumann algebra M, W*-rank(φ) = W*-rank(ψ)
if and only if τ ◦ φ = τ ◦ ψ for every normal tracial state τ on M if and only
if T ◦ φ = T ◦ ψ where T is the unique centre-valued trace on M. (Compare
with Proposition 2.1.)
For the convenience of the reader, we recall some notation (introduced
in the introduction). For a von Neumann algebra M and a normal linear
functional ρ ∈ M∗, recall that k.kw∗,ρ is the seminorm on M given by
kxkw∗,ρ =df ρ(x) for all x ∈ M.
AMENABILITY AND UNIQUENESS
9
Lemma 3.6. Let M be a countably decomposable properly infinite von Neu-
mann factor and let A be a C*-algebra. Suppose that either A is nuclear or
M is semidiscrete.
Then if
are two *-homomorphisms such that
φ, ψ : A → M
W*-rank(φ) = W*-rank(ψ)
then there exists a net {vα} of partial isometries in M such that for all
a ∈ A,
in the weak* topology.
v∗
αψ(a)vα → φ(a)
Proof. By [18], we may assume that M is a continuous properly infinite
If M is a type III factor then this follows from Corollary 3.5.
factor.
Hence, we may assume that M is a type II∞ factor.
That W*-rank(φ) = W*-rank(ψ) implies that ker(φ) = ker(ψ). Hence,
replacing A with A/ker(ψ) if necessary, we may assume that φ and ψ are
injective. We may further assume that A is a C*-subalgebra of M and
ψ : A → M is the natural inclusion map.
Let ǫ > 0 be given and let G ⊂ M∗ be a finite collection of normal states.
Let F ⊂ A be a finite set of elements. We may assume that the elements of
F have norm less than or equal to one.
Let δ > 0 be arbitrary.
Let P, Q ∈ M be the projections that are given by P =df 1A∩KM
strong and
Q =df 1
strong respectively. P is the strong limit of an approximate
unit for A ∩ KM which quasicentralizes A (see [3] Theorem 1). Hence,
choose a positive element e ∈ A ∩ KM with norm one such that the following
statements are true:
φ(A)∩KM
(3.1)
i. If p is the support projection of e then p ∈ KM.
ii. The elements ea, ae, eae are all within δ of each other, for all a ∈ F.
iii. There is a projection r ∈ KM with r ≤ e such that ρ(P − r), ρ(Q −
φ(r)) < δ, for all ρ ∈ G.
Since Q is the (strong) limit of an approximate unit for φ(A) ∩ KM that
quasicentralizes φ(A), we have that Qφ(a) = φ(a)Q for every a ∈ A. Hence,
we have a *-homomorphism
A → M : a 7→ (1 − Q)φ(a)(1 − Q).
Since W*-rank(φ) = W*-rank(ψ), φ(A ∩ KM) = φ(A) ∩ KM. (Indeed, for
every positive a ∈ A, a ∈ KM if and only if (a − µ)+ ∈ KM for all µ > 0 if
and only if the support projection of (a − µ)+ is in KM for all µ > 0 if and
only if the support projection of (φ(a) − µ)+ = φ((a − µ)+) is in KM for all
10
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
µ > 0 (since W*-rank(φ) = W*-rank(ψ)) if and only if (φ(a) − µ)+ ∈ KM
for all µ > 0 if and only if φ(a) ∈ KM.) Hence, the above map annihilates
A ∩ KM. Therefore, by Corollary 3.5, let v ∈ M be a partial isometry such
that vv∗ = 1 − Q, vp = 0 and
(3.2)
kvav∗ − (1 − Q)φ(a)(1 − Q)kw∗,ρ < δ
for all a ∈ F and all ρ ∈ G.
Both pMp and φ(p)Mφ(p) are type II1 factors. Hence, by Proposition
2.1 (taking eAe as the domain algebra and φeAe, ψeAe as the maps), there
exists a partial isometry w with p = w∗w and φ(p) = ww∗ such that
(3.3)
kweaew∗ − φ(eae)kw∗,ρ < δ
for all a ∈ F and all ρ ∈ G.
From (3.1), (3.2) and (3.3), for all a ∈ F and all ρ ∈ G,
k(v + w)a(v + w) − φ(a)kw∗,ρ
< 4δ + k(v + w)(eae + (1 − e)a(1 − e))(v∗ + w∗) − φ(eae + (1 − e)a(1 − e))kw∗,ρ
< 8δ + kweaew∗ − φ(eae)kw∗,ρ + kvav∗ − (1 − Q)φ(a)(1 − Q)kw∗,ρ
< 10δ.
Since δ is arbitrary, if we chose δ = ǫ/10 then we would have that φ is
(cid:3)
ǫ-approximately inner over F and with respect to G.
The above lemma generalizes the results of [26], [18] and [3] which proved
the case of type I codomains. In fact, in these papers, the convergence (for
the approximate unitary equivalence)is stronger (in the norm topology).
The next result seems standard, but we did not find an exact reference.
Lemma 3.7. Let M be a countably decomposable properly infinite von Neu-
mann algebra and let F ⊂ M be a finite subset. Let G ⊂ M∗ be a finite
collection of normal states. Then for every partial isometry v ∈ M, for
every ǫ > 0, there exists a unitary u ∈ M such that
for all a ∈ F and for all ρ ∈ G.
kvav∗ − uau∗kw∗,ρ < ǫ
Sketch of Proof. We may assume that the elements of F all have norm less
than or equal to one.
Let δ > 0 be arbitrary.
Since M is properly infinite, we can find a projection p ∈ M with p ≤ v∗v
such that ρ(v∗av − v∗papv) < δ for all a ∈ F and all ρ ∈ G, and 1 − p ∼
1 ∼ 1 − vpv∗.
Since 1− p and 1− vpv∗ is properly infinite, we can find a partial isometry
w ∈ M with initial projection 1 − p and range projection 1 − vpv∗ such that
for all a ∈ F and all ρ ∈ G,
ρ(vpaw∗) + ρ(wapv∗) + ρ(waw∗) < δ.
AMENABILITY AND UNIQUENESS
11
Let u ∈ M be the unitary given by
u =df up + w.
Then for all a ∈ F and all ρ ∈ G,
ρ(vav∗ − uau∗) < 2δ.
Since δ is arbitrary, if we had chosen δ = ǫ/2 then the proof would be
complete.
(cid:3)
Corollary 3.8. Let M be a countably decomposable properly infinite von
Neumann factor and let A be a C*-algebra. Suppose that either A is nuclear
or M is injective.
Let φ, ψ : A → M be two *-homomorphisms.
Then W*-rank(φ) = W*-rank(ψ) if and only if φ and ψ are weak* ap-
proximately unitarily equivalent.
Proof. The "only if" direction follows from Lemma 3.6.
For the "if" direction, if M is type III then use that two projections are
Murray-von Neumann equivalent if and only if they are both nonzero or both
zero. If M is type II∞, then the equivalence of projections is determined by
a normal semifinite trace τ . Suppose, for contradiction, there were a ∈ A+
such that the projection supp(φ(a)) is not Murray-von Neumann equiva-
lent to the projection supp(ψ(a)). Then, without loss of generality, one
may assume that supp(ψ(a)) is finite and τ (supp(φ(a))) > τ (supp(ψ(a))),
and hence there is a continuous positive function g such that τ (φ(g(a))) >
τ (ψ(g(a))), which contradicts to the assumption that φ(g(a)) is approxi-
mately conjugate to ψ(g(a)) in the weak* topology.
(cid:3)
Note that weak* approximate unitary equivalence is a relatively flexible
notion. In Corollary 3.8 we can have examples of unital A and weak* approx-
imately unitarily equivalent maps φ and ψ where φ(1A) = 1M but ψ(1A) is
a proper subprojection of 1M. On the other hand, σ-strong* approximate
unitary equivalence is a more rigid notion. In particular, for convergence in
the σ-strong* topology, we need for both maps to be unital.
Lemma 3.9. Let M be a countably decomposable von Neumann algebra
and let A be a unital C*-algebra. Let φ, ψ : A → M be two unital *-
homomorphisms.
Then φ and ψ are weak* approximately unitarily equivalent if and only if
φ and ψ are σ-strong* approximately unitarily equivalent.
In particular, if {uα} is a net of unitaries in M such that uαψ(a)u∗
α →
ψ(a) weak* for all a ∈ A, then uαψ(a)u∗
α → ψ(a) σ-strong* for all a ∈ A.
Proof. The "if" direction is clear.
The proof of the "only if" direction follows from the fact that on the uni-
tary group of M, the weak* topology is the same as the σ-strong* topology.
Also, A is the (norm-) closed linear span its unitaries.
(cid:3)
12
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
Corollary 3.10. Let M be a countably decomposable properly infinite von
Neumann factor and let A be a unital C*-algebra. Suppose that either A is
nuclear or M is injective.
Let φ, ψ : A → M be two unital *-homomorphisms.
Then the following statements are equivalent:
(1) W*-rank(φ) = W*-rank(ψ)
(2) φ and ψ are weak* approximately unitarily equivalent
(3) φ and ψ are σ-strong* approximately unitarily equivalent.
Proof. That (1) is equivalent to (2) follows from Corollary 3.8. That (2) is
equivalent to (3) follows from Lemma 3.9.
(cid:3)
4. Uniqueness for general von Neumann algebra codomains
In this section, we will generalize the results of section 3 to a general
von Neumann algebra M with separable predual by using its direct integral
decomposition along its centre. Without loss of generality, we can assume
that M acts on a separable Hilbert space H. We refer the reader to [21]
Chapter 14 for notation and preliminary results.
The centre Z(M) of M is isomorphic to L∞(T, µ) where (T, µ) is a (lo-
cally compact compete separable metric) measure space. Let {Ht}t∈T and
{Mt}t∈T be the corresponding direct integral decompositions of H and M
respectively.
For a positive element a in a von Neumann algebra N , let supp(a) denote
the support projection of a. The first lemma is a standard computation.
Lemma 4.1. Let M be a von Neumann algebra with separable predual and
T Mtdµ(t) be its central decomposition.
Suppose that a, b ∈ M are positive elements such that supp(a) is Murray-
let M =R ⊕
von Neumman equivalent to supp(b) in M.
Then a(t), b(t) ∈ Mt are positive elements and supp(a(t)) is Murray-von
Neumann equivalent to supp(b(t)), for µ-a.e. t ∈ T .
Proposition 4.2. Let M be a properly infinite von Neumann algebra with
separable predual, and let A be a separable C*-algebra. Suppose that either
A is nuclear or M is semidiscrete.
Let φ, ψ : A → M be injective *-homomorphisms.
Then W*-rank(φ) = W*-rank(ψ) if and only if φ and ψ are weak* ap-
proximately unitarily equivalent.
Proof. To prove the sufficiency of the condition (i.e., the "only if direction")
it is enough to prove the following statement:
Let ǫ > 0 be given and let ρ ∈ M∗ be a positive normal linear functional.
Let a1, a2, ..., an ∈ A be elements such that kaik ≤ 1 for 1 ≤ i ≤ n. Then
there exists a unitary u ∈ M such that
kuφ(ai)u∗ − ψ(ai)kw∗,ρ < ǫ
for 1 ≤ i ≤ n.
AMENABILITY AND UNIQUENESS
13
Keeping the above notation, we can moreover assume that H is infinite
dimensional and therefore that Ht is a separable infinite dimensional Hilbert
space and Mt is a properly infinite factor for µ-a.e. t ∈ T . Note that by [5],
if M is semidiscrete then Mt is semidiscrete for µ-a.e. t ∈ T .
Let H be a separable infinite dimensional Hilbert space. By [21] Lemma
14.1.23, let {xt}t∈T be a family of maps such that the following hold:
(1) xt : Ht → H is a unitary isomorphism for all t ∈ T .
(2) For each measurable field ξ : t ∈ T 7→ ξ(t) ∈ Ht, the map t ∈ T 7→
xtξ(t) ∈ H is measurable (i.e., for all ν ∈ H, the map t 7→ (xtξ(t)ν)
is measurable).
(3) For each measurable field y : t ∈ T → y(t) ∈ Mt, the map t ∈
t ∈ B( H) is measurable (i.e., for all ξ, ν ∈ H, the map
t ξν) is measurable).
T 7→ xty(t)x∗
t 7→ (xty(t)x∗
Let {wk}∞
k=1 be a countable set of unitaries which is strongly dense in
U (M). Hence, {wk(t))}∞
k=1 is strongly dense in U (Mt) for µ-almost every
t ∈ T . For all k ≥ 1, by changing wk on a Borel null subset of T if necessary,
we may assume that t 7→ xtwk(t)x∗
t is Borel.
Let {ρt}t∈T , {φt}t∈T , {ψt}t∈T be the corresponding direct integral decom-
positions of ρ, φ, ψ respectively. Changing {φt}t∈T and {ψt}t∈T on a Borel
null set if necessary, we may assume that for 1 ≤ i ≤ n and k ≥ 1, the maps
t 7→ xtφt(ai)x∗
t and t 7→ ρt(wk(t)φt(ai)wk(t)∗ − ψt(ai)) are
Borel.
t , t 7→ xtψt(ai)x∗
Recall (see [22] Theorem 14.12) that if X and Y are standard Borel spaces
then a map f : X → Y is Borel if and only if its graph is Borel.
Hence, for all k ≥ 1,
Sk
n
=df
{(t, xtwk(t)x∗
t ) ∈ T × U ( H) : ρt(wk(t)φt(ai)wk(t)∗ − ψt(ai)) < ǫ/2}
\i=1
= graph{t 7→ xtwk(t)x∗
k} ∩ (
{t ∈ T : ρt(wk(t)φt(ai)wk(t)∗ − ψt(ai)) < ǫ/2} × U ( H))
n
\i=1
is Borel.
Therefore, S =df S∞
k=1 Sk is Borel.
By Lemma 4.1, since W*-rank(φ) = W*-rank(ψ), W*-rank(φt) = W*-rank(ψt)
for µ-a.e. t ∈ T . Hence, by Corollary 3.8, For µ-a.e. t ∈ T , there exists a
unitary w′ ∈ U ( H) with x∗
t w′xt ∈ Mt such that (t, w′) ∈ S.
Therefore, by [21] Theorem 14.3.6 (measurable selection principle), there
exists a Borel null set N ⊂ T and a µ measurable map T − N → U ( H) :
t 7→ u(t) such that (t, u(t)) ∈ S for all t ∈ T − N . The map t 7→ x∗
t u(t)xt is
the decomposition of a unitary u ∈ M such that
kuφ(ai)u∗ − ψ(ai)kw∗,ρ < ǫ
for 1 ≤ i ≤ n.
14
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
The proof of the necessity of the condition (i.e., the "if" direction) is
(cid:3)
similar to that of Corollary 3.8.
Proposition 4.3. Let A be a C*-algebra and M a von Neumann algebra.
Suppose that either A is nuclear or M is semidiscrete.
Let φ, ψ : A → M be *-homomorphisms.
Then W*-rank(φ) = W*-rank(ψ) if and only if φ and ψ are weak* ap-
proximately unitarily equivalent.
Proof. Let us only prove the statement in the case that A is nuclear (the
proof for the case that M is semidiscrete is similar).
By Proposition 5 of [6], for any countable subset S ⊆ A, there is a sep-
arable nuclear sub-C*-algebra C with S ⊆ C. Therefore, we may assume
that A is separable. Hence, let F be a countable dense set in A+. For each
a ∈ F and n ≥ 1, let va,n be a partial isometry witnessing the Murray-
von Neumann equivalence of the support projections of φ((a − 1/n)+) and
ψ((a − 1/n)+). By considering the von Neumann algebra generated by
φ(A) ∪ ψ(A) ∪ {va,n : a ∈ F, n ≥ 1}, we may assume further that M is
countably generated. Then M is a direct product of von Neumann algebras
with separable predual. Since any von Neumann algebra is a direct sum of
a finite von Neumann algebra and a properly infinite von Neumann algebra,
the proof is reduced to the case that M is finite or properly infinite, and it
follows from Proposition 2.1 and Proposition 4.2 respectively.
(cid:3)
5. Uniqueness for C*-algebra codomains
We need some ideas that have been useful in recent studies of the Cuntz
semigroup. The following exposition follows [24]. Let A be a C*-algebra.
A tracial weight on A is an additive function τ : A+ → [0, ∞] satisfying
τ (λa) = λτ (a) and τ (x∗x) = τ (xx∗) for all a ∈ A+, x ∈ A and λ ∈ [0, ∞).
A tracial weight τ is lower semicontinuous if τ (a) = limτ (aα) whenever
{aα} is a norm-convergent increasing net with limit a. We let T (A) denote
the collection of lower semicontinuous tracial weights on A.
Each τ ∈ T (A) induces lower semicontinuous dimension function dτ :
A+ → [0, ∞] given by dτ (a) =df supǫ>0τ (fǫ(a)) for all a ∈ A+. Here, for
every ǫ > 0, fǫ : [0, ∞) → [0, ∞) is the unique continuous function which is
0 on 0, 1 on [ǫ, ∞), and linear on [0, ǫ].
Lemma 5.1. Let A be a C*-algebra and let a, b ∈ A be positive elements,
and let supp(a), supp(b) ∈ A∗∗ be their (respective support) projections.
Then we have the following:
(1) If supp(a) is Murray -- von Neumann subequivalent to supp(b) in A∗∗
then dτ (a) ≤ dτ (b) for all τ ∈ T (A); and if supp(a) is Murray -- von
Neumann equivalent to supp(b) in A∗∗ then dτ (a) = dτ (b) for all
τ ∈ T (A).
(2) Suppose, in addition, that A is separable. Then the converse of the
above statements hold. I.e., if dτ (a) ≤ dτ (b) for all τ ∈ T (A) then
AMENABILITY AND UNIQUENESS
15
supp(a) is Murray -- von Neumann subequivalent to supp(b) in A∗∗;
and if dτ (a) = dτ (b) for all τ ∈ T (A) then supp(a) is Murray -- von
Neumann equivalent to supp(b) in A∗∗
Proof. The above are [24] Corollary 5.4 and Theorem 5.8.
(cid:3)
Motivated by Lemma 5.1 and Definition 3.1, we define a notion of rank
for maps between C*-algebras.
Definition 5.1. Let A, B be C*-algebras and let φ, ψ : A → B be *-
homomorphisms.
We say that φ and ψ have the same C*-rank (and write "C*-rank(φ) =
C*-rank(ψ)") if for every positive element a ∈ A and for every τ ∈ T (B),
dτ (φ(a)) = dτ (ψ(a)).
With notation as in Definition 5.1, let i : B → B∗∗ be the natural inclu-
sion map. Note that by Lemma 5.1, if B is separable then C*-rank(φ) =
C*-rank(ψ) if and only if W*-rank(i ◦ φ) = W*-rank(i ◦ ψ). Moreover, if we
drop the hypothesis that B is separable, we still have the "if" direction.
Definition 5.2. Let A be a C*-algebra.
(a) A has the weak* uniqueness property if for every von Neumann
algebra M, for all *-homomorphisms φ, ψ : A → M, W*-rank(φ) =
W*-rank(ψ) if and only if φ, ψ are weak* approximately unitarily
equivalent.
(b) Suppose, in addition that A is separable. A has the weak unique-
ness property if for every unital separable C*-algebra B, for all *-
homomorphisms φ, ψ : A → B, C*-rank(φ) = C*-rank(ψ) if and only
if φ, ψ are weakly approximately unitarily equivalent.
Lemma 5.2. Let A be a separable C*-algebra.
Consider the following statements:
(1) A is nuclear.
(2) A has the weak* uniqueness property.
(3) A has the weak uniqueness property.
Then (1) ⇒ (2) ⇒ (3).
Proof. That (1) implies (2) follows from Proposition 4.3.
We now prove that (2) implies (3). We first prove the "only if" direction
of (3).
It suffices to prove the following:
Let ǫ > 0 and a positive linear functional ρ ∈ B∗ be given. Let a1, a2, ..., an ∈
A be elements with kaik ≤ 1 for 1 ≤ i ≤ n. Then there exists a unitary
u ∈ B such that
kuφ(ai)u∗ − ψ(ai)kw∗,ρ < ǫ
for 1 ≤ i ≤ n.
Let i : B → B∗∗ be the natural inclusion map. Now C*-rank(φ) =
C*-rank(ψ) and B is separable. Hence, by Lemma 5.1, rank(i ◦ φ) =
16
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
rank(i ◦ ψ). Also, ρ extends to a unique normal linear functional on B∗∗
(which we also denote by "ρ"). Hence, by (2), there exists a unitary v ∈ B∗∗
such that
kvφ(ai)v∗ − ψ(ai)kw∗,ρ < ǫ/10
for 1 ≤ i ≤ n. Let {uα} be a net of unitaries in B such that uα → v in
the strong operator topology. Hence, for 1 ≤ i ≤ n, uαφ(ai)u∗
α → vφ(ai)v∗
in the strong operator topology. So, choosing α0 sufficiently large, we have
that
kuα0φ(ai)u∗
α0 − ψ(ai)kw∗,ρ < ǫ
for 1 ≤ i ≤ n.
The "only if" direction for (3) follows from once more considering the
maps i◦φ, i◦ψ : A → B∗∗ and applying Proposition 4.3 and Lemma 5.1. (cid:3)
Remark 5.3. In section 7, we will show that all the statements in Lemma
5.2 are equivalent.
We also note that in Lemma 5.2, the proof of (1) ⇒ (2) does not require
the separability of A. Also, in the proof of (2) ⇒ (3), only separability of the
codomain algebra B is required. In other words, if we modify the definition
of the weak uniqueness property (i.e., Definition 5.2 (b)) to allow for nonsep-
arable domain C*-algebras but only separable codomain C*-algebras (i.e.,
allow for nonseparable A but only unital separable B) then (2) ⇒ (3) (and
hence all of Lemma 5.2) will hold without assuming that A is separable.
6. Injectivity and uniqueness
Recall that for a von Neumann algebra M and a positive normal linear
functional χ ∈ M∗, k.kχ is the seminorm on M that is given by kxkχ =df
χ(x∗x)1/2 for all x ∈ M. Recall also that k.k♯
χ is the seminorm on M that
is given by kxk♯
= χ(x∗x)1/2+χ(xx∗)1/2
for all x ∈ M.
2
χ =df
kxkχ+kx∗kχ
2
Lemma 6.1. Let M be a properly infinite von Neumann algebra and let χ ∈
M∗ be a normal state. For every ǫ > 0 and for each finite set x1, x2, ..., xm
of unitaries in M, there exist two orthogonal projections P, Q ∈ M and
finitely many elements z1, z2, ..., zm in the closed unit ball of P MP such
that the following hold:
(1) P ∼ Q ∼ 1M
(2) P + Q = 1M
(3) kzk − xkk♯
(4) kQk♯
χ < ǫ
χ < 6ǫ for 1 ≤ k ≤ m
Proof. We may assume that ǫ < 1.
As 1M is properly infinite, there exists (see, for example, [9] Corollary
III.8.6.2) a sequence {en} of pairwise orthogonal projections in M with
en ∼ 1 for all n and 1 = P en, where the sum converges in the strong
operator topology. Let S be the family of normal states S =df {χ, χ ◦
AMENABILITY AND UNIQUENESS
17
Ad(xk), χ ◦ Ad(x∗
k) : 1 ≤ k ≤ m}. Let N ≥ 1 be an integer such that
n=1 ρ(en) > 1 − ǫ2 for all ρ ∈ S.
Set
PN
N
Xn=1
(6.1)
P =df
en
and Q =df 1 − P.
Then ρ(P ) > 1 − ǫ2 and ρ(Q) < ǫ2 for ρ ∈ S. In particular, kQk♯
ǫ.
χ = kQkχ <
For all ρ ∈ S and all b ∈ M with 0 ≤ b ≤ 1, we have by the Cauchy --
Schwarz inequality that ρ(P bQ), ρ(QbP ), ρ(QbQ) all are strictly less
than ǫ2, and therefore
(6.2)
ρ(b) − ρ(P bP ) < 3ǫ2
Set zk =df P xkP for 1 ≤ k ≤ m. Then
kxk − zkk♯
χ ≤ kP xkQk♯
χ + kQxkP k♯
χ + kQxkQk♯
χ
As kaQkχ ≤ kakkQkχ for a ∈ M, we have for 1 ≤ k ≤ m, that
kxk − zkk♯
χ ≤ 2kQkχ + (1/2)(kQxk P kχ + kQx∗
kP kχ)
From this, (6.2) and (6.1), for 1 ≤ k ≤ m,
kxk − zkk♯
χ
< 2kQkχ + (1/2)(ρ(x∗
< 2ǫ + (1/2)(ǫ2 + ǫ2) + 3ǫ2
< 6ǫ.
k Qxk) + ρ(xkQx∗
k)) + 3ǫ2
(cid:3)
Theorem 6.2. Let M be a properly infinite von Neumann algebra. Then
the following statements are equivalent:
(1) M is semidiscrete.
(2) Let A be a C*-algebra and φ, ψ : A → M two *-homomorphisms.
Then W*-rank(φ) = W*-rank(ψ) if and only if φ and ψ are weak*
approximately unitarily equivalent.
Proof. That (1) implies (2) follows from Proposition 4.3.
We now prove that (2) implies (1).
We now prove the "if" direction. By [10] and [12], it suffices to prove that
for a normal state χ ∈ M∗, for any finite set of unitaries x1, x2, ..., xm ∈
U (M) and any ǫ > 0, there exists a finite dimensional unital subalgebra F
of M and elements y1, y2, ..., ym ∈ F such that
for 1 ≤ k ≤ m.
kxk − ykk♯
χ < ǫ
18
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
By Lemma 6.1, let P, Q ∈ M be orthogonal projections and zk (for 1 ≤
k ≤ m) an element of P MP with norm less than or equal to one such that
(6.3)
(1) P ∼ Q ∼ 1M
(2) P + Q = 1M
(3) kzk − xkk♯
(4) kQk♯
Let A be the unital C*-subalgebra of P MP that is generated by {P, z1, z2, ..., zm}.
χ < ǫ/2 for 1 ≤ k ≤ m and
χ < ǫ/2.
Let γ : A → QMQ be a full unital *-homomorphism.
(γ is full means
that for every positive a ∈ A − {0}, γ(a) is a full element of QMQ.)
Let φ : A → M be the unital full *-homomorphism that is given by
φ(a) =df a ⊕ γ(a) for all a ∈ A.
Since M is properly infinite, M = N ⊗B(H) where N is a von Neumann
algebra and H is a separable infinite dimensional Hilbert space. Hence, let
ψ : A → 1N ⊗ B(H) be any unital full *-homomorphism.
Since φ and ψ are both full, W*-rank(φ) = W*-rank(ψ). Hence, by
hypothesis, φ and ψ are weak* approximately unitarily equivalent. Hence,
by Lemma 3.9, φ and ψ are σ-strong* approximately unitarily equivalent.
Hence, there exists a net {uα}α∈I of unitaries in M such that for all a ∈ A,
uαψ(a)u∗
α → φ(a) in the σ-strong* topology. In particular, for 1 ≤ k ≤ m,
uαψ(zk)u∗
α → φ(zk) = zk ⊕ γ(zk) in the σ-strong* topology. From this and
(6.3), there exists α0 ∈ I such that for 1 ≤ k ≤ m,
χ < ǫ.
kxk − uα0ψ(zk)u∗
α0k♯
Now since 1N ⊗ B(H) is approximately finite dimensional, there exists a
sequence {Fn} of finite dimensional unital C*-subalgebras of 1N ⊗ B(H) and
for 1 ≤ k ≤ m, there exists an element wk,n ∈ Fn such that wk,n → ψ(zk) in
the σ-strong* topology. Hence, there exists N ≥ 1 such that for 1 ≤ k ≤ m,
kxk − uα0 wk,N u∗
α0k♯
χ < ǫ
and uα0wk,N u∗
α0 is an element of the finite dimensional C*-algebra uα0FN u∗
α0 .
(cid:3)
For finite von Neumann algebras, we have the following:
Theorem 6.3. Let M be a type II1 factor with separable predual, and
(unique) faithful normal tracial state τ . Then the following statements are
equivalent:
(1) M is injective.
(2) For every C ∗-algebra A, for all ∗-homomorphisms φ, ψ : A → M⊗M,
W*-rank(φ) = W*-rank(ψ) if and only if φ and ψ are σ-strong* ap-
proximately unitarily equivalent.
Proof. That (1) implies (2) follows from Proposition 2.1 and the fact that if
M is the injective type II1 factor then M⊗M is also the injective type II1
factor.
AMENABILITY AND UNIQUENESS
19
We now prove that (2) implies (1). By [7], it is enough to show that the
flip automorphism σ ∈ Aut(M⊗M), which is given by σ(x ⊗ y) = y ⊗ x
(x, y ∈ M), is approximately inner.
Let A be a separable unital C ∗-subalgebra of M⊗M such that A is k.kτ -
dense in M⊗M.
Note that on A, τ ◦ σ = τ . Hence, by assumption, there exists a sequence
{un}∞
n=1 of unitaries in M⊗M such that for all a ∈ A,
kσ(a) − unau∗
nkτ → 0
as n → ∞.
As M is type II1 and A is k.kτ -dense in M⊗M, it follows that σ is
(cid:3)
approximately inner, as required.
7. Nuclearity and uniqueness
Theorem 7.1. Let A be a separable C*-algebra. Then the following are
equivalent:
(1) A is nuclear.
(2) A has the weak* uniqueness property.
(3) A has the weak uniqueness property.
Proof. The directions (1) ⇒ (2) ⇒ (3) follow from Lemma 5.2.
We now prove that (3) implies (1).
Let H be a separable infinite dimensional Hilbert space. We will first
prove that the von Neumann algebra M =df A∗∗ ⊗ B(H) is injective (where
the tensor product is the spatial tensor product).
By [12] (see also [10]), it suffices to prove the following: Let ǫ > 0 and let
ρ ∈ M∗ be a normal state. Let x1, x2, ...., xm ∈ M be elements such that for
1 ≤ k ≤ m, xk = ak ⊗ bk where ak ∈ A, bk compact operators in B(H) and
kakk, kbkk ≤ 1. Then there exists a finite dimensional unital C*-subalgebra
F ⊂ M and yk ∈ F (for 1 ≤ k ≤ m) such that
kxk − ykk♯
ρ < ǫ
for 1 ≤ k ≤ m.
Since B(H) is injective, we may assume that there exists a finite dimen-
sional simple C*-subalgebra F0 ⊆ B(H) such that bk ∈ F0 for 1 ≤ k ≤ m
and 1B(H) − 1F0 is Murray-von Neumann equivalent to 1 in B(H). Let
P =df 1M − 1A∗∗ ⊗ 1F0. We may also assume that ρ(1M − P ) < ǫ/3.
Suppose that N ≥ 1 is such that F0 ∼= MN (N by N matrices over complex
numbers). Let A =df A + C1A∗∗ . (Note that if A is unital then A = A.) Let
{ei,j}1≤i,j≤N be a system of matrix units for MN (i.e., for F0). Consider the
natural map i : A ⊗ MN → A ⊗ F0 : a ⊗ ei,j 7→ a ⊗ ei,j for all a ∈ A and for
1 ≤ i, j ≤ N . Then xk ∈ i(A⊗MN ) for 1 ≤ k ≤ m. Let γ : A⊗MN → P MP
be a full unital *-homomorphism. (Recall that γ is full means that for all
positive a ∈ MN ( A) − {0}, γ(a) is a full element of P MP .)
20
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
Let φ : MN ( A) → M be the full unital *-homomorphism that is given by
φ =df i ⊕ γ. Note that
(7.1)
kφ(xk) − xkk♯
ρ < ǫ/3
for 1 ≤ k ≤ m.
and ψ A⊗e1,1
) = C*-rank(ψ A⊗e1,1
Let ψ : MN ( A) → 1A∗∗ ⊗ B(H) be a full unital *-homomorphism. For
1 ≤ i, j ≤ N , let pi,j =df φ(1A∗∗ ⊗ ei,j) and qi,j =df ψ(1A∗∗ ⊗ ei,j). Since φ
and ψ are both full maps, pi,i is Murray -- von Neumann equivalent to qi,i in
M for 1 ≤ i ≤ N . Thus, conjugating ψ (and hence conjugating 1A∗∗ ⊗B(H))
by a unitary if necessary, we may assume that pi,j = qi,j for 1 ≤ i, j ≤ N .
Let C ⊆ M be a separable C*-subalgebra such that the following hold:
(1) C is unital and 1C = p1,1,
(2) C is properly infinite,
(3) φ( A ⊗ e1,1) and ψ( A ⊗ e1,1) are both full C*-subalgebras of C.
Since φ A⊗e1,1
are both full maps into C and since C is prop-
erly infinite, we have that C*-rank(φ A⊗e1,1
) (as maps
into C). Hence, since A has the weak uniqueness property, φ A⊗e1,1
and
ψ A⊗e1,1
are weakly approximately unitarily equivalent (as maps into C).
Hence, as maps into M, φ and ψ are weak* approximately unitarily equiv-
alent.
Since φ and ψ are unital maps, by Lemma 3.9, φ and ψ are σ-strong*
approximately unitarily equivalent (as maps into M). From this and (7.1),
let u ∈ M be a unitary such that
(7.2)
for 1 ≤ k ≤ m.
kxk − uψ(xk)u∗k♯
ρ < ǫ/3
Since B(H) is an injective von Neumann algebra, there exists a sequence
{Fn} of finite dimensional C*-subalgebras of 1A∗∗ ⊗B(H) and elements cn,k ∈
Fn (for 1 ≤ k ≤ m) such that cn,k → ψ(xk) in the σ-strong* topology for
1 ≤ k ≤ m. Note that since Ad(u)ρ ∈ M∗, we must have that kuψ(xk)u∗ −
ucn,ku∗k♯
Ad(u)ρ → 0. From this and (7.2), there exists
n0 ≥ 1 such that
ρ = kψ(xk) − cn,kk♯
kxk − ucn0,ku∗k♯
ρ < ǫ
for 1 ≤ k ≤ m. Since uFn0u∗ is finite dimensional, we are done.
Since A∗∗ ⊗ B(H) is injective, A∗∗ is injective. Hence, by [6], A is nuclear.
(cid:3)
Remark 7.2. In Theorem 7.1, the direction (2) ⇒ (1) does not require
separability of A. The proof is a modification of the proof of (3) ⇒ (1).
Indeed, if we modify Definition 5.2 (b) (the definition of the weak unique-
ness property) to allow for nonseparable A (though still separable unital
codomains B), the proof of (3) ⇒ (1) would still work with minor modifi-
cations. From this and the remarks after Lemma 5.2, we have that, with a
AMENABILITY AND UNIQUENESS
21
modification of Definition 5.2, all the statements in Theorem 7.1 are equiv-
alent.
References
[1] C. A. Akemann, J. H. Anderson and G. K. Pedersen; Excising states of C ∗-algebras,
Canad. J. Math. 38 (1986), 1239 -- 1260.
[2] C. Anantharaman -- Delaroche and J. F. Havet, On approximate factorizations of com-
pletely positive maps, Journal of Functional Analysis, 90 (1990), 411 -- 428.
[3] W. Arveson, Notes on extensions of C*-algebras, Duke Math. J., 44 (1977), no. 2,
329 -- 355.
[4] B. Blackadar, Operator Algebras. Theory of C*-algebras and von Neumann algebras.
Encyclopedia of Mathematical Sciences, 122, Operator Algebras and Noncommuta-
tive Geometry, III. Springer -- Verlag, Berlin, 2006.
[5] M. D. Choi and E. Effros, Separable nuclear C*-algebras and injectivity, Duke Math.
J., 43 (1976), no. 2, 309 -- 322.
[6] M. D. Choi and E. Effros, Nuclear C*-algebras and injectivity:
the general case,
Indiana Univ. Math. J., 26 (1977), no. 3, 443 -- 446.
[7] A. Connes, A classification of injective factors. Cases I I1, I I∞, I I Iλ, λ 6= 1 Ann. of
Math. (2) 104 (1976), no. 1, 73 -- 115.
[8] E. Effros and C. Lance, Tensor products of operator algebras, Adv. Math., 25 (1977),
no. 1, 1 -- 34.
[9] J. Dixmier, Les algebres d'operateurs dans l'espace hilbertien (algebres de von Neu-
mann) Deuxieme edition, revue et augmentee. Gauthier -- Villars Editeur, Paris, 1969.
[10] G. A. Elliott, On approximately finite-dimensional von Neumann algebras. II, Canad.
Math. Bull. 21 (1978), no. 4, 415 -- 418.
[11] G. A. Elliott and D. Kucerovsky, An abstract Brown-Douglas-Fillmore absorption
theorem, Pacific J. of Math. 3 (2001), 1 -- 25.
[12] G. A. Elliott and E. J. Woods, The equivalence of various definitions for a properly
infinite von Neumann algebra to be approximately finite dimensional, Proc. Amer.
Math. Soc., 60 (1976), 175 -- 178.
[13] K. Davidson, C ∗-algebras by example Fields Institute Monographs, 6 (1996) Amer-
ican Mathematical Society, Providence, RI
[14] M. Gromov and V. D. Milman, A topological application of the isoperimetric inequal-
ity, Amer. J. Math. 105 (1983), no. 4, 843 -- 854.
[15] U. Haagerup, All nuclear C ∗C?-algebras are amenable., Invent. Math., 74 (1983),
no.2, 305 -- 319.
[16] U. Haagerup, A new proof of the equivalence of injectivity and hyperfiniteness for
factors on a separable Hilbert space, Journal of Functional Analysis, 62 (1985), 160 --
201.
[17] D. W. Hadwin, An operator-valued spectrum, Indiana Univ. Math. J., 26 (1977), no.
2, 329 -- 340.
[18] D. W. Hadwin, Nonseparable approximate equivalence, Trans. Amer. Math. Soc., 266
(1981), no. 1, 203 -- 231.
[19] H. Ding and D. W. Hadwin, Approximate equivalence in von Neumann algebras, Sci.
China Ser. A, 48 (2005), no. 2, 168 -- 210.
[20] R. V. Kadison and J. R. Ringrose, Fundamentals of the theory of operator algebras.
Vol. I. Elementary theory, Graduate studies in mathematics, 15 (1997), Reprint of
the 1983 original. American Mathematical Society, Providence, RI.
[21] R. V. Kadison and J. R. Ringrose, Fundamentals of the theory of operator algebras.
Vol. II. Advanced theory, Graduate studies in mathematics, 16 (1997), Corrected
reprint of the 1986 original. American Mathematical Society, Providence, RI.
22
A. CIUPERCA, T. GIORDANO, P. W. NG, AND Z. NIU
[22] A. S. Kechris, Classical descriptive set theory, Graduate Texts in Mathematics, 156.
Springer -- Verlag, New York, 1995.
[23] H. Lin, An introduction to the classification of amenable C ∗-algebras, World Scientific
Publishing Co., Inc.; ;2001; River Edge, NJ.
[24] E. Ortega, M. Rordam and H. Thiel, The Cuntz semigroup and comparison of open
projections, J. Funct. Anal., 260 (2011), 3474 -- 3493.
[25] M. Takesaki, Theory of Operator Algebras I, Encyclopedia of Mathematical Sciences ,
124. Operator Algebras and Non-commutative Geometry, 5. Springer-Verlag, Berlin,
2002.
[26] D. V. Voiculescu, A non-commutative Weyl -- von Neumann Theorem, Rev. Roumaine
Math. Pures Appl., 21 (1976), no. 1, 97 -- 113.
E-mail address: [email protected]
Department of Mathematics and Statistics, University of Ottawa, 585 King
Edward Avenue, Ottawa, Ontario, KiN 6N5, Canada
E-mail address: [email protected]
Department of Mathematics, University of Louisiana at Lafayette, 217
Maxim D. Doucet Hall, P. O. Box 41010, Lafayette, Louisiana, 70504-1010,
USA
E-mail address: [email protected]
Department of Mathematics and Statistics, Memorial University of New-
foundland, St. John's, NL, A1C 5S7, Canada
E-mail address: [email protected]
|
1005.3197 | 3 | 1005 | 2012-01-16T10:10:01 | The universal enveloping TRO of a JB*-triple system | [
"math.OA",
"math.FA"
] | We associate to every JB*-triple system a so-called universal enveloping TRO and show some functorial properties of this correspondence. We compute the universal enveloping TROs of the finite dimensional Cartan factors. | math.OA | math |
The universal enveloping TRO of a
J B ∗-triple system
Dennis Bohle, Wend Werner
Fachbereich Mathematik und Informatik
Westfalische Wilhelms-Universitat
Einsteinstrasse 62
48149 Munster
[email protected],
[email protected]
Abstract
We associate to every J B ∗-triple system a so called universal en-
veloping TRO. We compute the universal enveloping TROs of the finite
dimensional Cartan factors.
The content of the following is part of a project of the authors that
has as its objective to show that Cartan's classification of the (hermitian)
symmetric spaces has a K-theoretic background. This project will come to
a conclusion in the follow-up [BW11].
The symmetric spaces that will come up here consist of the open unit
balls of so called J B∗-triples, an important generalization of the concept
of a C ∗-algebra.
If the dimension is finite, their open unit balls coincide
exactly with the hermitian symmetric spaces of non-compact type so that
all of these spaces are obtained trough duality.
In the present paper we overcome a difficulty that is one of the main
obstacles for a direct generalization of the K-theory of C ∗-algebras: The
impossibility, in general, to define tensor products of a J B∗-triple with the
n× n matrices over the complex numbers. The J B∗-triples that do have
this property are precisely the ternary rings of operators which as spaces
coincide with the class of (full) Hilbert-C ∗-modules.
We will study a construction which allows the passage from an arbitrary
J B∗-triple to such a ternary ring in a way, that behaves so nicely that it will
pave the way for the program ahead. In the first section we will collect some
definitions and preliminary results, the second sections contains the actual
construction of the enveloping ternary ring of operators, whereas in the third
2000 Mathematics Subject Classification: 17C65, 46L70
Key words and phrases: J C ∗-triple system, J ∗-algebra, grid, Cartan factor, ternary
ring of operators, ternary algebra, universal enveloping TRO
The first author was supported by the Graduiertenkolleg fur analytische Topologie
und Metageometrie
1 PRELIMINARIES
2
section the enveloping ternary rings of all finite-dimensional Cartan factors
are calculated. We do this quite differently from the approach in [BFT10]
(the results of which were roughly obtained around the same time) in that
we use grids. These objects will be helpful in the sequel paper [BW11]
and are reminiscent of the root systems that are central to the classical
approach. Finally, in section 4, we slightly improve a result from [BFT10]
on the structure of the enveloping ternary ring in some special cases.
All the results in the present paper are taken from the first named au-
thor's PhD-thesis [Boh11].
1 Preliminaries
We will first provide some notation, definitions and well known facts of
triple theory. Our general references for the theory of J B∗-triple systems
are [Isi89] and [Upm85]. For n∈ N we denote by Mn the n× n-matrices over
the complex numbers and if Z is a Banach space, then B(Z) is the Banach
algebra of bounded linear operators on Z. A Banach space Z together with
a sesquilinear mapping
Z × Z ∋(x, y) ↦ x◻ y ∈ B(Z)
is called a JB∗-triple system, if for the triple product
{x, y, z} ∶=(x◻ y)(z)
and all a, b, x, y, z ∈ Z the following conditions are fulfilled:
The triple product {x, y, z} is continuous in (x, y, z), it is symmetric in
the outer variables and the C ∗-condition {x, x, x} = x3 is fulfilled. The
Jordan triple identity
{a, b,{x, y, z}} ={{a, b, x}, y, z}−{x,{b, a, y}, z}+{x, y,{a, b, z}}
holds. The operator x◻ x has non-negative spectrum in the Banach algebra
B(Z) and it is hermitian (i.e. exp(it(x◻ x)) is isometric for all t∈ R).
A closed subspace W of a J B∗-triple system Z which is invariant under
the triple product, and therefore is a J B∗-triple system itself, is called a
JB∗-subtriple (or subtriple for short) of Z.
A closed subspace I of a J B∗-triple system Z is called a JB∗-triple
ideal, if {Z, I, Z}+{I, Z, Z} ⊆ I. J B∗-triple ideals of Z are J B∗-subtriples
of Z and the kernel of a J B∗-triple homomorphism is always a J B∗-triple
ideal.
Every C ∗-algebra A becomes a J B∗-triple system under the product
{a, b, c} ∶= 1
2(ab∗c+ cb∗a).
(1)
2 UNIVERSAL OBJECTS
3
More generally every closed subspace of a C ∗-algebra which is invariant
under the product (1) is a J B∗-triple, called JC∗-triple system. A J B∗-
triple system Z which is a dual Banach space is called a JBW∗-triple
system.
Its predual is usually denoted by Z∗. The triple product of a
J BW ∗-triple is separately σ(Z, Z∗)-continuous and its predual is unique.
rings of operators). These are closed subspaces T ⊆ B(H) such that
An important example of J B∗-triples is given by the TROs (ternary
xy∗z ∈ T
(2)
Z = P e
TROs become J B∗-triples under the product (1).
non-zero tripotent, then e induces a decomposition of Z into the eigenspaces
is called a tripotent. The collection of all non-zero tripotents in Z is
for all x, y, z ∈ T . SubTROs are closed subspaces U ⊆ T closed under (2)
and TRO-ideals are subTROs I of T such that IT ∗T + T I ∗T + T T ∗I ⊆ I.
Let Z be a J B∗-triple system. An element e∈ Z that satisfies {e, e, e} = e
denoted by Tri(Z). A tripotent is called minimal if {e, Z, e} = Ce. If e is a
of e◻ e, the Peirce decomposition
0(Z)⊕ P e
k(Z) ∶= {z ∈ Z ∶ {e, e, z} = k
k-space, of e◻ e, for k = 0, 1, 2. Each Peirce-k-space, k = 0, 1, 2, is again
2(T)
becomes a unital C ∗-algebra under the product a ● b ∶= ae∗b, denoted by
2(T)(e).
1(Z)⊕ P e
2(Z),
2 z} is the k
a J B∗-triple system.
In the case of a TRO T , the Peirce-2-space P e
where P e
P e
Every finite-dimensional J B∗-triple system is the direct sum of so called
Cartan factors C1, . . . , C6. The two exceptional Cartan factors C5 and C6
can be realized as subspaces of the 3 × 3-matrices over the complex Cayley
algebra O and the other four types are treated in detail in Section 4.
2 -eigenspace, the Peirce-
2 Universal objects
We prove the existence of the universal enveloping TRO and the universal
enveloping C ∗-algebra of a J B∗-triple system. As a corollary we obtain a
new proof of one of the main theorems of J B∗-triple theory.
The following lemma and theorem are generalizations of classical results
for real J B-algebras (cf. [HOS84], Theorem 7.1.3 and [AS03], Theorem 4.36).
Lemma 2.1. Let Z be a J B∗-triple system. Then there exists a Hilbert
space H such that for every J B∗-triple homomorphism ϕ ∶ Z → B(K) the
C ∗-algebra Aϕ generated by ϕ(Z) can be embedded ∗-isomorphically into
B(H).
Proof. The cardinality of ϕ(Z) is less or equal to the cardinality of Z. One
can now proceed similar to the proof of [AS03], Lemma 4.35.
2 UNIVERSAL OBJECTS
4
Theorem 2.2. Let Z be a J B∗-triple system.
(a) There exist up to ∗-isomorphism a unique C ∗-algebra C ∗(Z) and a
J B∗-triple homomorphism ψZ ∶ Z → C ∗(Z) such that
(i) For every J B∗-triple homomorphism ϕ ∶ Z → A, where A is an
arbitrary C ∗-algebra, exists a ∗-homomorphism C ∗(ϕ) ∶ C ∗(Z) →
A with C ∗(ϕ) ○ ψZ = ϕ.
(b) There exists up to TRO-isomorphism a unique TRO T ∗(Z) and a J B∗-
(ii) C ∗(Z) is generated as a C ∗-algebra by ψZ(Z).
triple homomorphism ρZ ∶ Z → T ∗(Z) such that
(i) For every J B∗-triple homomorphism α ∶ Z → T , where T is an
arbitrary TRO, exists a TRO-homomorphism T ∗(α) ∶ T ∗(Z) → T
with T ∗(α) ○ ρZ = α.
(ii) T ∗(Z) is generated as a TRO by ρZ(Z).
universal properties.
If A is a C ∗-algebra and
Proof. Let H be the Hilbert space from Lemma 2.1 and I the family of
J B∗-triple homomorphisms from Z to B(H). Let ψZ ∶= ρZ ∶= ࣷψ∈I ψ and
H ∶= ࣷψ∈I Hψ be l2-direct sums with Hψ ∶= H. Then ψZ and ρZ are J B∗-
triple homomorphisms from Z to B( H). Let C ∗(Z) be the C ∗-algebra and
T ∗(Z) the TRO generated by ρ(Z) in B( H).
ϕ ∶ Z → A is a J B∗-triple homomorphism, where ϕ(Z) w.l.o.g. generates A
as a C ∗-algebra, then we can suppose that A is a subset of B(H). Therefore
ϕ can be regarded as an element of I. Let πϕ ∶ ࣷψ∈I B(Hψ) → B(Hϕ) be
the projection onto the ϕ-component, then πϕ(ψZ(z)) = πϕ(ρZ(z)) = ϕ(z)
for all z ∈ Z. We define C ∗(ϕ) resp. T ∗(ϕ) to be the restrictions of πϕ
to C ∗(Z) resp. T ∗(Z). Uniqueness is proved in the usual way using the
We call (T ∗(Z), ρZ) the universal enveloping TRO and (C ∗(Z), ψZ)
the universal enveloping C∗-algebra of Z respectively. Most of the time
Similar to the classical case [AS03], Proposition 4.40 there exists a TRO-
Proposition 2.3. Let Z be a J B∗-triple system. There exists a TRO-
we only use T ∗(Z) and C ∗(Z) as shorter versions.
antiautomorphism on T ∗(Z):
antiautomorphism θ (i.e. a linear, bijective mapping from T ∗(Z) to T ∗(Z)
such θ(xy∗z) = θ(z)θ(y)∗θ(x) for all x, y, z ∈ T ∗(Z)) of T ∗(Z) of order 2
such that θ ○ ρZ = ρZ .
Proof. Denote by T ∗(Z)op the opposite TRO of T ∗(Z), i.e. the TRO that
coincides with T ∗(Z) as a set and is equipped with the same norm, if γ ∶
T ∗(Z) → T ∗(Z)op, γ(a) = aop is the identity mapping then (xy∗z)op =
zop(yop)∗xop for all x, y, z ∈ T ∗(Z).
2 UNIVERSAL OBJECTS
5
The composed mapping γ ○ ρZ ∶ Z → T ∗(Z)op is a J B∗-triple homomor-
phism and thus lifts to a TRO-homomorphism T ∗(γ○ρZ) ∶ T ∗(Z) → T ∗(Z)op.
We put
θ ∶= γ−1 ○ T ∗(γ ○ ρZ) ∶ T ∗(Z) → T ∗(Z).
Easily it can be seen (since θ fixes by construction ρZ(Z) which generates
T ∗(Z) as a TRO) using the universal properties of T ∗(Z) that θ is a TRO-
antiautomorphism of order 2.
We refer to θ as the canonical TRO-antiautomorphism of order 2
on T ∗(Z).
Corollary 2.4. If the J B∗-triple system Z in Theorem 2.2 is a J C ∗-triple
then the mappings ψZ and ρZ are injective.
Obviously ψZ and ρZ are the 0 mappings if Z is purely exceptional.
Lemma 2.5. For every J B∗-triple ideal I in a J B∗-triple system Z and
every J B∗-triple homomorphism ϕ ∶ I → W , where W is a J BW ∗-triple
system, there exists a J B∗-triple homomorphism Φ ∶ Z → W which extends
ϕ.
Proof. We know by [Din86] that the second dual Z ′′ of Z is a J BW ∗-triple
system and the canonical embedding ι ∶ Z → Z ′′ is an isometric J B∗-triple
isomorphism onto a norm closed w∗-dense subtriple of Z ′′. By [BC92], Re-
mark 1.1 and since W is a J BW ∗-triple system there exists a unique, w∗-
continuous extension ϕ ∶ I ′′ → W of ϕ with ϕ(I ′′)= ϕ(I)w∗
. Let
I ⊥ ∶={x∈ Z ′′ ∶ y ↦{x, i, y} is the 0 mapping for all i∈ I ′′}
Theorem 4.2 (4)). If we denote the projection of Z ′′ onto I ′′ with π we get
be the w∗-closed orthogonal complement of I ′′ with Z ′′ = I ′′⊕I ⊥ (cf. [Hor87],
the desired extension of ϕ by defining Φ ∶= ϕ ○ π ○ ι.
We obtain a new proof of an important theorem of Friedman and Russo
(cf. [FR86], Theorem 2):
Corollary 2.6. Any J B∗-triple system Z contains a unique purely excep-
tional ideal J such that Z~J is J B∗-triple isomorphic to a J C ∗-triple system.
Proof. Let J be the kernel of the mapping ρZ ∶ Z → T ∗(Z), which is a
J B∗-triple ideal. We know that Z~J is a J B∗-triple system which is J B∗-
triple isomorphic to the J B∗-triple system ρZ(Z) ⊆ T ∗(Z) and hence to a
exists a non-zero J B∗-triple homomorphism ϕ from J into some B(H). This
Let us assume that J is not purely exceptional which means that there
J C ∗-triple system.
3 CARTAN FACTORS
6
a contradiction.
J B∗-triple homomorphism extends by Lemma 2.5 to a J B∗-triple homomor-
phism φ ∶ Z → B(H). Since φ= T ∗(φ) ○ ρZ holds, φ vanishes on J, which is
Now let I be another purely exceptional ideal such that Z~I is J B∗-triple
isomorphic to a J C ∗-triple system. On the one hand we have I ⊆ ker(ρZ)= J.
On the other hand let ϕ ∶ Z → B(H) be a J B∗-triple homomorphism with
kernel I. Then ϕ has to vanish on J and therefore J ⊆ I.
3 Cartan factors
In this section we compute the universal enveloping TROs of the finite-
dimensional Cartan factors. Since the universal enveloping TROs of the two
exceptional factors are 0, we have to compute the factors of type I -- IV. We
do so by using the grids spanning these factors (cf. [DF87] and Chapter 2).
We make heavy use of the elaborate work on grids in [NR03].
3.1 Factors of type IV
n + 1 (cf. [HOS84]). Every J C ∗-triple system which is J B∗-isomorphic to
such a J C ∗-algebra is called a spin factor. We now recall the definition of
A spin system is a subset S ={id, s1, . . . , sn}, n≥ 2, of self-adjoint elements
of B(H) which satisfy the anti-commutator relation sisj + sjsi = 2δi,j for all
i, j ∈{1, . . . , n}. The complex linear span of S is a J C ∗-algebra of dimension
a spin grid: A spin grid is a collection {uj ,uj j ∈ J} (or {uj ,uj j ∈ J}∪{u0}
in finite odd dimensions), where J is an index set with 0∉ J, for j ∈ J, uj,uj
are minimal tripotents and, if we let i, j ∈ J, i≠ j, then
2uj, {uj ,uj , ui}= 1
(SPG1) {ui, ui,uj}= 1
2 uj, {uj , uj , ui}= 1
(SPG2) {ui, ui, uj}= 1
(SPG3) {ui,ui,uj}= 1
2uj, {uj ,uj ,ui}= 1
2ui,
2uj,
(SPG4) {ui, uj ,ui}= − 1
(SPG5) {uj ,ui,uj}= − 1
2 ui,
2 ui,
2 ui,
(SPG6) All other products of elements from the spin grid are 0.
In the case of finite odd dimensions (where u0 is present) we have, for all
i∈ J, the additional conditions (as exceptions of (SPG6))
(SPG7) {u0, u0, ui}= ui, {ui, ui, u0}= 1
(SPG8) {u0, u0,ui}=ui, {ui,ui, u0}= 1
(SPG9) {u0, ui, u0}= −ui,{u0,ui, u0}= −ui.
2 u0,
2 u0,
3 CARTAN FACTORS
7
Let G ∶= {ui,ui
It is known (cf. [DF87]) that every finite-dimensional spin factor is linearly
spanned by a spin grid (but not necessarily by a spin system).
∶ i ∈ I} (resp. G ∶= G ∪ {u0}) be a spin grid which
spans the J C ∗-triple Z and 1∈ I an arbitrary index. If we define a tripotent
v ∶= i(u1+u1), Neal and Russo gave a method how to construct from G (resp.
G) and v a J C ∗-triple system in [NR03], which is J B∗-triple isomorphic to
2(Z) = Z and that, if A is any von Neumann algebra
2(Z) of v that P v
2(A)(v) is a C ∗-algebra TRO-isomorphic to P v
2(A)
2(Z)(v) is the linear span of a
spin grid. More precisely, let sj = uj +uj, j ∈ I ∖{1}; tj ∶= i(uj −uj), j ∈ I.
Z and contains a spin system. First they have shown for the Peirce-2-space
P v
containing Z, then P v
(the isomorphism is the identity mapping). Moreover, they proved:
Theorem 3.1 ([NR03], 3.1). The space P v
Then a spin system in the which linearly spans P v
2(Z)(v), is given by
{sj , tk, v ∶ j ∈ I ∖{1}, k ∈ I}
or, if the spin factor is of odd finite dimension
{sj, tk, v, u0 ∶ j ∈ I ∖{1}, k ∈ I}.
2(T).
(a) We have P v
If Z = P v
Lemma 3.2. Let T be a TRO and v ∈ Tri(T).
2(T)={z ∈ T ∶ v(vz∗v)∗ v = z}.
2(Z), then T = P v
(c) If v is a tripotent in the TRO T , then the Peirce-2-space P v
(b) Let Z ⊆ B(H) be a J C ∗-triple system and T the TRO generated by Z.
2(T) is a
Proof. (a) Let z ∈ T with vv∗z + zv∗v = 2z. Then vv∗ and v∗v are projec-
tions with vv∗zv∗v+zv∗v = 2zv∗v and vv∗zv∗v+vv∗z = 2vv∗z. Thus we
2(vv∗z + zv∗v)=
have vv∗zv∗v = zv∗v = vv∗z and therefore vv∗zv∗v = 1
If z ∈ Z with vv∗zv∗v = z, then vv∗zv∗v = zv∗v and vv∗zv∗v = vv∗z.
2(Z). By (a) we get
Thus vv∗xv∗v =
2nz2n+1 = x and it follows
then vv∗ab∗cv∗v = vv∗a(vv∗bv∗v)∗cv∗v =
2(vv∗z + zv∗v)= vv∗zv∗v = z.
2 z3 . . . z2nz2n+1 ∈ T , with zj ∈ Z = P v
(b) Let x = z1z∗
vv∗zj v∗v = zj and zj = vv∗zj = zj vv∗.
(vv∗z1)z∗
2 z3 . . . z∗
2(T).
that x∈ P v
∈ P v
(vv∗av∗v)b∗(vv∗cv∗v)= ab∗c.
2n(z2n+1v∗v) = z1z∗
2(T),
(c) Let a, b, c
2 z3 . . . z∗
subTRO of T .
z.
We get 1
3 CARTAN FACTORS
8
As a first result we get an upper bound for the dimension of the universal
enveloping TRO of a spin system:
Proposition 3.3. Let Z be a spin factor of dimension k + 1< ∞. Then
dim T ∗(Z)≤ 2k.
Proof. For k = 2n let G={u1,u1, . . . , un,un} (resp. G={u1,u1, . . . , un,un}∪
{u0} for k = 2n+1) be a spin grid generating Z. Then ρZ(G) is a spin grid in
2(T ∗(Z))=
ρZ(Z)⊆ T ∗(Z). By Lemma 3.2 we have for v ∶= i(u1+u1) that P v
2(T ∗(Z))(v). The unital C ∗-algebra
T ∗(Z), which is TRO-isomorphic to P v
2(T ∗(Z))(v) contains by Theorem 3.1 a spin system {id, s1, . . . , sk}, which
2(T ∗(Z))(v) is linearly spanned by the 2k elements si1 . . . sij , where
1≤ i1 < i2 < . . . < ij and 0≤ j ≤ k.
P v
generates it as a C ∗-algebra. It is easy to see (cf. [HOS84], Remark 7.1.12)
that P v
From the proof of Proposition 3.3 we can deduce that the universal en-
veloping TRO of a spin factor is TRO-isomorphic to its universal enveloping
C ∗-algebra, once we have shown that dim T ∗(Z)= 2k.
In Jordan-C ∗-theory the following famous spin system appears
(cf. [HOS84], 6.2.1):
Let
σ1 ∶= 1
0 −1 , σ2 ∶= 0 1
1 0 and σ3 ∶= 0
−i 0
0
i
be the Pauli spin matrices.
product of a with itself.
For matrices a = (αi,j) ∈ Mk and b ∈ Ml we define a ⊗ b ∶= (αi,j b) ∈
If a ∈ Mk and l ∈ N we denote by a⊗l the l-fold tensor
Mk(Ml) = Mkl.
The so called standard spin system, which is linearly generating a(k + 1)-
dimensional spin factor in M2n, when k ≤ 2n, is given via{id, s1, . . . , sk} with
s1 ∶= σ1 ⊗ id⊗(n−1)
s3 ∶= σ3 ⊗ σ1 ⊗ id⊗(n−2),
s2l+1 ∶= σ⊗l
for 1≤ l ≤ n − 1.
Lemma 3.4. Let S ={id, s1, . . . , sk} be the standard spin system. If k = 2n,
then the TRO generated by S in M2n is M2n. If k = 2n−1 then the generated
s2 ∶= σ2 ⊗ id⊗(n−1),
s4 ∶= σ3 ⊗ σ2 ⊗ id⊗(n−2),
s2l+2 ∶= σ⊗l
3 ⊗ σ1 ⊗ id⊗(n−l−1),
3 ⊗ σ1 ⊗ id⊗(n−l−1)
TRO is TRO-isomorphic to M
2n−1 ⊕ M
2n−1.
Proof. Let T be the TRO generated by S.
3 CARTAN FACTORS
9
Let k = 2n. It suffices to show that the 3k elements
aj ∶= id⊗(j−1) ⊗σ1 ⊗ id⊗(n−j),
bj ∶= id⊗(j−1) ⊗σ2 ⊗ id⊗(n−j),
cj ∶= id⊗(j−1) ⊗σ1 ⊗ id⊗(n−j)
C ⊗ . . . ⊗ C ⊗ M2 ⊗ C ⊗ . . . ⊗ C.
for every j = 1, . . . , k are elements of T , since aj , bj , cj and id ⊗ . . . ⊗ id span
Obviously a1 = s1 ∈ T . Suppose we have shown aj ∈ T for a fixed j ≥ 1,
s2js∗
2j+1aj = id⊗(j−1) ⊗σ2σ3σ1 ⊗ σ1 id⊗(n−j−1)
= iaj+1.
Similarly we have b1 = s2 ∈ T . If we have shown for a fixed j ≥ 1 that bj ∈ T ,
then
then
s2js∗
2j+2aj = ibj+1.
Another easy induction shows that cj ∈ T for all j = 1, . . . , n.
If k = 2n− 1 we have an ∈ T , bn, cn ∉ T . Since σ1 and id ⊗ . . . ⊗ id generate
the diagonal matrices, the statement is clear.
Alternatively we could argue that T contains the identity so T has to be
a C ∗-algebra. Then the statement follows from [HOS84], Theorem 6.2.2.
Theorem 3.5. For the universal enveloping TRO of a spin factor Z with
dim Z = k + 1 we have
T ∗(Z)=
M
2n−1 ⊕ M
2n−1
M2n
if k = 2n − 1,
if k = 2n.
⎧⎪⎪⎨⎪⎪⎩
Proof. The J C ∗-triple system Z is J B∗-isomorphic to the J C ∗-algebra J
linearly generated by the standard spin system {1, s1, . . . , sk}. By the univer-
2n−1 if k = 2n − 1
sal property of T ∗(Z) we get, since J generates M
(respectively M2n if k = 2n) as a TRO, a surjective TRO-homomorphism
2n−1 if k = 2n − 1 (respectively M2n if k = 2n).
from T ∗(Z) onto M
By Proposition 3.3 this has to be an isomorphism.
2n−1 ⊕ M
2n−1 ⊕ M
3.2 Factors of type III
A hermitian grid is a family {uij ∶ i, j ∈ I} of tripotents in Z such that for
all i, j, k, l ∈ I:
(HG1) uij = uji for all i, j ∈ I.
(HG2) {ukl, ukl, uij}= 0 if {i, j} ∩{k, l} = ∅.
(HG3) {uii, uii, uij}= 1
2 uij, {uij , uij , uii}= uii if i≠ j.
3 CARTAN FACTORS
10
tinct.
(HG4) {uij , uij , ujk} = 1
(HG5) {uij , ujk, ukl}= 1
(HG6) {uij , ujk, uki}= uii if at least two of these tripotents are distinct.
2 ujk, {ujk, ujk, uij} = 1
2 uil if i≠ l.
2 uij if i, j, k are pairwise dis-
(HG7) All other products of elements from the hermitian grid are 0.
Let Z be a finite-dimensional TRO. Then the direct sum
T = r
ࣷ
α=1
Mnα,mα
can be described by so called rectangular matrix units: Let E(α, i, j) ∶=
Ei,j ∈ Mnα,mα be the matrix in Mnα,mα which is 0 everywhere except 1 in
the (i, j)-component for all 1≤ i≤ nα, 1≤ j ≤ mα and α∈{1, . . . , r}. Put
e(α)
i,j
∶=(0, . . . , 0, E(α, i, j), 0, . . . , 0) ∈ T,
i,k .
i,j
where E(α, i, j) is in the αth summand. The rectangular matrix units satisfy
(i) e(α)
l,k = e(α)
i,j e(α)
l,j ∗
e(α)
p,q = 0 for j ≠ m, n≠ p, α≠ β or β ≠ γ.
n,m∗
i,j e(β)
e(γ)
(ii) e(α)
(iii) T = lin{e(α)
∶ 1≤ α≤ r, 1≤ i≤ nα, 1≤ j ≤ mα}.
If U is another TRO which contains elements f (β)
of (i) -- (iii) for 1 ≤ i ≤ nα, 1 ≤ j ≤ mα and α, β ∈ {1, . . . , r}, then it is easy
i,j
for 1 ≤ i ≤ nα, 1 ≤ j ≤ mα and
to see that the mapping sending e(α)
α∈{1, . . . , r} is a TRO-isomorphism.
i,j
grid {uij ∶ 1≤ i, j ≤ n} and T the TRO generated by this grid. Define
Let Z be a finite-dimensional J C ∗-triple system spanned by a hermitian
satisfying the analogues
to f (α)
i,j
eij ∶= uii( n
Q
k=1
ukk)∗uji ∈ T
for 1 ≤ i, j ≤ n,. From [NR03], Lemma 3.2 (a) we can conclude that {eij}
forms a system of rectangular matrix units in T . We get that
T ∗(Z)≃ Mn.
3 CARTAN FACTORS
11
3.3 Factors of type II
A symplectic grid is a family {uij ∶ i, j ∈ I, i ≠ j} of minimal tripotents such
that for all i, j, k, l ∈ I
(SYG1) uij = −uji for i≠ j.
(SYG2) {uij , uij , ukl}= 1
2 ukl, {ukl, ukl, uij}= 1
(SYG3) {ukl, ukl, uij}= 0 if {i, j} ∩{k, l} = ∅.
(SYG4) {uij , uil, ukl}= 1
2 uij for {i, j} ∩{k, l} ≠ ∅.
2 ukj for i, j, k, l pairwise distinct.
(SYG5) All other triple products in the symplectic grid are 0.
The standard example of a finite-dimensional symplectic grid is the collection
{Ui,j ∶ 1 ≤ i, j ≤ n, i ≠ j} ⊆ Mn, where Ui,j, for i < j, is a complex n × n-
matrix, which is 0 everywhere except for the (i, j)-entry, which is 1 and
the (j, i)-entry, which is −1. This grid spans linearly the J C ∗-triple system
{A∈ Mn ∶ At = −A} of skew-symmetric n × n matrices; its TRO span is Mn.
Let G ∶= {uij ∶ i, j ∈ I, i ≠ j} be a symplectic grid, Z the J C ∗-triple
system spanned by G and T the TRO generated by it. Since for dim Z = 3
Z is J B∗-triple isomorphic to a type I Cartan factor and for dim Z = 6 it
sections, let dim Z ≥ 10.
is J B∗-triple isomorphic to a type IV Cartan factor, both covered in other
If we define
for 1 ≤ i, j, k, l ≤ n pairwise distinct, we get with [NR03], Lemma 4.1 and
Lemma 4.3 that the elements eii and eij are well-defined and that for v ∶=
∑ ekk we have
ve∗
ij v = eji
and
eij v∗ekl = δjkeil.
Using this we see that
eij e∗
klemn = eij v∗elkv∗emn
= δjlδkmein,
which shows that {eij} is a set of rectangular matrix units.
with dim Z ≥ 10, then
Theorem 3.6. If Z is a J C ∗-triple system spanned by a symplectic grid
T ∗(Z)= Mn.
and
kluil
eii ∶= uiku∗
eij ∶= eiie∗
iiuij e∗
jjejj,
3 CARTAN FACTORS
12
3.4 Factors of type I
Let ∆ and Σ be two index sets. A rectangular grid is a family {uij ∶ i ∈
∆, j ∈ Σ} of minimal tripotents such that
(RG1) {uil, uil, ujk}= 0 if i≠ j, k ≠ l.
(RG2) {uil, uil, ujk} = 1
(RG3) {ujk, ujl, uil}= 1
2 ujk, {ujk, ujk, uil} = 1
2 uik if j ≠ i and k ≠ l.
2 uil if either j = i, k ≠ l or
j ≠ i, k = l.
(RG4) All other triple products in the rectangular grid equal 0.
Let Z be the J C ∗-triple system generated by a finite rectangular grid.
We assume that Z is finite-dimensional and hence J B∗-triple isomorphic to
We first exclude some candidates for T ∗(Z):
Mn,m with m= ∆ and n= Σ .
Lemma 3.7. For the J C ∗-triple system Z = Mn,m its universal enveloping
TRO T ∗(Z) is neither TRO-isomorphic to Mn,m nor to Mm,n.
Proof. Assume that T ∗(Z) is TRO-isomorphic to Mn,m. Let t ∶ Mn,m →
T ∗(Z) there is a mapping T ∗(t) such that
Mm,n be the transposition mapping. According to the universal property of
Mn,m
ρZ
;✈✈✈✈✈✈✈✈✈
Mn,m
t
T ∗(t)
❍❍❍❍❍❍❍❍❍
/ Mm,n
ticular ρZ is a complete isometry.
commutes. Since ρZ is bijective there is a TRO-isomorphism T ∗(ρZ) ∶
Mn,m → Mn,m with T ∗(ρZ) ○ ρZ = id. This means T ∗(ρZ) = ρ−1
Since ρZ and t are bijective the same holds for T ∗(t) and it follows
that t is a complete isometry. We get a contradiction because t is not even
completely bounded. The other statement can be proved analogously.
Z , in par-
angular grid spanning Z.
Lemma 3.8 ([NR03], Lemma 5.1 (b), Lemma 5.2 (b)). Let {uij} be a rect-
(a) If for i∈ ∆, k, l ∈ Σ, where k ≠ l, we have uilu∗
ik = 0 or for i, j ∈ ∆, k ∈ Σ,
ikujk = 0, then Z is TRO-isomorphic to Mn,m.
(b) If for i∈ ∆, k, l ∈ Σ, where k ≠ l, we have u∗
iluik = 0 or for i, j ∈ ∆, k ∈ Σ,
jk = 0, then Z is TRO-isomorphic to Mm,n.
where i≠ j, we have u∗
where i≠ j, we have uiku∗
By this we get
#
#
;
/
3 CARTAN FACTORS
13
Lemma 3.9. Let {eij} be a rectangular grid spanning ρZ(Z)⊆ T ∗(Z), then
il ≠ 0 and e∗
eike∗
jk ≠ 0 and e∗
ikeil ≠ 0 for all i∈ ∆, k, l ∈ Σ
ikejk ≠ 0 for all i, j ∈ ∆, k ∈ Σ.
eike∗
(3)
(4)
we have
as well as
Proof. If one of these conditions is not fulfilled we get by Lemma 3.8 and
since ρZ(Z) generates T ∗(Z) as a TRO, that ρZ(Z) = T ∗(Z) and hence
is isomorphic to Mn,m respectively Mm,n. But this is a contradiction to
Lemma 3.7.
Lemma 3.10. Let rank Z ≥ 2 and {eij} be a rectangular grid spanning
ρZ(Z), then
p ∶= Q
i∈∆
M
j∈Σ
eije∗
ij ∈ C ∗(Z)
is a sum of non-zero orthogonal projections. We have:
pT ∗(Z)⊆ T ∗(Z), (1 − p)T ∗(Z)⊆ T ∗(Z).
Proof. Since (3) and (4) hold we can use [NR03], Lemma 5.5 and get that
∏j∈Σ eij e∗
ij ≠ 0 are orthogonal projections for all i∈ ∆.
The fact that p leaves T ∗(Z) invariant is obvious.
Lemma 3.11. For all i, k, a ∈ ∆, j, l, b ∈ Σ we have
peij(pekl)∗peab = peij e∗
klpeab ∈ lin{peij}
and for q ∶=(1 − p)
qeij(qekl)∗qeab = qeije∗
klqeab ∈ lin{qeij}.
Proof. Since {eij} is a rectangular grid we know for i≠ k and j ≠ l that
kl = 0
and therefore, for i≠ k and j ≠ l,
eije∗
and
e∗
ij ekl = 0
peil(pekl)∗ = peile∗
klp= 0
as well as
ilpekl
(peil)∗pekl = e∗
il⎛
= e∗
⎝ Q
α∈∆
= e∗
ilei1e∗
eαβe∗
αβ⎞
M
⎠ekl
β∈Σ
i1 . . . ein e∗
inekl=0 if n≠l
(5)
3 CARTAN FACTORS
= 0,
14
(6)
since the range projections of collinear tripotents commute by [NR03], Lem-
ma 5.4.
i≠ j
k ≠ l
Equation (5) and (6) lead us to the fact that we only have to prove (for
Using (5) and (6) again, we have to prove this in the following cases:
● pejk(peik)∗peab
● peab(peil)∗peik
● peab(peil)∗peil
arbitrary a∈ ∆, b∈ Σ) that
● peik(peil)∗peab
k ≠ l
● peil(peil)∗peab
● peab(peil)∗pejl
i≠ j
are elements of lin{peij}.
● peik(peil)∗peib
● peik(peil)∗peil
● pejk(peik)∗peib
● pejk(peik)∗pejk
● peil(peil)∗peib
● peil(peil)∗peil.
products to show that {peij} is a rectangular grid (cf. the proof of [NR03],
Lemma 5.6) and it is true that all of them are elements of {peij}. One can
the rectangular grid {(1 − p)eij}.
Proposition 3.12. If rank Z ≥ 2 we have for the universal enveloping TRO
● peik(peil)∗peik
● peik(peil)∗peal
● pejk(peik)∗peik
● pejk(peik)∗peak
● peil(peil)∗peal
k ≠ l, k ≠ b≠ l
k ≠ l
b≠ k, i≠ j
i≠ j
b≠ l
k ≠ l
k ≠ l, a≠ i
i≠ j
i≠ j, a≠ i
a≠ i
show by similar methods that all products in the list for q are elements of
We obtain a similar list for q. Luckily, Neal and Russo calculated all these
Proof. The rectangular grid {eij} spans ρZ(Z) which generates T ∗(Z) as a
TRO, so an element x∈ T ∗(Z) has to be of the form
2n)∗eα
3 . . .(eα
x= n
2)∗eα
1(eα
2kα+1,
λαeα
Q
α=1
of Z
especially
T ∗(Z)= lin{peij ,(1 − p)eij ∶ 1≤ i≤ n, 1≤ i≤ m}
dim T ∗(Z)≤ 2nm.
3 CARTAN FACTORS
15
with eα
1 , . . . , eα
e1, . . . , e2n+1 and e ∶= e1e∗
2kα+1 ∈ {eij}, λα ∈ C and kα ∈ N for all 1 ≤ α ≤ n, n ∈ N. Let
2 e3 . . . e2ne∗
2n+1 ∈ T ∗(Z), then
e=(pe1 +(1 − p)e1)(pe2 +(1 − p)e2)∗ . . .(pe2n+1 +(1 − p)e2n+1)
= pe1(pe2)∗ . . . pe2n+1 +(1 − p)e1((1 − p)e2)∗ . . .(1 − p)e2n+1
+ mixed terms in p and (1 − p)
= pe1(pe2)∗ . . . pe2n+1 +(1 − p)e1((1 − p)e2)∗ . . .(1 − p)e2n+1,
since {peij} ⊥{(1 − p)eij} by Lemma [NR03], Lemma 5.6. An inductive use
of Lemma 3.11 gives us e∈{peij ,(1 − p)eij ∶ 1≤ i≤ n, 1 ≤ i≤ m}.
Theorem 3.13. Let Z be a J C ∗-triple system of rank ≥ 2 and isomorphic
to a finite-dimensional Cartan factor of type I. Let {uij; 1 ≤ i≤ n, 1≤ j ≤ m}
be a grid spanning Z. Then
T ∗(Z)= Mn,m ⊕ Mm,n.
isomorphism.
Proof. We identify Z with Mn,m. The mapping Φ ∶ Mn,m → Mn,m ⊕ Mm,n,
A ↦ A, At is a J B∗-triple isomorphism onto a J B∗-subtriple of Mn,m ⊕
Mm,n which generates Mn,m ⊕ Mm,n as a TRO. Since by 3.12 dim T ∗(Z) ≤
2nm the induced mapping T ∗(Φ) ∶ T ∗(Z) → Mn,m ⊕ Mm,n has to be a TRO
For the rest of this section we assume that rank Z = 1 and Z is of finite
dimensions. This implies, that if {uij; 1 ≤ i ≤ n, 1 ≤ j ≤ m} is a rectangular
A finite rectangular grid of rank 1 is a set {u1, . . . , un} of tripotents with
grid spanning Z then n or m have to be equal to 1. In this special case the
definition of a rectangular grid becomes simpler:
(RG'1) {ui, uj , ui}= 0 for i≠ j.
(RG'2) {ui, ui, uk}= 1
2 uk for i≠ k.
(RG'3) All other products are 0.
Let Z be a n-dimensional type 1 Cartan factor of rank 1. We fix a finite
rectangular grid {e1, . . . , en} of rank 1 spanning ρZ(Z)⊆ T ∗(Z).
Lemma 3.14. Let Z be as above, then
dim T ∗(Z)≤ n
Q
k=1 n
k − 1n
k .
Proof. Using the grid properties (RG'1),(RG'2),(RG'3) we show that
T ∗(Z)= lin{ei1 e∗
i2 ei3 . . . e∗
i2k
ei2k+1 ∶ ij < ij+2, 1≤ j ≤ 2k − 1,
3 CARTAN FACTORS
16
0≤ k ≤ 1
k − 1n
2(n − 1)}.
k choices for ei1 e∗
ej2k+1
i2 ei3 . . . e∗
i2k
ei2k+1 . This is
j2 ej3 . . . e∗
j2k
elements such that x is a sum of elements of the form ej1 e∗
k choices for i1 < i3 < . . . < i2k+1 and
tripotents we can assume that we do not have three equal indices in a row. If
we have the case eαe∗
For a fixed k we have n
true because ij < ij+2. We have n
n
k − 1 choices for i2 < i4 < . . . < i2k.
To prove that T ∗(Z) is the above mentioned linear span we give an induc-
ei2k+1 ∈ T ∗(Z) and rearranges the grid
tion which takes x = ei1 e∗
i2 ei3 . . . e∗
i2k
with j1 ≤ j3 ≤ . . . ≤ j2l+1 and j2 ≤ j4 ≤ . . . ≤ j2l. Since the grid elements are
βeα, where α≠ β this equals 0 by the minimality of the
tripotents (cf. (RG'1)). Therefore ja < ja+2 for all 1 ≤ a ≤ 2l − 1. Especially
2(n − 1).
l ≤ 1
ei2k+1 ∈ T ∗(Z). Since the eia are all minimal
So let x = ei1 e∗
tripotents we can assume eia ≠ eia+2.
For k = 0 nothing is to prove. Additionally we prove the case when k = 1.
Let x= ei1 e∗
If i1 < i3 we are done.
If i1 = i2 > i3 we can use (RG'2) and get x= ei3 − ei3 e∗
If i1 > i2 = i3 we can also use (RG'2) and get x= ei1 − ei2 e∗
If i1 ≠ i2 ≠ i3:
If i1 > i3 we can use (RG'3) and we deduce x= −ei3 e∗
i2 ei1.
Now we assume that we have shown the statement for 2k + 1∈ N, 2k + 3≤ n
and for all lesser indices. If we apply our induction statement to the first
2k + 1 grid elements in the product and then apply the beginning of the
induction to all the last three elements of the products in the resulting sum,
then one can easily convince himself that in at most three repetitions of this
procedure we get the desired form for x.
i1 ei1 .
i2 ei1 .
i2 ei3 . . . e∗
i2k
i2 ei3.
Again we have to give a faithful representation T of T ∗(Z). This happens
to be more complicated than in the other cases. Again we can use the work
of Neal and Russo. In [NR03] they showed that a J C ∗-triple system, which
is linearly spanned by a finite rectangular grid of rank 1 with n elements,
has to be completely isometric (especially J B∗-triple isomorphic) to one of
the spaces H k
column Hilbert space.
n, where k = 1, . . . , n, that are generalizations of the row and
4 THE RADICAL
17
n (cf. [NR03], Section 6 and
We recall the construction of the spaces H k
n − k = n
7 or [NR06], Section 1). Let 1 ≤ k ≤ n and I, J be subsets of {1, . . . , n}
such that I has k − 1 and J has n − k elements. There are qk ∶= n
k − 1
choices for I and pk ∶= n
k choices for J. We assume that the
collections I ∶= {I1, . . . , Iqk} and J ∶= {J1, . . . , Jpk} of such sets are ordered
lexicographically. Let eI1 , . . . , eIqk
of Cpk and Cqk. We can define an element in Mpk,qk by EI,J ∶= Ei,j, when
I = Ii ∈ I and J = Jj ∈ J . The space H k
n is the linear span of matrices bn,k
where 1≤ i≤ n, given by
∶=
sgn(I, i, J)EJ,I ,
be the canonical bases
and eJ1 , . . . , eJpk
bn,k
i
(7)
Q
i
,
I∩J=∅,(I∪J)c={i}
where
sgn(I, i, J)
is
the
signature
of
the
permutation
One can show that the TRO spanned by bn,k
(i1, . . . , ik−1, i, j1, . . . , jn−k) to (1, . . . , n), when I = {i1, . . . , ik−1}, where i1 <
i2 < . . . < ik−1, and J ={j1, . . . , jn−k} and where j1 < j2 < . . . < jn−k.
equals Mpk,qk, so if
we represent our J C ∗-triple system Z as ࣷn
n we get with Lemma 3.14:
Theorem 3.15. If Z is a J C ∗-triple system spanned by a finite rectangular
grid of rank 1, then
1 , . . . , bn,k
k=1 H k
n
taking
where pk =n
k and qk = n
ࣷ
k=1
Mpk,qk,
T ∗(Z)= n
k − 1 for all k = 1, . . . , n.
With this result the list of universal enveloping TROs of the finite-
dimensional Cartan factors is complete.
4 The Radical
We use the theory of reversibility developed in [BFT10] to prove some facts
for the universal enveloping TRO of a universally reversible TRO T . We
consider the case in which a universally reversible TRO T contains an ideal
of codimension 1 which is not covered in [BFT10]. We show that there
exists an ideal R(T) in T which is universally reversible and which does not
contain an ideal of codimension 1 itself, such that T~R(T) is an Abelian
J B∗-triple system. We obtain an exact sequence
0 Ð→ R(T) ⊕ θ(R(T)) Ð→ T ∗(T) Ð→ C T
0(Epi(T~R(T), C)) Ð→ 0,
where the notation is given below.
We adopt the following definition from [BFT10]. It is the generalization
of reversibility of J C ∗-algebras.
4 THE RADICAL
18
Definition 4.1. A J C ∗-triple system Z ⊆ B(H) is said to be reversible if
1
2(x1x∗
2 x3 . . . x∗
2nx2n+1 + x2n+1x∗
2n . . . x3x∗
2x1)∈ Z
for all x1, . . . , xn ∈ Z and n ∈ N. We call a J C ∗-triple system universally
reversible if it is reversible in every representation.
Obviously every TRO, and therefore every C ∗-algebra, is reversible (but
not necessarily universally reversible, since we have to cope with J B∗-triple
homomorphisms). A J C ∗-triple system is universally reversible if and only
if it is reversible when embedded in its universal enveloping TRO.
Lemma 4.2 ([BFT10], Theorem 4.4). Let Z be a universally reversible
J C ∗-triple system and let ϕ ∶ Z → B(H) be an injective triple homomor-
TRO(ϕ(Z)) such that Ψ ○ ϕ = ϕ, then T ∗(ϕ) ∶ T ∗(Z) → TRO(ϕ(Z)) is a
phism. Suppose there exists a TRO antiautomorphism Ψ of the TRO-span
TRO-isomorphism.
Lemma 4.3 ([BFT10], Corollary 4.5). Let T be a universally reversible TRO
in a C ∗-algebra A. Suppose T has no TRO-ideals of codimension 1 and there
is a TRO antiautomorphism θ ∶ A → A of order 2. Then T ∗(T) ≃ T ⊕ θ(T)
with universal embedding a ↦(a, θ(a)).
In order to establish the announced generalization of Lemma 4.3 we
define an ideal such that the quotient of T by this ideal is Abelian. We are
first recalling some known facts about Abelian J B∗-triple systems which
allow us to compute the universal enveloping TRO of a general Abelian
triple. Afterwards we show that every ideal of a universal reversible J C ∗-
triple system is universally reversible.
Recall that a J B∗-triple system Z is called Abelian, if
{{a, b, c}, d, e} ={a,{b, c, d}, e} ={a, b,{c, d, e}}
for all a, b, c, d, e ∈ Z. The importance of Abelian J B∗-triple systems derives
from the fact that every J B∗-triple system is locally Abelian, which means
that every element in a J B∗-triple system generates an Abelian subtriple.
Every commutative C ∗-algebra is an Abelian J B∗-triple system with the
product {a, b, c} = ab∗c. We call the elements of
Epi(Z, C) ∶={ϕ ∶ Z → C ∶ ϕ≠ 0 is a triple homomorphism}
the characters of Z. Following [Kau83], §1 we consider Epi(Z, C) as a sub-
space of Z ′ = B(Z, C) and endow it with the σ(Z ∗, Z) topology. Then
Epi(Z, C) becomes a locally compact space and a principal T-bundle for the
group T={t ∈ C ∶ t = 1}. The base space Epi(Z, C)~T can be identified with
4 THE RADICAL
19
C T
space
the set of all J B∗-triple ideals I ⊆ Z such that Z~I is isometric to C. The
0(Epi(Z, C)) ∶={f ∈ C0(Epi(Z, C)) ∀t ∈ T ∀λ∈ Epi(Z, C) ∶ f(tλ)= tf(λ)}
is a subtriple of the Abelian C ∗-algebra C0(Epi(Z, C)), the continuous func-
tions on Epi(Z, C) vanishing at infinity. The mapping
∶ Z → C T
0(Epi(Z, C))
(8)
defined by x(λ)= λ(x) for all x∈ Z and λ∈ Epi(Z, C) is called the Gelfand
transform of Z.
Theorem 4.4 ([Kau95], Theorem 6.2). For every J B∗-triple system Z the
following assertions are equivalent:
(a) Z is Abelian.
(b) Z is a subtriple of a commutative C ∗-algebra.
(c) The Gelfand transform of Z is a surjective isometry onto C T
Especially, every Abelian J B∗-triple system is a TRO.
0(Epi(Z, C)).
Lemma 4.5. Let Z be an Abelian J C ∗-triple. Then Z is a universally
reversible TRO.
Proof. We only have to show that every Abelian J C ∗-triple system is already
a TRO since every TRO is already reversible, but by Theorem 4.4 we know
that Z is a subtriple of an Abelian C ∗-algebra and therefore a TRO.
Proposition 4.6. Let Z be an Abelian J C ∗-triple system, then
and the universal embedding ρZ ∶ Z → C T
transform of Z.
T ∗(Z)≃ C T
0(Epi(Z, C))
0(Epi(Z, C)) is given by the Gelfand
Proof. The Abelian J C ∗-triple system Z is by Lemma 4.5 a universally re-
versible TRO. Let ∶ Z → C T
is by Theorem 4.4 a J B∗-triple isomorphism. The identity mapping id on
C T
0(Epi(Z, C)) be the Gelfand transform, which
0(Epi(Z, C)) is, since we are in the Abelian world, also an antiautomor-
0(Epi(Z, C)) as a TRO we
phism, satisfying id ○= . Since Z generates C T
obtain the statement from Lemma 4.2.
Definition 4.7. Let Z be an universally reversible J C ∗-triple system. De-
fine the radical of Z to be the set
R(Z) ∶=
ϕ∈Epi(Z,C)∪{0}
ker(ϕ).
4 THE RADICAL
20
In the case that Epi(Z, C) = ∅ we have R(Z)= Z.
The next proposition helps us to show that the radical of a universal
reversible J C ∗-triple system is universally reversible.
Proposition 4.8. Let Z be a universally reversible J C ∗-triple system and
I ⊆ Z a J B∗-triple ideal, then I is also universally reversible.
Proof. We assume that T ∗(I) ⊆ T ∗(T).
It suffices to show that ρZ(I) ⊆
T ∗(Z) is reversible. Since T ∗(I) is a TRO-ideal and ρZ(Z) is reversible by
definition, we know that ρZ(I) is reversible, if
Let x ∈ T ∗(I) ∩ ρZ(Z) and π ∶ ρZ(Z) → ρZ(Z)~ρZ(I) be the J B∗-quotient
ρZ(I)= T ∗(I) ∩ ρZ(Z).
homomorphism. It follows from [?] Theorem 4.2.4 that
T ∗(Z)~T ∗(I)≃ T ∗(ρZ(Z)~ρZ(I))
and therefore π(x) = τ(π)(x) = 0, which yields x∈ ρZ(I).
Since the radical is always a J B∗-triple ideal the next corollary follows
immediately.
Corollary 4.9. Let Z be a universally reversible J C ∗-triple system, then
R(Z) is universally reversible.
Theorem 4.10. Let T be a universally reversible TRO embedded in a C ∗-
algebra A such that there exists a TRO antiautomorphism θ ∶ A → A of order
2. Then we have an exact sequence of TROs
0 Ð→ R(T) ⊕ θ(R(T)) Ð→ T ∗(T) Ð→ C T
Proof. By Corollary 4.9 we know that the radical R(T) is universally re-
0 (Epi(T~R(T), C)) Ð→ 0.
versible and does not contain a TRO-ideal of codimension 1 by construction.
Using Lemma 4.3 we get
(9)
T ∗(R(T)) = R(T) ⊕ θ(R(T)).
The quotient T~R(T) is an Abelian J B∗-triple system and we get with
Proposition 4.6 that
The exactness of (9) follows now from the exactness of
T ∗(T~(R(T))) = C T
0 (Epi(T~R(T), C)) .
0 Ð→ R(T) Ð→ T Ð→ T~R(T) Ð→ 0,
and [?] Theorem 4.2.4.
Theorem 4.10 is a generalization of Lemma 4.3. If we add the additional
assumption that T does not contain a one codimensional TRO-ideal, then
R(T)= T and thus (9) becomes
0 Ð→ T ⊕ θ(T) Ð→ T ∗(T) Ð→ 0 Ð→ 0.
REFERENCES
References
21
[AS03] Erik M. Alfsen and Frederic W. Shultz. Geometry of state
spaces of operator algebras. Mathematics: Theory & Applications.
Birkhauser Boston Inc., Boston, MA, 2003.
[BC92] Leslie J. Bunce and Cho-Ho Chu. Compact operations, multipliers
and Radon-Nikod´ym property in JB∗-triples. Pacific J. Math.,
153(2):249 -- 265, 1992.
[BFT10] Leslie J. Bunce, Brian Feely, and Richard Timoney. Operator space
structure of JC∗-triples and TROs, I. Preprint, 2010.
[Boh11] Dennis Bohle. K-theory for ternary structures. PhD Thesis, 2011.
[BW11] Dennis Bohle and Wend Werner. A K-theoretic approach to the
classification of symmetric spaces. Preprint, 2011.
[DF87] Truong Dang and Yaakov Friedman. Classification of JBW∗-triple
factors and applications. Math. Scand., 61(2):292 -- 330, 1987.
[Din86] Se´an Dineen. Complete holomorphic vector fields on the second
dual of a Banach space. Math. Scand., 59(1):131 -- 142, 1986.
[FR86] Yaakov Friedman and Bernard Russo. The Gel′fand-Naımark the-
orem for JB∗-triples. Duke Math. J., 53(1):139 -- 148, 1986.
[Hor87] Gunther Horn. Characterization of the predual and ideal structure
of a JBW∗-triple. Math. Scand., 61(1):117 -- 133, 1987.
[HOS84] Harald Hanche-Olsen and Erling Størmer. Jordan operator alge-
bras, volume 21 of Monographs and Studies in Mathematics. Pit-
man (Advanced Publishing Program), Boston, MA, 1984.
[Isi89]
Jos´e-M. Isidro. A glimpse at the theory of Jordan-Banach triple
systems. Rev. Mat. Univ. Complut. Madrid, 2(suppl.):145 -- 156,
1989. Congress on Functional Analysis (Madrid, 1988).
[Kau83] Wilhelm Kaup. A Riemann mapping theorem for bounded sym-
metric domains in complex Banach spaces. Math. Z., 183(4):503 --
529, 1983.
[Kau95] Wilhelm Kaup.
On JB∗-triples defined by fibre bundles.
Manuscripta Math., 87(3):379 -- 403, 1995.
[NR03] Matthew Neal and Bernard Russo. Contractive projections and
operator spaces. Trans. Amer. Math. Soc., 355(6):2223 -- 2262 (elec-
tronic), 2003.
REFERENCES
22
[NR06] Matthew Neal and Bernard Russo. Representation of contractively
complemented Hilbertian operator spaces on the Fock space. Proc.
Amer. Math. Soc., 134(2):475 -- 485 (electronic), 2006.
[Upm85] Harald Upmeier.
Symmetric Banach manifolds and Jordan
C ∗-algebras, volume 104 of North-Holland Mathematics Stud-
ies. North-Holland Publishing Co., Amsterdam, 1985. Notas de
Matem´atica [Mathematical Notes], 96.
|
1609.09453 | 2 | 1609 | 2019-05-13T21:12:42 | Connective C*-algebras | [
"math.OA",
"math.KT"
] | Connectivity is a homotopy invariant property of separable C*-algebras which has three notable consequences: absence of nontrivial projections, quasidiagonality and a more geometric realization of KK-theory for nuclear C*-algebras using asymptotic morphisms. The purpose of this paper is to further explore the class of connective C*-algebras. We give new characterizations of connectivity for exact and for nuclear separable C*-algebras and show that an extension of connective separable nuclear C*-algebras is connective. We establish connectivity or lack of connectivity for C*-algebras associated to certain classes of groups: virtually abelian groups, linear connected nilpotent Lie groups and linear connected semisimple Lie groups. | math.OA | math |
CONNECTIVE C ∗-ALGEBRAS
MARIUS DADARLAT AND ULRICH PENNIG
Abstract. Connectivity is a homotopy invariant property of separable
C ∗-algebras which has three notable consequences: absence of nontriv-
ial projections, quasidiagonality and a more geometric realization of
KK-theory for nuclear C ∗-algebras using asymptotic morphisms. The
purpose of this paper is to further explore the class of connective C ∗-
algebras. We give new characterizations of connectivity for exact and for
nuclear separable C ∗-algebras and show that an extension of connective
separable nuclear C ∗-algebras is connective. We establish connectivity
or lack of connectivity for C ∗-algebras associated to certain classes of
groups: virtually abelian groups, linear connected nilpotent Lie groups
and linear connected semisimple Lie groups.
1. Introduction
Connectivity of separable C ∗-algebras was introduced in our earlier paper
[10] under different terminology, see Definition 2.1 below. The initial mo-
tivation for studying it stemmed from our search for homotopy-symmetric
C ∗-algebras. By a result of Loring and the first author [9], these are precisely
the separable C ∗-algebras for which one can unsuspend in the E-theory of
Connes and Higson [6]. Using a result of Thomsen [34], we proved in [10]
that connectivity is equivalent to homotopy-symmetry for all separable nu-
clear C ∗-algebras. Moreover, we showed that connectivity has a number of
important permanence properties. These facts allowed us to exhibit new
classes of homotopy-symmetric C ∗-algebras.
The purpose of this paper is to further explore the class of connective
C ∗-algebras. We are motivated by the following three properties they share:
(i) If A is a separable nuclear C ∗-algebra, then KK(A, B) is isomorphic
to the homotopy classes of completely positive and contractive (cpc) asymp-
totic morphisms from A to B ⊗ K for any separable C ∗-algebra B.
(ii) Connective C ∗-algebras are quasidiagonal. In fact, if A is connective,
(A ⊗ B denotes the
then A ⊗ B is quasidiagonal for any C ∗-algebra B.
minimal tensor product.)
M.D. was partially supported by NSF grant #DMS -- 1362824.
1
2
MARIUS DADARLAT AND ULRICH PENNIG
(iii) Connective C ∗-algebras do not have nonzero projections.
In fact,
if A is connective, then A ⊗ B does not have nonzero projections for any
C ∗-algebra B.
Connectivity is of particular interest in the case of group C ∗-algebras. A
countable discrete group G is called connective if the augmentation ideal
I(G) defined as the kernel of the trivial representation ι : C ∗(G) → C is
a connective C ∗-algebra.
In view of properties (ii) and (iii) connectivity
of G may be viewed as a stringent topological property that accounts si-
multaneously for the quasidiagonality of C ∗(G) and the verification of the
Kadison-Kaplansky conjecture for certain classes of groups. Examples of
nonabelian connective groups were exhibited in [10] and [11].
In this paper we give new characterizations of connectivity for exact and
nuclear separable C ∗-algebras, see Prop. 2.2, 2.3. We prove that connectivity
of separable nuclear C ∗-algebras is preserved under extensions, see Thm. 2.4.
This is a key permanence property which does not hold for quasidiagonal
C ∗-algebras.
There is a close connection between the topology of the spectrum and
connectivity, which we employ to reveal an obstruction to connectivity by
using work of Blackadar and Cuntz [1] and Pasnicu and Rørdam [30]. In
particular, we show that a countable discrete group G is not connective if
the trivial representation ι is a shielded point of the unitary dual of G in
the sense of Def. 2.9.
Motivated by this, we give a complete description of the neighborhood
of ι in the spectrum of the Hantzsche-Wendt group G, which is a torsion
free crystallographic group with holonomy Z/2Z × Z/2Z, in Sec. 3.1. This
allows us to prove that G is not connective in this case (Cor. 3.2). Moreover,
we show that this group provides a counterexample to a conjecture from
[8]. Specifically, we prove that the natural map [[I(G), K]] → K 0(I(G)) is
not an isomorphism (Lem. 3.3). In contrast, we show that all torsion free
crystallographic groups with cyclic holonomy are connective (Thm. 3.8).
Next, we investigate connectivity for C ∗-algebras associated to Lie groups.
We show that all noncompact linear connected nilpotent Lie groups have
connective C ∗-algebras (Thm. 4.3). Using classic results from representation
theory in conjunction with permanence properties of connectivity, we show
that if G is a linear connected complex semisimple Lie group, then C ∗
r (G)
is connective if and only if G is not compact (Thm. 4.5). Moreover, if G is
a linear connected real reductive Lie group, then C ∗
r (G) is connective if and
only if G does not have a compact Cartan subgroup (Thm. 4.6).
A common denominator of our results concerning group C ∗-algebras is
that in all the cases we analyzed, C ∗(G) contains a large connective ideal.
CONNECTIVE C ∗-ALGEBRAS
3
2. Connective C ∗-algebras
2.1. Definitions and background. For a C ∗-algebra A, the cone over A
is defined as CA = C0[0, 1) ⊗ A the suspension of A as SA = C0(0, 1) ⊗ A.
The first of the following two notions was introduced in [10, Def. 2.6
(i)], under a different terminology which we have abandoned. We use the
abbreviation cpc map for a completely positive and contractive map.
Definition 2.1. Let A be a C ∗-algebra.
(a) A is connective if there is a ∗-monomorphism
Φ : A →Yn
CL(H)/Mn
CL(H)
(b) A is almost connective, if there is a (not necessarily liftable) ∗-monomor-
which is liftable to a cpc map ϕ : A →Qn CL(H).
phism Φ : A →Qn CL(H)/Ln CL(H).
For a discrete group G, we define I(G) to be the augmentation ideal, i.e.
the kernel of the trivial representation C ∗(G) → C. We will sometimes say
that a discrete amenable group G is connective if the C ∗-algebra I(G) is
connective. Note that (almost) connective C ∗-algebras do not have nonzero
projections. Thus any connective C ∗-algebra is nonunital. Our definition
allows that the zero C ∗-algebra {0} is connective.
Let A and B be C ∗-algebras. An asymptotic morphism is a family of
maps {ϕt : A → B}t∈[0,∞) such that
a) for each a ∈ A the map t 7→ ϕt(a) is norm-continuous and bounded,
b) for all a, b ∈ A and λ ∈ C, we have
lim
t→∞
kϕt(a + λ b) − (ϕt(a) + λ ϕt(b))k = 0
lim
t→∞
kϕt(ab) − ϕt(a) ϕt(b)k = 0
kϕt(a∗) − ϕt(a)∗k = 0 .
lim
t→∞
A discrete asymptotic morphism (ϕn)n∈N between A and B is a family of
maps ϕn : A → B that satisfies the analogous conditions as a) and b) above
with the index set replaced by N. A homotopy between two (discrete) asymp-
totic morphisms (ϕ0
t )t∈I is a (discrete) asymptotic morphism
Ht : A → C[0, 1] ⊗ B, such that evi ◦ Ht = ϕi
t for all t ∈ I, where I either
denotes [0, ∞) or N. We will say that a (discrete) asymptotic morphism
(ϕt)t∈I is completely positive and contractive (cpc) if each of the maps ϕt
is cpc. The corresponding homotopy classes will be denoted as follows:
t )t∈I and (ϕ1
• [[A, B]] -- homotopy classes of asymptotic morphisms,
• [[A, B]]N -- homotopy classes of discrete asymptotic morphisms,
• [[A, B]]cp
N -- homotopy classes of discrete cpc asymp. morphisms
4
MARIUS DADARLAT AND ULRICH PENNIG
2.2. Characterizations of connectivity. In the following we give two
more characterizations of connectivity for exact and respectively nuclear
C ∗-algebras.
Proposition 2.2. Let A be a separable exact C ∗-algebra. Then A is connec-
tive if and only if there is an injective ∗-homomorphism π : A → O2 which
is null-homotopic as a discrete cpc asymptotic morphism. This means that
[[π]] = 0 in the set [[A, O2]]cp
N .
Proof. (⇒) By assumption, there is a cpc discrete asymptotic morphism
{ϕn : A → C[0, 1]⊗O2}n such that ϕ(0)
n = π is an injective ∗-homomorphism
and ϕ(1)
n = 0. Thus, we can view {ϕn}n as an injective discrete asymptotic
morphism {ϕn : A → C0[0, 1) ⊗ O2 ⊂ CL(H)}n and hence A is connective.
(⇐) Suppose that A is a separable exact connective C ∗-algebra. By [10,
Prop. 2.11] it follows that [[A, O2 ⊗K]]cp
N is an abelian group. By Kirchberg's
embedding theorem, there is an injective ∗-homomorphism π : A → O2 ⊗ K.
Moreover π⊕π : A → O2⊗K is unitarily homotopy equivalent to π. It follows
that [[π]]⊕ [[π]] = [[π]] in the group [[A, O2 ⊗ K]]cp
N and hence [[π]] = 0. After
embedding O2 ⊗ K into O2 we obtain the desired conclusion.
(cid:3)
Proposition 2.3. Let A be a separable nuclear C ∗-algebra. The following
properties are equivalent.
(i) A is connective.
(ii) A ⊗ B is connective for some C ∗-algebra B that contains a nonzero
projection.
(iii) A ⊗ B is connective for all C ∗-algebras B
(iv) [[A, O2 ⊗ K]] = 0.
(v) [[A, L(H) ⊗ K]] = 0.
Proof. The equivalences (i) ⇔ (ii) ⇔ (iii) were established in [10]. (For
(ii) ⇒ (i) observe that A is a subalgebra of A ⊗ B if B contains a nonzero
projection.)
(i) ⇒ (iv) and (i) ⇒ (v). Let B be a σ-unital C ∗-algebra such that
KK(A, B) = 0, for instance B = O2 or B = L(H). If A is connective, then
A is homotopy symmetric and hence [[A, B ⊗ K]] ∼= KK(A, B) = 0 by [10,
Thm. 3.1]. Note that even though [10, Thm. 3.1] was stated for separable
C ∗-algebras B it is routine to extend the result to general C ∗-algebras using
the separability of A.
(iv) ⇒ (i) and (v) ⇒ (i). Fix an embedding π : A → O2 ⊂ L(H) ⊗ K
and regard it as a constant asymptotic morphism {πt : A → L(H) ⊗ K}t. By
assumption, [[πt]] = 0 in [[A, L(H) ⊗ K]] and hence by restriction, [[πn]] = 0
in [[A, L(H) ⊗ K]]N. We shall view L(H) ⊗ K as a subalgebra of L(H). The
corresponding homotopy from the constant discrete morphism {πn}n to zero
CONNECTIVE C ∗-ALGEBRAS
5
will induce an embedding Φ : A →Qn CL(H)/Ln CL(H) which is liftable
to a cpc map ϕ : A →Qn CL(H) by the nuclearity of A.
2.3. Extensions of connective C ∗-algebras. Connectivity of C ∗-algebras
has a plethora of permanence properties as proven in [10, Thm. 3.3]. In par-
ticular, it is inherited by split extensions [10, Thm. 3.3 (d)]. In the following
theorem this result is extended to non-split extensions as well.
(cid:3)
Theorem 2.4. Let 0 → J → A → B → 0 be an exact sequence of separable
nuclear C ∗-algebras. If J and B are connective, then A is connective.
Proof. Since connectivity passes to nuclear subalgebras we may replace the
given extension by
0 → J ⊗ O2 ⊗ K → A ⊗ O2 ⊗ K → B ⊗ O2 ⊗ K → 0.
Adding to this extension a trivial absorbing extension, using the addition in
Ext-theory, we obtain an absorbing extension
(1)
0 → J ⊗ O2 ⊗ K → E → B ⊗ O2 ⊗ K → 0,
which by construction has the property that A ⊂ A ⊗ O2 ⊗ K ⊂ E, see
for instance [4, Lemma 2.2]. Since Ext(B ⊗ O2, J ⊗ O2) = 0 as O2 is KK-
contractible and we are dealing with an absorbing extension, it follows that
the extension (1) splits by [24, Sec. 7] and so E is connective by [10, Thm.
3.3]. We conclude that A ⊂ E is connective.
(cid:3)
In the sequel we will need to use the following result from [10], which is
based on [2].
Theorem 2.5 ([10], Cor. 3.4.). Let A be a separable continuous field of
nuclear C ∗-algebras over a compact connected metrizable space X. If one of
the fibers of A is connective, then A is connective.
Corollary 2.6. Let A be a separable continuous field of nuclear C ∗-algebras
over a locally compact metrizable space X that has no compact open subsets.
Then A is connective.
Proof. Let Y = X ∪ {y0} be the one-point compactification of X. Then Y
is a compact metrizable space which must be connected. Indeed, arguing
by contradiction, say that Y = U ∪ V with U , V open and nonempty with
y0 ∈ U and U ∩ V = ∅. Then V = V ∩ X is both an open and compact
subset of X.
We can view A as a continuous field over Y (see the remark on page 145
It follows that A is
(cid:3)
of [2]) and the fiber over y0 satisfies A(y0) = {0}.
connective by Thm. 2.5.
6
MARIUS DADARLAT AND ULRICH PENNIG
2.4. Obstructions to connectivity. If A is a C ∗-algebra, we denote by
irreducible representations and by Prim(A) the primitive spectrum of A
bA the spectrum of A, which consists of all unitary equivalence classes of
consisting of kernels of irreducible representations. The unitary dual bG of a
group G identifies with \C ∗(G). Recall that bA is topologized by pulling-back
the Jacobson topology of Prim(A) under the natural map bA → Prim(A),
topology of bA, see [14, Sec. 3.5].
π 7→ ker π, [14]. Let π and (πn)n be irreducible representations of A acting
on the same separable Hilbert space H. Suppose that kπn(a)ξ − π(a)ξk → 0
for all a ∈ A and ξ ∈ H. Then the sequence (πn)n converges to π in the
Proposition 2.7. Let A be a separable C ∗-algebra.
(i) If Prim(A) contains a non-empty compact open subset, then A is not
connective.
(ii) If A is nuclear and Prim(A) is Hausdorff, then A is connective if and
only if Prim(A) does not contain a non-empty compact open subset.
Proof. (i) Set X = Prim(A). If X has a non-empty compact open subset,
then A ⊗ O2 contains a nonzero projection by [30, Prop. 2.7] and hence A
cannot be connective.
(ii) One implication follows from (i). For the other implication suppose
that X = Prim(A) does not contain a non-empty compact open subset.
Since X is Hausdorff by assumption, A is a nuclear separable continuous
field over the locally compact space X, [16]. This is explained in detail in
[3, Sec. 2.2.2]. Now we apply Cor. 2.6.
(cid:3)
We would like thank Gabor Szabo for pointing out the following invariance
property of connectivity.
Proposition 2.8. Let A and B be separable nuclear C ∗-algebra with home-
omorphic primitive spectra. Then A is connective if and only if B is con-
nective.
Proof. Kirchberg's classification theorem [26] implies that if A and B are as
in the statement, then A ⊗ O2 ⊗ K ∼= B ⊗ O2 ⊗ K. The desired conclusion
follows now from Proposition 2.3.
(cid:3)
Definition 2.9. Let A be a separable C ∗-algebra. A point π ∈ bA is called
shielded, if bA \ {π} 6= ∅ and any sequence (πn)n in bA \ {π} which converges
to π has a subsequence that converges to another point η ∈ bA \ {π}.
Lemma 2.10. Let A be a unital separable C ∗-algebra. If a point π ∈ bA is
closed and shielded, then I = ker π is not connective.
CONNECTIVE C ∗-ALGEBRAS
7
as the preimage of the topology of Prim(A). Therefore, the lemma follows
Proposition 2.7 it suffices to show that Prim(I) = Prim(A) \ {I} is a
nonempty compact-open subset of Prim(A). Since {π} is closed, it follows
Proof. Observe that I 6= {0}, since bA \ {π} 6= ∅ and {π} is closed. By
that q−1(I) = {π} and hence q(bA \ {π}) = Prim(A) \ {I}. The quotient map
q : bA → Prim(A) is continuous and open, since the topology of bA is defined
if we show that bA \ {π} is compact and open. bA \ {π} is open because {π}
is closed. Since bA is compact and satisfies the second axiom of countability
[14], it suffices to show that bA \ {π} is sequentially compact, [25, p. 138].
Let (πn)n be a sequence in bA \ {π}. By compactness of bA it contains a
subsequence (πnk )k converging in bA. If it converges to π ∈ bA, then it has a
subsequence that converges to some other point η ∈ bA \ {π}, because π is
shielded. Hence bA \ {π} is also compact.
sentation ι ∈ bG is shielded, then I(G) is not connective.
closed in bG. Thus, the statement follows from Lemma 2.10.
Corollary 2.11. Let G be a countable discrete group. If the trivial repre-
Proof. Since ι is a one-dimensional representation,
it follows that {ι} is
(cid:3)
(cid:3)
3. Connectivity of crystallographic groups
It is known that there are precisely 10 closed flat 3-dimensional manifolds.
Conway and Rossetti [7] call these manifolds platycosms ("flat universes").
The Hantzsche-Wendt manifold [17], or the didicosm in the terminology of
[7], is the only platycosm with finite homology. Its fundamental group G,
known as the Hantzsche-Wendt group, is generated by two elements x and
y subject to two relations:
x2yx2 = y,
y2xy2 = x.
The group G is one of the classic torsion free 3-dimensional crystallographic
groups, [17, 7]. It is useful to introduce the notation z = (xy)−1.
A concrete realization of G as rigid motions of R3 is given by the following
transformations X, Y , Z that correspond to the group elements x, y and z.
X(ξ) = Aξ + a,
Y (ξ) = Bξ + b, Z(ξ) = Cξ + c,
ξ ∈ R3,
where
A =
1
0
0 −1
0
0
0
0 −1
, B =
−1 0
0
0
0
1
0 −1
0
, C =
0
−1
0
0 −1 0
1
0
0
,
8
and
MARIUS DADARLAT AND ULRICH PENNIG
a =
1/2
1/2
0
,
b =
0
1/2
1/2
,
c =
1/2
0
1/2
.
The transformations X 2, Y 2 and Z 2 are just translations by unit vectors
in the positive directions of the coordinate axes.
One shows (independently of the previous concrete realization) that the
elements x2, y2 and z2 commute. Moreover one has the following relations
in G:
xx2x−1 = x2,
yx2y−1 = x−2,
zx2z−1 = x−2,
xy2x−1 = y−2,
yy2y−1 = y2,
zy2z−1 = y−2,
xz2x−1 = z−2
yz2y−1 = z−2
zz2z−1 = z2
The subgroup N of G generated by x2, y2 and z2 is normal in G and it
is isomorphic to Z3 ∼= Zx2 ⊕ Zy2 ⊕ Zz2. Let q : G → H = G/N denote the
quotient map.
1
/ N
/ G
q
/ H
/ 1
H is isomorphic to Z/2 ⊕ Z/2 with generators are q(x) and q(y).
For later use, we will need the following identities that hold in G.
(2)
x−1y = yxy2z−2 = zx2z−2,
x−1z = yz2,
y−1x = zx2,
y−1z = x(x−2y2).
3.1. Induced representations and the unitary dual of G. Based on
Corollary 2.11 we will show that I(G) for the Hantzsche-Wendt group G is
Our basic reference for this section is the book of Kaniuth and Taylor [23].
The unitary dual of G consists of unitary equivalence classes of irreducible
not connective. This requires a thorough analysis of the spectrum bG.
unitary representations of G and is denoted by bG. G acts on bN ∼= T3
by g · χ = χ(g · g−1). If we identify the character χ ∈ bN with the point
(χ(x2), χ(y2), χ(z2)) = (u, v, w) ∈ T3, then the action of G is described as
follows:
x · (u, v, w) = (u, ¯v, ¯w),
y · (u, v, w) = (¯u, v, ¯w),
z · (u, v, w) = (¯u, ¯v, w).
The stabilizer of a character χ is the subgroup Gχ of G defined by Gχ = {g ∈
G : χ(g · g−1) = χ(·)}. It is clear that N ⊂ Gχ and that there is a bijection
from G/Gχ onto the orbit of χ. In particular, the orbits of the action of G
on bN can only have length 1, 2 or 4. Mackey has shown that each irreducible
representation π ∈ bG is supported by the orbit of some character χ ∈ bN ,
in the sense that the restriction of π to N is unitarily equivalent to some
/
/
/
/
CONNECTIVE C ∗-ALGEBRAS
9
multiple mπ of the direct sum of the characters in the orbit of χ.
πN ∼ mπ Mg∈G/Gχ
χ(g · g−1).
In the sum above g runs through a set of coset representatives.
Mackey's theory has a particularly nice form for virtually abelian discrete
(χ)
, χ ∈ Ω(cid:27) .
(χ)
verges to ι has a subsequence which is convergent to a point η 6= ι.
representations σ of Gχ such the restriction of σ to N is unitarily equivalent
to a multiple of χ. Then, according to [23, Thm. 4.28]
groups. Let Ω ⊂ bN be a subset which intersects each orbit of G exactly
once. For each χ ∈ bN , let cGχ be the unitary dual of the stabilizer group
the subset of cGχ consisting of classes of irreducible
Gχ and denote by cGχ
Theorem 3.1. bG =(cid:26)ind G
Gχ(σ) : σ ∈ cGχ
Let ι be the trivial representation of G. We will prove that ι ∈ bG is
shielded by showing that any sequence (πn)n of points in bG \ {ι} that con-
Let Rℓ ⊂ bG consist of those classes of irreducible representations which
lie over ℓ-orbits, i.e. the orbits of length ℓ. Write bG as the disjoint union
bG = R1 ∪ R2 ∪ R4. It suffices to assume that all the elements πn belong to
ε1, ε2, ε3 ∈ {±1}. These are precisely the points in bN which are fixed under
in R1 ⊂ bG and such that (πn)n is convergent to ι. Since the restriction of
πn to N is a multiple of a character χn = (ε1(n), ε2(n), ε3(n)), it follows
that ε1(n) = ε2(n) = ε3(n) = 1 for all sufficiently large n and hence since
πn is irreducible, there is m such that πn = ι for n ≥ m. Hence there is no
sequence in R1 \ {ι} which converges to ι.
1-orbits. Consider the characters of N of the form χ = (ε1, ε2, ε3), where
the action of G. In other words Gχ = G. Let (πn)n be a sequence of elements
the same subset Rℓ. We distinguish the three possible cases for ℓ:
(u, v, w) where precisely only one of the coordinates is not equal to ±1. Let
2-orbits. The characters χ ∈ bN with orbits of length two are those χ =
us argue first that if (πn)n is a sequence of elements in R2 ⊂ bG such that πn
lies over the orbit of χn = (un, vn, wn) and (πn)n is convergent to ι, then two
of the coordinates un, vn, wn must be equal to 1 for all sufficiently large n.
Suppose that each πn lies over the orbit of a character χn of the form
χn = (un, ε2(n), ε3(n)) where un 6= ±1 and ε2(n), ε3(n) ∈ {±1}. Then Gχn
is generated by x, y2, z2 and {e, y} are coset representatives for G/Gχn .
Since πnN ∼ mn(χn(·) ⊕ χn(y · y−1)), it follows that
πn(y2) ∼ mn(cid:18)ε2(n)
0
0
ε2(n)(cid:19) ,
πn(z2) ∼ mn(cid:18)ε3(n)
0
0
ε3(n)(cid:19)
10
MARIUS DADARLAT AND ULRICH PENNIG
and hence if (πn)n converges to ι, then we must have ε2(n) = ε3(n) = 1 for
all sufficiently large n. The cases χn = (ε1, vn, ε3) and χn = (ε1, ε2, wn) are
treated similarly.
In view of the discussion above, it suffices to focus on characters of N
the form χ = (u, 1, 1). The orbit of χ consists of two points, (u, 1, 1) and
(¯u, 1, 1). The corresponding stabilizer Gχ is generated by x, y2 and z2. In
particular Gχ = N ∪ xN and G = Gχ ∪ y Gχ. The exact sequence
1
/ N
/ Gχ
/ Z/2Z
/ 1
does not split since G is torsion free. The quotient G/Gχ is generated by
the coset of y. Let σ ∈ bGχ be an irreducible representation of Gχ whose
restriction to N is a multiple of χ. Since χ(y2) = χ(z2) = 1, it follows that
σ factors through Gχ/Z2. Moreover we have a nontrivial central extension
1
/ Z
/ Gχ/Z2
/ Z/2Z
/ 1
where the normal subgroup is generated by the image of x2 under the map
N → N/hy2, z2i and the quotient group is generated by the image of q(x)
under the map H → H/hq(y)i. Since Gχ/Z2 is an abelian group, σ must be
a character such that σ(x)2 = σ(x2) = χ(x2) = u. Thus σ(x) = a ∈ T with
a2 = u. Let us compute the representation π = ind G
Gχ(σ) of G induced by
σ. It acts on the Hilbert space
Hπ = {ξ : G → C : ξ(gh) = σ(h−1)ξ(g),
g ∈ G, h ∈ Gχ}.
Since G = Gχ ∪ y Gχ, we can identify Hπ with C2 via the isometry ξ 7→
(ξ(e), ξ(y)). Then π(g)ξ = ξ(g−1·) can be described using (2) as follows:
π(x)ξ(e) = ξ(x−1) = σ(x)ξ(e) = aξ(e)
π(x)ξ(y) = ξ(x−1y) = ξ(yxy2z−2) = σ(z2y−2x−1)ξ(y) = ¯aξ(y)
π(y)ξ(e) = ξ(y−1) = ξ(y · y−2) = σ(y2)ξ(y) = ξ(y)
π(y)ξ(y) = ξ(y−1 · y) = ξ(e)
which produces the following matrices with respect to the basis given above:
π(x) =(cid:18)a 0
0 ¯a(cid:19) ,
π(y) =(cid:18)0 1
1 0(cid:19) ,
π(z) =(cid:18)0 a
¯a 0(cid:19)
(3)
(4)
Corresponding to the characters (1, v, 1) and (1, 1, w) we obtain the irre-
ducible representations, where we use the isometries ξ 7→ (ξ(e), ξ(x)) and
ξ 7→ (ξ(e), ξ(y)) respectively:
π(x) =(cid:18)0 1
1 0(cid:19) ,
π(y) =(cid:18)b 0
0 ¯b(cid:19) ,
π(z) =(cid:18)0 ¯b
b 0(cid:19) ,
b2 = v,
/
/
/
/
/
/
/
/
CONNECTIVE C ∗-ALGEBRAS
11
and
(5)
π(x) =(cid:18)0 ¯c
c 0(cid:19) ,
1 0(cid:19) ,
π(y) =(cid:18)0 1
π(z) =(cid:18)c 0
0 ¯c(cid:19) ,
c2 = w.
Let (πn)n be a sequence in R2 that converges to ι in bG. Arguing by
symmetry, we may assume that each πn is given by the formulas (3) cor-
responding to a sequence of points un ∈ T with un /∈ {±1}. Since πn → ι
it follows from the equation (3) that un → 1. Again from (3) we can com-
pute the limits of the sequences πn(x) and πn(y) in U (2). This gives the
representation π : G → U (2):
π(x) =(cid:18)1 0
0 1(cid:19) ,
π(y) =(cid:18)0 1
1 0(cid:19)
It is clear that π is a representation of G that factors through the left
regular representation of Z/2. Decompose π into a direct sum of characters
π ∼ ι ⊕ η. Then η is not equivalent to ι and πn → η in bG.
4-orbits. Let χ = (u, v, w) ∈ T3 be a character of N with u, v, w /∈ {±1}.
Its orbit under the action of G consists of four points and Gχ = N . Let
us compute the representation π = ind G
It acts
on the Hilbert space Hπ = {ξ : G → C : ξ(gh) = χ(h−1)ξ(g),
g ∈
G, h ∈ N }. Thus one can identify Hπ with C4 via the isometry ξ 7→
(ξ(e), ξ(x), ξ(y), ξ(z)). Using the identities (2), we verify that π(g)ξ =
ξ(g−1·) is described as follows:
N (χ) of G induced by σ.
π(x)ξ(e) = ξ(x−1) = ξ(x · x−2) = χ(x2)ξ(x) = uξ(x)
π(x)ξ(x) = ξ(e)
π(x)ξ(y) = ξ(x−1y) = ξ(zx2z−2) = χ(z2x−2)ξ(z) = w¯uξ(z)
π(x)ξ(z) = ξ(x−1z) = ξ(yz2) = χ(z−2)ξ(y) = ¯wξ(y)
π(y)ξ(e) = ξ(y−1) = ξ(y · y−2) = χ(y2)ξ(y) = vξ(y)
π(y)ξ(x) = ξ(y−1x) = ξ(zx2) = χ(x−2)ξ(z) = ¯uξ(z)
π(y)ξ(y) = ξ(e)
π(y)ξ(z) = ξ(y−1z) = ξ(x · x−2y2) = χ(y−2x2)ξ(x) = ¯vuξ(x)
producing the matrices:
(6)
0 0 ¯w 0 !,
π(x) = 0 u 0 0
1 0 0 0
0 0 0 ¯uw
0 u¯v 0 0!
π(y) = 0 0 v 0
0 0 0 ¯u
1 0 0 0
12
MARIUS DADARLAT AND ULRICH PENNIG
It follows that
(7)
0 u 0 0
0 0 ¯u 0
0 0 0 ¯v!,
π(y2) = v 0 0 0
0 ¯v 0 0
0 0 v 0
0 0 0 ¯u!,
π(x2) = u 0 0 0
0 0 0 w!
π(z2) = w 0 0 0
Let (πn)n be a sequence in R4 converging to ι in bG. Each πn is given
by the formulas (6) corresponding to a sequence of points (uk, vk, wk) ∈ T3
with uk, vk, wk /∈ {±1}. Since πn → ι it follows from equation (7) that
uk, vk, wk → 1. Again from (6) we can compute the limits of the sequences
πn(x) and πn(y) in the space of unitary operators U (4). This gives the
representation π : G → U (4):
0 ¯w 0 0
0 0 ¯w 0
0 0 1 0!,
π(x) = 0 1 0 0
1 0 0 0
0 0 0 1
0 1 0 0!
π(y) = 0 0 1 0
0 0 0 1
1 0 0 0
It is clear that π is a representation of G that factors through H = Z/2×Z/2.
Decompose π into a direct sum of characters ηi. Since π is not equivalent to
a multiple of the trivial representation, it follows that at least one of these
characters is not equivalent to ι. On the other hand πn → ηi in bG for all i.
Combining the above analysis with Corollary 2.11 we obtain immediately
Corollary 3.2. If G is the Hantzsche-Wendt group, then I(G) is not con-
nective.
It was conjectured in [8] that if G is a torsion free discrete amenable
group, then [[I(G), K]] ∼= KK(I(G), C). We argue now that this conjecture
fails for the Hantzsche-Wendt group. Indeed this follows from the previous
corollary in conjunction with the following lemma.
Lemma 3.3. Let G be a residually finite torsion free discrete amenable
group which admits a classifying space with finitely generated K-homology
group K1(BG). Then [[I(G), K]] ∼= KK(I(G), C) if and only if I(G) is
connective.
Proof. Suppose first that [[I(G), K]] ∼= KK(I(G), C). Since G is amenable
and residually finite it follows that C ∗(G) is residually finite dimensional.
Since G is amenable, G satisfies the Baum-Connes conjecture and C ∗(G)
satisfies the UCT by results of Higson and Kasparov [22] and Tu [35]. In
particular we have a short exact sequence
0 → Ext1(K1(C ∗(G)), Z) → KK(C ∗(G), C) → Hom(K0((C ∗(G)), Z) → 0
Let πn : C ∗(G) → Md(n)(C) be a separating sequence of finite dimen-
sional representations. The restriction of πn to I(G) will be denoted by σn.
By [8, Prop. 3.2] (πn)∗ = d(n)ι∗ : K0(C ∗(G)) → Z and hence [σn] ∈
Ext1(K1(I(G)), Z) ⊂ KK(I(G), C) is a torsion element since K1(I(G)) ∼=
CONNECTIVE C ∗-ALGEBRAS
13
K1(BG) is finitely generated. After replacing πn by a suitable multiple of
itself we have arranged that [σn] = 0 in KK(I(G), C) and hence [[σn]] = 0 in
[[I(G), K]]. Since the sequence (σn) separates the elements of I(G) it follows
that I(G) is connective.
The converse is contained in the main result of [10] which shows that if A
is a separable nuclear connective C ∗-algebra, then [[A, K]] ∼= KK(A, C). (cid:3)
3.2. Crystallographic groups with cyclic holonomy. In this section
we are going to show that torsion free crystallographic groups with cyclic
holonomy are connective. Apart from this we isolate a lemma, which proves
that I(G) for a group G which is a finite extension of a connective group
always contains a "big" connective ideal. In particular, the lemma also holds
for the Hantzsche-Wendt group.
The proof of both results uses some tools from the index theory of C ∗-
subalgebras. A reference is [39]. Let Γ and G be discrete groups and let H
be a finite group. Suppose that they fit into an exact sequence
1
/ Γ
/ G
q
/ H
/ 1 .
Let E : C ∗(G) → C ∗(Γ) be the faithful conditional expectation [39, Ex. 1.2.3]
given on group elements by
E(g) =(g
0
if g ∈ Γ
else
.
Choose a lift gh ∈ G for each h ∈ H. The pairs (g−1
h , gh) form a quasi-basis
in the sense of [39, Def. 1.2.2]. Let E = C ∗(G) considered as a right Hilbert
C ∗(Γ)-module, where the right action is induced by the inclusion C ∗(Γ) →
C ∗(G) and the inner product is given by ha, bi = E(a∗b) [39, Sec. 2.1]. Note
that E is complete [39, Prop. 2.1.5]. The quasi-basis induces an isometric
isomorphism of right Hilbert C ∗(Γ)-modules u : E → ℓ2(H) ⊗ C ∗(Γ) with
u(a) =Xh
δh ⊗ E(gha)
and inverse u∗ : ℓ2(H) ⊗ C ∗(Γ) → E with u∗(δh ⊗ b) = g−1
h b. Let LC ∗(Γ)(E)
be the bounded adjointable operators on E and denote by KC ∗(Γ)(E) the
compact ones. Then we have LC ∗(Γ)(E) ∼= KC ∗(Γ)(E) ∼= K(ℓ2(H)) ⊗ C ∗(Γ).
The left multiplication of C ∗(G) on E induces a ∗-homomorphism
ψ : C ∗(G) → K(ℓ2(H)) ⊗ C ∗(Γ)
/
/
/
/
14
MARIUS DADARLAT AND ULRICH PENNIG
with matrix entries ψh′,h(a) = E(gh′ a g−1
with ψ(a) = 0. Then
h ). Suppose we have a ∈ C ∗(G)
a =
1
HXh,h′
h′ E(gh′ a g−1
g−1
h )gh = 0 .
Hence, ψ is injective.
Lemma 3.4. Let Γ be a connective group and let H be a finite group. Sup-
pose that the group G fits into a short exact sequence of the form
1
/ Γ
/ G
q
/ H
/ 1 .
Then I(G, H) = ker(I(q) : I(G) → I(H)) is connective as well.
Proof. Let ι : C ∗(Γ) → C be the trivial representation and let ψ be the
injective ∗-homomorphism constructed above. For all b ∈ C ∗(Γ) ⊂ C ∗(G) we
have ψh′,h(b) = δh′,h gh b g−1
h . In particular, ψ embeds the ideal J generated
by I(Γ) into ker(id ⊗ ι) = K(ℓ2(H)) ⊗ I(Γ), which is connective. Hence J is
connective as well. It is clear that J ⊆ I(G, H). Let x ∈ I(G, H). By the
property of the quasi-basis
q(E(x g−1
h )) h ⇒ q(E(x g−1
h )) = 0 ∀h ∈ H .
0 = q(x) = Xh∈H
Since I(G, H) ∩ C ∗(Γ) = I(Γ) we obtain that E(x g−1
h )) gh ∈ J, hence J = I(G, H).
h ) ∈ I(Γ) for all h ∈ H
(cid:3)
and therefore x =Ph E(x g−1
The proof of the second result uses an induction over the rank of the free
abelian subgroup based on the following observation.
Lemma 3.5. Let m > 1 and let Γ and G be countable discrete groups that
fit into the following short exact sequence
1
/ Γ
π
/ G
/ Z/mZ
/ 1 .
Suppose that Γ is connective and there are group homomorphisms ϕ : G → Z
and q : Z → Z/mZ, such that π = q ◦ ϕ. Then G is connective as well.
Proof. Let ψ : C ∗(G) → K(ℓ2(H))⊗C ∗(Γ) be the injective ∗-homomorphism
constructed above and let ι : C ∗(Γ) → C be the trivial representation. Ob-
serve that ρ = (id ⊗ ι) ◦ ψ satisfies
ρ(g γ)h′,h = ι(E(gh′ g γ g−1
h )) = ι(E(gh′ g g−1
h ) gh γ g−1
h ) = ι(E(gh′ g g−1
h ))
for all g ∈ G and γ ∈ Γ. In particular, ρ factors through the ∗-homomorphism
C ∗(G) → C ∗(Z/mZ) induced by π. By assumption this decomposes as
C ∗(G)
ϕ
/ C ∗(Z)
q
/ C ∗(Z/mZ) .
/
/
/
/
/
/
/
/
/
/
CONNECTIVE C ∗-ALGEBRAS
15
Altogether we obtain that ρ decomposes into a direct sum of one-dimensional
representations, each of which is homotopic through representations to the
trivial one. Hence, to show that G is connective, it suffices to construct a
path through discrete asymptotic morphisms connecting a faithful morphism
with a direct sum of copies of ρ.
Choose a path witnessing the connectivity of Γ, i.e. a discrete asymptotic
morphism
Hn : C ∗(Γ) → C([0, 1]) ⊗ Mn(C)
such that for H (t)
ful and H (1)
n
properties. Hence, G is connective.
n = evt ◦ Hn : C ∗(Γ) → Mn(C) we have that H (0)
n
is faith-
is a multiple of ι. Then (idMm(C) ⊗ Hn) ◦ ψ has the desired
(cid:3)
We need the following elementary fact:
Lemma 3.6. Let a, b > 1 be integers and consider the exact sequence
0
/ Z/aZ ·b
/ Z/abZ π
/ Z/bZ
/ 0
with π(x) = x mod b. Any generator of Z/bZ lifts to a generator of Z/abZ.
Proof. Let ¯y ∈ Z/bZ be a generator and let y ∈ {0, . . . , b−1} be a representa-
tive. Let p1, . . . , ps be the disctinct prime factors of ab, such that p1, . . . , pr
for r ≤ s are the ones not dividing y and pr+1, . . . , ps divide y. Since
gcd(y, b) = 1, the primes pr+1, . . . , ps do not divide b. Let x = y + p1 . . . pr b.
We have for i ∈ {1, . . . , r} and j ∈ {r + 1, . . . , s}
x ≡ y 6≡ 0 mod pi ,
x ≡ p1 . . . pr b 6≡ 0 mod pj .
Hence gcd(x, ab) = 1 and x ∈ Z/abZ is a generator with π(x) = ¯y.
(cid:3)
To start the induction we need the following lemma.
Lemma 3.7. Let G be a countable torsion free discrete group, which fits
into an exact sequence
0
/ Z
/ G
/ Z/mZ
/ 0
Then G is isomorphic to Z, hence in particular connective.
Proof. This can be proven by calculating H 2(Z/mZ, Z) for all Z/mZ-module
structures on Z, but we give a direct argument here.
Let x ∈ G be a lift of 1 ∈ Z/mZ. Then G is generated by x and Z.
Moreover, xm ∈ Z ∩ Z(G), where Z(G) denotes the center of G. We have
Aut(Z) ∼= GL1(Z) ∼= Z/2Z. If t ∈ Z denotes a generator, we therefore can
only have xtx−1 = t−1 or xt = tx. Suppose the first is true, then
xm = x xm x−1 = x−m ⇒ x2m = e
/
/
/
/
/
/
/
/
16
MARIUS DADARLAT AND ULRICH PENNIG
contradicting that G is torsion free. Thus, t and x commute and xm = tn
for some n ∈ Z. Without loss of generality we can assume gcd(m, n) = 1.
Indeed, if m = m′ ℓ and n = n′ ℓ with ℓ > 1, then (xm′
)ℓ = e and
therefore xm′
also holds in G. Consider
= tn′
t−n′
α : G → Z ;
xktℓ 7→ k n + ℓ m .
This is a well-defined group homomorphism, which is easily seen to be bi-
jective as a consequence of gcd(m, n) = 1.
(cid:3)
Theorem 3.8. Let G be a countable, torsion free, discrete group, which fits
into an exact sequence of the form
0
/ Zn
π
/ G
/ Z/mZ
/ 0
for some n, m ∈ N. Then G is connective.
Proof. This will be proven by induction over the rank of the free abelian
subgroup. The case n = 1 follows from Lemma 3.7.
Observe that Z(G) 6= {e}. Indeed, let x ∈ G be a lift of the generator
of Z/mZ. Then G is generated by Zn and x. Moreover, xm 6= e since G is
torsion free and π(xm) is trivial, hence xm ∈ Zn. Thus, xm commutes with
Zn and x, hence with all elements of G, i.e. xm ∈ Z(G).
This implies that the transfer homomorphism T : G → Zn associated to
the finite index subgroup Zn is non-trivial. Therefore there exists a surjective
group homomorphism ϕ : G → Z. Let q : Z → Z/mZ be the canonical
quotient homomorphism and let ¯ϕ = q ◦ ϕ. Let H = ker(ϕ). We have the
following commutative diagram with exact rows and columns:
0
H
0
Z/aZ
0
0
/ Zn
π
/ G
Z/mZ
ϕ
/ bZ
/ Z π′
Z/bZ
0
0
0
0
0
The value of a is chosen in such a way that Im( πH) ∼= Z/aZ and b satisfies
m = ab. The homomorphism π′ is surjective since π and Z/mZ → Z/bZ
are. The vertical arrow on the left hand side is induced by ϕZn.
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
CONNECTIVE C ∗-ALGEBRAS
17
Suppose H ⊂ ker(π) = Zn. Then a = 1, b = m and π = π′ ◦ ϕ. By
Lemma 3.5, G is then connective. So we may assume a > 1.
We claim that there is an element g ∈ G such that ϕ(g) = 1 and π(g)
is a generator of Z/mZ. This is constructed as follows: If b = 1, we can
choose g ∈ G, such that ϕ(g) = 1 and modify it by an element in H to
achieve that π(g) becomes a generator. Otherwise, choose g′ ∈ G such that
ϕ(g′) = 1 and note that π′(ϕ(g′)) is a generator of Z/bZ by surjectivity.
We can lift π′(ϕ(g′)) to a generator x ∈ Z/mZ by Lemma 3.6. Note that
π(g′)−x ∈ Z/aZ and lift this difference to an element h ∈ H. Let g = g′ h−1.
Then ϕ(g) = ϕ(g′) = 1 and π(g) = π(g′) − π(h) = π(g′) − π(g′) + x = x.
Let G′ = ker( ¯ϕ) = {g ∈ G ϕ(g) = ℓ · m for ℓ ∈ Z} ⊃ H. Hence, the
following diagram has exact rows.
(8)
1
1
/ G′
ϕG′
/ G
ϕ
/ mZ
/ Z
¯ϕ
q
Z/mZ
0
=
/ Z/mZ
/ 0
The group G is generated by Zn and the element g constructed above. We
have gm ∈ Zn ∩ Z(G) and ϕ(gm) = m. In particular, gm ∈ Z(G′). Let
ψ : H × Z → G′
;
(h, k) 7→ h · gmk .
This is a group homomorphism, since gm is central and it fits into the
commutative diagram with exact upper and lower part
1
/ H
Z
/ 0
G′
ϕG′ /m
;✇✇✇✇✇✇✇✇✇
#●●●●●●●●●
i
ψ
#❋❋❋❋❋❋❋❋❋
<①①①①①①①①①
pr
H × Z
proving that ψ is in fact an isomorphism. By the upper row in diagram
(8) and Lemma 3.5, the connectivity of G follows if H × Z, hence H, is
connective [10, Thm. 4.1]. But H fits into a short exact sequence of the
form
0 → A → H → Z/aZ → 0
where A is the free abelian kernel of the nonzero homomorphism Zn → bZ
from above, which has rank (n − 1). This completes the induction step. (cid:3)
/
/
/
/
/
/
/
/
/
/
#
/
#
;
/
<
O
O
18
MARIUS DADARLAT AND ULRICH PENNIG
4. Connectivity of Lie group C ∗-algebras
In this section we determine which linear connected nilpotent Lie groups
and which linear connected reductive Lie groups have connective reduced
C ∗-algebras. Let us recall that nilpotent connected Lie groups are liminary
as shown by Dixmier [13] and Kirillov [27] and semisimple connected Lie
groups are liminary as shown by Harish-Chandra [18].
4.1. Solvable and nilpotent Lie groups. A locally compact group N is
compactly generated if N =Sn V n for some compact subset V of N . Every
connected locally compact group is automatically compactly generated. The
structure of abelian compactly generated locally compact groups is known.
If N is such a group, then N ∼= Rn × Zm × K for integers n, m ≥ 0 and K
a compact group, [12, Thm. 4.4.2].
Proposition 4.1. If G is a second countable locally compact amenable group
(for example a solvable Lie group) whose center contains a noncompact
closed connected subgroup, then C ∗(G) is connective.
Proof. Let N be a noncompact closed connected subgroup of Z(G). Then,
by the structure theorem quoted above, N must have a closed subgroup
isomorphic to R. Consider the central extension:
0 → R → G → H → 0.
Since G is amenable, by [29, Thm. 1.2] (as explained in [15, Lemma 6.3]),
C ∗(G) has the structure of a continuous field of C ∗-algebras over bR ∼= R.
The desired conclusion follows from Cor. 2.6 since R has no compact open
subsets.
(cid:3)
Example 4.2. We give here two examples that complement Proposition 4.1.
(i) Simply connected solvable Lie groups can have discrete noncompact
centers. This is the case for G = C ⋊α R where α : R → Aut(C)
is defined by α(t)(z) = eitz for t ∈ R and z ∈ C.
In this case
Z(G) = {0} × 2πZ.
Nevertheless in this case C ∗(G) is connective. Consider the ex-
tension
0 → Z(G) → G → G/Z(G) ∼= C × T → 0.
Then C ∗(G) is a continuous C(T)-algebra whose fiber at 1 is the
algebra C ∗(C × T). Since C ∗(C × T) ∼= C0(R2) ⊗ c0(Z) is connective,
so is C ∗(G).
(ii) Both the real and the complex "ax + b" groups
G =(cid:26)(cid:18)a b
0 1(cid:19) : a ∈ F ×, b ∈ F(cid:27)
CONNECTIVE C ∗-ALGEBRAS
19
where F = R or F = C are solvable with trivial center and their
C ∗-algebras contain a copy of the compacts K, see [33], and so they
are not connective.
Theorem 4.3. Let G be a (real or complex) linear connected nilpotent Lie
group. Then C ∗(G) is connective if and only if G is not compact.
Proof. We view G as a real Lie group. By [37, Chap. 2, Thm. 7.3], if G
is a linear connected nilpotent Lie group, then G decomposes as a direct
product G = T × N of a torus T and a simply connected nilpotent group
N . If G is compact, then G = T and C ∗(G) is isomorphic to a direct sum of
C so that it is not connective. If G is noncompact, then N is nontrivial and
so the center of G is given by Z(G) = T × Z(N ), where the center Z(N ) of
N is isomorphic to Rn for some n ≥ 1. We conclude the proof by applying
Proposition 4.1.
(cid:3)
Remark 4.4. It is not true that a liminary (CCR) C ∗-algebra is connective
if and only if does not have nonzero projections. Indeed
A =(cid:26)f ∈ C([0, 1], M2(C)) : f (0) =(cid:18)λ 0
0 λ(cid:19) , f (1) =(cid:18)λ 0
0 0(cid:19) , λ ∈ C(cid:27)
does not contain nonzero projections but is not connective since Prim(A) is
homeomorphic to a circle S1 and hence it is compact (and open in itself).
4.2. Reductive Lie groups. A linear connected reductive group G is a
closed group of real or complex matrices that is closed under conjugate
transpose.
In other words G is a closed and selfadjoint subgroup of the
general linear group over either R or C. A linear connected semisimple
group is a linear connected reductive group with finite center [28].
Say G ⊂ GL(n, R) or GL(n, C). Define K = G ∩ O(n), or K = G ∩ U (n)
in the complex case. If G is linear connected reductive, then K is compact,
connected and is a maximal compact subgroup of G [28, Prop.1.2].
Let G = KAN be the Iwasawa decomposition of the linear connected
semisimple Lie group G. A is abelian and N is nilpotent and both are
closed simply connected subgroups of G, [28, Thm. 5.12].
First we consider the case of complex Lie groups.
Theorem 4.5. If G is a linear connected complex semisimple Lie group,
then C ∗
r (G) is connective if and only if G is not compact.
Proof. If G is compact, C ∗
bras and hence it is not connective as it contains nonzero projections.
r (G) is isomorphic to a direct sum of matrix alge-
Conversely, suppose now that G is a non-compact linear connected semi-
simple complex Lie group. Note that from the Cartan decomposition G =
20
MARIUS DADARLAT AND ULRICH PENNIG
KAK [28, Thm. 5.20] it follows that since G is non-compact, so is A and
therefore A ∼= Rn, for some n ≥ 1.
Let M be the centralizer of A in K. By Lemma 3.3 and Proposition 4.1
of [31], it follows that
C ∗
r (G) ⊂ C0(cM × bA, K).
Since cM × bA ∼= cM × Rn does not have nonempty compact open subsets,
it follows from Proposition 2.7(ii) that C0(cM × bA, K) is connective. This
completes the proof since connectivity passes to C ∗-subalgebras.
(cid:3)
Next we consider the case of linear connected real reductive Lie groups.
An element g ∈ G is semisimple if it can be diagonalized over C when viewed
as a matrix g ∈ Mn(C).
A closed subgroup H of G is called a Cartan subgroup if it is a maximal
abelian subgroup consisting of semisimple elements, [21, p.67]. If G is either
compact or a complex Lie group, then all Cartan subgroups of G are con-
nected and they are conjugated inside G. In the general case G has finitely
many Cartan subgroups up to conjugacy and Cartan subgroups can have
finitely many connected components.
We denote by bGd ⊂ bG the discrete series representations. It consists of
unitary equivalence classes of square-integrable representations
σ : G → U (Hσ).
Harish-Chandra has shown that the discrete series representations of a semi-
simple Lie group G are parametrized by compact Cartan subgroups and in
particular G has discrete series representations if and only if it has a compact
Cartan subgroup, [19, 20].
We recall the following facts from [21, p.72] concerning cuspidal parabolic
subgroups. Let H be a Cartan subgroup of G. Then H decomposes as a
direct product H = T A = T × A, where T is an abelian compact group and
A is a vector group isomorphic to Rn for n ≥ 0. The case n = 0 occurs when
H is a compact Cartan subgroup. The centralizer of A in G denoted by
L = CG(A) = {g ∈ G : ga = ag, ∀a ∈ A}
is a Levi subgroup of G. This means that there is a parabolic subgroup of
G of the form P = LN (not unique) with L as Levi subgroup. Since A is
central in L, H is a relatively compact Cartan subgroup of L, i.e. H/Z(L)
is compact. This implies that L has discrete series representations. Such a
parabolic subgroup P = LN is called cuspidal.
One can further decompose L = M A to obtain a Langlands decomposition
P = M AN = M A ⋉ N,
CONNECTIVE C ∗-ALGEBRAS
21
with N a unipotent group. If H is a compact Cartan subgroup, then L =
P = G by [21, p.72].
We will write P = MP AP NP whenever we want emphasize the compo-
nents of P .
The description of C ∗
r (G) relies on the analysis of the unitary principal
series representations of G associated to parabolic cuspidal subgroups P
(also called the P -principal series). They are of the form
IndG
P (σ ⊗ ω ⊗ 1N ),
where σ ∈ cMd and ω ∈ bA and 1N is the trivial representation of N .
Consider two pairs (Pi, σi), i = 1, 2 consisting of cuspidal parabolic sub-
groups of G and irreducible square-integrable unitary representations of the
subgroups Mi, i = 1.2. We say that the pairs are associated if there is g ∈ G
such that gP1g−1 = P2 and σ1(g · g−1) is unitarily equivalent to σ2. This
is an equivalence relation [5, Def. 5.2]. We denote by [P, σ] the equivalence
class of the pair (P, σ).
The following statement is based on the calculation of C ∗
r (G) by A. Wasser-
mann [38] although we don't really use the full strength of his results. An
expanded treatment of the structure of C ∗
r (G) appears in [5].
ker(π) ⊂ C ∗
Let J(G) =Tπ∈ bGd
r (G) be the common kernel of the discrete
series representations. The following theorem shows that the K-homology of
C ∗
r (G) can be described in terms of homotopy classes asymptotic morphisms
C ∗
r (G) → K which factor through J(G) and discrete series representations.
Theorem 4.6. Let G be a linear connected real reductive Lie group. Then
K(Hσ) and J(G) is a connective liminary C ∗-
C ∗
algebra. Moreover, the following assertions are equivalent:
r (G) ∼= J(G) ⊕Lσ∈ bGd
r (G) is connective,
(i) C ∗
(ii) G does not have discrete series representations,
(iii) G does not have a compact Cartan subgroup,
(iv) There are no nonzero projections in C ∗
r (G).
Proof. As explained in [38, p.560], [5, p.1306] the reduced C ∗-algebra of a
linear reductive connected Lie group admits an embedding
C ∗
r (G) ֒→M[P,σ]
C0(bAP , K(Hσ)) ⊕ Mσ∈ bGd
K(Hσ),
C ∗
r (G) ֒→M[P,σ]
C0(bAP , K(Hσ)),
where the direct sum is over equivalence classes [P, σ] as above. It is impor-
tant to emphasize that if G has a compact Cartan subgroup, then G itself
is one of the cuspidal parabolic subgroups and we have:
where
C ∗
r (G) ∼= J(G) ⊕ Mσ∈ bGd
J(G) ֒→M[P,σ]
C0(bAP , K(Hσ)) .
22
MARIUS DADARLAT AND ULRICH PENNIG
where the first direct sum involves proper cuspidal parabolic subgroups P =
MP AP NP and hence dim(bAP ) > 0. Moreover by [38], [5]:
K(Hσ)
(9)
Hence, J(G) is connective being a subalgebra of a connective C ∗-algebra.
The first part of the statement follows now from the decomposition (9).
The equivalence (ii) ⇔ (iii) is Harish-Chandra's result mentioned earlier.
In view of the decomposition (9), (ii) implies that C ∗
r (G) = J(G) and hence
(i) since J(G) is always connective. Connective C ∗-algebras do not contain
nonzero projections and hence (i) ⇒ (iv). Finally by using (9) again, we see
that (iv) ⇒ (ii) since K(Hσ) contains nonzero projections if Hσ 6= 0.
(cid:3)
4.3. A remark on full C ∗-algebras of Lie groups. The full C ∗-algebra
C ∗(G) of a property (T) Lie group G contains nonzero projections and hence
it is not connective, see [36]. Nevertheless, inspection of several classes of
examples indicates that C ∗(G) has interesting connective ideals that arise
naturally from the representation theory of G. We postpone a detailed
discussion of what is known for another time, but would like to mention two
examples.
If G is a connected semisimple Lie group with finite center, then C ∗(G)
is liminary (or CCR), see [40, p.115].
Proposition 4.7. (a) C ∗(SL2(C)) is connective.
(b) C ∗(SL3(C)) = I(SL3(C)) ⊕ C and I(SL3(C)) is connective.
now his result. Let Z be the subspace of R2 defined by Z =S∞
Proof. (a) C ∗(SL2(C)) was computed by Fell [16, Thm. 5.4]. We describe
n=0{n} × Ln
where L0 = (−1, ∞) and Ln = (−∞, ∞) for all n ≥ 1. Endow Z with
the induced topology from R2. Let H0 be a separable infinite dimensional
Hilbert space, let H = H0 ⊕ C and fix a unitary operator V : H0 → H.
Then C ∗(SL2(C)) is isomorphic to
{F ∈ C0(Z, K(H)) : F (0, −1) = V ∗F (2, 0)V ⊕ λ, for some λ ∈ C}
Since Z has no nonempty open compact subsets it follows that C0(Z, K(H))
is connective and therefore so is its subalgebra C ∗(SL2(C)).
(b) This will be obtained as a consequence of the following result on the
structure of C ∗(SL3(C)) obtained by Pierrot [32]. Let G = SL3(C) and
denote by λG : C ∗(G) → C ∗
r (G) the morphism induced by the left regular
representation and by ιG : C ∗(G) → C the trivial representation. Pierrot
CONNECTIVE C ∗-ALGEBRAS
23
proved that the kernel J of the morphism λG ⊕ ιG : C ∗(G) → C ∗
r (G) ⊕ C
is a contractible C ∗-algebra. The representation ιG is isolated since G has
property (T). Therefore there is an exact sequence
0 → J → I(G) → C ∗
r (G) → 0
where J is contractible and C ∗
clude that I(G) is connective by applying Theorem 2.4.
r (G) is connective by Theorem 4.5. We con-
(cid:3)
Acknowledgements We would like to thank Andrzej Szczepa´nski for
pointing out a glitch in the formulation of Definition 2.9.
References
[1] B. Blackadar and J. Cuntz. The stucture of stable algebraically simple
C ∗-algebras. American J. Math., 104:813 -- 822, 1982.
[2] ´E. Blanchard. Subtriviality of continuous fields of nuclear C ∗-algebras.
J. Reine Angew. Math., 489:133 -- 149, 1997.
[3] ´E. Blanchard and E. Kirchberg. Non-simple purely infinite C ∗-algebras:
the Hausdorff case. J. Funct. Anal., 207:461 -- 513, 2004.
[4] N. P. Brown and M. Dadarlat. Extensions of quasidiagonal C ∗-algebras
and K-theory. In Operator algebras and applications, volume 38 of Adv.
Stud. Pure Math., pages 65 -- 84. Math. Soc. Japan, Tokyo, 2004.
[5] P. Clare, T. Crisp, and N. Higson. Parabolic induction and restriction
via C ∗-algebras and Hilbert C ∗-modules. Compos. Math., 152(6):1286 --
1318, 2016. ISSN 0010-437X. doi: 10.1112/S0010437X15007824. URL
http://dx.doi.org/10.1112/S0010437X15007824.
[6] A. Connes and N. Higson. D´eformations, morphismes asymptotiques
et K-th´eorie bivariante. C. R. Acad. Sci. Paris S´er. I Math., 311(2):
101 -- 106, 1990.
[7] J. H. Conway and J. P. Rossetti. Describing the platycosms. Preprint
2003, arXiv:math/0311476v1.
[8] M. Dadarlat. Group quasi-representations and almost flat bundles. J.
Noncommut. Geom., 8(1):163 -- 178, 2014.
[9] M. Dadarlat and T. A. Loring. K-homology, asymptotic representa-
tions, and unsuspended E-theory. J. Funct. Anal., 126(2):367 -- 383,
1994.
[10] M. Dadarlat and U. Pennig.
and homotopy symmetric C ∗-algebras.
ISSN 1432-1807.
http://dx.doi.org/10.1007/s00208-016-1379-0.
doi:
Deformations of nilpotent groups
Mathem. Ann., 2016.
URL
10.1007/s00208-016-1379-0.
24
MARIUS DADARLAT AND ULRICH PENNIG
[11] M. Dadarlat, U. Pennig, and A. Schneider. Deformations of wreath
to appear in Bull. London. Math. Soc., Preprint (2016),
products.
arXiv:1609.00604, 2016.
[12] A. Deitmar and S. Echterhoff. Principles of harmonic analysis. Uni-
ISBN 978-3-319-
versitext. Springer, Cham, second edition, 2014.
05791-0; 978-3-319-05792-7. doi: 10.1007/978-3-319-05792-7. URL
http://dx.doi.org/10.1007/978-3-319-05792-7.
[13] J. Dixmier. Sur les repr´esentations unitaires des groupes de Lie nilpo-
tents. V. Bull. Soc. Math. France, 87:65 -- 79, 1959. ISSN 0037-9484.
[14] J. Dixmier. C ∗-algebras. North Holland, Amsterdam, 1982.
[15] S. Echterhoff and D. P. Williams. Crossed products by C0(X)-actions.
J. Funct. Anal., 158(1):113 -- 151, 1998. ISSN 0022-1236. doi: 10.1006/
jfan.1998.3295. URL http://dx.doi.org/10.1006/jfan.1998.3295.
[16] J. M. G. Fell. The structure of algebras of operator fields. Acta Math.,
106:233 -- 280, 1961.
[17] W. Hantzsche and H. Wendt. Dreidimensionale euklidische Raumfor-
men. Math. Ann., 110(1):593 -- 611, 1935. ISSN 0025-5831. doi: 10.1007/
BF01448045. URL http://dx.doi.org/10.1007/BF01448045.
[18] Harish-Chandra. Representations of semisimple Lie groups. III. Trans.
Amer. Math. Soc., 76:234 -- 253, 1954. ISSN 0002-9947.
[19] Harish-Chandra. Discrete series for semisimple Lie groups. I. Construc-
tion of invariant eigendistributions. Acta Math., 113:241 -- 318, 1965.
ISSN 0001-5962.
[20] Harish-Chandra. Discrete series for semisimple Lie groups. II. Explicit
determination of the characters. Acta Math., 116:1 -- 111, 1966. ISSN
0001-5962.
[21] R. A. Herb. An elementary introduction to Harish-Chandra's work.
In The mathematical legacy of Harish-Chandra (Baltimore, MD, 1998),
volume 68 of Proc. Sympos. Pure Math., pages 59 -- 75. Amer. Math.
Soc., Providence, RI, 2000. doi: 10.1090/pspum/068/1767892. URL
http://dx.doi.org/10.1090/pspum/068/1767892.
[22] Nigel Higson and Gennadi G. Kasparov, E-theory and KK-theory for
groups which act properly and isometrically on Hilbert space, Invent.
Math. 144 (2001), no. 1, 23 -- 74. MR 1821144
[23] E. Kaniuth and K. F. Taylor. Induced representations of locally compact
groups, volume 197 of Cambridge Tracts in Mathematics. Cambridge
University Press, Cambridge, 2013. ISBN 978-0-521-76226-7.
[24] G. G. Kasparov. The operator K-functor and extensions of C ∗-algebras.
Izv. Akad. Nauk SSSR Ser. Mat., 44(3):571 -- 636, 719, 1980. ISSN 0373-
2436.
CONNECTIVE C ∗-ALGEBRAS
25
[25] J. L. Kelley. General topology. D. Van Nostrand Company, Inc.,
Toronto-New York-London, 1955.
[26] E. Kirchberg. Das nicht-kommutative Michael-Auswahlprinzip und die
Klassifikation nicht-einfacher Algebren. In C ∗-algebras, pages 92 -- 141,
Berlin, 2000. Springer. (Munster, 1999).
[27] A. A. Kirillov. Unitary representations of nilpotent Lie groups. Uspehi
Mat. Nauk, 17(4 (106)):57 -- 110, 1962. ISSN 0042-1316.
[28] A. W. Knapp. Representation theory of semisimple groups, volume 36 of
Princeton Mathematical Series. Princeton University Press, Princeton,
NJ, 1986. ISBN 0-691-08401-7. An overview based on examples.
[29] J. A. Packer and I. Raeburn.
On the structure of
group C ∗-algebras.
718, 1992.
doi:
http://dx.doi.org/10.2307/2154478.
ISSN 0002-9947.
Trans. Amer. Math. Soc.,
10.2307/2154478.
twisted
334(2):685 --
URL
[30] C. Pasnicu and M. Rørdam.
real
rank zero.
ISSN 0075-4102.
http://dx.doi.org/10.1515/CRELLE.2007.091.
J. Reine Angew. Math.,
doi:
Purely infinite C ∗-algebras of
2007.
URL
613:51 -- 73,
10.1515/CRELLE.2007.091.
[31] M. G. Penington and R. J. Plymen. The Dirac operator and the prin-
cipal series for complex semisimple Lie groups. J. Funct. Anal., 53(3):
269 -- 286, 1983. ISSN 0022-1236. doi: 10.1016/0022-1236(83)90035-6.
URL http://dx.doi.org/10.1016/0022-1236(83)90035-6.
[32] F. Pierrot. La K-th´eorie de la C ∗-alg`ebre pleine de SL3(C). J. Funct.
Anal., 193(2):261 -- 277, 2002. ISSN 0022-1236. doi: 10.1006/jfan.2001.
3907. URL http://dx.doi.org/10.1006/jfan.2001.3907.
[33] J. Rosenberg. The C ∗-algebras of some real and p-adic solvable groups.
Pacific J. Math., 65(1):175 -- 192, 1976. ISSN 0030-8730.
[34] K. Thomsen. Discrete asymptotic homomorphisms in E-theory and
KK-theory. Math. Scand., 92(1):103 -- 128, 2003. ISSN 0025-5521.
[35] J.-L. Tu.
La conjecture de Baum-Connes pour les feuilletages
moyennables. K-Theory, 17(3):215 -- 264, 1999.
[36] A. Valette. Projections in full C ∗-algebras of semisimple Lie groups.
ISSN 0025-5831. doi: 10.1007/
Math. Ann., 294(2):277 -- 287, 1992.
BF01934326. URL http://dx.doi.org/10.1007/BF01934326.
[37] `E. B. Vinberg, editor. Lie groups and Lie algebras, III, volume 41 of
Encyclopaedia of Mathematical Sciences.
[38] A. Wassermann. Une d´emonstration de la conjecture de Connes-
Kasparov pour les groupes de Lie lin´eaires connexes r´eductifs. C. R.
Acad. Sci. Paris S´er. I Math., 304(18):559 -- 562, 1987. ISSN 0249-6291.
[39] Y. Watatani. Index for C ∗-subalgebras. Mem. Amer. Math. Soc., 83
(424):vi+117, 1990. ISSN 0065-9266. doi: 10.1090/memo/0424. URL
26
MARIUS DADARLAT AND ULRICH PENNIG
http://dx.doi.org/10.1090/memo/0424.
[40] J. A. Wolf, M. Cahen, and M. De Wilde, editors. Harmonic analysis
and representations of semisimple Lie groups, volume 5 of Mathematical
Physics and Applied Mathematics.
MD: Department of Mathematics, Purdue University, West Lafayette, IN
47907, USA
E-mail address: [email protected]
UP: School of Mathematics, Cardiff University, Senghennydd Road, Cardiff
CF24 4AG, UK
E-mail address: [email protected]
|
1604.04119 | 1 | 1604 | 2016-04-14T11:50:20 | Geometric implications of the M(r,s)-properties and the uniform Kadec-Klee property in JB*-triples | [
"math.OA",
"math.FA"
] | We explore new implications of the $M(r,s)$ and $M^*(r,s)$ properties for Banach spaces. We show that a Banach space $X$ satisfying property $M(1,s)$ for some $0<s\leq 1$, admitting a point $x_{0}$ in its unit sphere at which the relative weak and norm topologies agree, satisfies the generalized Gossez-Lami Dozo property. We establish sufficient conditions, in terms of the $(r, s)$-Lipschitz weak$^*$ Kadec-Klee property on a Banach space $X$, to guarantee that its dual space satisfies the UKK$^*$ property. We determine appropriate conditions to assure that a Banach space $X$ satisfies the $(r, s)$-Lipschitz weak$^*$ Kadec-Klee property. These results are applied to prove that every spin factor satisfies the UKK property, and consequently, the KKP and the UKK properties are equivalent for real and complex JB$^*$-triples. | math.OA | math |
GEOMETRIC IMPLICATIONS OF THE M(r, s)-PROPERTIES
AND THE UNIFORM KADEC-KLEE PROPERTY IN
JB∗-TRIPLES
LEI LI, EDUARDO NIETO, AND ANTONIO M. PERALTA
Abstract. We explore new implications of the M (r, s) and M ∗(r, s) prop-
erties for Banach spaces. We show that a Banach space X satisfying prop-
erty M (1, s) for some 0 < s ≤ 1, admitting a point x0 in its unit sphere at
which the relative weak and norm topologies agree, satisfies the generalized
Gossez-Lami Dozo property. We establish sufficient conditions, in terms of the
(r, s)-Lipschitz weak∗ Kadec-Klee property on a Banach space X, to guarantee
that its dual space satisfies the UKK∗ property. We determine appropriate
conditions to assure that a Banach space X satisfies the (r, s)-Lipschitz weak∗
Kadec-Klee property. These results are applied to prove that every spin factor
satisfies the UKK property, and consequently, the KKP and the UKK proper-
ties are equivalent for real and complex JB∗-triples.
1. Introduction
Banach spaces which are M-ideals in their bidual have been intensively studied
during decades due the rich geometric and isometric properties that they enjoy
(compare the monograph [23]). J.C. Cabello and the second author of this note
explore a weaker notion in [9]. Following the just quoted reference, a Banach
space X satisfies the M(r, s)-inequality (with r, s ∈]0, 1]) if
kx∗∗∗k ≥ rkπ(x∗∗∗)k + skx∗∗∗ − π(x∗∗∗)k,
for all x∗∗∗ in the third dual, X ∗∗∗, of X, where π is the canonical projection of
X ∗∗∗ onto X ∗. The M(1, 1)-inequality is the classical notion of M-ideal [23]. In-
spired by the so called properties (M) and (M ∗), studied in [23], J.C. Cabello and
the second author of this note introduce in [10] properties M(r, s) and M ∗(r, s).
For the concrete definition, we recall that, given r, s ∈ ]0, 1], a Banach space
X has property M(r, s) (resp. M ∗(r, s)) if whenever u, v ∈ X (resp. X ∗) with
kuk ≤ kvk and (xα) is a bounded weakly (resp. weak∗-) null net in X (resp. X ∗),
then
lim sup
α
kru + sxαk ≤ lim sup
α
kv + xαk.
Date: October 23, 2018.
2000 Mathematics Subject Classification. 46B20, 46B22, 17C65, 46B04.
Key words and phrases. property (M ); property M (r, s); Kadec-Klee property; Uniform
Kadec-Klee property; spin factors.
First author partially supported by NSF of China (11301285 and 11371201). Second and
third authors partially supported by Junta de Andaluc´ıa grant FQM375 and the Spanish Min-
istry of Economy and Competitiveness and European Regional Development Fund project no.
MTM2014-58984-P.
1
2
LI, NIETO, AND PERALTA
It should be noted here that properties M(1, 1) and M ∗(1, 1) are precisely prop-
erties (M) and (M ∗) in the terminology of [23].
Properties M(r, s) have been successfully applied in fixed point theory (com-
pare [20, 15, 10]).
The fixed point theory motivated the developing of many interesting properties
in Banach space theory. That is also the case of the Kadec-Klee (or Radon-Riesz)
property (KKP in the sequel). We recall that a Banach space has the KKP if any
sequence in the unit sphere whose weak limit is also in the unit sphere, is indeed
norm convergent. The uniform Kadec-Klee property for the weak topology on a
Banach space was introduced by R. Huff [24] as a useful substitute for uniform
convexity, especially in many non-reflexive spaces. D. van Dulst and B. Sims
[17]showed that the uniform Kadec-Klee property for weak and weak∗ topologies
implies weak (resp. weak∗) normal structure.
It is well-known (see [2, 34]) that the Schatten p-classes Cp, 1 ≤ p < ∞, have
the KKP. It has been shown by C. Lennard [30] that the direct argument given
by J. Arazy [2] for trace-class operators can be refined to show that L1(H), the
space of trace class operators on an arbitrary infinite-dimensional Hilbert space
H satisfies a stronger property called the uniformly Kadec-Klee in the weak∗
topology UKK∗ (see section 2 for the detailed definitions). From a somewhat
different viewpoint, it is a classical theorem of F. Riesz that norm convergence
for sequences in the unit sphere of L1[0, 1] coincides with convergence in measure.
Appropriate uniform version of this theorem may be found in [31].
Independently from the fixed point theory, The Kadec-Klee property has been
deeply studied in certain particular classes of Banach spaces including C∗-algebras
and JB∗-triples in connection with the Alternative Dunford-Pettis property (com-
pare [1, 6, 3] and [8]). Proposition 2.13 in [6] provides a complete description of
those JB∗-triples satisfying the KKP, namely, a JB∗-triple satisfies this property
if and only if it is finite-dimensional or a Hilbert space or a spin factor. A sim-
ilar conclusion holds for real JB∗-triples (compare [3, Proposition 3.13]). It is a
natural open problem to ask wether a JB∗-triple satisfying the KKP satisfies or
not the stronger UKK property. This is one of the motivations for this note.
In section 3 we explore new geometric implications of the M(r, s) and M ∗(r, s)
properties. In [20, Theorem 2.4 and Corollary 2.5], J. Garc´ıa-Falset and B. Sims
proved that if a Banach space X has property (M), and SX contains a point at
which the relative weak and norm topologies agree, then X has w-ns. We estab-
lish a generalization of this result by showing that a Banach space X satisfying
property M(1, s) for some 0 < s ≤ 1, and admitting a point x0 in its unit sphere
at which the relative weak and norm topologies agree, satisfies the generalized
Gossez-Lami Dozo property (see Proposition 3.1). In Theorem 3.2 we prove that
for a Banach space X satisfying property M ∗(1, s) for some 0 < s ≤ 1, reflexivity
of X admits many equivalent reformulations in terms of classical properties like
Radon-Nikod´ym property, PCP, and KKP.
Proposition 3.5 provides sufficient conditions, in terms of the (r, s)-Lipschitz
weak∗ Kadec-Klee property on a Banach space X, to guarantee that its dual space
M (r, s)-PROPERTIES AND THE UNIFORM KADEC-KLEE PROPERTY IN JB∗-TRIPLES 3
satisfies the UKK∗ property (see definitions below). Theorem 3.6 sets appropriate
conditions to assure that a Banach space X satisfies the (r, s)-Lipschitz weak∗
Kadec-Klee property. These two results, appropriately combined, allow us to
establish, in section 4, that every spin factor satisfies the UKK property (Theorem
4.1), and consequently, the KKP and the UKK properties are equivalent for real
and complex JB∗-triples (see Corollary 4.2 and Proposition 4.4). We also obtain,
with similar arguments and tools, another proof of the result, established by C.
Lennard, which asserts that L1(H) satisfies the UKK∗ property (Theorem 4.3).
2. Background and basic definitions
Throughout the paper, the closed unit ball and the unit sphere of a given
Banach space X will be denoted by BX and SX, respectively.
A sequence (xn) in a Banach space X is called separated (respectively, ε-
separated ) if sep(xn) := inf{kxn − xmk : n 6= m} > 0 (respectively, ≥ ε). It is
known that X has the Kadec-Klee property if and only if every separated weakly
convergent sequence (xn) in the closed unit ball of X converges to an element of
norm strictly less than one.
Following standard notation, we say that X satisfies the uniform Kadec-Klee
(UKK) property if for each ε > 0 there exists δ > 0 such that for every sequence
(xn) in the closed unit ball of X satisfying sep(xn) ≥ ε and (xn) → x weakly,
then it holds that kxk < 1 − δ (see [24]).
Clearly, if X has the Schur property (i.e. norm and weak convergent sequences
in X coincide) or if X is uniformly convex then X has the UKK property. While
uniformly convex spaces are necessarily reflexive, it turns out that many classical
non-reflexive spaces, e.g. the Hardy space H 1 of analytic functions on the ball or
on the polydisc in CN [5], the Lorentz space Lp,1(µ) [12, 14] and the trace class
operators L1(H) [24], satisfy the UKK or the UKK∗ property.
A Banach space X is said to be nearly uniformly convex if for every ε > 0,
there exists a δ, 0 < δ < 1 such that for any sequence {xn} in the closed unit ball
BX of X with sep(xn) ≥ ε, then conv({xn}) ∩ (1 − δ)BX 6= ∅.
We recall next the definition of the uniform weak∗ Kadec-Klee property, which
is somehow due to B. Sims [35]. Suppose that X is a dual Banach space and ε > 0.
X is ε-uniformly Kadec-Klee in the weak*-topology (ε-UKK∗ in the sequel) if there
exists δ ∈ (0, 1) such that whenever C is a weak∗ compact, convex subset of unit
ball BX with sup{sep(xn) : (xn) ⊂ C} ≥ ε it follows that C ∩ (1 − δ)BX 6= ∅.
X is said to have the uniform Kadec-Klee property in the weak*-topology (UKK∗
property) if it is ε-UKK∗ for all ε > 0.
If BX ∗ is weak∗ sequentially compact, the weak∗ uniform Kadec-Klee property
can be reformulated in terms of the definition given for UKK property in page 3,
but replacing weak convergence with weak∗ convergence (see [17, §3]).
3. Banach spaces satisfying the M(r, s)-properties
In this section we shall revisit the additional properties satisfied by Banach
spaces possessing M(r, s)-properties in connection with previous contributions.
D[(xn)] = lim sup
n (cid:18)lim sup
m
kxn − xmk(cid:19) .
4
LI, NIETO, AND PERALTA
We recall that a Banach space X has weak normal structure (in short w-ns)
if every weakly compact convex subset K of X containing more than one point
admits a nondiametral point, that is, there exists x in K such that
diam (K) > sup {kx − yk : y ∈ K}.
When X is a dual Banach space, if in the above definition, the weak topology
is replaced with the weak∗-topology we say that X has weak∗ normal structure
(w∗-ns).
Another related property is the generalized Gossez-Lami Dozo property (GGLD
property in short). A Banach space X satisfies the GGLD property if for every
weakly null sequence (xn) in X such that limn kxnk = 1, we have that D[(xn)] > 1,
where
It is known that every Banach space satisfying the GGLD property has w-ns (see
[26]).
In [20, Theorem 2.4 and Corollary 2.5], J. Garc´ıa-Falset and B. Sims proved
that if a Banach space X has property (M), and SX contains a point at which
the relative weak and norm topologies agree, then X has w-ns. Making use of
a different technique, we show next that the class of Banach spaces for which
the conclusion of the above result of Garc´ıa-Falset and Sims holds also includes
Banach spaces satisfying property M(1, s) for some 0 < s ≤ 1. We prove that
a stronger conclusion is also true. In Example 3.10 below we present a Banach
space satisfying the M(1, s) property for a fixed 0 < s < 1, but failing the (M)
property.
Proposition 3.1. Let X be a Banach space having property M(1, s) for some
0 < s ≤ 1. If there exists a point x0 ∈ SX at which the relative weak and norm
topologies agree, then X has the GGLD property.
Proof. Suppose that X fails to have the GGLD property. Then there exists a
weak null sequence (xn) in X satisfying
n kxnk = 1 and D[(xn)] ≤ 1.
lim
Let x ∈ X with kxk < 1. Then, for m large enough we have kxk ≤ kxmk. So,
we deduce, by property M(1, s), that
lim sup
n
kx − sxnk ≤ lim sup
n
kxm − xnk.
Hence, taking limit on m, we have that lim supn kx − sxnk ≤ 1. Therefore, by
the triangular inequality, lim supn kx − sxnk ≤ 1 for all x ∈ BX. In particular,
by the weak lower semi-continuity of the norm, lim supn kx0 − sxnk = kx0k = 1.
So, we deduce from the hypothesis on x0 that limn kxσ(n)k = 0, for a suitable
subsequence (xσ(n)) of (xn), which is a contradiction.
(cid:3)
We recall that a Banach space X has the point of continuity property (PCP
in short) if every non-empty (weakly) closed subset K of BX admits a point at
which the identity map on K is weak-norm continuous (compare [18], [21, §III]).
M (r, s)-PROPERTIES AND THE UNIFORM KADEC-KLEE PROPERTY IN JB∗-TRIPLES 5
A wide list of geometric properties are equivalent for Banach spaces satisfying
M ∗(1, s) property for some 1 ≥ s > 0 (compare [20, Corollary 2.5]).
Theorem 3.2. Let X be a Banach space having property M ∗(1, s) for some 0 <
s ≤ 1. Then the following assertions are equivalent:
(i) X is reflexive;
(ii) X has the Radon-Nikod´ym property;
(iii) X has the PCP;
(iv) There is a point in SX at which the relative weak and norm topologies se-
quentially agree;
(v) X has the KKP.
(vi) X doesn't contain an isomorphic copy of c0.
Proof. Let us start with an observation. By [10, Proposition 3.1], X satisfies
property M(1, s) and the M(1, s)-inequality.
It is well known that every reflexive space has the Radon-Nikod´ym property
so (i) ⇒ (ii). The implications (ii) ⇒ (iii) ⇒ (iv), (v) ⇒ (iv), and (i) ⇒ (vi)
are also well known.
(iv) ⇒ (v) Let x0 be a point in SX at which the relative weak and norm
topologies on SX sequentially agree. Let x ∈ SX and let (xn) be a sequence in
X with (xn) w−→ x and kxαk −→ 1. By property M(1, s) and the weak lower
semi-continuity of the norm,
1 ≤ lim sup
n
kx0 + s(xn − x)k ≤ lim sup
n
kx + (xn − x)k = 1.
Thus, by assumptions, limn kxn − xk = 0.
If X is non-reflexive then, by [9, Corollary 3.4(1)], X contains an isomorphic
copy of c0. Thus, we have (vi) ⇒ (i). If we assume (iv), Proposition 3.1 implies
that X has the GGLD property. However, by [16, Theorem 8], X fails to have
the GGLD property, which is a contradiction. This proves (iv) ⇒ (i).
(cid:3)
Remark 3.3. The arguments given in the proof of (iv) ⇒ (v) in Theorem 3.2
above, actually show that a Banach space X satisfying the M(1, s)-property and
admitting a point in SX at which the relative weak and norm topologies sequen-
tially agree always satisfies the KKP.
Following [22, Definition 2.3], given r, s ∈]0, 1], we will say that a Banach space
X satisfies the (r, s)-Lipschitz weak∗ Kadec-Klee property (in short, (r, s)-LKK∗
property) if for every x∗ ∈ X ∗ and every weak∗ null sequence (x∗
n) in X ∗,
lim sup
n
kx∗
nk.
kx∗ + x∗
nk ≥ rkx∗k + s lim sup
n
It is obvious that the (r, s)-LKK∗ property implies property M ∗(r, s) (property
M ∗(r, s) and its sequential version are equivalent for separable spaces -see [23,
pp. 300, 301]).
Corollary 3.4. Let X be a Banach space satisfying the (1, s)-LKK∗ property.
Then the relative weak and norm topologies on SX do not coincide at any point
in SX.
6
LI, NIETO, AND PERALTA
Proof. Arguing by contradiction, we suppose the existence of a point x0 ∈ SX at
which the relative weak and norm topologies on SX coincide.
By [9, Proposition 2.5], X is an Asplund space. Thus, X does not satisfy Schur
property. So, there exists a weakly null sequence (xn) in SX. We deduce, by the
weak lower semi-continuity of the norm and [22, Lemma 2.5], that
lim sup
n
x0 + sxn = 1,
which implies, from our assumptions, that kxσ(n)k → 0, for a certain subsequence
(xσ(n)), which is impossible.
(cid:3)
Let us fix some notation. Given ε > 0 we denote
1 − r
s
1,ε =(cid:26)(r, s) ∈ (0, 1]2 :
π+
< ε(cid:27) .
For a Banach space X, we write
LKK∗(X) = {(r, s) ∈ (0, 1]2 : X satisfies the (r, s)-LKK∗ property}.
We establish now some results related to the UKK∗ property. We recall that for
every separable Banach space X, the closed unit ball of X ∗ is weak∗ sequentially
compact. It is also known that BX ∗ is weak∗ sequentially compact whenever X ∗
does not contain a copy of ℓ1. In particular, BX is weak∗ sequentially compact
whenever X is reflexive (cf. [13, Chapter XIII]).
Proposition 3.5. Let X be a Banach space. If LKK∗(X)∩ π+
1,ε 6= ∅ for all ε > 0
(in particular, if X satisfies the (1, s)-LKK∗ property), then X ∗ has the UKK∗
property.
Proof. We observe that the hypothesis LKK∗(X) ∩ π+
1,ε 6= ∅ for all ε > 0 implies
that BX ∗ is weak∗ sequentially compact. Indeed, since the (r, s)-LKK∗ property
implies property M ∗(r, s), we deduce that, for every ε > 0, there exist r, s ∈ (0, 1]
with 1−r
s < ε such that X satisfies property M ∗(r, s). In particular, we can assure
that X satisfies the M ∗(r, s) property for certain r, s with r + s > 1. Proposition
3.1 [10] and Corollary 2.8 in [9] assure that X does not contain an isomorphic
copy of ℓ1, and hence BX ∗ is weak∗ sequentially compact.
Let us fix an arbitrary ε > 0. Let (x∗
n) be a sequence in BX ∗ converging to x∗
n) ≥ ε and consider (r, s) ∈ LKK∗(X) ∩ π+
.
1, ε
2
in the weak∗ topology with sep(x∗
Then we have
1 ≥ lim sup
On the other hand, since sep(x∗
nk = lim sup
kx∗
n
n
Therefore,
kx∗ + (x∗
n − x∗)k ≥ rkx∗k + s lim sup
n) ≥ ε, we have that lim supn kx∗
n − x∗k.
kx∗
n − x∗k ≥ ε/2.
n
Since
1 − r
s
<
kx∗k ≤
, we have that δ = 1 −
ε
2
1 − s ε
r
1 − s ε
2
r
2
.
> 0.
(cid:3)
M (r, s)-PROPERTIES AND THE UNIFORM KADEC-KLEE PROPERTY IN JB∗-TRIPLES 7
We can establish now a key tool to deal with the (r, s)-LKK∗ property. We
recall that lim sup and lim inf of a bounded net of real numbers can be defined
in a similar manner as for sequences.
Theorem 3.6. Let X be a Banach space and r, s ∈]0, 1]. If there exists a net
j (x∗)− x∗k = 0 for all x∗ ∈ X ∗,
(Kj) of compact operators on X satisfying lim
j kK ∗
and
lim sup
j
x,y∈BX krKj(x) + s(y − Kj(y))k ≤ 1,
sup
(1)
then X satisfies the (r, s)-LKK∗ property.
Proof. Let us first observe that, by (1), we can easily deduce that (Kj) is a
bounded net. Since for each x∗ ∈ BX ∗ we have
j x∗k + skx∗ − K ∗
rkK ∗
x,y∈BX rx∗Kj(x) + sx∗(y − Kj(y)) ≤ sup
= sup
j x∗k = sup
y∈BX sx∗(y − Kj(y))
x,y∈BX krKj(x) + s(y − Kj(y))k,
x∈BX rx∗Kj(x) + sup
it can be easily seen that
lim sup
j
sup
x∗∈BX ∗(cid:0)rkK ∗
j (x∗)k + skx∗ − K ∗
j (x∗)k(cid:1) ≤ 1.
(2)
The equivalence of (1) and (2) was already established in [32, Proposition 2.3]
for non-necessarily compact operators. The argument above is included here for
completeness reasons.
Now, let us fix x∗ ∈ X ∗ and a weak∗ null sequence (x∗
arbitrary ε > 0, there exists a natural m0 such that
n) in X ∗. Given an
m≥m0 kx∗ + x∗
sup
mk ≤ lim sup
m
kx∗ + x∗
mk + ε.
Applying (2), we have
lim sup
j
sup
m≥m0
rkK ∗
j (x∗ + x∗
m)k + sk(x∗ + x∗
mk + ε.
kx∗ + x∗
≤ lim sup
m) − K ∗
j (x∗ + x∗
m)k
m
To simplify the notation, let us set
j (x∗ + x∗
aj := sup
m≥m0
rkK ∗
m)k + sk(x∗ + x∗
m) − K ∗
j (x∗ + x∗
m)k.
Since lim supj aj ≤ lim supm kx∗ + x∗
j ≥ j0 we have supl≥j al < lim supj aj + ε ≤ lim supm kx∗ + x∗
the inequality
mk + ε, there exists j0 such that for each
mk + 2ε. That is,
kx∗ + x∗
m)k < lim sup
mk + 2ε,
j (x∗ + x∗
m
rkK ∗
j (x∗ + x∗
m)k + sk(x∗ + x∗
m) − K ∗
holds for every j ≥ j0 and m ≥ m0.
and taking limit lim supm in the above inequality, we get
Since for a fixed subindex j, limm kK ∗
m) − K ∗
j (x∗)k + s lim sup
m)k = 0, fixing an arbitrary j ≥ j0
j (x∗)k ≤ lim sup
k(x∗ + x∗
kx∗ + x∗
mk + 2ε,
rkK ∗
j (x∗
m
m
8
LI, NIETO, AND PERALTA
for every j ≥ j0. Finally, taking lim supj≥j0 we deduce from the hypothesis that
kx∗ + x∗
the desired conclusion follows from the arbitrariness of ε.
rkx∗k + s lim sup
kx∗
mk + 2ε,
mk ≤ lim sup
m
m
(cid:3)
Remark 3.7. We observe that, accordingly to the arguments given in the proof
of Theorem 3.6, the assumption in (1) can be replaced with the inequality (2)
and the conclusion of the Theorem remains unaltered (c.f. [32, Proposition 2.3]).
Let us observe that a bounded net (Kj) of compact operators on a Banach
j (x∗) − x∗k = 0
space X satisfying lim
for all x∗ ∈ X ∗, is termed a shrinking compact approximation of the identity in
many references (see for example [23, Definition VI.4.16]).
We complete this section with a series of example that illustrate the optimality
j kKj(x) − xk = 0 for all x ∈ X and lim
j kK ∗
and novelty of our previous results.
Example 3.8. For 1 < p < +∞, consider the equivalent renorming of ℓp, X =
C ⊕∞ ℓp, where the usual norm is considered on ℓp. Then X satisfies the (r, s)-
LKK∗ property for all positive r and s with rp + sp ≤ 1. Indeed, Let πn, (n ∈
N), denote the natural projection of X onto the first nth coordinates. Given
(α, x), (β, y) ∈ BX and n ∈ N, we have
krπn+1(α, x) + s((β, y) − πn+1(β, y))kp =
= max((rα)p, rp
xip + sp
Xi=1
Xj=n+1
n
+∞
yjp) ≤ rp + sp ≤ 1.
Theorem 3.6 implies that X satisfies the (r, s)-LKK∗ property for evert r, s ∈ (0, 1]
with rp + sp ≤ 1.
We claim that X fails to have the KKP and property M(1, s) for every 0 <
s ≤ 1. To see this, let {en} denote the canonical basis in ℓp, and we define
w−→ x0 := (1, 0). However
xn = (1, en), its is clear that kxnk = 1 and (xn)
kxn − x0k = kenk = 1, for every natural n. Clearly, X is reflexive.
If X had
property M(1, s) for a real s ∈ (0, 1], Theorem 3.2 would imply that X has KKP,
which is impossible.
We present next a non-reflexive example. It is well known [19, Theorem 3.b.9
and Theorem 3.d.4] that the James tree space JT has w-ns and the KKP. Ap-
plying Proposition 3.5, we improve these facts, obtaining stronger conclusion.
Example 3.9. Let B denote the predual of the James tree space JT (compare
[19, page 175]). Then the following statements hold:
(a) B satisfies the (r, s)-LKK∗ property for all r, s > 0 with r2 + s2 ≤ 1;
(b) JT has the UKK∗ property;
(c) B doesn't satisfy property M ∗(1, s) for any 0 < s ≤ 1.
To see the first statement, we recall that JT is separable and admits a (bound-
edly complete) basis (en) (compare [19, Definition 3.a.2]). We denote by πn,
M (r, s)-PROPERTIES AND THE UNIFORM KADEC-KLEE PROPERTY IN JB∗-TRIPLES 9
n ∈ N, the natural projections onto the nth first coordinates of this basis. By
[19, Lemma 3.a.3] the inequality
kπn(x)k2 + kx − πn(x)k2 ≤ kxk2,
holds for every n ∈ N and x ∈ JT . So, for every n ∈ N and x∗, y∗ ∈ BJ T ∗,
n(y∗))k2 ≤ r2kx∗k2 + s2ky∗k2 ≤ r2 + s2 ≤ 1.
n(x∗) + s(y∗ − π∗
krπ∗
Theorem 3.6 assures that B satisfies the (r, s)-LKK∗ property.
Having in mind that JT does not contain ℓ1 (see [19, Theorem 3.a.8]), and
hence BJ T is weak∗ sequentially compact, statement (b) follows from Proposition
3.5.
(c) It is well known that B has the PCP (so, B doesn't contain an isomorphic
copy of c0) and fails to have the Radon-Nikod´ym property (see [19, 4.b.2, 4.b.5,
and 3.c.10]). Since B is non-reflexive and JT does not contain a copy of ℓ1, an
application of [10, Proposition 3.1] and [9, Corollary 3.4(2)] gives the desired
conclusion.
We consider next a series of examples inspired in [11] and [10].
Example 3.10. (Compare [11, Example 4.4] and [10, Proposition 4.1 and Corol-
αn of positive real numbers such that α1 6= α2 and
∞
lary 4.2(2.)]) Fix a series Xn
Xn=1
by
αn = a ∈ (0, 1). Let X be the Banach space c0 equipped with the norm given
kxk := sup(xn +
xjαj : n ∈ N) .
n
Xj=1
By [10, Proposition 4.1 and Corollary 4.2(2.)] we know that X satisfies the
M ∗(1, 1 − a) property, and hence the M(1, 1 − a) property (see [10, Proposi-
tion 3.1]). The arguments in the latest reference also prove that X satisfies the
hypothesis of Theorem 3.6, and hence X satisfies the (1, 1 − a)-LKK∗ property.
e1
e2, where (en) is the canonical basis of c0. It is not hard to check
and ky + enk =
, which shows that X does not satisfy property (M).
and y = 1
that kxk = kyk = 1, kx + enk = 1 + α1
1 + α2
1+α2
We claim that X does not satisfy the (M)-property. Indeed, let x = 1
+ αn → 1 + α1
1+α2
1+α1
1+α1
1+α1
+ αn → 1 + α2
1+α2
Example 3.11. (Compare [10, Proposition 4.1 and Corollary 4.3]) Let X be the
Banach space C × c0 equipped with the norm defined by
k(α, x)k := max{α + λkxk0,kxk0} ,
where 0 < λ < 1 is a fixed number. Corollary 4.3 in [10] and its proof show that
X satisfies the M ∗(1 − λ, 1) property, and hence the M(1 − λ, 1) property (see
[10, Proposition 3.1]).
We shall finally prove that X does not satisfy the M(1, s)-property for any
0 < s ≤ 1. Indeed, let x = (1, 0) and y = (0, e1), where (en) is the canonical basis
10
LI, NIETO, AND PERALTA
of c0. Clearly, kxk = kyk = 1. We can easily see that kx + s(0, en)k = 1 + λs and
ky + (0, en)k = 1, for every n ≥ 2, which proves the desired statement.
4. Uniform Kadec-Klee property in JB∗-triples
A spin factor is a JB∗-triple V whose norm and triple product are given by
the following rules. V is a Hilbert space with respect to an Hilbert product h·,·i,
there exists a conjugation ·
: V → V (i.e. a conjugate linear isometry of period
2) such that
and
{a, b, c} =
1
2
(ha, bic + hc, bia − ha, cib),
kak2 := ha, ai + (ha, ai2 − ha, ai2)
1
2 .
(3)
(4)
(5)
for all a, b, c ∈ V . Let k · k2 denote the Hilbert norm of V . Clearly,
∀ a ∈ V.
We note that the spin factor V is not strictly convex.
kak2 ≤ kak ≤
√2kak2,
There is an undoubted advantage of regarding spin factors as projective tensor
products of certain Hilbert spaces. Given two Banach spaces X and Y , the
symbol X ⊗π Y will denote the projective tensor product of X and Y , while k.kπ
will stand for the projective norm. It is known that
kukπ = inf( m
Xi=1
kxikkyik :
xi ⊗ yi) ,
m
Xi=1
for every u ∈ X ⊗ Y , and (X ⊗π Y )∗ = L(X, Y ∗) (see [33, §2.2]). It follows from
this identification that, given a Hilbert space H, the projective tensor product
H ⊗π H satisfies (H ⊗π H)∗ = B(H), the space of bounded linear operators on
H. It follows from the uniqueness of the predual of every von Neumann algebra
that H ⊗π H = L1(H), the trace class operators on H.
Spin factors can be represented as real projective tensor products of certain
Hilbert spaces. More concretely, by Lemma 3.5 in [3], for every spin factor V there
exists a real Hilbert space K such that V is JB∗-triple isometrically isomorphic
to C ⊗R
π K, the real projective tensor product of K and C, when the latter is
regarded as a real space. Under this point of view, the UKK property in spin
factors can be easily handle.
Theorem 4.1. Every spin factor satisfies the Uniform Kadec-Klee property.
Proof. Let V = C ⊗R
known that (cid:0)C ⊗R
bounded real linear operators from C
V as a dual Banach space with V∗ = V ∗ = BR(C
For each finite dimensional subspace F ⊆ K, let pF denote the orthogonal
π K be a spin factor, where K is a real Hilbert space. It is
R, K), where the latter denotes the space of all
R into K. Since V is reflexive, we can regard
π K(cid:1)∗ ∼= BR(C
projection of K onto F . We define a finite range operator
R, K).
KF : BR(C
R, K) → BR(C
R, K),
M (r, s)-PROPERTIES AND THE UNIFORM KADEC-KLEE PROPERTY IN JB∗-TRIPLES 11
given by KF (T ) = pF T . Clearly kKFk ≤ 1 for every F as above. Let F (K)
denote the set of all finite dimensional subspaces of K ordered by inclusion. If we
consider the net (KF )F ∈F (K), it can be easily checked that, for each T ∈ BR(C
R, K),
the net (kKF (T ) − Tk)F tends to zero.
Fix now S, T ∈ BR(C
R, K), r, s ∈ (0, 1], and λ ∈ C with λ ≤ 1. The inequality
krKF (S)(λ) + s(T − KF (T ))(λ)k2 = krpF S(λ) + s(Id − pF )T (λ)k2
holds for every F ∈ F (K). Therefore we prove that
= r2kpF S(λ)k2 + s2k(Id − pF )T (λ)k2 ≤ r2kTk2 + s2kSk2,
√r2 + s2 ≤ 1,
kSk,kT k≤1krKF (S) + s(T − KF (T ))k ≤
lim sup
F ∈F (K)
sup
for every r, s ∈ (0, 1] with r2 + s2 ≤ 1.
that
Pick now 1 ⊗ x + i ⊗ y in C ⊗R
π K and T ∈ BR(C
R, K). It can be easily seen
K ∗
F (1 ⊗ x + i ⊗ y)(T ) = (1 ⊗ x + i ⊗ y)KF (T ) = (1 ⊗ x + i ⊗ y)(pF T )
= hx, pF T (1)i + hy, pF T (i)i = hpF (x), T (1)i + hpF (y), T (i)i
= (1 ⊗ pF (x) + i ⊗ pF (y))(T ),
and hence K ∗
Consequently,
F (1⊗x+i⊗y) = 1⊗pF (x)+i⊗pF (y), for every 1⊗x+i⊗y ∈ C⊗R
π K.
lim
and hence
Applying Theorem 3.6 with X = BR(C
(cid:13)(cid:13)K ∗
F (1 ⊗ x + i ⊗ y) − (1 ⊗ x + i ⊗ y)(cid:13)(cid:13)π ≤ kpF (x) − xkK kpF (y) − ykK,
F ∈F (K)(cid:13)(cid:13)K ∗
F (1 ⊗ x + i ⊗ y) − (1 ⊗ x + i ⊗ y)(cid:13)(cid:13)π = 0.
R, K) and the net (KF )F ∈F (K), we deduce
that X satisfies the (r, s))-LKK∗ property for every r, s ∈ (0, 1] with r2 + s2 ≤ 1.
Proposition 3.5 proves that X ∗ = V satisfies the UKK∗ property. The proof
concludes by observing that for a reflexive space the UKK property and the
UKK∗ property are equivalent.
(cid:3)
Corollary 4.2. The Kadec-Klee property and the uniform Kadec-Klee property
are equivalent for JB∗-triples.
Proof. Let E be a JB∗-triple. Suppose that E has the Kadec-Klee property, by [6,
Proposition 2.13], we can derive that E is finite-dimensional or a spin factor or a
Hilbert space. It is known that a Hilbert space satisfies the uniform Kadec-Klee
property. If E is a finite dimensional JB∗-triple, then it is nearly uniform convex,
and hence E satisfies the uniform Kadec-Klee property (see, for example, [24,
p.744]). Finally, if E is a spin factor, Theorem 4.1 proves that E has the uniform
Kadec-Klee property.
(cid:3)
We can actually show that similar techniques to those employed above can be
also applied to give and alternative proof of a result due to C. Lennard [30]. As in
the proof given by Lennard in the just quoted paper, we rely on previous results
of J. Arazy [2]. We have already commented that the projective tensor product,
L1(H) = H ⊗π H, of a Hilbert space H with itself coincides with the predual of
12
LI, NIETO, AND PERALTA
B(H). In this setting L1(H) = H ⊗π H also is the dual of the space K(H) of
compact operators on H.
Theorem 4.3. [30, Theorem 2.4] For every Hilbert space H, the space L1(H) =
H ⊗π H of all trace class operators on H satisfies the UKK∗ property.
Proof. Keeping in mind the notation in the proof of Theorem 4.1, For each F ∈
F (H), let KF : K(H) → K(H), denote the finite rank (contractive) projection
given by KF (T ) = pF T pF , where pF denotes the orthogonal projection of H
onto F , and by an abuse of notation we regard T pF as mapping from F into H.
The net (KF )F ∈F (H) satisfies that, for each T ∈ K(H), the net (kKF (T ) − Tk)F
converges to zero.
Let us take an element u ∈ L1(H) = H ⊗π H, with kukπ ≤ 1, r, s ∈ (0, 1], and
F ∈ F (H). By [2, page 48 (iii)] (or [30, Proposition 2.2] or [3, Lemma 3.1]) it
follows that
r(cid:13)(cid:13)rK ∗
≤ r kpF upFkπ + s (kpF u(1 − pF )kπ + k(1 − pF )upFkπ + k(1 − pF )u(1 − pF )kπ)
On the other hand, it can be easily checked that for each ξ, η ∈ H, we have
K ∗
F (ξ ⊗ η) = pF (ξ) ⊗ pF (η), and hence
F (u))(cid:13)(cid:13)π = r kpF upFkπ + sk(u − pF upF )kπ
π − kpF upFk2
π(cid:1)
F (u)(cid:13)(cid:13)π + s(cid:13)(cid:13)(u − K ∗
≤ r + s√3(cid:0)kuk2
(cid:13)(cid:13)K ∗
F (ξ ⊗ η) − (ξ ⊗ η)(cid:13)(cid:13)π ≤ kpF (ξ) − ξkH kpF (η) − ηkH,
2 ≤ r + √6s.
1
lim
F ∈F (H)(cid:13)(cid:13)K ∗
We deduce by Theorem 3.6 (see also Remark 3.7) that K(H) satisfies the (r, s)-
F (u) − (u)(cid:13)(cid:13)π = 0, for each u in
which can be apply to prove that
H ⊗π H.
LKK∗-property for every r, s ∈ (0, 1] with r + √6s ≤ 1. Since the unit ball of
L1(H) = H ⊗π H is weak∗ sequentially compact (see [30, Lemma 2.3]), we are in
position to apply Proposition 3.5 to assure that L1(H) = H ⊗π H satisfies the
UKK∗.
(cid:3)
Let us briefly recall that a real JB∗-triple is a closed real subspace A of a
JB∗-triple E which is closed for the triple product of E (see [25]). There is
an equivalent definition asserting that every real JB∗-triple is precisely the real
subspace of all fixed points of a conjugation on a complex JB∗-triple. More
concretely, a conjugation on a complex Banach space X is a conjugate linear
isometry of order 2 τ : X → X. The associated real form of X is the set of all
τ -fixed points in X X τ := {x ∈ X : τ (x) = x}. Let us observe that X τ is the
image of the real contractive projection 1
2(id + τ ), and that
X = X τ ⊕ iX τ .
(6)
The Kaup-Banach-Stone theorem asserts that every linear (or conjugate linear)
surjective isometry on a complex JB∗-triple E is a triple isomorphism (see [27,
Proposition 5.5]). In particular, for every conjugation τ on E, Eτ is a real JB∗-
subtriple of E. It is established in [25, Proposition 2.2] that every real JB∗-triple
is of the form Eτ , where E is a complex JB∗-triple and τ is a conjugation on E.
M (r, s)-PROPERTIES AND THE UNIFORM KADEC-KLEE PROPERTY IN JB∗-TRIPLES 13
It is clear that if X is a complex Banach space satisfying the KKP or the UKK,
then X τ satisfies the same property for every conjugation τ on X. The example
given in [7, Example 1.3] provides a complex Banach space X and a conjugation
τ on X such that X τ is an infinite dimensional real Hilbert space, and hence
satisfies the KKP and the UKK property, while X does not satisfy the KKP.
Having in mind the obstacle raised by the previous example, the study of real
JB∗-triples satisfying the UKK requires of an appropriate strategy and does not
follow from Theorem 4.1 nor Corollary 4.2.
Proposition 4.4. The Kadec-Klee property and the uniform Kadec-Klee property
are equivalent for real JB∗-triples.
Proof. Let A be a real JB∗-triple satisfying the KKP. Proposition 3.13 in [3]
assures that A is finite dimensional or a real, complex, or quaternionic Hilbert
space or a real or complex spin factor. Applying the previous results, we can
assume that A is a real spin factor (i.e. a real form of a complex spin factor).
By [28, Theorem 4.1(viii)], there exits a real Hilbert space H, and closed linear
⊥, and A is the ℓ1-sum A = H1⊕1H2.
subspaces H1 and H2 of H such that H2 = H1
Since A∗ = H1⊕ℓ∞ H2. We consider the set F (H1)×F (H2) with the order given
by inclusion (i.e. (F1, F2) ≤ (G1, G2) if and only if Fi ⊆ Gi). For each (F1, F2) ∈
F (H1)×F (H2) we define a mapping K(F1,F2) : H1⊕ℓ∞ H2 → H1⊕ℓ∞ H2, defined by
is the orthogonal projection of Hi
K(F1,F2)(x1, x2) = (pF1
onto Fi. Clearly, K(F1,F2) is a finite rank projection on H1⊕ℓ∞ H2 with kK(F1,F2)k ≤
1, and for each (x1, x2) ∈ H1⊕ℓ∞ H2 we have lim
(F1,F2) kK(F1,F2)(x1, x2)−(x1, x2)k = 0.
It is also clear that for each (x1, x2) ∈ H1 ⊕1 H2 we have lim
(x1, x2) −
(x1, x2)k = 0.
and r, s ∈ (0, 1], we have
It is not hard to see that for each (x1, x2), (y1, y2) in the closed unit ball of A∗,
(x2)), where pFi
(F1,F2) kK ∗
(x1), pF2
(F1,F2)
rK(F1,F2)(x1, x2) + s((y1, y2) − K(F1,F2)(y1, y2))(cid:13)(cid:13)(cid:13)
(x2) + s(y2 − pF2
(y1))kH1
,krpF2
(x1) + s(y1 − pF1
(cid:13)(cid:13)(cid:13)
√r2 + s2.
(y2))kH2(cid:9) ≤
= max(cid:8)krpF1
Finally, combining Theorem 3.6 and Proposition 3.5 above we obtain that A
(cid:3)
satisfies the UKK property.
We have already commented that, F. Rambla and J. Becerra proved in [4]
that when X is a real or complex JB∗-triple or the predual of a real or complex
JBW∗-triple, then following are equivalent:
(a) X satisfies the fixed point property;
(b) X has normal structure;
(c) X is reflexive.
It is further known that when X is a real or complex JB∗-triple, then X has normal
structure if and only if X has weak normal structure. It was known that the class
of real or complex JB∗-triples satisfying the KK property is strictly smaller than
the class of reflexive real or complex JB∗-triples (compare [6, Proposition 2.13]).
14
LI, NIETO, AND PERALTA
We can conclude now that the class of JB∗-triples satisfying the UKK property is
precisely the class determined in [6, Proposition 2.13] and [3, Proposition 3.13].
References
[1] M.D. Acosta, A. M. Peralta, An alternative Dunford-Pettis property for JB∗-triples, Quart.
J. Math., 52, 391-401 (2001).
[2] J. Arazy, More on convergence in unitary matrix spaces, Proc. Amer. Math. Soc., 83, 44-48
(1981).
[3] J. Becerra Guerrero, A. M. Peralta, The Dunford-Pettis and the Kadec-Klee properties on
tensor products of JB∗-triples, Math. Z., 251, 117-130 (2005).
[4] J. Becerra Guerrero, F. Rambla-Barreno, The fixed point property in JB∗-triples and
preduals of JBW∗-triples, J. Math. Anal. Appl. 360, no. 1, 254-264 (2009).
[5] M. Besbes, S.J. Dilworth, P.N. Dowling, C.J. Lennard, New convexity and fixed point
properties in Hardy and Lebesgure-Bochner spaces, J. Funct. Anal., 119, 340-357 (1994).
[6] L.J. Bunce, A.M. Peralta, On weak sequential convergence in JB*-triple duals, Studia
Math., 160(2), 117-127 (2004).
[7] L.J. Bunce, A.M. Peralta, The alternative Dunford-Pettis property, conjugations and real
forms of C∗-algebras, J. London Math. Soc. (2) 71, no. 1, 161-171 (2005).
[8] L.J. Bunce, A.M. Peralta, Dunford-Pettis properties, Hilbert spaces and projective tensor
products, J. Funct. Anal. 253, no. 2, 692-710 (2007).
[9] J.C. Cabello, E. Nieto, On Properties of M -ideals, Rocky Mountain J. Math. 28, 61-93
(1998).
[10] J.C. Cabello, E. Nieto, On M -type structures and the fixed point property, Houston J.
Math. 26, 549-560 (2000).
[11] J.C. Cabello, E. Nieto, E. Oja, On ideals of compact operators satisfying the M (r, s)-
inequality, J. Math. Anal. Appl. 220, 334-348 (1998).
[12] N.L. Carothers, S.J. Dilworth, C.J. Lennard, D.A. Trautman, A fixed point property for
the Lorentz space Lp,1(µ), Indiana Univ. Math. J., 40, 345-352 (1991).
[13] J. Diestel, Sequences and series in Banach spaces, Graduate Texts in Mathematics, 92.
Springer-Verlag, New York, 1984.
[14] S.J. Dilworth, Y.-P. Hsu, The uniform Kadec-Klee property for the Lorentz spaces Lω,1,
J. Aust. Math. Soc., 60, no. 1, 7-17 (1996).
[15] T. Dom´ınguez-Benavides, J. Garc´ıa Falset, M.A. Jap´on, The τ -fixed point property for
nonexpansive mappings, Abstr. Appl. Anal. 3, No.3-4, 343-362 (1998).
[16] P.N. Dowling, Asymptotically isometric copies of c0 and renormings of Banach spaces, J.
Math. Anal. Appl. 228, 265-271 (1998).
[17] D. van Dulst, B. Sims, Fixed points of nonexpansive mappings and Chebyshev centers
in Banach spaces with norms of type (KK), in Banach Space Theory and its Applica-
tions (Proc. Bucharest 1981), Lecture Notes in Math., vol. 991, Springer-Verlag, Berlin,
Heidelberg and New York, 1983.
[18] G.A. Edgar, R.F. Wheeler, Topological properties of Banach spaces, Pacific J. Math. 115,
317-350 (1984).
[19] H. Fetter, B. Gamboa, The James Forest. London Mathematical Society, Lecture Note
Series 236. Cambridge University Press, Cambridge 1997.
[20] J. Garc´ıa-Falset, B. Sims, Property (M ) and the weak fixed point property, Proc. Amer.
Math. Soc. 125, 2891-2896 (1997).
[21] N. Ghoussoub, G. Godefroy, B. Maurey, W. Schachermayer, Some topological and geomet-
rical structures in Banach spaces, Mem. Amer. Math. Soc. 70, no. 378 (1987).
[22] G. Godefroy, N. Kalton, G. Lancien, Subspaces of c0(N) and Lipschitz isomorphisms,
Geom. Funct. Anal. 10, 798-820 (2000).
[23] P. Harmand, D. Werner, W. Werner, M -ideals in Banach Spaces and Banach Algebras.
Lecture Notes in Math. 1547, Springer, Berlin, 1993.
M (r, s)-PROPERTIES AND THE UNIFORM KADEC-KLEE PROPERTY IN JB∗-TRIPLES 15
[24] R. Huff, Banach spaces which are nearly uniformly convex, Rocky Mountain J. Math., 10,
743-749 (1980).
[25] J.M. Isidro, W. Kaup, A. Rodr´ıguez, On real forms of JB∗-triples, Manuscripta Math. 86,
311-335 (1995).
[26] A. Jim´enez-Melado, Stability of weak normal structure in James quasi reflexive space, Bull.
Austral. Math. Soc. 46, 367-372 (1992).
[27] W. Kaup, A Riemann mapping theorem for bounded symmetric domains in complex Ba-
nach spaces, Math. Z. 183, 503-529 (1983).
[28] W. Kaup, On real Cartan factors, Manuscripta Math. 92, 191-222 (1997).
[29] W.A. Kirk, An abstract fixed point theorem for nonexpansive mappings, Proc. Amer.
Math. Soc., 82, 640-642 (1981).
[30] C. Lennard, C1 is uniformly Kadec-Klee, Proc. Amer. Math. Soc., 109, 71-77 (1990).
[31] C. Lennard, A new convexity property that implies a fixed point property for L1, Studia
Math., 100, 95 -- 108 (1991).
[32] E. Nieto, On M-structure and weakly compactly generated Banach spaces, Proc. Edinb.
Math. Soc. II. Ser. 46, No. 3, 679-686 (2003).
[33] R.A. Ryan, Introduction to Tensor Products of Banach Spaces, Springer, 2002.
[34] B. Simon, Convergence in trace ideals, Proc. Amer. Math. Soc., 83, 39-43 (1981).
[35] B. Sims, The existence question for fixed points of nonexpansive maps, Lecture Notes, Kent
State University, 1986.
(Li) School of Mathematical Sciences and LPMC, Nankai University, Tianjin
300071, China
E-mail address: [email protected]
(Nieto and Peralta) Departamento de An´alisis Matem´atico, Facultad de Ciencias,
Universidad de Granada, 18071 Granada, Spain
E-mail address: [email protected], [email protected]
|
1608.02746 | 1 | 1608 | 2016-08-09T09:59:04 | Holomorphic Hilbert Bundles and the Frame Existence Problem | [
"math.OA"
] | We show that if $A=K(l^2)+ \mathbb{C}I_{l^2}$, then there exists a Hilbert $A$-module that possess no frames. | math.OA | math |
HOLOMORPHIC HILBERT BUNDLES AND THE
FRAME EXISTENCE PROBLEM
MOHAMMAD B. ASADI AND ZAHRA HASSANPOUR-YAKHDANI
Abstract. We show that if A = K(l2) + CIl2 , then there exists
a Hilbert A-module that possess no frames.
1. Introduction
The classical frame theory has been generalized to the setting of
Hilbert C∗-modules by Frank and Larson [5]. They concluded from
Kasparov's stabilization theorem that every countably generated Hilbert
C∗-module over a unital C∗-algebra has a standard frame. They asked
in [5, Problem 8.1], for which C∗-algebra A, every Hilbert A-module
has a frame? In 2010, Li solved this problem in the commutative uni-
tal case, and characterized the unital commutative C∗-algebra A that
every Hilbert A-module admits a frame as finite dimensional one [6].
In general case, the conjecture is as follow:
"Every Hilbert C∗-module over a C∗-algebra A admits a frame if and
only if A is a dual algebra."
In the commutative case, Li applied the Serre-Swan theorem. This
theorem states that there is an one to one correspondence between
finitely generated projective modules over a commutative C∗-algebra
C(Ω) and complex vector bundles over Ω [7].
In [4], Elliott and Kawamura shown that the vector space of bounded
uniformly continuous holomorphic sections of every holomorphic uni-
formly continuous Hilbert bundle of dual Hopf type over pure states of
a C∗-algebra A admits a unique structure of right Hilbert A-module.
In this paper, we give a partially affirmative response to the above
conjecture. Indeed, we have applied the Elliott-Kawamura approach
and concluded the following result:
"If H is a separable infinite dimensional Hilbert space and A =
K(H) + CIH , where K(H) is the C∗-algebra of compact operators on
H, then there exists a Hilbert A-module that possess no frames."
2010 Mathematics Subject Classification. Primary 46L08; Secondary 42C15,
46L05.
Key words and phrases. Hilbert C*-modules, frames, holomorphic bundles.
1
2
M. B. ASADI AND Z. HASSANPOUR-YAKHDANI
2. Holomorphic Hilbert Bundle
Let A be a C∗-algebra, A the spectrum of A and P (A) be the set
of pure states of A. In general, P (A) is not compact, in this case we
consider P0(A) = P (A) ∪ {0}. However, we set P0(A) = P (A), when
P (A) is compact.
For any π : A −→ B(Hπ) in A, and each unite vector h ∈ Hπ, the
linear functional f , given by f (·) = hπ(·)h, hi, is a pure state. In this
case, we use the notations π = [f ] and f = (π, e), where e = h ⊗ h.
Also, the unitary equivalence class of f (as a set) is equal to
R1(Hπ) := {e ∈ B(Hπ) : e is a rank one projection}.
R1(Hπ) has a natural holomorphic manifold structure that is indepen-
dent of the chosen representative element in each equivalence class in
P (A) [4]. Therefore, we can identify P (A) as the disjoint union of
projective spaces, i. e.,
P (A) = [
π∈ A
{π} × R1(Hπ).
Then P0(A) has a natural holomorphic manifold structure and it has
a natural uniform structure determined by the seminorms arising from
evaluation at the elements of A.
In [4], Elliott and Kawamura introduce the concept of uniformly
continuous holomorphic Hilbert bundle of dual Hopf type over pure
states of a C∗-algebra. In fact, we set
H = {B(Hπ, Kπ)e}(π∈ A∪{0},e∈R1(Hπ )),
where Kπ is a Hilbert space, for all π ∈ A. If X(H), the vector space of
bounded uniformly continuous holomorphic sections of H, exhausting
fibres then the pair (H, X(H)) is a uniformly continuous holomorphic
Hilbert bundle of dual Hopf type. In this case, for any S ∈ X(H) and
any π ∈ A there exists a operator Sπ ∈ B(Hπ, Kπ) such that
S((π, e)) = Sπ ◦ e
(e ∈ R1(Hπ))
As shown in [4], X(H) is a Hilbert A-module. In fact for any S, T ∈
X(H), the A-valued inner product is defined by S∗T , where
S∗(π, e) = e ◦ S∗
π ∈ eB(Kπ, Hπ),
for all (π, e) ∈ P0(A).
Since (S∗
consider S∗T belongs to A, by a Brown's result [2].
π ◦ Tπ)π∈ A ∈ Qπ∈ A(B(Hπ)) is uniformly continuous, we can
FRAME EXISTENCE PROBLEM
3
3. Frame Existence Problem
Theorem 3.1. Suppose that A is a C∗-algebra, f0 ∈ P (A), π0 = [f0],
Hπ0 is a separable Hilbert space and W is a countable subset of P (A)
such that f0 ∈ W \ W . If there exists a uniformly continuous holomor-
phic Hilbert bundle of dual Hopf type H = (B(Hπ, Kπ)eπ)(π,eπ)∈P0(A)
such that for any π ∈ [W ], Kπ is separable and Kπ0 is nonseparable,
then the Hilbert A-module X(H) possess no frames.
Proof. Assume that {Sj}j∈J is a frame for X(H). Hence, there exist
positive numbers C, D such that for any section S ∈ X(H) the following
inequality holds
CS∗S ≤ Σj∈J S∗SjS∗
j S ≤ DS∗S.
Hence, for every π ∈ A, eπ ∈ R1(Hπ) and S ∈ X(H), we have
j S(π, eπ)) ≤ DS∗S((π, eπ)),
CS∗S((π, eπ)) ≤ Σj∈J S∗SjS∗
so,
Ceπ ◦S∗
π ◦Sπ ◦eπ ≤ Σj∈J eπ ◦S∗
π ◦Sjπ ◦eπ ◦S∗
jπ ◦Sπ ◦eπ ≤ Deπ ◦S∗
π ◦Sπ ◦eπ.
In particular, for any nonzero element xπ ∈ Hπ we have
CkSπ(xπ)k2 ≤ Σj∈J hSπ(xπ), Sjπ(xπ)i 2≤ DkSπ(xπ)k2.
Since bounded holomorphic sections exhaust fibers, so for any yπ ∈ Kπ
there exists a section S ∈ X(H) such that Sπ(xπ) = yπ. Thus
Ckyπk2 ≤ Σj∈J hyπ, Sjπ(xπ) 2≤ Dkyπk2.
(3.1)
According to inequality 3.1, for all π ∈ A, 0 6= xπ ∈ Hπ, 0 6= yπ ∈ Kπ
the following set has to be countable
Fxπ,yπ := {j ∈ J : hyπ, Sjπ(xπ)i 6= 0}.
In particular, if π ∈ [W ] then Kπ is separable and so it has a count-
able orthonormal basis as Eπ. Hence, for each π ∈ [W ], the following
set has to be countable
Fπ,xπ := {j ∈ J : Sjπ(xπ) 6= 0} = [
{j ∈ J : hyπ, Sjπ(xπ)i 6= 0}.
yπ∈Eπ
Consequently, if we write W = {(πn, en) : n ∈ N}, then F = Sn∈N Fπn,xn
is a countable set, where for any n ∈ N, xn ∈ Hπn and en = xn ⊗ xn.
Also, we use the notation f0 = (π0, e0), where e0 = x0 ⊗ x0 for some
x0 ∈ Hπ0.
For each j ∈ F , Im(Sjπ0) is a separable space, since Hπ0 is sep-
arable. Then, K0 = hSj∈F Im(Sjπ0)i is a separable subspace of the
4
M. B. ASADI AND Z. HASSANPOUR-YAKHDANI
non-separable Hilbert space Kπ0, hence there exists a unit element
yπ0 ∈ Kπ0 that is orthogonal to K0. Then for any j ∈ F , S∗
jπ0(yπ0) = 0.
On the other hands, for any j ∈ J \ F , we have Sjπ0(x0) = 0,
since (π0, e0) ∈ W and Sj is continuous. Thus for any j ∈ J, we have
hyπ0Sjπ0(x0)i = 0. By (3.1), yπ0 is equal to zero, that is a contradiction.
Therefore, the Hilbert A-module X(H) admits no frames.
(cid:3)
4. K(l2) + C
In the following, we consider A = K(H) + CIH , where H is a sep-
arable infinite dimensional Hilbert space. Also, let {hn}n∈N be an or-
thonormal basis for H and en = hn ⊗ hn, for all n ∈ N.
We recall that, A = {π0, π1}, where π1 = id and π0(T + λIH) = λ,
for every T ∈ K(H) and λ ∈ C. Thus, we can consider
P (A) = ({π1} × R1(H)) ∪ {(π0, 1)}.
Note that in this case, P (A) is a compact hausdorff space and also
(π0, 1) ∈ W \ W , where W = {(π1, en) : n ∈ N}.
Theorem 4.1. There exists a uniformly continuous holomorphic vec-
tor bundle of dual Hopf type over P (A) satisfying the conditions of
Theorem 3.1.
Proof. Li shown that in [6, Lemma 2.1], there exists an uncountable set
F of injective maps from N to N such that for any distinct f, g ∈ F ,
f (n) 6= g(n) for all but finitely many n 6= N, and f (n) 6= g(m) for all
n 6= m.
Let Kπ1 = l2 with the standard basis {zn}n∈N and Kπ0 be a non-
separable Hilbert space with an orthonormal basis {hf }f ∈F indexed by
F . For each f ∈ F consider the isometry uf : H −→ l2, given by
uf (hn) = zf (n) for all n ∈ N. Also, we consider vf : C −→ Kπ0 by
vf (λ) = λhf .
Now, we can define Sf : P (A) −→ (Se∈R1(H) B(H, l2)e)∪(B(C, Kπ0)1)
by
Sf ((π, e)) = {
uf ◦ e
vf ◦ 1
π = π1
π = π0
Set V = {Pn
function (π, e) 7→ S(π, e) is continuous on P (A) for every S ∈ V .
i=1 λiSfi : n ∈ N, λi ∈ C, fi ∈ F }. We claim that the
For this, we note that if S = Pm
i=1 λiSfi ∈ V , then there is a finite
subset J of N such that fi(n) 6= fj(n), for all n ∈ J c and i 6= j. Hence,
FRAME EXISTENCE PROBLEM
5
if e = x ⊗ x, for some unite element x ∈ H, then we have
S(π1, e)2 =
λiufi(x)2 =
m
X
i=1
m
X
i=1
m
∞
λi(
X
hx, hnizfi(n))2
n=1
m
=
X
X
λihx, hnizfi(n)2 +
X
X
λihx, hnizfi(n)2
n∈J
i=1
m
X
X
i=1
n∈J
=
λihx, hnizfi(n)2 +
i=1
m
X
i=1
n∈J c
λi2(1 − X
hx, hnizfi(n)2).
n∈J
Now, if a net {(π1, eα)}α∈I is convergent to (π1, e) (or (π0, 1)) and
for every α ∈ I, eα = xα ⊗ xα for some unite element xα ∈ H, then
hxα, yi → hx, yi (or hxα, yi → 0), for all y ∈ H. Consequently, for
every f ∈ S and y1, · · ·, yN ∈ H, we have
N
X
n=1
hxα, ynizf (n)(= (
N
X
n=1
hxα, yni2)
1
2 ) →
N
X
n=1
hx, ynizf (n)
(or
N
X
n=1
hxα, ynizf (n) → 0).
Thus, S(π1, eα) → S(π1, e) (or S(π1, eα) → S(π0, 1)). This
proves the claim.
Therefore, V is a linear space of bounded holomorphic sections with
uniformly continuous norm and it exhausts each fibre. Now, as men-
tioned in [4], by Zorn's lemma, we can extend it to a linear space X(H)
of the bounded holomorphic sections with uniformly continuous norm,
maximal with this property, and exhausting each fibre. Clearly, X(H)
satisfies the conditions of Theorem 3.1.
(cid:3)
The following result can be obtained from Theorems 3.1 and 4.2.
Corollary 4.2. If A = K(l2) + CIl2, then there exists a Hilbert A-
module that possess no frames.
References
[1] M. Amini, M. B. Asadi, G. A. Elliott, Frames in Hilbert C∗-modules and Morita
equivalent C∗-algebras, Glassgow Math., to appear.
[2] L. G. Brown, Complements to various Stone-Weierstrass theorems for C∗-
algebras and a theorem of Shultz, Comm. Math. Phys. 143 (1992), 405- 413.
[3] J. Dixmier, C∗-algebras, North-Holland Publishing Company, 1977.
[4] G. A. Elliott, K. Kawamura, A Hilbert bundle characterization of Hilbert C∗-
modules, Trans. Amer. Math. Soc. 360 (2008), 4841- 4862.
[5] M. Frank, D. R. Larson, Frames in Hilbert C∗-modules and C∗-algebras, J.
Operator Theory 48 (2000), 273- 314.
6
M. B. ASADI AND Z. HASSANPOUR-YAKHDANI
[6] H. Li, A Hilbert C∗-module admitting no frames, Bull. London Math. Soc. 42
(2010), 388- 394.
[7] R. G. Swan, Vector bundles and projective modules, Trans. Amer. Math. Soc.
105 (1962), 264-277.
School of Mathematics, Statistics and Computer Science, College
of Science, University of Tehran, Tehran, Iran, and School of Math-
ematics, Institute for Research in Fundamental Sciences (IPM), P.O.
Box: 19395-5746, Tehran, Iran
E-mail address: [email protected]
School of Mathematics, Statistics and Computer Science, Collage
of Science, University of Tehran, Tehran, Iran
E-mail address: [email protected]
|
1711.05981 | 3 | 1711 | 2018-09-05T08:00:38 | Maximum modulus principle for "holomorphic functions" on the quantum matrix ball | [
"math.OA"
] | We describe the Shilov boundary ideal for a q-analog of the algebra of holomorphic functions on the unit ball in the space of $n\times n$ matrices and show that its $C^*$-envelope is isomorphic to the $C^*$-algebra of continuous functions on the quantum unitary group $U_q(n)$. | math.OA | math |
Maximum modulus principle for "holomorphic functions"
on the quantum matrix ball
Olga Bershtein
Department of Mathematics
University of Copenhagen
Universitetsparken 5, 2100 København Ø
email: [email protected]
Olof Giselsson and Lyudmila Turowska
Department of Mathematical Sciences,
Chalmers University of Technology and the University of Gothenburg,
Gothenburg SE-412 96, Sweden
email: [email protected], email: [email protected]
Abstract
We describe the Shilov boundary ideal for a q-analog of the algebra of holomorphic
functions on the unit ball in the space of n × n matrices and show that its C ∗-envelope
is isomorphic to the C ∗-algebra of continuous functions on the quantum unitary group
Uq(n).
1
Introduction
In 80s S.L. Woronowicz introduced the notion of a compact quantum group within the
framework of C ∗-algebras.
It was clear from the beginning that one can consider the
appearing theory as a counterpart to the theory of compact groups and their represen-
tations. For instance, the key Peter-Weyl theorem was proved for the case of compact
quantum groups.
The concept of a non-compact quantum group was much less clear, the corresponding
theory was in need of the key statements (for instance, it was not clear whether there
exists a Haar measure on a noncompact quantum group). At that point L. Vaksman
and his collaborators suggested another approach to representation theory of noncompact
quantum groups. The idea was to construct quantum analogs for homogeneous spaces X
of noncompact Lie groups G0 and then develop representation theory in connection with
these quantum spaces.
The object they started with were Hermitian symmetric spaces of non-compact type.
Let us briefly recall some facts about them and representation theory in connection.
1
Under the Harish-Chandra embedding, an irreducible Hermitian symmetric space X of
non-compact type allows a realization as a unit ball D in a certain normed vector space.
The group of biholomorphic automorphisms G0 = Aut(D) is a non-compact real Lie group,
and X is a homogeneous space of G0. One of the classical approaches to Harish-Chandra
modules for non-compact real groups is to derive them from a G0-orbits in certain flag
variety Xc (see, e.g.
[29]). The flag variety Xc is namely the dual of X (in the sense of
symmetric space), which is a Hermitian symmetric space of compact type. An important
fact is that there is a unique closed G0-orbit in Xc. This orbit namely corresponds to the
Shilov boundary S(D) of the bounded symmetric domain D. In particular, if we consider
the unit matrix ball X = Dn = {Z ∈ MatnI −ZZ ∗ > 0}, its group of symmetries G0 is the
group of pseudounitary matrices SU (n, n), the dual Hermitian symmetric space Xc is the
Grassmannian Grn(C2n) and the Shilov boundary is the group of unitary n × n-matrices
S(Dn) = Un. The series of Harish-Chandra modules related to the Shilov boundary are
called the principal degenerate series.
Quantum bounded symmetric domains were introduced in [22] and studied in the
series of papers (see [27] and references therein). The authors associated to a bounded
symmetric domain D in a complex vector space V non-commutative algebras C[V ]q and
Pol(V )q which they treated as the algebras of holomorphic resp. arbitrary polynomials on
the quantum vector space V ; the algebra of continuous functions on the quantum domain
D are derived then from Pol(V )q via some completion procedure. The main obstacle
for developing representation theory for quantum G0 in this setting was that there were
no quantum analogs for G0-orbits on flag varieties. The Shilov boundary was a happy
exception because it itself is a compact symmetric space. Although the construction of
quantum Shilov boundary for an arbitrary quantum bounded symmetric domain is rather
nontrivial, in some simpler cases the object was already known.
In particular, for the
quantum analog of the unit matrix ball Dn, the corresponding quantum Un, more precisely
a q-analog C[Un]q of the algebra of functions on Un, was well studied. In [25], using a
purely algebraic approach, Vaksman defined a q-analog of polynomials Pol(S(Dn))q on
the Shilov boundary Un = S(Dn) as a ∗-algebra isomorphic to C[Un]q and produced a
∗-homomorphism
jq : Pol(Matn)q → Pol(S(Dn))q;
the latter can be understood as a q-analog of the operator that restricts the polynomi-
als on Dn to its Shilov boundary; we refer to the kernel Jn := ker jq as algebraic Shilov
boundary ideal. The principal degenerate series of the quantum SU (n, n) related to the
Shilov boundary were studied in [4].
In the following we want to clarify the connec-
tion of the constructed quantum Shilov boundary with its topological (non-commutative)
counterpart.
Recall that the classical topological notion of Shilov boundary is used in the study of
uniform algebras A, that is, closed subalgebras of C(X) of all continuous functions on a
compact X that separate points and contain constants; this is the smallest subset S ⊂ X
where the functions in A attain their maximum; the latter means that the restriction of
the operator j : C(X) → C(S), f 7→ f S, to the subalgebra A is an isometry. In complex
analysis the uniform algebras are typically the algebra A(X 0) of functions holomorphic in
the interior X 0 of a compact domain X in a complex vector space; for instance A(Dn) ⊂
C(Dn) for the unit matrix ball Dn.
2
In the late 1960s W.Arveson initiated in his influential paper [1] the study of non-
commutative uniform algebras as (non-selfadjoint) subalgebras A of C ∗-algebras B and
introduced a non-commutative analog of the Shilov boundary; the latter is the largest ideal
J ⊂ B such that the quotient map j : B → B/J is a complete isometry when restricted to
A. A question, which was raised by Vaksman, is whether the "algebraically" constructed
Shilov boundary for quantum unit matrix ball coincides with the Arveson one. We give an
affirmative answer to this question for general value of n. The cases n = 1 and n = 2 were
treated in [26] and [19], respectively. More precisely, considering a completion CF (Dn)q
of Pol(Matn)q, a q-analog of continuous functions in Dn, we prove that the closure of the
algebraic Shilov boundary ideal Jn is the Shilov boundary ideal in the sense of Arveson
for the subalgebra A(Dn)q which is the closure of the holomorphic polynomials C[Matn]q
in the quantum space of n × n matrices. The pair (A(Dn)q, CF (Dn)q) is a q-analog of the
pair (A(Dn), C(Dn)). Note that the quotient CF (Dn)q/J n provides a realization of the
C ∗-envelope of A(Dn)q. We refer the reader to [2, 8, 16] for the background and recent
development concerning Shilov boundary and C ∗-envelope.
The paper is organized as follows. After finishing this section by introducing some
general notational conventions, in section 2.1-2.3 we introduce and collect some properties
of the main objects of our study, the algebras C[Matn]q, Pol(Matn)q and C[SUn]q. The
∗-algebra Pol(Matn)q possesses a C[SUn × SUn]q-comodule structure; it plays a crucial
role in our consideration and is presented in Lemma 3.
In section 2.4 we discuss ∗-
representations of Pol(Matn)q. In particular, we propose a new construction of the Fock
∗-representation πF,n of the ∗-algebra; the representation is known to be the only faithful
irreducible ∗-representation by bounded operators (see [21]). The construction allows to
derive a number of consequences about other ∗-representations of Pol(Matn)q which are
discussed in sections 2 and 3. The C ∗-algebra CF (Dn)q is defined to be the completion of
Pol(Matn)q with respect to the norm kf k = kπF,n(f )k, f ∈ Pol(Matn)q, and shown to be
a C ∗-subalgebra of the (spatial) tensor product C ∗(S)⊗n2
of n2 copies of the C ∗-algebra
generated by the unilateral shift S on ℓ2(Z+). In section 3 we prove the main results of
the paper: Theorems 3 and 4; they state that J n is the Shilov boundary ideal of CF (Dn)q
relative the subalgebra A(Dn)q. As a corollary we obtain that the C ∗-envelope of A(Dn)q is
isomorphic to the C ∗-algebra of continuous functions on the quantum unitary group Uq(n),
i.e. the C ∗-enveloping algebra of C[Un]q. Our approach is based on dilation-theoretic
arguments. Namely, we construct a ∗-representation of Pol(Matn)q which annihilates
the ideal Jn and compresses to the Fock representation when both are restricted to the
holomorphic part C[Matn]q. We refer the reader to [8] and [2] for the dilation ideas to the
Shilov boundary ideal and C ∗-envelopes.
We finish this section by recalling standard notation and notions that are used in
the paper. For a Hilbert space H we let B(H) denote the space of all bounded linear
operators on H. We shall write H ⊗ K for the Hilbertian tensor product of two Hilbert
spaces H and K and H ⊗n for the tensor product of n-copies of H. For an index set I,
{ei : i ∈ I} will always stand for the standard orthonormal basis in the Hilbert space
ℓ2(I). If A ∈ B(H), B ∈ B(K), then A ⊗ B stands for the operator in B(H ⊗ K) given
by A ⊗ B(ξ ⊗ η) = Aξ ⊗ Bη, ξ ∈ H, η ∈ K. If W ⊂ H is a closed subspace and A ∈ B(H)
leaves W invariant then we write AW for the restriction of A to W . If A, B are ∗-algebras
we write, as usual, A ⊗ B for the algebraic tensor product of the algebras; if A and B
3
are C ∗-algebras by A ⊗min B we denote their minimal C ∗-tensor product. Even though
we shall always have one of the C ∗-algebras A and B nuclear and hence all C ∗-norms on
A ⊗ B will be the same, we shall keep the notation A ⊗min B in order to distinguish the
latter from the algebraic tensor product of A and B. We write A⊗n for the minimal tensor
product of n copies of A.
For a set V , we denote, as usual, by Mn(V ) the set of all n × n matrices with entries
in V . It is clearly a vector space if V is such. For a map φ : V → W between vector
spaces V and W we let φ(n)((ai,j)i,j) = (φ(ai,j))i,j for each (ai,j)i,j ∈ Mn(V ).
If A is a ∗-algebra, any ∗-homomorphism π : A → B(H) is called a bounded ∗-
representation. As in this paper we mostly deal with bounded ∗-representation we shall
often omit the word bounded and write simply a ∗-representation. The set of all bounded
∗-representations of A will be denoted by Rep(A).
Let πi
: Ai → B(Hi) be a ∗-representation of Ai, i = 1, 2. We write π1 ⊗ π2 :
A1⊗A2 → B(H1⊗H2) for the ∗-representation given by π1⊗π2(a1 ⊗a2) = π1(a1)⊗π2(a2),
a1 ∈ A1, a2 ∈ A2.
It should not be confused with the tensor product π1 ⊗ π2 of ∗-
representations of a Hopf ∗-algebra A (used in section 2.3), which is actually the ∗-
homomorphism (π1 ⊗ π2) ◦ ∆ : A → B(H1 ⊗ H1), where ∆ : A → A ⊗ A is the co-product
on A.
2 The ∗-algebras Pol(Matn)q and C[SUn]q and their
representations
2.1 The ∗-algebra Pol(Matn)q
In what follows C is a ground field and q ∈ (0, 1). We assume that all the algebras
under consideration are unital. Consider the well known algebra C[Matn]q defined by its
generators zα
a , α, a = 1, . . . , n, and the commutation relations
a zβ
zα
a zβ
zα
a zβ
zα
b − qzβ
b zα
a = 0,
b − zβ
b zα
a = 0,
b − zβ
b zα
a − (q − q−1)zβ
a zα
b = 0,
a = b & α < β, or a < b & α = β, (1)
α < β & a > b,
α < β & a < b.
(2)
(3)
This algebra is a quantum analogue of the polynomial algebra C[Matn] on the space of n×n
matrices. It follows from the Bergman diamond lemma (see [3]) that the lexicographically
ordered monomials (zn
a ∈ Z+, α, a = 1, . . . , n, form
a basis of the vector space C[Matn]q. Hence C[Matn]q admits a natural grading given by
degzα
a = 1.
In a similar way, introduce the algebra C[Matn]q, defined by its generators (zα
n−1)γn−1
n−1 . . . (zn
1 . . . (z1
n (zn−1
1 )γn
n)γn
1 , γα
1 )γ1
a )∗,
α, a = 1, . . . , n, and the relations
(zβ
(zβ
(zβ
b )∗(zα
b )∗(zα
b )∗(zα
a )∗ − q(zα
a )∗ − (zα
a )∗ − (zα
a )∗(zβ
a )∗(zβ
a )∗(zβ
b )∗ = 0,
b )∗ = 0,
b )∗ − (q − q−1)(zα
a = b & α < β,
b )∗(zβ
a )∗ = 0,
or
a < b & α = β, (4)
α < β & a > b, (5)
α < β & a < b. (6)
4
A grading in C[Matn]q is given by deg(zα
a )∗ = −1.
Finally, consider the algebra Pol(Matn)q whose generators are zα
a , (zα
a )∗, α, a = 1, . . . , n,
and the list of relations is formed by (1) -- (6) and
(zβ
b )∗zα
a = q2 ·
n
Xa′,b′=1
with δab, δαβ being the Kronecker symbols, and
βα · zα′
a′ (zβ′
b′ )∗ + (1 − q2)δabδαβ,
(7)
Rkl
n
ij =
ba Rβ′α′
Rb′a′
q−1,
1,
−(q−2 − 1),
0,
Xα′,β′=1
detq z = Xs∈Sn
i 6= j & i = k & j = l
i = j = k = l
i = j & k = l & l > j
otherwise.
The involution in Pol(Matn)q is introduced in the obvious way: ∗ : zα
Recall a standard notation for the q-determinant of the matrix z = (zα
a 7→ (zα
a )n
a )∗.
α,a=1:
(−q)l(s)zs(1)
1
zs(2)
2
. . . zs(n)
n
,
with l(s) = card{(i, j) i < j & s(i) > s(j)}.
center of C[Matn]q.
It is well-known that detq z is in the
In order to introduce the Shilov boundary for the quantum matrix ball we will need
the algebra of regular functions on the quantum GLn, denoted by C[GLn]q (see [13]). It
is the localization of C[Matn]q with respect to the multiplicative system (detq z)N. The
algebra C[GLn]q possesses a unique involution ∗ given by
(zα
a )∗ = (−q)a+α−2n(detq z)−1 detq zα
a
(8)
with zα
Furthermore,
a being the matrix derived from z by deleting the α-th row and a-th column.
detq z(detq z)∗ = (detq z)∗ detq z = q−n(n−1)
(9)
(see [25, Lemma 2.1]).
Theorem 1 ([25]) There exists a unique ∗-homomorphism
ψ : Pol(Matn)q → (C[GLn]q, ∗)
such that ψ : zα
a 7→ zα
a , α, a = 1, . . . , n.
Note that the ∗-algebra (C[GLn]q, ∗) is isomorphic to the ∗-algebra C[Un]q = (C[GLn]q, ⋆)
of regular functions on the quantum Un (see [14]) with the isomorphism ι : (C[GLn]q, ∗) →
(C[GLn]q, ⋆) given by ι : zα
a , a, α = 1, . . . , n; the involution ⋆ in C[Un]q satisfies
(zα
a )⋆ = (−q)a−α(detq z)−1 detq zα
We finish the section by a lemma that makes a connection between Pol(Matn)q for
a , a, α = 1, . . . , n.
a → qα−nzα
different values of n. For ϕ ∈ [0, 2π) let
Πϕ(zi
j) =
q−1zi
j,
eiϕ,
0,
5
i, j < n,
i = j = n,
(10)
otherwise.
Lemma 1
1. The map Πϕ extends uniquely to a ∗-homomorphism from Pol(Matn)q
to Pol(Matn−1)q.
2. The map zi
j
Pol(Mat2n)q.
7→ zi+n
j+n, i, j = 1, . . . , n defines an embedding of Pol(Matn)q into
1.
Proof.
It is enough to check that all relations (1)-(7) turn to correct identities
under Πϕ. Relations (1)-(3) are checked easily. Indeed, if such a relation does not involve
elements of the last row or column of (zi
i,j=1, then it remains the same. If a relation
contains zn
n, then either it is a relation of type (1) and under Πϕ it transforms to 0 = 0, or
it is a relation of type (3) and under Πϕ it transforms to q−1eiϕzi
j. If, finally, a
relation involves some element from the last column or row but not zn
n, then it transforms
to 0 = 0.
j = q−1eiϕzi
j)n
Now let us check the relations between holomorphic and antiholomorphic generators,
namely, relations of type (7). Let us consider several cases. The first case is a 6= b and
α 6= β. Then the explicitly written commutation relation has the form (zβ
b )∗,
and this relation either survives under Πϕ or turns to the identity 0 = 0.
a = zα
b )∗zα
a (zβ
The second case is when one of the pairs of indices (a, b) and (α, β) coincides and the
other does not. These cases are completely similar, so we will elaborate here only one. Let
us suppose that a = b. Then the commutation relation can be rewritten in a more explicit
way as follows: (zβ
a′ )∗. If a = n, then Πϕ maps this relation to
the identity 0 = 0, since α 6= β. Let us now assume that a < n. If either α or β equals
n, then also Πϕ maps this relation to the identity 0 = 0. Finally, if all indices are less
than n, then Πϕ maps this relation to the relation q−2(zβ
a′ )∗
which holds in Pol(Matn−1)q.
a = qPn
a′=1 Ra′a′
a′=1 Ra′a′
a′q−2(zβ
aa zα
aa zα
a′(zβ
a )∗zα
a )∗zα
Now let us consider the last case where a = b and α = β. Then the relation can be
a = qPn−1
written more explicitly as
(zα
a )∗zα
a = q2
n
Xa′=1
n
Xα′=1
Ra′a′
aa Rα′α′
αα zα′
a′ (zα′
a′ )∗ + 1 − q2.
If a = α = n, then we have the standard relation (zn
n )∗zn
n = q2zn
n(zn
n )∗ + 1 − q2 which
is mapped by Πϕ to the identity 1 = q2 + 1 − q2.
If a = n and α < n (the other case is analogous), we have the relation
(zα
n )∗zα
n = q2
Rα′α′
αα zα′
n (zα′
n )∗ + 1 − q2.
n
Xα′=1
Πϕ maps this relation to the identity 0 = −q2(q−2 − 1) + 1 − q2.
Finally, if a < n and α < n, then we can rewrite the relation in the most explicit way
as follows:
(zα
a )∗zα
a = q2zα
a (zα
a )∗ − q2
n
Xa′=a+1
(q−2 − 1)zα
a′ (zα
a′ )∗ − q2
n
Xα′=α+1
(q−2 − 1)zα′
a (zα′
a )∗+
(q−2 − 1)2zα′
a′ (zα′
a′ )∗ + 1 − q2.
n
Xα′=α+1
q2
n
Xa′=a+1
6
Applying Πϕ, we get the relation
q−2(zα
a )∗zα
a = zα
a (zα
a )∗ −
(q−2 − 1)zα
a′ (zα
a′)∗ −
n−1
Xa′=a+1
Xa′=a+1
n−1
n−1
Xα′=α+1
n−1
Xα′=α+1
(q−2 − 1)zα′
a (zα′
a )∗+
(q−2 − 1)2zα′
a′ (zα′
a′ )∗ + q2(q−2 − 1)2 + 1 − q2.
Obviously, q2(q−2 − 1)2 + 1 − q2 = q−2(1 − q2), so after multiplying the relation by the
common factor q2 we get the corresponding relation in Pol(Matn−1)q. So, we also checked
all the relations between holomorphic and antiholomorphic generators of Pol(Matn)q and
the map Πϕ admits an extension to an algebra morphism. Its uniqueness is obvious.
2. It is a consequence of a similar easy inspection of the commutation relations; we
leave the details to the reader.
2.2 The quantum universal enveloping algebra UqslN and its
action on Pol(Matn)q
The Drinfeld-Jimbo quantum universal enveloping algebra is among the basic notions of
the quantum group theory. Recall the definition of the Hopf algebra Uq slN [11]. Let
(ai,j)N −1
i,j=1 be the Cartan matrix of slN :
2,
−1,
0,
i − j = 0,
i − j = 1,
otherwise.
ai,j =
The algebra Uq slN is determined by the generators Ei, Fi, Ki, K −1
the relations
i
, i = 1, . . . , N − 1, and
KiKj = KjKi, KiK −1
i = K −1
i Ki = 1, KiEj = qaij EjKi, KiFj = q−aij FjKi,
EiFj − FjEi = δij (Ki − K −1
)/(q − q−1),
i Ej − (q + q−1)EiEjEi + EjE2
E2
i Fj − (q + q−1)FiFjFi + FjF 2
F 2
EiEj − EjEi = FiFj − FjFi = 0,
i
i = 0,
i = 0,
i − j = 1,
i − j = 1,
i − j 6= 1.
The comultiplication ∆, the antipode S, and the counit ε are determined by
∆(Ei) = Ei ⊗ 1 + Ki ⊗ Ei, ∆(Fi) = Fi ⊗ K −1
S(Fi) = −FiKi,
S(Ei) = −K −1
i Ei,
ε(Ei) = ε(Fi) = 0,
ε(Ki) = 1.
i + 1 ⊗ Fi, ∆(Ki) = Ki ⊗ Ki,
S(Ki) = K −1
i
,
Let Uq sun,N −n denotes the ∗-Hopf algebra (Uq slN , ∗) given by
(K ±1
j
)∗ = K ±1
j
,
E∗
j =(KjFj,
−KjFj,
j 6= n,
j = n,
7
F ∗
j =(EjK −1
,
−EjK −1
j
j
j 6= n,
j = n,
,
with j = 1, . . . , N − 1.
An important fact from the theory of quantum bounded symmetric domains is that the
∗-algebra Pol(Matn)q possesses a quantum symmetry with respect to Uq sun,n. Recall a
general notion of a module algebra. Namely, let A be a Hopf algebra, B an A-module and
an algebra. Then B is called an A-module algebra if the multiplication and embedding
of the unit are morphisms of A-modules. If both A and B possess involutions, then they
should satisfy the compatibility condition (ab)∗ = (S(a))∗b∗ for each a ∈ A, b ∈ B.
Proposition 1 ([22]) The ∗-algebra Pol(Matn)q possesses a Uq sun,n-module algebra stru-
cture.
From here on we will focus on the ∗-subalgebra of Uq sun,n generated by all K ±1
, Ej, Fj
for j 6= n. This subalgebra will be denoted as Uq sun ⊗ Uq sun. The formulas below
completely determine the Uq sun ⊗ Uq sun-action on Pol(Matn)q.
j
For a, α = 1, . . . , n and k < n we have
a , a = k,
a , a = k + 1,
otherwise,
K ±1
k zα
q±1zα
q∓1zα
zα
a ,
a =
a = q1/2 ·(zα
a = q−1/2 ·(zα
Fkzα
Ekzα
a+1, a = k,
0,
otherwise,
a−1, a = k + 1,
0,
otherwise,
while for k > n we have
a , α = 2n − k,
a , α = 2n − k + 1,
otherwise,
K ±1
k zα
q±1zα
q∓1zα
zα
a ,
a =
a = q1/2 ·(zα+1
a = q−1/2 ·(zα−1
a
0,
a
0,
Fkzα
Ekzα
, α = 2n − k,
otherwise,
, α = 2n − k + 1,
otherwise.
(11)
(12)
(13)
(14)
(15)
(16)
Recall that the action on other elements of Pol(Matn)q can be obtained from the property
(g) and S(ξ)∗(f ∗) = (ξ(f ))∗ for ξ ∈ Uq sun ⊗ Uq sun, f , g ∈
i ⊗ ξ(2)
(in the Sweedler notation).
i
i
The above formulas show that the action of Uq sun ⊗ Uq sun preserves the degree of
(f )ξ(2)
that ξ(f g) = Pi ξ(1)
Pol(Matn)q and ∆(ξ) =Pi ξ(1)
i
each element f ∈ Pol(Matn)q. As a simple corollary, we have
Lemma 2 The Uq sun ⊗ Uq sun-action in Pol(Matn)q is locally finite.
8
2.3 The ∗-algebra C[SUn]q and its coaction
Recall the definition of the Hopf algebra C[SLn]q. It is defined by the generators {ti,j :
i, j = 1, . . . , n} and the relations
tα,atβ,b − qtβ,btα,a = 0,
tα,atβ,b − tβ,btα,a = 0,
tα,atβ,b − tβ,btα,a − (q − q−1)tβ,atα,b = 0,
detq t = 1.
a = b & α < β, or a < b & α = β,
α < β & a > b,
α < β & a < b,
Here detq t is the q-determinant of the matrix t = (tα,a)n
α,a=1.
It is well known (see [13] or other standard book on quantum groups) that C[SUn]q
def
=
(C[SLn]q, ⋆) is a Hopf ∗-algebra; the comultiplication ∆, the counit ε, the antipode S and
the involution ⋆ are defined as follows
∆(ti,j) =Xk
and
ti,k ⊗ tk,j,
ε(ti,j) = δij, S(ti,j) = (−q)i−j detq tji,
t⋆
i,j = (−q)j−i detq tij,
where tij is the matrix derived from t by discarding its i-th row and j-th column. We
have, in particular,
S(ti,j) = t⋆
j,i and S2(ti,j) = q2(i−j)ti,j,
i, j = 1, . . . , n
(17)
([13, Proposition 9.10]). From the relations it easily follows that the mapping ti,j 7→
qi−jti,j extends to an automorphism α of C[SUn]q such that α(t⋆
i,j. Combining,
for example, this result with (17), one can see that
i,j) = qj−it⋆
θ : ti,j 7→ qj−itj,i
(18)
gives a ∗-automorphism of C[SUn]q: we have θ(ti,j) = α(S(ti,j)⋆).
There is a canonical isomorphism C[SUn]q ≃ C[Un]q/hdetq z − 1i, given by ti,j 7→
zi
j + hdetq z − 1i, i, j = 1, . . . , n; here hdetq z − 1i denotes the two-sided ∗-ideal generated
by detq z − 1 (see [13]).
Theorem 1 and the remark after it give a ∗-homomorphism φ : Pol(Matn)q → C[SUn]q
such that
φ(zi
j ) = qi−nti,j, i, j = 1, . . . n.
(19)
Let us recall here the standard construction of ∗-representations of C[SUn]q. Consider
the Hilbert space l2(Z+) with the standard basis {en : n ∈ Z+}. The formulas
Π(t1,1)en = (1 − q2n)1/2en−1,
Π(t2,1)en = −qnen,
Π(t1,2)en = qn+1en,
Π(t2,2)en = (1 − q2n+2)1/2en+1, n ∈ Z+
define an irreducible ∗-representation of C[SU2]q, (see e.g. [13] or [15]).
9
For j ∈ {1, . . . , n − 1} let ψj : C[SUn]q → C[SU2]q denote the morphism of ∗-Hopf
algebras defined on the generators as follows:
ψj(ta,b) =(ta−j+1,b−j+1, j ≤ a, b ≤ j + 1,
δab, otherwise.
Recall that the symmetric group Sn is the Weyl group of sln, and denote by s1, . . . , sn−1
the transpositions (1, 2), . . . , (n − 1, n), respectively. For each sj one can associate an
irreducible ∗-representation of C[SUn]q via πsj = Π ◦ ψj.
Let s be an arbitrary element of the Weyl group, and s = sj1 . . . sjk its reduced
decomposition. Then one associates to it the ∗-representation of the Hopf ∗-algebra
C[SUn]q by the rule:
πs = πsj1
⊗ . . . ⊗ πsjk
.
A remarkable result of Soibelman [23, 24] and Soibelman and Vaksman [28] (see also
[15, chapter 3.6] for a general result) says that πs is irreducible and does not depend on
the reduced decomposition, i.e. two representations obtained through this construction
are unitarily equivalent if and only if they correspond to the same element of the Weyl
group; moreover, any irreducible representation of C[SUn]q differs from some πs by tensor
multiplication by a one-dimensional representation.
Recall a general fact on the connection between modules and comodules of algebras
and their Hopf duals. In the following proposition A◦ denotes the Hopf dual (or the finite
dual) of an algebra A (the definition can be found in [5][p.82]).
Proposition 2 ([5][p.86-87])
• Let A be an algebra and V be a left A-module. Then
V can be made into a right A◦-comodule whose associated left A-module is V if and
only if V is a locally finite A-module.
• V possesses a left A-module locally finite algebra structure if and only if there exists
a right A◦-comodule algebra structure on it.
Lemma 3 The map
Dn : zi
j 7→
n
Xa,b=1
za
b ⊗ tb,j ⊗ ta,i, i, j = 1, . . . , n
extends uniquely to a ∗-homomorphism
Dn : Pol(Matn)q → Pol(Matn)q ⊗ C[SUn]q ⊗ C[SUn]q.
Proof. Let us focus on the first part Uq sln ⊂ Uq sln ⊗ Uqsln, arguments for the second one
are the same. Recall that C[SLn]q is the (finite) dual for the quantum universal enveloping
algebra Uq sln. As the Uq sln-action on Pol(Matn)q defined by (11)-(13) is locally finite, we
can apply the previous proposition to obtain a right C[SLn]q-comodule algebra structure
on Pol(Matn)q. We claim that this coaction is given by
zi
j 7→
n
Xb=1
zi
b ⊗ tb,j,
1 ≤ i, j ≤ n.
(20)
10
To see this it is enough to verify that
ξ · zi
j =
n
Xb=1
zi
btb,j(ξ) for any ξ ∈ Uq sln, 1 ≤ i, j ≤ n.
(21)
For the generators ξ ∈ Uq sln this can be easily recovered from the explicit formulas for
the linear functionals tb,j ∈ C[SLn]q ⊂ (Uq sln)∗. One has
ti,i+1(Ei) = q−1/2,
ti+1,i(Fi) = q1/2,
ti,i(Ki) = q,
ti,i(Ki−1) = q−1,
i = 1, . . . , n,
and all other evaluations of tb,j on the generators of Uq sln are zero (see e.g. [13]).
To see that (21) holds for any ξ ∈ Uq sln it is enough to observe that whenever
b=1 zi
btb,j(ξm) = ξm · zi
j, m = 1, 2, then
n
Pn
Xb=1
zi
btb,j(ξ1ξ2) =
=
n
n
Xb=1
Xk=1
(
n
Xb=1
n
Xb=1
n
n
zi
b
Xk=1
Xk=1
zi
b∆(tb,j)(ξ1 ⊗ ξ2) =
tb,k(ξ1)tk,j(ξ2)
zi
btb,k(ξ1))tk,j(ξ2) =
(ξ1 · zi
k)tk,j(ξ2) = ξ1 · (
zi
ktk,j(ξ2))
n
Xk=1
= ξ1 · (ξ2 · zi
j) = (ξ1ξ2) · zi
j.
Since the algebra Pol(Matn)q is a Uq sun-module algebra, and the involutions in Uq sun
and C[SUn]q are compatible, one deduces that the map (20) extends to a ∗-homomorphism.
2.4 On ∗-representation of Pol(Matn)q
In this section we shall discuss ∗-representations of the ∗-algebra Pol(Matn)q. One of
the most important ∗-representations of Pol(Matn)q is the so-called Fock representation
which we now define.
Let HF,n be a Pol(Matn)q-module with one generator v0 and the defining relations
πF,n(zi
j)∗v0 = 0,
i, j = 1, . . . , n.
It was proved in [21, Corollary 2.3, Proposition 2.4, Proposition 2.5] that there ex-
ists a unique sesquilinear form (·, ·) on HF,n such that (v0, v0) = 1 and (πF,n(f )u, v) =
(u, πF,n(f ∗)v) for any f ∈ Pol(Matn)q, u, v ∈ HF,n. Moreover, the sesquilinear form
is positive definite and the linear mappings πF,n(f ) define a bounded ∗-representation,
called the Fock representation, of Pol(Matn)q; it acts on the Hilbert space HF,n which is
the completion of HF,n with respect to (·, ·).
We say that a ∗-representation π : Pol(Matn)q → B(H) has a vacuum vector if there
exists a unit vector v in H such that π(zi
j)∗v = 0 for any i, j = 1, . . . , n; any such vector
v is called a vacuum vector for π. The Fock representation is the only (up to unitary
equivalence) irreducible ∗-representation of Pol(Matn)q that possesses a vacuum vector
(see [21]).
11
Let CF (Dn)q be the C ∗-subalgebra of B(HF,n) generated by the operators πF,n(zi
j),
i, j = 1, . . . , n.
By [21], πF,n is a faithful irreducible representation of the ∗-algebra Pol(Matn)q. When
n = 1 and n = 2 it was proved in [17, 26] and [18] that kπ(a)k ≤ kπF,n(a)k for any bounded
∗-representation of Pol(Matn)q and a ∈ Pol(Matn)q, giving that CF (Dn)q is isomorphic to
the universal enveloping C ∗-algebra of Pol(Matn)q. Recently, using, in particular, results
from the present work, the second author proved in [9] that the statement holds for all
values of n.
In order to prove the faithfulness of the Fock representation of the ∗-algebra Pol(Matn)q
the authors of [21] gave an explicit construction of the Fock representation: for this they
embed first Pol(Matn)q in a localization of C[SL2n]q with an involution and then use
a concrete well understood ∗-representation of the latter ∗-algebra. The embedding is
highly non-trivial that makes the construction very difficult to work with. In this section
we propose a new simpler construction of the Fock representation of Pol(Matn)q built
out of a representation of C[SU2n]q. Using this construction we shall derive a number of
consequences concerning the C ∗-algebra CF (Dn)q and its ∗-representations.
Recall the ∗-representation Π of C[SU2]q from the previous section and let Cq, S, d(q) :
ℓ2(Z+) → ℓ2(Z+) be the operators defined as follows:
Sen = en+1, Cqen =p1 − q2nen,
We have
d(q)en = qnen.
(22)
Π(t1,1) = S∗Cq, Π(t1,2) = qd(q), Π(t2,1) = −d(q), Π(t2,2) = CqS.
As in [21] consider the element
u =(cid:18) 1
. . .
n + 1 n + 2 . . .
2
n n + 1 n + 2 . . .
2n
. . .
2
1
2n
n (cid:19)
(23)
of the symmetric group S2n. This element is the product of n cycles: u = c1 · c2 . . . · cn,
where ck = sk+n−1sk+n−2 . . . sk and sk is the transposition (k, k + 1). The concatenation
of the expressions for the cycles ck gives a reduced decomposition of u = σ1σ2 . . . σn2,
σi ∈ {s1, . . . , s2n−1}. Let
be the corresponding representation of C[SU2n]q (see section 2.3).
Consider the map
πu = πσ1 ⊗ πσ2 ⊗ . . . πσn2
ι : zi
j 7→ qi−ntn+i,n+j ∈ C[SU2n]q, 1 ≤ i, j ≤ n,
defined on the generators {zi
j : 1 ≤ i, j ≤ n} of Pol(Matn)q. It admits a unique extension
to a ∗-homomorphism ι : Pol(Matn)q → C[SU2n]q as the composition of the embedding
j to zi+n
of Pol(Matn)q into Pol(Mat2n)q that sends zi
j+n, 1 ≤ i, j ≤ n (see Lemma 1), and
the ∗-homomorphism φ : Pol(Mat2n)q → C[SU2n]q, zi
j 7→ qi−2nti,j (see (19)).
Theorem 2 Let T := πu ◦ ι : Pol(Matn)q → B(ℓ2(Z+)⊗n2
to the Fock representation πF,n.
). Then T is unitary equivalent
12
In order to understand the action of T we shall associate to it a collection of square
box diagrams with directed routes in the following way.
For 1 ≤ j, k ≤ n we have
T (zj
k) = qj−n
2n
Xk1,...,kn2−1=1
πσ1(tn+j,k1) ⊗ · · · ⊗ πσn2 (tkn2−1,n+k).
(24)
Recall from section 2.3 that the only non-zero factors in (24) are
πsi(ti,i+1) = qd(q), πsi(ti+1,i) = −d(q), πsi(ti,i) = S∗Cq,
πsi(ti+1,i+1) = CqS, πsi(tk,k) = I, k 6= i, i + 1,
where d(q), Cq, S : ℓ2(Z+) → ℓ2(Z+) are given by (22) and I stands for the identity
operator on ℓ2(Z+).
To each non-zero term πσ1(tn+j,k1)⊗· · ·⊗πσn2 (tkn2−1,n+k) in (24) we associate a square
tableau consisting of n2 equal boxes. Each box will represent a factor in the term. The
non-zero factors πsi(ti,i+1), πsi(ti+1,i), πsi(ti,i), πsi(ti+1,i+1), πsi(tk,k), k 6= i, i + 1, will be
represented by the following boxes with arrows:
πsi(ti,i+1)
πsi(ti+1,i)
πsi(ti,i)
πsi(ti+1,i+1)
πsi(tk,k)
(25)
(26)
(27)
(28)
(29)
the i-th column of the tableau will correspond to the part related to the cycle ci =
si+n−1si+n−2 . . . si in the decomposition u = c1 · c2 · . . . · cn, i.e.
πσn(i−1)+1 (tkn(i−1),kn(i−1)+1) ⊗ . . . ⊗ πσni(tkn(i−1),kn(i−1)+1)
= πsi+n−1(tkn(i−1),kn(i−1)+1) ⊗ . . . ⊗ πsi(tkn(i−1),kn(i−1)+1)
where the reading from the left to the right in the tensor product will correspond to the
moving up along the column. For instance, if the part of a non-zero term corresponding
to the first cycle is given by
πsn(tn+1,n) ⊗ πsn−1(tn,n−1) ⊗ . . . ⊗ πsk (tk+1,k) ⊗ πsk−1(tk,k) ⊗ πsk−2(tk,k) ⊗ . . . ⊗ πs1(tk,k)
we obtain the column:
πs1(tk,k)
...
πsk−1(tk,k)
...
πsn−1(tn,n−1)
πsn(tn+1,n)
13
It is not difficult to see that following the above algorithm applied to a non-zero term of
(24) we obtain in the corresponding n by n box tableau a path, called admissible and built
out of arrow boxes (25)-(29) with the start and end positions (n, j) and (k, n), respectively.
In fact, as πsl(tm,p) = δm,pI for n ≥ m > l + 1, we obtain that the elementary tensor
πσ1(tn+j,k1) ⊗ · · · ⊗ πσn(j−1)+1 (tkn(j−1),kn(j−1)+1) =
πsn(tn+j,k1) ⊗ . . . ⊗ πs1(tkn−1,kn) ⊗ πsn+1(tkn+1,kn+2) ⊗ . . . ⊗ πsn+j−1(tkn(j−1),kn(j−1)+1)
is non-zero if and only if n + j = k1 = k2 = . . . = kn(j−1) and kn(j−1)+1 is either n + j − 1
or n + j so that the last factor is either πsn+j−1(tn+j,n+j−1) or πsn+j−1(tn+j,n+j) while the
others are the identity operators. Hence the first arrow in the diagrams corresponding to
a summand in T (zj
, respectively.
Similarly, we observe that
k) occurs at the position (n, j) and it is either
or
πσn2 −k+1
(tkn2−k,kn2−k+1
= πsn+k−1(tkn2−k,kn2−k+1
) ⊗ · · · ⊗ πσn2 (tkn2−1,n+k)
) ⊗ . . . ⊗ πsn(tkn2−1,n+k)
is non-zero if and only if n + k = kn2−1 = . . . = kn2−k+1 and kn2−k is either n + k or
n + k − 1 so that the first factor is either πsn+k−1(tn+k,n+k) or πsn+k−1(tn+k−1,n+k) and
the others are the identity operators. Hence the last arrow in the diagrams corresponding
to a summand in T (zj
, respectively.
With a similar analysis it is not difficult to convince oneself that one gets a connected
path from (n, j) to (k, n); the details are left to the reader.
k) occurs at the position (k, n) and it is either
or
We have, in particular, that for n = 3 and j = k = 1 the diagrams representing the
non-zero terms in T (z1
1) are given by
The term corresponding, for example, to the first diagram can be easily recovered as
d(q) ⊗ d(q) ⊗ CqS ⊗ I ⊗ I ⊗ d(q) ⊗ I ⊗ I ⊗ d(q).
With the diagram approach the following result becomes evident.
Lemma 4 The vector v0 = e0 ⊗ . . . ⊗ e0 ∈ ℓ2(Z+)⊗n2
representation T .
is a vacuum vector for the ∗-
Proof. The statement follows from the simple observation that any route corresponding
to a non-zero term in (24) must contain the right hook arrow box (28) that represents the
factor CqS in the term. Recalling that (CqS)∗e0 = 0, we get the claim.
To proceed to the next claim, that can also be easily derived from the associated
square box diagrams, we enumerate the boxes in n by n tableaux by numbers from 1 to
n2 counting the boxes in each column from the bottom up and moving gradually from
the first column to the last one, i.e. we have the following ordering for the box positions
in the tableaux:
(n, 1) < (n − 1, 1) < . . . < (1, 1) < (n, 2) < (n − 1, 2) . . . < (1, 2) < . . .
< (n, n) < (n − 1, n) < . . . < (1, n)
14
Let h(k, l) be the number in {1, . . . , n2} corresponding to the box in the position (k, l).
We have h(k, l) = n(l − 1) + n − k + 1 although we will not use this expression.
Lemma 5 The subspace {T (a)v0 : a ∈ Pol(Matn)q} is dense in ℓ2(Z+)⊗n2
.
Proof. Consider all admissible routes from (n, j) to (k, n) in our set of tableaux. We
observe that the only route which does not contain the upper hook arrow box (27),
corresponding to the factor S∗Cq, is of the form
k
j
The route represents an operator of the form Rj
k ⊗ CqS
h(k,j)
↑
⊗ F j
k , where Rj
k, F j
k are tensor
products of the factors d(q) and I. Furthermore, S∗Cq can occur as a factor in a non-zero
term for T (zj
k) at the position with index which is strictly larger than h(k, j). Recalling
now that S∗Cq annihilates e0 we obtain that for v = em1 ⊗ . . . ⊗ emh(k,j) ⊗ e0 ⊗ . . . ⊗ e0
T (zj
k)v = qj−n(Rj
k ⊗ CqS
h(k,j)
↑
⊗ F j
k )v,
i.e. the only summand in T (zj
T (zj
k) to the vector v. Similarly, for m ≥ 1,
k) that corresponds to the above route survives after applying
T (zj
k)mv = q(j−n)m((Rj
k)m ⊗ (CqS)m
⊗ (F j
k )m)v
↑
h(k,j)
= βj
k(m)em1 ⊗ . . . ⊗ emh(k,j)−1 ⊗ emh(k,j)+m
⊗ e0 ⊗ . . . ⊗ e0
↑
h(k,j)
for some non-zero constant βj
By letting Xh(k,j) = T (zj
k(m) (depending on m1, . . . , mh(k,j)).
k) we can now easily obtain that
X
mn2
n2
. . . X m2
2 X m1
= β1(m1)β2(m1, m2)X
1 v0 = β1(m1)X
. . . X m3
mn2
n2
3
. . . X m2
mn2
n2
(em1 ⊗ em2 ⊗ e0 ⊗ . . . ⊗ e0) = . . .
(em1 ⊗ e0 ⊗ e0 ⊗ . . . ⊗ e0)
2
=
n2
Yi=1
βi(m1, . . . , mi)em1 ⊗ em2 ⊗ em3 ⊗ . . . ⊗ emn2
for non-zero βi(m1, . . . , mi), i = 1, . . . , n2. The proof is done.
The proof above gives more, namely if Xi, i = 1, . . . , n2, are as in the proof then we
obtain
15
Corollary 1 {X
mn2
n2
. . . X m2
2 X m1
1 v0, mi ∈ Z+} is an orthogonal basis of ℓ2(Z+)n2
.
We can now complete the proof of the theorem.
Proof of Theorem 2. As the Fock representation is the only ∗-representation (up
to unitary equivalence) that has a cyclic vacuum vector (see [21]), the statement follows
from Lemma 4 and Lemma 5.
We proceed with deriving a number of useful corollaries from our construction of the
Fock representation.
Let C ∗(S) be the C ∗-algebra generated by the isometry S. It is easy to see that the
operators Cq, S, d(q) ∈ B(ℓ2(Z+)) satisfy the following relations
Cq = (1 − q2)1/2(
q2nSn+1(S∗)n+1)1/2,
∞
Xn=0
d(q) =
∞
Xn=0
qn(Sn(S∗)n − Sn+1(S∗)n+1), S0 := 1,
(30)
(31)
where the series converge in the operator norm topology. Consequently, Cq, d(q) ∈ C ∗(S).
Identifying πF,n with the representation T of Theorem 2, we have
Corollary 2 CF (Dn)q ⊂ C ∗(S)⊗n2
.
Proof. Each T (zj
factor of which is either d(q), CqS, S∗Cq or I, the latter are elements of C ∗(S).
k) is a linear combination of elementary tensors in B(ℓ2(Z+))⊗n2
each
Let K be the space of compact operators on ℓ2(Z+) and C(T) be the C ∗-algebra of
continuous functions on the torus T = {z ∈ C : z = 1}. It is known that
φ : C ∗(S) → C(T), S 7→ z,
(32)
is a surjective ∗-homomorphism with kernel ker φ = K (see e.g.
(30)-(31)
[7]). In particular, by
φ(d(q)) = 0 and φ(Cq) = 1.
(33)
This is essential for the proof of the following lemma.
k → eiϕk ql−nδkl, k, l = 1 . . . , n.
Lemma 6 Let ϕ = (ϕ1, . . . , ϕn) ∈ [0, 2π)n and χϕ : zl
Then χϕ extends uniquely to a ∗-representation of Pol(Matn)q. Moreover, χϕ(a) ≤
kπF,n(a)k for any a ∈ Pol(Matn)q, i.e.
the map πF,n(a) 7→ χϕ(a), a ∈ Pol(Matn)q,
extends to a ∗-representation of CF (Dn)q.
Proof.
If we compose the ∗-homomorphism φ given by (32) with the evaluation map
at eiα, α ∈ [0, 2π), we get a ∗-homomorphism C ∗(S) → C determined by S 7→ eiα. For
τ = (τ1, . . . , τn2) let ∆τ : C ∗(S)⊗n2
→ C be a ∗-homomorphism defined by
I ⊗ ... ⊗ I ⊗ S
⊗ I ⊗ ... ⊗ I 7→ eiτi,
↑
i'th place
i ∈ {1, 2, . . . , n2}. As φ(d(q)) = 0 we obtain that ∆τ annihilates any elementary tensor
containing d(q) as a factor. Next we observe that the only admissible paths from (n, l)
16
to (k, n) that do not contain the arrow boxes
consisting of
and
following one after another; for example if n = 3, k = 1 we have
and
are the paths from (n, k) to (k, n)
and if n = 3, k = 2 we have
k) = eick ql−nδkl for some constants ck ∈ [0, 2π). Moreover, it is easy
k) =
k). As ∆τ is a ∗-homomorphism and hence contractive, we obtain that χϕ extends
Therefore, ∆τ ◦ T (zl
to see that we can choose the set {τ1, . . . , τn2} such that ck = ϕk and hence ∆τ ◦ T (zl
χϕ(zl
to a ∗-representation of Pol(Matn)q and
χϕ(a) = k∆τ ◦ T (a)k ≤ kT (a)k
for any a ∈ Pol(Matn)q.
In what follows we shall also need the so-called coherent representations which are
defined in a way similar to the Fock representation: consider a Pol(Matn)q-module HΩ
determined by a cyclic vector Ω such that
(zi
j)∗Ω = 0, if (i, j) 6= (1, 1) and (z1
1)∗Ω = e−iϕΩ,
for ϕ ∈ [0, 2π). Next we shall prove that the module action gives rise to a bounded ∗-
representation of Pol(Matn)q on a Hilbert space HΩ which is a completion of HΩ with
respect to some inner product. For ψ ∈ [0, 2π) let ωψ : C(T) → C be the evaluation map
f 7→ f (eiψ).
Proposition 3 Let T be the Fock representation given in Theorem 2. Then
Tψ := (id ⊗ . . . ⊗ (ωψ ◦ φ)
⊗ id ⊗ . . . ⊗ id) ◦ T : Pol(Matn)q → C ∗(S)⊗(n2−1)
↑
n
is unitary equivalent to a coherent representation of Pol(Matn)q.
Proof. Any admissible route from the position (n, 1) to (1, n) is either
1
1
17
or contains both the right hook and the upper hook arrow boxes in positions different
from (1, 1). Hence the corresponding summand in the expression for T (z1
1) is either
(−1)n−1d(q) ⊗ . . . ⊗ d(q) ⊗ CqS
↑
n
⊗ I ⊗ . . . ⊗ I ⊗ d(q)
. . . ⊗ I ⊗ . . . ⊗ I ⊗ d(q)
↑
2n
↑
n2
or an elementary tensor that contains both the factors CqS and S∗Cq in positions different
from n = h(1, 1) and has the identity operator as the n-th factor. Hence applying (id ⊗
. . . ⊗ (ωψ ◦ φ)
⊗ id ⊗ . . . ⊗ id) we obtain
↑
n
Tψ(z1
1)Ω = (−1)n−1eiψΩ
for Ω := e0 ⊗ . . . ⊗ e0 ∈ ℓ2(Z+)⊗(n2−1). That Tψ(zj
k)∗Ω = 0 for (j, k) 6= (1, 1) follows from
the fact that any admissible path from (n, j) to (k, n) contains the right hook arrow box
in a position different from (1, 1). That Ω is a cyclic vector can be seen using arguments
similar to one in Lemma 5.
It follows from [12, Proposition 1.3.3] that a coherent representation corresponding to
ϕ ∈ [0, 2π) is unique, up to unitary equivalence. We shall denote it by ρn
ϕ.
The following technical result will be needed in the next section and can be easily
derived from our box diagrams.
Lemma 7 Let ϕ ∈ [0, 2π). There exist operators Ajk, B ∈ B(ℓ2(Z
and ρn
ℓ2(Z
(n2−1)
+
ϕ are unitary equivalent to the following representations on ℓ2(Z
(n2−1)
+
), respectively:
)) such that πF,n
) ⊗ ℓ2(Z+) and
(n2−1)
+
πF,n : z1
ρn
ϕ : z1
1 7→ B ⊗ CqS + A11 ⊗ I,
zj
k 7→ Ajk ⊗ I, (j, k) 6= (1, 1)
1 7→ eiϕB + A11,
zj
k 7→ Ajk, (j, k) 6= (1, 1)
Proof. Let ψ ∈ [0, 2π), (−1)n−1eiψ = eiϕ and let T and Tψ be the Fock and the coherent
representations from Theorem 2 and Proposition 3, respectively. It follows from the proof
of the previous proposition that
T (z1
1) = R1
1 ⊗ CqS
↑
n
⊗ F 1
1 +Xi
A1
1(i) ⊗ I
↑
n
⊗ B1
1(i)
and
T (zj
k) =Xi
Aj
k(i) ⊗ I
↑
n
⊗ Bj
k(i), (k, j) 6= (1, 1),
for some elementary tensors R1
Applying (id ⊗ . . . ⊗ (ωψ ◦ φ)
1, F 1
k(i), Bj
1 , Aj
⊗ id ⊗ . . . ⊗ id) to (34)-(35) gives
k(i).
(34)
(35)
↑
n
Tψ(z1
1) = eiψR1
1 ⊗ F 1
1 +Xi
A1
1(i) ⊗ B1
1(i)
18
and
Tψ(zj
k) =Xi
Aj
k(i) ⊗ Bj
k(i), (k, j) 6= (1, 1).
Set Ajk = Pi Aj
ℓ2(Z+)⊗n2
→ ℓ2(Z+)⊗n2
k(i) ⊗ Bj
k(i) and B = R1
given by
1 ⊗ F 1
1 and consider the unitary operator U :
U : f1 ⊗ . . . ⊗ fn−1 ⊗ fn ⊗ fn+1 ⊗ . . . ⊗ fn2 7→ f1 ⊗ . . . ⊗ fn−1 ⊗ fn+1 ⊗ . . . ⊗ fn2 ⊗ fn.
Clearly, T is unitary equivalent via the operator U to πF,n giving the statement.
We conclude this section with a couple of lemmas on ∗-representations of Pol(Matn)q
induced from the coaction map, defined in the previous section.
Recall the coaction Dn : Pol(Matn)q → Pol(Matn)q ⊗ C[SUn]q ⊗ C[SUn]q defined in
Lemma 3 and the ∗-homomorphism Πϕ : Pol(Matn)q → Pol(Matn−1)q, ϕ ∈ [0, 2π), given
by (10). Clearly, if ρ is a ∗-representation of Pol(Matn−1)q, τ is a ∗-representation of
Pol(Matn)q, and π1, π2 are ∗-representations of C[SUn]q then ρ ◦ Πϕ and (τ ⊗ π1 ⊗ π2) ◦ Dn
are ∗-representations of Pol(Matn)q.
Lemma 8 Given ∗-representations π1, π2 of C[SUn]q, the ∗-representation (πF,n ⊗ π1 ⊗
π2) ◦ Dn of Pol(Matn)q is a direct sum ⊕iπF,n of copies of the Fock representation. In
particular, k(πF,n ⊗ π1 ⊗ π2) ◦ Dn(a)k ≤ kπF,n(a)k for any a ∈ Pol(Matn)q.
Proof.
Let {fi}i∈I , {gj }j∈J be orthonormal bases for the representation spaces Hπ1
and Hπ2 respectively and let v0 denote a vacuum vector for πF,n. Recall that C[Matn]q
stands for the subalgebra of Pol(Matn)q generated by the holomorphic generators zk
l ,
k, l = 1, . . . , n. We have that each v0 ⊗ fi ⊗ gj, i ∈ I, j ∈ J, is a vacuum vector for
τ := (πF,n ⊗ π1 ⊗ π2) ◦ Dn. Moreover, as πF,n(a)v0 ⊥ v0 for any a ∈ C[Matn]q with
deg a > 0, one can easily see that v0 ⊗ fk ⊗ gl is orthogonal to span{τ (a)(v0 ⊗ fi ⊗ gj) :
a ∈ C[Matn]q} whenever (i, j) 6= (k, l). As the latter subspace is invariant with respect to
τ (a), a ∈ Pol(Matn)q, we have
span{τ (a)(v0 ⊗ fi ⊗ gj) : a ∈ C[Matn]q} = span{τ (a)(v0 ⊗ fi ⊗ gj) : a ∈ Pol(Matn)q}
and the subspaces
span{τ (a)(v0 ⊗ fi ⊗ gj) : a ∈ Pol(Matn)q} and span{τ (a)(v0 ⊗ fk ⊗ gl) : a ∈ Pol(Matn)q}
are orthogonal whenever (i, j) 6= (k, l). This implies the statement.
Let w = sn−1 . . . s1 ∈ Sn and let πw denote the irreducible representation of C[SUn]q
corresponding to w.
Lemma 9 The coherent ∗-representation ρn
of τϕ,w,w := ((πF,n−1 ◦ Πϕ) ⊗ πw ⊗ πw) ◦ Dn.
ϕ is unitary equivalent to a sub-representation
It is enough to prove that τϕ,w,w possesses a vector Ω such that τϕ,w,w(zi
j)∗Ω = 0
Proof.
if (i, j) 6= (1, 1) and τϕ,w,w(z1
1)∗Ω = e−iϕΩ.
19
The representation τϕ,w,w acts on HF,n−1 ⊗ ℓ2(Z+)⊗(n−1) ⊗ ℓ2(Z+)⊗(n−1). Let e(n−1)
=
, where v0 is a vacuum vector for πF,n−1. Then
0
e0 ⊗ . . . ⊗ e0
and Ω = v0 ⊗ e(n−1)
0
⊗ e(n−1)
0
n−1
{z
}
τϕ,ω,ω(z1
1)∗Ω =
πF,n−1(Πϕ(za
b ))∗v0 ⊗ πw(t∗
b,1)e(n−1)
0
⊗ πw(t∗
a,1)e(n−1)
0
.
n
Xa,b=1
It follows from the definition of the Fock representation and the ∗-homomorphism Πϕ that
πF,n−1(Πϕ(zb
a))∗v0 = 0 whenever (a, b) 6= (n, n) and πF,n−1(Πϕ(zn
n ))∗v0 = e−iϕv0. As
πw(t∗
n,i)e(n−1)
0
=
n
Xk1,...,kn−2=1
πsn−1(t∗
n,k1)e0 ⊗ πsn−2(t∗
k1,k2)e0 ⊗ . . . ⊗ πs1(t∗
kn−2,i)e0,
πsk−1(t∗
k,l)e0 6= 0 if and only if l = k − 1 and πsk−1(t∗
n,1)e(n−1)
n,n−1)e0 ⊗ πsn−2(t∗
= πsn−1(t∗
0
πw(t∗
k,k−1)e0 = −e0, we obtain
n−1,n−2)e0 ⊗ . . . ⊗ πs1(t∗
2,1)e0 = (−1)n−1e(n−1)
0
and πw(t∗
Hence
n,i)e(n−1)
0
= 0 if i > 1.
τϕ,ω,ω(z1
1)∗Ω = e−iϕv0 ⊗ πw(t∗
n,1)e(n−1)
0
⊗ πw(t∗
n,1)e(n−1)
0
= e−iϕΩ.
Similarly
τϕ,ω,ω(zj
i )∗Ω = e−iϕv0 ⊗ πw(t∗
n,i)e(n−1)
0
⊗ πw(t∗
n,j)e(n−1)
0
= 0 if (i, j) 6= (1, 1).
3 Shilov boundary
The Shilov boundary ideal is a non-commutative analog of the Shilov boundary of a
compact Hausdorff space X relative to a uniform subalgebra A of C(X), which is, by
definition, the smallest closed subset K of X such that every function in A attains its
maximum modulus on K. The notion goes back to the fundamental work by W. Arveson,
[1], which gave birth to several directions in mathematics.
A typical "commutative" example that we should keep in mind is the algebra A of
holomorphic functions on the open unit ball Dn = {z ∈ Mn(C) : zz∗ < 1} which are
continuous on its closure. The Shilov boundary of Dn relative to A is known to be the
space of unitary n by n matrices.
To introduce the non-commutative version of the Shilov boundary recall that if V is
a subspace of a C ∗-algebra B, Mn(V ) is the space of n by n matrices in V with norm
induced by the C ∗-norm on Mn(B), then any linear map φ from V to another C ∗-algebra
C induces a linear map φ(n) : Mn(V ) → Mn(C) by letting
φ(n)((ai,j)i,j) = (φ(ai,j))i,j, (ai,j)i,j ∈ Mn(V ).
The linear map is called a complete isometry if φ(n) is an isometry for any n.
Assume V contains the identity of B and generates B as a C ∗-algebra. The following
definition was given by Arveson [1].
20
Definition 1 A closed two-sided ideal J in B is called a boundary ideal for V if the
canonical quotient map q : B → B/J is a complete isometry when restricted to V . A
boundary ideal is called a Shilov boundary or a Shilov boundary ideal for V if it contains
every other boundary ideal.
Arveson demonstrated the existence of the Shilov boundary in several situations of
particular interests.
In [10] Hamana proved that the boundary always exists for any
such subspace V . Another proof using dilation arguments was given by Dritschel and
McCullough in [8] (see also [2]). As the operator space structure on commutative C ∗-
algebras is minimal one has that a closed subset K of a compact Hausdorff space X is
the Shilov boundary relative to a uniform subalgebra of C(X) if and only if the ideal
J = {f ∈ C(X) : f K = 0} is the Shilov boundary ideal. This shows that the above
definition indeed generalises the commutative notion.
If J is the Shilov ideal for a unital subalgebra V of C ∗(V ), then C ∗(V )/J provides
e (V ) of V , a C ∗-algebra which is determined by the
a realization of the C ∗-envelope C ∗
property: there exists a completely isometric representation γ : V → C ∗
e (V ) such that
C ∗
e (V ) = C ∗(γ(V )) and if ρ is any other completely isometric representation, then there
exists an onto representation π : C ∗(ρ(V )) → C ∗(γ(V )) such that π(ρ(a)) = γ(a) for all
a ∈ V .
In this section we shall describe the Shilov boundary ideal for the closed subalgebra
of CF (Dn)q generated by the "holomorphic" generators πF,n(zi
In Pol(Matn)q consider the two-sided ideal Jn generated by
j), i, j = 1, . . . , n.
q2n−α−βzα
j (zβ
j )∗ − δαβ , α, β = 1, . . . , n,
n
Xj=1
where δαβ is the Kronecker symbol. The ideal Jn is a ∗-ideal, i.e. Jn = J ∗
n. The quo-
tient algebra Pol(S(Dn))q := Pol(Matn)q/Jn is a Uq sun,n-module ∗-algebra called the
polynomial algebra on the Shilov boundary of a quantum matrix ball. The canonical
homomorphism
jq : Pol(Matn)q → Pol(S(Dn))q
is a q-analog of the restriction operator which maps a polynomial on the ball Dn = {z ∈
Matn : zz∗ < 1} to its restriction to the Shilov boundary S(Dn) = {z ∈ Matn : zz∗ = 1}.
The ∗-algebra Pol(S(Dn))q was introduced by L.Vaksman in [25] and shown to be
isomorphic to the ∗-algebra (C[GLn]q, ∗) ≃ C[Un]q introduced in section 2.1; the isomor-
phism Ψ : Pol(S(Dn))q → (C[GLn]q, ∗) is given by zi
j, i, j = 1, . . . , n (see
[25, Theorem 2.2, Proposition 6.1]). The author of [25] used an algebraic approach to
introduce the q-analog of the Shilov boundary and posed a question whether this notion
would coincide with the "analytic" Shilov boundary of Arveson. In this section we shall
give an affirmative answer to that question for general value of n.
j + Jn → zi
Let A(Dn)q be the closed (non-involutive) subalgebra of CF (Dn)q generated by πF,n(zi
j),
i, j = 1, . . . , n. Let ¯Jn be the closure of Jn in CF (Dn)q and write jq also for the canonical
quotient map CF (Dn)q → CF (Dn)q/ ¯Jn.
We are now ready to state the main theorems of the paper.
Theorem 3 The ideal ¯Jn is a boundary ideal for A(Dn)q.
21
Theorem 4 The ideal ¯Jn is the Shilov boundary ideal for A(Dn)q.
We begin by proving several auxiliary lemmas.
Lemma 10 Let τ be a ∗-representation of Pol(Matn−1)q that annihilates the ideal Jn−1
of Pol(Matn−1)q. Then ((τ ◦ Πϕ) ⊗ π1 ⊗ π2) ◦ Dn is a ∗-representation of Pol(Matn)q that
annihilates the ideal Jn of Pol(Matn)q for any ∗-representations π1, π2 of C[SUn]q.
Proof. Write Φn : Pol(Matn)q → C[Un]q for the ∗-homomorphism given by
Φn : zi
j 7→ qi−nzi
j, i, j = 1, . . . , n,
(see Theorem 1 and the remark after it). It is straightforward to check that
Φn−1 ◦ Πϕ = Ψϕ ◦ Φn, ϕ ∈ [0, 2π],
where Ψϕ : C[Un]q → C[Un−1]q is the ∗-homomorphism such that
zi
j,
eiϕ,
0,
i, j < n,
i = j = n,
otherwise.
Ψϕ(zi
j) =
Therefore, as ker Φn = Jn, we have Πϕ(Jn) ⊂ Jn−1, giving that τ ◦ Πϕ annihilates the
ideal Jn whenever τ is a ∗-representation of Pol(Matn−1)q such that τ (Jn−1) = 0.
Next we observe that if Υ : C[Un]q → C[SUn]q is the canonical ∗-homomorphism given
by Υ : zi
j 7→ ti,j, then, for any a ∈ Pol(Matn)q,
(Υ ⊗ id ⊗ Υ) ◦ (∆ ⊗ id) ◦ ∆ ◦ Φn(a) = (σ ⊗ id) ◦ (id ⊗ σ) ◦ (id ⊗ id ⊗ θ) ◦ (Φn ⊗ id ⊗ id) ◦ Dn(a),
where σ is the flip map that sends a ⊗ b to b ⊗ a, ∆ is the comultiplication on C[Un]q
and θ : C[SUn]q → C[SUn]q is the ∗-automorphism defined by (18). Hence, (Φn ⊗ id ⊗
id) ◦ Dn(Jn) = 0. It is now immediate that if ρ is a ∗-representation of Pol(Matn)q that
annihilates Jn , then (ρ ⊗ π1 ⊗ π2) ◦ Dn(Jn) = 0 and, by the first part of the proof,
((τ ◦ Πϕ) ⊗ π1 ⊗ π2) ◦ Dn(Jn) = 0, for any τ , π1, π2 satisfying the assumptions of the
lemma.
The dilation technique will play a crucial role in the rest of the paper. Recall that if
V is a vector space and φi : V → B(Hi), i = 1, 2, are linear maps then φ2 is said to be
a dilation of φ1 on V and write φ1 ≺ φ2 if H1 ⊂ H2 and φ1(a) = PH1 φ2(a)H1 for a ∈ V .
Clearly, the relation ≺ on V is transitive.
The next statement shows that the Fock representation is a dilation of a "sum" of
coherent representations when considered as maps on C[Matn]q.
Lemma 11 There is a ∗-representation Ψ of Pol(Matn)q such that Ψ is a direct integral
of coherent representations ρn
ϕ and πF,n ≺ Ψ on C[Matn]q.
Proof. Let S and Cq be the shift and the diagonal operators on ℓ2(Z+) given by (22).
By Lemma 7,
πF,n(z1
1) = B ⊗ CqS + A11 ⊗ 1, πF,n(zi
j) = Aij ⊗ 1, (i, j) 6= (1, 1),
22
for some Aij, B ∈ B(ℓ2(Z+)⊗(n2−1)), while the images under the coherent representations
are:
ρn
ϕ(z1
1) = (−1)n−1eiϕB + A11, ρn
ϕ(zi
j) = Aij (i, j) 6= (1, 1).
As CqS is a contraction, by Sz.-Nagy's theorem (see e.g.
[16]), there exists a unitary
operator U on a larger space containing ℓ2(Z+) such that (CqS)k = Pℓ2(Z+)U kℓ2(Z+) for
all k ∈ N. Let
Ψ(z1
1) = B ⊗ U + A11 ⊗ 1, Ψ(zi
j) = Aij ⊗ 1 if (i, j) 6= (1, 1).
Clearly, πF,n ≺ Ψ on C[Matn]q and Ψ is a ∗-representation of Pol(Matn)q which is a direct
integral of those ρn
ϕ such that (−1)n−1eiϕ is in the spectrum of U .
Lemma 12
1. Any ∗-representation of Pol(Matn)q that annihilates the ideal Jn is a
7→ eiϕk qk−nρ(tk,l), k, l = 1, . . . , n,
direct integral of ∗-representations given by zk
l
where ρ is a ∗-representation of C[SUn]q and ϕk ∈ [0, 2π).
2. If π is a ∗-representation of Pol(Matn)q such that π(Jn) = 0 then kπ(a)k ≤ kπF,n(a)k,
a ∈ Pol(Matn)q, and
kj(k)
q
((πF,n(ai,j))i,j)k = sup{k(π(ai,j ))i,jk : π ∈ Rep(Pol(Matn)q), π(Jn) = 0} (36)
for any k ∈ N and (ai,j)i,j ∈ Mk(Pol(Matn)q).
Proof. 1. Let π be a ∗-representation of Pol(Matn)q that annihilates Jn. By [25, Proposi-
tion 6.1, Theorem 2.2], the family {π(zi
j ) : i, j = 1, . . . , n} determines a ∗-representation of
(C[GLn]q, ∗). Since detq z is central in (C[GLn]q, ∗) and satisfies detq z(detq z)∗ = q−n(n−1)
(see (9)), by the spectral theorem
[0,2π]
π(detq z) =Z ⊕
π =Z ⊕
[0,2π]
eiϕq−n(n−1)/2Iϕdµ(ϕ)
πϕdµ(ϕ),
and
where {πϕ : ϕ ∈ [0, 2π)} is a field of ∗-representations of Pol(Matn)q such that πϕ(detq z) =
eiϕq−n(n−1)/2Iϕ with Iϕ being the identity operator on the representation space of πϕ.
l ) = eiϕk qk−nzk
Fix now ϕ ∈ [0, 2π) and let ι : (C[GLn]q, ∗) → C[Un]q be the ∗-isomorphism given
l , k, l = 1, . . . , n, where ϕ1 + . . . + ϕn = ϕ. Then πϕ ◦ ι−1 is a
by ι(zk
∗-representation of C[Un]q such that πϕ ◦ ι−1(detq z) = I and hence ρϕ := πϕ ◦ ι−1 ◦ j−1 is
a ∗-representation of C[SUn]q ≃j C[Un]q/hdetq z − 1i, where j(zk
l ) = tk,l, k, l = 1, . . . , n.
We obtain
πϕ(zk
l ) = ρϕ ◦ j ◦ ι(zk
l ) = eiϕk qk−nρϕ(tk,l), k, l = 1, . . . , n.
2. Let χϕ, ϕ = (ϕ1, . . . , ϕn), be the one dimensional representation of Pol(Matn)q
defined in Lemma 6. By the lemma the mapping ψ : πF,n(a) 7→ χϕ(a), a ∈ Pol(Matn)q ex-
tends to a ∗-homomorphism from CF (Dn)q to C. Given representations π1, π2 of C[SUn]q
and a ∈ Pol(Matn)q, by Lemma 8, we obtain
k(χϕ ⊗ π1 ⊗ π2) ◦ Dn(a)k = k(ψ ⊗ id ⊗ id)((πF,n ⊗ π1 ⊗ π2) ◦ Dn(a))k
(37)
≤ k(πF,n ⊗ π1 ⊗ π2) ◦ Dn(a)k ≤ kπF,n(a)k.
23
If π2 is the one-dimensional representation given by π2(tk,l) = δkl, k, l = 1, . . . , n, then
(χϕ ⊗ π1 ⊗ π2) ◦ Dn(zk
l ) =
n
Xa,b=1
χϕ(za
b ) ⊗ π1(tb,l) ⊗ π2(ta,k) = eiϕk qk−nπ1(tk,l).
Hence, applying the first statement of the lemma and (37) we get
kπ(a)k ≤ kπF,n(a)k, a ∈ Pol(Matn)q
for any ∗-representation π of Pol(Matn)q such that π(Jn) = 0.
The equality (36) holds by the fact that kj(k)
((πF,n(ai,j))i,j)k is the supremum over
k(π(ai,j)i,jk, where π runs over all ∗-representations of Pol(Matn)q that annihilate the
ideal Jn and such that kπ(x)k ≤ kπF,n(x)k, x ∈ Pol(Matn)q.
q
We are now in a position to prove the main theorems.
Proof of Theorem 3.
Since jq is a ∗-homomorphism between C ∗-algebras, it is a complete contraction. To
see that jq is a complete isometry when restricted to A(Dn)q it is enough to prove that
(πF,n(ai,j))i,j ≤ j(k)
q
((πF,n(ai,j))i,j)
for any (ai,j)i,j ∈ Mk(A(Dn)q).
The strategy is to find a ∗-representation Π : Pol(Matn)q → B(K), K ⊃ HF,n such
that Π(Jn) = 0 and
πF,n(a) = PHF,nΠ(a)HF,n, a ∈ C[Matn]q,
(38)
i.e. πF,n ≺ Π when considered as maps on C[Matn]q. As in this case
k(πF,n(ai,j))i,jk ≤ k(Π(ai,j))i,jk, (ai,j )i,j ∈ Mk(C[Matn]q), k ∈ N,
Lemma 12 would give immediately the statement.
We proceed by induction. If n = 1 this was proved in [26]. For the reader's convenience
we reproduce the arguments. We have that Pol(C)q is generated by a single element z
subject to the relation z∗z = q2zz∗ + (1 − q2) and πF,1(z) = CqS, a contraction (see
e.g. the proof of Lemma 6). By Sz.-Nagy's theorem there exist a Hilbert space K ⊃
HF,1 and a unitary operator U ∈ B(K) such that πF,1(p(z)) = PHF,1p(U )HF,1 for all
holomorphic polynomials p. As the map z 7→ U extends to a ∗-representation of Pol(C)q
that annihilates the ideal J1 = hzz∗ − 1i, we obtain a necessary dilation.
Assume (38) holds for some n ≥ 1. By Lemma 11 there exists a ∗-representation Ψ of
and such that
Pol(Matn+1)q, which is a direct integral of coherent representations ρn+1
πF,n+1 ≺ Ψ on C[Matn+1]q.
ϕ
By the remark after Lemma 9, ρn+1
ϕ
is a ∗-subrepresentation of τϕ,w,w and hence
ρn+1
ϕ (a) = PLτϕ,w,w(a)L, a ∈ Pol(Matn+1)q, for a subspace L.
As τϕ,w,w = ((πF,n ◦ Πϕ) ⊗ πw ⊗ πw) ◦ Dn+1 and by induction πF,n ≺ Π on C[Matn]q
with Π(Jn) = 0 we obtain
τϕ,w,w ≺ ((Π ◦ Πϕ) ⊗ πw ⊗ πw) ◦ Dn+1 on C[Matn+1]q.
24
Finally, by Lemma 10 ((Π◦Πϕ)⊗πw⊗πw)◦Dn+1 annihilates the ideal Jn+1 of Pol(Matn+1)q.
Combining all these steps and using transitivity of the relation ≺ we obtain the desired
statement.
Proof of Theorem 4.
Assume that I is a boundary ideal for A(Dn)q with I ⊃ ¯Jn and identify Pol(Matn)q
with its image under the Fock representation. Let a ∈ C[Matn]q ⊂ Pol(Matn)q be a
polynomial in "holomorphic" generators zi
j, 1 ≤ i, j ≤ n.
By assumption,
ka + Ik = kak = ka + ¯Jnk.
(39)
Let x ∈ Pol(Matn)q. As, by (8) and (9),
(zi
j)∗ + Jn = (−q)i+j−2n(detq z)−1 detq zi
j + Jn,
detq z + Jn is a central element in Pol(Matn)q/Jn and
(detq z)∗ detq z + Jn = q−n(n−1) + Jn,
there exist k ∈ Z+ and a ∈ C[Matn]q such that x + ¯Jn = (detq z)−ka + ¯Jn. Hence, by (39)
and the fact that I ⊃ ¯Jn, we obtain
kx + ¯Jnk2 = k(x∗ + ¯Jn)(x + ¯Jn)k
= ka∗((detq z)−k)∗(detq z)−ka + ¯Jnk
= ka∗qkn(n−1)a + ¯Jnk = kqkn(n−1)/2a + ¯Jnk2
= kqkn(n−1)/2a + Ik2 = ka∗qkn(n−1)a + Ik
= ka∗((detq z)−k)∗(detq z)−ka + Ik
= kx + Ik2.
This implies that CF (Dn)q/ ¯Jn = CF (Dn)q/I and hence ¯Jn = I.
Let C(Un)q be the C ∗-enveloping algebra of C[Un]q. As the matrix (zi
j)i,j formed by
the generators of C[Un]q is unitary one has that the norm of each generator is not larger
than 1 in each ∗-representation by bounded operators on a Hilbert space and hence the
C ∗-enveloping algebra is well-defined.
Corollary 3 The C ∗-envelope C ∗
e (A(Dn)q) is isomorphic to C(Un)q.
Let ψ : Pol(Matn)q → (C[GLn]q, ∗) be the surjective ∗-homomorphism from
Proof.
l ) = qk−nzk
Theorem 1 and ι : (C[GLn]q, ∗) → C[Un]q the ∗-isomorphism given by ι(zk
l ,
k, l = 1, . . . , n. As ker ψ = Jn, any ∗-representation of Pol(Matn)q such that π(Jn) = 0
is given by π(a) = ρ(ι ◦ ψ(a)), a ∈ Pol(Matn)q, for some ∗-representation ρ of C[Un]q.
Moreover, the correspondence π ↔ ρ is one-to-one. Consider a ∗-representation ρ of
C[Un]q such that ρ(C[Un]q) ≃ C(Un)q. In what follows we identify the latter two algebras.
Let Ψ : πF,n(Pol(Matn)q) → C(Un)q, πF,n(a) 7→ ρ(ι ◦ ψ(a)). By Lemma 12,
kρ(ι ◦ ψ(a))k ≤ kπF,n(a)k, a ∈ Pol(Matn)q
25
and hence Ψ extends to a surjective ∗-homomorphism (denoted by the same letter) from
CF (Dn)q to C(Un)q. As any representation of C(Un)q gives rise to a representation of
CF (Dn)q that annihilates the ideal Jn, we have ker Ψ = ¯Jn and hence
CF (Dn)q/ ¯Jn ≃ C(Un)q.
As ¯Jn is the Shilov boundary ideal of A(Dn)q, CF (Dn)q/ ¯Jn ≃ C ∗
statement.
e (A(Dn)q) giving the
Acknowledgements. We would like to thank Daniil Proskurin for valuable conversa-
tions during the preparation of this paper. We are grateful to the referee for a number of
suggestions that improved the exposition.
References
[1] W. Arveson, Subalgebras of C∗-algebras, Acta Math. 123 (1969), 122 -- 141
[2] W. Arveson, Notes on the unique extension property, Unpublished, 2003 Available
from http://math.berkley.edu/∼arveson/Dvi/unExt.pdf
[3] G. Bergman, The diamond lemma for ring theory, Adv. Math., 29 (1978), 178-218
[4] O. Bershtein, Degenerate principal series of quantum Harish-Chandra modules,
J.Math.Phys. 45 (2004), no. 10 3800-3827
[5] K.A. Brown, K.R. Goodearl, Lectures on Algebraic Quantum Groups, Birkhau-
ser, 2002
[6] V. Chari, A. Pressley, A Guide to Quantum Groups, Cambridge Univ. Press,
1995
[7] K. Davidson, C ∗-algebras by Examples, American Mathematical Society, Provi-
dence, RI 6, Fields Institute Monographs, 1996
[8] M.A. Dritschel, S.A. McCullough, Boundary Representations for families of
representations of operator algebras and spaces, J.Operator Theory 53 (2005), 159-
167
[9] O. Giselsson, The universal C ∗-algebra of the quantum matrix ball and its irre-
ducible ∗-representations, arXiv:1801.10608
[10] M. Hamana, Injective envelopes of operator systems, Publ.RIMS Kyoto Univ.
(1979), v.15, 773 -- 785
[11] J.C. Jantzen, Lectures in Quantum Groups, Graduate Studies in Mathematics,
American Mathematical Society, Providence, RI, 1996
[12] P.E.T. Jørgensen, L.M. Schmitt, R.F. Werner, Positive representations of
general commutation relations allowing Wick ordering, J. Funct. Anal. 134 (1995),
no. 1, 33-99
[13] A. Klimyk, K. Schmudgen, Quantum Groups and Their Representations, Sprin-
ger-Verlag, 1997
26
[14] H.T. Koelnik, On ∗-representations of the Hopf ∗-algebra associated with quantum
group Uq(n), Compositio Math., 77 (1991), 199-231
[15] L.I. Korogodski, Y.S. Soibelman, Algebra of Functions on Quantum Groups:
Part I, American Mathematical Society, Providence, RI, 1998
[16] V. Paulsen, Completely bounded maps and operator algebras, Cambridge University
Press, 78 (2003)
[17] D. Proskurin, Yu.S. Samoilenko, Stability of the C ∗-algebra associated with
twisted CCR, Algebr. Represent. Theory 5 (2002), no. 4, 433-444
[18] D. Proskurin, L. Turowska, On the C ∗-algebra associated with Pol(Mat2,2)q,
Methods Funct. Anal. Topology 7 (2001), no. 1, 88 -- 92
[19] D. Proskurin, L. Turowska, Shilov boundary for "holomorphic functions" on a
quantum matrix ball, Algebr. Represent. Theory 18 (2015), no. 4, 917-929
[20] W. Pusz, S.L. Woronowicz, Twisted second quantization, Rep. Math. Phys. 27
(1989), no. 2, 231-257
[21] D. Shklyarov, S. Sinelshchikov, L. Vaksman, Fock representations and quan-
tum matrices, Internat. J. Math. 15, (2004), no. 9, 855 -- 894
[22] S. Sinel'shchikov, L. Vaksman, On q-analogues of bounded symmetric domains
and Dolbeault complexes, Mat. Fiz. Anal. Geom. 1 (1998), 75 -- 100
[23] Ya.S. Soibelman, Irreducible representations of the algebra of functions on the
quantum group SU(n) and Schubert cells. (Russian), Dokl. Akad. Nauk SSSR 307
(1989), no. 1, 41 -- 45; translation in Soviet Math. Dokl. 40 (1990), no. 1, 34-38
[24] Ya.S. Soibelman, Algebra of functions on a compact quantum group and its rep-
resentations. (Russian), Algebra i Analiz 2 (1990), no. 1, 190 -- 212; translation in
Leningrad Math. J. 2 (1991), no. 1, 161-178
[25] L. Vaksman, Quantum matrix ball: the Cauchy-Szego kernel and the Shilov bound-
ary, Mat. Fiz. Anal. Geom. 8 (2001), no. 4, 366-384
[26] L. Vaksman, Maximum principle for "holomorphic functions" in the quantum ball,
Mat. Fiz. Anal. Geom. 10 (2003), no. 1, 12-28
[27] L. Vaksman, Quantum Bounded Symmetric Domains, American Mathematical So-
ciety, Providence, RI, 2010
[28] L.L. Vaksman, Ya.S. Soibelman, Algebra of functions on the quantum group
SU (n + 1), and odd-dimensional quantum spheres. (Russian), Algebra i Analiz 2
(1990), no. 5, 101 -- 120; translation in Leningrad Math. J. 2 (1991), no. 5, 1023-1042
[29] J. Wolf, The action of a real semisimple Lie group on a complex flag manifold I.
Orbit structure and holomorphic arc components, Bull. Amer. Math.Soc. 75 (1969),
1121-1237
27
|
1209.5838 | 1 | 1209 | 2012-09-26T06:21:16 | $E_0$-semigroups: around and beyond Arveson's work | [
"math.OA"
] | We give an account of the theory of $E_0$-semigroups. We first focus on Arveson's contributions to the field and related results. Then we present the recent development of type II and type III $E_0$-semigroups. We also include a short note in Appendix, based on Arveson's observation, on noncommutative Poisson boundaries. | math.OA | math |
E0-SEMIGROUPS: AROUND AND BEYOND ARVESON'S WORK
MASAKI IZUMI
Dedicated to the memory of Bill Arveson
Abstract. We give an account of the theory of E0-semigroups. We first focus
on Arveson's contributions to the field and related results. Then we present the
recent development of type II and type III E0-semigroups. We also include a short
note in Appendix, based on Arveson's observation, on noncommutative Poisson
boundaries.
1. Introduction
Arveson opened up new frontiers in several areas in operator algebras, or more
generally noncommutative analysis. He often provided essential ideas at the very
beginning of new subjects, and then generously left further development to others,
as probably seen from Kenneth Davidson's article, also in this volume. One specific
example is the theory of E0-semigroups, the main subject of this article, and the
author is one of Arveson's followers who has luckily obtained substantial benefit
from his ideas. These ideas were summarized as a monograph [23] by himself, which
is the standard reference for E0-semigroups. Two main subjects of the monograph,
among others, are product systems (infinite tensor product systems of Hilbert spaces)
and the dilation theory of semigroups of completely positive maps, which reflects the
importance of these subjects in the field well. Therefore it is natural for us to discuss
them in this article too.
Definition 1.1. An E0-semigroup α = {αt}t≥0 is a family of unital normal ∗-
endomorphisms of a von Neumann algebra M satisfying the following conditions:
(i) α0 = id,
(ii) αs ◦ αt = αs+t for all s, t ≥ 0,
(iii) For every a ∈ M and every normal functional ϕ ∈ M∗, the function [0,∞) ∋
t 7→ ϕ(αt(a)) ∈ C is continuous.
Throughout the article, we concentrate on E0-semigroups acting on B(H), the set
of bounded operators of a separable infinite dimensional complex Hilbert space H.
In this particular case the normality condition of αt is redundant (see [70, Chapter
V, Theorem 5.1]). Since B(H) is believed to be the easiest infinite dimensional
factor, one may wonder if something significant comes out of such objects. Yet, it
turned out that E0-semigroups on B(H) are already rich and difficult, and they have
deep connections with various areas in mathematics such as probability theory and
classical analysis (not to mention functional analysis).
1
2
MASAKI IZUMI
Soon after Powers [56] initiated the systematic analysis of E0-semigroups in the
late 80's, Arveson [5] introduced the notion of a product system, which is the key
concept for later development of the theory.
In particular, it inspired the proba-
bilistic approach of Tsirelson [72]. Arveson [6],[9] showed that there is a one-to-one
correspondence between the cocycle conjugacy classes of E0-semigroups and the iso-
morphism classes of product systems, which reduced the classification problem of
the former to that of the latter. Then he completely classified so-called type I E0-
semigroups by an invariant called index, which is the first substantial classification
result in the theory. There are a lot of other important invariants for E0-semigroups
defined via the corresponding product systems now.
In general, it is not so easy to construct new examples of E0-semigroups. However,
it is relatively easy to construct semigroups of unital completely positive (ucp) maps,
and they give rise to E0-semigroups through Bhat's dilation theorem [29]. This
approach was extensively developed by Arveson [18] and Powers [59], [60], and it is
one of the most important constructions of E0-semigroups.
For the proofs and related results of the statements we present in this article, the
reader is referred to Arveson's monograph [23]. Another standard reference of E0-
semigroups is the conference proceedings [64] with several valuable articles, including
Arveson's survey [24], which appeared at the right time when the field sufficiently
matured. [24] can be read as an introduction to [23].
This article ends with a short note in Appendix on noncommutative Poisson bound-
aries, a notion introduced by the author for normal ucp maps acting on von Neumann
algebras. At the occasion of a workshop taking place at the Fields Institute in 2007,
Arveson briefly mentioned to the author that the noncommutative Poisson boundary
of a normal ucp map is identified with the fixed point algebra of the minimal dilation
of the given normal ucp map. This observation has potentially useful consequences,
and we include them.
2. Around Arveson's work
2.1. Basic equivalence relations. We first introduce two basic equivalence rela-
tions of E0-semigroups, conjugacy and cocycle conjugacy. One of the main goals in
the theory of E0-semigroups is to classify them up to cocycle conjugacy.
Definition 2.1. Let α and β be E0-semigroups acting on B(H) and B(K) respec-
tively.
(i) We say that α and β are conjugate if there exists a unitary V from H onto
K satisfying Ad V ◦ αt = βt ◦ Ad V for all t ≥ 0, where Ad V (A) = V AV ∗.
(ii) We say that a weakly continuous family of unitaries U = {Ut}t≥0 in B(H)
is an α-cocycle if they satisfy the 1-cocycle relation Us+t = Usαt(Ut) for all
s, t ≥ 0. When U is an α-cocycle, we have a new E0-semigroup defined by
αU = {Ad Ut ◦ αt}t≥0, which we call the perturbation of α by U.
(iii) We say that α and β are cocycle conjugate if a cocycle perturbation of α is
conjugate to β.
E0-SEMIGROUPS
3
To see the difference of the two equivalence relations above, we start with the
trivial class, namely the class of E0-semigroups α with αt ∈ Aut(B(H)) for all t ≥ 0.
In this case, a classical result due to E. Wigner says that there exists a (possibly
unbounded) self-adjoint operator A, uniquely determined up to additive constant,
satisfying αt = Ad eitA. Therefore such α is always in the cocycle conjugacy class
of id, the semigroup consisting of only the identity map, while the classification of
such α up to conjugacy is equivalent to the classification of A up to additive constant
and unitary equivalence. Therefore we see that the classification up to conjugacy is
already complicated in the trivial case, and we cannot expect reasonable classification
results for more general classes of E0-semigroups.
There is another reason why cocycle conjugacy is so natural. Let α be an E0-
semigroup acting on B(H). The generator δ of α is defined by the limit
lim
t→+0
1
t
(αt(A) − A),
in the strong operator topology, where the domain of δ is the set of A ∈ B(H) for
which the limit exists. Let D ∈ B(H) be a self-adjoint operator, and let δ′(A) =
δ(A) + i[D, A]. Then δ′ generates an E0-semigroup etδ′ that is a cocycle perturbation
of α.
2.2. Product systems. Before introducing a product system, we first discuss the
Hilbert space consisting of the intertwining operators for a unital endomorphism ρ of
B(H). Recall that H is a separable infinite dimensional complex Hilbert space. The
endomorphism ρ, regarded as a normal representation of B(H) on H, is unitarily
equivalent to a direct sum of copies of the identity representation of B(H). This
means that there exists an orthogonal decomposition H = ⊕i∈IHi with ρ(B(H))-
invariant closed subspaces Hi ⊂ H such that the restriction of ρ to Hi is unitar-
ily equivalent to the identity representation of B(H). Thus there exist isometries
Vi ∈ B(H) whose range is Hi, satisfying ViA = ρ(A)Vi for all A ∈ B(H), and in
consequence
(2.1)
ρ(A) =Xi∈I
ViAV ∗
i ,
∀A ∈ B(H),
where the right-hand side converges in the strong operator topology.
Let Hρ be the space of intertwining operators between id and ρ, that is
Hρ = {V ∈ B(H); V A = ρ(A)V, ∀A ∈ B(H)}.
For V, W ∈ Hρ, the product W ∗V is a scalar as it belongs to the center of B(H).
Equipped with inner product hV, WiHρ1 = W ∗V , the intertwiner space Hρ is a Hilbert
space with an orthonormal basis {Vi}i∈I. It is easy to see that the right-hand side of
(2.1) does not depend on the choice of the orthonormal basis {Vi}i∈I of Hρ. In fact
for an arbitrary orthonormal system {Vi}i∈I of Hρ, the equation (2.1) characterizes
its completeness.
4
MASAKI IZUMI
Now we consider two unital endomorphisms ρ, σ ∈ B(H). Then we have an inclu-
sion relation Hρ · Hσ ⊂ Hρ◦σ, and for V, V ′ ∈ Hρ and W, W ′ ∈ Hσ,
hV W, V ′W ′iHρ◦σ = W ′∗V ′∗V W = hV, V ′iHρW ′∗W ′ = hV, V ′iHρhW, W ′iHσ
= hV ⊗ W, V ′ ⊗ W ′iHρ⊗Hσ.
Moreover if {Vi}i∈I and {Wj}j∈J are orthonormal bases of Hρ and Hσ respectively,
then {ViWj}i∈I,j∈J is an orthonormal basis of Hρ◦σ as we have
ρ ◦ σ(A) = Xi∈I,j∈J
ViWjAW ∗
j V ∗
i .
This means that we can identify Hρ◦σ with the Hilbert space tensor product Hρ⊗Hσ
under the identification of the product V W in B(H) and the simple tensor V ⊗ W
in Hρ ⊗ Hσ.
We get back to our original situation and consider an E0-semigroup α acting on
B(H). Then we have a 1-parameter family of Hilbert spaces Eα(t) := Hαt for t > 0,
with identification Eα(s + t) = Eα(s) ⊗ Eα(t), where the usual product in B(H)
corresponds to tensor product. Moreover the association (0,∞) ∋ t 7→ Eα(t) should
be measurable (or more strongly continuous) in an appropriate sense. Arveson [5]
axiomatized this situation and introduced the notion of product systems.
Definition 2.2. A product system is a family of separable Hilbert spaces p : E →
(0,∞) over the half-line (0,∞), with fiber Hilbert spaces E(t) = p−1(t), endowed
with a bilinear associative multiplication E(s) × E(t) ∋ (x, y) 7→ xy ∈ E(s + t)
satisfying the following conditions:
(i) hxu, yvi = hx, yihu, vi for all x, y ∈ E(s), u, v ∈ E(t),
(ii) The linear span of E(s) · E(t) is dense in E(s + t),
(iii) E has the structure of a standard Borel space that is compatible with the
projection p : E → (0,∞), multiplication, the vector space operations, and
the inner product. Moreover, there exist a separable infinite dimensional
Hilbert space H0 and a Borel isomorphism from E onto (0,∞)×H0 compatible
with the projection p.
Remark 2.3. (1) The measurability condition (iii) is equivalent to the following con-
dition: there exists a countable family of cross sections {ξn(t)}n∈N of p such that the
linear span of {ξn(t)}n∈N is dense in E(t) for all t and the functions (0,∞) ∋ t 7→
hξm(t), ξn(t)i and (0,∞)2 ∋ (s, t) 7→ hξm(s)ξn(t), ξl(s + t)i are Borel measurable for
all m, n, l. See [47, Lemma 7.39] for the proof.
(2) V. Liebscher [47, Corollary 7.16] showed that for a given E satisfying (i) and (ii),
there exists at most one Borel structure satisfying (iii). This justifies the following
definition of isomorphisms of product systems: an isomorphism θ from E to F is a
family {θt}t>0 of unitaries θt : E(t) → F (t) satisfying θs(x)θt(y) = θs+t(xy) for all
x ∈ E(s) and y ∈ E(t).
In what follows, we always consider the Borel structure of B(H) given by the weak
operator topology. Then B(H) is a standard Borel space with this Borel structure.
When α is an E0-semigroup, Arveson [5] showed that
E0-SEMIGROUPS
5
Eα = {(t, T ) ∈ (0,∞) × B(H); T ∈ Eα(t)}
is a product system with p(t, T ) = t. We will often identify p−1(t) with Eα(t) and
(t, T ) with T .
Cocycle conjugate E0-semigroups give isomorphic product systems. Indeed, it is
obvious that conjugate E0-semigroups give rise to isomorphic product systems. If U
is an α-cocycle and αU is the cocycle perturbation of α by U, then the family of maps
Eα(t) ∋ T 7→ UtT ∈ EαU (t) gives an isomorphism of Eα and EαU .
Arveson [5] showed that the converse is also true.
Theorem 2.4. Two E0-semigroups α and β are cocycle conjugate if and only the
corresponding product systems Eα and Eβ are isomorphic.
To sketch the proof of the other implication, we assume that both α and β act on
B(H) for simplicity, and assume that Eα and Eβ are isomorphic by θt : Eα(t) → Eβ(t).
We choose an arbitrary orthonormal basis {Vn}∞
n=1 of Eα(t) and set
θt(Vn)V ∗
n ,
Ut =
∞Xn=1
which converges to a unitary operator in the strong operator topology. Then Ut is
independent of the choice of {Vn}∞
n=1, which enables us to show that {Ut}t>0 satisfies
the 1-cocycle relation. Moreover the condition (iii) implies that the map (0,∞) ∋
t 7→ Ut is Borel. Now it follows from a standard trick that {Ut}t>0 is continuous, and
it is an α-cocycle. By construction, we have θt(V ) = UtV for all V ∈ Eα(t) and β is
the cocycle perturbation of α by U thanks to (2.1).
More strongly, Arveson [9] showed that the association α 7→ Eα induces a one-to-
one correspondence between the set of cocycle conjugacy classes of E0-semigroups and
the set of isomorphism classes of product systems. The only issue now is surjectivity
of α 7→ Eα.
Theorem 2.5. For any product system E, there exists an E0-semigroup α whose
product system Eα is isomorphic to E.
Arveson's original proof is really involved and it was the only proof for a while
(see [25], [47], [66], [67] for simpler proofs). In order to prove the theorem, Arveson
developed the representation theory of product systems, which is interesting in its
own right.
A representation φ of a product system E is a Borel map π : E → B(H) satisfying
φ(x)φ(y) = φ(xy) for all x ∈ E(s), y ∈ E(t), and φ(v)∗φ(u)1 = hu, vi for all u, v ∈
E(t). Arveson constructed the regular representation of E by analogy with the regular
representation of a locally compact group. If φ is a representation of E, then the
following formula with an orthonormal basis {en}∞
n=1 of E(t),
αt(A) =
φ(en)Aφ(en)∗, A ∈ B(H),
∞Xn=1
6
MASAKI IZUMI
does not depend on the particular choice of {en}∞
n=1, and α = {αt}t>0 is a semigroup
of endomorphisms of B(H) with appropriate continuity.
In order to construct an
E0-semigroup whose product system is isomorphic to E, the only problem is that
αt may not be unital. The representations with αt being unital are called essential
representations. As in the case of locally compact groups, Arveson introduced the
spectral C ∗-algebra C ∗(E) of E having the universal property with respect to the
representations of E, and then he constructed a state of C ∗(E) giving rise to an
essential representation of E through the GNS construction.
Arveson [7] showed that C ∗(E) is a nuclear C ∗-algebra for any product system E.
For the structure of C ∗(E), see [38], [39], [77], [78].
For attempts to generalize product systems to those for Hilbert W ∗ and C ∗-
modules, see, for example, [2], [28], [36], [52], [53], [68].
2.3. CCR flows. The most fundamental examples of E0-semigroups are CCR flows
acting on B(H), where H is the symmetric Fock space over the test function space
L2((0,∞), K).
by eG the symmetric Fock space
For a complex Hilbert space G, which will be L2((0,∞), K) for later use, we denote
eG =
∞Mn=0
Gn,
where Gn is the n-fold symmetric tensor product of G, and G0 is interpreted as the
1-dimensional space spanned by a unit vector Ω, called the vacuum. The exponential
vector exp(f ) ∈ eG for f ∈ G is defined by
∞Xn=0
1
√n!
exp(f ) =
f ⊗n,
and we have hexp(f ), exp(g)i = ehf,gi. The set of exponential vectors form an inde-
pendent and total subset of G.
When G is decomposed into the direct sum of two closed subspaces G1 and G2, we
have
hexp(f1 ⊕ f2), exp(g1 ⊕ g2)i = ehf1⊕f2,g1⊕g2i = ehf1,g1i+hf2,g2i
= hexp(f1), exp(g1)ihexp(f2), exp(g2)i
= hexp(f1) ⊗ exp(f2), exp(g1) ⊗ exp(g2)i,
for f1, g1 ∈ G1 and f2, g2 ∈ G2. This shows that the map
exp(f1 ⊕ f2) 7→ exp(f1) ⊗ exp(f2)
extends to a unitary from eG1⊕G2 onto eG1 ⊗ eG2. Therefore forming the symmetric
Fock space is a functor from the category of Hilbert spaces into itself transforming
In what follows we always identify eG1⊕G2 with
direct sums into tensor products.
eG1 ⊗ eG2.
E0-SEMIGROUPS
7
We denote by W (f ) ∈ B(eG) the Weyl operator for f ∈ G, which is the unitary
operator defined by
W (f ) exp(g) = e− 1
2 kf k2−hg,f i exp(g + f ).
The Weyl operators satisfy the canonical commutation relation in the Weyl form
W (f )W (g) = eiℑhf,giW (f + g),
and their linear span is a dense ∗-subalgebra of B(eG) in the weak operator topology.
Now we specify the test function space G to be the set of square integrable functions
L2((0,∞), K) on the half-line with values in a complex Hilbert space K, called the
multiplicity space. We denote by S = {St}t≥0 the (forward) shift semigroup acting
on L2((0,∞), K):
Stf (x) =(cid:26) 0,
f (x − t),
for 0 < x < t
for
t ≤ x
.
Definition 2.6. There exists a unique E0-semigroup αK acting on B(H) with H =
eL2((0,∞),K)) satisfying
αK
∀f ∈ L2((0,∞), K).
t (W (f )) = W (Stf ),
We call αK the CCR flow of rank dim K.
Remark 2.7. The CAR flows are defined in the same way except for replacing the
symmetric Fock space with the antisymmetric Fock space. Powers-Robinson [63]
showed that the CCR flow and CAR flow of the same rank are conjugate.
Arveson [5] identified the product systems corresponding to the CCR flows. For
0 ≤ a < b ≤ ∞, we regard L2((a, b), K) as a closed subspace of L2((0,∞), K) in
a natural way, and eL2((a,b),K) as a closed subspace of H = eL2((0,∞),K) generated
by {exp(f )}f ∈L2((a,b),K). We denote by Γ(St) the isometry in B(H) determined by
Γ(St) exp(f ) = exp(Stf ) for all f ∈ L2((0,∞), K). For t > 0, we set EK(t) =
eL2((0,t),K) ⊂ H. Then
is a product system with p(t, ξ) = t and multiplication
EK = {(t, ξ) ∈ (0,∞) × H; ξ ∈ EK(t)}
where we use the following identification
(s, ξ) · (t, η) = (s + t, ξ ⊗ Γ(Ss)η),
eL2((0,s+t),K) = eL2((0,s),K)⊕SsL2((0,t),K) = eL2((0,s),K) ⊗ Γ(Ss)eL2((0,t),K).
We call EK the exponential product system, which is isomorphic to the product
system EαK associated with the CCR flow αK, via the representation φ : EK → B(H)
given by φ((t, ξ))η = ξ ⊗ Γ(St)η for η ∈ H.
Remark 2.8. The exponential product systems can be interpreted as product systems
associated with white noise. For simplicity, we consider the case K = C. Let {Bt}t≥0
be the standard Brownian motion defined on the probability space (Ω,F , W ). For
0 ≤ s < t ≤ ∞, we denote by Fs,t the σ-algebra generated by the increments
Bv − Bu for all s ≤ u < v ≤ t. We may assume F = F0,∞. Then the well-known
8
MASAKI IZUMI
Wiener-Ito chaos decomposition (see [51, Chapter IV]) says that the nested system
of subspaces {L2(Fs,t)}0≤s<t≤∞ of L2(Ω,F , W ) is identified with that of subspaces
{eL2(s,t)}0≤s<t≤∞ of eL2(0,∞), and the identification goes along with the tensor product
factorizations L2(Fr,t) = L2(Fr,s)⊗ L2(Fs,t) and eL2(r,t) = eL2(r,s) ⊗ eL2(s,t) for r < s <
t. Moreover, the time shift of the Brownian motion induces an isometry from L2(F0,∞)
onto L2(Ft,∞). Therefore we can completely describe the exponential product system
in terms of so called white noise, which consists of (Ω,F , W ), {Fs,t}0≤s<t≤∞, and the
time shift (strictly speaking, white noise is the two-sided version of it). White noise is
only a special example of Tsirelson's notion of noises, and this interpretation opened
up Tsirelson's probabilistic approach to E0-semigroups (see Section 3).
2.4. Index. It is natural to ask whether one can distinguish the CCR flows with
different ranks up to cocycle conjugacy. To answer the question in the positive, we
need an isomorphism invariant for product systems, and the first such invariant was
provided by Arveson [5].
Definition 2.9. Let E be a product system. A unit of E is a non-zero measurable
section (0,∞) ∋ t 7→ u(t) ∈ E(t) that is multiplicative,
We denote by UE the set of units of E.
u(s + t) = u(s)u(t),
∀s, t > 0.
For the product system Eα associated with an E0-semigroup α, a unit is nothing
but a continuous semigroup V = {Vt}t≥0 of isometries, up to normalization, satisfying
the intertwining property VtA = αt(A)Vt. We say that α is spatial if UEα is not empty.
The multiplicative property of units implies that for u, v ∈ UE, there exists a unique
complex number cE(u, v) satisfying hu(t), v(t)i = etcE (u,v) for all t > 0. For each fixed
t > 0, the function etcE (u,v) on UE × UE is positive definite by definition, and the
Schoenberg theorem shows that cE is conditionally positive definite. Let C0UE be
the set of functions ξ : UE → C with finite support and Pu∈UE
ξ(u) = 0. Then cE
being conditionally positive definite means
∀ξ ∈ C0UE.
cE(u, v)ξ(u)ξ(v) ≥ 0,
Xu,v
Therefore cE gives a positive semi-definite inner product of C0UE, and we get a sep-
arable Hilbert space, denoted by H(UE, cE), by the usual procedure. The dimension
of this Hilbert space is an isomorphism invariant of the product system E. Note
that H(UE, cE) = {0} is possible, not as in the case of positive definite functions.
Intuitively dim H(UE, cE) is " dimUE" − 1.
Definition 2.10. The index of a product system E with UE 6= ∅ is
ind(E) = dim H(UE, cE).
The index of a spatial E0-semigroup α is ind(α) = ind(Eα).
Remark 2.11. The first attempt to introduce a numerical invariant for E0-semigroups
was made by Powers [56]. To define his index, he constructed what is now called the
boundary representation by an infinitesimal argument. However, his definition a
E0-SEMIGROUPS
9
priori depends on the choice of a normalized unit. Powers-Price [62] clarified the
precise relationship between Arveson's index and the boundary representation, and
Alevras [1] showed that the boundary representation does not depend on the choice
of a normalized unit.
Arveson [6] showed that the addition formula ind(α ⊗ β) = ind(α) + ind(β) holds
For a product system with UE 6= ∅, we fix a unit e ∈ UE with normalization
for spatial E0-semigroups α and β. When one of them is non-spatial, so is α ⊗ β.
he(t), e(t)i = 1, and set
U e
E = {u ∈ UE; hu(t), e(t)i = 1, ∀t > 0}.
Let Lu = δu − δe ∈ C0UE for u ∈ U e
and H(UE, cE) is spanned by {Lu}u∈U e
E
For the exponential product systems EK, we set
E. Then one can show that hLu, Lvi = cE(u, v)
.
u(a,ζ)(t) = eat exp(1(0,t) ⊗ ζ) ∈ EK(t),
for a ∈ C and ζ ∈ K, where 1(0,t) is the indicator function of the interval (0, t). Then
u(a,ζ) is a unit and
cEK (u(a1,ζ1), u(a2,ζ2)) = a1 + a2 + hζ1, ζ2i.
Arveson [5] showed that there are no other units. We can choose e = u(0,0), the
vacuum vector, and in this case U e
EK = {u(0,ζ); ζ ∈ K}. Now the Hilbert space
H(UEK , cEK) is identified with the multiplicity space K.
Theorem 2.12. Every unit of the exponential product system EK is of the form u(a,ζ)
for a ∈ C and ζ ∈ K. The correspondence K ∋ ζ 7→ δu(0,ζ) − δu(0,0) ∈ H(UEK , cEK )
gives a unitary operator from K onto H(UEK , cEK ). In particular, the index ind(αK)
of the CCR flow αK is dim K.
2.5. Type classification and the classification of type I product systems.
Product systems are classified according to how abundant the set UE is.
Definition 2.13. Let E be a product system.
(i) We say that E is of type I if the linear span of
{u1(t1)u2(t2)· · · un(tm) ∈ E(t); u1, u2, . . . , um ∈ UE, t1 + t2 + · · · tm = t}
is dense in E(t) for all (or equivalently, some) t > 0. Type I product systems
are further divided into type In, n = 1, 2, . . . ,∞, according to the value
n = ind(E) of the index.
(ii) We say that E is of type II if UE 6= ∅ and the condition in (i) is not satisfied.
Type II product systems are further divided into type IIn, n = 0, 1, . . . ,∞,
according to the value n = ind(E) of the index.
(iii) We say that E is of type III if UE = ∅.
We use the same terms for E0-semigroups α through the product systems Eα.
Remark 2.14. Since we assume that E(t) is infinite dimensional for each t > 0,
type I0 never occurs while type II0 product systems actually occur and they form an
10
MASAKI IZUMI
important subclass of type II product systems. One could define trivial E0-semigroups
(i.e. αt ∈ Aut(B(H))) to be of type I0.
Arveson [5] completely classified type I E0-semigroups.
Theorem 2.15. Let E be a type I product system, and let K = H(UE, cE). Then E is
isomorphic to the exponential product system EK. In particular, there exists exactly
one cocycle conjugacy class of E0-semigroups of type In for each n = 1, 2, . . . ,∞.
To prove that two given E0-semigroups are cocycle conjugate, in general it is not
so easy to construct a cocycle explicitly, and Arveson's proof really makes use of the
advantage of introducing the abstract notion of product systems.
We sketch how to construct the isomorphism in Theorem 2.15. For a given type I
product system E, we fix a normalized unit e ∈ UE, and define U e
E and Lu for u ∈ U e
E
as before. Let u1, u2, . . . , um ∈ U e
E, and let t1, t2, . . . , tm > 0 with summation t. We
set s0 = 0, and si = t1 + t2 + · · · + ti. Then the isomorphism in Theorem 2.15 takes
u1(t1)u2(t2)· · · um(tm) ∈ E(t) to
mXi=1
1(si−1,si) ⊗ Lui) ∈ eL2((0,t),K).
exp(
Theorem 2.15 shows that even if we try more general L´evy processes instead of the
Brownian motion in Remark 2.8, we still get exponential product systems. To obtain
non-type I product systems, we need truly non-classical noises.
Later, Arveson [15] strengthened Theorem 2.15. Let E be a product system. We
say that a non-zero vector x ∈ E(t) is decomposable if for every 0 < s < t, there
exist y ∈ E(s) and z ∈ E(t − s) satisfying x = yz. We denote by D(t) the set
of decomposable vectors in E(t). We say that E is decomposable if D(t) is a total
subset of E(t) for all t > 0. A typical example of a decomposable vector is the
product of units u1(t1)u2(t2)· · · um(tm) as above, and so type I product systems are
decomposable. Arveson [15] showed that every decomposable product system is of
type I, whose proof is much more involved than that of Theorem 2.15.
2.6. Gauge groups. It is often true that significant information of a mathematical
object is carried by the structure of its automorphism group. For an E0-semigroup α,
the automorphism group Aut(Eα) of the associated product system Eα is isomorphic
to the gauge group.
Definition 2.16. A gauge cocycle of an E0-semigroup α is an α-cocycle U satisfying
αU = α, that is Ut ∈ αt(B(H))′ for all t > 0. We denote by G(α) the group of gauge
cocycles, and call it the gauge group of α.
Since U(H) is a Polish group in the weak operator topology, so is the gauge group
G(α) in the topology of uniform convergence on compact subsets.
For a type I product system, automorphisms are determined by their actions on
units. Arveson [5] completely determined the structure of the gauge groups of the
CCR flows.
E0-SEMIGROUPS
11
Theorem 2.17. For the CCR flow αK, the gauge group G(αK) is isomorphic to the
central extension of the semi-direct product group K ⋊ U(K) by R. More precisely
G(αK) = R × K × U(K) as a topological space, and the group operation is given by
(λ, ξ, U)(µ, η, V ) = (λ + µ + ω(ξ, Uη), ξ + Uη, UV ),
where ω is the symplectic form ω(ξ, η) = ℑhξ, ηi, ξ, η ∈ K.
Since every type I E0-semigroup is cocycle conjugate to one of the CCR flows, The-
orem 2.17 is often very useful in order to show type I criteria in specific constructions
of E0-semigroups (see, for example, [40],[42],[43]).
2.7. Dilation theory. One of the richest sources of E0-semigroups is semigroups
of unital normal completely positive maps, which are often easier to construct than
E0-semigroups.
Definition 2.18. A CP0-semigroup is a family of unital normal completely positive
maps P = {Pt}t≥0 of a von Neumann algebra N satisfying the following conditions:
(i) P0 = id,
(ii) Ps ◦ Pt = Ps+t for all s, t ≥ 0,
(iii) For every a ∈ N and every normal functional ϕ ∈ N∗, the function [0,∞) ∋
t 7→ ϕ(Pt(a)) ∈ C is continuous.
A corner N of a von Neumann algebra M is a von Neumann subalgebra of the
particular form N = pMp, where p is a projection. The central carrier of p in M is
the smallest projection in the center Z(M) of M dominating p.
Definition 2.19. A dilation of a CP0-semigroup P acting on N consists of a von
Neumann algebra M, a projection p ∈ M, and an E0-semigroup α acting on M
satisfying
(i) N is the corner pMp of M,
(ii) Pt(a) = pαt(a)p for any a ∈ N and t ≥ 0.
If moreover the following two conditions are satisfied, we say that the dilation is
minimal:
(iii) M is generated by ∪t≥0αt(N),
(iv) the central carrier c(p) of p in M is 1M .
Note that (ii) implies that {αt(p)}t≥0 is an increasing family of projections, and if
(iii) is satisfied, it converges to the unit of M in the strong operator topology as t
tends to ∞. In the case of a minimal dilation, if N is a factor, so is M.
We identify two dilations (M, p, α) and (R, q, β) if there exists an isomorphism θ
from M onto R such that the restriction of θ to pMp = qRq = N is the identity map
and θ ◦ αt = βt ◦ θ holds for any t ≥ 0.
Theorem 2.20. There exists a unique minimal dilation for any CP0-semigroup act-
ing on a von Neumann algebra with separable predual.
Bhat [29] proved Theorem 2.20 in the case of type I factors based on his previous
results [34],[35], and he computed the product systems of the dilations. The existence
12
MASAKI IZUMI
in the general case was obtained in [30],[36] (see also [52],[53]). These works use the
fact that the map [0,∞) ∋ t 7→ Pt(a) ∈ N is continuous in the strong operator
topology, which was proved by Markiewicz and Shalit [50] later. Arveson [13],[23,
Section 8.9] showed the uniqueness in the general case. The uniqueness in the non-
unital case appears very subtle (see [31]).
When P is a CP0-semigroup acting on a type I factor, Arveson [14] described the
units of the minimal dilation in terms of P . Answering a question raised in [29] about
the dilations of CP0-semigroups acting on matrix algebras, Arveson [18] showed the
following result (cf. [58]).
Theorem 2.21. Let P be a CP0-semigroup acting on B(H0), with H0 possibly finite
dimensional, which is not a semigroup of automorphisms. If the generator L of P
is bounded, then the minimal dilation α of P is an E0-semigroup of type I, and the
index ind(α) is the rank of L.
For the definition of the rank of L, see [23, Chapter 10].
Arveson [16],[19] applied dilation theory to what is called interaction theory, which,
roughly speaking, deals with coupling of two E0-semigroups, one for the past and the
other for the future, with prescribed invariant normal states.
Markiewicz [48] computed the product systems for the minimal dilations of concrete
CP0-semigroups acting on B(L2(R)) arising from a modified Weyl-Moyal quantization
of convolution semigroups of probability measures on R2, including the CCR heat flow
discussed by Arveson [20] as a special case. Despite that the generators of these CP0-
semigroups are unbounded, the resulting product systems are still of type I (in fact
type I2).
Shalit-Solel [65] and Bhat-Mukherjee [33] recently introduced essentially the same
notion, called subproduct systems in [65], and inclusion systems in [33], which had
been implicitly used in [36], [48], [52]. Their role in product systems is somewhat
similar to the role of CP0-semigroups in E0-semigroups. See [32], [54], [61], [69] for
related results.
3. Beyond Arveson: Type II case
The first example of an E0-semigroup of type II was constructed implicitly by
Tsirelson-Vershik [76] in 1998 via the noise theory, and about the same time by Powers
[57] via the boundary representation. Later on, both Powers [59] and Tsirelson [72]
constructed uncountably many type II0 examples. Thanks to Arveson's addition
formula, we have In ⊗ II0 = IIn, and so type IIn examples exist for n = 0, 1, . . . ,∞.
3.1. Tsirelson's probabilistic method. Tsirelson introduced the following con-
cept of a noise in probability theory (see [72],[74]).
Definition 3.1. A noise consists of a probability space (Ω,F , P ), sub-σ-fields Fs,t ⊂
F for s, t ∈ R with s < t, and a measure preserving action T of R on the probability
space satisfying
(i) Fr,s ⊗ Fs,t = Fr,t for any r < s < t,
(ii) Th sends Fs,t to Fs+h,t+h,
E0-SEMIGROUPS
13
(iii) F is generated by ∪s<tFs,t,
(iv) P (A ⊖ T −1
difference of A and B.
h (A)) → 0 as h → 0 for any A ∈ F , where A ⊖ B is the symmetric
A typical example of a noise is white noise already discussed in Remark 2.8. Every
noise gives rise to a product system by E(t) = L2(F0,t) and ξ · η = ξ(η ◦ T−s) for
ξ ∈ E(s), η ∈ E(t). The resulting product system has at least one unit given by the
constant function 1. Using the factorization L2(F−∞,∞) = L2(F−∞,0)⊗ L2(F0,∞) and
the 1-parameter unitary group {Ut}t∈R arising from T , we can directly construct the
corresponding E0-semigroup α acting on the type I factor B(F0,∞) by 1 ⊗ αt(A) =
Ut(1 ⊗ A)U ∗
A noise arising from a L´evy process is called a classical noise, and Tsirelson showed
that every noise contains the maximal classical noise, called the classical part, which
corresponds to the subspaces generated by decomposable vectors in E(t). A black
noise is a noise with trivial classical part, which gives rise to a product system of
type II0. Tsirelson-Vershik [76] showed
t for A ∈ B(L2(F0,∞)).
Theorem 3.2. There exists a black noise.
A black noise is a singular object, and is not so easy to construct. Very few
examples are known.
Like the Wiener-Ito chaos decomposition, the space L2(Ω,F , P ) has a canonical
decomposition. However, subspaces for a finite number of particles do not generate
the whole space unless the noise is classical. From this decomposition, random sets
arise as an invariant of the noise, and it also makes sense as an invariant of product
systems of type II0. A variant of this invariant adapted to type II product systems
was extensively studied by Liebscher [47]. Among others he showed that there exists
a type IIn product system for each n = 0, 1, . . . ,∞, that never splits as a tensor
product of two product systems.
Tsirelson [72] introduced the notion of homogeneous continuous products of mea-
sure classes (HCPMC), more general objects than noises. This idea originated from
Vershik according to Tsirelson. HCPMCs are more flexible than noises, and are still
good enough to produce E0-semigroups. The main difference of an HCPMC from a
noise is that the independence Fr,s ⊗ Fs,t = Fr,s does not necessarily hold for P , but
it is required to hold for a measure equivalent to P . Typical examples of HCPMCs
arise from random sets associated with Markov processes. Tsirelson [72] showed that
they give uncountably many type II0 product systems that are not anti-isomorphic
to themselves.
Tsirelson [75] recently constructed a type II1 product system, by using the noise
theory, whose automorphism group does not act on the set of normalized units tran-
sitively. This shows that choosing an arbitrary unit is not always justified in order
to define an invariant of product systems.
3.2. Powers CP -flows. Powers [59],[60] found a systematic way to construct E0-
semigroups of type II by using dilation theory. For simplicity, we assume that CP -
flows are unital in this note, though non-unital ones also play important roles in
Powers's argument.
14
MASAKI IZUMI
Definition 3.3. Let K be a separable complex Hilbert space, and let H0 = L2((0,∞), K).
We denote by S = {St}t≥0 the shift semigroup acting on H0. A CP -flow P is a CP0-
semigroup acting on B(H0) satisfying StA = Pt(A)St for all A ∈ B(H0) and t ≥ 0.
The minimal dilation of a CP -flow always has a unit. On the other hand, any
spatial E0-semigroup is cocycle conjugate to an E0-semigroup that is a CP -flow.
Thus it is important to understand the structure of CP -flows. Powers [59] showed
that all of the information of a CP -flow is encoded in its boundary weight map.
We define Λ : B(K) → B(H0) by
Λ(A)f (x) = e−xAf (x),
f ∈ H0.
For simplicity, we include complete positivity and the unitality condition in the def-
inition of boundary weight maps.
Definition 3.4. Let A(H0) = (1H0 − Λ(1K))1/2B(H0)(1H0 − Λ(1K))1/2. A boundary
weight µ is a linear functional of A(H0) such that the linear functional
B(H0) ∋ A 7→ µ((1H0 − Λ(1K))1/2A(1H0 − Λ(1K))1/2) ∈ C,
is bounded and normal. We denote by A(H0)∗ the set of boundary weights. A
boundary weight map ω is a completely positive map ω : B(K)∗ → A(H0)∗ satisfying
ω(ρ)(1H0 − Λ(1K)) = ρ(1K) for any ρ ∈ B(K)∗.
For a normal map Φ between von Neumann algebras, we denote by Φ the map
between the preduals induced by Φ. Being a semigroup, a CP -flow P is determined
by its resolvent
RP (A) =Z ∞
0
e−tPt(A)dt.
On the other hand, since we have
Pt(A) = StAS∗
t )Pt(A)(1 − StS∗
t ), A ∈ B(H0),
the first approximation of RP is
t + (1 − StS∗
Γ(A) =Z ∞
0
e−tStAS∗
t dt.
Our task is to describe the difference RP − Γ, which is a completely positive map.
Powers [59] showed the following.
Theorem 3.5. Let the notation be as above.
(i) For any CP -flow P , there exists a unique boundary weight map ω satisfying
RP (η) = Γ(ω(Λη) + η),
∀η ∈ B(H0)∗.
(ii) For a boundary weight map ω, we set ωt(ρ)(A) = ω(ρ)(StS∗
The map ω is called the boundary weight map associated with P .
t AStS∗
t ). If id +Λωt
is invertible and πt := ωt◦ (id +Λωt)−1 is a completely positive contraction for
any t > 0, then ω is the boundary weight map associated with a CP -flow.
E0-SEMIGROUPS
15
When K = C, a boundary weight map ω is identified with the boundary weight
ω(1K), which is a normal semifinite weight of B(H0) satisfying
The condition in (ii) is automatically satisfied for such a weight. In particular, any
function f with
ω(1K)(1H0 − Λ(1K)) = 1.
Z ∞
f (x)2(1 − e−x)dx = 1,
0
gives rise to a CP -flow. Already this case provides uncountably many type II0 E0-
semigroups. More precisely, Powers showed that the resulting E0-semigroup is of
type II0 unless f ∈ L2(0,∞), and that for any such functions f1, f2 as above, the
resulting E0-semigroups are cocycle conjugate if and only if c1f1 + c2f2 ∈ L2(0,∞)
for some c1, c2 ∈ C \ {0}.
See [3],[44],[45],[46],[49] for recent progress in this approach. There are several
E0-semigroups of type II whose gauge groups are known.
It is desirable to unify the two approaches presented in this section, and we propose
two problems, just to start with.
(i) Compute Tsirelson's random sets invariant for an E0-semigroup of type II0
arising from a CP -flow in terms of its boundary weight map.
(ii) Give a description of the boundary representation for an E0-semigroup of type
II arising from an HCPMC.
4. Beyond Arveson: Type III case
Powers [55] constructed the first example of a type III E0-semigroup using the CAR
algebra in 1987, just after the theory of E0-semigroups was initiated. It had been the
only example of a non-type I E0-semigroup for a while. Much later, Tsirelson [72]
constructed the first continuous family of type III product systems using HCPMCs
coming from off-white noises.
4.1. CAR construction. Recall that the CAR flows are conjugate to the CCR
flows, which are necessarily of type I, and they are constructed in the vacuum rep-
resentation. To construct the first type III example, Powers used a quasi-free repre-
sentation of the CAR algebra instead of the vacuum representation.
Let G := L2((0,∞), CN ). We denote by A the CAR algebra over the test function
space, which is the universal C ∗-algebra generated by {a(f ); f ∈ G}, depending
linearly on f , and satisfying the CAR relations:
a(f )a(g) + a(g)a(f ) = 0,
a(f )a(g)∗ + a(g)∗a(f ) = hf, gi1.
Let S = {St}t≥0 be the shift semigroup acting on G. Since the CAR relation
involves only the inner product, there exists a continuous semigroup γ of unital
endomorphisms of A given by γt(a(f )) = a(Stf ). If π is a type I factor representation
of A such that π ◦ γt is quasi-equivalent to π for all t ≥ 0, then γ extends to an E0-
semigroup acting on π(A)′′. The vacuum representation is an example of such a
representation, giving the CAR/CCR flow of rank N.
16
MASAKI IZUMI
A quasi-free state ωA on A associated with a positive contraction A ∈ B(G) is a
unique state determined by the formula
ωA(a(fn)· · · a(f1)a(g1)∗ · · · a(gm)∗) = δn,m det(hAfi, gji).
If a positive contraction A satisfies the condition
(4.1)
tr(A − A2) < ∞, S∗
t ASt = A,
∀t ≥ 0,
the GNS representation for ωA has the desired property, and we can construct an
E0-semigroup.
To present the positive contraction Powers constructed, we need to introduce
Toeplitz operators. We regard G as a closed subspace of G := L2(R, CN ), and
we denote by P+ the projection from G onto G. For Φ ∈ L∞(R)⊗ MN (C), we define
the corresponding Fourier multiplier CΦ ∈ B( G) by
\(CΦf )(p) = Φ(p)bf (p).
Then the Toeplitz operator TΦ ∈ B(G) with a symbol Φ is defined by TΦf = P+CΦf ,
f ∈ G.
Powers [55] came up with a mysterious symbol giving a type III example.
Theorem 4.1. Let N = 2, and let
Φ(p) =
1
2(cid:18) 1
e−iθ(p)
eiθ(p)
1 (cid:19) ,
θ(p) = (1 + p2)−1/5.
Then A = TΦ satisfies the condition (4.1), and the quasi-free representation for
A = TΦ gives a type III E0-semigroup.
Arveson [23, Section 13.3] determined the most general form of a positive contrac-
tion A ∈ B(G) satisfying the condition (4.1), and showed that such an operator must
be a Toeplitz operator TΦ with a symbol Φ satisfying a certain condition. We call
the resulting E0-semigroup the Toeplitz CAR flow arising from the symbol Φ.
Let Φν be the matrix valued function given by the same formula as the Powers
symbol except for θ(p) = (1 + p2)−ν. Then TΦν satisfies the condition (4.1) for all
ν > 0, and we denote by αν the resulting E0-semigroup. Recently Srinivasan and the
author [43] showed the following.
Theorem 4.2. Let the notation be as above.
(i) If ν > 1/4, then αν is of type I2.
(ii) If 0 < ν ≤ 1/4, then αν is of type III.
(iii) If 0 < ν1 < ν2 ≤ 1/4, then αν1 and αν2 are not cocycle conjugate.
To distinguish αν in the type III region 0 < ν ≤ 1/4, we used the type I factor-
izations of Araki-Woods [4] arising from local von Neumann algebras for the product
systems (see the next subsection). In general Toeplitz CAR flows are either of type
I or of type III.
ZR2
log σ(λ1) − log σ(λ2)2
λ1 − λ22
dλ1dλ2 < ∞,
E0-SEMIGROUPS
17
4.2. Off-white noises and generalized CCR flows. Let {Bt}t∈R be the (two-
sided) Brownian motion, and let X(t) be the formal derivative dB(t)
. Then {X(t)}t∈R
dt
is a stationary Gaussian generalized (i.e. distribution valued) random process with
correlation function E(X(s)X(t)) = δ(s − t). There is no relation between the past
and the future at all. Let Fs,t be the σ-field generated by hX, fi with test functions
f supported in (s, t). Then we get white noise.
An off-white noise is an HCPMC, not really a noise, constructed in the same
way by replacing X(t) with a stationary Gaussian generalized random process ξ(t)
having a slight correlation of the past and the future. The correlation function
C(s−t) = E(ξ(s)ξ(t)) is now a positive definite distribution, whose Fourier transform
C is a measure. Tsirelson [71],[73] showed that if C has a density σ(λ) with respect
to the Lebesgue measure dλ, and σ satisfies
then we get an HCPMC, which is called an off-white noise. The function σ(λ) is
called the spectral density function of the off-white noise, and all information about
the off-white noise is encoded in it. In the case of white noise, it is a constant function.
Tsirelson [72] showed the following.
Theorem 4.3. For r > 0, let σr be a smooth positive even function with σr(λ) =
log−r λ for large λ. Then σr is a spectral density function of an off-white noise, and
the family {σr(λ)}r>0 gives rise to mutually non-isomorphic type III product systems.
Tsirelson's construction has many faces. Bhat-Srinivasan [37] systematically inves-
tigated the product systems arising from so-called sum systems by a purely functional
analytic method, which recaptures Tsirelson's construction as a special case.
Srinivasan and the author [42] showed that the E0-semigroups corresponding to
the product systems arising from sum systems are generalized CCR flows. Let G be
a real Hilbert space, and let S = {St}t≥0 and T = {Tt}t≥0 be C0-semigroups acting
on G such that T ∗
t St = 1G and St − Tt is a Hilbert-Schmidt operator for any t ≥ 0.
Then we can construct an E0-semigroup α acting on B(eG⊗C) by
f, g ∈ G.
αt(W (f + ig)) = W (Stf + iTtg),
E0-semigroups constructed in this way are called generalized CCR flows. Generalized
CCR flows are either of type I or of type III (see [37],[41]). In the case of off-white
noises, we can choose G = L2((0,∞), R) and S to be the shift semigroup. The author
[40] gave the precise relationship between T and the spectral density function σ in
this case. Using this correspondence, Srinivasan and the author [42] showed that
there exists a continuous family of off-white noises whose spectral density functions
converge to 1 at infinity, such that the family still gives mutually non-isomorphic
type III product systems.
There is a certain similarity between the Toeplitz CAR flows and generalized CCR
flows (for example, they are never of type II), and it is desirable to clarify their
relationship. As a first step, we propose the following problem: determine the gauge
groups of the Toeplitz CAR flows and generalized CCR flows.
18
MASAKI IZUMI
Before ending this final section, we emphasize the importance of local von Neumann
algebras associated with product systems in the results discussed in this section. Let
E be a product system. For 0 < s < t < 1, we denote by Us,t,1 the unitary map
determined by
Us,t,1 : E(s) ⊗ E(t − s) ⊗ E(1 − t) ∋ x ⊗ y ⊗ z → xyz ∈ E(1).
For an interval J = (s, t) ⊂ [0, 1], we define the von Neumann algebra AE(J) ⊂
B(E(1)) associated with J by
AE(J) = Us,t,1(1E(s) ⊗ B(E(t − s)) ⊗ 1E(1−t))U ∗
s,t,1.
We apply a similar definition in the case with s = 0 or t = 1. Then AE(J) is
a type I subfactor of B(E(1)), and when two intervals J1 and J2 are disjoint, the
corresponding algebras AE(J1) and AE(J2) commute with each other. For an open
subset O ⊂ [0, 1], we define AE(O) to be the von Neumann algebra generated by
∪J⊂OAE(J), which may not be of type I. The system of von Neumann algebras
{AE(J)}J⊂[0,1] is an important isomorphism invariant of the product system E, and
it is an analogue of local observable algebras in algebraic quantum field theory. For
example, Liebscher [47] showed the following useful theorem, which Srinivasan and
the author used in [43] to obtain a type I criterion.
Theorem 4.4. Let E and F be product systems. If there exists an isomorphism θ
from B(E(1)) onto B(F (1)) satisfying θ(AE((0, t))) = AF ((0, t)) for all 0 < t < 1,
then E and F are isomorphic.
The system of von Neumann algebras {AE(I)}I⊂[0,1] and the isomorphism classes
of von Neumann algebras AE(O) are employed as isomorphism invariants of E in
[42],[43],[72] to differentiate continuous families of product systems.
Acknowledgements
The author would like to thank Alexis Alevras, Kenneth Davidson, Daniel Markiewicz,
and Gilles Pisier for useful comments. This work is supported in part by the Grant-
in-Aid for Scientific Research (B) 22340032, JSPS.
References
[1] A. Alevras, A note on the boundary representation of a continuous spatial semigroup of
[2] A. Alevras, One parameter semigroups of endomorphisms of factors of type II1, J. Operator
∗-endomorphisms of B(H), Proc. Amer. Math. Soc., 123(1995), 3129 -- 3133.
Theory, 51(2004), 161 -- 179.
[3] A. Alevras, R. T. Powers, G. Price, Cocycles for one-parameter flows of B(H), J. Funct.
Anal., 230(2006), 1 -- 64.
[4] H. Araki, E. J. Woods, Complete Boolean algebras of type I factors, Publ. Res. Inst. Math.
Sci. Ser. A, 2(1966), 157-242.
[5] W. Arveson, Continuous analogues of Fock space, Mem. Amer. Math. Soc. 80 (1989), no.
409.
[6] W. Arveson, An addition formula for the index of semigroups of endomorphisms of B(H),
Pacific J. Math., 137(1989), 19 -- 36.
[7] W. Arveson, Continuous analogues of Fock space. III. Singular states, J. Operator Theory,
22(1989), 165 -- 205.
E0-SEMIGROUPS
19
[8] W. Arveson, Continuous analogues of Fock space. II. The spectral C ∗-algebra, J. Funct.
Anal., 90(1990), 138 -- 205.
[9] W. Arveson, Continuous analogues of Fock spaces IV: Essential states, Acta Math.,
164(1990), 265-300.
[10] W. Arveson, Quantizing the Fredholm index, Operator Theory: Proceedings of the 1988
GPOTS-Wabash Conference (Indianapolis, IN, 1988), 1 -- 31, Pitman Res. Notes Math. Ser.
225, Longman Sci. Tech., Harlow, 1990.
[11] W. Arveson, The spectral C ∗-algebra of an E0-semigroup, Operator theory: operator algebras
and applications, Part 1 (Durham, NH, 1988), 1 -- 15, Proc. Sympos. Pure Math., 51, Part 1,
Amer. Math. Soc., Providence, RI, 1990.
[12] W. Arveson, E0 -semigroups in quantum field theory, Quantization, nonlinear partial dif-
ferential equations, and operator algebra (Cambridge, MA, 1994), 1 -- 26, Proc. Sympos. Pure
Math., 59, Amer. Math. Soc., Providence, RI, 1996.
[13] W. Arveson, Minimal E0-semigroups, Operator algebras and their applications (Waterloo,
ON, 1994/1995), 1 -- 12, Fields Inst. Commun., 13, Amer. Math. Soc., Providence, RI, 1997.
[14] W. Arveson, The index of a quantum dynamical semigroup, J. Funct. Anal., 146(1997),
557 -- 588.
[15] W. Arveson, Path spaces, continuous tensor products, and E0-semigroups, Operator algebras
and applications (Samos, 1996), 1 -- 111, NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 495,
Kluwer Acad. Publ., Dordrecht, 1997.
[16] W. Arveson, Pure E0-semigroups and absorbing states, Comm. Math. Phys., 187(1997),
19 -- 43.
[17] W. Arveson, Dynamical invariants for noncommutative flows, Operator algebras and quantum
field theory (Rome, 1996), 476 -- 514, Int. Press, Cambridge, MA, 1997.
[18] W. Arveson, On the index and dilations of completely positive semigroups, Internat. J. Math.,
10(1999), 791 -- 823.
[19] W. Arveson, Interactions in noncommutative dynamics, Comm. Math. Phys., 211(2000),
63 -- 83.
[20] W. Arveson, The heat flow of the CCR algebra, Bull. London Math. Soc., 34(2002), 73 -- 83.
[21] W. Arveson, The domain algebra of a CP-semigroup, Pacific J. Math., 203 (2002), 67 -- 77.
[22] W. Arveson, Generators of non-commutative dynamics, Ergodic Theory Dynam. Systems,
22(2002), 1017 -- 1030.
[23] W. Arveson, Non-commutative dynamics and E-semigroups, Springer Monograph in Math.
(Springer 2003).
[24] W. Arveson, Four lectures on noncommutative dynamics, Advances in quantum dynamics
(South Hadley, MA, 2002), 1 -- 55, Contemp. Math., 335, Amer. Math. Soc., Providence, RI,
2003.
[25] W. Arveson, On the existence of E0-semigroups, Infin. Dimens. Anal. Quantum Probab.
Relat. Top., 9(2006), 315 -- 320.
[26] W. Arveson, A. Kishimoto, A note on extensions of semigroups of ∗-endomorphisms, Proc.
[27] W. Arveson, G. Price, Infinite tensor products of completely positive semigroups, J. Evol.
Amer. Math. Soc., 116(1992), 769 -- 774.
Equ. 1(2001), 221 -- 242.
[28] S. D. Barreto, B. V. R. Bhat, V. Liebscher, M. Skeide, Type I product systems of
Hilbert modules, J. Funct. Anal., 212 (2004), 121 -- 181.
[29] B. V. R. Bhat, An index theory for quantum dynamical semigroups, Trans. Amer. Math.
Soc., 348(1996), 561 -- 583.
[30] B. V. R. Bhat, Minimal dilations of quantum dynamical semigroups to semigroups of endo-
morphisms of C ∗-algebras, J. Ramanujan Math. Soc., 14(1999), 109 -- 124.
[31] B. V. R. Bhat, Atomic dilations, Advances in quantum dynamics (South Hadley, MA, 2002),
99 -- 107, Contemp. Math., 335, Amer. Math. Soc., Providence, RI, 2003.
20
MASAKI IZUMI
[32] B. V. R. Bhat, V. Liebscher, M. Mukherjee, M. Skeide, The spatial product of Arveson
systems is intrinsic, J. Funct. Anal., 260(2011), 566 -- 573.
[33] B. V. R. Bhat, M. Mukherjee, Inclusion systems and amalgamated products of product
systems. Infin. Dimens. Anal. Quantum Probab. Relat. Top., 13(2010), 1 -- 26.
[34] B. V. R. Bhat, K. R. Parthasarathy, Kolmogorov's existence theorem for Markov pro-
cesses in C ∗-algebras, K. G. Ramanathan memorial issue. Proc. Indian Acad. Sci. Math. Sci.,
104(1994), 253 -- 262.
[35] B. V. R. Bhat, K. R. Parthasarathy, Markov dilations of nonconservative dynamical
semigroups and a quantum boundary theory, Ann. Inst. H. Poincare Probab. Statist., 31(1995),
601 -- 651.
[36] B. V. R. Bhat, M. Skeide, Tensor product systems of Hilbert modules and dilations of
completely positive semigroups, Infin. Dimens. Anal. Quantum Probab. Relat. Top., 3(2000),
519 -- 575.
[37] B. V. R. Bhat, R. Srinivasan, On product systems arising from sum systems, Infin. Dimens.
Anal. Quantum Probab. Relat. Top. 8(2005), 1 -- 31.
[38] I. Hirshberg, C ∗-algebras of Hilbert module product systems, J. Reine Angew. Math. 570
(2004), 131 -- 142.
[39] I. Hirshberg, J. Zacharias, On the structure of spectral algebras and their generalizations,
Advances in quantum dynamics (South Hadley, MA, 2002), 149 -- 162, Contemp. Math., vol.
335, Amer. Math. Soc., Providence, RI, 2003.
[40] M. Izumi, A perturbation problem for the shift semigroup, J. Funct. Anal., 251(2007), 498 --
545.
[41] M. Izumi, Every sum system is divisible, Trans. Amer. Math. Soc., 361(2009), 4247 -- 4267.
[42] M. Izumi, R. Srinivasan, Generalized CCR flows, Comm. Math. Phys. 281(2008), 529 -- 571.
[43] M. Izumi, R. Srinivasan, Toeplitz CAR flows and type I factorizations, Kyoto J. Math.,
50(2010), 1 -- 32.
[44] C. Jankowski, On type II0 E0-semigroups induced by boundary weight doubles, J. Funct.
Anal., 258(2010), 3413 -- 3451.
[45] C. Jankowski, D. Markiewicz, Gauge groups of E0-semigroups obtained from Powers
weights, Int. Math. Res. Notices, (2012), 3278 -- 3310.
[46] C. Jankowski, D. Markiewicz, R. T. Powers, E0-semigroups of type II0 and q-purity:
boundary weight maps of range rank one and two, J. Funct. Anal., 262(2012), 3006 -- 3061.
[47] V. Liebscher, Random sets and invariants for (type II) continuous tensor product systems of
Hilbert spaces, Mem. Amer. Math. Soc. 199 (2009), no. 930.
[48] D. Markiewicz, On the product system of a completely positive semigroup, J. Funct. Anal.,
200(2003), 237 -- 280.
[49] D. Markiewicz, R. T. Powers, Local unitary cocycles of E0-semigroups, J. Funct. Anal.,
256(2009), 1511 -- 1543.
[50] D. Markiewicz, O. M. Shalit, Continuity of CP -semigroups in the point-strong operator
topology, J. Operator Theory, 64(2010), 149 -- 154.
[51] P.-A. Meyer, Quantum probability for probabilists. Lecture Notes in Mathematics, 1538.
Springer-Verlag, Berlin, 1993.
[52] P. S. Muhly, B. Solel, Quantum Markov processes (correspondences and dilations), Inter-
nat. J. Math., 13 (2002), 863 -- 906.
[53] P. S. Muhly, B. Solel, Quantum Markov semigroups: product systems and subordination,
Internat. J. Math. 18(2007), 633 -- 669.
[54] M. Mukherjee, Index computation for amalgamated products of product systems, Banach
J. Math. Anal., 5(2011), 148 -- 166.
Inst. Math. Sci. 23(1987), 1053 -- 1069.
[55] R. T. Powers, A nonspatial continuous semigroup of ∗-endomorphisms of B(H), Publ. Res.
[56] R. T. Powers, An index theory for semigroups of ∗-endomorphisms of B(H) and type II1
factors, Can. J. Math., 40(1988), 86 -- 114.
E0-SEMIGROUPS
21
[57] R. T. Powers, New examples of continuous spatial semigroups of ∗-endomorphisms of B(H),
[58] R. T. Powers, Induction of semigroups of endomorphisms of B(H) from completely positive
Internat. J. Math. 10(1999), 215 -- 288.
[59] R. T. Powers, Continuous spatial semigroups of completely positive maps of B(h), New York
semigroups of (n × n) matrix algebras, Internat. J. Math., 10(1999), 773 -- 790.
J. Math., 9(2003), 165 -- 269.
[60] R. T. Powers, Construction of E0-semigroups of B(h) from CP-flows. Advances in quan-
tum dynamics (South Hadley, MA, 2002), 57 -- 97, Contemp. Math., 335, Amer. Math. Soc.,
Providence, RI, 2003.
[61] R. T. Powers, Addition of spatial E0-semigroups, Operator algebras, quantization, and non-
commutative geometry, 281 -- 298, Contemp. Math., 365, Amer. Math. Soc., Providence, RI,
2004.
Amer. Math. Soc. 321(1990), 347 -- 361.
[62] R. T. Powers, G. Price, Continuous spatial semigroups of ∗-endomorphisms of B(H), Trans.
[63] R. T. Powers, D. W. Robinson, An index for continuous semigroups of ∗-endomorphisms
[64] G. L. Price, B. M. Baker, P. E. T. Jorgensen, P. S. Muhly (Editors), Advances in
Quantum Dynamics (South Hadley, MA, 2002), Contemp.Math. 335, Amer. Math. Society,
Providence, RI (2003).
of B(H), J. Funct. Anal. 84(1989), 85 -- 96.
[65] O. Shalit, B. Solel, Baruch, Subproduct systems, Doc. Math. 14(2009), 801 -- 868.
[66] M. Skeide, A simple proof of the fundamental theorem about Arveson systems, Infin. Dimens.
Anal. Quantum Probab. Relat. Top., 9(2006) 305 -- 314.
[67] M. Skeide, Existence of E0-semigroups for Arveson systems: Making two proofs into one,
Infin. Dimens. Anal. Quantum Probab. Relat. Top. 9(2006), 373 -- 378.
[68] M. Skeide, Product systems; a survey with commutants in view, Quantum stochastics and
information, 47 -- 86, World Sci. Publ., Hackensack, NJ, 2008.
[69] M. Skeide, The Powers sum of spatial CP D-semigroups and CP -semigroups, Noncommutative
harmonic analysis with applications to probability II, 247 -- 263, Banach Center Publ., 89, Polish
Acad. Sci. Inst. Math., Warsaw, 2010.
[70] M. Takesaki, Theory of Operator Algebras. I. Encyclopaedia of Mathematical Sciences, 124.
Operator Algebras and Non-commutative Geometry, 5. Springer-Verlag, Berlin, 2002.
[71] B. Tsirelson, Spectral densities describing off-white noises, Ann. Inst. H. Poincar Probab.
Statist., 38(2002), 1059 -- 1069.
[72] B. Tsirelson, Non-isomorphic product systems, Advances in Quantum Dynamics (South
Hadley, MA, 2002), 273 -- 328, Contemp. Math., 335, Amer. Math. Soc., Providence, RI, 2003.
[73] B. Tsirelson, Scaling limit, noise, stability, Lectures on probability theory and statistics, 1 --
106, Lecture Notes in Math., 1840, Springer, Berlin, 2004.
[74] B. Tsirelson, Nonclassical stochastic flows and continuous products, Probab. Surv., 1(2004),
173 -- 298.
[75] B. Tsirelson, On automorphisms of type II Arveson systems (probabilistic approach), New
York J. Math., 14 (2008), 539 -- 576.
[76] B. Tsirelson, A. Vershik, Examples of nonlinear continuous tensor products of measure
spaces and non-Fock factorizations, Rev. Math. Phys., 10(1998), 81 -- 145.
[77] J. Zacharias, Continuous tensor products and Arveson's spectral C ∗-algebras, Mem. Amer.
Math. Soc. 143 (2000), no. 680.
[78] J. Zacharias, Pure infiniteness of spectral algebras, J. Funct. Anal. 178(2000), 381 -- 395.
22
MASAKI IZUMI
5. Appendix: Dilation theory and noncommutative Poisson boundary
The notion of noncommutative Poisson boundaries for normal ucp maps was intro-
duced by the author [I1]. Let N be a von Neumann algebra, and let L be a weakly
closed operator system in N, i.e. L is a self-adjoint linear subspace of N including
the identity. It is known that if there exists a completely positive projection E from
N onto L, then L is a von Neumann algebra with respect to the Choi-Effros product
x ◦ y = E(xy) (see [CE]).
Let P be a normal ucp map of N. We denote by H ∞(N, P ) the fixed point set
{x ∈ N; P (x) = x} of P , whose members are called harmonic elements. Then
H ∞(N, P ) is a weakly closed operator system and it is the image of a completely
positive projection from N. Indeed, we choose a free ultra-filter ω ∈ βN \ N and set
E(x) = w − lim
n→ω
1
n
n−1Xk=0
P k(x),
x ∈ N.
Then E is the desired projection. Although E depends on the choice of ω, the
Choi-Effros product of H ∞(N, P ) does not because an operator system may have at
most one von Neumann algebra structure. Concrete realization of the von Neumann
algebra structure of H ∞(N, P ) is called the noncommutative Poisson boundary for
P . This notion has proved to be particularly useful to capture structure that appears
only after taking weak closure (see, for example, [I1]).
When I invited Bill Arveson to Kyoto in 2004, he showed his interest in noncom-
mutative Poisson boundaries. However, he did not seem to be happy about the fact
that the Choi-Effros product is defined after the choice of the ultra-filter ω is made
(even though it does not depend on ω). When we met in 2007 at the Fields Institute,
Bill told me in a brief conversation that the noncommutative Poisson boundary for P
is nothing but the fixed point algebra of the minimal dilation of P . It is so natural an
idea. In fact, the usual measure theoretical construction of the Poisson boundary for
a Markov operator in the commutative situation essentially uses the corresponding
Markov process (see [K]), which is more or less the minimal dilation of the Markov
operator. I felt a little embarrassed because such an idea had never occurred to me.
I include a few consequences of Bill's observation here.
Recall that the minimal dilation of P consists of a von Neumann algebra M, a
projection p ∈ M whose central carrier is 1M , and a unital normal endomorphism
α of M such that N = pMp, M is generated by ∪n≥0αn(N), and P n(a) = pαn(a)p
for all a ∈ N and n ≥ 1. In this situation {αn(p)}∞
n=0 is an increasing sequence of
projections whose limit is 1M . Without any modification, Theorem 2.20 holds in the
discrete time case too.
Theorem 5.1. Let M α = {x ∈ M; α(x) = x} be the fixed point algebra of α. Then
the map θ : M α ∋ x 7→ pxp ∈ H ∞(N, P ) is a completely positive order isomorphism
between the two operator systems. In particular, the von Neumann algebra M α gives
a concrete realization of the von Neumann algebra structure of H ∞(N, P ).
Proof. It suffices to show that θ is a completely positive isometry of M α onto H ∞(N, P )
because the inverse of a unital completely positive isometry between two operator
E0-SEMIGROUPS
23
systems is automatically completely positive. Since the proof does not change af-
ter taking the tensor product with a matrix algebra, and θ is obviously completely
positive, it suffices to show that θ is an isometry onto H ∞(N, P ).
For x ∈ M α, we have P (pxp) = pα(pxp)p = pα(p)xα(p)p. Since p ≤ α(p),
we obtain θ(x) ∈ H ∞(N, P ). The map θ is obviously a contraction. Since the
sequence {αn(p)}∞
n=1 converges to 1M in the strong operator topology, the sequence
{αn(θ(x))}∞
n=1 converges to x ∈ M α in the strong operator topology too, and so θ is
an isometry. It remains to show that θ is a surjection. For a ∈ H ∞(N, P ), we set
an =
1
n
n−1Xk=0
αk(a),
and we choose an accumulation point x of the sequence {an}∞
topology. Then x ∈ M α. Since
1
n
pαk(a)p =
panp =
P k(a) = a,
1
n
n−1Xk=0
n−1Xk=0
n=1 in the weak operator
we obtain θ(x) = a.
(cid:3)
The following statement is known in concrete examples where the martingale
convergence theorem (commutative or noncommutative) is available (see [I2],[KV]).
Thanks to the dilation theory, we are able to prove it in the general case.
Corollary 5.2. For any a, b ∈ H ∞(N, P ), the sequence {P n(ab)}∞
the Choi-Effros product a ◦ b in the strong operator topology.
Proof. Let x = θ−1(a), y = θ−1(b). Then
n=1 converges to
P n(ab) = pαn(pxpyp)p = pαn(p)xαn(p)yαn(p)p = pxαn(p)yp,
which converges to pxyp = θ(xy) in the strong operator topology.
(cid:3)
The following example was suggested by Bill. Let H 2 be the Hardy space of the unit
disk, and let p be the projection from L2(T) = ℓ2(Z) onto H 2. Let v ∈ B(H 2) be the
unilateral shift, and let u ∈ B(L2(T)) be the bilateral shift. We introduce a normal
ucp map P of N = B(H 2) by P (a) = v∗av. Then the space of harmonic elements
H ∞(N, P ) consists of the Toeplitz operators Tf , f ∈ L∞(T) (see [A, Theorem 4.2.4]).
The minimal dilation of P is given by (M = B(L2(T)), p, α = Ad u∗), and the fixed
point algebra M α is L∞(T). In summary, the map θ : L∞(T) ∋ f 7→ Tf ∈ H ∞(N, P )
is a completely positive order isomorphism between the two operator systems, and
{v∗nTf Tgvn}∞
n=1 converges to Tf g in the strong operator topology for any f, g ∈
L∞(T).
We end this note with an example coming from random walks on discrete groups
discussed in [I2]. Let G be a discrete group, and let µ be a probability measure on
G whose support S generates G as a semigroup. We define a ucp map Pµ acting
on ℓ∞(G) by the right convolution operator Pµ(f ) = f ∗ µ where µ(g) = µ(g−1).
Then Pµ gives rise to a random walk on G with the transition probability p(g, h) :=
Pr(Xn+1 = hXn = g) given by µ(g−1h). We denote by H ∞(G, µ) the space of
24
MASAKI IZUMI
bounded harmonic functions for Pµ and by (∂G, ν) the Poisson boundary with the
harmonic measure for Pµ (see [K],[KV]) .
Qµ(a) =Xg∈G
As in [I2], we extend Pµ to N = B(ℓ2(G)) by
µ(g)ρgaρ−1
g ,
where ρ is the right regular representation. Then the noncommutative Poisson bound-
ary for Qµ is the boundary crossed product L∞(∂G, ν)⋊G (see [I2],[JN]). Let λ be the
left regular representation of G. Then H ∞(B(ℓ2(G)),Qµ) is spanned by H ∞(G, µ)λG.
In the rest, we identify the minimal dilation (M, p, α) of P = Qµ, and give a new
description of the boundary crossed product. Let θ : M α ∋ x 7→ pxp ∈ H ∞(N, P ) be
as before. Since λg is a unitary in the multiplicative domain of P , it is easy to show
the following lemma.
Lemma 5.3. Let ug = θ−1(λg). Then {ug}g∈G is a unitary representation of G in
M α commuting with p.
We may assume that M = B(H) and p is a projection onto a closed subspace H0
of H identified with ℓ2(G). By minimality, the Hilbert space H is spanned by
αn(egn,hn)αn−1(egn−1,hn−1)· · · α(eg1,h1)δg0,
where {eg,h}g,h∈G is the system of matrix units in B(ℓ2(G)) with respect to the
orthonormal basis {δg}g∈G. Thanks to Lemma 5.3 and eg,h = δgλgh−1, we have
αn(eg,h) = αn(δg)ugh−1. Thus we see that H is spanned by
ζ n(g0, g1, . . . , gn) = αn(δgn)αn−1(δgn−1)· · · α(δg1)δg0.
Direct computation shows
hζ n(g0, g1, . . . , gn), ζ n(h0, h1, . . . , hn)i
= δgn,hnhαn−1(pα(δgn)p)ζ n−1(g0, g1, . . . , gn−1), ζ n−1(h0, h1, . . . , hn−1)i
= δgn,hnhαn−1(δhn−1P (δgn)δgn−1)ζ n−2(g0, g1, . . . , gn−2), ζ n−2(h0, h1, . . . , hn−2)i
= δgn,hnδgn−1,hn−1p(gn−1, gn)
× hαn−1(δgn−1)ζ n−2(g0, g1, . . . , gn−2), ζ n−2(h0, h1, . . . , hn−2)i,
and therefore
We also have
hζ n(g0, g1, . . . , gn), ζ n(h0, h1, . . . , hn)i
= δg0,h0δg1,h1 · · · δgn,hnp(g0, g1)p(g1, g2)· · · p(gn−1, gn).
ζ n(g0, g1, . . . , gn) =Xg∈G
ζ n+1(g0, g1, . . . , gn, g).
Now we can see that H is identified with the L2-space over the path space G∞. Let
C n(g0, g1, . . . , gn) be the cylinder set
C n(g0, g1, . . . , gn) = {(xn)∞
n=0 ∈ G∞; xi = gi, 0 ≤ i ≤ n},
E0-SEMIGROUPS
25
and let m be the measure on G∞ determined by m(C 0(g)) = 1 and
m(C n(g0, g1, . . . , gn)) = p(g0, g1)p(g1, g2)· · · p(gn−1, gn).
Then L2(G∞, m) ∋ 1C n(g0,g1,...,gn) 7→ ζ n(g0, g1, . . . , gn) ∈ H gives a unitary operator,
and we identify the two Hilbert spaces in what follows. Note that δg ∈ H0 is identified
with 1C0(g). Since
ugζ n(g0, g1, . . . , gn) = ζ n(gg0, gg1, . . . , ggn),
we have ugξ((xn)) = ξ((g−1xn)) for all ξ ∈ L2(G∞, m).
Let Hn be the closed subspace of H spanned by {1C n(g0,g1,...,gn)}g0,g1,...,gn∈G, and
let pn be the projection onto Hn. Then we have p = p0. Let jk : ℓ∞(G) → B(H)
be the representation of ℓ∞(G) defined by jk(f )ξ((xn)) = f (xk)ξ((xn)). We identify
f ∈ ℓ∞(G) and λg in N with j0(f )p and ugp in M. Then we have αk(j0(f )p) =
jk(f )pk. Since M is generated by ∪k≥0αk(N), and N is spanned by ℓ∞(G)λG, the
homomorphism α is determined by the condition α(ug) = ug and α(jk(f )pk) =
jk+1(f )pk+1.
Since α is a unital homomorphism of M = B(H), it is implemented by a Cuntz
algebra representation. Let T : G∞ → G∞ be the time shift (T x)n = xn+1. For each
g ∈ S, we set
Sgξ((xn)) =
ξ ∈ L2(G∞, m).
Then Sg ∈ B(H) is an isometry with the adjoint operator given by
ξ ◦ T ((xn)),
δx0g,x1pµ(g)
S∗
g ξ((xn)) =pµ(g)ξ(x0g−1, x0, x1,· · · ),
ξ ∈ L2(G∞, m),
g ξ((xn)) = δx0g,x1ξ((xn)). The range projections {SgS∗
and so we have SgS∗
g}g∈S are
mutually orthogonal, and the summation converges to 1M . Thus {Sg}g∈S satisfy the
Cuntz algebra On relation with n = #S. Let
SgxS∗
g ,
x ∈ M.
β(x) =Xg∈S
Then it is easy to see β(ug) = ug, β(η) = η◦ T for η ∈ L∞(G∞, m), and β(pk) = pk+1.
Thus we get α = β.
The above argument shows that the fixed point algebra M α is the commutant of the
Cuntz algebra On = C ∗{Sg}g∈S. We specify a state of On giving this representation.
We claim that 1C0(e) is a separating vector for M α. Recall that H ∞(N, P ) is spanned
by H ∞(G, µ)λG. For f ∈ H ∞(G, µ), we have
hθ−1(f λg)1C0(e), 1C0(e)i = lim
= lim
k→∞hjk(f )pk1C0(g), 1C0(e)i = δg,0 lim
p(k)(e, h)f (h) = δg,0f (e),
= δg,0 lim
k→∞hαk(j0(f )p)ug1C0(e), 1C0(e)i
k→∞hjk(f )1C0(e), 1C0(e)i
k→∞Xh∈G
where p(k)(e, g) is the k-step transition probability. This shows that 1C0(e) induces a
faithful normal state of the boundary crossed product L∞(∂G, ν) ⋊ G corresponding
26
MASAKI IZUMI
to the harmonic measure ν (see [I2]), and so the claim is proved.
In consequence
1C0(e) is cyclic for On = C ∗{Sg}g∈S. We denote by ωµ the state of On given by 1C0(e).
Note that we have S∗
Corollary 5.4. There exists a state ωµ of the Cuntz algebra On = C ∗{Sg}g∈S given
by
g 1C0(h) =pµ(g)1C0(hg).
ωµ(Sg1Sg2 · · · SgkS∗
hlS∗
hl−1 · · · S∗
h1) =
kYi=1
µ(gi)1/2
lYj=1
µ(hj)1/2δg1g2···gk,h1h2···hl,
and the boundary crossed product L∞(∂G, ν) ⋊ G is isomorphic to πωµ(On)′, where
πωµ is the GNS representation of ωµ.
References for Appendix
[A] W. Arveson, A short course on spectral theory, Graduate Texts in Mathematics, 209. Springer-
Verlag, New York, 2002.
[CE] M. D. Choi, E. G. Effros, Injectivity and operator spaces, J. Functional Analysis, 24(1977),
156 -- 209.
[I1] M. Izumi, Non-commutative Poisson boundaries and compact quantum group actions, Adv.
Math. 169(2002), 1 -- 57.
[I2] M. Izumi, Non-commutative Poisson boundaries, Discrete geometric analysis, Contemp. Math.,
347, Amer. Math. Soc., Providence, RI, 2004, 69 -- 81.
[JN] W. Jaworski, M. Neufang, The Choquet-Deny equation in a Banach space, Canad. J.
Math., 59(2007), 795 -- 827.
[K] V. A. Kaimanovich, Measure-theoretic boundaries of Markov chains, 0-2 laws and entropy,
Harmonic Analysis and Discrete Potential Theory, (M. A. Picardello, Ed.), 145 -- 180, Plenum
Press, New York, 1992.
[KV] V. A. Kaimanovich, A. M. Vershik, Random walks on discrete groups: boundary and
entropy, Ann. Probab. 11(1983), 457 -- 490.
MASAKI IZUMI, Department of Mathematics, Graduate School of Science, Kyoto
University, Sakyo-ku, Kyoto 606-8502, Japan
E-mail address: [email protected]
|
1609.08920 | 2 | 1609 | 2017-02-20T07:35:48 | Amenability of locally compact quantum groups and their unitary co-representations | [
"math.OA",
"math.FA"
] | We prove that amenability of a unitary co-representation $U$ of a locally compact quantum group passes to unitary co-representations that weakly contain $U$. This generalizes a result of Bekka, and answers affirmatively a question of B\'edos, Conti and Tuset. As a corollary, we extend to locally compact quantum groups a result of the first-named author, which characterizes amenability of a locally compact group $G$ by nuclearity of the reduced group $C^{*}$-algebra $C_{r}^{*}(G)$ and an additional condition. | math.OA | math | AMENABILITY OF LOCALLY COMPACT QUANTUM GROUPS AND THEIR UNITARY
CO-REPRESENTATIONS
CHI-KEUNG NG AND AMI VISELTER
ABSTRACT. We prove that amenability of a unitary co-representation U of a locally compact quan-
tum group passes to unitary co-representations that weakly contain U. This generalizes a result
of Bekka, and answers affirmatively a question of Bédos, Conti and Tuset. As a corollary, we ex-
tend to locally compact quantum groups a result of the first-named author, which characterizes
amenability of a locally compact group G by nuclearity of the reduced group C ∗-algebra C ∗
r (G)
and an additional condition.
7
1
0
2
b
e
F
0
2
]
.
A
O
h
t
a
m
[
2
v
0
2
9
8
0
.
9
0
6
1
:
v
i
X
r
a
INTRODUCTION
A well-known result of Lance says that if a locally compact group G is amenable then its re-
duced group C ∗-algebra C ∗
r (G) is nuclear, and that the converse holds when G is discrete [Lan73,
Proposition 4.1 and Theorem 4.2] (but not generally; see Connes [Con76, Corollary 7]). It is
thus interesting to look for a condition whose combination with nuclearity of C ∗
r (G) is equiv-
alent to amenability of G for an arbitrary locally compact group G. This problem was solved
recently by the first-named author [Ng15, Theorem 8], who proved that G is amenable if and
r (G) is nuclear and possesses a tracial state. In Section 3 we extend this theorem to
only if C ∗
locally compact quantum groups in the sense of Kustermans and Vaes, replacing traciality with
a suitable "noncommutative" condition. Our result bears some resemblance to recent charac-
terizations of amenability in terms of various notions of injectivity, which have proven to admit
many applications [SV14, CN16, Cra16]. However, it focuses on the dual reduced C ∗-algebra
rather than the dual L∞-algebra.
A key tool used in our proof is that of amenability of (unitary) co-representations of locally
compact quantum groups. This notion was introduced for groups in the fundamental work of
Bekka [Bek90]. It is related to group amenability by the fact that a locally compact group G
is amenable if and only if every representation of G is amenable, if and only if the left regular
representation of G is amenable. Another useful result of [Bek90] asserts that if π1, π2 are repre-
sentations of G such that π1 is amenable and is weakly contained in π2, then π2 is also amenable.
Bédos, Conti and Tuset [BCT05] and Bédos and Tuset [BT03] introduced amenability of co-
representations of locally compact quantum groups, generalizing Bekka's notion. One question
they left open was whether amenability was well-behaved with respect to weak containment
as proved for groups by Bekka. We provide an affirmative answer to this question in Section
2. It is then employed to establish the main result of Section 3. We remark that amenability
of co-representations of Kac algebras was called "weak Bekka amenability" in [Ng01] up to a
difference in the convention of what a co-representation is.
1
2
CHI-KEUNG NG AND AMI VISELTER
1. PRELIMINARIES
For a (complex) Hilbert space H we denote by B(H), respectively K(H), the C ∗-algebra of
all bounded, respectively compact, operators on H. For ζ, η ∈ H we define ωζ,η ∈ B(H)∗ by
ωζ,η(x) := hxζ, ηi (x ∈ B(H)) and ωζ := ωζ,ζ. Representations of C ∗-algebras are assumed to be
nondegenerate. For a C ∗-algebra A, we write id for the identity map on A and 1 for the unit of
A, if exists. We denote by M(A) the multiplier algebra of A. For details on multiplier algebras,
the strict topology and related topics, consult [Lan95]. The symbols ⊗min and ⊗ stand for the
minimal tensor product of C ∗-algebras and the normal spatial tensor product of von Neumann
algebras, respectively. We will use terminology and results from operator space theory; see
[ER00] as a general reference.
A locally compact quantum group (abbreviated LCQG) is a pair G = (L∞(G), ∆), where L∞(G)
is a von Neumann algebra and ∆ is a co-multiplication, namely a normal unital ∗-homomorphism
from L∞(G) to L∞(G) ⊗ L∞(G) that is co-associative: (∆ ⊗ id)∆ = (id ⊗ ∆)∆, admitting a left-
invariant weight and a right-invariant weight [KV00, KV03, VD14]. The precise definition of
left/right invariance will not be needed here explicitly, and so we refer the reader to the above
references for details, as well as for the following facts. Each LCQG G has a dual LCQG, denoted
by G. The von Neumann algebras L∞(G) and L∞( G) act standardly on the same Hilbert space
L2(G). A very important object is the left regular co-representation of G, which is a multiplicative
unitary W ∈ L∞(G) ⊗ L∞( G) that satisfies ∆(x) = W ∗(1 ⊗ x)W for all x ∈ L∞(G). The
von Neumann algebra L∞(G) has a canonical weakly dense C ∗-subalgebra C0(G), and we have
W ∈ M(C0(G) ⊗min C0( G)). We write L1(G) for the predual L∞(G)∗.
A (unitary) co-representation of a LCQG G on a Hilbert space H is a unitary operator U ∈
M(C0(G) ⊗min K(H)) that satisfies (∆ ⊗ id)(U) = U13U23. The universal picture of G involves
0 ( G)) of G
another C ∗-algebra, C u
with the property that there is a bijection between co-representations U of G and representations
π of C u
0 (G). There exists a co-representation W ∈ M(C0(G) ⊗min C u
0 ( G) given by U = (id ⊗ π)(W). For this, see [Kus01].
The simplest examples of LCQGs are given by locally compact groups G. The associated L∞,
0 algebras are just L∞(G), C0(G) and C0(G), respectively, and ∆ maps a function
C0 and C u
f ∈ L∞(G) to the function ∆(f ) ∈ L∞(G) ⊗ L∞(G) ∼= L∞(G × G) given by (t, s) 7→ f (ts),
t, s ∈ G. The dual of G is the LCQG G whose associated algebras L∞( G), C0( G) and C u
0 ( G) are
r (G) and the full
the (left) group von Neumann algebra VN(G), the reduced group C ∗-algebra C ∗
group C ∗-algebra C ∗(G) of G, respectively, and the co-multiplication of G maps λg to λg ⊗ λg for
every g, where (λg)g∈G is the left regular representation of G.
A LCQG G is called compact if C0(G) is unital. This is equivalent to the left- and right-invariant
weights being equal and finite. A LCQG is called discrete if its dual is compact. See [Wor98, ER94,
VD96], and also [Run08] for the equivalence of different characterizations.
2. AMENABILITY OF CO-REPRESENTATIONS AND WEAK CONTAINMENT
In this section we extend an important result of Bekka [Bek90] to LCQG co-representations.
We begin with two definitions from [Ng01, BCT05, BT03].
AMENABILITY OF LOCALLY COMPACT QUANTUM GROUPS AND THEIR UNITARY CO-REPRESENTATIONS
3
Definition 2.1. A co-representation U of a LCQG G on a Hilbert space H is called left amenable,
respectively right amenable, if there exists a state m of B(H) such that m ((ω ⊗ id) (U ∗(1 ⊗ x)U)) =
ω(1)m(x), respectively m ((ω ⊗ id) (U(1 ⊗ x)U ∗)) = ω(1)m(x), for every x ∈ B(H) and ω ∈
L1(G). The state m is said to be a left-invariant, respectively right-invariant, mean of U.
Definition 2.2. For i = 1, 2, let Ui be a co-representation of a LCQG G on a Hilbert space Hi, and
0 ( G) on Hi. We say that U1 is weakly contained in
write πi for the associated representation of C u
U2 if π1 is weakly contained in π2 [Dix77, Section 3.4], that is, ker π2 ⊆ ker π1.
The following result is a generalization of [Bek90, Corollary 5.3]. It was proved for strong
instead of weak containment in [BCT05, Proposition 7.14], and for discrete quantum groups in
[BCT05, Corollary 9.7]. It is mentioned in [BCT05, p. 49] that the general case is open and of
interest. We will use it below to establish Theorem 3.2.
Theorem 2.3. Let U1, U2 be co-representations of a LCQG G. Suppose that U1 is weakly contained
in U2. If U1 is left (respectively, right) amenable, then so is U2.
We require the next lemma, which follows as a particular case from [Neu04]. For the reader's
convenience, we give its short proof.
Lemma 2.4. Let A, B, C be von Neumann algebras and Φ : B → C a completely bounded map.
Then there exists a unique completely bounded linear map id ⊗ Φ : A ⊗ B → A ⊗ C such that
(ω ⊗ id)((id ⊗ Φ)(X)) = Φ((ω ⊗ id)(X))
(∀X ∈ A ⊗ B, ω ∈ A∗).
(2.1)
It satisfies kid ⊗ Φkcb = kΦkcb.
Proof. Uniqueness is clear. The map CB(A∗, B) → CB(A∗, C) given by T 7→ Φ ◦ T for T ∈
CB(A∗, B) is evidently well defined and with cb-norm at most kΦkcb. Using the natural completely
isometric identifications A ⊗ B ∼= CB(A∗, B) and A ⊗ C ∼= CB(A∗, C) as operator spaces [ER00,
Theorem 7.2.4 and Proposition 7.1.2] we get a linear map id ⊗ Φ : A ⊗ B → A ⊗ C with cb-norm
at most kΦkcb that satisfies (2.1). Since (id ⊗ Φ)(a ⊗ b) = a ⊗ Φ(b) for all a ∈ A and b ∈ B, this
implies that kid ⊗ Φkcb = kΦkcb.
(cid:3)
Recall that a unital linear map between operator systems is completely positive if and only if
it is completely contractive.
Proof of Theorem 2.3. Let Ui be a co-representation of G on a Hilbert space Hi, and write πi
0 ( G) on Hi (i = 1, 2). By assumption, there exists a ∗-
for the associated representation of C u
homomorphism π : Im π2 → Im π1 given by π ◦ π2 = π1. Denote by π the (unique) extension
of π to a unital ∗-homomorphism M(Im π2) → M(Im π1) ⊆ B(H1) (actually, the extension of π
to the trivial unitization of Im π2 would suffice). Viewing π as a representation of M(Im π2) on
H1, we extend it to a unital completely positive map Φ : B(H2) → B(H1) by Arveson's extension
theorem. Consider now the map id ⊗ Φ : L∞(G) ⊗ B(H2) → L∞(G) ⊗ B(H1) given by Lemma
2.4, which is unital and completely positive as Φ is. The unitaries Ui = (id ⊗ πi)(W), i = 1, 2,
satisfy (id ⊗ Φ)(U2) = U1, because for every ω ∈ L1(G),
(ω ⊗id) ((id ⊗ Φ)(U2)) = Φ((ω ⊗id)(U2)) = (Φ◦π2)((ω ⊗id)(W)) = π1((ω ⊗id)(W)) = (ω ⊗id)(U1).
4
CHI-KEUNG NG AND AMI VISELTER
Hence U2 belongs to the multiplicative domain of id ⊗ Φ [Pau02, Theorem 3.18]. As a result, for
every x ∈ B(H2),
(id ⊗ Φ) (U ∗
2 (1 ⊗ x)U2) = (id ⊗ Φ)(U ∗
2 )(1 ⊗ Φ(x))(id ⊗ Φ)(U2) = U ∗
1 (1 ⊗ Φ(x))U1.
If now mU1 is a left-invariant mean for U1, then mU1 ◦ Φ is a left-invariant mean for U2, because
for every x ∈ B(H2) and ω ∈ L1(G),
(mU1 ◦ Φ) [(ω ⊗ id) (U ∗
2 (1 ⊗ x)U2)] = mU1(ω ⊗ id) [(id ⊗ Φ) (U ∗
2 (1 ⊗ x)U2)]
The proof for right amenability is similar.
(cid:3)
= mU1(ω ⊗ id) (U ∗
1 (1 ⊗ Φ(x))U1) = ω(1)(mU1 ◦ Φ)(x).
3. AMENABILITY OF LOCALLY COMPACT QUANTUM GROUPS
This section is devoted to Theorem 3.2 below, which generalizes the main result of [Ng15]. It
provides a condition that sits between amenability of a LCQG G and co-amenability of its dual.
When G is discrete (or a group), all three conditions are equivalent.
Definition 3.1 ([DQV02, BT03]). Let G be a LCQG.
(1) We say that G is amenable if it has a left-invariant mean, namely a state m of L∞(G) that
satisfies
m ((ω ⊗ id)∆(x)) = ω(1)m(x)
(∀x ∈ L∞(G), ω ∈ L1(G)).
(2) We say that G is co-amenable if there exists a state ǫ of C0(G) that satisfies (ǫ⊗id)(W ) = 1.
A locally compact group G is amenable if and only if it is amenable in the above sense when
viewed as a LCQG. For every LCQG G, co-amenability of G implies amenability of G [BT03,
Theorem 3.2]. The converse holds when G is a locally compact group (by Leptin's theorem) and
when G is discrete (see for instance [Tom06]). Whether it is true in general is arguably the most
important open question in LCQG amenability theory.
Theorem 3.2. Let G be a LCQG.
(1) Consider the following conditions:
(a) G is co-amenable;
(b) C0( G) is nuclear, and there exists a state ρ of C0( G) that is invariant under the left
C ∗-algebraic action of G on C0( G), i.e.,
(id ⊗ ρ) (W ∗(1 ⊗ x)W ) = ρ(x)1
(∀x ∈ C0( G));
(3.1)
(c) G is amenable.
Then (a) =⇒ (b) =⇒ (c).
(2) Moreover, if G has trivial scaling group (for instance, if G is a Kac algebra), then (a) =⇒
(b') =⇒ (b), where
(b') C0( G) is nuclear and admits a tracial state.
Remark 3.3. Observe that in contrast to the specific case of (locally compact) groups, the second
half of condition (b) is not intrinsic to the C ∗-algebra C0( G). When G is a locally compact group,
AMENABILITY OF LOCALLY COMPACT QUANTUM GROUPS AND THEIR UNITARY CO-REPRESENTATIONS
5
(3.1) is equivalent to ρ being tracial. When G is a discrete quantum group, the Haar state ρ of G
satisfies (3.1) if and only if G is a Kac algebra by [Izu02, Corollary 3.9 and its proof] (note the
difference in the conventions). Also, one cannot deduce from Theorem 3.2 that for discrete G,
nuclearity of C0( G) implies amenability of G -- whether this is true remains an open question.
Recall that the antipode of G is a generally unbounded operator κ over L∞(G) that satisfies
(id ⊗ ρ)(W ) ∈ D(κ) and κ((id ⊗ ρ)(W )) = (id ⊗ ρ)(W ∗)
(∀ρ ∈ L1( G)).
(3.2)
The antipode has a "polar decomposition" κ = R ◦ τ−i/2, where R is the unitary antipode, which
is an anti-automorphism of L∞(G), and τ is the scaling group. Thus, when τ is trivial, κ = R.
Lemma 3.4. Let G be a LCQG with trivial scaling group. For every θ ∈ C0( G)∗,
κ((id ⊗ θ)(W )) = (id ⊗ θ)(W ∗).
(3.3)
Proof. Fix θ ∈ C0( G)∗. Let θ ∈ L∞( G)∗ be a Hahn -- Banach extension of θ and (θi)i∈I be a net in
L1( G) that converges to θ in the σ(L∞( G)∗, L∞( G))-topology. For every ω ∈ L1(G), we have
θi (((ω ◦ κ) ⊗ id)(W ) − (ω ⊗ id)(W ∗)) −−→
i∈I
θ (((ω ◦ κ) ⊗ id)(W ) − (ω ⊗ id)(W ∗)) .
(3.4)
But by (3.2), the left-hand side of (3.4) is zero for all i ∈ I. We conclude that (ω◦κ) ((id ⊗ θ)(W )) =
ω ((id ⊗ θ)(W ∗)) for every ω ∈ L1(G). This gives (3.3).
(cid:3)
Remark 3.5. Essentially the same proof, replacing L1(G) by L1
♯ (G) and using [BDS13, Proposition
A.1], shows that for an arbitrary LCQG G (not necessarily with trivial scaling group), (3.2) holds
for all ρ ∈ C0( G)∗.
The next result generalizes one direction of [Izu02, Corollary 3.9] alluded to above.
Proposition 3.6. Let G be a LCQG with trivial scaling group. Then every tracial state ρ of C0( G)
satisfies (3.1).
Proof. Write (Hρ, πρ, ξρ) for the GNS construction of ρ. Since ρ is a trace, the formula Jπρ(a)ξρ :=
πρ(a∗)ξρ, a ∈ C0( G), defines an involutive anti-unitary J on Hρ. Let (ηi)i∈I be an orthonormal
basis of Hρ. Denote by F (I) the set of finite subsets of I directed by inclusion. By a standard
argument [KV00, Lemma A.5 and its proof], for every X, Y ∈ M(C0(G) ⊗min C0( G)), the net
Xi∈F
((id ⊗ ωξρ,ηi)(id ⊗ πρ)(Y ))∗(id ⊗ ωξρ,ηi)(id ⊗ πρ)(X)!F ∈F (I)
= Xi∈F
((id ⊗ (ωηi,ξρ ◦ πρ))(Y ∗))(id ⊗ (ωξρ,ηi ◦ πρ))(X)!F ∈F (I)
in M(C0(G)) ⊆ B(L2(G)) is bounded, and converges strongly to (id ⊗ ωξρ)(id ⊗ πρ)(Y ∗X) =
(id ⊗ ρ)(Y ∗X). Let x ∈ C0( G). Taking Y := (1 ⊗ x∗)W and X := W , we deduce that the bounded
net
Xi∈F
((id ⊗ (ωηi,ξρ ◦ πρ))(W ∗(1 ⊗ x)))(id ⊗ (ωξρ,ηi ◦ πρ))(W )!F ∈F (I)
6
CHI-KEUNG NG AND AMI VISELTER
converges strongly to (id ⊗ ρ)(W ∗(1 ⊗ x)W ). Thus, for every ω ∈ L1(G),
lim
F ∈F (I)
(ω◦κ)"Xi∈F
((id ⊗ (ωηi,ξρ ◦ πρ))(W ∗(1 ⊗ x)))(id ⊗ (ωξρ,ηi ◦ πρ))(W )# = (ω◦κ◦(id⊗ρ))(W ∗(1⊗x)W ).
By traciality of ρ, for every b, c ∈ C0( G), we have
(ωξρ,πρ(c)ξρ ◦ πρ)(b) = ρ(c∗b) = ρ(bc∗) = (ωJπρ(c)ξρ,ξρ ◦ πρ)(b).
As πρ(C0( G))ξρ is dense in Hρ, this entails that ωξρ,η ◦ πρ = ωJη,ξρ ◦ πρ for all η ∈ Hρ. Consequently,
by Lemma 3.4, for every i ∈ I,
κ(cid:2)((id ⊗ (ωηi,ξρ ◦ πρ))(W ∗(1 ⊗ x)))(id ⊗ (ωξρ,ηi ◦ πρ))(W )(cid:3)
= κ(cid:2)(id ⊗ (ωξρ,ηi ◦ πρ))(W )(cid:3) κ(cid:2)((id ⊗ (ωηi,ξρ ◦ πρ))(W ∗(1 ⊗ x)))(cid:3)
= (id ⊗ (ωξρ,ηi ◦ πρ))(W ∗)((id ⊗ (ωηi,ξρ ◦ πρ))(W (1 ⊗ x)))
= (id ⊗ (ωJηi,ξρ ◦ πρ))(W ∗)((id ⊗ (ωξρ,Jηi ◦ πρ))(W (1 ⊗ x))).
Since (ηi)i∈I is an orthonormal basis of Hρ and J is anti-unitary, (Jηi)i∈I is also an orthonormal
basis of Hρ. Using the analog of the above reasoning with (Jηi)i∈I in lieu of (ηi)i∈I and taking
Y := W and X := W (1 ⊗ x), the bounded net
Xi∈F
(id ⊗ (ωJηi,ξρ ◦ πρ))(W ∗)((id ⊗ (ωξρ,Jηi ◦ πρ))(W (1 ⊗ x)))!F ∈F (I)
converges strongly to (id ⊗ ρ)(W ∗W (1 ⊗ x)) = ρ(x)1. All in all, (ω ◦ κ ◦ (id ⊗ ρ))(W ∗(1 ⊗ x)W ) =
ρ(x)ω(1) for every ω ∈ L1(G), hence (κ ◦ (id ⊗ ρ))(W ∗(1 ⊗ x)W ) = ρ(x)1 = κ(ρ(x)1), proving
(3.1) by the injectivity of κ.
(cid:3)
We give two proofs of implication (b) =⇒ (c). The first one uses Theorem 2.3. The second is
shorter and more direct, but it basically uses the same idea.
Proof of Theorem 3.2. (a) =⇒ (b): assume that G is co-amenable. Then G is amenable, so that
C0( G) is nuclear by [BT03, Theorem 3.3]. Let ρ ∈ C0( G) be a state such that (id ⊗ ρ)(W ) = 1.
By a multiplicative domains argument, ρ is a character of C0( G) [BT03, Proof of Theorem 3.1],
from which (3.1) readily follows.
(b) =⇒ (c), first proof: suppose that such ρ exists. Writing (Hρ, πρ, ξρ) for the GNS construc-
tion of ρ and Wρ := (id ⊗ πρ)(W ), we get
(id ⊗ ωξρ)(cid:2)W ∗
ρ (1 ⊗ y)Wρ(cid:3) = ωξρ(y)1
for every y ∈ πρ(C0( G)), hence for every y in the von Neumann algebra M := πρ(C0( G))
⊆
B(Hρ). Since C0( G) is nuclear, M is injective by [Bla06, Theorem IV.2.2.13]. Let E be a condi-
tional expectation from B(Hρ) onto M. Notice that Wρ ∈ M(C0(G)⊗minπρ(C0( G))) ⊆ L∞(G)⊗M.
Precisely as in [SV14, Theorem 2.4] (or the relevant part of the proof of Theorem 2.3), for all
z ∈ B(Hρ) and ω ∈ L1(G) we have
weak
E(cid:0)(ω ⊗ id)(cid:2)W ∗
ρ (1 ⊗ z)Wρ(cid:3)(cid:1) = (ω ⊗ id)(cid:2)W ∗
ρ (1 ⊗ Ez)Wρ(cid:3) ,
AMENABILITY OF LOCALLY COMPACT QUANTUM GROUPS AND THEIR UNITARY CO-REPRESENTATIONS
7
thus
(ωξρ ◦ E)(cid:0)(ω ⊗ id)(cid:2)W ∗
ρ (1 ⊗ z)Wρ(cid:3)(cid:1) = (ω ⊗ ωξρ)(cid:2)W ∗
ρ (1 ⊗ Ez)Wρ(cid:3) = (ωξρ ◦ E)(z)ω(1).
In conclusion, the co-representation Wρ is left amenable with ωξρ ◦ E a left-invariant mean. By
definition, Wρ is weakly contained in W , so from Theorem 2.3 we infer that W is left amenable.
This evidently implies that G is amenable (in fact, the converse is also true by [BT03, Theorem
4.1]).
(b) =⇒ (c), second proof: embed C0( G) in C0( G)∗∗ canonically, and view ρ as a normal state
of C0( G)∗∗. We will regard W ∈ M(C0(G) ⊗min C0( G)) both as an element of C0(G)∗∗ ⊗ C0( G)∗∗
and as an element of C0(G)∗∗ ⊗ B(L2(G)) depending on the context. By a weak∗-continuity
argument, we have
(id ⊗ ρ) (W ∗(1 ⊗ y)W ) = ρ(y)1
(∀y ∈ C0( G)∗∗).
The assumption that C0( G) is nuclear is equivalent to C0( G)∗∗ being an injective operator system
[Bla06, Theorem IV.3.1.12]. Therefore, the embedding M(C0( G)) ֒→ C0( G)∗∗ extends to a unital
completely positive map Φ : B(L2(G)) → C0( G)∗∗. Consider the unital completely positive map
id ⊗ Φ : C0(G)∗∗ ⊗ B(L2(G)) → C0(G)∗∗ ⊗ C0( G)∗∗ given by Lemma 2.4. For every ω ∈ C0(G)∗,
(ω ⊗ id) ((id ⊗ Φ)(W )) = Φ ((ω ⊗ id)(W )) = (ω ⊗ id)(W )
because (ω ⊗ id)(W ) ∈ M(C0( G)). This means that (id ⊗ Φ)(W ) = W . Thus W belongs to
the multiplicative domain of id ⊗ Φ, so that (id ⊗ Φ) (W ∗XW ) = W ∗(id ⊗ Φ)(X)W for every
X ∈ C0(G)∗∗ ⊗ B(L2(G)). Let θ := ρ ◦ Φ. For every x ∈ B(L2(G)),
(id ⊗ θ) (W ∗(1 ⊗ x)W ) = (id ⊗ ρ) (W ∗(1 ⊗ Φ(x))W ) = θ(x)1.
Consequently, θL∞(G) is a left-invariant mean of G.
(a) =⇒ (b') for every LCQG G because every character is a tracial state. When G has trivial
(cid:3)
scaling group, (b') =⇒ (b) by Proposition 3.6.
For discrete quantum groups, amenability is equivalent to co-amenability of the dual, and we
thus have the following consequence of Theorem 3.2.
Corollary 3.7. Let G be a discrete quantum group. Then G is amenable if and only if condition (b)
of Theorem 3.2 holds.
We conjecture that condition (b) is, in fact, equivalent either to condition (a) or to condition
(c) for arbitrary LCQGs. However, we were not able to verify this.
Acknowledgement. The second-named author is indebted to Paweł Kasprzak and Adam Skalski
for intriguing conversations on the content of this paper and related topics, which took place
when he was visiting Adam Skalski in Warsaw, and for their helpful remarks. The problem of
extending the equivalence between amenability of G and nuclearity of C ∗
r (G) for discrete groups
G to discrete quantum groups was suggested to the second-named author several years ago by
Piotr M. Sołtan (a similar problem was subsequently addressed in [SV14]), and for that he is
grateful to him.
8
CHI-KEUNG NG AND AMI VISELTER
REFERENCES
[BCT05] E. Bédos, R. Conti, and L. Tuset, On amenability and co-amenability of algebraic quan-
tum groups and their corepresentations, Canad. J. Math. 57 (2005), no. 1, 17 -- 60.
[BT03] E. Bédos and L. Tuset, Amenability and co-amenability for locally compact quantum
groups, Internat. J. Math. 14 (2003), no. 8, 865 -- 884.
[Bek90] M. E. B. Bekka, Amenable unitary representations of locally compact groups, Invent.
Math. 100 (1990), no. 2, 383 -- 401.
[Bla06] B. Blackadar, Operator algebras, Encyclopaedia of Mathematical Sciences, vol. 122,
Springer-Verlag, Berlin, 2006, Theory of C ∗-algebras and von Neumann algebras, Op-
erator Algebras and Non-commutative Geometry, III.
[BDS13] M. Brannan, M. Daws, and E. Samei, Completely bounded representations of convolution
algebras of locally compact quantum groups, Münster J. Math. 6 (2013), 445 -- 482.
[Con76] A. Connes, Classification of injective factors. Cases II1, II∞, IIIλ, λ 6= 1, Ann. of Math.
(2) 104 (1976), no. 1, 73 -- 115.
[Cra16] J. Crann, Amenability and covariant injectivity of locally compact quantum groups II,
preprint, arXiv:1507.03296, 2016.
[CN16] J. Crann and M. Neufang, Amenability and covariant injectivity of locally compact quan-
tum groups, Trans. Amer. Math. Soc. 368 (2016), no. 1, 495 -- 513.
[DQV02] P. Desmedt, J. Quaegebeur, and S. Vaes, Amenability and the bicrossed product con-
struction, Illinois J. Math. 46 (2002), no. 4, 1259 -- 1277.
[Dix77] J. Dixmier, C ∗-algebras, North-Holland Mathematical Library, vol. 15, North-Holland
Publishing Co., Amsterdam-New York-Oxford, 1977.
[ER94] E. G. Effros and Z.-J. Ruan, Discrete quantum groups. I. The Haar measure, Internat. J.
Math. 5 (1994), no. 5, 681 -- 723.
[ER00]
, Operator spaces, London Mathematical Society Monographs. New Series,
vol. 23, Oxford University Press, 2000.
[Izu02] M. Izumi, Non-commutative Poisson boundaries and compact quantum group actions,
Adv. Math. 169 (2002), no. 1, 1 -- 57.
[Kus01] J. Kustermans, Locally compact quantum groups in the universal setting, Internat. J.
Math. 12 (2001), no. 3, 289 -- 338.
[KV00] J. Kustermans and S. Vaes, Locally compact quantum groups, Ann. Sci. École Norm. Sup.
(4) 33 (2000), no. 6, 837 -- 934.
[KV03]
, Locally compact quantum groups in the von Neumann algebraic setting, Math.
Scand. 92 (2003), no. 1, 68 -- 92.
[Lan73] C. Lance, On nuclear C ∗-algebras, J. Funct. Anal. 12 (1973), 157 -- 176.
[Lan95] E. C. Lance, Hilbert C ∗-modules. A toolkit for operator algebraists, London Mathematical
Society Lecture Note Series, vol. 210, Cambridge University Press, Cambridge, 1995.
[Neu04] M. Neufang, Amplification of completely bounded operators and Tomiyama's slice maps,
J. Funct. Anal. 207 (2004), no. 2, 300 -- 329.
AMENABILITY OF LOCALLY COMPACT QUANTUM GROUPS AND THEIR UNITARY CO-REPRESENTATIONS
9
[Ng01] C.-K. Ng, Amenable representations and Reiter's property for Kac algebras, J. Funct. Anal.
187 (2001), no. 1, 163 -- 182.
[Ng15]
, Strictly amenable representations of reduced group C ∗-algebras, Int. Math. Res.
Not. IMRN (2015), no. 17, 7853 -- 7860.
[Pau02] V. Paulsen, Completely bounded maps and operator algebras, Cambridge Studies in Ad-
vanced Mathematics, vol. 78, Cambridge University Press, Cambridge, 2002.
[Run08] V. Runde, Characterizations of compact and discrete quantum groups through second
duals, J. Operator Theory 60 (2008), no. 2, 415 -- 428.
[SV14] P. M. Sołtan and A. Viselter, A note on amenability of locally compact quantum groups,
Canad. Math. Bull. 57 (2014), no. 2, 424 -- 430.
[Tom06] R. Tomatsu, Amenable discrete quantum groups, J. Math. Soc. Japan 58 (2006), no. 4,
949 -- 964.
[VD96] A. Van Daele, Discrete quantum groups, J. Algebra 180 (1996), no. 2, 431 -- 444.
[VD14]
, Locally compact quantum groups. A von Neumann algebra approach, SIGMA
Symmetry Integrability Geom. Methods Appl. 10 (2014), Paper 082, 41 pp.
[Wor98] S. L. Woronowicz, Compact quantum groups, Symétries quantiques (Les Houches,
1995), North-Holland, Amsterdam, 1998, pp. 845 -- 884.
CHERN INSTITUTE OF MATHEMATICS AND LPMC, NANKAI UNIVERSITY, TIANJIN 300071, CHINA
E-mail address: [email protected]
DEPARTMENT OF MATHEMATICS, UNIVERSITY OF HAIFA, 31905 HAIFA, ISRAEL
E-mail address: [email protected]
|
1512.00093 | 1 | 1512 | 2015-11-30T23:46:12 | Cocycle superrigidity for coinduced actions | [
"math.OA",
"math.DS"
] | We prove a cocycle superrigidity theorem for a large class of coinduced actions. In particular, if $\Lambda$ is a subgroup of a countable group $\Gamma$, we consider a probability measure preserving action $\Lambda\curvearrowright X_0$ and let $\Gamma\curvearrowright X$ be the coinduced action. Assume either that $\Gamma$ has property (T) or that $\Lambda$ is amenable and $\Gamma$ is a product of non-amenable groups. Using Popa's deformation/rigidity theory we prove $\Gamma\curvearrowright X$ is $\mathcal U_{fin}$-cocycle superrigid, that is any cocycle for this action to a $\mathcal U_{fin}$ (e.g. countable) group $\mathcal V$ is cohomologous to a homomorphism from $\Gamma$ to $\mathcal V.$ | math.OA | math |
COCYCLE SUPERRIGIDITY FOR COINDUNCED ACTIONS
DANIEL DRIMBE
Abstract. We prove a cocycle superrigidity theorem for a large class of coinduced actions.
In
particular, if Λ is a subgroup of a countable group Γ, we consider a probability measure preserving
action Λ y X0 and let Γ y X be the coinduced action. Assume either that Γ has property (T) or
that Λ is amenable and Γ is a product of non-amenable groups. Using Popa’s deformation/rigidity
theory we prove Γ y X is Uf in-cocycle superrigid, that is any cocycle for this action to a Uf in (e.g.
countable) group V is cohomologous to a homomorphism from Γ to V.
1. Introduction and statement of main results
1.1. Introduction. The goal of this article is to prove a general cocycle superrigidity theorem for
coinduced actions (see Definition 1.1) and derive several consequences to orbit equivalence and von
Neumann algebras.
The classification of probability measure preserving (pmp) actions of countable groups on standard
probability spaces up to orbit equivalence has attracted a lot of interest in the last 15 years (see
the surveys [Po07, Fu09, Ga10, Va10a, Io12a]). Two pmp actions Γ y (X, µ) and Λ y (Y, ν) are
orbit equivalent (OE) if there exists a measure space isomorphism f : X → Y which sends orbits
to orbits, i.e. f (Γx) = Λf (x), for almost every x ∈ X.
If the groups are amenable, the classification up to orbit equivalence is done. More precisely,
Orstein and Weiss proved in [OW80] (see also [Dy58, ?]) that all free ergodic pmp actions of
countable amenable groups are orbit equivalent. In contrast, the non-amenable case is much more
challenging and complex. Remarkably, several classes which are rigid in the sense that one can
deduce conjugacy from OE, have been discovered. The most extreme form of rigidity for orbit
equivalence is OE-superigidity: Γ y X is OE-superrigid if every free ergodic pmp action which is
OE with Γ y X is conjugate with it. The first OE-superrigidity result was obtained by Furman in
the late 1990s by building on Zimmer’s cocycle superrigidity [Zi84]. He showed that many actions
of higher rank lattices, including the action SLn(Z) y Tn, for n ≥ 3 is OE-superrigid [Fu98, Fu99].
After this, a number of striking OE-superrigidity results were obtained [MS02, Po05, Po06, Ki06,
Io08, PV08, Ki09, PS09, Io14, TD14, CK15].
In particular, in his breakthrough work [Po05, Po06], Popa used his deformation/rigidity theory to
prove a remarkable cocycle superrigidity theorem for Bernoulli actions of groups with property (T)
and of products of non-amenable groups. More precisely, if Γ y X is such an action, Popa obtained
that every cocycle with values in a countable (and more generally, in a Uf in) group is cohomologous
with a group homomorphism. By applying his cocycle superrigidity theorem to cocycles arising
from orbit equivalence, he proved that the action Γ y X is OE-superrigid.
1.2. Statement of the main results. Our main result provides a generalization of Popa’s cocycle
superrigidity theorem to coinduced actions. We first review some basic concepts starting with the
construction of coinduced actions (see e.g. [Io06b]).
1
2
DANIEL DRIMBE
Definition 1.1. Let Γ be a countable group and let Λ be a subgroup. Let φ : Γ/Λ → Γ be a
section. Define the cocycle c : Γ × Γ/Λ → Λ by the formula c(g, x) = φ−1(gx)gφ(x), for all g ∈ Γ
and x ∈ Γ/Λ.
Let Λ
Γ
σ0y X0 be a pmp action, where (X0, µ0) is a standard probability space. We define an action
σ
y X Γ/Λ
0
, called the coinduced action of σ0, as follows:
σg((xh)h∈Γ/Λ) = (x′
h)h∈Γ/Λ,
where (xh)h∈Γ/Λ ∈ X Γ/Λ
0
probability space X Γ/Λ
0
, where X Γ/Λ
0
is endowed with the product measure µΓ/Λ
.
0
and x′
h = c(g, h)xg−1h. Note that σ is a pmp action of Γ on the standard
Remark 1.2. If we consider the trivial action of Λ = {e} on X0, then the coinduced action of Γ
on X Γ/{e}
0 is the Bernoulli action.
= X Γ
0
We say that the inclusion Γ0 ⊂ Γ of countable groups has the Kazhdan’s relative property (T) if for
every ǫ > 0, there exist δ > 0 and F ⊂ Γ finite such that if π : Γ → U (K) is a unitary representation
and ξ ∈ K is a unit vector satisfying kπ(g)ξ − ξk < δ, for all g ∈ F , then there exists ξ0 ∈ K such
that kξ − ξ0k < ǫ and π(h)ξ0 = ξ0, for all h ∈ Γ0. The group Γ has the property (T) if the inclusion
Γ ⊂ Γ has the relative property (T). To give some example, Z2 ⊂ Z2 ⋊ SL2(Z) has the relative
property (T) and SLn(Z), n ≥ 3, has the property (T) [Ka67, Ma82].
An infinite subgroup H of Γ is w-normal in Γ if there exist an ordinal β and intermediate subgroups
H = H0 ⊂ H1 ⊂ · · · ⊂ Hβ = Γ such that for all 0 < α ≤ β, the group ∪α′<αHα′ is normal in Hα.
Denote by Uf in the class of Polish groups which arise as closed subgroups of the unitary groups
of II1 factors. In particular, all countable discrete groups and all compact Polish groups belong to
Uf in. These two notions are due to Popa [Po05].
For a Polish group G, a measurable map w : Γ × X → G is called a cocycle if it satisfies the relation
w(γ1γ2, x) = w(γ1, γ2x)w(γ2, x), for all γ1, γ2 ∈ Γ and for almost every x ∈ X. Two cocycles
w, w′ : Γ × X → G are cohomologous if there exists a measurable map φ : X → G such that
w′(γ, x) = φ(γx)w(γ, x)φ(x)−1, for all γ ∈ Γ and for almost every x ∈ X. An action Γ y (X, µ)
is called Uf in-cocycle superrigid if every cocycle with values in a group from Uf in is cohomologous
with a group homomorphism.
The following theorem is our first main result, which generalizes Popa’s cocycle superrigidity
theorem for Bernoulli actions of property (T) groups to coinduced actions (see [Po05] and also
[Fu06, Va06]).
Theorem A (Groups with relative property (T)). Let Γ be a countable group and Λ be a subgroup.
Let H ⊂ Γ be a subgroup with relative property (T). Assume that there does not exist a finite index
subgroup H0 of H which is contained in a conjugate g−1Λg of Λ, for some g ∈ Γ.
Take V ∈ Uf in. Let σ0 be a pmp action of Λ on a standard probability space (X0, µ0) and σ the
coinduced action of Γ on X := X Γ/Λ
Then, any cocycle w : Γ × X → V for the restriction of σ to H is cohomologous to a group
homomorphism d : H → V.
Moreover, if H is w-normal in Γ, then w is cohomologous to a group homomorphism d : Γ → V
and therefore Γ y X is Uf in-cocycle superrigid.
.
0
In particular, Theorem A implies that if Γ has property (T) (e.g. Γ = SLn(Z), n ≥ 3) and Λ is an
infinite index subgroup of Γ (e.g. Λ is cyclic), then any coinduced action of Γ from Λ is Uf in-cocycle
superrigid.
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
3
In [Po06, Corollary 1.2], Popa proved a cocycle superrigidity theorem for the Bernoulli action of
product groups analogous with [Po05, Corollary 5.4]. The next theorem generalizes this result to
coinduced actions.
Theorem B (Product groups). Let Γ be a countable group and Λ be an amenable subgroup. Let
H and H ′ be infinite commuting subgroups of Γ such that H ′ is non-amenable. Assume that there
does not exist a finite index subgroup H0 of H which is contained in a conjugate g−1Λg of Λ, for
some g ∈ Γ.
Take V ∈ Uf in. Let σ0 be a pmp action of Λ on a standard probability space (X0, µ0) and σ the
coinduced action of Γ on X := X Γ/Λ
Then, any cocycle w : Γ × X → V for the restriction of σ to HH ′ is cohomologous to a group
homomorphism d : HH ′ → V.
Moreover, if H is w-normal in Γ, then w is cohomologous to a group homomorphism d : Γ → V
and therefore Γ y X is Uf in-cocycle superrigid.
.
0
The proof of Theorem B goes along the same lines as the proof of [Po06, Theorem 4.1]. First, we
untwist the cocycle on H using the rigidity gained from the non-amenability of H ′ (instead of using
property (T) as in Theorem A). Then, using weak mixing properties of coinduced actions and the
fact that H is normal in HH ′, we are able to untwist the cocycle on HH ′.
We will prove in this paper a more general version of Theorems A and B dealing with coinduced
actions of Γ on AΓ/Λ that arise from actions of Λ on arbitrary tracial von Neumann algebras A.
As an immediate consequence of Theorems A and B, we deduce the following OE-superrigidity
result for coinduced actions.
Corollary 1.3 (OE-superrigidity). Let Γ be a countable subgroup with no non-trivial finite normal
subgroups and Λ a subgroup. Let H ⊂ Γ be a w-normal subgroup. Assume that there does not exist
a finite index subgroup H0 of H which is contained in a conjugate g−1Λg of Λ, for some g ∈ Γ.
Assume either that H has the relative property (T) or that Λ is amenable and there exists a non-
amenable subgroup of Γ which commutes with H.
Let σ0 be a pmp action of Λ on a standard probability space (X0, µ0) and σ the coinduced action of
Γ on X := X Γ/Λ
σ
y X is free, then it is OE-superrigid.
. If Γ
0
We need in Corrolary 1.3 the freeness assumption of the coinduced action since the proof uses
Proposition 5.2. See Lemma 5.3 for a large class of coinduced actions that are free. In particular,
if ∩g∈ΓgΛg−1 = {e} and (X0, µ0) is non-atomic, then Γ y X is free.
Corrolary1.3 proves for example that any coinduced action of SL3(Z) from a cyclic subgroup is
OE-superrigid. We contrast this with Bowen’s OE-flexibility results for coinduced actions [B10].
In particular, he proved that any two coinduced actions of F2 = Z ∗ Z from one of the copies of Z
are OE. Thus, any coinduced action of F2 from one of the copies of Z is not OE-superrigid.
1.3. Applications to W∗-superrigidity. For every measure preserving action Γ y X of a count-
able group Γ on a standard probability space, we associate the group measure space von Neumann
algebra L∞(X) ⋊ Γ [MvN36]. If the action Γ y X is free, ergodic and pmp, then L∞(X) ⋊ Γ is
a II1 factor which contains L∞(X) as a Cartan subalgebra, i.e. a maximal abelian von Neumann
algebra whose normalizer generates L∞(X) ⋊ Γ.
Two pmp actions Γ y (X, µ) and Λ y (Y, ν) on two standard probability spaces (X, µ) and
(Y, ν) are said to be W∗-equivalent if L∞(X) ⋊ Γ is isomorphic with L∞(Y ) ⋊ Λ. It can be seen
that orbit equivalence is stronger than W∗-equivalence. Moreover, Singer proved in [Si55] that
4
DANIEL DRIMBE
two free ergodic pmp actions Γ y (X, µ) and Λ y (Y, ν) are orbit equivalent if and only if
they are W∗-equivalent via an isomorphism which identifies the Cartan subalgebras L∞(X) and
L∞(Y ). The action Γ y (X, µ) is W∗-superrigid if whenever Λ y (Y, ν) is a free ergodic measure
preserving action W∗-equivalent with Γ y (X, µ), then the two actions are conjugate. Therefore,
W∗-superrigidity for an action Γ y X integrates two different rigidity aspects, which are hard to
obtain: OE-superrigidity and uniqueness of group measure space Cartan subalgebras. The latter
means that whenever M = L∞(X) ⋊ Γ = L∞(Y ) ⋊ Λ, then the Cartan subalgebras L∞(X) and
L∞(Y ) are unitarily conjugate in M.
A few years ago, the first example of virtually W∗-superrigid actions (i.e. conjugacy is obtained up
to finite index subgroups) were found in [Pe09] building on results of [Io08,OP08]. Soon after, Popa
and Vaes discovered the first concrete families of W∗-superrigid actions [PV09] and Ioana proved
that Bernoulli actions of icc property (T) groups are W∗-superrigid [Io10]. Subsequently, several
other classes of W∗-superrigid actions have been found in [FV10,CP10,HPV10,Io10,IPV10,Va10b,
CS11, CSU11, PV11, PV12, Bo12, CIK13, CK15]. By applying Theorems A and B we will deduce
W∗-superrigidity for a large class of coinduced actions. To obtain these examples, we will use
several results in the literature which prove uniqueness of group measure space Cartan subalgebras
for various classes of groups.
We denote by C the class of all countable groups Γ which satisfy one of the following conditions:
(1)
[CP10] Γ = Γ1 × Γ2, where Γi is icc and admits an unbounded cocycle into a mixing
representation and a non-amenable icc subgroup with the relative property (T), for i ∈
{1, 2};
(2) [PV11,PV12] Γ = Γ1 × Γ2 × ... × Γn is a finite product of non-elementary hyperbolic groups
with n ≥ 2;
[Io12b] Γ is a finite product of groups of the form Γ1 ∗Σ Γ2, each one of them satisfying:
(3)
• [Γ1 : Σ] ≥ 2, [Γ2 : Σ] ≥ 3;
• there exist g1, g2, ..., gn ∈ Γ such that ∩n
i=1giΣg−1
is finite.
1 ∗Σ0 Γ0
In addition, we assume than one of the factors Γ0
2 of Γ satisfies the conditions:
1 has property (T) and Σ0 is a normal subgroup of Γ0
Γ0
2.
i
If Γ ∈ C satisfies condition (i), we say that Γ ∈ Ci, whenever i ∈ {1, 2, 3}. For Γ ∈ C, we fix a
subgroup Λ satisfying the following:
(1) If Γ ∈ C1, take Λ an amenable subgroup of Γ1;
(2) If Γ ∈ C2, take Λ an amenable subgroup of one of the factors which appears in Γ;
(3) If Γ ∈ C3, take Λ such that Σ0 does not have a finite index subgroup which is contained in
a conjugate of Λ (e.g. Λ can be taken to be the commutant of Σ0 in Γ0
2).
Theorems A and B combined with [CP10, Corollary 5.3] [PV12, Theorem 1.1] [Io12b, Theorem 1.1]
give us the following W∗-superrigidity result.
Corollary 1.4. Let Γ ∈ C a group with no non-trivial finite normal subgroups and Λ a subgroup
chosen as before. Let Λ y X0 be a pmp action on a standard probability space X0 and let Γ y X
be the coinduced action of Λ y X0. If Γ y X is free, then it is W∗-superrigid.
Example 1.5. If we take Γ = Γ1 ∗Σ (Σ × Λ) ∈ C3, Corollary 1.4 gives another proof of W∗-
superrigidity for the coinduced action proved in [PV09, Example 6.9].
1.4. Acknowledgements. I am very grateful to my advisor Adrian Ioana for suggesting this
problem to me and for all the help given through many valuable discussions, important advice and
great support. I would also like to thank R´emi Boutonnet for important remarks and to Daniel
Hoff for helpful comments about the paper.
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
5
2. Preliminaries and cocycle rigidity
At the beginning of this section we review some basic tools of Popa concerning cocycles and continue
by introducing the free malleable deformation for Bernoulli actions. The last point will be a cocycle
rigidity result of Popa adapted to the context of the free malleable deformations.
2.1. Notations. Recall that a tracial von Neumann algebra (A, τ0) is a von Neumann algebra A
endowed with a normal faithful tracial state τ0. We denote by kxk2 = τ0(x∗x)
2 the induced Hilbert
norm on A, for an element x ∈ A. A von Neumann algebra is finite if and only if is tracial.
1
• We denote by ⊕i∈IHi, the Hilbert direct sum of a family {Hi}i∈I of Hilbert spaces and by
H1 ⊗ H2, the Hilbert tensor product of two Hilbert spaces H1 and H2.
• Let I be a nonempty set and (A, τ0) be a tracial von Neumann algebra. We denote by AI , the
tensor product ¯⊗I A, which is again a tracial von Neumann algebra. If we take ai ∈ A, i ∈ I,
such that {i ∈ Iai 6= 1} is finite, we use the notation ⊗i∈I ai for an elementary tensor. We
denote by supp(⊗i∈I ai) the set {i ∈ Iai 6= 1}, called the support of ⊗i∈I ai. AI has the
trace τ I
0 (⊗i∈I ai) = Qi∈I τ0(ai), where ⊗i∈Iai is an elementary tensor.
• For J ⊂ I, we have the canonical embedding AJ ⊂ AI , which takes an elementary tensor
0 given by τ I
⊗j∈J aj to ⊗i∈I ai, where ai = 1 if i /∈ J.
2.2. Perturbation of cocycles, property (T) and extensions. Let σ be a trace preserving
action of Γ on a tracial von Neumann algebra P . A map w : Γ → U (P ) is called a cocycle if
wgh = wgσg(wh), for all g, h ∈ Γ. Two cocycles w, w′ : Γ → U (P ) are called cohomologous if there
exists a unitary v ∈ P such that wgσg(v) = vw′
Lemma 2.1. ( [Po05, Lemma 2.12]) Let w, w′ be cocycles for a trace preserving action σ of a
group Γ on a tracial von Neumann algebra Q. The following statements are true:
g, for all g ∈ Γ.
(1) If there exists δ > 0 such that kwg − w′
gk2 ≤ δ, for all g ∈ Γ, then there exists a partial
isometry v ∈ Q such that kv − 1k2 ≤ 4δ1/2 and wgσg(v) = vw′
g, for all g ∈ Γ.
(2) If for any ǫ > 0 there exists u ∈ U (Q) such that kwgσg(u) − uw′
gk2 ≤ ǫ, for all g ∈ Γ, then
w and w′ are cohomologous.
(3) If w and w′ are cohomologous and v ∈ Q is a partial isometry satisfying wgσg(v) = vw′
g, ∀g ∈
Γ, then there exists u ∈ U (Q) such that uv∗v = v and wgσg(u) = uw′
g, ∀g ∈ Γ.
Let σ be a trace preserving action of a countable group Γ on a tracial von Neumann algebra Q.
Take w : Γ → U (Q) a cocycle. Let δ be a positive real number and a finite subset F of Γ. Denote
Ωw(δ, F ) = {w : Γ → U (Q)kwg − w′
gk2 ≤ δ, ∀g ∈ F }. Assuming this context, we have the following
result:
Lemma 2.2. ( [Po05, Lemma 4.2]) Let H ⊂ Γ be a subgroup with the relative property (T). Then
for every cocycle w : Γ → U (Q) and ǫ > 0, there exist δ > 0 and F a finite subset of Γ such that
∀w′ ∈ Ωw(δ, F ), ∃v ∈ Q partial isometry satisfying kv − 1k2 ≤ ǫ and w′
hσh(v) = vwh, ∀h ∈ H.
Definition 2.3. Let Γ be a countable group and σ be a trace preserving action on a tracial von
Neumann algebra (P, τ ). The action σ is weak mixing if for every ǫ > 0 and finite subset F of P ⊖C,
there exists g ∈ Γ such that τ (y∗σg(x) ≤ ǫ, for all x, y ∈ F.
Proposition 2.4. ( [Po05, Proposition 3.6]) Let σ and σ′ be trace preserving actions of a countable
group Γ on tracial von Neumann algebras P and N and let w be a cocycle for σ ⊗ σ′. Let H ⊂ Γ
be an infinite normal subgroup and assume that σ is weak mixing on H. If wh ∈ N, for all h ∈ H,
then wg ∈ N, for all g ∈ Γ.
6
DANIEL DRIMBE
2.3. Coinduced actions for tracial von Neumann algebras and the free product defor-
mation. The coinduced action for tracial von Neumann algebras is defined as in Section 1.2. More
precisely, let Γ be a countable group and let Λ be a subgroup. Let φ : Γ/Λ → Γ be a section. Define
the cocycle c : Γ × Γ/Λ → Λ by the formula
c(g, x) = φ−1(gx)gφ(x),
for all g ∈ Γ and x ∈ Γ/Λ.
Let Λ
define an action Γ
σ0y (A, τ0) be a trace preserving action, where (A, τ0) is a tracial von Neumann algebra. We
σ
y AΓ/Λ, called the coinduced action of σ0, as follows:
σg((ah)h∈Γ/Λ) = (a′
h)h∈Γ/Λ,
where a′
Note that σ is a trace preserving action of Γ on the tracial von Neumann algebra AΓ/Λ.
h = c(g, h)ag−1h.
σ0y (X0, µ0) be a pmp action, where (X0, µ0) is a standard probability space.
Remark 2.5. Let Λ
We consider the associated action of Λ on L∞(X0, µ0). On one hand, we obtain an coinduced action
Γ
. Note that σ is precisely
0
the usual coinduced action of Γ obtained from the action of Λ on X0.
σ
y L∞(X0, µ0)Γ/Λ. We also call σ, the associate action of Γ on X Γ/Λ
In [Io06a], Ioana introduced a malleable deformation for general Bernoulli actions, where the base
is any tracial von Neumann algebra. This is a variant of the malleable deformation discovered by
Popa [Po03] in the case of Bernoulli actions with abelian or hyperfinite base. Here we adapt the
deformation of [Io06a] to the context of general coinduced actions.
σ0y A be a trace preserving action. Take Γ
Let Γ be a countable group and Λ be a subgroup. Let A be a tracial von Neumann algebra and
σ
y AΓ/Λ the corresponding coinduced action. Let σ′
Λ
be a trace preserving action of Γ on another tracial von Neumann algebra (N, τ ′).
Denote by A the tracial von Neumann algebra A ∗ L(Z), which is the free product of A and L(Z).
Take u ∈ L(Z) the canonical generating Haar unitary. Let h = h∗ ∈ L(Z) be such that u = exp(ih)
and set ut = exp(ith) for all t ∈ R. Denote by P = AΓ/Λ and P = AΓ/Λ the tensor product von
Neumann algebras and define θ : R → Aut( P ) by
θt(⊗h∈Γ/Λah) = ⊗h∈Γ/Λ Ad(ut)(ah),
where ⊗h∈Γ/Λah ∈ P is an elementary tensor.
We observe that θt extends naturally as an automorphism of P ¯⊗N . Define also β ∈ Aut( P ¯⊗N ) by
βP ¯⊗N = idP ¯⊗N and β(⊗h∈F u) = ⊗h∈F u∗, for all finite subsets F of Γ/Λ.
Notice that the action σ extends naturally to an action σ on P by letting σg(⊗h∈F u) = ⊗h∈F u, for
all finite subsets F of Γ/Λ. We denote by ρ the tensor product action σ ⊗ σ′ of Γ on P ¯⊗N and by
ρ the tensor product action σ ⊗ σ′ of Γ on P ¯⊗N .
Remark 2.6. Notice that ρ commutes with the automorphims β and θt for all t. Thus, we can
consider β and θt as automorphisms of (P ¯⊗N ) ⋊ Γ and ( P ¯⊗N ) ⋊ Γ, by extending them in a natural
way. Also note that βθt = θ−tβ and β2 = id.
2.4. Finite union of translates of a subgroup and a fixed point lemma.
Lemma 2.7. Let H be a group and Hi subgroups, for 1 ≤ i ≤ n. Suppose that there exist finite
subsets Fi of H such that
H = ∪n
i=1FiHi.
Then there exists i ∈ {1, 2, ..., n} such that Hi is a subgroup of finite index in H.
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
7
Proof. We will proceed by induction over n. For n = 1 it is clear. Let us suppose the statement is
true for n − 1 and prove it for n. We consider the case where Hn is a subgroup of infinite index in
H, otherwise we are done.
Let us write a partition of H via the infinite index subgroup Hn:
H = FnHn ∪ (∪∞
k=1hkHn),
where h−1
j hi /∈ Hn, for all i 6= j and h−1
i=1hiHn ⊂ ∪n−1
k h0 /∈ Hn for all k ≥ 1 and h0 ∈ F.
Then, ∪∞
obtain that H can be also written as finite union of translates of ∪n−1
i=1 FiHi. Since H can be written as finite union of translates of ∪∞
i=1 FiHi. Thus,
i=1hiHn, we
H = ∪n−1
i=1 F ′
i Hi,
with F ′
least one of the Hi’s is a subgroup of finite index in H for an i ∈ {1, 2, ..., n}.
i some finite subsets of H. Now we can apply the induction hypothesis and conclude that at
(cid:3)
Remark 2.8. The following proposition is a consequence of [PV06, Lemma 2.4], but we include a
proof for the reader’s convenience.
Proposition 2.9. Let Γ be a countable group and Λ a subgroup. Let H be another subgroup of Γ.
Then there exists a finite set F ⊂ Γ/Λ such that gF ∩ F 6= ∅ for all g ∈ H if and only if there exists
a subgroup H0 of finite index of H such that H0 is contained in a conjugate g−1Λg of Λ.
j
.
i,j=1fiΛf −1
Proof. Let us suppose that there exists a finite set F ⊂ Γ/Λ such that gF ∩ F 6= ∅ for all g ∈ H.
Let F = {f1, f2, . . . fn}. Then for all h ∈ H, there exist i, j ∈ {1, 2, ..., n} such that hfjΛ = fiΛ.
We obtain that H ⊂ ∪n
Let Hij := {h ∈ HhfjΛ = fiΛ} and notice that H = ∪n
i,j=1Hij. For i 6= j, if Hij 6= ∅, take gij ∈ Hij
an arbitrary element. Observe that Hij = gijHjj. For i 6= j, if Hij = ∅, choose gij to be the neutral
element. This allows us to write H in the form H = ∪n
i,j=1gijHjj which is sufficient for applying
Lemma 2.7, where gii is the neutral element for all i ∈ {1, 2, ..., n}.
Notice that Hii = H ∩ fiΛf −1
of Lemma 2.7.
and at least one of these subgroups is of finite index in H because
i
The converse is easy. This finishes the proof.
(cid:3)
For the following lemma we use the notations from Section 2.3.
Lemma 2.10. Let H be a subgroup of Γ. Assume that there does not exist a subgroup H0 of finite
index in H such that H0 is contained in a conjugate g−1Λg of Λ. Let wh and w′
h be arbitrary elements
in P ¯⊗N , for all h ∈ H, and define the map α : H → B(L2( P ¯⊗N )) by αh(x) = γ(w′
h)ρh(x)wh,
where γ ∈ {id, θ1}. Let S be the k · k2-closed linear subspace of P generated by γ(P )P . Then
{ξ ∈ P ¯⊗N )αh(ξ) = ξ, ∀h ∈ H} ⊂ S ⊗ L2(N ).
Proof. We begin the proof with a claim which will prove the lemma.
Claim. For any ǫ > 0 and ξ, η ∈ P ¯⊗N with ξ, η ⊥ S ¯⊗N , there exists h ∈ H such that
hξ, αh(η)i ≤ ǫkξk2kηk2.
To prove the claim, we can assume kξk2 = kηk2 = 1. Let us take ξ0, η0 ∈ P ¯⊗N with k · k2 norm
smaller than 1 and F a finite subset of Γ/Λ such that
kξ − ξ0k2 ≤ ǫ/2,
ξ0 =
n
X
i=1
pi ⊗ ni,
pi ∈ AF ⊂ P , ni ∈ N,
pi ⊥ S, ∀i ∈ {1, 2, ..., n}.
8
and
DANIEL DRIMBE
kη − η0k2 ≤ ǫ/2,
η0 =
n
X
i=1
qi ⊗ mi,
qi ∈ AF ⊂ P , mi ∈ N,
qi ⊥ S, ∀i ∈ {1, 2, ..., n}.
Proposition 2.9 allows us to take h ∈ H, such that hF ∩ F = ∅. By the triangle inequality we have
hξ, αh(η)i ≤ hξ − ξ0, αh(η)i + hξ0, αh(η − η0)i + hξ0, αh(η0)i
≤ ǫ/2 + ǫ/2 + hξ0, αh(η0)i.
We will prove the claim if we show that
hξ0, αh(η0)i = hξ0, γ(w′
h)ρh(η0)whi = 0.
By linearity and continuity (weak operator topology) we may suppose that wh = ⊗F ′aj ⊗ n, w′
h =
j ⊗ n′ ∈ P ¯⊗N are elementary tensors with F ′ ⊂ Γ/Λ a finite subset and ⊗F ′aj, ⊗F ′a′
⊗F ′a′
j ∈
AΓ/Λ = P, n, n′ ∈ N . By the above we may assume that ξ0 = p0 ⊗ n0, η0 = q0 ⊗ m0 ∈ P ¯⊗N , p0
and q0 orthogonal to S and n0, m0 ∈ N . Moreover, p0 and q0 can be considered to have support
contained in F .
This scalar product will be proven to be 0 by computing it more explicitly. First, notice that the
support of the elements from P which appear in the scalar product is contained in F ∪ hF ∪ F ′.
Denote by τ the trace on P . Then, since F ∩ hF = ∅, we have the decomposition
hξ0, γ(w′)ρh(η0)wi = τ (b1)τ (b2),
j γ(a′∗
j )p0 ∈ AF and b2 ∈ A(hF ∪F ′)\F ¯⊗N.
where b1 = ⊗F ∩F ′a∗
The first factor is 0 because p0 is orthogonal to S. This proves the claim.
(cid:3)
Now, we can finish the proof of the lemma. Take v ∈ P ¯⊗N such that αh(v) = v, for all h ∈ H.
Write v = v0 + v⊥ with v0 ∈ S ⊗ L2(N ) and v⊥ ⊥ S ⊗ L2(N ). Since S ⊗ L2(N ) is α-invariant, we
get that v0 and v⊥ are α-invariant. The claim gives us that v⊥ = 0, which implies that v ∈ S ¯⊗N .
This ends the lemma.
(cid:3)
2.5. Cocycle rigidity. The following proposition is the first part of [Po05, Proposition 3.2]. Before
writing the result, let us introduce some terminology.
Let Γ be a countable group and σ be a trace preserving action of Γ on a tracial von Neumann
algebra Q. We recall that a local cocycle for the action σ is a map w on Γ with values in the set
of partial isometries of Q which satisfies wgσg(wh) = wgh, for all g, h ∈ Γ.
Let σ′ be a trace preserving action of Γ on another tracial von Neumann algebra N and denote
by ρ the tensor product action σ ⊗ σ′. For a cocycle w : Γ → U (Q ¯⊗N ) we denote by wl : Γ →
U (Q ¯⊗Q ¯⊗N ) the image of w via the canonical isomorphism and inclusion Q ¯⊗N ≃ Q ¯⊗1 ¯⊗N ⊂
Q ¯⊗Q ¯⊗N. Similarly, we denote by wr the image of w via the canonical isomorphism and inclusion
Q ¯⊗N ≃ 1 ¯⊗Q ¯⊗N ⊂ Q ¯⊗Q ¯⊗N.
Proposition 2.11. [Po05, Proposition 3.2] Let σ be a weak mixing trace preserving action of Γ
on a tracial von Neumann algebra Q and σ′ a trace preserving action of Γ on another tracial von
Neumann algebra N . Let w : Γ → U (Q ¯⊗N ) be a cocycle for the action ρ. Let b ∈ L2(Q ¯⊗Q ¯⊗N ) be a
∗ =
non-zero element and p ∈ P(Q ¯⊗1 ¯⊗N ) be a non-zero projection such that pb = b and wl
b, for all g ∈ Γ, where ¯σ := σ ⊗ σ ⊗ σ′. Then, there exist a partial isometry v ∈ Q ¯⊗N and a local
cocycle w′
g, for all g ∈ Γ.
g(v∗v)) such that vv∗ ≤ p, v∗v ∈ N and wg(σ ⊗ σ′
g ∈ U (v∗vN σ′
g ¯σg(b)wr
g
g)(v) = vw′
Remark 2.12. Let us explain why the first part of [Po05, Proposition 3.2] can be written as above.
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
9
• In [Po05] the tracial von Neumann algebra (Q, τ ) is extended to a larger tracial von Neu-
mann algebra ( Q, τ ) satisfying the following properties: it exists a trace preseving action σ
of Γ on Q which extends σ and an automorphism α1 of Q which satisfies spw Qα1(Q) = Q
and τ (xα1(y)) = τ (x)τ (y), for all x, y ∈ Q.
In particular, it can be chosen Q = Q ¯⊗Q.
• Notice that b can be chosen in L2( Q ¯⊗N ) in [Po05, Proposition 3.2], not necessary in Q ¯⊗N ,
since the proof uses only this information.
From now on until the end of the section, we assume the following context. Let Λ be a subgroup
of a countable group Γ. Let σ0 be a trace preserving action of Λ on a tracial von Neumann algebra
A and σ the coinduced action of Γ on P := AΓ/Λ. Let us consider a trace preserving action σ′ of Γ
on another tracial von Neumann algebra N .
Denote by ρ the tensor product action σ ¯⊗σ′ of Γ on P ¯⊗N , by ρ the tensor product action σ ⊗ σ′
of Γ on P ¯⊗N and by ¯σ the tensor product action σ ⊗ σ ⊗ σ′ of Γ on P ¯⊗P ¯⊗N.
Let w : Γ → U (P ¯⊗N ) be a cocycle for ρ. Define the representations π : Γ → U (L2(P ¯⊗P ¯⊗N )) and
∗ and γg(c) = wg ρg(c)θ1(wg)∗. Here we have
γ : Γ → U (sp P θ1(P ) ⊗ L2(N )), by πg(b) = wl
denoted by sp P θ1(P ) the k · k2-closed linear subspace generated by {xθ1(y)x, y ∈ P }.
Notice that L2(P ¯⊗P ¯⊗N ) and sp P θ1(P )⊗L2(N ) may be viewed as left P ¯⊗N Hilbert modules with
the actions (p⊗n)·(x⊗y ⊗n′) := px⊗y ⊗nn′ and, respectively, (p⊗n)·xθ1(y)⊗n′ := pxθ1(y)⊗nn′,
for all p, x, y ∈ P and n, n′ ∈ N. The following lemma makes Proposition 2.11 useful in our context
in which we work with the free product deformation. The proof is a straightforward verification.
g ¯σg(b)wr
g
Lemma 2.13. The map U : L2(P ¯⊗P ¯⊗N ) → sp P θ1(P ) ⊗ L2(N ) defined by U (p1 ⊗ p2 ⊗ n) =
p1θ1(p2) ⊗ n, with p1, p2 ∈ P, n ∈ N, is an isomorphism of Hilbert spaces which intertwines the
representations π and γ. Moreover, U intertwines the left P ¯⊗N - module structures of these Hilbert
spaces.
In order to apply Proposition 2.11, we need the weak mixing property for the coinduced action.
Lemma 2.14. Let H be a subgroup of Γ with the property that there is no finite index subgroup H0
of H which is contained in a conjugate gΛg−1 of Λ. Then the coinduced action σ is weak mixing
on H.
Ioana proved this result for coinduced actions on standard probability spaces in [Io06b, Lemma
2.2], but the proof also works for tracial von Neumann algebras.
Using the same arguments as in the second part of the proof of [Po05, Proposition 3.2], we obtain
the following result:
Theorem 2.15. Let Γ be a countable group and Λ be a subgroup. Let H be a subgroup of Γ
with the property that there is no finite index subgroup H0 of H such that H0 is contained in a
conjugate gΛg−1 of Λ. Let w : Γ → U (P ¯⊗N ) be a cocycle for the action ρ. If wH and θ1(w)H are
cohomologous, then wH is cohomologous with a cocycle with values in N .
Proof. We will use Proposition 2.11 and a maximality argument.
Denote by W the set of pairs (v, w′) with v ∈ P ¯⊗N partial isometry satisfying v∗v ∈ N and
w′ : Γ → U (v∗vN σ′(v∗v)) local cocycle for ρ such that vw′
We endow W with the order: (v0, w′
W is an inductive set and let (v0, w′
0) ≤ (v1, w′
0v0, v∗
0) ∈ W be a maximal element.
g = wgρg(v), for all g ∈ Γ.
1) iff v0 = v1v∗
0v0w′
1(g) = w′
0(g), for all g ∈ Γ.
10
DANIEL DRIMBE
Claim. v0 is a unitary.
Note that the claim finishes the proof. Let us prove the claim by contradiction. Suppose v0 is
not a unitary. Denote by v = v0θ1(v∗
0 and a direct computation gives us that
0(g−1)∗) = w′
wg ρg(v) = vθ1(wg). Indeed, since ρg(w∗
0(g)ρg(θ1(v∗
0). Then vv∗ = v0v∗
g−1) = wg and ρg(w′
0)) = v0w′
wg ρg(v) = wg ρg(v0)ρg(θ1(v∗
0(g), we have
0))
= v0θ1(ρg(v0w′
= v0θ1(v∗
0ρg(w∗
= vθ1(wg).
0(g−1))∗) = v0θ1(ρg(wg−1ρg−1(v0))∗)
g−1)) = v0θ1(v∗
0wg)
Since w and θ1(w) are cohomologous, by Lemma 2.1, we obtain the existence of a partial isometry
v′ ∈ P ¯⊗N such that wg ρg(v′) = v′θ1(wg) and v′v′∗ = 1 − vv∗, v′∗v′ = 1 − v∗v.
Next, Lemma 2.10 implies that v′ ∈ sp P θ1(P ) ¯⊗N , which allows us to use Lemma 2.13. Since v′ is
a fixed point for γ, U −1(v′) is a fixed point for π. Now we can apply Proposition 2.11 to obtain the
existence of a partial isometry v1 ∈ P ¯⊗N with the left support majorized by l(U −1(v′)) and right
support in N which satisfies v1w′
1)).
Here we denote by l(U −1(v′)) the left support of U −1(v′).
Notice that l(U −1(v′)) is majorized by v′v′∗ = 1 − v0v∗
P ¯⊗N left module structure. Now, since v′v′∗ = 1 − vv∗ = 1 − v0v∗
U −1(v′v′∗v′) = v′v′∗U −1(v′), which proves the claim.
Thus, in the finite von Neumann algebra P ¯⊗N we have v∗
the first and the last projection lies in N , we obtain that v∗
the central trace).
0. Indeed, by Lemma 2.13, U intertwines the
0 ∈ P ¯⊗N , we have U −1(v′) =
0v0. Since
0v0 in N (by working with
1(g) = wg ρg(v1) for some local cocycle w′
1 : Γ → U (v1v∗
1v1 (cid:22) 1 − v∗
1 ≤ 1 − v0v∗
1v1 ∼ v1v∗
0 ∼ 1 − v∗
1N σ′(v1v∗
Now, we conclude as in the proof of [Po05, Proposition 3.2]. By multiplying v1 to the right with
a partial isometry in N and conjugate w′
0v0. But
then, (v0 + v1, w′
0), which contradicts the maximality
assumption.
(cid:3)
1) ∈ W and strictly majorizes (v0, w′
1 appropriately, we may assume v∗
1v1 ≤ 1 − v∗
0 + w′
3. Proof of Theorem A
We will prove the following theorem, which is the general version of Theorem A dealing with
coinduced actions of Γ on AΓ/Λ that arise from actions of Λ on arbritrary tracial von Neumann
algebras A.
Theorem 3.1 (Groups with relative property (T)). Let Γ be a countable group and Λ be a subgroup.
Let H ⊂ Γ be a subgroup with relative property (T). Assume that there does not exist a subgroup
H0 of finite index in H such that H0 is contained in a conjugate g−1Λg of Λ.
Let σ0 be a trace preserving action of Λ on a tracial von Neumann algebra A and σ the coinduced
action on P := AΓ/Λ. Let us consider another action σ′ on a tracial von Neumann algebra N .
Denote by ρ the tensor product action σ ¯⊗σ′ of Γ on P ¯⊗N.
Then, any cocycle w : Γ → U (P ¯⊗N ) for the restriction of ρ to H is cohomologous with a cocycle
of the form w′ : H → U (N ).
Moreover, if H is w-normal in Γ, then w is cohomologous with a cocycle of the form w′ : Γ → U (N ).
From now on, in this section we use the same notations as in Section 2.3. The first step of the
proof of Theorem 3.1 is to prove that wH and θ1(w)H are cohomologous. This is obtained by the
following result which is [Po05, Lemma 4.6] adapted to the free product deformation.
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
11
Proposition 3.2. [Po05, Lemma 4.6] Let Λ be a subgroup of Γ. Let H ⊂ Γ be a subgroup with
relative property (T) such that there does not exist a subgroup H0 of finite index in H which is
contained in a conjugate g−1Λg of Λ.
Let σ0 be a trace preserving action of Λ on a tracial von Neumann algebra A and σ the coinduced
action on P = AΓ/Λ. Consider a trace preserving action σ′ on a tracial von Neumann algebra N .
Let w : Γ → P ¯⊗N be a cocycle for the action ρ on P ¯⊗N . Then wH and θ1(w)H are cohomologous
as cocycles for the action ρH on P ¯⊗N .
The proof of Proposition 3.2 is almost identical to that of [Po05, Lemma 4.6], but we include it for
completeness. At the end of the proof of [Po05, Lemma 4.6] it is used the weak mixing property
and therefore is obtained that a certain element is in a smaller algebra. The difference is that in
the proof of Proposition 3.2 is used Lemma 2.10 to obtain the same result.
Proof of Proposition 3.2. It is enough to prove that ∀ǫ > 0, ∃v ∈ P ¯⊗N partial isometry such that
kv∗v − 1k2 ≤ ǫ and
wh ρh(v) = vθ1(wh), ∀h ∈ H.
Indeed, if this holds, take a unitary u ∈ U ( P ¯⊗N ) satisfying uv∗v = v. By triangle inequality, we
get that
kwh ρh(u) − uθ1(wh)k2 ≤ 2ku − vk2 = 2k1 − v∗vk2 ≤ 2ǫ, ∀h ∈ H.
Using now Lemma 2.1, we get that wH and θ1(w)H are cohomologous.
We now prove the first statement of this proof in two steps.
Step 1. For all ǫ > 0, there exist v0 ∈ P ¯⊗N and n ∈ N such that kv∗
(3.1)
wh ρh(v0) = v0θ1/2n(wh), ∀h ∈ H.
0v0 − 1k2 ≤ ǫ and
This is just an application of Lemma 2.2. Indeed, the lemma gives us the existence of a partial
isometry v0 ∈ P ¯⊗N and n ∈ N, satisfying kv0 − 1k2 ≤ ǫ/2 such that formula 3.1 holds. Using the
triangle inequality, we get that kv∗
Step 2. Assume that there exists a partial isometry v ∈ P ¯⊗N and t ∈ (0, 1) satisfying
(3.2)
Then there exists a partial isometry v′ ∈ P ¯⊗N satisfying kvk2 = kv′k2 and
wh ρh(v) = vθt(wh), ∀h ∈ H.
0v0 − 1k2 ≤ ǫ.
wh ρh(v′) = v′θ2t(wh), ∀h ∈ H.
For proving Step 2, we will use the properties of the automorphism β. Since βθt = θ−tβ and
βP ¯⊗N = idP ¯⊗N we get that
wh ρh(β(v)) = β(v)θ−t(wh), ∀h ∈ H.
By taking the adjoint in 3.2, we obtain
v∗wh = θt(wh)ρh(v∗), ∀h ∈ H.
Define now v′ = θt(β(v)∗v). We get
which implies that
v′∗wh = θt(v∗β(v)θ−t(wh))
= θt(v∗wh ρh(β(v)))
= θt(θt(wh)ρh(v∗β(v)))
= θ2t(wh)ρh(v′∗),
wh ρh(v′) = v′θ2t(wh), ∀h ∈ H.
12
DANIEL DRIMBE
Let us prove now that kvk2 = kv′k2. Since kv′k2 = kβ(v)∗vk2, it’s enough to prove that β(vv∗) =
vv∗. Using the equation 3.2 and applying the adjoint to it, we get that
wh ρh(vv∗)w∗
h = vv∗, ∀h ∈ H.
By Lemma 2.10, we obtain that vv∗ ∈ P ¯⊗N , so β(vv∗) = vv∗. This ends the proof.
(cid:3)
The proof of Theorem 3.1 is now an easy consequence of Proposition 3.2 and Theorem 2.15.
Proof or Theorem 3.1
By Proposition 3.2, there exists a unitary v ∈ P ¯⊗N such that
whρh(v) = vθ1(wh), ∀h ∈ H.
Theorem 2.15 gives us the existence of a cocycle w′ : H → U (N ) cohomogous with w. More
precisely, we have
wh = uw′
hρh(u∗),
∀h ∈ H,
for a unitary u ∈ U (P ¯⊗N ).
For the moreover part, notice that Lemma 2.14 implies that the coinduced action is weak mixing
on H. Thus, we can apply Proposition 2.4 and obtain that u∗wgρg(u) ∈ N , for all g ∈ Γ. This
allows us to define w′ on Γ and obtain that w is cohomologous with a cocycle with values in N on
Γ.
(cid:3)
Remark 3.3. Theorem 3.1 implies Theorem A. Indeed, this is true by Remark 2.5 and [Po05,
Proposition 3.5], which allows us to untwist a cocycle into a Uf in group once is unwisted into U (N ).
4. Proof of Theorem B
In this section we prove Theorem 4.1, which is a more general version of Theorem B dealing with
coinduced actions of Γ on AΓ/Λ that arise from actions of Λ on arbitrary tracial von Neumann
algebras A.
Theorem 4.1 (Product groups). Let Γ be a countable group and Λ be an amenable subgroup. Let
H and H ′ be infinite commuting subgroups of Γ such that H ′ is non-amenable. Assume that H
does not have a subgroup H0 of finite index in H such that H0 is contained in a conjugate g−1Λg
of Λ.
Let σ0 be a trace preserving action of Λ on a tracial von Neumann algebra A and σ the coinduced
action on P := AΓ/Λ. Let us consider another action σ′ on a tracial von Neumann algebra N .
Denote by ρ the tensor product action σ ¯⊗σ′ of Γ on P ¯⊗N.
Then, any cocycle w : Γ → U (P ¯⊗N ) for the restriction of ρ to HH ′ is cohomologous with a cocycle
of the form w′ : HH ′ → U (N ).
Moreover, if H is w-normal in Γ, then w is cohomologous with a cocycle of the form w′ : Γ → U (N ).
We use the same notations as in section 2.3. We still consider σ the coinduced action on P , σ′ a
trace preserving action on a tracial von Neumann algebra N and the free product deformation θt.
The following result is known as Popa’s transvesality lemma.
Lemma 4.2. ( [Po06, Lemma 2.1]) For every s ∈ (0, 1/2) and x ∈ P ¯⊗N , we have
kθ2s(x) − xk2 ≤ 2kθs(x) − EP ¯⊗N (θs(x))k2.
Lemma 4.3. Let Γ be a countable group and Λ an amenable subgroup. Let F be a finite subset
of Γ/Λ. Denote NF = {g ∈ ΓgF = F }, where Γ acts on Γ/Λ by left multiplication. Then NF is
amenable.
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
13
Proof. The action of NF on F , by left multiplication, is well defined. Denote by SF the group of
bijections on the finite set F. We obtain a homomorphism φ : NF → SF , defined by φ(g)f = gf,
for all g ∈ NF and f ∈ F .
Notice that kerφ, the kernel of φ, is amenable. Indeed, if f Λ ∈ F , kerφ ⊂ f Λf −1. Since Λ is
amenable, ker φ is amenable. Note that φ(NF ), the image of φ, is amenable, being a finite group.
Since kerφ and φ(NF ) are amenable groups, we conclude that NF is amenable.
Theorem 4.4. Let Γ be a countable group and Λ an amenable subgroup. Let H and H ′ be infinite
commuting subgroups of Γ such that H ′ is non-amenable. Denote M = ( P ¯⊗N ) ⋊ H and M =
(P ¯⊗N ) ⋊ H.
Let w : H ′ → U (P ¯⊗N ) be a cocycle for ρ and define the representation π : H ′ → U (L2( M )⊖L2(M ))
by πg(x) = wg ρg(x)w∗
g. Then π has spectral gap.
(cid:3)
Remark 4.5. In Theorem 4.4 the action ρH ′ is considered to be extended in a natural way to
( P ¯⊗N ) ⋊ H. This is possible since H and H ′ commute.
Proof of Theorem 4.4. Let B = {1 = η0, η1, ..} ⊂ A be an orthonormal basis of L2(A). Denote by u
the canonical Haar unitary of L(Z). Thus, we obtain an orthonormal basis for L2(A ∗ L(Z)) given
by
B = {un1ηj1uη2...ηjk j1, ..jk−1 ≥ 1, k ∈ N} = {1 = η0, η1, ..},
as in [Io06a, Proposition 2.3]. Also, we have that
and
N = {⊗f ∈Γ/Ληif {f if 6= 0} is finite}
N = {⊗f ∈Γ/Λ ηif {f if 6= 0} is finite}
are orthonormal bases for L2(P ) and, respectively, for L2( P ).
Let x = ⊗f ∈Γ/Λ ηif ∈ N . Denote Fx = {f ∈ Γ/Ληif ∈ B \ B} and K 0
that K 0
F ′, whenever F 6= F ′ are finite subsets of Γ/Λ. This implies that
F ⊥ K 0
F = sp{x ∈ N Fx = F }. Notice
where the direct sum runs over all finite non empty subsets F ⊂ Γ/Λ.
L2( P ) ⊖ L2(P ) = sp N \ N = ⊕K 0
F ,
Thus,
L2( P ¯⊗N ) ⊖ L2(P ¯⊗N ) = ⊕K 1
F ,
where the direct sum runs over all finite non empty subsets F ⊂ Γ/Λ and K 1
F = K 0
F ⊗ L2(N ).
Finally, we get the decomposition
L2( M ) ⊖ L2(M ) = ⊕KF ,
where the direct sum runs over all finite non empty subsets F ⊂ Γ/Λ and KF = sp{K 1
F uhh ∈ H}.
Claim 1. We can decompose L2( M ) ⊖ L2(M ) = ⊕i∈I sp π(H ′)M ξiM, where {ξi}i∈I is a family of
vectors from L2( P ) and each ξi ∈ KF for some non empty finite set F ⊂ Γ/Λ.
Proof of the claim 1. Let S be the set of elementary tensors ⊗i∈F ηi, with F finite subset of Γ/Λ
such that each ηi is an element of A which starts and ends with a non-trivial power of u. Then Γ
acts on S and choose T to be a set of representatives for this action.
Then, L2( M ) ⊖ L2(M ) = ⊕ξ∈T sp π(H ′)M ξM.
Denote by λH ′ the left regular representation of H ′ on l2(H ′).
Claim 2. π (cid:22) λH ′, i.e. π is weakly contained in λH ′.
(cid:3)
14
DANIEL DRIMBE
We suppose the claim holds and we prove it after the end of this theorem. For finishing the proof,
note that the non-amenability of H ′ implies 1H ′ (cid:14) λH ′. Thus, 1H ′ (cid:14) π, which means that π has
spectral gap on H. This proves the theorem.
(cid:3)
We now prove Claim 2 from the proof of Theorem 4.4 using the same notations.
Lemma 4.6. π (cid:22) λH ′.
Proof. For every F ⊂ Γ/Λ, non empty finite subset, denote H ′
on Γ/Λ by left multiplication. Since F is finite and Λ is amenable, Lemma 4.3 implies that H ′
an amenable group.
F = {h′ ∈ H ′h′F = F }, where Γ acts
F is
Let us take a family of vectors {ξi}i∈I as in the first claim of Theorem 4.4. Fix i ∈ I and let F be
a finite subset of Γ/Λ such that ξi ∈ KF .
Note that
(4.1)
hπg(x), xi = 0, ∀x ∈ KF , g /∈ H ′
F .
F )ηij are mutually orthogonal, we get the decomposition sp π(H ′
Let us observe that we can decompose sp π(H ′)M ξiM = ⊕j∈J sp π(H ′)ηij in cyclic subspaces,
with ηij ∈ KF . Indeed, by taking a maximal family of vectors {ηij}j∈J with the property that
sp π(H ′
Since ξi ∈ KF and KF is a π(H ′
F is a
subgroup of H ′, the decomposition sp π(H ′)M ξiM = ⊕j∈J sp π(H ′)ηij also holds. Indeed, (4.1)
implies that sp π(H ′)ηij is orthogonal on sp π(H ′)ηij ′, for all j, j′ ∈ J, with j 6= j′. This proves the
claim.
Fix j ∈ J. Define the cyclic representations θ : H ′ → U (sp π(H ′)ηij ) and θF : H ′
as the restrictions of π, respectively of πH ′
Let
F ) invariant subspace, we obtain that ηij ∈ KF . Since H ′
to the coresponding cyclic subspaces.
F )M ξiM = ⊕j∈J sp π(H ′
F → U (sp π(H ′
F )ηij)
F
F )ηij.
θ := IndH ′
H ′
F
θF : H ′ → U (l2(H ′/H ′
F ) ⊗ sp π(H ′
F )ηij)
F and η ∈ sp π(H ′
θg(δx ⊗ η) = δgx ⊗ [θF (c(g, x))η],
be the induced representation of θF defined by
(4.2)
for all g ∈ H ′, x ∈ H ′/H ′
cocycle defined as in section 1.2. Recall that c(g, x) = φ−1(gx)gφ(x), for all g ∈ H ′ and x ∈ H ′/H ′
with φ : H ′/H ′
with e the neutral element of H ′. This implies c(g, H ′
Claim. The induced representation θ contains θ as a subrepresentation.
Proof of the Claim. Define the positive definite function ϕ : H ′ → C by ϕ(g) =< θ(g)ηij, ηij > for
g ∈ H ′. The formula (4.1) implies that ϕ is zero on H ′ \ H ′
Denote by η := δeH ′
F → H ′ an arbitrary fixed section. Moreover, φ can be chosen such that φ(H ′
F is the canonical
F ,
F ) = e,
F )ηij. A direct computation gives us that
F )ηij, where c : H ′ × H ′/H ′
F ) = g, for all g ∈ H ′
F .
F , since ηij ∈ KF .
⊗ ηij ∈ l2(H ′/H ′
F ) ⊗ sp π(H ′
F → H ′
F
< θ(g)η, η >=< θ(c(g, eH ′
F )ηij, ηij) >=< θ(g)ηij, ηij) >= ϕ(g)
for all g ∈ H ′
Finally, we have obtained that
F . For g /∈ H ′
F , the formula (4.2) gives us that< θ(g)η, η >= 0.
< θ(g)η, η >= ϕ(g)
for all g ∈ H ′. Since θ is a cyclic representation, we get that θ is contained in θ. This ends the
claim.
(cid:3)
Now, we can finish the proof of the lemma. Since H ′
Theorem G.3.2], for example). By Fell absorbing principle, we get that θF (cid:22) λH ′
F is amenable, we have 1H ′
(see [BHV08,
. [BHV08,
(cid:22) λH ′
F
F
F
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
15
Theorem F.3.5] gives us continuity of weak containment with respect to induction. This implies
that θ = IndH ′
H ′
F
H. Since θ is contained in θ, we get that θ (cid:22) λH ′.
θF (cid:22) IndH ′
H ′
F
= λ′
λH ′
F
Denote by θi : H ′ → U (sp π(H ′)M ξiM ) the restriction of π to the subspace sp π(H ′)M ξiM. The
decomposition sp π(H ′)M ξiM = ⊕sp π(H ′)ηij gives us that θi (cid:22) λH ′.
The decomposition given by the first claim implies that π = ⊕i∈I θi. Thus, π (cid:22) λH ′, which ends
the proof of the lemma.
(cid:3)
Proof of Theorem 4.1
Define the representation π : Γ → U (L2(( P ¯⊗N ) ⋊ H) ⊖ L2((P ¯⊗N ) ⋊ H)) by πg(x) = wg ρg(x)w∗
g
and denote M = ( P ¯⊗N ) ⋊ H and M = (P ¯⊗N ) ⋊ H as in the previous theorem.
Theorem 4.4 implies that π has spectral gap on H ′. Thus, for all ǫ > 0, exists δ > 0 and F ′ ⊂ H ′
finite, such that if u ∈ U ( M ) satisfies kπh(u) − uk2 ≤ δ, ∀h′ ∈ F ′, then ku − EM (u)k2 ≤ ǫ.
Let us proceed now as in [ [Po06],Theorem 4.1] . Denote by ¯ug := wgug, g ∈ Γ. Since the map
s → θs(¯us) is continuous in k · k2 for all h′ ∈ F ′, it follows that for small enough s, we get that
kθ−s/2(¯uh) − ¯uhk2 ≤ δ/2,
for all h′ ∈ F ′. Because H and H ′ commute, ¯uh′ and ¯ug commute for all h′ ∈ F ′ and g ∈ H. Thus,
we get that
k[θs/2(¯ug), ¯uh′]k2 = k¯ug, θ−s/2(¯uh′)k2 ≤ 2kθ−s/2(¯uh′) − ¯uh′k ≤ δ,
for all h′ ∈ F ′ and g ∈ H.
Notice that π′
h(x) = ¯u′
hx¯u′∗
h , for all g ∈ H ′. A direct computation gives us that
for all h′ ∈ F ′ and g ∈ H, which implies that
kπh′(θs/2(¯ug)) − θs/2(¯ug)k2 = k[θs/2(¯ug), ¯uh′]k2 ≤ δ,
kθs/2(¯ug) − EM (θs/2(¯ug))k2 ≤ ǫ.
Using Lemma 4.2, we get that kθs(¯ug)− ¯ugk2 ≤ 2ǫ for all g ∈ H. The set K := cow{¯ugθs(¯ug)∗g ∈ H}
is convex weakly compact and for all ξ ∈ K and g ∈ H, we have ¯ugξθs(¯ug)∗ ∈ K. Thus, if we denote
by ξ0 ∈ K the unique element of minimal norm kk2, then we get that ¯ugξ0θs(¯ug)∗ = ξ0 for all
g ∈ H. This is equivalent to
for all g ∈ H. Taking v ∈ P ¯⊗N , to be the partial isometry of ξ0, we get that
wg ρg(ξ0) = ξ0θs(wg),
wg ρg(v) = vθs(wg),
for all g ∈ H.
Since
k¯ugθs(¯ug)∗ − 1k2 = k¯ug − θs(¯ug)k2 ≤ 2ǫ,
we get that kξ0 −1k2 ≤ 2ǫ, which implies that kv −1k2 ≤ 4(2ǫ)1/2. This proves Step 1 of the proof of
Proposition 3.2. In combination with Step 2 from the proof of Proposition 3.2, the conclusion follows
as in the proof of Proposition 3.2. Meaning, we obtain that wH and θ(w)H are cohomologous.
As in the proof of Theorem 3.1, we use Theorem 2.15 to deduce the existence of a unitary u ∈
U (P ¯⊗N ) and of a cocycle w′ : H → U (N ) such that
hρh(u∗),
wh = uw′
∀h ∈ H.
16
DANIEL DRIMBE
We have H ⊳ HH ′, because H and H ′ commute. Since the restriction of ρ to H is weakly mixing,
by using Proposition 2.4 we obtain a cocycle w′ with values in N which is cohomologous to w on
HH ′.
For the moreover part, we apply again Proposition 2.4 as in the proof of Theorem 3.1.
(cid:3)
5. Applications to W∗-superrigidity
We record the results [CP10, Corollary 5.3], [PV12, Theorem 1.1], [Io12b, Theorem 1.1] in the
following theorem, which give uniqueness of group measure space Cartan subalgebras for groups in
C.
Theorem 5.1. Let Γ ∈ C and let Γ y X be a free ergodic pmp action on a standard probability
space X. Suppose there exists Λ y Y a free ergodic pmp action on a standard probability space
Y such that M = L∞(X) ⋊ Γ = L∞(Y ) ⋊ Λ. Then there exists a unitary u ∈ M such that
uL∞(X)u∗ = L∞(Y ).
The following result is a particular case of [Po05, Theorem 5.6] (see also [Fu06, Theorem 1.8]).
Proposition 5.2. [Po05, Theorem 5.6] Let Γ be a countable group with no non-trivial finite normal
subgroups. Let Γ y (X, µ) be a free pmp action, where (X, µ) is a standard probability space. If
Γ y (X, µ) is Uf in-cocycle superrigid, then Γ y (X, µ) is OE-superrigid.
The following lemma gives a sufficient condition for coinduced actions to be free.
Lemma 5.3. [Io06b, Lemma 2.1] Let Γ be a countable group and Λ a subgroup of infinite index.
σ0y (X0, µ0) be a pmp action on the standard probability space (X0, µ0) which has no atoms
Let Λ
σ
y (X, µ) := (X0, µ0)Γ/Λ be the coinduced action. Suppose ∩g∈ΓgΛg−1 is finite and
and let Γ
∩g∈ΓgΛg−1 ∩ F ix(Λ y X0) = {e}, where F ix(Λ y X0) consists of those elements g ∈ Λ for which
{x0 ∈ X0gx0 = x0} has measure 1. Then Γ y X is free.
Proof. Define Ag = {(xh)h∈Γ/Λ ∈ Xσg((xh)h) = (xh)h} for g ∈ Γ. Recall that σg((xh)h) = (x′
where x′
If g0 /∈ ∩g∈ΓgΛg−1, there exists g1 ∈ Γ such that g−1
h = φ−1(gh)gφ(h)xg−1 h and φ : Γ/Λ → Γ is a section.
0 g1Λ 6= g1Λ. Then
h)h,
Ag0 = {(xh)h ∈ Xxh = φ(g0h)−1g0φ(h)xg−1
0 h, ∀h ∈ Γ/Λ}
⊂ {(xh)h ∈ Xxg1Λ = φ(g0g1Λ)−1g0φ(g1Λ)xg−1
0 g1Λ}
has measure 0 since X0 is non-atomic.
Now, if g0 ∈ Σ := ∩g∈ΓgΛg−1 \ {e}, we have g−1g0g ∈ Σ \ {e}, for all g ∈ Γ. The hypothesis implies
that Cλ := {x0 ∈ X0λx0 = x0} has measure less than 1, for all λ ∈ Σ \ {e}. Then,
Ag0 = {(xh)h∈Γ/Λxh = φ(h)−1g0φ(h)xh, ∀h ∈ Γ/Λ}
= Qh∈Γ/Λ Cφ(h)−1g0φ(h)
has measure 0. Indeed, since Σ is finite, there exists g1 ∈ Σ\{e} such that {h ∈ Γ/Λφ(h)−1g0φ(h) =
g1} is an infinite set. This implies Ag0 has measure 0 since µ0(Cg1) < 1.
(cid:3)
Notice that the proof of Lemma 5.3 also proves that if we coinduce from free actions, we obtain
free actions. The following result proves cocycle superrigidity for coinduced actions of groups from
C.
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
17
Theorem 5.4. Let Γ ∈ C and Λ a subgroup defined as in Corollary 1.4. Let Λ y X0 be a measure
preserving action on a standard probability space X0 and let Γ y X be the coinduced action from
Λ y X0. Then Γ y X is Uf in-cocycle superrigid.
Proof. We apply Theorems A and B and let us use the notations from these theorems.
For Γ ∈ C1, we want to apply Theorem B. If we take H ′ = Γ1 and H = Γ2, the conditions of
Theorem B are satisfied, so we obtain the claim.
If Γ ∈ C2, consider Γ = Γ1 × Γ2 × ... × Γn, with all the Γi’s non-elementary hyperbolic groups.
Without loss of generality suppose that Λ ⊂ Γ1. We apply again Theorem B. By taking H ′ = Γ1
and H = Γ2 × ... × Γn we notice that the conditions of this theorem are again satisfied.
Let Γ ∈ C3. Since Σ0 is contained in Γ0
subgroups which are contained in a conjugate of Λ. We want to apply Theorem A for H = Γ0
1.
Take V ∈ Uf in and w : Γ × X → V a cocycle for Γ y X. Theorem A implies that there exist
φ : X → V such that φ(gx)−1w(g, x)φ(x) is independent of x on Γ0
1. Since Σ0 is normal in Γ0
2
1, Lemma 2.14 combined with Proposition 2.4 implies that φ(gx)−1w(g, x)φ(x)
and contained in Γ0
is independent of x on Γ0
2. This proves that w is cohomologous with a group homomorphism on
1 ∗Σ0 Γ0
Γ0
2.
Now, since Γ0
φ(gx)−1w(g, x)φ(x) is independent of x on Γ. This ends the proof.
2 is normal in Γ, we apply again Lemma 2.14 and Proposition 2.4 to obtain that
(cid:3)
1, the hypothesis implies that Γ0
1 does not have finite index
1 ∗Σ0 Γ0
Proof of the Corollary 1.4. Combining Proposition 5.2 with Theorem 5.4, we obtain that Γ y X is
OE-superrigid. Lemma 2.14 proves that Γ y X is ergodic and we conclude using Theorem 5.1. (cid:3)
References
L. Bowen: Orbit equivalence, coinduced actions and free products, Groups Geom. Dyn. 5 (2011), no. 1, 115.
[B10]
R. Boutonnet: W∗-superrigidity of mixing Gaussian actions of rigid groups, Adv. Math. 244 (2013) 6990.
[Bo12]
[BHV08] B.Bekka, P. de la Harpe, A.Valette: Kazhdan’s Property (T), (New Mathematical Monographs, 11), Cam-
bridge University Press, Cambridge, 2008.
[CFW81] A. Connes, J. Feldman, B. Weiss: An amenable equivalence relations is generated by a single transformation,
[CIK13]
[CK15]
[CP10]
[CS11]
Ergodic Th. Dynam. Sys. 1(1981), 431-450.
I. Chifan, A. Ioana, Y. Kida: W∗-superrigidity for arbitrary actions of central quotients of braid groups,
Math. Ann. 361 (2015), 925959.
I. Chifan, Y. Kida: OE and W* superrigidity results for actions by surface braid groups, Proceedings of the
London Mathematical Society, in press.
I. Chifan, J. Peterson: Some unique group-measure space decomposition results, Duke Math. J. 162 (2013),
no. 11, 1923-1966.
I. Chifan, T. Sinclair: On the structural theory of II1 factors of negatively curved groups, Ann. Sci. c. Norm.
Supr. (4) 46 (2013), 133.
[CSU11] I. Chifan, T. Sinclair, B. Udrea: On the structural theory of II1 factors of negatively curved groups, II.
Actions by product groups, Adv. Math. 245 (2013), 208236.
[Dy58] H. Dye: On groups of measure preserving transformation. I, Amer. J. Math. 81 (1959), 119-159.
[Fu98]
[Fu99]
A. Furman: Orbit equivalence rigidity, Ann. of Math. 150 (1999), 1083-1108.
A. Furman: Gromov’s measure equivalence and rigidity of higher rank lattices, Ann. of Math. 150 (1999),
1059-1081.
A. Furman: On Popa’s Cocycle Superrigidity Theorem, Inter. Math. Res. Notices IMRN, 2007 (2007), 1 -
46.
A. Furman: A survey of Measured Group Theory, Geometry, Rigidity, and Group Actions, 296-374, The
University of Chicago Press, Chicago and London, 2011.
[Fu06]
[Fu09]
[FV10] P. Fima, S. Vaes: HNN extensions and unique group measure space decomposition of II1 factors Trans.
Amer. Math. Soc. 354 (2012) 26012617.
[Ga10] D. Gaboriau: Orbit equivalence and measured group theory, In Proceedings of the ICM (Hyderabad, India,
2010), Vol. III. Hindustan Book Agency, 2010, pp. 1501-1527.
18
DANIEL DRIMBE
[HPV10] C. Houdayer, S. Popa, S. Vaes: A class of groups for which every action is W∗-superrigid, Groups Geom.
Dyn. 7 (2013), 577590.
[Io06a] A. Ioana: Rigidity results for wreath product of II1 factors, J. Funct. Anal. 252 (2007), 763-791.
[Io06b] A. Ioana: Orbit inequivalent actions for groups containing a copy of F2, Invent. Math. 185 (2011), 55-73.
A. Ioana: Cocycle superrigidity for profinite actions of property (T) groups, Duke Math J. 157, (2011),
[Io08]
337–367.
A. Ioana: W∗-superrigidity for Bernoulli actions of property (T) groups J. Amer. Math. Soc. 24 (2011),
1175–1226.
[Io10]
[Io12a] A. Ioana: Classification and rigidity for von Neumann algebras, European Congress of Mathematics, EMS
(2013), 601-625.
[Io12b] A. Ioana: Cartan subalgebras of amalgamated free product of II1 factors, Ann. Sci. ´Ec. Norm. Sup´er 48,
[Io14]
fascicule 1 (2015), 71-130.
A. Ioana: Strong ergodicity, property (T), and orbit equivalence rigidity for translation actions, J. Reine.
Angew. Math.
[IPV10] A. Ioana, S. Popa, S. Vaes: A Class of superrigid group von Neumann algebras, Ann. of Math. (2) 178
(2013), 231286
[Ka67] D. Kazhdan: On the connection of the dual space of a group with the structure of its closed subgroups, Funct.
Anal. and its Appl. 1(1967), 6365.
Y. Kida: Measure equivalence rigidity of the mapping class group, Ann. of Math. (2) 171 (2010), 1851-1901.
Y. Kida: Rigidity of amalgamated free products in measure equivalence, J. Topol. 4 (2011), 687-735.
[Ki06]
[Ki09]
[Ma82] G. Margulis: Finitely-additive invariant measures on Euclidian spaces, Ergodic Theory Dynam. Systems
2(1982), 383396.
[MS02] N.Monod, Y. Shalom: Y., Orbit equivalence rigidity and bounded cohomology, Ann. of Math. 164 (2006),
825-878.
[MvN36] F. J. Murray, J. von Neumann, On rings of operators. Ann. of Math. 37 (1936). 116229.
[OP08] N. Ozawa, S. Popa: On a class of II1 factors with at most one Cartan subalgebra, II, Amer. J. Math. 132
(2010), 841-866.
[OW80] D. Ornstein, B. Weiss: Ergodic theory of amenable group actions. I. The Rohlin lemma, Bull. Amer. Math.
[Pe09]
[Po03]
[Po04]
[Po05]
[Po06]
[Po07]
[PS09]
[PV06]
[PV08]
[PV09]
[PV11]
[PV12]
Soc. (N.S.), 2(1):161-164, 1980.
J. Peterson: Examples of group actions which are virtually W∗-superrigid, Preprint. arXiv:1002.1745.
S. Popa: Strong rigidity of II1 factors arising from malleable actions of w-rigid groups, I. Invent. Math. 165
(2006), 369-408.
S. Popa: Some computations of 1-cohomology groups and construction of non-orbit-equivalent actions, J.
Inst. Math. Jussieu 5(2) (2006), 309-322.
S. Popa: Cocycle and orbit equivalence superrigidity for malleable actions of w-rigid groups, Invent. Math.
170 (2007), 243-295.
S. Popa: On the superrigidity of malleable actions with spectral gap, J. Amer. Math. Soc. 21 (2008), 981-
1000.
S. Popa: it Deformation and rigidity for group actions and von Neumann algebras, In Proceedings of the
ICM (Madrid, 2006), Vol. I, European Mathematical Society Publishining House, 2007, 445-477.
J. Peterson, T. Sinclair: On cocycle superrigidity for Gaussian actions Erg. Th. and Dyn. Sys.32 (2012),
no. 1, 249-272.
S. Popa, S. Vaes: Strong rigidity of generalized Bernoulli actions and computations of their symmetry groups,
Advanced in Mathematics 217 (2008), 833-872.
S.Popa, S. Vaes: Cocycle and orbit superrigidity for lattices in SL(n, R) acting on homogeneous spaces, In
Geometry, rigidity and group actions, eds. B. Farb and D. Fisher, The University of Chicago Press, 182
(2011), 419-541.
S.Popa, S. Vaes: Group measure space decomposition of II1 factors and W -superrigidity, Invent. Math. 182
(2010), no. 2, 371-417.
S. Popa, S. Vaes: Unique Cartan decomposition for II1 factors arising from arbitrary actions of free groups,
Acta Mathematica 212 (2014), 141-198.
S. Popa, S. Vaes: Unique Cartan decomposition for II1 factors arising from arbitrary actions of hyperbolic
groups, Journal fur die reine und angewandte Mathematik, 690 2014, 433-458.
I. M. Singer: Automorphism of finite factors, Amer. J. Math. 77 (1955), 117-133.
[Si55]
[TD14] R. D. Tucker-Drob: Invariant means and the structure of inner amenable groups, Preprint. arXiv:1407.7474.
S. Vaes: Rigidity results for Bernoulli actions and their von Neumann algebras (after Sorin Popa), Sminaire
[Va06]
Bourbaki, expos 961. Astrisque 311 (2007), 237-294.
COCYCLE SUPERRIGIDITY FOR THE COINDUCED ACTION
19
[Va10a] S. Vaes: Rigidity for von Neumann algebras and their invariants, Proceedings of the ICM (Hyderabad,
India, 2010), Vol. III, Hindustan Book Agency (2010), 1624-1650.
[Va10b] S Vaes: One-cohomology and the uniqueness of the group measure space decomposition of a II1 factor, Math.
[Zi84]
Ann. 355 (2013), 661-696.
R. Zimmer: Ergodic theory and semisimple groups, Monographs in Mathematics, 81. Birkhauser Verlag,
Basel, 1984. x+209 pp.
Mathematics Department; University of California, San Diego, CA 90095-1555 (United States).
E-mail address: [email protected]
|
1605.07435 | 2 | 1605 | 2016-05-30T15:29:59 | Weight theory for ultraproducts | [
"math.OA",
"math.FA"
] | For a family of von Neumann algebras $\mathcal{M}_j$ equipped with normal weights $\varphi_j$ we define the ultraproduct weight $(\varphi_j)_\omega$ on the Groh--Raynaud ultrapower $\prod_{j, \omega} \mathcal{M}_j$. We prove results about Tomita-Takesaki modular theory and consider ultraproducts of spatial derivatives. This extends results by Ando--Haagerup and Raynaud for the state case. We give some applications to noncommutative $L^p$-spaces and indicate how ultraproducts of weights appear naturally in transference results for Schur and Fourier multipliers. Using ideas from complex interpolation with respect to ultraproduct weights, we give a new proof of a theorem by Raynaud which shows that $\prod_{j, \omega} L^p(\mathcal{M}_j) \simeq L^p(\prod_{j, \omega} \mathcal{M}_j )$. We complement the paper by showing that spatial derivatives take a natural form in terms of noncommutative $L^p$-spaces. | math.OA | math |
WEIGHT THEORY FOR ULTRAPRODUCTS
MARTIJN CASPERS
Abstract. For a family of von Neumann algebras Mj equipped with normal weights ϕj
we define the ultraproduct weight (ϕj)ω on the Groh -- Raynaud ultrapower Qj,ω Mj. We
prove results about Tomita-Takesaki modular theory and consider ultraproducts of spatial
derivatives. This extends results by Ando -- Haagerup and Raynaud for the state case. We
give some applications to noncommutative Lp-spaces and indicate how ultraproducts of
weights appear naturally in transference results for Schur and Fourier multipliers. Using
ideas from complex interpolation with respect to ultraproduct weights, we give a new
proof of a theorem by Raynaud which shows that Qj,ω Lp(Mj) ≃ Lp(Qj,ω Mj ). We
complement the paper by showing that spatial derivatives take a natural form in terms of
noncommutative Lp-spaces.
1. Introduction
Ultraproducts form a very useful tool in the study of Banach spaces. This was first
observed in [DaCa72] after which many applications were found. We mention a couple
here. Firstly in von Neumann algebra theory ultraproducts play a very prominent role. This
was for example the case in the classification of hyperfinite factors [Con76], [Haa87]. Also
Connes' famous -- still unresolved -- embedding problem (equivalently the QWEP conjecture)
is formulated in terms of ultraproducts. Recall that it states that every separable II1-factor
embeds in an (Ocneanu) ultraproduct of the hyperfinite II1-factor. There are (further)
strong connections of ultraproducts with free probabiltiy theory, see e.g.
[Jun05], [Pop14]
or [Hou15] or model theory/descriptive set theory, see e.g. [FHS13], [BCI16].
A second application can be found in noncommutative harmonic analysis, i.e. harmonic
analysis for nonabelian groups, harmonic analysis on group von Neumann algebras or vec-
tor valued semi-commutative analysis. It is often very useful to asymptotically intertwine
Fourier multipliers to obtain sharp bounds for them or to discretize them. This asymp-
totic intertwining technique or "transference method" was applied in many places, see for
instance [NeRi11], [CaSa15], [CPPR15], [Ric15], [Gon16]. We may view such an asymptotic
intertwining property as a (non-asymptotic) intertwining property between ultraproduct
maps (we omit further details here but refer to these references and further discussions
below).
Whereas the definition of ultraproducts of Banach spaces is rather elementary it is often
quite non-obvious what properties the ultraproduct Banach space has. For example, if
Date: August 15, 2018.
1
2
MARTIJN CASPERS
one takes an ultraproduct of Lp-spaces, it is a non-trivial result that the resulting space
is again an Lp-space. The latter fact is due to Raynaud [Ray02].
In [Ray02] Raynaud
also proved that the ultraproduct of the Tomita -- Takesaki modular automorphism group
of states localizes to the modular automorphism group of the ultraproduct state, i.e. the
bounded case of Theorem A below. In addition Raynaud proves a similar result for the
Connes cocycle derivative.
More recently, in [AnHa14] Ando and Haagerup gave a complete outline of different ul-
traproducts of von Neumann algebras and their relations. In particular they considered the
Groh-Raynaud ultraproduct and the Ocneanu ultraproduct and showed that the Ocneanu
ultraproduct is isomorphic to a corner algebra of the Groh-Raynaud ultraproduct. Ando and
Haagerup proved many things about ultraproducts. For example they extended Raynaud's
results on Tomita -- Takesaki theory to ultraproducts of possibly different von Neumann al-
gebras with a new proof. They were also able to answer questions on asymptotic central
sequences, type and factoriality of ultraproducts.
In this paper we shall only be dealing with the Groh-Raynaud ultraproduct whose name
originates from [Gro84] and [Ray02]. Recall that it is defined as follows. Let ω be a
nonprincipal ultrafilter on N and for every j ∈ N let Mj be a von Neumann algebra.
Assume that Mj is represented on the standard Hilbert space Hj. Let H = Qj,ω Hj be
the ultraproduct Hilbert space. A bounded sequence (xj)ω with xj ∈ Mj acts on H by
(xj)ω(ξj)ω = (xjξj)ω. The closure in the strong topology of such (xj)ω in B(H) is called
the Groh-Raynaud ultraproduct. We denote this ultraproduct by M. In [Ray02] (see also
[Gro84]) it was proved that M∗ ≃ Qj,ω(Mj)∗ by extending the pairing h(ρj)ω, (xj)ωi =
limj,ω ρj(xj).
The aim of this paper is the study of ultraproduct weights. Whereas one can easily define
ultraproducts of (normal) functionals it is a priori not clear how ultraproducts of normal
weights should be defined. The problem lies in the fact that there is no bound on a sequence
(ϕj)ω and therefore a naive definition h(ϕj )ω, (xj)ωi = limj,ω ϕj(xj) does not make sense.
Indeed it could very well be that the right hand side of this expression is non-zero whereas
(xj)ω is (the equivalence class of) the 0 element.
In [AnHa14] ultraproducts of weights were discussed briefly for the first time. The ap-
proach is defined in the case that ϕj is a fixed weight on a fixed von Neumann algebra (i.e.
the weight and von Neumann algebra do not depend on j). In this case one sets (ϕ)ω = ϕ◦E
where E is the conditional expectation of the ultraproduct on the von Neumann subalgebra
of fixed sequences. This gives a different notion of ultraproduct weights than Definition 1.1,
see Remark 4.6. The weights from [AnHa14] are sometimes insufficient for applications. For
example one cannot prove Corollary 5.3 because there is very few control over the supports
of the ultraproduct weights from [AnHa14]. Moreover, typical weights appearing in the
literature do depend on the index. For example the weights occuring in [CaSa15, Section 5]
WEIGHT THEORY FOR ULTRAPRODUCTS
3
are given by Tr(PF · PF ) on B(L2(G)) where F are Følner sets on a locally compact group
G (the Følner sets give the net structure).
In this paper we introduce ultraproducts of weights in the following way.
Definition 1.1. Let ϕj be a normal (not necessarily faithful or semi-finite) weight on Mj.
Then we define the ultraproduct weight ϕ = (ϕj )ω as
(ϕj )ω := sup {ρ = (ρj)ω ρj ∈ (Mj)+
∗ , ρj ≤ ϕj,kρjk bounded in j}.
Note that the definition agrees with the ultraproduct of bounded normal functionals. We
now summarize the main results of this paper. Firstly we prove that the modular theory of
an ultraproduct localizes in the sense of the following theorem.
Theorem A. Let Mj be von Neumann algebras with normal, semi-finite, faithful weights
ϕj. Let M =Qj,ω Mj and let ϕ = (ϕj)ω be the ultraproduct weight on M. Assume that
ϕ is semi-finite. Let pϕ be the support projection of ϕ. Then,
t (pϕ(xj)ωpϕ) = pϕ(σϕj
σϕ
t (xj))ωpϕ.
We also prove the analogue of Theorem A for cocycle derivatives, see Theorem 4.17.
Theorem A generalizes the result in [Ray02] and [AnHa14] where it was proved for the case
that ϕj's are bounded uniformly in j. Note that [Ray02] and [AnHa14] give different proofs.
In this paper we give a third proof through the theory of spatial derivatives. In fact we
prove the following, which is (opposed to Theorem A) new even in the case that the ϕj's
are bounded functionals.
Theorem B. Let Mj be von Neumann algebras with normal, semi-finite, faithful weights
projection pϕ. Assume that ϕ is semi-finite. Let Jj be the modular conjugation acting on the
ϕj. Let M =Qj,ω Mj and let ϕ = (ϕj )ω be the ultraproduct weight on M with support
standard Hilbert space Hj of Mj and let Jω = (Jj)ω which acts on H =Qj,ω Hj. Suppose
that ψj are normal, semi-finite, faithful weights on the commutant M′j with ultraproduct
ψ = (ψj)ω. Assume that ψ is semi-finite. Let rψ be the support of ψ and set pψ =
JωrψJω, qψ = pψrψ. Assume that pϕ ≤ pψ. Then qψ(H) is an invariant subspace of(cid:16) dϕj
dψj(cid:17)ω
and,
(1.1)
dϕ
dψ
= qψ(cid:18) dϕj
dψj(cid:19)ω
qψ.
The statement of Theorem B uses ultraproducts of positive self-adjoint operators which
we define in this paper and takes values in the extended positive cone. So (1.1) needs to be
interpreted in the extended positive cone of M. Along the way of proving Theorem B we
also discuss whether or not every normal weight on M can be written as an ultraproduct of
weights on Mj. For states this is true, but for weights it no longer holds, see Proposition
4
MARTIJN CASPERS
4.18. In fact it seems that normal, semi-finite and faithful weights on an ultraproduct von
Neumann algebras rarely appear as an ultraproduct.
Using Theorem B we can show that ultraproduct von Neumann algebras admit locally
a nice interpolation structure. As a consequence we reprove Raynaud's theorem [Ray02,
Theorem 3.1].
Theorem C. We have,
(1.2)
Lp(Yj,ω
Mj) ≃Yj,ω
Lp(Mj).
The crucial novelty lies in how Theorem C was proved, see Theorem 5.2 and Corollary 5.3
(we found it more suitable to surpress these slightly more technical statements in Theorem
C making them more explicit in Section 5). We show that the isomorphism of (1.2) is
canonical and allows us to see Qj,ω Lp(Mj ) locally as complex interpolation spaces in a
natural way through this isomorphism. In many situations this complex interpolation is
useful. For example in [CaSa15] isometric embeddings of Lp-spaces were considered.
It
follows from our result that these isometries may be obtained through interpolation as well.
Similar situations/ideas occur in [CPPR15] and [Gon16].
In the last part of this paper, in Appendix A, we show that spatial derivatives take a
natural form in terms of noncommutative Lp-spaces. These results are probably known
amongst experts but have not appeared in the literature so far.
Structure. In Section 2 we recall preliminaries on ultraproducts and Tomita-Takesaki theory.
Section 3 contains preliminaries on noncommutative Lp-spaces associated with an arbitrary
von Neumann algebra. In Section 4 we introduce and study ultraproducts of weights. In
particular we prove Theorem A and Theorem B. Section 5 is devoted to a proof of Theorem
C. The Appendix A contains explicit forms of spatial derivatives on the standard form. We
use these results in the main part of this paper.
General notation. We use [x] for the closure of an operator x and x · y = [xy].
2. Preliminaries
For standard results on von Neumann algebras we refer to [Tak79] and [Tak00] and
adopt most of its notation. Von Neumann algebras will typically be denoted by M and
N and Hilbert spaces by H. N is usually an arbitrary von Neumann algebra whereas M
shall be an ultraproduct. We use N∗ for the predual and it induces the σ-weak topology
on N . We freely use the various other topologies on N summarized in [Tak79, Theorem
II.2.6]. N + is the cone of positive (bounded) operators in N and N +
∗ the space of positive
normal functionals. For x ∈ N and ρ ∈ N∗ we write ρx for the functional (ρx)(y) = ρ(xy)
and similarly (xρ)(y) = ρ(yx). For ξ ∈ H we write ρξ,ξ for the inner product functional
ρξ,ξ(x) = hxξ, ξi.
WEIGHT THEORY FOR ULTRAPRODUCTS
5
2.1. Weight theory. Let N be a von Neumann algebra. A weight on N is a map ϕ :
N + → [0,∞] that satisfies:
(1) ϕ(λx) = λϕ(x) for x ∈ N + and λ ≥ 0;
(2) ϕ(x + y) = ϕ(x) + ϕ(y) for x, y ∈ N +.
Alternatively one can say that ϕ preserves convex combinations on the cone N +. We set,
nϕ = {x ∈ N ϕ(x∗x) < ∞}.
From the inequality x∗y∗yx ≤ kyk2x∗x it follows that ϕ(x∗y∗yx) ≤ kyk2ϕ(x∗x) and there-
fore nϕ is a left ideal. We set mϕ to be the span of n∗ϕnϕ. Also set m+
ϕ = mϕ ∩ N +. Then
mϕ is the linear span of m+
ϕ . Recall the following definitions.
(1) The weight ϕ is called faithful if ϕ(x) = 0, x ∈ N + implies that x = 0;
(2) The weight ϕ is called semi-finite if nϕ is σ-weakly dense in N . This is equivalent
to σ-weak density of mϕ in N or σ-weak density if m+
ϕ in N +;
(3) The weight ϕ is called normal if ϕ(supk xk) = supk ϕ(xk) for every bounded increas-
ing net {xk}k∈K in N +.
Every weight in this paper shall be normal and from this point we assume that so is ϕ.
Let e ∈ N be the projection such that N e equals the σ-strong closure of nϕ. Let f ∈ N
be the projection such that N f equals {x ∈ N ϕ(x∗x) = 0}. ϕ is semi-finite on eN e and
faithful on fN f . The projection e − f is called the support of ϕ and we denote it by pϕ.
Note that in this paper we also consider supports of weights ψ on the commutant N ′ which
shall be denoted by rψ to distinguish. If N is represented on the standard Hilbert space we
set the opposite support pψ = JrψJ where J is the modular conjugation of the standard
form defined below.
Now assume that ϕ is normal, semi-finite and faithful. We have a non-degenerate inner
product,
nϕ × nϕ → C : (x, y) 7→ ϕ(y∗x),
from which we may complete nϕ to a Hilbert space Hϕ. We write Λϕ : nϕ → Hϕ for the
identification map. nϕ is a σ-weak/norm closed map. We have a normal representation of
N on Hϕ given by πϕ(x)Λϕ(y) = Λϕ(xy) where x ∈ N and y ∈ nϕ. We may identify N
with its image πϕ(N ) and shall omit πϕ in the notation. We let,
S0 : (⊆ Hϕ) → Hϕ : Λϕ(x) 7→ Λϕ(x∗),
x ∈ nϕ ∩ n∗ϕ.
S0 is preclosed and we set is closure to be S. S has polar decomposition S = J∆
2 . Here
the anti-linear isometry J is called the modular conjugation and the positive operator ∆ is
called the modular operator. We set σϕ
t (x) = ∆itx∆−it, t ∈ R, x ∈ N which is the modular
automorphism group of ϕ. Tomita-Takesaki theory shows that σϕ is a strongly continuous
1 -- parameter group of automorphisms of N (i.e. the non-trivial part is that it preserves N ).
1
6
MARTIJN CASPERS
We need a couple of standard tools from modular theory. Firstly we set,
Tϕ = {x ∈ N x is analytic with respect to σϕ and ∀z ∈ C : σϕ
z (x) ∈ nϕ ∩ n∗ϕ}.
If we view Tϕ as a subset of Hϕ then Tϕ is a Tomita algebra as explained in [Tak00]. For
a, b ∈ Tϕ there exists a normal state ϕab that is defined by
ϕab(x) = ϕ(bxσϕ
−i(a)).
The set of such states is dense in N∗. If ϕ is a state then ϕab = ϕab. The following lemma is
a standard approximation argument. Recall that on bounded sets the strong and σ-strong
topology agree so that the lemma can in fact be derived from (the proof of) [Ter82, Lemma
9].
Lemma 2.1. There exists a net aj ∈ Tϕ such that for every z ∈ C the net kσϕ
bounded and such that aj → 1 and σϕ
2.2. Standard forms. We first recall the definition of the standard form as it was reduced
to in [AnHa14, Lemma 3.19]. See also [Haa75].
i/2(aj) → 1 in the σ-strong topology.
z (aj)k is
Definition 2.2. Consider a 4-tuple (N ,H, J,P) consisting of a von Neumann algebra N
that is represented on a Hilbert space H, an anti-linear isometry J on H with J 2 = 1 and
P ⊆ H a closed convex cone which is self-dual, i.e. P = P 0 where
P 0 = {ξ ∈ H hξ, ηi ≥ 0, η ∈ P}.
Then (N ,H, J,P) is called a standard form if JN J = N ′, Jξ = ξ, ξ ∈ P, xJxJ(P) ⊆ P, x ∈
N .
Standard forms are unique in the sense that if (N ′,H′, J′,P′) is another standard form
for which N and N ′ are isomorphic, then there exists a unitary U : H → H′ such that
N = U∗N ′U, J = U∗J′U and UP = P′. Moreover, this unitary is unique. For ρ ∈ N∗ there
exists a unique vector ξ ∈ P such that ρ = ρξ,ξ. We will denote this vector with D
ρ , see
below.
1
2
2.3. The extended positive cone. Let N be a von Neumann algebra. The extended
positive cone N +
ext is defined as the space of all lower semi-continuous mappings N +
∗ → [0,∞]
that preserve convex combinations. Note that N +
ext is itself a cone, meaning that it is closed
under taking convex combinations and multiplication by scalars ≥ 0. Each x ∈ N +
ext admits
a spectral resolution
where eλ, λ ∈ [0,∞) is an increasing net of projections and p = 1 − supλ eλ. We take the
convention 0 · ∞ = 0. We may naturally view N + inside N +
ext as the set of all elements
for which the projection p on the infinite part is 0 and eλ is the spectral resolution of a
(2.1)
hx, ρiN +
ext,N +
∗
=Z ∞
0
λ dρ(eλ) + ρ(p) · ∞,
WEIGHT THEORY FOR ULTRAPRODUCTS
7
bounded operator (that is, there exists λ such that eλ = 1). For x ∈ N +
ext and q ∈ N a
projection we say that x and q commute if all eλ, λ ≥ 0 and p in (2.1) commute with q. We
shall also write P∞(x) for p and P<∞(x) for 1 − p.
2.4. Spatial derivatives. Spatial derivatives for von Neumann algebras have been intro-
duced by Connes in [Con80]. A comprehensive summary may also be found in [Ter82]. Let
N be a von Neumann algebra acting on a Hilbert space H. We emphasize that in general H
is not necessarily the standard form Hilbert space, but in this paper this is always the case.
We fix a normal, semi-finite, faithful weight ψ on the commutant N ′. Let (Hψ, Λψ, πψ) be
a GNS-construction for ψ. For ξ ∈ H we define Rψ(ξ) as the densely defined operator given
by the closure of the (preclosed) mapping:
x ∈ nψ.
Λψ(x) 7→ xξ,
(2.2)
Let D(H, ψ) denote the vectors for which the operator Rψ(ξ) is bounded. The vectors in
D(H, ψ) are called the ψ-bounded vectors. It is easy to check that Rψ(xξ) = xRψ(ξ), x ∈ N
and in particular that D(H, ψ) is invariant under N . Also Rψ(ξ)πψ(y) = yRψ(ξ), y ∈ N ′ so
that it follows that for ξ ∈ D(H, ψ) the operator Rψ(ξ)Rψ(ξ)∗ ∈ N . More generally there
exists θψ(ξ, ξ) ∈ N +
ext determined by
hρη,η, θψ(ξ, ξ)i =( kRψ(ξ)∗ηk2
Hψ
∞
η ∈ Dom(Rψ(ξ)∗)
otherwise.
Let ϕ be a normal (not necessarily semi-finite or faithful) weight on N . We define a
closed quadratic form qϕ(ξ) = hϕ, θψ(ξ, ξ)i which has domain all ξ ∈ D(H, ψ) such that
θψ(ξ, ξ) ∈ m+
dψ is defined
as the element in B(H)+
ϕ . For all other ξ we have qϕ(ξ) = ∞. Then the spatial derivative dϕ
ext determined by:
2
dψ(cid:19) 1
h(cid:18) dϕ
2
dψ(cid:19) 1
ξ,(cid:18) dϕ
ξi = qϕ(ξ).
If ϕ is semifinite then dϕ
of ϕ equals the support of dϕ
is invertible as an unbounded operator -- we have
dψ is an unbounded positive self-adjoint operator on H. The support
dψ . Recall also that if ϕ is semi-finite and faithful -- so that dϕ
dψ
(2.3)
(cid:18) dϕ
dψ(cid:19)it
x(cid:18) dϕ
dψ(cid:19)−it
=σϕ
t (x),
x ∈ N .
In Appendix A we relate spatial derivatives to noncommutative Lp-spaces.
3. Noncommutative Lp-spaces
Noncommutative Lp-spaces associated with an arbitrary von Neumann algebra were in-
troduced by various people.
In [Haa77], [Ter81] Haagerup defined them as subspaces of
weak Lp-spaces of the core of a von Neumann algebra (recall that the core is defined as
8
MARTIJN CASPERS
N ⋊σϕ R with ϕ a normal, semi-finite, faithful weight on N ). Later Hilsum [Hil81] gave an-
other definition based on Connes' spatial derivative [Con80]. Note that Hilsum's approach
in [Hil81] in fact uses Haagerup's earlier construction to show that his Lp-spaces form a
vector space. Other approaches through the complex interpolation method can be found in
the papers by Kosaki [Kos84b], Terp [Ter82] and Izumi [Izu97].
Here we use Hilsum's definition as it is closest to the spatial theory we already studied.
Of course our results translate into any of the other definitions mentioned.
3.1. Definitions. Let N be a von Neumann algebra acting on a Hilbert space H. Let ψ
be a normal, semi-finite, faithful weight on the commutant N ′. A closed densely defined
operator x on H is called γ-homogeneous with γ ∈ R if,
xa ⊆ σψ
iγ(a)x,
for every a ∈ N ′ analytic for σψ.
We now have the following theorem, see [Con80].
Theorem 3.1. Let x be a closed, densely defined operator on a Hilbert space H. Write
x = ux for the polar decomposition. Then the following are equivalent:
(1) x is (−1)-homogeneous.
(2) u ∈ N and x is (−1)-homogeneous.
(3) u ∈ N and x equals the spatial derivative dϕ
ϕ on N .
dψ for some normal, semi-finite weight
In particular every positive self-adjoint (−1)-homogeneous operator occurs as a spatial
derivative. Also note that if x ≥ 0 is (−1/p)-homogeneous then xp is (−1)-homogeneous.
This allows us to set the following definition.
Definition 3.2. Let N be a von Neumann algebra. Let ψ be a normal, semi-finite, faithful
weight on the commutant N ′. The Connes -- Hilsum Lp-space Lp(N , ψ) is defined as the
space of all closed, densely defined (−1/p)-homogeneous operators x such that if x = ux
is the polar decomposition then xp = dρ
∗ . We define,
dψ for some ρ ∈ N +
kxkp = ρ(1)1/p.
Connes -- Hilsum Lp-spaces are Banach spaces sharing many of the properties of classical
Lp-spaces, such as Holder estimates, reflexivity, interpolation, et cetera.
It is proved in
[Ter82] that for two choices of normal, semi-finite, faithful weights ψ1 and ψ2 we have that
Lp(N , ψ1) and Lp(N , ψ2) are isometrically isomorphic. We will write Lp(N ) in case it is
clear which weight ψ is chosen.
3.2. Interpolation. In [Ter82] Terp proved that Lp(N ) is a complex interpolation space
between N and its predual N∗. We give a brief description here. More details on the complex
interpolation method can be found in [BeLo76]. We also refer the reader to [Cas13] for a
slightly more elaborate discussion of what we introduce in this section.
WEIGHT THEORY FOR ULTRAPRODUCTS
9
Fix again a normal, semi-finite, faithful weight ψ on N ′ and denote the associated Lp-
spaces by Lp(N ). Let ϕ be a normal, semi-finite, faithful weight on N . We define the
space
K = {x ∈ N ∃ϕ(−1/2)
x
∈ N∗ s.t. ϕ(−1/2)
x
(y∗z) = hJx∗JΛϕ(z), Λϕ(y)i}.
It is proved in [Ter82] that mϕ ⊆ K. By definition there is a map j∞ : K ֒→ N : x 7→ x.
Also there exists a map j1 : K ֒→ N∗ : x 7→ ϕx. If we equip K with the norm kxkK =
max{kxk,kϕxk} then the embeddings j1 and j∞ are contractions. Dualizing them yields a
commutative diagram:
(3.1)
K
j1
=⑤⑤⑤⑤⑤⑤⑤⑤
!❈❈❈❈❈❈❈❈
j∞
∞
j ∗
"❉❉❉❉❉❉❉❉
<②②②②②②②②
j ∗
1
K∗,
M∗
M
Within K∗ it makes sense to speak about the intersection of M and M∗ and in fact it turns
out that M ∩ M∗ = K. Applying the complex intpolation method at parameter θ ∈ [0, 1]
to this diagram yields a Banach space (N ,N∗)[θ]. Let jp : K ֒→ (N ,N∗)[1/p], p ∈ [1,∞) be
the natural inclusion. It is proved in [Ter82] that for p ∈ [1,∞) we have
(N ,N∗)[ 1
p ] ≃ Lp(N ).
1
2
1
2
Moreover under this isomorphism the mapping j1 is given by sendingPj y∗j zj with yj, zj ∈
nϕ to
ϕy∗j · [zjd
d
Xj
ϕ with x = Pj(y∗j zj) for this expression as it does not depend
We simply write d
on the decomposition of x into summands (see [GoLi99] for similar considerations). For
completeness we mention that in [Izu97] also embeddings of suitable subspaces of N into
N∗ were considered that are different from K. Also we note that, for x, y ∈ K with 0 ≤ x ≤ y
we have
dϕ
dψ
where
dϕ =
ϕxd
ϕ],
1
2
1
2
.
,
and
ϕx ≤ kxkϕ.
(3.2)
ϕ(−1/2)
ϕ we have kϕ(−1/2)
≤ ϕ(−1/2)
k = ϕ(x).
x
y
x
For x ∈ m+
Remark 3.3 (Standard forms, Lp-spaces and the Dϕ-notation). In this paper we shall
use noncommutative L2-spaces in order to explicitly write down GNS-representations of
different weights on the same Hilbert space. Throughout the paper κ′ shall be a fixed
normal, semi-finite, faithful weight on a von Neumann algebra N ′. (In Section 4 onwards
we may assume that κ′j is a fixed nsf weight on M′j and κ′ is a fixed nsf weight on M, with
no relation between κ′j and κ′). We will use Lp(N ) = Lp(N , κ′). For a normal weight ϕ on
p
"
.
=
p
!
-
<
10
MARTIJN CASPERS
N we write
Dϕ =
The map ϕ → Dϕ is order preserving. We set,
for
Tr(Dϕ) = ϕ(1),
dϕ
dκ′
.
Dϕ ∈ L1(N )+,
and extend linearly to L1(N ). Suppose that ϕ is normal, semi-finite and faithful. For
ρ := ϕ(−1/2)
ϕ ⊆ K the associated D-operator is given by
with x ∈ m+
x
Dρ = D
1
2
ϕ√x(D
1
2
ϕ√x)∗,
1
2
1
2
1
2
ϕ xD
for which we shall write D
ϕ as explained above, see [Ter82, Eqn. (38)]. For z ∈ nϕ
ϕ is in L2(N ). We shall again omit such closures in the
we have that the closure of zD
notation as elements of noncommutative L2-spaces are closed by definition. We also have
that zDϕz∗ = Dzϕz ∗, see [Ter81, Theorem III.14]. In fact we can extend our notation a
bit in the following way. Suppose that ϕ is a normal, semi-finite (not necessarily faithful)
weight on N and let x ∈ m+
for the functional
ϕ then we write ϕ(−1/2)
N ∋ y 7→ ϕ(−1/2)
x
pϕxpϕ (pϕypϕ)
and note that the latter functional was defined in the corner pϕN pϕ. We have that ϕ(−1/2)
ϕ(−1/2)
y
if and only if pϕxpϕ ≤ pϕypϕ. Finally recall that [Ter82, Theorem II.36],
x
≤
is a standard form. For the inner product we have hξ, ηi = Tr(η∗ξ).
(N , L2(N ), J : x 7→ x∗, L2(N )+)
4. Ultraproducts of weights
Given a series of normal weights ϕj on von Neumann algebras Mj we define its ultraprod-
uct in a way that generalizes the ultraproduct of normal states. We also develop the spatial
theory of ultraproducts. Consequently we are able to prove results on Tomita -- Takesaki
theory (such as Theorem A in the introduction).
4.1. Groh-Raynaud ultraproducts. Let ω be a non-principal ultrafilter on the natural
numbers N. All of the constructions in this section will work for an arbitrary set instead of
N, but we take this assumption for simplicity. Another reason for making this convention is
that also the results in [AnHa14] were proved under the same assumption (with again the
remark that most constructions go through immediately for arbitrary sets). Indices over N
will typically be denoted by j and therefore we omit N in the notation from now on.
Suppose that Xj are Banach spaces. Then consider the space A of all series (xj)j with
xj ∈ Xj and supj kxjkXj < ∞. We put a semi-norm on A by setting k(xj)jk = limj,ω kxjkXj .
Let A0 be the space of all x ∈ A with kxk = 0. Then k · k descends to the quotient
space A/A0 and the latter space is called the Banach space ultraproduct which we denote
WEIGHT THEORY FOR ULTRAPRODUCTS
11
of von Neumann algebras differently.
denoted by (xj)ω. This notation is taken from [AnHa14]. Note however that [Ray02] uses
the notation (xj)• for (xj)ω.
by Qj,ω Xj. The equivalence class of a sequence (xj)j in Qj,ω Xj will from this point be
If each Xj is a Hilbert space then so isQj,ω Xj in a natural way. If each Xj is a C∗-algebra
then so isQj,ω Xj. It is not true that if each Xj is a von Neumann algebra then the Banach
space ultraproductQj,ω Xj is a von Neumann algebra. Therefore we define ultraproducts
Aj is a C∗-algebra acting on a Hilbert space Hj then (xj)ω ∈ Qj,ω
Qj,ω Hj by (xj)ω(ξj)ω = (xjξj)ω.
We recall the Groh-Raynaud ultraproduct of von Neumann algebras. First note that if
Aj acts naturally on
(1) For a family of von Neumann algebras Mj that are represented on a Hilbert space
Hj we define the abstract ultraproduct Qj,ω(Mj,Hj) as the closure in the strong
operator topology of the ∗-algebra of operators (xj)ω acting onQj,ω Hj.
Qj,ω Mj asQj,ω(Mj,Hj) where Hj is the standard form Hilbert space.
(2) For a family of von Neumann algebras Mj we define the Groh-Raynaud ultraproduct
As shown in [AnHa14, Remark 3.6] the strong closure in the above definitions is necessary.
In this paper if we write x = (xj)ω ∈ M we mean that x ∈ M is an element that can be
written as a sequence (xj)ω. That is, x is contained in the Banach space ultraproduct of
Mj.
Recall that besides the Groh-Raynaud ultraproduct there are other definitions of ul-
traproducts (such as the Ocneanu ultraproduct) which will not be used in this paper. A
complete and detailed account of different ultraproducts and their relations was given by
Ando -- Haagerup in [AnHa14].
It is proved in [Ray02] that there is a natural pairing between Qj,ω Mj (ultraproduct
of von Neumann algebras) andQj,ω(Mj)∗ (ultraproduct of Banach spaces) determined by
h(ρj)ω, (xj)ωi = limj,ω ρj(xj) and moreover through this pairingQj,ω Mj =(cid:16)Qj,ω(Mj)∗(cid:17)∗.
Recall also [AnHa14] that for the commutants we have (cid:16)Qj,ω Mj(cid:17)′ = Qj,ω M′j. We use
these facts without further reference.
4.2. Ultraproducts of unbounded operators. In this section we define the ultraproduct
of elements in the extended positive cone (Mj)+
ext. We do this using a bounded transform.
We define the continuous increasing and operator monotone function
F (s) =
s
1 + s
,
s ∈ [0,∞).
For x ∈ M+
ext with decomposition (2.1) there is a unique F (x) ∈ M+ given by
hF (x), ρiM,M∗ =Z ∞
0
F (λ) dρ(eλ) + ρ(p).
12
MARTIJN CASPERS
Conversely for x ∈ M+ with kxk ≤ 1 and spectral decomposition x = R ∞0 λ deλ we set
F −1(x) ∈ M+
ext by
hF −1(x), ρiM,M∗ =Z ∞
0
F −1(λ) dρ(eλ) + ρ(p) · ∞,
where p = R ∞0 δ1(λ)deλ, i.e. the projection on the eigenspace of the eigenvalue 1. As F
is operator monotone and strongly continuous it preserves suprema of bounded increasing
nets. Now if Mj are von Neumann algebras and xj ∈ M+
j,ext then we set its ultrapower by
(xj)ω = F −1((F (xj ))ω).
(4.1)
Note that if xj ∈ M+
equals zero so that our notation (xj)ω is unambiguous.
j,ext is contained in M+ and its uniform norm converges to 0 then (4.1)
We need some results on spectral calculus on ultraproducts. Firstly for a continuous
function f and x = (xj)ω ∈ M we have
f ((xj)ω) = (f (xj))ω,
(xj)ω ∈ M.
Indeed this can be seen by approximating f with polynomials. In the unbounded situation
we need two lemmas.
Lemma 4.1. Let f : [0,∞) → [0,∞) be a continuous function for which f (t) has a limit
in [0,∞] as t → ∞. Let xj ∈ (Mj)+
Proof. As the limit f (t), t → ∞ exists we may view F ◦ f ◦ F −1 as a continuous function
[0, 1] → [0, 1]. Then we have,
ext. Then f ((xj)ω) = (f (xj))ω.
f ((xj)ω) = f (F −1((F (xj))ω)) = F −1 ◦ (F ◦ f ◦ F −1)(F (xj )ω)
=F −1(((F ◦ f ◦ F −1) ◦ F (xj))ω) = F −1((F (f (xj)))ω) = (f (xj))ω.
(cid:3)
The statement of the previous lemma does not hold for the function f (t) = eit, see
[AnHa14, Example 4.7]. However if we restrict to a suitable spectral subspace we get the
following lemma. We use χA for the indicator function on a set A.
Lemma 4.2 (Lemma 4.4 of [AnHa14]). Let xj be positive self-adjoint operators affiliated
with a von Neumann algebra Mj. Let x = (xj)ω ∈ M+
ext be its ultrapower. Let K ⊆ H be
λ ,λ)(x)H. Then for every t ∈ R,
the space given by ∪λ>0χ( 1
(xK)it = (xit
j )ωK.
For an operator A ∈ B(H)+
ext we set P∞(A) for the projection p defined in the spectral
resolution (2.1). We also set P<∞(A) = 1− p. The next lemma shows that we may compute
values of ultraproducts in a natural way.
Lemma 4.3. Let Aj be (unbounded) positive self-adjoint operators acting on a Hilbert space
Hj with domain Dom(Aj). Let A := (Aj)ω be its ultraproduct which is contained in B(H)+
ext
WEIGHT THEORY FOR ULTRAPRODUCTS
13
with H =Qj,ω Hj. Let ξj ∈ Dom(Aj) and suppose that kξjk2 is bounded so that we may
put ξ = (ξj)ω ∈ H. We have,
(4.2)
hA, ωξ,ξi ≤ lim
j,ω kA
1
2
j ξjk2
2.
Moreover, let ξ ∈ P<∞(A)H be such that supj kA
(4.3)
j ξjk2
2 (ξj)ω = P<∞(A)(A
A
1
2
1
j ξj)ω.
1
2
2 is finite. Then we get an equality,
Proof. Let Gλ be a continuous function supported on [0, 2λ] such that 0 ≤ Gλ ≤ 1 and
Gλ(s) = 1 for all s ∈ [0, λ]. The projection P<∞(A) is given by the supremum of the
projections Gλ(A), λ ≥ 0. Write ξ1 = P<∞(A)ξ, ξ2 = (1 − P<∞(A))ξ. Then, writing
ξ1 = (ξ1,j)ω,
(4.4)
hA, ρξ1,ξ1i = sup
λ>0 hA, ρGλ(A)ξ1,Gλ(A)ξ1i = sup
j ξ1,jk2
j Gλ(Aj)ξ1,jk2
2.
2 ≤ lim
j,ω kA
1
2
1
2
= sup
λ>0
j,ω kA
lim
λ>0 h(AjGλ(Aj))ω, ρξ1,ξ1i
If ξ2 6= 0 then write ξ2 = (ξ2,j)ω. There exists 0 <
This shows (4.2) in case ξ2 = 0.
ǫ < kξ2k such that for every λ > 0 there exists U ∈ ω such that for all j ∈ U we have
2 = ∞. The inequality (4.2)
kχ(λ,∞)(Aj)ξ2,jk > kξ2k − ǫ. But then clearly limj,ω kA
then follows.
j ξjk2
1
2
1
2
It remains to prove the final statement. Take ξ = (ξj)ω ∈ P<∞(A)H, suppose that
2 is finite. Firstly note that A acts on P<∞(A)H as an unbounded operator
2 by (4.2). Note that (as in
supj kA
and that the assumptions ensure that ξ is in the domain of A
(4.4)) we get again,
j ξjk2
1
1
2
P<∞(A)(A
j ξj)ω = lim
λ→∞
Gλ(A)(A
j ξj)ω
1
2
(Gλ(Aj)A
1
2
= lim
λ→∞
j ξj)ω = lim
λ→∞
Gλ(A)A
1
2 (ξj)ω = A
1
2 (ξj)ω.
This concludes the proof.
(cid:3)
Note that in (4.3) the projection P<∞(A) is necessary. For example, let H = L2(R).
Take ξj = j−1/2χ[j,j+1] and let Aj be the multiplication operator with the identify function
s 7→ s. Then (ξj)ω = 0 whereas (Ajξj)ω 6= 0.
4.3. Ultraproducts of weights. For two normal weights ρ and ϕ on a von Neumann
algebra M we say that ρ ≤ ϕ if ρ(x) ≤ ϕ(x) for every x ∈ M+. For an increasing net (ρk)k
of normal weights we say that ϕ = supk ρk if ϕ is the smallest weight majorizing each ρk.
In fact for an increasing net (ρk)k with ρk ∈ M∗ we may define a normal weight ϕ as,
ϕ(x) = sup
k hρk, xiM∗,M,
x ∈ M+.
14
MARTIJN CASPERS
Note that ϕ is not necessarily semi-finite. If the net (ρk)k were to be bounded then ϕ ∈ M∗
and the convergence ρk → ϕ is in norm (as kρk − ϕk = ϕ(1) − ρk(1) → 0).
Now we recall the following theorem, which is a strengthening of [Tak00, Theorem
VII.1.11 (iv)]. We give a proof here.
Theorem 4.4. Let ϕ be a normal, semi-finite, faithful weight on a von Neumann algebra
N . Consider the set
Fϕ = {ρ ∈ N∗ ρ ≤ ϕ} .
Then ϕ = supρ∈Fϕ ρ.
Proof. We need to prove that Fϕ is a net in N∗ (note that this is stronger than [Tak00,
Theorem VII.1.11]). Take ρ ∈ N∗ with ρ ≤ ϕ. Let Dρ and Dϕ be the associated densities.
As Dρ ≤ Dϕ we have D− 1
In fact xρ := [D− 1
ϕ ] ∈ N as xρ is 0-
Indeed, take y, z ∈ nϕ.
homogeneous. We claim moreover that x ∈ K (see Section 3).
As,
ϕ DρD− 1
ϕ DρD− 1
ϕ ≤ 1.
2
2
2
2
[zD
2
2
1
2
ϕ ] · [D− 1
ϕ ][D− 1
ϕ DρD− 1
ϕ DρD− 1
ϕ ] · D
ϕ ]D
1
2
1
2
1
2
ϕ y∗ ⊇ zD
1
2
ϕ D− 1
ϕ DρD− 1
ϕ D
2
2
1
2
ϕ y∗ ⊇ zDρy∗,
we have that [zD
is actually an equality by [Hil81, Theorem 4]. Then,
2
2
ϕ y∗ ⊇ z · Dρ · y∗ and as z · Dρ · y∗ ∈ L1(N ) the inclusion
(4.5)
y∗z 7→ hJx∗ρJΛ(z), Λ(y)i = hxρJΛ(y), JΛ(z)i = Tr([zD
=Tr([zD
ϕ DρD− 1
ϕ y∗) = Tr(z · Dρ · y∗) = ρ(y∗z),
ϕ ] · [D− 1
ϕ ] · D
1
2
1
2
2
2
ϕ ] · xρ · D
1
2
1
2
ϕ y∗)
xρ
extends boundedly to the normal functional ϕ(−1/2)
= ρ, so that xρ ∈ K. Now let ρ1, ρ2 ∈
Fϕ. Let xρ1 and xρ2 be the associated operators as before. The function F (s) = s(1+s)−1 is
operator monotone and operator concave. We let x ∈ N + be such that F −1(x) = F −1(xρ1)+
F −1(xρ2 ) (note that this equality takes values in N +
ext on which we have well-defined spectral
calculus). We have xρ1 ≤ x, xρ2 ≤ x by operator monotonicity of F and further x ≤ 1.
Furthermore as F is operator concave and F (2F −1(s)) = 2s(1 + s)−1 ≤ 2, s ∈ R, we have,
x =F (F −1(xρ1) + F −1(xρ2)) = F(cid:18) 1
≤F (2F −1(xρ1)) + F (2F −1(xρ2)) ≤ 2(xρ1 + xρ2).
(2F −1(xρ1) + 2F −1(xρ2))(cid:19)
2
x
ej
xρi
≤ ϕ(−1/2)
ր ϕ. So ϕ = supρ∈Fϕ ρ.
≤ ϕ. This shows that Fϕ is a
ϕ , ej ≤ 1 converging to 1 strongly and such that kejk ≤ 1 gives
So x ∈ K. Therefore for i = 1, 2 we get ϕ(−1/2)
net. Taking an net ej ∈ m+
that ϕ(−1/2)
Definition 4.5. As before let Mj be von Neumann algebras and let M = Qj,ω Mj be
its ultraproduct. Let ϕj be a normal weight on the von Neumann algebra Mj. Consider
the set F := Fϕj,j∈N of all ultraproducts (ρj)j ∈ M∗ with ρj ∈ Fϕj and kρjk bounded in
(cid:3)
WEIGHT THEORY FOR ULTRAPRODUCTS
15
j. Then set ϕ = supρ∈F ρ. We call ϕ the ultraproduct weight and we denote it by (ϕj)ω.
Clearly ϕ is normal.
Remark 4.6. In [AnHa14] another ultraproduct of weights was considered. Suppose that
N is a von Neumann algebra equipped with a fixed normal weight ϕN . Let Mj = N and
let M =Qj,ω N be its ultraproduct. There exists a conditional expectation E : M → N :
(xj)ω 7→ limj,ω xj, where the limit is understood in the σ-weak topology. The normal weight
ϕM = ϕN ◦ E is in [AnHa14] defined as the ultraproduct of the fixed series of weights ϕN .
This definition does not agree with ours. Take for example N = L∞(R) with weight ϕN the
integral against the standard Lebesgue measure. Take fj the indicator function on [j, j + 1].
Then fj → 0 in the σ-weak topology. So ϕN ◦ E((fj)ω) = 0. Whereas h(ϕ)ω, (fj)ωi = 1. It
is easy to check that in general ϕM ≤ (ϕ)ω.
4.4. Spatial theory of ultraproducts. Let again Mj be a sequence of von Neumann
algebras and let (Mj,Hj, Jj, Pj ) be its standard form which we may and will identify
(uniquely, see discussion below Definition 2.2) with (Mj, L2(Mj), Jj , L2(Mj)+). Let M =
Qj,ω Mj be its Groh-Raynaud ultrapower which acts on the ultraproduct Hilbert space
H =Qj,ω Hj. By [AnHa14] (M,H, Jω,P) is a standard form where P is the positive cone
given by the vectors (ξj)ω with ξj ∈ Pj. Therefore we may and will identify (M,H, Jω,P)
with (M, L2(M), J : x 7→ x∗, L2(M)+) and again there is a unique such identification.
Recall also that for a normal, semi-finite, faithful weight ϕj on Mj the map z 7→ zD
ϕj is
a GNS-construction and the analogous result holds for weights on M. If ρ = (ρj)ω is a
vector in M∗ we get vectors D
ρj in the respective positive cones P and Pj. Then
ρ = (D
D
For the next lemma we use the notation kzk2,ϕ = ϕ(z∗z). Also recall again the conven-
ρj )ω. In the weight case we get Lemma 4.10 below.
ρ and D
tions on closures of operators in Lp-spaces from Remark 3.3.
1
2
1
2
1
2
1
2
1
2
Lemma 4.7. Let N be a von Neumann algebra. Suppose that (ρk)k∈K is an increasing net in
N∗ with supremum a normal, semi-finite weight ϕ. Let z ∈ nϕ. Then, kzD
ϕk2 → 0.
Proof. Without loss of generality we may assume that ϕ is faithful, because if it is not then
we may consider ϕ + ϕ′ instead where ϕ is a normal, semi-finite weight with support equal
to 1 − pϕ and for which z ∈ nϕ′. Then ϕ′ = supρ′∈Fϕ′ ρ′ so that the ρk + ρ′ with k ∈ K and
ρ′ ∈ Fϕ′ form an increasing net converging to ϕ + ϕ′. Then, assuming that we can prove
the lemma for ϕ + ϕ′, we have kzD
ϕ+ϕ′k2
ϕ+ϕ′k2 → 0. By orthogonality,
ϕk2
ρk − zD
ρk+ρ′ − zD
2 = kzD
2 + kzD
ρk − zD
ρ′ − zD
ϕk2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
kzD
ρk − zD
ρk+ρ′ − zD
ϕk2 → 0.
1
2
1
2
so that kzD
2
Because ϕ is faithful we may consider its Tomita algebra Tϕ. We first prove the lemma
ϕ . So that xρk =
under the assumption that z ∈ Tϕ. As ρk ≤ ϕ we have that D
ρk ≤ D
1
2
1
2
16
MARTIJN CASPERS
4
4
1
2
ϕ
ρk D− 1
D− 1
is an element of N with norm ≤ 1. Moreover (xρk )k∈K is an increasing net
ϕ D
with supremum equal to 1 (see the proof of Theorem 4.4). In particular xρk → 1 in the
strong topology. Now,
(4.6)
kzD
=hzD
=hzD
1
2
ρk − zD
ρk , zD
1
2
1
2
1
2
ρk , zD
1
2
2
ϕk2
ρki + hzD
ρki + hzD
1
2
1
2
1
2
ϕ , zD
ϕ , zD
1
2
ϕi − hzD
ϕi − hzD
1
2
1
2
1
2
1
2
ϕ , zD
ϕ , zD
ρki − hzD
ϕ xρk D
1
4
1
4
ϕi − hzD
1
2
ρk , zD
1
2
ϕi
ϕ xρk D
1
4
1
4
ϕ , zD
1
2
ϕi.
We have, c.f. Remark 3.3,
1
2
ρk , zD
1
2
hzD
ρki = Tr (zDρk z∗) = Tr (Dzρkz ∗) → Tr (Dzϕz ∗) = Tr (zDϕz∗) = hzD
1
2
ϕ , zD
1
2
ϕi.
Then we get as xρk → 1 strongly (and as z ∈ Tϕ each of the following expressions is well
defined),
1
2
hzD
1
4
ϕ , zD
ϕ xρk D
=Tr(cid:18)xρk D
1
4
ϕ z∗zD
1
4
1
4
ϕi = Tr(cid:18)D
ϕ(cid:19) → Tr(cid:18)D
3
4
ϕ xρk D
1
4
ϕ z∗zD
1
4
ϕ z∗zD
1
2
ϕ(cid:19)
ϕ(cid:19) = hzD
3
4
1
2
ϕ , zD
1
2
ϕi.
And similarly,
This shows that (4.6) converges to 0.
1
4
hzD
1
4
ϕ xρk D
ϕ , zD
1
2
ϕi → hzD
1
2
ϕ , zD
1
2
ϕi.
Next we treat a general z ∈ nϕ. Let zl ∈ Tϕ be such that kz − zlk2,ϕ → 0. Recall again
that kz − zlk2,ρk ≤ kz − zlk2,ϕ. So let ǫ > 0 and take l such that kz − zlk2,ϕ < ǫ. Then let
k0 be such that for k ≥ k0 we have kzlD
1
2
1
2
ρk − zlD
ρkk2 + kzlD
1
2
ϕk < ǫ. We get that,
ρk − zlD
ϕk2 + kzlD
1
2
1
2
1
2
1
2
ϕ − zD
ϕk2 < 3ǫ.
1
2
kzD
ρk − zD
1
2
1
2
ϕk2 ≤ kzD
ρk − zD
1
2
ρk − zlD
ϕk2 → 0.
1
2
This shows that kzD
Lemma 4.8. For every j let pj be an atomic projection in Mj. Then (pj)ω is atomic in
M.
Proof. The set (pj)ωM(pj)ω is the closure of all series (pj)ω(xj)ω(pj)ω = (pjxjpj)ω where
(cid:3)
Lemma 4.9. Let ϕ = (ϕj)ω be an ultrapower of normal, semi-finite, faithful weights ϕj.
(xj)ω ∈ M. So it equals the Banach space ultraproductQj,ω pjMjpj ≃Qj,ω
Assume that ϕ is semi-finite. For ρj ∈ Fϕj ⊆ (Mj)∗ set xρj = D− 1
the supremum over ρ = (ρj)ω ∈ F := Fϕj ,j∈N of (xρj )ω is 1.
Proof. Firstly note that the elements (xj)ω with xj ∈ m+
net that converges to 1 (see the proof of Theorem 4.4). Recall that for xj ∈ m+
xj = xρxj
ϕj and xj ≤ 1 form an increasing
ϕj we have
(see (4.5)), but there is not necessarily a uniform bound
where ρxj := ϕ(−1/2)
ϕj . We have that
ϕj Dρj D− 1
C ≃ C.
j,xj
(cid:3)
2
2
WEIGHT THEORY FOR ULTRAPRODUCTS
17
1
ϕj and xj ≤ 1. Take a spectral decomposition
on the ρxj . Now fix some (xj)ω with xj ∈ m+
xj =R λ dExj (λ). Note that the support of the spectral measure is contained in [0, 1]. Let
Sj ⊆ (0, 1] be the set of (non-zero) atoms, i.e. the set of s ∈ (0, 1] such thatR δsdExj (λ) is
non-zero, say xj,s. For later use let exj = xj −Ps∈Sj
decompose xj,s further into xj,s =Pk∈Kj,s
closed intervals Ij,t ⊆ (0, 1] indexed by t such that yj,t =RIj,t
xj,s be the non-atomic part of xj. We
xj,s,k in such a way that either kρxj,s,kk ≤ 1 or
xj,s,k is an atomic projection in Mj. For the non-atomic part for every j we may take
λ dEexj (λ) has the property
kxj,s,kk
that kρyj,tk ≤ 1.
Now let A be the set of all ultraproducts (zj)ω where zj is one of the xj,s,k's or yj,t's
defined above. We claim that limj,ω ϕj(zj) < ∞. Indeed if this were not to be the case
then for every C > 1 there exists U ∈ ω such that for every j ∈ U we have ϕj(zj) > C.
But this means that for every j ∈ U , we had chosen zj to be a multiple of an atomic
projection, i.e. pj := 1
zj. Then (pj)ω is an atomic projection on M (see Lemma 4.8)
kzjk
and (pj)ωM(pj)ω ≃ C. But then h(ϕj )ω, (pj)ωi = limj,ω ϕj(pj) > C. We may therefore
take decreasing sets U1 ⊇ U2 ⊇ . . . in ω such that we may pick for every j ∈ Uk an
atomic projection pj such that ϕj(pj) > k. Then h(ϕj )ω, (pj)ωi = limj,ω ϕj(pj) = ∞ which
contradicts that ϕ is semi-finite (as (pj)ω is atomic).
The previous paragraph shows that for (zj)ω ∈ A we have that ρzj is bounded in j (as
kρzjk = ϕj(zj)) and so (ρzj )ω ∈ F. By construction (xj)ω is the smallest element that is
bigger then all (zj)ω ∈ A. Then as all elements (xj)ω with xj ∈ m+
ϕj have supremum 1 we
see that also the supremum of (xρj ) with (ρj)ω ∈ F equals 1.
(cid:3)
Lemma 4.10. Let ϕ = (ϕj )ω be an ultrapower of normal, semi-finite, faithful weights ϕj.
Assume that ϕ is semifinite. Let zj ∈ nϕj be bounded in j such that also kzjk2,ϕj is bounded.
Then (zj)ω ∈ nϕ and,
1
2
(zjD
ϕj )ω = (zj)ωD
ϕ .
1
2
Proof. Firstly, with F as in Definition 4.5, as,
j,ω hρj, z∗j zji ≤ lim
lim
h(ϕj )ω, (z∗j )ω(zj)ωi = sup
ρ∈F
j hϕj, z∗j zji < ∞,
it follows that (zj )ω ∈ nϕ. Now by Lemma 4.10,
ρ = lim
ρ∈F
ϕ = lim
ρ∈F
(zj )ωD
(zj)ωD
1
2
1
2
(zj)ω(D
1
2
ρj )ω = lim
ρ∈F
1
2
(zjD
ρj )ω,
ϕj )ω. To see this let yj ∈ nϕj be bounded in j such
and we claim that this limit equals (zjD
that also ky∗jk2,ϕj is bounded. Then, using the notation of Lemma 4.9 and using [Ray02,
Theorem 5.1] in the last equality,
1
2
lim
ρ∈F
= lim
ρ∈F
(zjD
ρj )ω(y∗j )ω = lim
ρ∈F
(zjD
1
2
ρj y∗j )ω
1
2
1
4
(zjD
ϕj xρj D
1
4
ϕj y∗j )ω = lim
ρ∈F
1
4
(zjD
ϕj )ω(xρj )ω(D
1
4
ϕj y∗j )ω.
18
MARTIJN CASPERS
1
2
1
2
1
2
ϕj y∗j )ω = (zjD
ϕj )ω(y∗j )ω by Lemma 4.9. Then let ξ be the norm
which converges to (zjD
ρj )ω. We see that for all (yj)ω with ky∗jk2,ϕj bounded we have ξ(y∗j )ω =
limit limρ∈F (zjD
ϕj )ω (indeed we can take a net of
(zjD
(yj)ω's that are positive and increasing to 1 as follows from the argument given in the proof
of Lemma 4.9). This concludes the proof.
(cid:3)
ϕj )ω(y∗j )ω. But this can only happen if ξ = (zj D
1
2
1
2
Lemma 4.11. Let ϕ = (ϕj )ω be an ultrapower of normal, semi-finite, faithful weights ϕj.
Assume that ϕ is semifinite. Let yj, zj ∈ nϕj be bounded in j such that also kyjk2,ϕj and
kzjk2,ϕj are bounded. Then,
(ϕ(−1/2)
y∗
j zj
)ω = ϕ(−1/2)
j zj)ω
(y∗
Proof. We have using Lemma 4.10 for the second equality,
)ω, (xj)ωi = hJω(xj)∗ωJω(yjD
1
2
ϕj )ω, (zj D
1
2
ϕj )ωi
y∗
j zj
h(ϕ(−1/2)
=hJω(xj)∗ωJω(yj)ωD
1
2
ϕ , (zj )ωD
1
2
ϕi = hϕ(−1/2)
j zj)ω
(y∗
, (xj)ωi.
(cid:3)
Lemma 4.12. Let ϕ = (ϕj )ω be an ultrapower of normal, semi-finite, faithful weights
ϕj. Assume that ϕ is semifinite. Let z ∈ nϕ. Then there exists a net zk = (zk,j)ω with
zk,j ∈ nϕj such that kzk,jk2,ϕj is bounded in j and pϕ(zk,j)ωpϕ is bounded in k. Moreover
pϕ(zk,j)ωpϕ → pϕzpϕ strongly and Λϕ(pϕ(zk,j)ωpϕ) → Λϕ(pϕzpϕ) in norm.
Proof. To start with take z ∈ m+
and such that k√ajk2,ϕj is bounded and ϕ(−1/2)
4.4 shows that A is a net with respect to the order on positive elements. We claim that
(4.7)
ϕ . Consider the set A of elements a = (aj)ω with aj ∈ m+
ϕj
. The same proof as in Theorem
(aj )ω ≤ ϕ(−1/2)
pϕapϕ = pϕzpϕ.
z
sup
a∈A
j,b
j,bj
ϕj such that ϕ(−1/2)
z−y = (ϕ(−1/2)
j,bj
k ≤ kϕ(−1/2)
z−y
ϕj are dense in (Mj)+
k. That is ϕj(bj) ≤ kϕ(−1/2)
z−y
∗ . Therefore we may find bj ∈ m+
In particular this shows using Lemma 4.11 that ϕ(−1/2)
Indeed let y be such that supa∈A pϕapϕ = pϕypϕ. Note that pϕypϕ ≤ pϕzpϕ so that pϕypϕ ∈
ϕ . As the weights ϕj are normal semi-finite and faithful, the functionals ϕ(−1/2)
m+
with
b ∈ m+
We may assume moreover that kϕ(−1/2)
ebj = bj(1 + bj)−1. We get thatebj ≤ bj and kqebjk2,ϕj ≤ ϕj(bj) is bounded. Seteb = (ebj)ω.
We conclude thateb := (ebj)ω ∈ A. This also implies that pϕebpϕ ≤ pϕ(z − y)pϕ. But then for
We see that for every a ∈ A we have a +eb ∈ A. But this can only happen ifeb = 0 in which
pϕ(a +eb)pϕ ≤ pϕypϕ + pϕ(z − y)pϕ = pϕzpϕ.
case pϕ(z − y)pϕ = 0. This concludes (4.7). As suprema are attained in the strong toplogy
= ϕ(−1/2)
j,(ebj )ω ≤ (ϕ(−1/2)
)ω.
k. Then set
every a ∈ A we have
)ω = ϕ(−1/2)
z−y
j,bj
eb
.
WEIGHT THEORY FOR ULTRAPRODUCTS
19
we get,
pϕapϕ = pϕzpϕ.
lim
a∈A
We also get that,
(4.8) kΛϕ(pϕzpϕ − pϕapϕ)k2
Note that √pϕzpϕ ∈ nϕ and we may write
2 = ϕ((pϕzpϕ)2) + ϕ((pϕapϕ)2) − ϕ(pϕzpϕapϕ) − ϕ(pϕapϕzpϕ)
pϕapϕ = (pϕzpϕ)1/2((pϕzpϕ)−1/2(pϕapϕ)(pϕzpϕ)−1/2)(pϕzpϕ)1/2,
and xa := (pϕzpϕ)−1/2(pϕapϕ)(pϕzpϕ)−1/2 is increasing over a ∈ A with supremum equal
to the support of pϕzpϕ. By normality of ϕ we see that,
ϕ((pϕapϕ)2) → ϕ((pϕzpϕ)2),
ϕ(pϕzpϕapϕ) = ϕ((pϕzpϕ)
3
and
2 xa(pϕzpϕ)
1
2 ) → ϕ((pϕzpϕ)2).
Similarly ϕ(pϕapϕzpϕ) → ϕ((pϕzpϕ)2). So (4.8) converges to 0.
Next we need to prove the lemma for general z ∈ nϕ through a standard core argument.
Firstly note that as every element in mϕ can be written as the sum of four elements in m+
ϕ ,
the result follows for mϕ. Now take z ∈ pϕnϕpϕ. Let {el}l be a bounded net in pϕnϕpϕ of
positive elements approximating 1 in the strong topology. Note that elz ∈ mϕ. We have
elz → z strongly and Λϕ(pϕelzpϕ) → Λ(pϕzpϕ) in norm.
In turn the beginning of the
proof shows that we may approximate pϕelzpϕ in the graph norm of Λϕ with bounded nets
of elements pϕzkpϕ = pϕ(zk,j)ωpϕ such that kzk,jk2,ϕj is bounded in j and pϕ(zk,j)ωpϕ is
bounded in k.
(cid:3)
Lemma 4.13. Let ϕ = (ϕj )ω be an ultrapower of normal, semi-finite, faithful weights ϕj.
Assume that ϕ is semifinite. Let pϕ be the support of ϕ. Then there exists a bounded net
zk = (zk,j)ω on M such that zk → pϕ in the σ-weak topology and such that zk,j are analytic
for σϕj with σϕj
i/2(zk,j))ω → pϕ in the σ-weak
topology.
i/2(zk,j) bounded (in both j and k) and with (σϕj
t )ω, xi. Note that t 7→ (σϕj
t )ω. For x ∈ M
t )ω is not necessarily
strongly continuous (in fact in [AnHa14] it is proved it is usually not) but it is measurable
t )ω(x)dt as the element determined by
Proof. Firstly for t ∈ R consider the ultraproduct of ∗-homomorphisms (σϕj
we have h(ρj )ω, (σϕj
in the sense that for f ∈ L1(R) we defineRR f (t)(σϕj
h(ρj)ω,RR f (t)(σϕj
with kxjk and kx∗jk2,ϕ bounded. Then, applying Lemma 4.10 twice,
t (xj))ω = D
t )ω(x)i = h(ρj ◦ σϕj
t )ω(x)dti = h(RR f (t)ρj ◦ σϕj
Now we claim that pϕ is invariant for this homomorphism. Indeed, take x = (xj)ω ∈ M
t )ω, xi.
ϕ x) = (∆it
ϕj xj)ω = (D
t (xj))ω.
ϕ (σϕj
1
2
ϕj σϕj
j )ω(D
(∆it
j )ω(D
1
2
1
2
1
2
1
2
j )ω(D
Hence (∆it
pϕH. Then it follows that (σϕj
ϕ x) ⊆ pϕH. Since ((∆it
t )ω(pϕ) = (∆it
j )ω)−1 = (∆−it
j )ωpϕ(∆−it
j
j
)ω we see that actually (∆it
)ω = pϕ.
j )ωpϕH =
20
MARTIJN CASPERS
By Kaplansky's density theorem let yk = (yk,j)ω be a bounded net converging in the
v (zk,j) =
t (yk,j)dt. Then we have σϕj
e−t2 σϕj
t (yk,j)dt which is bounded. Also, for ρ = (ρj)ω ∈ M∗,
−∞
−∞
e−(t−v)2 σϕj
σ-weak topology to pϕ. We set zk,j = 1√πR ∞
1√πR ∞
√πZ ∞
√πZ ∞
=h(ρj)ω,Z ∞
hρ, (zk,j )ωi =h(
→h(
−∞
−∞
−∞
1
1
e−t2
e−t2
=hρ, pϕi,
so that (zk,j)ω → pϕ σ-weakly. Similarly,
t ( · )dt)ω, (yk,j)ωi
t ( · )dt)ω, pϕi
ρj ◦ σϕj
ρj ◦ σϕj
(σϕj
e−t2
t )ω(pϕ)dti
hρ, (σϕj
i/2(zk,j))ωi =h(
→h(
1
√πZ ∞
√πZ ∞
−∞
−∞
1
e−(t−i/2)2
e−(t−i/2)2
ρj ◦ σϕj
ρj ◦ σϕj
t ( · )dt)ω, (yk,j)ωi
t ( · )dt)ω, pϕi = hρ, pϕi.
This concludes the proof.
(cid:3)
Lemma 4.14. Let ϕ be a normal, semi-finite, faithful weight on a von Neumann algebra N .
Let x ∈ nϕ. Let a ∈ N be analytic for σϕ. Then xa ∈ nϕ and kxak2,ϕ ≤ kσϕ
Proof. That xa ∈ nϕ follows from [Tak00, Lemma VIII.2.4]. We have
i/2(a)kkxk2,ϕ.
hΛϕ(xa), Λϕ(xa)i = hJσϕ
=hJσϕ
i/2(a)σϕ
i/2(a)∗JΛϕ(x), Λϕ(x)i ≤ kσϕ
−i/2(a∗)JΛϕ(x), Jσϕ
i/2(a)k2kxk2
−i/2(a∗)JΛϕ(x)i
2,ϕ.
(cid:3)
Theorem 4.15. Let ϕj be normal, semi-finite, faithful weights on Mj with ultraproduct
ϕ = (ϕj)ω. Let ψj be normal, semi-finite, faithful weights on M′j with ultraproduct ψ =
(ψj)ω. Let pϕ be the support of ϕ and let rψ be the support of ψ. Put pψ = JωrψJω and
qψ = pψrψ. Assume that both ϕ and ψ are semi-finite and pϕ ≤ pψ. Then we have,
qψ(cid:18) dϕj
dψj(cid:19)ω
qψ =
dϕ
dψ
.
dψj(cid:17)ω
Furthermore qψ and(cid:16) dϕj
commute.
Proof. Let A be the set of all vectors a = (aj)ω such that kajk2,ϕj is bounded. Let B be the
set of all vectors b = (bj)ω such that kbjk2,ψj is bounded. Firstly we claim that the linear
dψ(cid:17) 1
ψ forms a core for(cid:16) dϕ
2 . Indeed by Lemma A.5 it suffices
span of the vectors pψA∗pψBpψD
to show that every element y ∈ pψ n∗ϕpψ nψpψ, say y = pψy∗1pψy2pψ with y1 ∈ nϕ, y2 ∈ nψ,
dψ(cid:17) 1
ψ in the graph norm of (cid:16) dϕ
2 .
may be approximated by elements in span pψA∗pψBpψD
1
2
1
2
WEIGHT THEORY FOR ULTRAPRODUCTS
21
Now it follows from Lemma 4.12 that we may find nets {ak}k in A and {bl}l in B such that
pψakpψ and pψblpψ are bounded, such that a∗k → y∗1 and bk → y2 strongly and such that
ϕ pψa∗kpψ → D
D
ϕ pψy∗1pψ and pψblpψD
ψ → pψy2pψD
ψ in norm. Then,
1
2
1
2
1
2
1
2
1
2
1
2
kpψa∗kpψblpψD
≤kpψa∗kpψblpψD
≤kpψa∗kpψkkpψblpψD
ψ − pψy∗1pψy2pψD
ψ − pψa∗kpψy2pψD
ψ − pψy2pψD
1
2
1
2
1
2
ψk2
ψk2 + kpψa∗kpψy2pψD
ψk2 + kpψa∗kpψy2pψD
1
2
1
2
1
2
ψ − pψy∗1pψy2pψD
ψ − pψy∗1pψy2pψD
1
2
ψk2
ψk2,
1
2
which goes to 0. Similarly kD
claim, see Theorem A.3.
1
2
1
2
ϕ pψa∗kpψblpψ − D
ϕ pψy∗1pψy2pψk2 → 0. This concludes the
Now fix a = (aj)ω ∈ A and b = (bj)ω ∈ B. Let zk = (zk,j)ω be a bounded net in M
converging to pψ strongly. Assume moreover that each zk,j is analytic for σϕj and that
i/2(z∗k,j) is bounded uniformly in j and that (σϕj
σϕj
i/2(z∗k,j))ω → pψ strongly, see Lemma 4.13.
This ensures that (ajz∗k,j)ω ∈ A by Lemma 4.14. Set xk,l = zka∗zlb so that xk,l = (xk,l,j)ω =
(zk,ja∗j zl,jbj)ω. We get by Theorem A.3, Lemma 4.10 and again Theorem A.3,
(4.9)
2
1
2
1
2
1
2
=(D
ψ = D
pψxk,lpψD
(cid:18) dϕ
dψ(cid:19) 1
ϕj xk,l,j)ωpψ = (cid:18) dϕj
dψj(cid:19) 1
2 ≥ h(cid:18) dϕj
dψj(cid:19)ω
ψk2
(xk,l,jD
1
2
ψj
, ρ
1
2
2
ϕ pψxk,lpψ = D
ϕ xk,lpψ
1
2
xk,l,jD
1
2
pψ.
ψj!ω
dψj(cid:19)ω
)ωi = h(cid:18) dϕj
So that, by Lemma 4.3 and Lemma 4.10,
2
1
2
1
2
1
2
1
2
, ρ
1
2
ψj
xk,lD
ψ i.
)ω,(xk,l,jD
1
2
ψ ,xk,lD
ψ and D
pψxk,lpψD
Taking the limit in k, l gives xk,lpψD
k(cid:18) dϕ
dψ(cid:19) 1
dψj(cid:17)ω
in norm. This shows that (using closedness of the quadratic form associated with(cid:16) dϕj
ψ → pψa∗pψbpψD
dψ(cid:19) 1
k,l k(cid:18) dϕ
dψj(cid:19)ω
ψ i = h(cid:18) dϕj
dψj(cid:17)ω(cid:17). Therefore, from (4.9) and Lemma
in pψHpψ = qψ(H). We see that qψ ≤ P<∞(cid:16)(cid:16) dϕj
k(cid:18) dϕ
dψ(cid:19) 1
dψj(cid:19)ω
k,l h(cid:18) dϕj
By Lemma 4.12 the span of the set of vectors pψa∗pψbpψD
ψ with a ∈ A and b ∈ B is dense
ϕ pψa∗pψbpψ
),
ϕ pψxk,lpψ → D
1
2
ψ ,pψa∗pψbpψD
pψa∗pψbpψD
pψxk,lpψD
1
2
ψ ,xk,lD
≥ lim
2 = lim
pψa∗pψbpψD
ψk2
ψk2
ψ i.
xk,lD
, ρ
, ρ
1
2
1
2
1
2
1
2
2
1
2
2
2
1
2
22
4.3,
MARTIJN CASPERS
pψxk,lpψD
2
(cid:18) dϕ
dψ(cid:19) 1
=pψ (cid:18) dϕj
dψj(cid:19) 1
2
pψxk,lpψD
2
1
2
ψ = pψ(cid:18) dϕ
dψ(cid:19) 1
ψj!ω
dψj(cid:19) 1
pψ = qψ(cid:18) dϕj
1
2
2
ω
1
2
ψ pψ
(xk,l,jD
1
2
ψj
)ω.
xk,l,jD
and taking limits in k, l gives (using closedness of the operators),
(4.10)
2
1
2
2
1
2
1
2
2
ω
2
ω
2
ω
2
ω
1
2
pψa∗pψbpψD
(pψa∗pψbpψD
We proved in the first paragraph that the span of the vectors pψa∗pψbpψD
(cid:18) dϕ
dψj(cid:19) 1
dψj(cid:19) 1
dψ(cid:19) 1
ψ == qψ(cid:18) dϕj
ψ ) = qψ(cid:18) dϕj
dψj(cid:17) 1
dψ(cid:17) 1
dψ(cid:17) 1
and b ∈ B forms a core for(cid:16) dϕ
2 ⊆ qψ(cid:16) dϕj
2 so that we get from (4.10) that (cid:16) dϕ
dψ(cid:19) 1
(cid:18) dϕ
and as both sides are self-adjoint we actually have
= qψ(cid:18) dϕj
dψj(cid:19) 1
qψ(pψa∗pψbpψD
ψ with a ∈ A
qψ
ψ ).
qψ.
the third equality, and using that we saw that qψ ≤ P<∞,
. In order to do so take
a bounded net zk = (zk,j) such that zk → pψ strongly and such that each zk,j is analytic
for σϕj with (σ
−i/2(zk,j))ω bounded and strongly convergent to pψ, c.f. Lemma 4.13. Set
It remains to prove that qψ(H) is an invariant subspace of(cid:16) dϕj
dψj(cid:17)ω
dψj(cid:17)ω(cid:17). Then, using closedness of the operators involved, Lemma 4.3 for
P<∞ = P<∞(cid:16)(cid:16) dϕj
(cid:18) dϕj
dψj(cid:19) 1
k,l,m(cid:18) dϕj
dψj(cid:19) 1
=P<∞(σϕj
=pψ(D
ϕj zk,ja∗j zl,jbjzm,j(cid:19)ω
ϕj a∗j )ωzl(bj)ωzm = P<∞pψ(D
dψj(cid:19) 1
k,l,m(cid:18) dϕj
−i/2(zk,j))ω(D
ϕj a∗j )pψ(bj)ωpψ ∈ qψ(H).
P<∞(cid:18)(D
(zk,ja∗j zl,jbjzm,jD
1
zka∗zlbzmD
2
ψ
ϕj a∗j )ωpψ(bj)ωpψ
pψa∗pψbpψD
)ω = lim
k,l,m
ψ = lim
= lim
1
2
ψj
ϕj
1
2
1
2
1
2
1
2
2
ω
2
ω
2
ω
1
2
So we conclude that(cid:16) dϕj
dψj(cid:17) 1
ω
2
preserves qψ(H).
(cid:3)
4.5. Modular theory. In this section we show that Theorem 4.15 implies results on mod-
ular theory and Radon-Nikodym derivatives. We first prove Theorem A of the introduction.
Theorem 4.16. Let ϕj be normal, semi-finite, faithful weights on Mj. Let ϕ = (ϕj)ω be
the ultraproduct weight on M with support pϕ. Assume that ϕ is semi-finite. Then,
t (pϕ(xj)ωpϕ) = pϕ(σϕj
σϕ
t (xj))ωpϕ.
WEIGHT THEORY FOR ULTRAPRODUCTS
23
Proof. Let ψj be the opposite weight of ϕj (see [Tak00]) and let ψ = (ψj)ω be its ultraprod-
uct. Letting again rψj be the support of ψj and pψj = Jjrψj Jj. We have pϕj = pψj and as
the support of ψ equals JωpϕJω we also have pϕ = pψ. Hence we may apply Theorem 4.15
to find that,
(4.11)
dϕ
dψ
= qψ(cid:18) dϕj
dψj(cid:19)ω
qψ.
dψ acting on qψ(H) is 0. Therefore (4.11) shows that the
to the invariant subspace qψ(H) has kernel equal to 0 and qψ is disjoint
(see (2.1)). This shows that
)H. Applying Theorem 4.15 and Theorem 4.2 and
commute then shows,
Now since pϕ = pψ the kernel of dϕ
qψ(H) is contained in ∪λ>0χ[ 1
restriction of(cid:16) dϕj
dψj(cid:17)ω
from the infinite projection in the spectral resolution of(cid:16) dϕj
dψj(cid:17)ω
λ ,λ]((cid:16) dϕj
dψj(cid:17)ω
dψj(cid:17)it(cid:19)ω
noting in addition that qψ and(cid:18)(cid:16) dϕj
qψ(cid:19)it
=(cid:18)qψ(cid:18) dϕj
dψj(cid:19)ω
qψ(xj)ωqψ(cid:18) dϕ
dψ(cid:19)−it
dψj(cid:19)−it!ω
qψ = qψ (cid:18) dϕj
dψ(cid:19)it
(cid:18) dϕ
dψ(cid:19)it
t (qψ(xj)ωqψ) =(cid:18) dϕ
=qψ (cid:18) dϕj
qψ(xj)ωqϕ (cid:18) dϕj
dψj(cid:19)it!ω
=qψ (cid:18) dϕj
dψj(cid:19)−it!ω
xj(cid:18) dϕj
dψj(cid:19)it
= qψ (cid:18)dϕj
Then we apply (2.3) twice to get,
qψ = qψ(σϕj
t (xj))ωqψ.
σϕ
dψj(cid:19)it!ω
qψ.
dψj(cid:19)it!ω
(xj)ω (cid:18) dϕj
dψj(cid:19)−it!ω
qψ
This also yields that σϕ
t (xj))ωpϕ as pϕMpϕ → qϕMqϕ : x 7→ qϕxqϕ is
a (non-spatial) isomorphism (see [AnHa14, Lemma 3.26] or [Haa75, Corollary 2.5, Lemma
2.6]).
(cid:3)
t (pϕ(xj)ωpϕ) = pϕ(σϕj
We refer to [Tak00, Section VIII.3] for the definition of the cocycle derivative.
Theorem 4.17. Let ϕj and ρj be normal, semi-finite, faithful weights on Mj. Let ϕ =
(ϕj)ω and ρ = (ρj)ω be their ultraproduct weights on M. Assume that ρ and ϕ are semi-
finite with equal support p. Then,
(4.12)
p(cid:18)(cid:18) Dϕj
Dψj(cid:19)t(cid:19)ω
=(cid:18)(cid:18) Dϕj
Dψj(cid:19)t(cid:19)ω
p =(cid:18) Dϕ
Dψ(cid:19)t
.
Proof. Let ψ = (ψj )ω be an ultraproduct of normal, semi-finite, faithful weights ψj on Mj
such that ψ is semi-finite with support r = JωpJω. Set q = pr. Then by [Tak00, Theorem
IX.3.8], Theorem 4.15, Theorem 4.2 (which is applicable, c.f. the proof of Theorem 4.16)
24
MARTIJN CASPERS
and once more [Tak00, Theorem IX.3.8], we see that,
(cid:18) Dϕ
Dρ(cid:19)t
dψj(cid:19)it
=q(cid:18) dϕj
ω
q
dψ(cid:19)it(cid:18)dψ
=(cid:18) dϕ
dρ(cid:19)it
= q (cid:18) dϕj
dρj(cid:19)it
ω(cid:18) dψj
(cid:18) Dϕ
Dρ(cid:19)t
q(cid:18) dψj
= q(cid:18) dϕj
dψj(cid:19)it
dρj(cid:19)it
dρj(cid:19)it!ω
= q(cid:18)(cid:18)Dϕj
dψj(cid:19)it(cid:18)dψj
Dρj(cid:19)t(cid:19)ω
=(cid:18)(cid:18) Dϕj
q.
ω
ω
Dρj(cid:19)t(cid:19)ω
.
Adapting the second line of this equation in the obvious way also gives
Under the isomorphism pMp ≃ qMq : x 7→ qxq (c.f.
Corollary 2.5, Lemma 2.6]), this yields (4.12).
[AnHa14, Lemma 3.26] or [Haa75,
(cid:3)
4.6. Related results. We end this section with a discussion about the following question.
can a weight on M always be decomposed in this way? We answer this question in the
negative.
If ρ ∈ M∗ then it can always be written in the form ρ = (ρj)ω as M∗ ≃Qj,ω(Mj)∗. But
Proposition 4.18. There exists an ultraproduct M =Qj,ω Mj with normal, semi-finite,
faithful weight ϕ on M such that ϕ cannot be written as the ultraproduct of normal, semi-
finite, faithful weights on each of the Mj's.
Proof. Suppose that there exists x ∈ M+ that is not of the form (xj)ω, i.e.
it is not
contained in the Banach space ultraproduct of the Mj's (see [AnHa14, Remark 3.6] for an
example). Suppose that ϕ is a normal, semi-finite, faithful weight on M such that x ∈ m+
ϕ
(note that such a weight always exists), then ϕ cannot be written as an ultraproduct (ϕj)ω of
∈ M+
normal, semi-finite and faithful weights ϕj on Mj. Indeed as x ∈ m+
∗
which we write as ϕ(−1/2)
ϕj . Let Gλ be a continuous function
such that 0 ≤ Gλ ≤ 1 with Gλ(s) = 1 for s ∈ [0, λ] and Gλ(s) = 0 for s ∈ [2λ,∞). Because
by Lemma 4.11 we have ϕ(−1/2)
and ϕ is faithful it
Gλ(xj)ω
follows that Gλ(xj)ω ≤ x and for λ > kxk we get equality (Gλ(xj))ω = x. This contradicts
that x cannot be written as an element (xj)ω.
)ω with xj ∈ m+
j,Gλ(xj ))ω ≤ (ϕ(−1/2)
ϕ we have ϕ(−1/2)
)ω = ϕ(−1/2)
= (ϕ(−1/2)
= (ϕ(−1/2)
j,xj
(cid:3)
x
j,xj
x
x
The following proposition shows that sometimes ultraproduct weights are faithful. Note
that we do not claim that τ in the next proposition is semi-finite, in fact this is not true as
for suitable ultrafilters it would contradict [Ray02, Proposition 1.14].
Proposition 4.19. Let Mj be type I von Neumann algebras. Let τj : Mj → [0,∞] be the
trace that is determined by τj(p) = 1 for every atomic projection p ∈ Mj. Then τ := (τj)ω
is faithful onQj,ω Mj.
Proof. We first claim that for elementary elements 0 ≤ x := (xj)ω ∈ Qj,ω Mj we have
τ (x) = 0 implies x = 0. Indeed, we may assume that xj ≥ 0. If τ (x) = 0 then this implies
WEIGHT THEORY FOR ULTRAPRODUCTS
25
that for every series of atomic projections (pj)ω we have h(pjτjpj)ω, (xj )ωi = 0. But this
implies that limj,ω pjxjpj = 0. So that (pj)ω(xj)ω(pj)ω = 0. Let λj be the largest eigenvalue
of xj and let p′j be the corresponding spectral projection. Let pj be an atomic subprojection
of p′j. Then 0 = (pj)ω(xj)ω(pj)ω = (pjxjpj)ω = (λjpj)ω. So limj,ω λj = 0 and therefore
(xj)ω = 0.
(pj)ω be a series of atomic projections such that x(pj)ω is non-zero. Then also (pj)ωx(pj)ω is
Now let 0 ≤ x ∈ Qj,ω Mj be arbitrary such that τ (x) = 0. Suppose that x 6= 0. Let
non-zero and moreover this element is contained in (pj)ω(cid:16)Qj,ω Mj(cid:17) (pj)ω ≃Qj,ω pjMjpj,
so it falls in the case of the first paragraph. Then 0 = τ (x) ≥ τ ((pj)ωx(pj)ω) 6= 0, which is
a contradiction. Hence x = 0.
(cid:3)
Remark 4.20. In view of Theorem 5.2 it would be interesting to know if for every ul-
traproduct M = Qj,ω Mj we can find some normal, semi-finite, faithful weights ϕj on
Mj such that the ultraproduct (ϕj)ω is again normal, semi-finite and faithful. Taking into
account the previous propositions we do not expect that this holds.
5. Applications to noncommutative Lp-spaces
In this section we prove Theorem C of the introduction. We need the following version
of the Powers -- Størmer inequality. A proof can be found in [CPPR15, Lemma 2.2]. Note
that this lemma only holds for semi-finite von Neumann algebras. The general case can be
deduced using [Kos84a, Proposition 7 and Lemma B].
Proposition 5.1 (Powers -- Størmer inequality). Let N be an aribtrary (not necessarily semi-
finite) von Neumann algebra. We have for x, y ∈ L1(N )+ that
1
kx
p − y
1
pkp
p ≤ kx − yk1.
Let ψ be a weight on the commutant N ′ of a von Neumann algebra N which we assume
is in standard form. Recall that the opposite weight is defined as the weight ψop on N that
is given by ψop(x∗x) = ψ(z∗z) where z = Jx∗J.
Theorem 5.2. Let ψj be normal, semi-finite, faithful weights on M′j with ultraproduct
ψ = (ψj)ω which we assume to be semi-finite. Let rψ be the support of ψ and let pψ = JωrψJω
and qψ = pψrψ. We have an isomorphism,
that is determined by
pψYj,ω
Lp(Mj, ψj) pψ →≃ Lp(qψMqψ, ψ),
(cid:18) dρj
dψj(cid:19) 1
ω 7→(cid:18) dρ
dψ(cid:19) 1
,
p
p
where ρ = (ρj)ω with ρj ∈ (Mj)∗ uniformly bounded in j and such that pρ ≤ pψ.
26
MARTIJN CASPERS
Proof. Throughout this proof we shall refer to the following special types of elements.
(∗) Let Aj ∈ Lp(Mj)+ be a sequence bounded in j in the k · kp-norm and of the form
dψj(cid:19) 1
Aj =(cid:18) dρj
p
.
with moreover ρj ∈ (Mj)∗ bounded uniformly in j and such that ρ = (ρj)ω has
support pρ ≤ pψ. Set A = (Aj)ω.
jk1/p
is bounded uniformly in j.
We may view A as in (∗) as an element of Qj,ω Lp(Mj, ψj) because it follows that
kAjkp = kAp
which shows that qψ(H) is an invariant subspace for (Ap
unbounded operator, namely dρ
j )ω by Lemma 4.1. We may apply Theorem 4.15
j )ω acts as an
On the other hand we have Ap = (Ap
dψ . This shows that qψ(Aj)ωqψ ∈ Lp(qψMqψ, ψ) with
j )ω on which (Ap
1
ρj(1) = lim
xj
a
j,ω kAjkp
Lp(M,ψ) = ρ(1) = lim
j,ω
the latter is true by assumption, see (∗).
Proof of Claim 1. Firstly let us show that the elements A from (∗) are indeed contained
So the proof so far shows that A 7→ qψAqψ gives an isometry from the closed linear span
kqψ(Aj)ωqψkp
Lp(Mj ,ψj) = k(Aj)ωkpQj,ω Lp(Mj ,ψj).
inQj,ω Lp(Mj, ψj) of operators A as in (∗) to the space Lp(qψMqψ, ψ). We need to prove
that such operators A are dense in pψ(cid:16)Qj,ω Lp(Mj, ψj)(cid:17) pψ and that the range equals
Lp(qψMqψ, ψ).
Claim 1. We claim that the A's from (∗) are densely contained in pψ(cid:16)Qj,ω Lp(Mj, ψj )(cid:17) pψ.
in Qj,ω Lp(Mj, ψj). To do so it suffices to show that (Ap
j )ω is in pψQj,ω L1(Mj, ψj)pψ,
which is equivalent to showing that (ρj)ω is in pψ(cid:16)Qj,ω(Mj)∗(cid:17) pψ (as the correspondence
Qj,ω(Mj)∗ ≃Qj,ω L1(Mj) preserves the left and right module action of Qj,ω Mj ). But
density of the elements Ap with A as in (∗) in pψ(cid:16)Qj,ω L1(Mj, ψj)(cid:17) pψ. Let ϕ be the
opposite weight of ψ. We have pϕ = pψ for the supports. The set ϕ(−1/2)
with x ∈ pϕm+
ϕ pϕ
is dense in (pψMpψ)+
ϕ pϕ. Let A be the set of all series
(xj )ω ≤ ϕ(−1/2)
is bounded uniformly in j and ϕ(−1/2)
(xj)ω with xj ∈ m+
.
Recall also that (ϕ(−1/2)
by Lemma 4.11. We have pψ(xj)ωpψ ≤ pψxpψ and
in fact exactly as in the proof of Lemma 4.12 one can show that supa∈A pψapψ = pψxpψ.
Then also supa∈A ϕ(−1/2)
with a ∈ A
are dense in (pψMpψ)+
∗ ≃
pψ(cid:16)Qj,ω(Mj)+
, a =
∗(cid:17) pψ ≃ pψ(cid:16)Qj,ω L1(Mj, ψj)+(cid:17) pψ under which the functional ϕ(−1/2)
∗ . Now use that we have an isometric correspondence (pψMpψ)+
dψj (cid:17). As A = (Aj)ω clearly satisfies (∗) we conclude
∗ . So fix such an element x ∈ pϕm+
ϕj such that ϕ(−1/2)
j,xj
)ω = ϕ(−1/2)
(xj )ω
It remains to show density. By Powers -- Størmer, c.f. Proposition 5.1, it suffices to show
(aj)ω ∈ A corresponds to Ap
our claim.
. This shows that the functionals ϕ(−1/2)
j :=(cid:16) dϕj,aj
= ϕ(−1/2)
x
a
a
x
x
WEIGHT THEORY FOR ULTRAPRODUCTS
27
Claim 2. We claim that the qψAqψ's from (∗) are dense in Lp(qψMqψ, pψ).
Proof of Claim 2. The proof is essentially the same as in Claim 1. By the Powers --
Størmer inequality it suffices to show that (qψAqψ)p = qψApqψ with A as in (∗) is dense in
L1(qψMqψ, ψ)+. Then it suffices to show the density in (qψMqψ)+
∗ of all positive normal
functionals on qψMqψ that are of the form (ϕ(−1/2)
ϕj are such that both
xj and ϕ(−1/2)
are bounded uniformly in j. This is exactly what the previous paragraph
shows.
)ω where xj ∈ m+
j,xj
j,xj
In particular we give a new proof of the following result due to Raynaud, see [Ray02,
Theorem 3.1].
(cid:3)
Corollary 5.3. Let Mj be von Neumann algebras and let M = Qj,ω Mj. For every
p ∈ [1,∞] we have an isomorphism,
Lp(M) ≃Yj,ω
Lp(Mj).
Proof. For p = ∞ this is by definition so assume that p ∈ [1,∞). The proof follows by an
inductive limit argument if we could show that the isomorphisms of Theorem 5.2 agree on
intersection taking into account the isomorphism given by a change of weight.
dψ2(cid:17) 1
p 7→(cid:16) dρ
p with ρ ∈ N∗ having support pρ ≤ pψ1.
Let N be an arbitrary von Neumann algebra. If ψ1, ψ2 ∈ N +
∗ are such that pψ1 ≤ pψ2 then
we have Φψ1,ψ2 : Lp(pψ1N pψ1, ψ1) ⊆ Lp(pψ2N pψ2, ψ2) isometrically and the isomorphism is
dψ1(cid:17) 1
given by(cid:16) dρ
Going back to ultraproducts take ψ1 = (ψ1,j)ω ∈ M′∗ and ψ2 = (ψ2,j)ω ∈ M′∗ with sup-
ports rψ1 and rψ2 and opposite supports pψ1 and pψ2. We denote the supports of the ψk,j's
similarly. Assume that pψ1 ≤ pψ2. We may and will assume that for every j we have sup-
ports pψ1,j ≤ pψ2,j (indeed one may replace ψ1,j by rψ2,j ψ1,jrψ2,j and as rψ1 ≤ rψ2 ≤ (rψ2,j )ω
we have (rψ2,j ψ1,jrψ2,j )ω = (rψ2,j )ω(ψ1,j)ω(rψ2,j )ω = (ψ1,j)ω). Now the isomorphisms so far
give a commutative diagram (as is clear from their explicit prescriptions),
Thm5.2
Lp(qψ1Mqψ1, ψ1)
Φψ1 ,ψ2
pψ1(cid:16)Qj,ω Lp(pψ1,jMjpψ1,j , ψ1,j )(cid:17) pψ1
pψ2(cid:16)Qj,ω Lp(pψ2,jMjpψ2,j , ψ2,j )(cid:17) pψ2
Qj,ω Φψ1,j ,ψ2,j
Thm5.2
/ Lp(qψ2Mqψ2, ψ2).
So take an increasing net {ψk}k in M∗ with pψk → 1. Then taking the inductive limit of
the isomorphisms pψk(cid:16)Qj,ω Lp(Mj, ψk,j)(cid:17) pψk ≃ Lp(qψkMqψk , ψk) of Theorem 5.2 yields
the result.
(cid:3)
/
/
/
28
MARTIJN CASPERS
Appendix A. Spatial derivatives on the standard form
We compliment this paper by showing that the spatial derivative attains a nice represen-
tation on the standard form, at least if the standard form is represented as
(A.1)
where J : x 7→ x∗. See [Ter82] for a discussion on standard forms and Lp-spaces.
(N , L2(N ), J, L2(N )+),
We fix again notation, keeping the appendix self-contained. Let N be a von Neumann al-
gebra. Let κ′ be some normal, semi-finite, faithful weight on N ′ and set Lp(N ) = Lp(N , κ′).
We assume that N is represented in standard form, ie. (A.1). For a normal weight ϕ on N
we write again Dϕ for dϕ/dκ′. For a normal, semi-finite, faithful weight ϕ on N we set
Iϕ = {ω ∈ N∗ The map Λϕ(x) 7→ ω(x∗), x ∈ nϕ extends boundedly} .
By the Riesz theorem for ω ∈ Iϕ there exists a vector ξ(ω) ∈ Hϕ such that
hξ(ω), Λϕ(x)i = ω(x∗).
It follows (see also [Cas13]) that ξ(yϕ) = Λϕ(y), y ∈ (Tϕ)2 and as ϕy =σϕ
that
−i(y) ϕ it follows
ξ(ϕy) = Λϕ(σϕ
y ∈ (Tϕ)2.
1
2
1
2
−i(y)),
ρ ∈ L2(N ) with kxD
(A.2)
Fix a normal, semi-finite, faithful weight ψ on N ′ and let ψop be the opposite weight on
N . Recall that it is defined as ψop(x∗x) = ψ(y∗y), for x = Jψy∗Jψ with y ∈ N ′ so that
ρ k2 = ρ(x∗x). We shall also
x ∈ N . Recall that if x ∈ nρ then xD
write Dψ for Dψop (as ψ is a weight on N ′ there can be no confusion). Note that we have
ψ z, where z ∈ n∗ψop .
nψ = Jψ n∗ψop Jψ and a GNS-construction for ψ is given by Jz∗J 7→ D
For an element D ∈ L1(N )+ we recall that Tr(D) = ρ(1) where ρ is such that D = dρ
dψ . We
extend this integral/trace to L1(N ) by linearity. Now we have the following.
Lemma A.1. Let ξ ∈ L2(N ) and suppose that the mapping f : N∗ → C : ω 7→ hξ(ω), ξi is
bounded. Then there exists x ∈ nψop such that ξ = Λψop (x).
Proof. Because f is bounded there exists x∗ ∈ N such that f (ω) = ω(x∗). We claim that
x ∈ nψop. Let y ∈ (Tψop )2 and let aj ∈ Tψop be a bounded net converging to 1 in the σ-
strong topology and such that also σψop
i/2 (aj) is bounded and convergent to 1 in the σ-strong
topology, c.f. Lemma 2.1. Then xaj ∈ nψop and xaj → x in the σ-strong topology. Also we
have,
1
2
y ), Λψop (xaj)i = ψop(a∗j x∗σψop
−i (y)) = hξ(ψop
−i (ya∗j )), ξi = hJσψop
i/2 (aj)JΛψop (σψop
ya∗
j
), ξi
−i (y)), ξi
hξ(ψop
=hΛψop (σψop
=hJσψop
i/2 (aj)Jξ(ψop
y ), ξi.
WEIGHT THEORY FOR ULTRAPRODUCTS
29
So we see that Λψop (xaj) converges weakly to ξ as the vectors ξ(ψop
y ) are dense in L2(N )
(c.f. (A.2) and [Cas13, Lemma A.2]). As Λψop is σ-weak/weak closed this implies that
x ∈ nψop and moreover that Λψop (x) = ξ. This proves the claim.
(cid:3)
Lemma A.2. The set of ψ-bounded D(L2(N ), ψ) equals {xD
we have Rψ(xD
1
2
ψ ) = x.
1
2
ψ x ∈ nψop}. Furthermore
Proof. Write πr(z), z ∈ N for the right multipliciation mapping L2(N ) → L2(N ) : x 7→ xz.
In particular πr(z) ∈ N ′ and πr(z) = Jz∗J. Recall that Λψ : πr(z) 7→ D
ψ z with z ∈ n∗ψop is
ψ ∈ L2(N ) we have for z ∈ n∗ψop,
a GNS-construction for ψ. As for x ∈ nψop we have that xD
ψ z = Λψ(πr(z)) 7→ πr(z)xD
ψ = xD
Rψ(xD
ψ ) : D
ψ z.
1
2
1
2
1
2
1
2
1
2
1
2
In particular xD
ψ is ψ-bounded and Rψ(xD
ψ ) = x.
1
2
1
2
Conversely, let y ∈ (Tψop )2 and take ξ ∈ D(L2(N ), ψ) ⊆ L2(N ) and set x = Rψ(ξ). Then
we have,
hξ(yψop), ξi = Tr(ξ∗yD
=Tr(D
ψ ξσψ
1
2
−i/2(y∗)) = Tr(D
ψ σψ
1
2
=Tr(D
1
2
ψ xD
1
2
ψ ) = Tr(D
ψ y∗ξ) = Tr(σψ
1
2
−i/2(y∗)D
1
2
ψ ξ)
1
2
ψ Rψ(ξ)Λψ(πr(σψ
−i/2(y∗))))
−i/2(y∗)) = Tr(xDψy∗) = Tr(yDψx∗) = yψop(x∗).
This shows that yψop 7→ hξ(yψop), ξi extends boundedly to a map N∗ → C (using that
functionals yψop with y ∈ (Tψop)2 are dense in N∗). Then applying Lemma A.1 shows that
ξ = Λψop (x) for some x ∈ nψop and therefore ξ = xD
ψ .
(cid:3)
1
2
Theorem A.3. Let ϕ be a normal, semi-finite, faithful weight on N and let ψ be a normal,
semi-finite, faithful weight on N ′. Let Dom(G) = {xD
ψ x ∈ n∗ϕ ∩ nψop}. Consider the
mapping,
1
2
(A.3)
G : Dom(G) → L2(N ) : xD
ψ 7→ D
1
2
1
2
ϕ x.
by (A.3).
2 . In particular G is closed and positive self-adjoint on the domain defined
Then G =(cid:16) dϕ
dψ(cid:17) 1
Proof. The domain of(cid:16) dϕ
dψ(cid:17) 1
2 equals all ψ-bounded vectors for which hϕ, θψ(xD
ψ )i is
ψ with x ∈ nψop for
finite. So by Lemma A.2 these are exactly all vectors of the form xD
which ϕ(xx∗) is finite, i.e. x ∈ n∗ϕ. From Lemma A.2 we see that for x, y ∈ nψop we have
ψ , xD
1
2
1
2
1
2
30
MARTIJN CASPERS
1
2
1
2
1
2
1
2
(xD
2
ψ , xD
θψ(yD
ψ ) : D
ψ z 7→ yx∗D
dψ(cid:19) 1
h(cid:18) dϕ
=hϕ, θψ(xD
=hJ(x∗D
dψ(cid:17) 1
This implies that G ⊆(cid:16) dϕ
1
2
1
2
2 .
ψ , xD
ϕ ), J(x∗D
2
1
2
1
2
(xD
ψ z. Consequently for x ∈ n∗ϕ ∩ nψop we have,
dψ(cid:19) 1
ψ ),(cid:18) dϕ
ψ )i = ϕ(xx∗) = hx∗D
ϕ )i = hD
ϕi
ϕ xi = hG(xD
ψ ), G(xD
ϕ , x∗D
ϕ x, D
ψ )i
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
ψ )i.
Corollary A.4. Using the notation of Theorem A.3, for the Connes cocycle derivative we
have:
(cid:18) Dϕ
Dψop(cid:19)t
= Dit
ϕ D−it
ψ .
(cid:3)
Proof. For any normal semi-finite faithful weight ρ on N ′ we have (see [Tak00, Theorem
IX.3.8])
Dϕ (cid:19)t
(cid:18) Dψop
=(cid:18) dϕ
dρ(cid:19)it(cid:18) dψop
dρ (cid:19)−it
,
(cid:3)
so that the proposition follows from Theorem A.3.
Finally we need an approximation argument for one of our proofs.
Lemma A.5. Let ϕ be a normal, semi-finite, faithful weight on N and let ψ be a normal,
ψ with x ∈ span n∗ϕnψop is a core for(cid:16) dϕ
dψ(cid:17) 1
2 .
semi-finite, faithful weight on N ′. The set xD
Proof. Let {ej}j be a net in n∗ϕ converging to 1 strongly and such that also σϕ
converges to 1 strongly. Let x ∈ n∗ϕ ∩ nψop . Then ejx ∈ n∗ϕnψop . Then we have that,
−i/2(ej )
1
2
1
2
ejxD
ψ → xD
1
2
ψ ,
in norm. By Theorem A.3 we see that therefore the linear span of n∗ϕnψ is a core for
1
2
D
ϕ ejx = σϕ
1
2
−i/2(ej)D
ϕ x → D
1
2
ϕ x,
References
(cid:3)
in norm and
dψ(cid:17) 1
(cid:16) dϕ
2 .
[AnHa14] H. Ando, U. Haagerup, Ultraproducts of von Neumann algebras, J. Funct. Anal. 266 (2014),
6842 -- 6913.
[BeLo76] J. Bergh, J. Lofstrom, Interpolation spaces, Springer 1976.
[BCI16] R. Boutonnet, I. Chifan, A. Ioana, II1 factors with non-isomorphic ultrapowers, R. Boutonnet, I.
Chifan, A. Ioana.
[Cas13] M. Caspers, The Lp-Fourier transform on locally compact quantum group, J. Operator Theory 69
(2013), 161 -- 193.
WEIGHT THEORY FOR ULTRAPRODUCTS
31
[CPPR15] M. Caspers, J. Parcet, M. Perrin, E. Ricard, Noncommutative de Leeuw theorems, Forum of
Mathematics - Sigma 3 (2015), e21.
[CaSa15] M. Caspers, M. de la Salle, Schur and Fourier multipliers of an amenable group acting on non-
commutative Lp-spaces, Trans. Amer. Math. Soc. 367 (2015), no. 10, 6997 -- 7013.
[Con76] A. Connes, Classification of injective factors. Cases II1, II∞, IIIλ,
6= 1, Ann. of Math. (2) 104
(1976), no. 1, 73 -- 115.
[Con80] A. Connes, On the spatial theory of von Neumann algebras, J. Funct. Anal. 35 (1980), 153 -- 164.
[DaCa72] D. Dacunha-Castelle, J.-L. Krivine, Application des ultraproduits `a l´etude des espaces et alg`ebres
de Banach, Studia Math. 41 (1972), 315 -- 334.
[FHS13] I. Farah, B. Hart, D. Sherman, Model theory of operator algebras I: stability, Bull. Lond. Math.
Soc. 45 (2013), 825 -- 838.
[GoLi99] S. Goldstein, J.M. Lindsay, Markov semigroups KMS-symmetric for a weight, Math. Ann. 313
(1999), no. 1, 39 -- 67.
[Gon16] A. Gonz´alez-P´erez, PhD thesis, ICMAT Madrid 2016.
[Gro84] U. Groh, Uniform ergodic theorems for identity preserving Schwarz maps on W∗-algebras, J. Oper-
ator Theory. 11 no. 2 (1984), 395 -- 404.
[Haa77] U. Haagerup, Lp-spaces associated with an arbitrary von Neumann algebra, Alg`ebres d'op´erateurs
et leurs applications en physique math´ematique, Proc. Colloq., Marseille 1977, 175 -- 184.
[Haa87] U. Haagerup, Connes' bicentralizer problem and uniqueness of the injective factor of type III1, Acta
Math. 158 (1987), no. 1 -- 2, 95 -- 148.
[Haa75] U. Haagerup, The standard form of von Neumann algebras, Math. Scand. 37 (1975), no. 2, 271 -- 283.
[Hil81] M. Hilsum, Les espaces Lp d'une alg`ebre de von Neumann d´efinies par la deriv´ee spatiale, J. Funct.
Anal. 40 (1981), 151 -- 169.
[Hou15] C. Houdayer, Y. Isono, Free independence in ultraproduct von Neumann algebras and applications,
J. Lond. Math. Soc. (2) 92 (2015), no. 1, 163 -- 177.
[Izu97] H. Izumi, Constructions of non-commutative Lp-spaces with a complex parameter arising from mod-
ular actions, Internat. J. Math. 8 (1997), 1029 -- 1066.
[Jun05] M. Junge, Embedding of the operator space OH and the logarithmic 'little Grothendieck inequality',
Invent. Math. 161 (2005), no. 2, 225 -- 286.
[Kos84a] H. Kosaki, On the continuity of the map φ 7→ φ from the predual of a W ∗-algebra, J. Funct. Anal.
59 (1984), no. 1, 123 -- 131.
[Kos84b] H. Kosaki, Applications of the complex interpolation method to a von Neumann algebra: noncom-
mutative Lp-spaces, J. Funct. Anal. 56 (1984), no. 1, 29 -- 78.
[NeRi11] S. Neuwirth, E. Ricard, Transfer of Fourier multipliers into Schur multipliers and sumsets in a
discrete group, Canad. J. Math. 63 (2011), no. 5, 1161 -- 1187.
[Pop14] S. Popa, Independence properties in subalgebras of ultraproduct II1 factors, J. Funct. Anal. 266
(2014), no. 9, 5818 -- 5846.
[Ray02] Y. Raynaud, On ultrapowers of non commutative Lp spaces, J. Operator Theory 48 (2002), no. 1,
41 -- 68.
[Ric15] E. Ricard, Lp-multipliers on quantum tori, arXiv: 1512.01142.
[Tak79] M. Takesaki, Theory of operator algebras I, Springer 1979.
[Tak00] M. Takesaki, Theory of operator algebras II, Springer 2000.
[Tak03] M. Takesaki, Theory of operator algebras III, Springer 2003.
[Ter81] M. Terp, Lp spaces associated with von Neumann algebras, Notes, Report No. 3a + 3b, Københavns
Universitets Matematiske Institut, 1981.
32
MARTIJN CASPERS
[Ter82] M. Terp, Interpolation spaces between a von Neumann algebra and its predual, J. Operator Theory
8 (1982), 327 -- 360.
M. Caspers, Universiteit Utrecht, Budapestlaan 6, 3584 CD Utrecht, The Netherlands
E-mail address: [email protected]
|
1002.2276 | 1 | 1002 | 2010-02-11T13:58:05 | Noncommutative topological entropy of endomorphisms of Cuntz algebras II | [
"math.OA"
] | A study of noncommutative topological entropy of gauge invariant endomorphisms of Cuntz algebras began in our earlier work with Joachim Zacharias is continued and extended to endomorphisms which are not necessarily of permutation type. In particular it is shown that if H is an N-dimensional Hilbert space, V is an irreducible multiplicative unitary on the tensor product of H with itself and F is the tensor flip, then the Voiculescu entropy of the Longo's canonical endomorphism associated with the unitary VF is equal to log N. | math.OA | math |
NONCOMMUTATIVE TOPOLOGICAL ENTROPY OF
ENDOMORPHISMS OF CUNTZ ALGEBRAS II
ADAM SKALSKI
Abstract. A study of noncommutative topological entropy of gauge invariant
endomorphisms of Cuntz algebras began in our earlier work with J. Zacharias
is continued and extended to endomorphisms which are not necessarily of per-
mutation type. In particular it is shown that if H is an N -dimensional Hilbert
space, V is an irreducible multiplicative unitary on H⊗H and F : H⊗H → H⊗H
is the tensor flip, then the Voiculescu entropy of the Longo's canonical endo-
morphism ρV F ∈ End(ON ) is equal to log N .
Noncommutative topological entropy for transformations of C ∗-algebras intro-
duced by D. Voiculescu in [Voi] is an interesting invariant generalising the classical
topological entropy of continuous transformations of compact spaces ([NS]);
its
value for an endomorphism ρ will be denoted by ht(ρ). The most studied from the
point of view of the Voiculescu entropy class of transformations is the family of var-
ious noncommutative generalisations of classical shifts, in particular the canonical
shift on the Cuntz algebra ON . In [SZ] together with J. Zacharias we computed the
Voiculescu entropy of certain permutation endomorphisms of ON generalising the
canonical shift.
The basic method of obtaining the lower estimates for the entropy ht(ρ) of a
given endomorphism ρ of a C ∗-algebra A is based on finding a commutative C ∗-
subalgebra left invariant by ρ on which ρ is induced by a classical transformation
whose entropy can be computed by an application of standard dynamical systems'
techniques. Having said that, the examples of bitstream shifts studied in [NS] show
that it can happen that
ht(ρ) > htc(ρ) = sup{ht(ρC) : C is a commutative ρ-invariant C ∗-subalgebra of A}
(see [Ska]). Each permutation endomorphism of ON leaves the so-called canonical
masa CN ⊂ ON invariant. Although the analysis in [SZ] showed that there exist
examples when ht(ρ) > ht(ρCN ), in each case we still had ht(ρ) = htc(ρ) and the
actual value of entropy was achieved already on some ρ-invariant standard masa
(i.e. a masa arising from CN by a change of coordinates). Moreover the only values
of entropy obtained explicitly for endomorphisms of ON were equal to log k, where
k ∈ N.
In view of the above discussion it is interesting to investigate on one hand the
existence of invariant standard masas for endomorphisms of a Cuntz algebra and on
the other to seek new ways of establishing lower bounds for Voiculescu entropy for
such endomorphisms. The first problem was studied by J.H. Hong, W. Szyma´nski
and the author in the recent paper [HSS]. The current work is devoted to making
Permanent address of AS: Faculty of Mathematics and Computer Science, University of L´od´z,
ul. Banacha 22, 90-238 L´od´z, Poland.
2000 Mathematics Subject Classification. Primary 46L55, Secondary 37B40.
Key words and phrases. Noncommutative topological entropy, Cuntz algebra, polynomial
endomorphisms.
1
progress in the second. The way forward was suggested by connections between the
values of index of permutation endomorphisms and their entropy discussed in [CS2]
and resembling the connections between the index of a subfactor of a finite factor
and the Connes-Størmer entropy of a canonical shift of Ocneanu (see for example
[Hia]).
We begin the note by introducing the notations and recalling basic statements
needed in what follows. In Section 2 we analyse a class of examples of endomor-
phisms of ON introduced by M. Izumi in [Izu]. Each endomorphism ρ in this class,
although not a permutation endomorphism itself, is at the same time a square
root of a permutation endomorphism and a composition of a permutation endo-
morphism with a Bogolyubov automorphism β. Moreover ht(ρ) = 1
2 log N and
ht(ρ ◦ β) = log N (note that in [HSS] it was shown that when N = 2 the corre-
sponding endomorphism leaves no standard masa invariant). In Section 3 we prove
that the computation of the entropy of the Izumi's examples can be interpreted as
a special instance of a general fact: each irreducible multiplicative unitary V on
a finite-dimensional Hilbert space H ([BaS]) leads to an endomorphism of a corre-
sponding Cuntz algebra whose Voiculescu entropy is equal to the logarithm of the
dimension of H.
1. Notations and preliminaries
Let N in N and let ON denote the Cuntz algebra, i.e. the C ∗-algebra gener-
ated by N isometries with orthogonal ranges summing to 1 ([Cu1]). The isome-
tries generating ON will be usually denoted by S1, . . . , SN ; composition of several
generating isometries will be expressed via a multi-index type notation (see for
example [SZ]). For each k ∈ N we denote the set of multi-indices of length k by
well-known 1 − 1 correspondence between the unital endomorphisms of ON (de-
noted further by End(ON )) and unitaries in ON , first observed in [Cu2]: given
U ∈ U(ON ) the associated endomorphism is determined by
Jk := {(j1, . . . , jk) : j1, . . . , jk ∈ {1, . . . , N}} and put J = Sk∈N Jk. There is a
ρU (Si) = U Si,
i = 1, . . . , N,
and conversely given ρ ∈ End(ON ) its associated unitary is given by
Uρ =
ρ(Si)S ∗
i .
nXi=1
If a unitary U ∈ ON is a linear combination of elements of the form SiS ∗
j (i, j =
1, . . . , N ), the associated transformation ρU is an automorphism, called a Bo-
golyubov automorphism (its action corresponds to the change of coordinates in the
Hilbert space CN underlying the Cuntz algebra in the approach due to R. Longo
and J. Roberts). In particular the family of automorphisms {ρt : t ∈ T} provides
an action of T on ON , called the gauge action (for Bogolyubov automorphisms
we will usually write αU instead of ρU ). The fixed point space of this action is
denoted by FN and is generated as a normed space by the unionSk∈N F k
N , where
F k
N = Lin{SJ S ∗
N is isomorphic to
MN k ≈ M ⊗k
are compatible with
the usual unital embeddings on the matrix level, so that FN is a UHF algebra. The
canonical masa (maximal abelian subalgebra) in ON is the algebra CN generated
N and the corresponding embeddings F k
It is easy to check that F k
K : J, K ∈ Jk}.
N ⊂ F k+1
N
2
by {sI s∗
some Bogolyubov automorphism α.
I : I ∈ J }; a standard masa is a C ∗-subalgebra of ON equal to α(CN ) for
An interesting class of permutation endomorphisms of ON having a relatively
simple combinatorial description was introduced in [Kaw] and was subsequently
studied for example in [Szy] and [CS1]. Let Pk denote the set of all permutations of
the set Jk. The permutation endomorphism given by σ ∈ Pk is the endomorphism
J . The canonical shift on ON is the
i S ∗
j
and denoted further by Φ. Every standard masa is isomorphic to the algebra of
continuous functions on C, the full shift on N letters ([Wal]).
associated to a unitary Uσ = PJ ∈Jk Sσ(J)S ∗
permutation endomorphism associated with the flip unitary F =PN
i,j=1 SiSjS ∗
The standard topological entropy of a continuous transformation T of a compact
space ([Wal]) will be denoted by htop(T ). For the definition and basic properties
of the Voiculescu's noncommutative topological entropy of an endomorphism (or a
completely positive map) of a (nuclear) C ∗-algebra we refer to the original paper
[Voi] or to the monograph [NS]. All the information needed to read this note can
be also found in [SZ].
2. Entropy of an endomorphism coming from a real sector -- the
square root of a canonical endomorphism
In [Izu] M. Izumi studied certain explicit examples of endomorphisms of Cuntz
algebras motivated by the subfactor theory. One class of them (Example 3.7 of [Izu])
was constructed in the following way: let G be a finite abelian group of cardinality
N ≥ 2 with the (symmetric) duality bracket h·,·i : G × G → T satisfying the usual
conditions (g, g′, h ∈ G)
hh, gi =(0
hg, hi = h−g, hi, Xh∈G
hg, hihg′, hi = hg + g′, hi, hg, hi = hh, gi
if g 6= e
N if g = e
,
(the group operation in G will be written additively). If G = Z/N Z one can put
hk, li := exp( 2πi(kl)
N ). We will use elements of G as indices of generating isometries
in ON . Define unitaries U (g) ∈ FN ⊂ ON (g ∈ G) by
hg, hiShS ∗
h,
and the endomorphism ρ ∈ End(ON ) by
ρ(Sg) =
hg, hiShU (g)∗.
U (g) =Xh∈G
√N Xh∈G
1
The endomorphism ρ was studied in detail in [HSS]. It does not leave CN invariant.
The unitary associated with ρ is equal to
(2.1)
V =
Define for each h ∈ G
hg, h − liShSlS ∗
l S ∗
g .
1
√N Xg,h,l∈G
eSh =
1
√N Xa∈G
3
hh, aiSa
h) =Xg∈G
h) =Xg∈G
ρ′(ShSkS ∗
kS ∗
SgSh+gSk+h+gS ∗
k+h+gS ∗
h+gS ∗
g ,
and let β ∈ Aut(ON ) be given by
β(Sh) = eSh, h ∈ G.
It is easy to see that β is a Bogolyubov automorphism.
Lemma 2.1. The endomorphism ρ′ := ρ ◦ β is a permutation endomorphism of
ON given by the formula (notations as above)
ρ′(Sh) =Xg∈G
SgSh+gSh+g
∗, h ∈ G.
Moreover ht(ρ′) = log N .
Proof. The first statement is shown in [HSS]. As the endomorphism ρ′ is induced
by a unitary in F 2
N , Theorem 2.2 of [SZ] implies that ht(ρ′) ≤ log N . As ρ′ is
a permutation endomorphism, it leaves CN invariant and ρ′CN is induced by a
continuous transformation Tρ′ of C. Note that we have (h, k ∈ G)
ρ′(ShS ∗
SgSh+gSh+g
∗S ∗
g ,
and so on. The analysis of the transformations on cylinder sets (see [SZ] or [Szy])
implies that Tρ′ is given by the formula
(Tρ′ (w))k = wk+1 − wk, w := (wn)∞
n=1 ∈ C, k ∈ N.
Hence an application of Lemma 3.2 of [SZ] yields that htop(Tρ′ ) = log N so that
ht(ρ′) ≥ ht(ρ′CN ) = htop(Tρ′ ) = log N .
(cid:3)
Theorem 2.2. The Voiculescu entropy of ρ is equal to 1
2 log N .
Proof. It is easily checked that ht(ρ) = 1
2 ht(ρ2). Write γ := ρ2. It suffices to show
that ht(γ) = log N . This will follow from the general result in Theorem 3.1, but
here we can provide a direct proof, as γ is a permutation endomorphism given by
the formula (g ∈ G)
γ(Sg) =Xk∈G
SkSg+kS ∗
k
(see [HSS]). As the associated unitary Vγ :=Pg,h∈G SgSh+gS ∗
Theorem 2.2 of [SZ] implies that ht(γ) ≤ log N .
Examine the action of γ on CN . For each n ∈ N and g1, . . . , gn ∈ G
gn+h ··· S ∗
ShSg1+h ··· Sgn+hS ∗
γ(Sg1 ··· Sgn S ∗
gn ··· S ∗
g S ∗
g1+hS ∗
h,
h belongs to F 2
N ,
g1 ) =Xh∈G
so that once again analysing the cylinder sets we see that γCN is induced by the
continuous transformation defined by
(Tγ(w))k = w1 + wk+1, w = (wn)∞
n=1 ∈ C, k ∈ N.
Thus appealing to Lemma 3.2 in [SZ] yields ht(γCN ) = htop(Tγ) = log N. This ends
the proof.
(cid:3)
4
Note that Lemma 2.1 and Theorem 2.2 yield an example of an endomorphism ρ
of ON and a Bogolyubov automorphism β such that ht(ρ ◦ β) 6= ht(ρ) (although,
as follows from [DS], the Voiculescu entropy of each Bogolyubov automorphism of
ON is equal 0).
Let G = Z/2Z = {0, 1} with the natural duality bracket (h1, 1i = −1, all other
brackets take value 1). The Izumi endomorphism discussed above is then given by
(2.2) ρ(S0) =
1
√2
(S0 + S1), ρ(S0) =
1
√2
(S0S0S ∗
0 + S1S1S ∗
1 − S1S0S ∗
0 − S0S1S ∗
1 ).
In Section 6 of [HSS] we showed that ρ defined via the formulas in (2.2) does
not leave any standard masa invariant. This naturally leads to the following closely
connected questions: can one characterise masas in ON left invariant by ρ? Do we
have ht(ρ) = htc(ρ)?
It would be very interesting to investigate the entropy of other examples of real
sectors given in [Izu]. This would require completely new methods even for obtain-
ing upper estimates, as Theorem 2.2 of [SZ] applies only to the endomorphisms
associated to unitaries in FN and other examples of Section 3 of [Izu] are not of
this type.
3. Entropy of canonical endomorphisms associated to multiplicative
unitaries
Consider again the endomorphism ρ associated to the unitary defined in (2.1) via
a symmetric duality bracket on a finite abelian group G discussed in the last section.
Let τ be the faithful trace on FN and let φ = τ ◦ E, where E : ON → FN is the
canonical conditional expectation (given by integrating the gauge action). It follows
from Lemma 2.1 of [Lo2] that the endomorphism ρ preserves φ, so also extends to an
endomorphism of M := πφ(ON )′′, where πφ denotes the GNS representation with
respect to φ (we will denote the extension by eρ). It follows from the easily checked
condition in Corollary 4.3 of [CP] that eρ is irreducible, i.e. eρ(M)′ ∩ M = CIM.
As discussed in Section 2, in [Izu] it is observed (as a consequence of results
in [Lo1] and [Wat]) that ρ is a restriction of a 'square root' of a canonical endo-
morphism of III 1
-factor generated by ON in the GNS representation with respect
to the state φ.
It follows from Proposition 2.5 in [Izu] that if we consider the
conditional expectation Eρ : ON → ρ(ON ) defined by
e ρ(x)Se),
Eρ(x) = ρ(S ∗
N
where e ∈ G is the neutral element, then Ind Eρ = N and it is a minimal index in
the sense of Kosaki ([Kos]). Moreover the endomorphism γ = ρ2 is related to the
left regular representation of the group G. Indeed, it is easy to check that
(3.1)
γ2 = Φ ◦ γ,
the flip unitary F =Pg,h∈G ShSgS ∗
which due to Proposition 2.1 of [Cu3] is equivalent to the fact that the unitary
associated with γ is a product of a multiplicative unitary on Cn ⊗ Cn ([BaS]) and
As suggested by the above discussion, there is a connection between the entropy
computation in Section 2, index values and the fact that ρ2 is related to the left
regular representation of a finite group. Indeed, Theorem 2.2 may be viewed as
a special instance of a general entropy result related to interaction between finite-
dimensional Kac algebras and index for subfactors stated in the following theorem.
hS ∗
g .
5
Theorem 3.1. Let V be an irreducible multiplicative unitary on H⊗ H, where H is
an N -dimensional Hilbert space; view it as a matrix in MN ⊗ MN and further via
the usual isomorphism MN ⊗ MN ≈ F 2
N ⊂ ON as a unitary in ON . Let F be the
flip unitary in F 2
N . The topological entropy of the endomorphism of ON associated
with V F is equal to log N .
Proof. Denote R = V F , γ := ρR. As R ∈ F 2
ht(γ) ≤ log N .
N , Theorem 2.2 of [SZ] implies that
As mentioned above, Lemma 2.1 of [Lo2] implies that γ preserves φ, so γ extends
. Due to Proposition 4.7 in [Lo2], M(0)
(Corollary 4.3, [Lo2]), where N is the fixed point algebra for the coaction associated
serving a state ω (see [NS] for the precise definitions -- although Theorem 3.2.2
(ii) is stated for the automorphisms, its proof is valid also when ρ is just an en-
domorphism). As by Theorem 6.2.2 (ii) in [NS] ht(γ) ≥ hφ(γ), to finish the proof
to an endomorphism of M := πφ(ON )′′, denoted by eγ. Irreducibility of V implies
that eγ is a canonical endomorphism for an irreducible inclusion of factors N ⊂ M
to V ([Lo2], [Cu3]). By Proposition 3.1 in [Lo2] Ind(eγ) = N 2.
N(cid:1)′′
Let M(0) := πφ(FN )′′ = πφ(cid:0)Sn∈N F n
is equal to the strong closure of the algebra Sn∈Neγn(M)′ ∩ M. As V ∈ FN , the
endomorphism eγ leaves M(0) invariant. By Theorem 3.2.2 (ii) in [NS] we have
hφ(eγ) = hφ(γ), where hω(α) denotes the CNT entropy of an endomorphism α pre-
it suffices to show that hφ(eγ) ≥ log N . Note that as the conditional expectation
E extends to a φ-preserving normal conditional expectation eE : M → M(0), by
Theorem 3.2.2 (v) in [NS] we have hφ(eγ) ≥ hφ(eγM(0) ).
Let N(0) := N ∩ M(0). Note that because both γ and canonical shift Φ commute
with the gauge action, we have N(0) = eE(N). As the canonical expectation Eγ :
M → eγ(M) discussed in Lemma 4.6 in [Lo2] preserves the trace on M(0), due to
of the same paper1 implies [M(0), N(0)] = (Ind(eγ))
eγM(0) is conjugate to the Ocneanu's canonical shift associated with the inclusion
N(0) ⊂ M(0); note that now we are in the framework of finite factor inclusions.
Corollary 4.6 of [Hia] gives then hφ(eγM(0) ) = log([M(0), N(0)]) = log N , provided
the inclusion N(0) ⊂ M(0) is extremal and strongly amenable. Extremality follows
from the equality (N(0))′∩M(0) = CIM, which itself is a consequence of Theorem 6.6
in [LoR] and Proposition 3.2 of [Lo2]. By Theorem 1 in [Pop] strong amenability
of the inclusion in question will follow if we can only show that it has finite depth,
as M(0) is a hyperfinite factor. As the Jones tunnel corresponding to the inclusion
N(0) ⊂ M(0) is given by
Lemma 7.3.5 in [Cho] we see that N(0) is a subfactor of M(0); Proposition 7.3.6
1
2 = N . Further we can use
the observation in Theorem 7.3.7 of [Cho] (see also Corollary 4.3 in [Lo2]) that
··· ⊂ γ(N(0)) ⊂ γ(M(0)) ⊂ N(0) ⊂ M(0) ⊂ ··· ,
it suffices to show that γ(N(0))′ ∩ M(0) is a factor. The earlier observation that
(N(0))′ ∩ M(0) = CIM and a suitably adapted argument from Corollary 3.3 of [Lo2]
delivers precisely that (one can show that γ(N(0))′ ∩ M(0) = F 1
N ).
(cid:3)
1Note that in Lemma 7.2.1 in [Cho] the operator V should be defined as V = 1
λ γ(e)f e and is
only a partial isometry -- this does not affect further reasoning and the main results of that paper
remain valid.
6
Conceptually the reason for which we get ht(ρV F ) = log N is that ρV F is indeed
a map closely related to the canonical shift on ON , as is suggested by the formula
(3.1); several instances of such analogies can be found in [LoR].
Acknowledgment. The work on this note was started during a visit of the author
to University of Tokyo in October-November 2009 funded by a JSPS Short Term
Postdoctoral Fellowship.
References
[BaS] S. Baaj and G. Skandalis, Unitaires multiplicatifs et dualit´e pour les produits crois´es de
C ∗-alg`ebres, Ann. Sci. ´Ecole Norm. Sup. (4), 26 (1993), no. 4, 425 -- 488.
[Cho] M. Choda, A C ∗-dynamical entropy and applications to canonical endomorphisms,
J. Funct. Anal. 173 (2000), no. 2, 453 -- 480.
[CP] R. Conti and C. Pinzari, Remarks on the index of endomorphisms of Cuntz algebras,
J. Funct. Anal. 142 (1996), no. 2, 369 -- 405.
[CS1] R. Conti and W. Szyma´nski, Labeled Trees and Localized Automorphisms of the Cuntz
Algebras, Transactions of AMS, to appear (available at arXiv:0805.4654).
[CS2] E. Conti and W. Szyma´nski, Computing the Jones index of quadratic permutation endo-
morphisms of O2, J. Math. Phys. 50 (2009), no. 1, 012705, 5 pp.
[Cu1] J. Cuntz, Simple C ∗-algebras generated by isometries. Comm. Math. Phys. 57 (1977), no.
2, 173 -- 185.
[Cu2] J. Cuntz, Automorphisms of certain simple C ∗-algebras, in 'Quantum fields -- algebras, pro-
cesses' (Proc. Sympos., Univ. Bielefeld, Bielefeld, 1978), pp. 187 -- 196, Springer, 1980.
[Cu3] J. Cuntz, Regular actions of Hopf algebras on the C ∗-algebra generated by a Hilbert space,
in "Operator algebras, mathematical physics, and low-dimensional topology (Istanbul,
1991)", 87 -- 100, Res. Notes Math., Wellesley, 1993.
[DS] K. Dykema and D. Shlyakhtenko, Exactness of Cuntz-Pimsner C ∗-algebras, Proc. Ed-
inb. Math. Soc. (2) 44 (2001), no. 2, 425 -- 444.
[Hia] F. Hiai, Entropy for canonical shifts and strong amenability, Internat. J. Math. 6 (1995),
no. 3, 381 -- 396.
[HSS] J.H. Hong, A. Skalski and W. Szyma´nski, On Invariant MASAs for Endomorphisms of the
Cuntz Algebras, preprint, arXiv:1001.1899.
[Izu] M. Izumi, Subalgebras of infinite C ∗-algebras with finite Watatani indices I. Cuntz algebras,
Comm. Math. Phys. 155 (1993), no. 1, 157 -- 182.
[Kaw] K. Kawamura, Polynomial endomorphisms of the Cuntz algebras arising from permutations.
I. General theory., Lett. Math. Phys. 71 (2005), no. 2, 149 -- 158.
[Kos] H. Kosaki, "Type III factors and index theory," Lecture Notes Series, 43, Seoul National
University, Seoul, 1998.
[Lo1] R. Longo, Index of subfactors and statistics of quantum fields. II. Correspondences, braid
group statistics and Jones polynomial, Comm. Math. Phys. 130 (1990), no. 2, 285 -- 309.
[Lo2] R. Longo, A duality for Hopf algebras and for subfactors, Comm. Math. Phys. 159 (1994),
no. 1, 133 -- 150.
[LoR] R. Longo and J.E. Roberts, A theory of dimension, K-Theory 11 (1997), no. 2, 103 -- 159.
[NS] S.Neshveyev and E.Størmer, "Dynamical entropy in operator algebras," Ergebnisse der
Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics,
50. Springer-Verlag, Berlin, 2006.
[Pop] S. Popa, Classification of amenable subfactors of type II, Acta Math. 172 (1994), no. 2,
163 -- 255.
[Ska] A. Skalski, On automorphisms of C ∗-algebras whose Voiculescu entropy is genuinely non-
commutative, Ergodic Th. Dynam. Systems, to appear (available at arXiv:0911.3951).
[SZ] A. Skalski and J. Zacharias, Noncommutative topological entropy of endomorphisms of
Cuntz algebras, Lett. Math. Phys. 86 (2008), no. 2-3, 115 -- 134.
[Szy] W. Szyma´nski, On localized automorphisms of the Cuntz algebras which preserve the di-
agonal subalgebra, in 'New Development of Operator Algebras', R.I.M.S. Kokyuroku 1587
(2008), 109 -- 115.
7
[Voi] D. Voiculescu, Dynamical approximation entropies and topological entropy in operator al-
gebras, Comm. Math. Phys. 170 (1995), no. 2, 249 -- 281.
[Wal] P. Walters, "An introduction to ergodic theory, Graduate Texts in Mathematics," 79.
Springer-Verlag, New York-Berlin, 1982.
[Wat] Y. Watatani, Index for C ∗-subalgebras, Mem. Amer. Math. Soc. 83 (1990), no. 424.
Department of Mathematics and Statistics, Lancaster University, Lancaster, LA1
4YF
E-mail address: [email protected]
8
|
1603.08758 | 3 | 1603 | 2016-05-04T16:45:53 | Mixtures of classical and free independence | [
"math.OA",
"math.PR"
] | We revive the concept of Lambda-freeness of Mlotkowski, which describes a mixture of classical and free independence between algebras of random variables. In particular, we give a description of this in terms of cumulants; this will be instrumental in the subsequent paper [SW] where the quantum symmetries underlying these mixtures of classical and free independences will be considered. | math.OA | math |
MIXTURES OF CLASSICAL AND FREE
INDEPENDENCE
ROLAND SPEICHER AND JANUSZ WYSOCZANSKI
Abstract. We revive the concept of Λ-freeness of M lotkowski
[10], which describes a mixture of classical and free independence
between algebras of random variables. In particular, we give a de-
scription of this in terms of cumulants; this will be instrumental
in the subsequent paper [14] where the quantum symmetries un-
derlying these mixtures of classical and free independences will be
considered.
1. Introduction
In the context of non-commutative probability spaces there are only
very few possibilities for universal notions of independence. If we re-
quire that this notion is commutative (i.e., x independent from y is
the same as y independent from x) and that constants are indepen-
dent from everything then there are only two such concepts, namely
the classical independence and the free independence. On the level of
algebras, equipped with a state, this means that there are only two uni-
versal kind of product constructions, namely the tensor product and
the reduced free product. We refer the reader to [13, 11, 1] for more
details on this.
So if we have a collection of variables which are independent (in this
univeral sense) then there are only two possibilities; they are either all
classically independent or they are all freely independent. On the other
hand, we can gain some more flexibility if we do not ask for the same
kind of independence between all of them. This raises the question
about mixtures of the two forms of independences. Of course, one can
create quite easily such situations by starting with two sets of variables
X and Y which are free; then split each of them into two subsets
X = X1 ∪ X2 and Y = Y1 ∪ Y2, such that X1 and X2 are classically
independent and Y1 and Y2 are freely independent. One can continue
in this fashion and get so a collection of variables where some pairs
of them are free and other pairs are classically independent. However,
R.S. is supported by the ERC-Advanced Grant "Non-commutative Distributions
in Free Probability".
1
2
ROLAND SPEICHER AND JANUSZ WYSOCZANSKI
this is restricted to situations where we can group our variables in sets
with specific kind of independence among them. We are interested in a
generalization of this, by trying to prescribe arbitrarily free or classical
independence for any pair. An example for this would be to ask for
five variables x1, x2, x3, x4, x5 such that
• x1 and x2 are free,
• x2 and x3 are free,
• x3 and x4 are free,
• x4 and x5 are free,
• x5 and x1 are free,
• but all other pairs are independent.
(Here and in the following we will always mean "classically indepen-
dent" when we say "independent".)
Such a situation cannot be generated by the above dividing into
groups, and it is not clear apriori whether such a requirement can
be satisfied in any meanigful way.
In [10] M lotkowski showed that
this can, indeed, be achieved for any prescription of the mixture of
free and classical independence. For this he introduced the general
notion of Λ-freeness. It seems that his work did not get the attention
it deserves and we hope that our work will stimulate new interest in
this concept. To his original results we will add here a description
of the combinatorial structure of Λ-freeness, featuring in particular a
formula for mixed moments in terms of free cumulants. This will be
taken up in the subsequent paper [14] and will lead to new forms of
quantum groups, with partial commutation relations.
On the level of groups or semi-groups the prescription of commuta-
tion relations for some fixed pairs of generators is of course not new;
in the group case this goes, among others, under the names of "right
angled Artin groups" (see [3]), "free partially commutative groups"
or "trace groups", in the case of semi-groups one talks about "Cartier-
Foata monoids" (see [5]) or "trace monoids". Actually, there is also the
notion of a corresponding mixed product of groups, which is usually
called the "graph product of groups" and was introduced by Green in
[6]. In a sense Λ-freeness reveals the notion of "independence" for the
group algebras of such graph products of groups with respect to their
canonical trace. We will make this connection precise in Proposition
4.2.
Our interest in Λ-freeness arouse out of discussions on similar con-
structions of the second author, on mixtures between monotone and
boolean [16] and boolean and free independences [8]. Much motivation
is also taken from recent work on bi-freeness [15, 9, 2]. Bifreeness does
MIXTURES OF CLASSICAL AND FREE INDEPENDENCE
3
not fit in the frame presented here, but there are some similarities, in
particular, concerning the underlying combinatorics.
2. The setting
The notion of Λ-freeness is defined in terms of a matrix which spec-
ifies the choice which pairs should be free and which should be in-
dependent. M lotkowski denoted this matrix by Λ; we prefer here to
call it ε, and hence we will also speak of ε-freeness or, alternatingly,
ε-independence.
So let I be an index set (finite or infinite). For any given collection of
algebras Ai, for all i ∈ I, we want to embed the Ai in a bigger algebra
A, such that for each pair of algebras we have that they are either free
or independent. In order to specify this choice we will use a symmetric
matrix ε = (εij)i,j∈I with non-diagonal entries either 0 or 1. This ε
should specify our mixture according to:
• Ai and Aj are free if εij = 0, and
• Ai and Aj are independent if εij = 1 (which includes in partic-
ular, that Ai and Aj commute
It will be convenient to set εii = 0 for all i ∈ I.
In the following such a matrix ε will be fixed. Of course, we can
identify such a matrix with the adjacency matrix of a simple (i.e.,
no loops, no multiple edges) graph; then the edges of the graph give
us the independence relations between the involved algebras, which
correspond to the vertices of the graph.
For the basic notions and facts about non-commutative probability
spaces, non-crossing partitions or free cumulants we refer to [12].
3. The definition of ε-independence
Notation 3.1. Let us use the following notation. Given some subal-
gebras Ai (i ∈ I) and an index-tuple i = (i(1), . . . , i(n)) ∈ I n we write
(a1, . . . , an) ∈ Ai for: ak ∈ Ai(k) for k = 1, . . . , n.
Definition 3.2. 1) By I ε
n we denote those n-tuples of indices from I for
which neigbours are different modulo our ε-relations; more precisely,
i = (i(1), . . . , i(n)) ∈ I ε
if we have i(k) = i(l) for
1 ≤ k < l ≤ n then there is a p with k < p < l such that i(p) 6= i(k)
and εi(k)i(p) = 0.
n if and only if:
2) Let (A, ϕ) be a non-commutative probability space. We say that
unital subalgebras Ai (i ∈ I) are ε-independent, if we have the follow-
ing.
• Ai and Aj commute for all (i, j) for which εij = 1 and
4
ROLAND SPEICHER AND JANUSZ WYSOCZANSKI
• whenever n ∈ N and (a1, . . . , an) ∈ Ai such that ϕ(ak) =
n, then we have
0 for all k = 1, . . . , n and such that i ∈ I ε
ϕ(a1 · · · an) = 0.
Note that we can use the usual centering trick to reduce any mixed
moment to mixed moments of the above form; hence if we know ϕ
restricted to each of the Ai and we know that the Ai are ε-independent,
then ϕ is uniquely determined on the algebra generated by all the Ai.
Namely, consider an arbitrary mixed moment of the form ϕ(a1 · · · an)
with (a1, . . . , an) ∈ Ai. We can also assume that i ∈ I n
ε (otherwise, by
using the commutation relations among the algebras, we bring elements
from the same algebra together and replace them by their product).
Then we write each ak as ak = ϕ(ak)1 + a◦
k. We plug this in for a1 · · · an
and multiply out. We get one term of length n, namely a◦
n plus
many other terms with fewer factors. By induction we can assume that
we already know how to calculate ϕ applied to those smaller terms, and
for the longest term we have ϕ(a◦
n) = 0, by our definition of
ε-independence.
2 · · · a◦
1a◦
1a◦
2 · · · a◦
It is also clear that if εij = 1 for all i 6= j, then ε-independence
is the same as classical independence; and if εij = 0 for all i, j, then
ε-independence is the same as free independence.
What might be not so clear from this definition is whether, given non-
commutative probability spaces (Ai, ϕi) for all i ∈ I, one can embed
them in a bigger non-commutative probability space (A, ϕ) such that
ϕ restricted to Ai yields ϕi and such that the (Ai)i∈I are ε-independent
in (A, ϕ). That this is indeed the case, for any choice of ε, as well as the
fact that positivity and traciality of the the involved linear functionals
is preserved under such a construction, was one of the main results of
[10].
4. ε-independence and the ε-products of groups
As we already mentioned in the Introduction, on the level of groups,
the notion of groups with partial commutation relations is a well-known
one. Actually, there is also the notion of an ε-product of groups, which
is usually called the graph product of groups (corresponding to the
graph with adjacency matrix ε) and was introduced by Green in [6],
see also [7].
Definition 4.1. Let Gi (i ∈ I) be groups. Then the ε-product (or the
graph product) ⋆εGi is the quotient of the free product group ⋆i∈I Gi by
the relations that Gi and Gj commute whenever εij = 1.
MIXTURES OF CLASSICAL AND FREE INDEPENDENCE
5
As expected, the notion of ε-independence is adapted to this setting
of an ε-product of groups.
Proposition 4.2. 1) Let G = ⋆εGi be the ε-product of subgroups Gi.
Denote by τ : CG → C the canonical state on the group algebra CG,
which gives the coefficient of the neutral element in a linear combination
of group elements. Then, the group algebras of the subgroups, CGi
(i ∈ I), are ε-independent in the non-commutative probability space
(CG, τ ).
2) In particular, in the group algebra of a right angled Artin group
G = hsi(i ∈ I) sisj = sjsi for all i 6= j with εij = 1i the generators si
(i ∈ I) are ε-independent.
Proof. 1) The ε-commutation relations are clear. So it remains to
show that a product g1 · · · gn of group elements with gj ∈ Gi(j) and
(i(1), . . . , i(n)) ∈ I ε
n, cannot be the neutral element if none of the gj
is the neutral element. But this follows from the description of graph
groups in [6]. In the notation of Definition 3.5 of [6], (g1, . . . , gn) is a re-
duced sequence, and then the above statement is contained in Theorem
3.9 of [6]. See also [3, 7].
2) This follows from the previous part, because our right angled Artin
(cid:3)
group is the ε-product of I-many copies of Z.
5. Description of ε-independence via free cumulants
We come now to the main result of this note. Namely, we want to
see that we can also describe our notion of ε-independence by some
cumulant machinery. Note, however, that we do not introduce some
kind of new cumulants, but the moment-cumulant formula will always
involve the usual free cumulants. What makes the difference is the set
of partitions over which we sum.
Definition 5.1. Let us define, for each i = (i(1), . . . , i(n)), N C ε[i] as
those partitions π ∈ P(n) for which we have π ≤ ker i (i.e., π connects
only k and l for which we have i(k) = i(l)) and which can be reduced
to the empty partition by iteration of the following two operations:
• remove "interval"-blocks, which consist just of neighbouring ele-
ments; i.e., if π = π∪{(r, r+1, r+2, . . . , r+p)}, then π ∈ N C ε[i]
if and only if π ∈ N C ε[i(1), . . . , i(r − 1), i(r + p + 1), . . . , i(n)]
• exchange the points k and k + 1 if we have εi(k)i(k+1) = 1; i.e., if
we denote by πl↔k the partition which we get from π by swaping
the points k and l, then
π ∈ N C ε[i(1), . . . , i(k), i(k + 1), . . . , i(n)]
6
ROLAND SPEICHER AND JANUSZ WYSOCZANSKI
if and only if
πk↔k+1 ∈ N C ε[i(1), . . . , i(k + 1), i(k), . . . , i(n)].
Recall that on the diagonal we have set ε to 0, i.e., we have
εii = 0 for all i ∈ I.
Another way of saying this is
N C ε[i] = {π ∈ P(n) π ≤ ker i
and π is (ε, i)-non-crossing},
where (ε, i)-non-crossing for a π with π ≤ ker i means that if there are
1 ≤ p1 < q1 < p2 < q2 ≤ n such that p1 ∼π p2, q1 ∼π q2, p1 6∼π q1, then
εi(p1)i(q1) = 1.
Note that in the case where all i-indices are the same, i(1) = i(2) =
· · · = i(n) = i, the second operation comes never into effect and hence,
for any choice of ε, we have
N C ε[(i, i, . . . , i)] = N C(n).
Let us also check the two extremes in ε. First, assume that all εij are
zero. Then (ε, i)-non-crossing is the same as non-crossing and hence
we have:
(1) N C ε[i] = {π ∈ N C(n) π ≤ ker i}
if εij = 0 for all i, j.
On the other hand, when εij = 1 for all i 6= j, then all blocks of ker i
can be commuted and N C ε[i] factorizes into a product of non-crossing
lattices, one for each block of ker i,
(2)
N C ε[i] = Y
N C(V )
if εij = 1 for all i 6= j.
V ∈ker i
Theorem 5.2. Let Ai (i ∈ I) be ε-independent in (A, ϕ). Consider
i ∈ I n and (a1, . . . , an) ∈ Ai. Then we have
(3)
ϕ(a1 · · · an) = X
κπ(a1, . . . , an),
π∈N C ε[i]
where κπ(a1, . . . , an) is the product of the free cumulants for each block,
κπ(a1, . . . , an) = Y
κV ((ak)V ),
V ∈π
where for V = (r1 < · · · < rp) ∈ π we set
κV ((ak)V ) = κp(ar1, . . . , arp).
Let us first check that this formula is the correct one in the two
extreme cases where all pairs have the same kind of independence.
Assume first that all εij = 0. Then N C ε[i] is always [0, ker i] ⊂ N C(n)
and the formula is just the moment-cumulant formula in the free case,
MIXTURES OF CLASSICAL AND FREE INDEPENDENCE
7
combined with the fact that our restriction to the summation π ≤ ker i
amounts to the vanishing of mixed free cumulants. This gives then the
rule for the calculation of free random variables.
Consider now the other extreme that εij = 1 for all i 6= j. Then
N C ε[i] factorizes as in (2), and (3) is then
ϕ(a1 · · · an) = X
Y
κπV ((ai)V ) = Y
ϕ((ak)V ),
π=(πV )V ∈ker i
V ∈ker i
V ∈ker i
i.e., ϕ(a1 · · · an) factorizes into the product of the expectations of the
product of the variables belonging to the same algebra. This is the rule
for the calculation of independent random variables.
Note that for the previous calculation we actually only needed that
all algebras for which we have a crossing in ker i commute. Hence the
same arguments prove also the following (which was also shown in [10]).
Corollary 5.3. Let Ai (i ∈ I) be ε-independent in (A, ϕ). Consider
a mixed moment ϕ(a1 · · · an) for (a1, . . . , an) ∈ Ai with i ∈ I n.
If
ker i ∈ N C ε[i] then the mixed moment factorizes into the product
ϕ(a1 · · · an) = Y
ϕ((ak)V ).
V ∈ker i
Proof. For each i ∈ I, let (Bi, ψi) be a copy of (Ai, ϕi) and define B as
the free product of the Bi with amalgamation over C1; i.e., we identify
the units of the Bi, but have no further relation among different Bi's.
Hence in B we have that Bi ∩ Bj = C1 for all i 6= j.
We define
ψ(n) : [
Bi → C
i∈I n
by
(4) ψ(n)(b1, . . . , bn) = X
π∈N C ε[i]
κπ(b1, . . . , bn)
for (b1, . . . , bn) ∈ Bi.
The κπ(b1, . . . , bn) are here as before the product of the free cumu-
lants corresponding to the blocks of π, and for each block we use the
free cumulants given by the corresponding ψi. The only ambiguity
in the definition (4) might occur when some of the bk belong to sev-
eral Bi. However, this can only happen for multiples of 1. Let us
check the case where b1 = 1, so that we have (b1, b2, . . . , bn) ∈ Bi
for i = (i, i(2), . . . , i(n) for arbitrary i. We have to see that the for-
mula in (4) is independent of i. But this follows from the fact that
κπ(1, b2, . . . , bn) is zero unless the first element is a singleton, hence π
must be of the form π = (1) ∪ σ, where σ ∈ P(2, . . . , n). But in the
constraint (1) ∪ σ ∈ N C ε[(i, i(2), . . . , i(n)] the value of i does not play
8
ROLAND SPEICHER AND JANUSZ WYSOCZANSKI
a role, since the block (1) cannot have any crossings. Thus our ψ(n)'s
are well-defined. We will use them to define a functional ψ on B by
putting
ψ(b1 · · · bn) := ψ(n)(b1, . . . , bn)
for (b1, . . . , bn) ∈ Bi
and extend this linearly. Again we have to make sure that this is well-
defined; we have to check that in the situation where two neighbouring
bk's, say b1 and b2 come from the same algebra, both possible definitions
give the same, i.e., for (b1, b1, b2 . . . , bn) ∈ B[(i(1),i(1),i(2),...,i(n))] we must
have
ψ(n+1)(b1, b1, b2 . . . , bn) = ψ(n)(b1b1, b2, . . . , bn).
The left hand side is given by
(5)
ψ(n+1)(b1, b1, b2 . . . , bn) =
X
κπ(b1, b1, b2, . . . , bn),
π∈N C ε[(i(1),i(1),i(2),...,i(n))]
whereas the right hand side is given by
X
b1, b2, . . . , bn) =
(6)
ψ(n)(b1
κσ(b1
b1, b2, . . . , bn).
σ∈N C ε[(i(1),i(2),...,i(n))]
The cumulant corresponding to the first block V = (1 < r(1) < · · · <
r(p)) of σ is now, by the formula for free cumulants with products as
arguments (see Theorem 11.12 in [12]), the same as
κp+1(b1b1, br(1), . . . , br(p)) = κp+2(b1, b1, br(1), . . . , br(p))
+
r
X
q=0
κp−q+1(b1, br(q+1), br(q+2), . . . , br(p)) · κq+1(b1, br(1), . . . , br(q))
These terms correspond exactly to the contributions of those π in (5),
which collapse to σ under the identification of the first two elements.
This shows that (5) and (6) agree and our ψ is well-defined on B.
We claim now that this ψ satisfies the defining property of ε-independence.
Assume we have (b1, . . . , bn) ∈ Bi with i ∈ I ε
n and such that ψ(bk) = 0
for all k = 1, . . . , n. But then the definition of I ε
n and N C ε[i] imply
that every π ∈ N C ε[i] must have at least one singleton, which means
that the corresponding contribution κπ in (4) is zero; hence
ψ(b1 · · · bn) = ψ(n)(b1, . . . , bn) = 0.
Since ε-independence and the distribution on the individual algebras
determines the distribution on the generated algebra, ψ must agree, via
the canonical identification Bi → Ai, with ϕ on the algebra generated
by the Ai; hence the formula (4) is also valid for ϕ.
(cid:3)
MIXTURES OF CLASSICAL AND FREE INDEPENDENCE
9
Remark 5.4. One might wonder about the apparent unsymmetry of
Theorem 5.2 with respect to free and classical independence, as only
free cumulants show up. However, as was pointed out to us by Guil-
laume Cebron this is due to our choice that on the diagonal εii is always
zero; which results in the fact that each variable is described in terms
of its free cumulants. We could also change this convention and put all
εii = 1; then each variable goes with classical cumulants and we get a
version of Theorem 5.2 where the classical cumulants instead of the free
cumulants show up. Of course, the set N C ε is then different, in partic-
ular, with this definition we would have N C ε[(i, i, . . . , i)] = P(n). Also
mixtures between free and classical cumulants are possible, by choosing
some εii = 0 and other εjj = 1.
Acknowledgements
We thank Franz Lehner and Guillaume Cebron for discussions; in
particular, the former for pointing out the relevance of Coxeter and
Artin groups in this context and the latter for Remark 5.4.
References
[1] Anis Ben Ghorbal and Michael Schurmann. Non-commutative notions of sto-
chastic independence. In Mathematical Proceedings of the Cambridge Philo-
sophical Society, volume 133, pages 531 -- 561. Cambridge Univ Press, 2002.
[2] Ian Charlesworth, Brent Nelson, and Paul Skoufranis. On two-faced families
of non-commutative random variables. arXiv preprint arXiv:1403.4907, 2014.
[3] Ruth Charney. An introduction to right-angled Artin groups. Geometriae Ded-
icata, 125(1):141 -- 158, 2007.
[4] Gero Fendler. Central limit theorems for Coxeter systems and Artin systems
of extra large type. Infinite Dimensional Analysis, Quantum Probability and
Related Topics, 6(04):537 -- 548, 2003.
[5] Dominique Foata and Pierre Cartier. Probl`emes combinatiores de commutation
et r´earrengements. Springer-Verlag, 1969.
[6] Elisabeth Ruth Green. Graph products of groups. PhD thesis, University of
Leeds, 1990.
[7] Susan Hermiller and John Meier. Algorithms and geometry for graph products
of groups. Journal of Algebra, 171(1):230 -- 257, 1995.
[8] Anna Kula and Janusz Wysocza´nski. An example of a Boolean-free type central
limit theorem. Probab. Math. Statist., 33(2):341 -- 352, 2013.
[9] Mitja Mastnak and Alexandru Nica. Double-ended queues and joint moments
of left -- right canonical operators on full Fock space. International Journal of
Mathematics, 26(02):1550016, 2015.
[10] Wojciech M lotkowski. Λ-free probability. Infinite Dimensional Analysis, Quan-
tum Probability and Related Topics, 7(01):27 -- 41, 2004.
[11] Naofumi Muraki. The five independences as natural products. Infinite Dimen-
sional Analysis, Quantum Probability and Related Topics, 6(03):337 -- 371, 2003.
10
ROLAND SPEICHER AND JANUSZ WYSOCZANSKI
[12] Alexandru Nica and Roland Speicher. Lectures on the combinatorics of free
probability, volume 335 of London Mathematical Society Lecture Note Series.
Cambridge University Press, Cambridge, 2006.
[13] Roland Speicher. On universal products. Free probability theory (Waterloo,
ON, 1995), 12:257 -- 266, 1997.
[14] Roland Speicher and Moritz Weber. Quantum groups with partial commuta-
tion relations. preprint, 2016.
[15] Dan-Virgil Voiculescu. Free probability for pairs of faces I. Communications
in Mathematical Physics, 332(3):955 -- 980, 2014.
[16] Janusz Wysocza´nski. bm-independence and bm-central limit theorems asso-
ciated with symmetric cones. Infin. Dimens. Anal. Quantum Probab. Relat.
Top., 13(3):461 -- 488, 2010.
|
1004.0851 | 1 | 1004 | 2010-04-06T13:43:02 | Groupoid normalisers of tensor products: infinite von Neumann algebras | [
"math.OA"
] | The groupoid normalisers of a unital inclusion $B\subseteq M$ of von Neumann algebras consist of the set $\mathcal{GN}_M(B)$ of partial isometries $v\in M$ with $vBv^*\subseteq B$ and $v^*Bv\subseteq B$. Given two unital inclusions $B_i\subseteq M_i$ of von Neumann algebras, we examine groupoid normalisers for the tensor product inclusion $B_1\ \overline{\otimes}\ B_2\subseteq M_1\ \overline{\otimes}\ M_2$ establishing the formula $$ \mathcal{GN}_{M_1\,\overline{\otimes}\,M_2}(B_1\ \overline{\otimes}\ B_2)''=\mathcal{GN}_{M_1}(B_1)''\ \overline{\otimes}\ \mathcal{GN}_{M_2}(B_2)'' $$ when one inclusion has a discrete relative commutant $B_1'\cap M_1$ equal to the centre of $B_1$ (no assumption is made on the second inclusion). This result also holds when one inclusion is a generator masa in a free group factor. We also examine when a unitary $u\in M_1\ \overline{\otimes}\ M_2$ normalising a tensor product $B_1\ \overline{\otimes}\ B_2$ of irreducible subfactors factorises as $w(v_1\otimes v_2)$ (for some unitary $w\in B_1\ \overline{\otimes}\ B_2$ and normalisers $v_i\in\mathcal{N}_{M_i}(B_i)$). We obtain a positive result when one of the $M_i$ is finite or both of the $B_i$ are infinite. For the remaining case, we characterise the II$_1$ factors $B_1$ for which such factorisations always occur (for all $M_1, B_2$ and $M_2$) as those with a trivial fundamental group. | math.OA | math |
GROUPOID NORMALISERS OF TENSOR PRODUCTS: INFINITE VON
NEUMANN ALGEBRAS
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
Abstract. The groupoid normalisers of a unital inclusion B ⊆ M of von Neumann algebras
consist of the set GNM (B) of partial isometries v ∈ M with vBv∗ ⊆ B and v∗Bv ⊆ B. Given
two unital inclusions Bi ⊆ Mi of von Neumann algebras, we examine groupoid normalisers
for the tensor product inclusion B1 ⊗ B2 ⊆ M1 ⊗ M2 establishing the formula
GNM1 ⊗ M2 (B1 ⊗ B2)′′ = GNM1 (B1)′′ ⊗ GNM2 (B2)′′
when one inclusion has a discrete relative commutant B′
1 ∩ M1 equal to the centre of B1
(no assumption is made on the second inclusion). This result also holds when one inclusion
is a generator masa in a free group factor. We also examine when a unitary u ∈ M1 ⊗ M2
normalising a tensor product B1 ⊗ B2 of irreducible subfactors factorises as w(v1 ⊗ v2) (for
some unitary w ∈ B1 ⊗ B2 and normalisers vi ∈ NMi (Bi)). We obtain a positive result
when one of the Mi is finite or both of the Bi are infinite. For the remaining case, we
characterise the II1 factors B1 for which such factorisations always occur (for all M1, B2
and M2) as those with a trivial fundamental group. normalisers and groupoid normalisers
and tensor products and von Neumann algebras and fixed point algebras
1. Introduction
This paper is concerned with the behaviour of structrural properties of inclusions of von
Neumann algebras obtained from tensor products. This important construction has a rich
history, and we mention two particularly significant results. Given two inclusions Bi ⊆ Mi
(i = 1, 2) of von Neumann algebras, Tomita's commutation theorem (see [17]) determines
the relative commutant of B1 ⊗ B2 in M1 ⊗ M2 showing that
(1.1)
(B1 ⊗ B2)′ ∩ (M1 ⊗ M2) = (B′
1 ∩ M1) ⊗ (B′
2 ∩ M2).
In a different direction, Ge and Kadison [8] showed that when M1 and M2 are factors, any
von Neumann algebra lying between M1 ⊗ 1 and M1 ⊗ M2 must have the form M1 ⊗ B2 for
a von Neumann algebra B2 ⊆ M2. Equation (1.1) can be interpreted as saying that the two
operations of taking tensor products and passing to relative commutants commute. In [7],
Wiggins and the authors examined commutation questions of this form for the operations
of taking tensor products and passing to the algebra generated by the groupoid normalisers
in the context of inclusions of finite von Neumann algebras. This paper examines these
questions for general inclusions of von Neumann algebras.
The normalisers of an inclusion B ⊆ M of von Neumann algebras consist of those unitaries
u ∈ M with uBu∗ = B. These elements form a group denoted NM (B) and the von Neumann
algebra NM (B)′′ they generate is the normalising algebra of B in M. Normalisers were
first studied by Dixmier [2] who used them to distinguish various types of maximal abelian
subalgebras (masas). He defined a masa B ⊆ M to be singular, regular, or semiregular
respectively if the normalising algebra is B, M, or a proper subfactor. For two inclusions
Date: 8 March, 2010.
1
2
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
Bi ⊆ Mi of masas in finite von Neumann algebras, Chifan [1] showed that the normalising
algebra of the tensor product is related to the tensor product of the individual normalising
algebras by
(1.2)
NM1 ⊗ M2(B1 ⊗ B2)′′ = NM1(B1)′′ ⊗ NM2(B2)′′,
a formula that was established earlier in [14] for singular masas. The commutation identity
(1.2) also holds when each Bi is an irreducible subfactor of a II1 factor, [15], and in this
situation any normaliser u is of the form w(v1 ⊗ v2) for normalisers vi ∈ NMi(Bi) and some
unitary w ∈ B1 ⊗ B2.
In general (1.2) fails. Consider the inclusion of B = C ⊕ M2 as a subalgebra of the 3 × 3
matrices M3 = M. Dimension considerations show that every normaliser of B lies in B. Since
B⊗B ∼= C⊕M2⊕M2⊕M4, we can find a unitary in M3⊗M3 which interchanges the two copies
of M2. This produces a non-trivial normaliser of the tensor product inclusion B⊗B ⊆ M ⊗M.
The obstruction here is the presence of partial isometries v ∈ M \ B with vBv∗ ⊆ B and
v∗Bv ⊆ B. Such partial isometries form a groupoid denoted GNM (B) and are referred to
as the groupoid normalisers of B in M. For masas in finite von Neumann algebras and
irreducible inclusions of subfactors of II1 factors the groupoid normalisers and normalisers
generate the same von Neumann algebra and so this obstruction does not occur, but in
general GNM (B)′′ can be larger than NM (B)′′. There are two related problems regarding
the form of the groupoid normalising algebra of a tensor product of inclusions.
Question 1.1. Consider two unital inclusions Bi ⊆ Mi of von Neumann algebras.
(i) Under what conditions on both inclusions do we have
(1.3)
GNM1 ⊗ M2(B1 ⊗ B2)′′ = GNM1(B1)′′ ⊗ GNM2(B2)′′?
(ii) Under what conditions on B1 ⊆ M1 is (1.3) valid for all choices of B2 ⊆ M2?
i ∩ Mi lie in Bi for i = 1, 2.
The main result of [7] established (1.3) when M1 and M2 are finite and the relative
commutants B′
In general some assumption on the relative
commutant is necessary for (1.3) to hold (see [7, Example 1.1] for a 3 × 3 matrix example
demonstrating this). In the case when one of M1 or M2 is infinite, additional hypotheses
will be needed to ensure that (1.3) holds. Indeed, the following easy example shows that it
is not possible to extend Chifan's result [1] for masas to the infinite setting.
Example 1.2. Let H = L2[0, 1] and let A = L∞[0, 1] acting on H by left multiplication.
This is the unique (up to unitary conjugacy) diffuse masa in B(H) and direct computations
show that NB(H)(A)′′ = GNB(H)(A)′′ = B(H). Now let H1 = H ⊕ C and let A1 = A ⊕ C.
This is a masa in B(H1) and
(1.4)
GNB(H1)(A1)′′ = NB(H1)(A1)′′ = B(H) ⊕ B(C) $ B(H1).
Now consider A ⊗ A1 acting on H ⊗ H1. Since this is a diffuse masa, it is unitarily conjugate
to the original inclusion A ⊆ B(H). In particular
(1.5)
GNB(H) ⊗ B(H1)(A ⊗ A1)′′ = NB(H) ⊗ B(H1)(A ⊗ A1)′′ = B(H) ⊗ B(H1).
Thus
(1.6)
GNB(H) ⊗ B(H1)(A ⊗ A1)′′ % GNB(H)(A)′′ ⊗ GNB(H1)(A1)′′.
In this paper we obtain positive answers to Question 1.1 (i) in several different contexts,
and also to Question 1.1 (ii) in the following two situations:
GROUPOID NORMALISERS
3
(1) When one inclusion has an atomic relative commutant B′
1 ∩ M1 which is equal to the
centre of B1 (Theorem 4.6).
(2) When one inclusion is the generator masa in a free group factor (Theorem 4.8).
Unlike [7], both cases above make no assumption on the other inclusion. The methods
employed there make extensive use of basic construction techniques and so are specialised
to inclusions of finite von Neumann algebras. Here we develop new techniques which reach
outside the finite setting and so the paper can be read independently of [7]. The second
case described above is obtained by a direct combinatorial calculation examining the group
elements supporting a groupoid normaliser. This is a self-contained argument found in
Lemma 4.7 and Theorem 4.8 below. In particular, this gives the first example of a masa --
factor inclusion B1 ⊆ M1 with diffuse relative commutant B′
1 ∩ M1 so that (1.3) holds for all
inclusions B2 ⊆ M2.
We study groupoid normalisers by examining the fixed point algebra of certain groups of
automorphisms. These techniques have their origins in [5, 6] and are based on the simple
observation that any normaliser u ∈ NM (B) gives rise to an automorphism Ad(u) of the
commutant of B (in some faithful representation) which fixes M ′ pointwise. In Section 3
we examine all automorphisms of B′ which fix M ′ pointwise and show that the subalgebra
of B′ fixed by these automorphisms is precisely the commutant of the groupoid normalisers
GNM (B) (Theorem 3.3). This enables us to characterise those inclusions B ⊆ M for which
NM (B)′′ = GNM (B)′′. In Section 4, we use the description of GNM (B) in terms of these
automorphisms to establish instances of (1.3). A key idea is to tensor by copies of B(H)
to ensure that all our von Neumann algebras are properly infinite as, in the countably
decomposable situation, this forces NM (B)′′ = GNM (B)′′ (Lemma 2.4).
In Section 5 we examine two inclusions of irreducible subfactors Bi ⊆ Mi. In [15] it was
shown that when both M1 and M2 are finite, then a normaliser of B1 ⊗ B2 in M1 ⊗ M2 must
factorise as w(v1 ⊗ v2) for w ∈ B and vi ∈ NMi(Bi). Like [7], this relies on basic construction
techniques. In Section 5 we study such normalisers u by means of the induced automorphism
Ad(u) of the commutant B′
2. We show that such an automorphism necessarily splits
as a tensor product θ1 ⊗ θ2 of automorphisms of B′
i . From this we recover [15,
Theorem 4.2] and extend it to cover the following situations:
i which fix M ′
1 ⊗ B′
(1) Only one of M1 or M2 is finite;
(2) Both B1 and B2 are infinite;
(3) B1 is finite with trivial fundamental group and M1 is infinite.
We provide examples that show that this result can fail whenever the algebra B1 of (3)
has a nontrivial fundamental group. This enables us to characterise II1 factors with trivial
fundamental group in terms of normalising unitaries of tensor product inclusions.
Many results in the paper rely on Lemma 2.3 which requires a hypothesis of countable
decomposability, and so attention is generally resticted to this class of algebras. The excep-
tions are the results from Lemma 4.7 through to Theorem 5.8. Finally, all inclusions B ⊆ M
of von Neumann algebras in the paper are assumed to share the same unit unless explicitly
stated otherwise.
Acknowledgements. The work in this paper originated in the Workshop in Analysis and
Probability, held at Texas A&M in 2008.
It is a pleasure to record our thanks to the
organisers of the workshop and to the NSF for providing financial support to the workshop.
4
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
2. Preliminaries
In this section we establish some preliminary results, the most important of which is
Lemma 2.6 which describes a useful connection between groupoid normalisers and the fixed
point algebras of certain automorphism groups. We begin with the following observation,
which is a version of [6, Lemma 2.11].
Proposition 2.1. Let B ⊆ M be an inclusion of von Neumann algebras and let x ∈ M
satisfy xBx∗ ⊆ B and x∗Bx ⊆ B. Let p, q ∈ B be central projections and let pxq = vpxq
be the polar decomposition of pxq. Then pxq ∈ B and v ∈ GNM (B).
For a masa A in a II1 factor M, a theorem of Dye [4] (see also [13, Lemma 6.2.3]) states that
every v ∈ GNM (A) is of the form ue for some projection e ∈ A and a unitary normaliser u ∈
NM (A). This cannot hold in the infinite setting: the unilateral shift is a groupoid normaliser
of the diagonal masa in B(ℓ2(N)) which cannot be extended to a unitary normaliser. However
the unilateral shift does lie in the algebra generated by the normalisers. The lemma below
expands this observation to a general setting. We adopt the notation Z(N) for the centre of
any von Neumann algebra N.
Lemma 2.2. Let B ⊆ M be an inclusion of von Neumann algebras and let v ∈ GNM (B)
satisfy vv∗, v∗v ∈ Z(B). Then v ∈ NM (B)′′.
Proof. Note that the hypotheses ensure that both v and v∗ lie in GNM (Z(B)). Indeed, given
z ∈ Z(B) and b ∈ B, we have
(2.1)
vzv∗b = vzv∗bvv∗ = vv∗bvzv∗ = bvzv∗,
so that vZ(B)v∗ ⊆ Z(B). Similarly v∗Z(B)v∗ ⊆ Z(B).
Let p be the maximal projection in Z(B) with p ≤ v∗v and vp ∈ NM (B)′′. Suppose
that p 6= v∗v and write p0 = v∗v − p ∈ Z(B) and q0 = vp0v∗ ∈ Z(B). If p0 = q0, then
u = vp0 + (1 − p0) is a unitary normaliser of B and vp0 = up0 ∈ NM (B)′′, contradicting
maximality of p. Hence p0 6= q0. Now suppose that p1 = p0(1 − q0) 6= 0, and write
q1 = vp1v∗ ∈ Z(B) so q1p0 = 0. Define a unitary by u = vp1 + (vp1)∗ + (1 − p1 − q1). For
b ∈ B, we have
(2.2)
ubu∗ = u(p1b + q1b + (1 − p1 − q1)b)u∗ = vp1bv∗ + v∗q1bv + (1 − p1 − q1)b ∈ B
so that u is a unitary normaliser of B with vp1 = up1. Thus v(p + p1) ∈ NM (B)′′, contradict-
ing maximality of p. Finally, if q1 = (1 − p0)q0 6= 0, then by interchanging the roles of v and
v∗, there is some u ∈ NM (B) with u∗q1 = v∗q1 so that q1v = v(v∗(1 − p0)q0v) ∈ NM (B)′′. As
v∗(1 − p0)q0v ∈ Z(B), we can adjoin v∗(1 − p0)q0v ≤ p0 to p, contradicting the maximality
of p.
(cid:3)
For inclusions of properly infinite von Neumann algebras, standard techniques allow us
to adjust groupoid normalisers to have central initial and final projections so the previous
lemma applies. We record the details in slightly greater generality for use in Section 4.
Lemma 2.3. Let M be a von Neumann algebra and let B1, B2 ⊆ M be properly infinite,
countably decomposable von Neumann subalgebras of M. Any partial isometry v ∈ M with
vB1v∗ ⊆ B2 and v∗B2v ⊆ B1 factorises as v = b2wb1, where b1, b2, w are partial isometries
with wB1w∗ ⊆ B2, w∗B2w ⊆ B1, w∗w ∈ Z(B1), ww∗ ∈ Z(B2), b1 ∈ B1 and b2 ∈ B2.
GROUPOID NORMALISERS
5
Proof. Let e = v∗v ∈ B1 and let p be the central support of e in B1. As B1 is properly
infinite, standard arguments enable us to find a sequence (en)∞
n=1 (necessarily countable from
the hypothesis of countable decomposability) of pairwise orthogonal, equivalent projections
in B1, and which sum to p and with e1 = e. Similarly, let vv∗ = f ∈ B2, let q be the
central support of f in B2, and find a sequence (fn)∞
n=1 of pairwise orthogonal, equivalent
projections in B2 which sum to q with f1 = f . For each n, find partial isometries b1,n ∈ B1
with b∗
2,nb2,n = f and
b2,nb∗
1,n = en and partial isometries b2,n ∈ B2 with b∗
2,n = fn. Then define a partial isometry
1,nb1,n = e and b1,nb∗
(2.3)
w =
∞
X
n=1
b2,nvb∗
1,n ∈ M,
noting that the series converges in the strong operator topology. By construction ww∗ = q ∈
Z(B2), w∗w = p ∈ Z(B1), wB1w∗ ⊆ B2, w∗B2w ⊆ B1 and v = b∗
(cid:3)
Lemma 2.4. Let B ⊆ M be an inclusion of von Neumann algebras. Suppose that B is
properly infinite and countably decomposable. Then GNM (B)′′ = NM (B)′′.
Proof. Using Lemma 2.3, any groupoid normaliser v of B can be factorised in the form b2wb1
so that w is a groupoid normaliser of B with ww∗, w∗w ∈ Z(B) and b1, b2 ∈ B. By Lemma
2.2, w lies in NM (B)′′ and hence so too does v.
(cid:3)
2,1wb1,1.
Now we turn to the connections between normalisers, groupoid normalisers and certain
automorphisms of the commutant inclusion.
Definition 2.5. Given an inclusion B ⊆ M of von Neumann algebras, we define the B-
bimodular automorphisms of M, denoted AutB(M), to be those θ ∈ Aut(M) satisfying
θ(b) = b for all b ∈ B. Given a subgroup G of the automorphism group of M, let M G denote
the fixed point algebra {x ∈ M : θ(x) = x, θ ∈ G}. This is a von Neumann subalgebra of M.
In particular we can apply this to AutB(M) to obtain M AutB (M ) which is a von Neumann
subalgebra of M containing B.
A number of the von Neumann algebras and groups that will appear below have compli-
cated notations, so we will sometimes adopt the expression Fix (M, G) for M G.
On occasion we shall need to assume that certain von Neumann algebras are represented
in standard form. In full generality, the standard form of a von Neumann algebra was set out
by Haagerup, [9]. The key fact we will need is that every automorphism of a von Neumann
algebra in standard form is spatially implemented.
Lemma 2.6. Let B ⊆ M ⊆ B(H) be an inclusion of von Neumann algebras. Then
(2.4)
and equality holds if B (or equivalently B′) is in standard form on H.
Proof. Let x ∈ (B′)AutM ′ (B ′). Given u ∈ NM (B), we have uBu∗ = B so uB′u∗ = B′.
Thus Ad(u)B ′ defines an automorphism of B′ which is M ′-bimodular as u ∈ M. Hence
Ad(u)(x) = x and so x ∈ NM (B)′.
(B′)AutM ′ (B ′) ⊆ NM (B)′,
Now suppose that B′ lies in standard form on H and take x ∈ NM (B)′ ⊆ B′. Given
θ ∈ AutM ′(B′), there is a unitary u ∈ B(H) such that θ = Ad(u)B ′ as B′ is in standard
form on H. Since θ(m′) = m′ for m′ ∈ M ′, the double commutant theorem gives u ∈ M.
Now uB′u∗ = B′, so we can take commutants to see that uBu∗ = B, placing u ∈ NM (B). It
follows that θ(x) = uxu∗ = x, since x ∈ NM (B)′, and so x ∈ (B′)AutM ′ (B ′), as required. (cid:3)
6
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
3. Groupoid normalisers and fixed point algebras
Our objective in this section is to characterise the groupoid normalising algebra GNM (B)′′
of an inclusion of von Neumann algebras as the commutant of the fixed point algebra
(B′)AutM ′ (B ′). The first step is to show that the inclusions of an atomic masa inside B(K),
and the inclusion B(K) ⊆ B(K) give positive answers to Question 1.1 (ii). We proceed by
direct computations with the matrix units as in [6, Section 2.2].
Lemma 3.1. Let K be a Hilbert space and let A be the atomic masa in B(K). Given any
inclusion B ⊆ M of von Neumann algebras, we have
(3.1)
GNM ⊗ B(K)(B ⊗ A)′′ = GNM ⊗ B(K)(B ⊗ B(K))′′ = GNM (B)′′ ⊗ B(K).
Proof. We consider a more general situation which will imply containment of each of the
first two algebras from (3.1) into the third simultaneously. Thus take a partial isometry
v ∈ M ⊗ B(K) with
(3.2)
v(B ⊗ A)v∗ ⊆ B ⊗ B(K) and v∗(B ⊗ A)v ⊆ B ⊗ B(K),
and note that groupoid normalisers from the first two algebras in (3.1) will satisfy (3.2). Use
the minimal projections in A to identify operators in M ⊗ B(K) with matrices over M, and
write v = (vi,j)i,j∈I. For each j ∈ I and x ∈ B, let y ∈ B ⊗ A be the operator with x in
the (j, j)-position and 0 elsewhere. By considering the (i, i)-th component of vyv∗, we see
that vi,jxv∗
i,jBvi,j ⊆ B for all i, j. Let wi,jhi,j be the polar decomposition
of vi,j, so wi,j ∈ GNM (B) and hi,j ∈ B by Proposition 2.1. Thus each vi,j lies in GNM (B)′′
so that v ∈ GNM (B)′′ ⊗ B(K). This shows that the first and second algebras in (3.1)
are contained in GNM (B)′′ ⊗ B(K). Since the reverse inclusions are immediate, the result
follows.
(cid:3)
i,j ∈ B. Similarly v∗
Lemma 3.2. Let B ⊆ M be an inclusion of von Neumann algebras such that B is countably
decomposable, and let K be an infinite dimensional separable Hilbert space. Then
(3.3)
NM ⊗ B(K)(B ⊗ B(K))′′ = GNM (B)′′ ⊗ B(K).
Proof. Since B ⊗ B(K) is properly infinite and countably decomposable, Lemma 2.4 gives
(3.4)
NM ⊗ B(K)(B ⊗ B(K))′′ = GNM ⊗ B(K)(B ⊗ B(K))′′.
The result then follows from the second equality of Lemma 3.1.
(cid:3)
We can now characterise those inclusions for which NM (B)′′ = GNM (B)′′ in terms of the
fixed points of the automorphism group AutM ′(B′).
Theorem 3.3. Let B ⊆ M ⊆ B(H) be an inclusion of von Neumann algebras such that B
is countably decomposable, and let K be an infinite dimensional separable Hilbert space. The
following statements are equivalent:
(i) NM (B)′′ = GNM (B)′′;
(ii) NM ⊗ B(K)(B ⊗ B(K))′′ = NM (B)′′ ⊗ B(K);
(iii) NM (B)′ = (B′)AutM ′ (B ′).
Proof. The equivalence between statements (i) and (ii) follows from Lemma 3.2 and is implicit
in [6, Theorem 2.4].
(iii) =⇒ (ii). Applying Lemma 2.6 to B ⊗ B(K) ⊆ M ⊗ B(K) gives
(3.5)
Fix (B′ ⊗ C1, AutM ′⊗C1(B′ ⊗ C1)) ⊆ NM ⊗ B(K)(B ⊗ B(K))′.
GROUPOID NORMALISERS
7
We have an isomorphism between AutM ′(B′) and AutM ′⊗C1(B′ ⊗ C1) given by θ 7→ θ ⊗ id
which demonstrates that
(3.6)
Fix (B′ ⊗ C1, AutM ′⊗C1(B′ ⊗ C1)) = (B′)AutM ′ (B ′) ⊗ C1.
Applying the hypothesis of (3) gives
(3.7)
and the inclusion
(3.8)
NM (B)′ ⊗ C1 ⊆ NM ⊗ B(K)(B ⊗ B(K))′,
NM ⊗ B(K)(B ⊗ B(K))′′ ⊆ NM (B)′′ ⊗ B(K)′′
follows by taking commutants in (3.7). Since the reverse inclusion is immediate, condition
(2) holds.
(2) =⇒ (3). Let π be a standard representation of B′ on some Hilbert space H1. By the
general theory of representations of von Neumann algebras [3, p. 61], we can assume that
π is obtained by an amplification of the representation on H followed by a compression.
Therefore, we can find another Hilbert space K and a projection in e ∈ B ⊗ B(K) =
(B′ ⊗ 1K)′ such that π(x) = e(x ⊗ 1K)e acting on e(H ⊗ K). Writing B1 = π(B′)′ and M1 =
π(M ′)′ with the commutants taken on H1 = e(H ⊗ K), we have B1 = e(B ⊗ B(K))e and
M1 = e(M ⊗ B(K))e. Since π is a faithful representation of B′, we have π((B′)AutM ′ (B ′)) =
1). Furthermore, since π(B′) is in standard form, so too is B1 = π(B′)′. Thus
(B′
Lemma 2.6 gives
1)AutM ′
(B ′
1
(3.9)
π((B′)AutM ′ (B ′)) = (B′
1)AutM ′
1
(B ′
1) = NM1(B1)′.
Every normaliser v of e(B ⊗ B(K))e in e(M ⊗ B(K))e is a groupoid normaliser of
B ⊗ B(K) in M ⊗ B(K) and so lies in NM ⊗ B(K)(B ⊗ B(K))′′ by Lemma 2.4. This gives
the inclusion
(3.10)
Now
(3.11)
so that
(3.12)
NM1(B1)′′ = Ne(B ⊗ B(K))e(e(M ⊗ B(K))e)′′ ⊆ e(NM ⊗ B(K)(B ⊗ B(K))′′)e.
π(NM (B)′)′ = [(NM (B)′ ⊗ C1K)e]′ = e(NM (B)′′ ⊗ B(K))e,
π(NM (B)′)′ = e(NM (B)′′ ⊗ B(K))e
= e(NM ⊗ B(K)(B ⊗ B(K))′′)e
⊇ Ne(B ⊗ B(K))e(e(M ⊗ B(K))e)′′
= NM1(B1)′′ = π((B′)AutM ′ (B ′))′,
by (3.11), condition (2), (3.10) and (3.9). Taking commutants gives
(3.13)
and so
(3.14)
π(NM (B)′) ⊆ π((B′)AutM ′ (B ′)),
NM (B)′ ⊆ (B′)AutM ′ (B ′)
since π is a faithful representation of B′. The reverse inclusion (B′)AutM ′ (B ′) ⊆ NM (B)′ is
Lemma 2.6 so condition (3) holds.
(cid:3)
8
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
Finally in this section we can express the groupoid normalising algebra GNM (B)′′ in terms
of the automorphism group AutM ′(B′). We will use this result repeatedly in the next section
to obtain further instances of (1.3).
Theorem 3.4. Let B ⊆ M ⊆ B(H) be an inclusion of von Neumann algebras where B is
countably decomposable. Then
(3.15)
GNM (B)′ = (B′)AutM ′ (B ′).
Proof. Let K be an infinite dimensional separable Hilbert space and define B1 = B ⊗ B(K)
and M1 = M ⊗ B(K). Then
(3.16)
NM1(B1)′′ = GNM (B)′′ ⊗ B(K),
by Lemma 3.2. Since B1 is properly infinite, Lemma 2.4 gives GNM1(B1)′′ = NM1(B1)′′ so
the inclusion B1 ⊆ M1 satisfies condition (1) of Theorem 3.3. By (3.16) and condition (3)
of this theorem,
(3.17)
GNM (B)′ ⊗ C1K = NM1(B1)′ = (B′
1)AutM ′
1
(B ′
1).
Finally note that θ 7→ θ ⊗ idB(K) gives an isomorphism between AutM ′(B′) and AutM ′
so that (B′
1) = (B′)AutM ′ (B ′) ⊗ C1K. Thus GNM (B)′ = (B′)AutM ′ (B ′).
1)AutM ′
1(B′
1)
(cid:3)
(B ′
1
4. Groupoid normalisers and tensor products
In this section, we establish our positive answers to Question 1.1. We being with separably
acting inclusions of the form B ⊗ A ⊆ M ⊗ A, where A is abelian. When B is singular in
M, [6, Lemma 4.1] showed that B ⊗ A is singular in M ⊗ A by a direct integral argument
based on [16, Lemma 6.6]. The proof given in [6] shows that any normaliser u of B ⊗ A in
M ⊗ A is a direct integral of normalisers of B in M over the spectrum of A. The analogous
result also holds for groupoid normalisers, giving the following lemma. We omit the proof,
which is a routine modification of [16, Lemma 6.6] and [6, Lemma 4.1].
Lemma 4.1. Let B ⊆ M be an inclusion of separably acting von Neumann algebras. If A
is a separable abelian von Neumann algebra, then
(4.1)
NM ⊗ A(B ⊗ A)′′ = NM (B)′′ ⊗ A, GNM ⊗ A(B ⊗ A)′′ = GNM (B)′′ ⊗ A.
The next lemma follows the same pattern as the deduction of [6, Theorem 4.3] from [6,
Theorem 3.1, Lemma 4.1].
Lemma 4.2. Let B ⊆ M be a unital inclusion of separably acting von Neumann algebras
and let L be a separably acting von Neumann algebra. Then
(4.2)
GNM ⊗ L(B ⊗ L)′′ = GNM (B)′′ ⊗ L.
Proof. Fix a faithful representation of M on a separable Hilbert space H1 and use this to
define the commutants B′ ⊇ M ′. Write A = Z(L) and take a standard representation of A
on a separable Hilbert space K so that A is a masa in B(K). Work on the Hilbert space
H1 ⊗ K so that (B ⊗ A)′ = B′ ⊗ A. Given θ ∈ AutM ′ ⊗ A(B′ ⊗ A), Theorem 3.4 gives
θ(x) = x for all x ∈ GNM ⊗ A(B ⊗ A)′ = GNM (B)′ ⊗ A, where the last identity is obtained
by taking commutants in Lemma 4.1.
GROUPOID NORMALISERS
9
Now suppose that L is faithfully represented on a separable Hilbert space H2 and work
on H1 ⊗ H2. Given an automorphism θ ∈ AutM ′ ⊗ L′(B′ ⊗ L′), we have
θ(x)(1B ′ ⊗ ℓ′) = θ(x)θ(1B ′ ⊗ ℓ′) = θ(x(1B ′ ⊗ ℓ′)) = θ((1B ′ ⊗ ℓ′)x)
(4.3)
= θ(1B ′ ⊗ ℓ′)θ(x) = (1B ′ ⊗ ℓ′)θ(x),
ℓ′ ∈ L′, x ∈ B′ ⊗ A,
since θ(1B ′ ⊗ ℓ′) = 1B ′ ⊗ ℓ′ and 1B ′ ⊗ ℓ′ commutes with B′ ⊗ A ⊆ B′ ⊗ L′. Thus
θ(x) ∈ (C1B ′ ⊗ L′)′ ∩ (B′ ⊗ L′) = B′ ⊗ Z(L) = B′ ⊗ A,
(4.4)
and so every element θ of AutM ′ ⊗ L′(B′ ⊗ L′) restricts to an element of AutM ′ ⊗ A(B′ ⊗ A).
It then follows from the first paragraph of the proof that θ(x) = x for all x ∈ GNM (B)′ ⊗ A.
Since θ was arbitrary, it follows that
(4.5)
GNM (B)′ ⊗ A ⊆ Fix (B′ ⊗ L′, AutM ′ ⊗ L′(B′ ⊗ L′)) = GNM ⊗ L(B ⊗ L)′,
where the second equality is Theorem 3.4. Take commutants to obtain
(4.6)
on H1 ⊗ H2. Then
GNM (B)′′ ⊗ A′ ⊇ GNM ⊗ L(B ⊗ L)′′
(4.7)
GNM ⊗L(B ⊗ L)′′ ⊆ (GNM (B)′′ ⊗ A′) ∩ (M ⊗ L) = GNM (B)′′ ⊗ L.
Since the reverse inclusion is immediate, the result follows.
(cid:3)
We now start work on our main result. Given two unital inclusions Bi ⊆ Mi of von
Neumann algebras, we will show that
(4.8)
GNM1 ⊗ M2(B1 ⊗ B2)′′ = GNM1(B1)′′ ⊗ GNM2(B2)′′,
when B′
1∩M1 is atomic and lies in the centre of B1. The inclusion from right to left is immedi-
ate. We establish the inclusion from left to right by demonstrating that GNM1 ⊗ M2(B1 ⊗ B2)′′
is contained in M1 ⊗ GNM2(B2)′′ and in GNM1(B1)′′ ⊗ M2 separately. The next lemma, which
is based on [6, Theorem 4.4], handles the first, and easier, of these two inclusions.
Lemma 4.3. For i = 1, 2, let Bi ⊆ Mi be unital inclusions of separably acting von Neumann
algebras and suppose that B′
1 ∩ M1 = Z(B1) is atomic. Then
(4.9)
GNM1 ⊗ M2(B1 ⊗ B2)′′ ⊆ M1 ⊗ GNM2(B2)′′.
Proof. Represent M1 and M2 faithfully on Hilbert spaces H1 and H2 respectively and consider
1) and
an automorphism θ ∈ AutM ′
y2 ∈ B′
2). Write A = Z(B1). For y1 ∈ A = Z(B′
1 ⊗ B′
(B′
1 ⊗ M ′
2, we have
2
(4.10)
Thus
θ(y1 ⊗ y2)(x ⊗ 1) = θ(y1x ⊗ y2) = θ(xy1 ⊗ y2)
= (x ⊗ 1)θ(y1 ⊗ y2),
x ∈ M ′
1.
1 ⊗ C1)′ ∩ (B′
θ(y1 ⊗ y2) ∈ (M ′
2 = A ⊗ B′
2,
2). Let A be represented
and so it follows that θ restricts to an element of AutC1⊗M ′
as the diagonal operators on some Hilbert space K, so that working on K ⊗ H2 we have
(A ⊗ B2)′ = A ⊗ B′
2. Lemma 3.1 gives GNB(K) ⊗ M2(A ⊗ B2)′′ = B(K) ⊗ GNM2(B2)′′ so that
Fix (A ⊗ B′
1 ∩ M1) ⊗ B′
2(A ⊗ B′
2)) = C1 ⊗ GNM2(B2)′,
2, AutC1⊗M ′
2(A ⊗ B′
1 ⊗ B′
2) = (B′
(4.11)
(4.12)
10
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
by Theorem 3.4. Thus θ(1 ⊗ z) = 1 ⊗ z for all z ∈ GNM2(B2)′. Theorem 3.4 also gives
(4.13)
so that
(4.14)
The inclusions
GNM1 ⊗ M2(B1 ⊗ B2)′ = Fix (B′
1 ⊗ B′
2, AutM ′
1 ⊗ M ′
2
(B′
1 ⊗ B′
2))
C1 ⊗ GNM2(B2)′ ⊆ GNM1 ⊗ M2(B1 ⊗ B2)′.
GNM1 ⊗ M2(B1 ⊗ B2)′′ ⊆ (B(H1) ⊗ GNM2(B2)′′) ∩ (M1 ⊗ M2)
(4.15)
= M1 ⊗ GNM2(B2)′′
follow by taking commutants.
(cid:3)
Now we turn to the second inclusion required for (4.8). Recall that a von Neumann algebra
in standard form has the property that its automorphisms are all spatially implemented.
Lemma 4.4. Let B ⊆ M be a unital inclusion of separably acting von Neumann algebras
with B′ ∩ M ⊆ B and suppose that p and q are minimal central projections in B. Let L
be another separable von Neumann algebra acting in standard form on the Hilbert space K.
Suppose that v ∈ GNM ⊗ B(K)(B ⊗ L) satisfies v∗v, vv∗ ∈ Z(B ⊗ L) with v∗v ≤ p ⊗ 1 and
vv∗ ≤ q ⊗ 1. Then v ∈ GNM (B)′′ ⊗ B(K).
Proof. We represent M on a Hilbert space H. Since p and q are minimal central projections
in Z(B), there are central projections p1, q1 ∈ L with v∗v = p ⊗ p1 and vv∗ = q ⊗ q1. Thus
x 7→ vxv∗ gives an isomorphism from (Bp) ⊗ (Lp1) onto (Bq) ⊗ (Lq1). As a consequence,
we obtain a surjective isomorphism θ : (B′p) ⊗ (L′p1) → (B′q) ⊗ (L′q1) by θ(y) = vyv∗.
Note that v commutes with M ′ ⊗ C1 so
(4.16)
θ(M ′ ⊗ p1) = v(M ′ ⊗ p1)v∗ = v(M ′ ⊗ 1)v∗ = (M ′ ⊗ 1)vv∗ = M ′q ⊗ q1.
We claim that θ(p ⊗ L′p1) = q ⊗ L′q1. Indeed, for m′ ∈ M ′ and ℓ′ ∈ L′,
(4.17)
(m′p ⊗ p1)(p ⊗ ℓ′p1) = (p ⊗ ℓ′p1)(m′p ⊗ p1),
so that applying θ and using (4.16), we obtain
(4.18)
θ(p ⊗ ℓ′p1) ∈ (M ′q ⊗ q1)′ ∩ (B′q ⊗ L′q1) = (B′ ∩ M)q ⊗ L′q1 = q ⊗ L′q1,
from the minimality of q ∈ Z(B). Interchanging the roles of v and v∗ shows that θ−1(q ⊗
L′q1) = p ⊗ L′p1, establishing the claim.
Since both L′p1 and L′q1 act in standard form on p1(K) and q1(K) respectively, there is
a partial isometry w ∈ B(K) with w∗w = p1 and ww∗ = q1 so that
(4.19)
v(p ⊗ ℓ′p1)v∗ = θ(p ⊗ ℓ′p1) = (1 ⊗ w)(q ⊗ ℓ′p1)(1 ⊗ w)∗,
ℓ′ ∈ L′.
Define v1 = (1 ⊗ w)∗v so that
(4.20)
and
(4.21)
v1v∗
1 = (1 ⊗ w)∗vv∗(1 ⊗ w) = (1 ⊗ w)∗(q ⊗ q1)(1 ⊗ w) = q ⊗ p1
1v1 = v∗(1 ⊗ w)(1 ⊗ w)∗v = v∗(1 ⊗ q1)v = p ⊗ p1.
v∗
GROUPOID NORMALISERS
11
As (1 ⊗ w)(1 ⊗ w∗) = 1 ⊗ q1 ≥ v1v∗
and direct computations give
1, we have v = (1 ⊗ w)v1. Certainly v1 ∈ M ⊗ p1B(K)p1,
v1(1 ⊗ ℓ′) =v1v∗
1v1(1 ⊗ ℓ′)v∗
1v1 = v1(p ⊗ l′p1)v∗
1v1
(4.22)
Thus
=(1 ⊗ w)∗θ(p ⊗ ℓ′p1)(1 ⊗ w∗)v1
=(q ⊗ w∗wℓ′w∗w)v1 = (1 ⊗ ℓ′)v1,
ℓ′ ∈ L′.
v1 ∈ (C1 ⊗ L′)′ ∩ (M ⊗ p1B(K)p1) = M ⊗ Lp1 ⊆ M ⊗ L.
1 is an isomorphism from B′p ⊗ L′p1 = (B ⊗ L)′ ∩ B(pH ⊗ p1K) onto
(4.23)
Since x 7→ v1xv∗
B′q ⊗ L′p1 = (B ⊗ L)′ ∩ B(qH ⊗ p1K) we can take commutants to see that
(4.24)
and similarly that v∗
v1 ∈ GNM (B)′′ ⊗ L by Lemma 4.2. It follows that
(4.25)
1 ⊆ B ⊗ L. Consequently v1 ∈ GNM ⊗ L(B ⊗ L) and so
v = (1 ⊗ w)v1 ∈ GNM (B)′′ ⊗ B(K),
1(B ⊗ L)v∗
v1(B ⊗ L)v∗
1 ⊆ B ⊗ L
as required.
(cid:3)
Lemma 4.5. Let B ⊆ M be an inclusion of separably acting von Neumann algebras so that
B′ ∩ M = Z(B) is atomic. Let L be another von Neumann algebra, acting in standard form
on a separable Hilbert space K. Then
(4.26)
GNM ⊗ B(K)(B ⊗ L)′′ ⊆ GNM (B)′′ ⊗ B(K).
Proof. We first show that it suffices to prove the lemma under the additional assumption
that B is properly infinite. If B is not properly infinite, consider the inclusion
(4.27)
B0 = B(ℓ2(N)) ⊗ B ⊆ B(ℓ2(N)) ⊗ M = M0,
which also satisfies the hypotheses of the lemma and B0 is properly infinite. Lemma 3.1
gives
(4.28)
and
GNM0(B0)′′ = B(ℓ2(N)) ⊗ GNM (B)′′
(4.29)
GNM0 ⊗ B(K)(B0 ⊗ L)′′ = B(ℓ2(N)) ⊗ GNM ⊗ B(K)(B ⊗ L)′′.
Thus, assuming that the lemma holds for the inclusion B0 ⊆ M0, we obtain
B(ℓ2(N)) ⊗ GNM ⊗ B(K)(B ⊗ L)′′ = GNM0 ⊗ B(K)(B0 ⊗ L)′′
(4.30)
⊆ GNM0(B0)′′ ⊗ B(K)
= B(ℓ2(N)) ⊗ GNM (B)′′ ⊗ B(K),
and so (4.26) holds for the original inclusion B ⊆ M.
Now assume that B is properly infinite. Fix v ∈ GNM ⊗ B(K)(B ⊗ L) and minimal projec-
tions p, q ∈ Z(B). We will show that (p ⊗ 1)v(q ⊗ 1) ∈ GNM (B)′′ ⊗ B(K), from which the
result follows immediately. Let (p⊗1)v(q⊗1) = w(p⊗1)v(1⊗1)q be the polar decomposition
of (p ⊗ 1)v(q ⊗ 1), so that (p ⊗ 1)v(q ⊗ 1) ∈ B ⊗ L and w ∈ GNM ⊗ B(K)(B ⊗ L) by Propo-
sition 2.1. We must show that w ∈ GNM (B)′′ ⊗ B(K). Write Ap = (p ⊗ 1)(B ⊗ L)(p ⊗ 1)
and Aq = (q ⊗ 1)(B ⊗ L)(q ⊗ 1) so that wAqw∗ ⊆ Ap and w∗Apw ⊆ Aq. Note that both
Ap and Aq are central cutdowns of B ⊗ L and so properly infinite. By Lemma 2.3, we can
12
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
factorise w = aquap, where ap and aq are partial isometries in Ap and Aq respectively and u
is a partial isometry with uAqu∗ ⊆ Ap, u∗Apu ⊆ Aq and u∗u ∈ Z(Aq), uu∗ ∈ Z(Ap). Lemma
4.4 then shows that u, and hence w, lies in GNM (B)′′ ⊗ B(K), exactly as required.
(cid:3)
We can now establish (4.8), giving a general class of inclusions with a positive answer to
Question 1.1 (ii).
Theorem 4.6. Let B1 ⊆ M1 be an inclusion of separably acting von Neumann algebras
where B′
1 ∩ M1 = Z(B1) is atomic and let B2 ⊆ M2 be another unital inclusion of separably
acting von Neumann algebras. Then
(4.31)
GNM1 ⊗ M2(B1 ⊗ B2)′′ = GNM1(B1)′′ ⊗ GNM2(B2)′′.
Proof. Take faithful representations of M1 and M2 on Hilbert spaes H1 and H2 respectively.
Fix θ ∈ AutM ′
2). We will show that θ(x ⊗ 1) = x ⊗ 1 for all x ∈ GNM1(B1)′.
1 ⊗ B′
(B′
1 ⊗ M ′
2
(4.32)
Define L = B′
θ(b′
2 ∩ M2. For m′
1 ⊗ ℓ)(1 ⊗ m′
1 ⊗ ℓ) ∈ (C1 ⊗ M ′
2 ∈ M ′
2) = θ(b′
2)′ ∩ (B′
2, b′
1 ⊗ ℓm2) = θ(b′
1 ⊗ B′
1 ∈ B′
1 and ℓ ∈ L = B′
2 ∩ M2 we have
1 ⊗ m2ℓ) = (1 ⊗ m′
2)θ(b′
1 ⊗ ℓ),
so that θ(b′
of AutM ′
Applying Theorem 3.4 to the inclusion B1 ⊗ L′ ⊆ M1 ⊗ B(K), we have
1 ⊗ L. In particular θ restricts to an element
1 ⊗ L). Now take a standard representation of L on K and work on H1 ⊗ K.
1⊗C1(B′
2) = B′
(4.33)
Fix (B′
1 ⊗ L, AutM ′
1⊗C1(B′
1 ⊗ L)) = GNM1 ⊗ B(K)(B ⊗ L′)′.
In particular θ(z) = z for all z ∈ GNM1 ⊗ B(K)(B ⊗ L′)′. Lemma 4.5 gives
(4.34)
GNM1 ⊗ B(K)(B1 ⊗ L′)′′ ⊆ GNM1(B1)′′ ⊗ B(K),
and the inclusion
(4.35)
GNM1(B1)′ ⊗ C1 ⊆ GNM1 ⊗ B(K)(B1 ⊗ L′)′
follows by taking commutants in (4.34). Thus θ(x ⊗ 1) = x ⊗ 1 for all x ∈ GNM1(B1)′.
Applying Theorem 3.4 to the original inclusion B1 ⊗ B2 ⊆ M1 ⊗ M2 gives
(4.36)
and so
(4.37)
GNM1 ⊗ M2(B1 ⊗ B2)′ = Fix (B′
1 ⊗ B′
2, AutM ′
1 ⊗ M ′
2
(B′
1 ⊗ B′
2))
GNM1(B1)′ ⊗ C1 ⊆ GNM1 ⊗ M2(B1 ⊗ B2)′.
Taking commutants gives
(4.38)
GNM1 ⊗ M2(B1 ⊗ B2)′′ ⊆ GNM1(B1)′′ ⊗ B(H2).
As Lemma 4.3 states that
(4.39)
we obtain
(4.40)
GNM1 ⊗ M2(B1 ⊗ B2)′′ ⊆ M1 ⊗ GNM2(B2)′′,
GNM1 ⊗ M2(B1 ⊗ B2)′′ ⊆ (GNM1(B1)′′ ⊗ B(H2)) ∩ (M1 ⊗ GNM2(B2)′′)
= GNM1(B1)′′ ⊗ GNM2(B2)′′.
As the reverse inclusion is immediate, the result follows.
(cid:3)
GROUPOID NORMALISERS
13
We end this section by giving an example of a diffuse masa-factor inclusion A ⊆ M for
which
GNM ⊗ N (A ⊗ B)′′ = GNM (A)′′ ⊗ GNN (B)′′,
(4.41)
for all inclusions B ⊆ N. Fix k ∈ N with k ≥ 2 and let Fk be the free group on the
k generators a, b1, . . . , bk−1 (the argument given below will also be valid for k = ∞). Let
M = L(Fk) be the group von Neumann algebra generated by Fk. We identify elements of
the group with the corresponding elements in M and let A be the von Neumann subalgebra
of M generated by a. This is a masa in M, known as a generator masa so has A′ ∩ M = A.
The subgroup generated by a satisfies Dixmier's criterion for singularity of the masa A [2]
(see also [13, p. 22]) and so GNM (A)′′ = A. The inclusion A ⊆ M fits into the framework
of [7] so (4.41) holds whenever B ⊆ N is an inclusion of finite von Neumann algebras with
B′ ∩ N ⊆ B. Our objective is to establish (4.41) without making any assumption on the
inclusion B ⊆ N.
We denote the standard orthonormal basis for ℓ2(Fk) by {δg : g ∈ Fk}. When we view
an operator x ∈ L(Fk) as a vector in the underlying Hilbert space, then it has a square
summable Fourier series Pg∈Fk αgδg. The support of x is then {g ∈ Fk : αg 6= 0}. When
viewing x as an operator, we write x = Pg∈Fk αgg.
Lemma 4.7. With the notation above, let H be a Hilbert space and x ∈ M ⊗ B(H) satisfy
x(A ⊗ C1)x∗ ⊆ A ⊗ B(H). Then x ∈ A ⊗ B(H).
Proof. With respect to some choice of matrix units for B(H), we may write x = (xij)i,j∈Λ
with xij ∈ L(Fk). The hypothesis implies that Pj xijatx∗
ji ∈ A for all t ∈ Z and all i ∈ Λ,
from which we wish to conclude that each xij lies in A. Thus it suffices to consider operators
yi ∈ L(Fk) so that Pj yjaty∗
j ∈ A for all t ∈ Z, and deduce that yj ∈ A for all j. Taking
t = 0 and applying the trace shows that Pj kyjk2
2 < ∞, so by scaling we may assume that
this sum is 1. We will argue by contradiction, so suppose that there is some j0 such that
yj0 /∈ A. Then the support of yj0 contains a word g which is not a power of a. Then g may
be written in reduced form as anw0am where m, n ∈ Z and w0 is a non-trivial reduced word
whose first and last letters lie in {b±1
k−1}. We will examine the coefficient of gatg−1
in S(t) := Pj yjaty∗
j and show that it is nonzero for sufficiently large values of t. This will
give the desired contradiction.
1 , . . . , b±1
Let c = hyj0δe, δgi 6= 0 be the the coefficient of g in yj0. For any fixed group element h ∈ Fk
and any t ∈ Z, the coefficient of h in yjaty∗
(4.42)
hyjaty∗
j is
j δe, δhi = haty∗
j δe, Jh−1Jy∗
j δei,
where Jh−1J is the unitary operator of right-convolution by h. This is bounded in absolute
value by kyjk2
2 using the Cauchy-Schwarz inequality. Consequently, the contribution of
Pj∈Λ0 yjaty∗
2, for any
subset Λ0 of the index set Λ. Choose a finite set Λ0 ⊆ Λ so that Pj∈Λ\Λ0 kyjk2
2 < c2/4. For
t ∈ Z, write SΛ0(t) = Pj∈Λ0 yjaty∗
j .
j to the coefficient of gatg−1 in S(t) is bounded above by Pj∈Λ0 kyjk2
For each j ∈ Λ0, we may write yj as an orthogonal sum zj + z′
j where each zj is a finite
jk2 < c2/4. Then
linear combination of group elements, g appears in zj0, and Pj∈Λ0 kz′
gatg−1 can appear in SΛ0(t) from the four terms Pj∈Λ0 zjatz∗
jatz∗
j
and Pj∈Λ0 z′
jk2 ≤ kyjk2 ≤ 1 and kzjk2 ≤ kyjk2 ≤ 1, we may argue as above to
see that the total contribution (in absolute value) of the latter three terms to the coefficient
j . Since kz′
j , Pj∈Λ0 zjatz′∗
j , Pj∈Λ0 z′
jatz′∗
14
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
of gatg−1 is at most 3c2/4. For example, the contribution of the term Pj∈Λ0 zjatz′∗
bounded above in absolute value by
j
is
(4.43)
X
kzjk2kz′∗
j k2 ≤ X
kz′∗
j k2 < c2/4.
j∈Λ0
j∈Λ0
We now examine the contribution from Pj∈Λ0 zjatz∗
j , recalling that each zj is supported on
finitely many group elements. Thus there is an integer K > 0 which bounds the number
of a±1's in any word in the support of zj for any j ∈ Λ0. There are two forms for such
words. The first is ap where p ≤ K while the second is apvaq, where p, q ∈ Z and the
first and last letters of v lie in {b±1
k−1}. Moreover, we will have that p, q ≤ K,
and that v can contain at most K a±1's. We now restrict attention to t ≥ 5K. In order to
avoid a proliferation of cases, we write the typical term in the expansion of Pj∈Λ0 zjatz∗
j as
apvaqata−sw−1a−r, where p, q, s, r ≤ K, each of v and w is either e or begins and ends
with letters from {b±1
k−1} and contains at most K a±1's. We wish to know when this
equals anw0amata−mw−1
0 a−n. This is impossible when both v and w are e,
so we first consider the degenerate case when v = e and the first and last letters of w lie in
{b±1
0 a−n = anw0atw−1
1 , . . . , b±1
1 , . . . , b±1
k−1}. Then the equation
1 , . . . , b±1
ap+q+t−sw−1a−r = anw0atw−1
(4.44)
is false since w0 starts with a letter from {b±1
k−1} while p + q + t − s − n ≥ K. A
similar argument disposes of the possibility that w = e. Thus we need only consider the case
when both v and w begin and end with letters from {b±1
k−1}, giving p = r = n. This
question then reduces to examining the equation
1 , . . . , b±1
1 , . . . , b±1
0 a−n
vat+q−sw−1 = w0atw−1
0 .
(4.45)
The last letter in v either lies in w0 or in w−1
0 . In the second case v must contain at least
w0at, contradicting the bound of K on the number of a's in v since t ≥ 5K. Thus the first
case must hold, and we can write w0 = vw1 where no cancelations can occur between v and
w1. Then (4.45) becomes
at+q−sw−1 = w1atw−1
0 .
1 , . . . , b±1
(4.46)
k−1}, there is no cancelation between at and w−1
Since the last letter of w0 lies in {b±1
0 .
Thus there are two possibilities: w1 = e or w1 ends with a letter from {b±1
k−1}. If the
second one holds then no cancelations occur on the right hand side of (4.46), so w1 begins
with at+q−s, showing that w0 contains this power. But t + q − s ≥ 3K and a contradiction
ensues. Thus w1 = e and w0 = v. Returning to (4.45), we conclude that q = s and w0 = w.
It follows that the contributions to the coefficient of gatg−1 in Pj∈Λ0 zjatz∗
j only arise from
products datd−1 where d has the form anw0aq. Thus this coefficient is a sum of positive
terms including c2 (which arises from the coefficient canw0am term in zΛ0). In conclusion,
for t sufficiently large, the coefficient of gatg−1 in Pj∈Λ yjaty∗
j has absolute value at least
c2 − c2/4 − 3c2/4, and is thus nonzero. Consequently these elements cannot lie in A, and
this contradiction completes the proof.
(cid:3)
1 , . . . , b±1
Theorem 4.8. Let k ≥ 2 and let A be a generator masa in M = LFk. Given any unital
inclusion B ⊆ N of von Neumann algebras, we have
(4.47)
GNM ⊗ N (A ⊗ B)′′ = A ⊗ GNN (B)′′.
GROUPOID NORMALISERS
15
Proof. Take a faithful representation of N on some Hilbert space H. Let v be a groupoid
normaliser of A ⊗ B in M ⊗ N. Then v(A ⊗ C1)v∗ ⊆ A ⊗ B ⊆ A ⊗ B(H), so by Lemma
4.7, v lies in A ⊗ B(H). By applying slice maps, it follows that v ∈ A ⊗ N. Thus
(4.48)
GNM ⊗ N (A ⊗ B)′′ = GNA ⊗ N (A ⊗ B)′′,
as the inclusion from right to left is immediate. Lemma 4.1 then gives
(4.49)
GNM ⊗ N (A ⊗ B)′′ = A ⊗ GNN (B)′′,
and so the result follows.
(cid:3)
5. Irreducible subfactors
Recall that an inclusion B ⊆ M of factors is irreducible if B′ ∩ M = C1. In this section
we consider two irreducible inclusions Bi ⊆ Mi (i = 1, 2) of factors and the resulting tensor
product inclusion B = B1 ⊗ B2 ⊆ M1 ⊗ M2 = M. Our objective is to examine normalisers
of this latter inclusion and to address the question of whether they always factorise in the
form v(w1 ⊗ w2) for some normalisers wi ∈ NMi(Bi) and some unitary v ∈ B.
In [15] a
positive answer was obtained when both M1 and M2 are type II1 using basic construction
methods (the special case of finite index inclusions follows from the earlier paper [11]). Here
we examine the M ′-bimodular automorphism groups used in Sections 3 and 4 to answer this
question beyond the finite setting. To this end, we regard each Mi as faithfully represented on
a Hilbert space Hi so that M is represented on H = H1⊗H2. Tomita's commutation theorem
ensures that the commutants M ′ and B′ are obtained as the tensor products M ′
2 and
B′
2 respectively. Given any unitary normaliser u ∈ NM (B), we obtain an automorphism
θ = Ad(u) ∈ AutM ′(B′). In the setting of irreducible subfactors, these automorphisms must
factorise, as we now show.
1 ⊗ M ′
1 ⊗ B′
Lemma 5.1. With the notation above, let θ ∈ AutM ′(B′). Then there exist automorphisms
θi ∈ AutM ′
i), i = 1, 2, such that θ = θ1 ⊗ θ2.
i (B′
Proof. For y1 ∈ B′
1 and x2 ∈ M ′
2, we have
(1 ⊗ x2)θ(y1 ⊗ 1) = θ((1 ⊗ x2)(y1 ⊗ 1)) = θ((y1 ⊗ 1)(1 ⊗ x2))
= θ(y1 ⊗ 1)(1 ⊗ x2)
(5.1)
so that
(5.2)
θ(y1 ⊗ 1) ∈ (B′
Similarly, θ−1 also maps B′
to give an automorphism of B′
restricts to an automorphism θ2 of B′
x1 ∈ M ′
1 and x2 ∈ M ′
2, we have
2) ∩ (C1 ⊗M ′
1 ⊗ B′
1 ⊗ (B′
1 ⊗C1 into itself, so if we identify B′
2)′ = B′
2 ∩ M2) = B′
1 with B′
1 ⊗C1, then θ restricts
1. By interchanging the roles of B1 and B2, we see that θ also
2, allowing us to conclude that θ = θ1 ⊗ θ2. Given
1 ⊗ C1.
(5.3)
x1 ⊗ x2 = θ(x1 ⊗ x2) = θ1(x1) ⊗ θ2(x2),
so that each θi is M ′
i -bimodular for i = 1, 2.
(cid:3)
The next lemma enables us to recover groupoid normalisers from elements commuting
with bimodular automorphisms.
16
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
Lemma 5.2. Let B ⊆ M be an irreducible inclusion of subfactors, faithfully represented
on some Hilbert space H and let ψ be an element of AutM ′(B′). Let a ∈ B(H) have polar
decomposition a = va and also satisfy
(5.4)
Then aa∗, a∗a ∈ B and v is a groupoid normaliser of B in M satisfying
ψ(y)a = ay,
y ∈ B′.
(5.5)
ψ(y)v = vy,
y ∈ B′.
Proof. Applying (5.4) and its adjoint gives
(5.6)
so that a∗a ∈ B. Similarly
(5.7)
so that aa∗ ∈ B. Now,
(5.8)
which implies that
a∗ay = a∗ψ(y)a = ya∗a,
y ∈ B′,
aa∗ψ(y) = ψ(y)aa∗,
y ∈ B′
ψ(y)va = vay = vya,
y ∈ B′,
(5.9)
since v∗v is the range projection of a (and so lies in B). By hypothesis ψ(x) = x for x ∈ M ′,
so the double commutant theorem ensures that v ∈ M. For b ∈ B and y ∈ B′, we have
ψ(y)v = vyv∗v = vv∗vy = vy,
y ∈ B′
(5.10)
so that vbv∗ ∈ B. Similarly
vbv∗ψ(y) = vbyv∗ = vybv∗ = ψ(y)vbv∗,
(5.11)
so v∗bv ∈ B and we have proved that v ∈ GNM (B).
v∗bvy = v∗bψ(y)v = v∗ψ(y)bv = yv∗bv,
(cid:3)
We briefly recall from [18] some facts about slice maps on tensor products that will be
used subsequently.
If P and Q are two von Neumann algebras then the algebraic tensor
product P∗ ⊗ Q∗ is a norm dense subspace of the predual of P ⊗Q. Thus any fixed φ ∈ Q∗
defines a left slice map Lφ : P ⊗ Q → P by
(5.12)
Lφ(x)(ψ) = (ψ ⊗ φ)(x),
x ∈ P ⊗ Q, ψ ∈ P∗.
This definition shows that each Lφ is a normal map and, when restricted to elementary
tensors, it has the form
(5.13)
Lφ(p ⊗ q) = φ(q)p,
p ∈ P, q ∈ Q.
It is immediate from (5.13) that each Lφ is a P -bimodule map. Moreover, if N is a von
Neumann subalgebra of P and x ∈ P ⊗ Q is such that Lφ(x) ∈ N for all φ ∈ Q∗, then
x ∈ N ⊗ Q. In the next lemma we note that the automorphism θ splits as θ = θ1 ⊗ θ2, by
Lemma 5.1.
Lemma 5.3. For i = 1, 2, let Bi ⊆ Mi be irreducible inclusions of factors. Write M =
M1 ⊗ M2 and B = B1 ⊗ B2 and let u ∈ NM (B). Write θ = θ1 ⊗ θ2 for the induced element
Ad(u) of AutM ′(B′). Then there exist vi ∈ GNMi(Bi), i = 1, 2, such that each vi is either an
isometry or a co-isometry and
(5.14)
θi(yi)vi = viyi,
y ∈ B′
i, i = 1, 2.
GROUPOID NORMALISERS
17
Proof. By symmetry, we only need examine the case i = 1. Consider the set
(5.15)
S1 = {v ∈ GNM1(B1) : θ1(y)v = vy,
y ∈ B′
1},
which we equip with the partial ordering v ≤ w iff wv∗v = v (equivalent to the requirement
that the partial isometry w be an extension of v). It is clear that any chain in S1 has a least
upper bound so, by Zorn's Lemma, there is some maximal element v ∈ S1. Let e = 1 − v∗v
and f = 1 − vv∗. Suppose that v is neither an isometry nor a co-isometry, so that both e
and f are non-zero.
Since u(e ⊗ 1) 6= 0, there is a normal state φ on B(H2) inducing a slice map id ⊗ φ :
If we define a ∈ B(H1) by a =
B(H1) ⊗ B(H2) → B(H1) with (id ⊗ φ)(u(e ⊗ 1)) 6= 0.
(id ⊗ φ)(u(e ⊗ 1)), then the module property of the slice map shows that
(5.16)
Now
(5.17)
a = (id ⊗ φ)(u(e ⊗ 1)) = (id ⊗ φ)(u))e.
(θ1(y) ⊗ 1)u = θ(y ⊗ 1)u = u(y ⊗ 1),
y ∈ B′
1,
so applying id ⊗ φ gives
(5.18)
θ1(y)(id ⊗ φ)(u) = (id ⊗ φ)(u)y,
y ∈ B′
1.
Then multiplication on the right by e ∈ B1 leads to
(5.19)
θ1(y)a = ay,
y ∈ B′
1.
Letting a = v1a be the polar decomposition of a, we obtain v1 ∈ S1 from Lemma 5.2. Define
f1 = v1v∗
1, and note that some non-zero subprojection of f1 is equivalent to a subprojection
of f since B1 is a factor. Let w ∈ B1 be a nonzero partial isometry with w∗w ≤ f1 and
ww∗ ≤ f , and consider the partial isometry wv1. This is a groupoid normaliser of B1 with
(5.20)
θ1(y)wv1 = wv1y,
y ∈ B′
1,
since both v1 and w have these properties. Accordingly, v + wv1 lies in S1 and is strictly
greater than v, contradicting maximality. Hence v must be either an isometry or a co-
isometry.
(cid:3)
In the following corollary we recapture the normaliser results for tensor products of irre-
ducible inclusions of II1 factors from [15].
Corollary 5.4. For i = 1, 2, let Bi ⊆ Mi be irreducible inclusions of II1 factors. Then any
unitary normaliser of B1 ⊗ B2 in M1 ⊗ M2 is of the form w(v1 ⊗ v2) for some unitary
w ∈ B1 ⊗ B2 and normalisers vi ∈ NMi(Bi).
1 ⊗ B′
2)
Proof. Given a normaliser u of B1 ⊗ B2 in M1 ⊗ M2, let θ = Ad(u) ∈ AutM ′
(formed with respect to some faithful actions of M1 and M2 on Hilbert spaces H1 and H2).
By Lemma 5.1, we have a factorisation θ = θ1 ⊗ θ2 for some θi ∈ AutM ′
i). Lemma 5.3
gives groupoid normalisers vi ∈ GNMi(Bi) satisfying
i (B′
(B′
1 ⊗ M ′
2
(5.21)
θi(yi)vi = yivi,
yi ∈ B′
i,
and such that v1 and v2 are either isometries or co-isometries. Since both M1 and M2 are
finite, v1 and v2 are unitary normalisers. Define w = u(v∗
2). Since
1 ⊗ v∗
(5.22)
θ(y1 ⊗ y2)w = u(y1v∗
1 ⊗ y2v∗
2) = u(v∗
1θ1(y1) ⊗ v∗
2θ2(y2)) = wθ(y1 ⊗ y2),
yi ∈ B′
i,
18
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
the double commutant theorem ensures that w ∈ B1 ⊗ B2 and u = w(v1 ⊗ v2), as required.
(cid:3)
We now return to the situation where we do not assume that M1 and M2 are finite.
Lemma 5.5. Let B ⊆ M be an irreducible inclusion of infinite factors and suppose that
ψ ∈ AutM ′(B′) is given and v ∈ GNM (B) is an isometry or co-isometry with
(5.23)
ψ(y)v = vy,
y ∈ B′.
Then there exists a normaliser v1 ∈ NM (B) with
(5.24)
ψ(y)v1 = v1y,
y ∈ B′.
Proof. We may assume that v∗v = 1 and vv∗ = f < 1. The map x 7→ vxv∗ implements a
surjective ∗-isomorphism between B and f Bf . Hence f Bf is an infinite factor so f is infinite
in B. Then there exists a partial isometry w ∈ B with w∗w = f and ww∗ = 1. Define
v1 = wv. This is certainly a unitary normaliser of B and (5.24) follows from multiplying
(5.23) on the left by w (which commutes with ψ(y)).
(cid:3)
Theorem 5.6. For i = 1, 2, let Bi ⊆ Mi be irreducible inclusions of factors. Suppose
that either one of M1 or M2 is finite or that both B1 and B2 are infinite. Then any unitary
normaliser u of B1 ⊗ B2 in M1 ⊗ M2 is of the form w(v1⊗v2) for some unitary w ∈ B1 ⊗ B2
and normalisers vi ∈ NMi(Bi).
Proof. As in the proof of Corollary 5.4, write Ad(u) = θ1 ⊗ θ2 for some θi ∈ AutM ′
find isometries or co-isometries vi ∈ GNMi(Bi) with
i (B′
i), and
(5.25)
θi(yi)vi = yivi,
yi ∈ B′
i.
If both B1 and B2 are infinite, then Lemma 5.5 enables us to replace the vi by unitary
normalisers satisfying the same condition. Just as in Corollary 5.4, w = u(v∗
2) is a
unitary in B1 ⊗ B2 and in this case u has the desired form. Now consider the case where
M1 is finite. Then v1 is automatically a unitary normaliser. If B2 is infinite, then we can
use Lemma 5.5 to replace v2 by a normaliser, and the result follows. If B2 is finite, then
the element w, defined by w = u(v∗
2), lies in B1 ⊗ B2 and is either an isometry or
co-isometry (since u is a unitary). In this case, the tensor product B1 ⊗ B2 is finite, so w,
and hence v2, is a unitary as required.
(cid:3)
1 ⊗ v∗
1 ⊗ v∗
To consider the last remaining case when both inclusions Bi ⊆ Mi consist of an irreducible
inclusion of a II1 factor inside an infinite factor, we use Murray and von Neumann's funda-
mental group from [10]. Let B be a II1 factor with trace τ . Recall that the fundamental
group, F(B), of B is the subgroup of R+ of those quotients τ (p)/τ (q), where p and q are
projections in B such that pBp and qBq are isomorphic. Many subgroups of R+ can oc-
cur, in particular all countable subgroups of R+ [12]. In our situation, it is enough for one
subfactor Bi to have trivial fundamental group; this enables us to convert the isometries or
co-isometries appearing in the proof of Theorem 5.6 into unitary normalisers.
Lemma 5.7. For i = 1, 2, let Bi ⊆ Mi be irreducible inclusions of factors. Suppose that
B1 is finite and has trivial fundamental group. Then any unitary normaliser of B1 ⊗ B2
in M1 ⊗ M2 is of the form w(v1 ⊗ v2) for some unitary w ∈ B1 ⊗ B2 and normalisers
vi ∈ NMi(Bi).
GROUPOID NORMALISERS
19
Proof. As in the proof of Corollary 5.4, write Ad(u) = θ1 ⊗ θ2 for some θi ∈ AutM ′
find isometries or co-isometries vi ∈ GNMi(Bi) with
i (B′
i), and
(5.26)
θi(yi)vi = yivi,
yi ∈ B′
i.
If v1 is an isometry, then x 7→ v1xv∗
1).
As F(B1) = {1}, we have v1v∗
If v1 is a co-isometry, we can
consider x 7→ v∗
1xv1 to see that v1 is again a unitary. The rest of the argument now follows
exactly the proof of Theorem 5.6, considering the cases where B2 is infinite and B2 is finite
separately.
(cid:3)
1 implements an isomorphism of B onto (v1v∗
1 = 1 so v1 is a unitary.
1)B(v1v∗
In Theorem 5.9 we will establish a converse to Lemma 5.7. First fix a finite factor B and
then consider arbitrary irreducible inclusions B ⊆ M. We will show that if the conclusion
of Lemma 5.7 holds for the tensor product with every other separable irreducible inclusion
B2 ⊆ M2, then B has trivial fundamental group.
Recall that if B is a II1 factor with trace τB, then hx, yi = τB(y∗x) defines an inner product
on B, inducing a norm kxk2 = τB(x∗x)1/2 for x ∈ B. The completion is denoted L2(B), we
write ξ for the image of 1 in this Hilbert space, and {xξ : x ∈ B} is a dense subspace. The
representation of B by b(xξ) = (bx)ξ for b, x ∈ B, extends by continuity to a representation
of B on L2(B) and puts B into standard form on this Hilbert space.
Now suppose that B is a II1 factor with non-trivial fundamental group. Fix a projection
e ∈ B with e 6= 0, 1 such that B ∼= eBe and choose an isomorphism θ from B onto eBe. The
trace on eBe is given by τeBe(x) = τB(x)/τB(e) and since this trace is unique, we must have
(5.27)
Thus
(5.28)
τ (x) = τeBe(θ(x)),
x ∈ B.
kτB(e)−1/2θ(x)k2
2 =
τB(θ(x∗x))
τB(e)
= τeBe(θ(x∗x)) = τB(x∗x) = kxk2
2,
x ∈ B.
Therefore, we can define an isometry v on L2(B), by extending
(5.29)
v(xξ) =
1
τ (e)1/2 θ(x)ξ,
x ∈ B,
by continuity. This operator implements θ, in that θ(x) = vxv∗ for all x ∈ B and as θ is a
surjective isomorphism, ebe 7→ v∗bv must be the inverse of θ. Since vv∗ = e, this shows that
v∗Bv = B so v is a groupoid normaliser of B in B(L2(B)). Now let H = L2(B) ⊗ ℓ2(Z).
Let v1 be the bilateral shift operator on ℓ2(Z) satisfying v1en = en+1 for n ∈ Z and define
u = v ⊗ v1 ∈ B(H). Let M be the von Neumann subalgebra of B(H) generated by B ⊗ 1
and u. This algebra resembles a crossed product; indeed it would be B ⋊α Z for the action
induced by Ad(u) if u were a unitary.
Theorem 5.8. In the situation described above, B ⊗ C1 ⊆ M is a singular inclusion of
factors with GNM (B ⊗ 1)′′ = M.
Proof. By construction v is a groupoid normaliser of B with v(B′)v∗ = B′e. Since B′ is
isomorphic to B′e via b′ 7→ b′e, we can define an automorphism φ of B′ so that φ(b′)e = vb′v∗.
Given some b′ ∈ B′, define a bounded operator y = (yi,j)i,j on H with respect to the matrix
units for ℓ2(Z) by setting yi,i = φi(b′) for all i and setting the off-diagonal elements to be
20
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
zero. Certainly y commutes with (B ⊗ C1). Now, for n ∈ Z and η ∈ L2(B),
yu(η ⊗ en) = y(vη ⊗ en+1) = φn+1(b′)evη ⊗ en+1
= vφn(b′)v∗vη ⊗ en+1 = vφn(b′)η ⊗ en+1
= uy(η ⊗ en),
(5.30)
so y commutes with u. The elements y that we are considering form a self-adjoint set, and
so they also commute with u∗. Thus these elements commute with the generators of M and
so lie in M ′.
Now take x ∈ M, let x = (xi,j) with respect to the matrix units of ℓ2(Z). Since 1 ⊗ v1
commutes with B ⊗ C1 and u = v ⊗ v1, 1 ⊗ v1 lies in M ′. Thus x = (1 ⊗ v1)∗x(1 ⊗ v1). This
shows that x has constant diagonal entries, i.e. xi,i = x0,0 for all i. Given b′ ∈ B′, define
y ∈ M ′ as above. Comparing the (0, 0)-th entries of xy = yx, we see that x0,0b′ = b′x0,0 so
that x0,0 = b0 for some b0 ∈ B. For n ≥ 0, consider x(v∗ ⊗ v∗
1)n which lies in M. Thus this
operator has constant entries, say b−n ∈ B, down the diagonal and so xi+n,i = b−nvn for all
i. Similarly we can find bounded operators bn ∈ B for n > 0 such that xi−n,i = (v∗)nbn.
Then the general form of an operator x ∈ M is
(5.31)
x =
X
(b−nvn ⊗ vn
1 ) +
∞
n=0
∞
X
n=1
((v∗)nbn ⊗ (v∗
1)n),
with convergence in the weak operator topology. For n ≥ 1, define en = θ(en) = vn(v∗)n.
Since b−nvn = b−nenvn, we may assume that b−n = b−nen in (5.31) for n > 0. Similarly, we
can assume that bn = enbn for n > 0.
Now suppose that x is a normaliser of B ⊗ C1 in M and write x in the form (5.31). Define
ψ : B → B by ψ(b) ⊗ 1 = x∗(b ⊗ 1)x so that (b ⊗ 1)x = x(ψ(b) ⊗ 1) for b ∈ B. Substituting
this into (5.31) gives
(5.32)
∞
X
n=0
(bb−nvn − b−nvnψ(b)) ⊗ vn
1 +
∞
X
n=1
(b(v∗)nbn − (v∗)nbnψ(b)) ⊗ (v∗
1)n) = 0
for each b ∈ B. Comparing matrix elements in (5.32) leads to
(5.33)
and
(5.34)
bb−nvn = b−nvnψ(b), n ≥ 0, b ∈ B,
b(v∗)nbn = (v∗)nbnψ(b), n > 1, b ∈ B.
After multiplying (5.33) on the right by (v∗)n, we see that
(5.35)
bb−n = b−nθn(ψ(b)), n ≥ 0, b ∈ B.
When b = b∗, we can apply this twice to obtain
−n = b−nθ(ψ(b))b∗
−n = b−nb∗
bb−nb∗
−nb, n ≥ 0, b = b∗ ∈ B,
(5.36)
so that b−nb∗
−n ∈ B′ ∩ B = C1. Thus b−n is a scalar multiple of a unitary in B. Then the
relations b−n = b−nen and en 6= 1 for n > 0 imply that b−n = 0 for n > 0. Replacing x by x∗
shows that bn = 0 for n < 0 and so x = b0 ⊗ 1 ∈ B ⊗ C1. Thus B ⊗ C1 is singular in M, so
(B ⊗ C1)′ ∩ M ⊆ (B ⊗ C1)′ ∩ B ⊗ C1 = C1. In particular, M has trivial centre and so we
have an inclusion of factors. Finally, M is generated by B ⊗ C1 and the groupoid normaliser
u = v ⊗ v1 so that M = GNM (B ⊗ C1)′′.
(cid:3)
GROUPOID NORMALISERS
21
Our final result characterises the property of having trivial fundamental group in terms
of normalisers of tensor products.
Corollary 5.9. Let B be a II1 factor with a separable predual. The following statements are
equivalent.
(i) The fundamental group of B is trivial.
(ii) Whenever B ⊆ M is an inclusion of factors, GNM (B)′′ = NM (B)′′.
(iii) Whenever B ⊆ M is an irreducible inclusion of factors and B2 ⊆ M2 is another
irreducible inclusion of factors, then every unitary normaliser u ∈ NM ⊗ M2(B ⊗ B2)
factorises as u = w(u1 ⊗ u2), where w ∈ B ⊗ B2, u1 ∈ NM (B) and u2 ∈ NM2(B2).
Proof. Lemma 5.7 shows that (i) ⇒ (iii) and the previous theorem establishes (ii) ⇒ (i), so
it remains to show (iii) ⇒ (ii). Take B2 = M2 = B(ℓ2(N)) so that condition (iii) implies that
(5.37)
NM ⊗ B(ℓ2(N))(B ⊗ B(ℓ2(N))) = NM (B)′′ ⊗ B(ℓ2(N)).
Thus GNM (B)′′ = NM (B)′′ by Theorem 3.3.
(cid:3)
References
[1] I. Chifan. On the normalizing algebra of a masa in a II1 factor. arXiv:math.OA/0606225, 2006.
[2] J. Dixmier. Sous-anneaux ab´eliens maximaux dans les facteurs de type fini. Ann. of Math. (2), 59:279 --
286, 1954.
[3] J. Dixmier. von Neumann algebras, volume 27 of North-Holland Mathematical Library. North-Holland
Publishing Co., Amsterdam, 1981.
[4] H. A. Dye. On groups of measure preserving transformations. II. Amer. J. Math., 85:551 -- 576, 1963.
[5] J. Fang. On maximal injective subalgebras of tensor products of von Neumann algebras. J. Funct. Anal.,
244(1):277 -- 288, 2007.
[6] J. Fang. On completely singular von Neumann subalgebras. Proc. Edinb. Math. Soc., 52:607 -- 618, 2009.
[7] J. Fang, R. R. Smith, S. A. White, and A. D. Wiggins. Groupoid normalizers and tensor products. J.
Funct. Anal., 258:20 -- 49, 2010.
[8] L. Ge and R. V. Kadison, On tensor products for von Neumann algebras. Invent. Math., 123:453 -- 466,
1996.
[9] U. Haagerup. The standard form of von Neumann algebras. Math. Scand., 37(2):271 -- 283, 1975.
[10] F. J. Murray and J. von Neumann. On rings of operators. IV. Ann. of Math. (2), 44:716 -- 808, 1943.
[11] M. Pimsner and S. Popa. Entropy and index for subfactors. Ann. Sci. ´Ecole Norm. Sup. (4), 19(1):57 --
106, 1986.
[12] S. Popa. Strong rigidity of II1 factors arising from malleable actions of w-rigid groups, I. Invent. Math.,
165:369 -- 408, 2006.
[13] A. M. Sinclair and R. R. Smith. Finite von Neumann algebras and masas, volume 351 of London
Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 2008.
[14] A. M. Sinclair, R. R. Smith, S. A. White, and A. Wiggins. Strong singularity of singular masas in II1
factors. Illinois J. Math., 51(4):1077 -- 1083, 2007.
[15] R. R. Smith, S. A. White, and A. Wiggins. Normalisers of irreducible subfactors. J. Math. Anal. Appl.,
352:684 -- 695, 2009.
[16] S¸. Stratila and L. Zsid´o. The commutation theorem for tensor products over von Neumann algebras. J.
Funct. Anal., 165(2):293 -- 346, 1999.
[17] M. Takesaki. Tomita's theory of modular Hilbert algebras and its applications. Lecture Notes in Mathe-
matics, Vol. 128, Springer-Verlag, Berlin-New York 1970.
[18] J. Tomiyama. Tensor products and projections of norm one in von Neumann algebras. Unpublished
lecture notes, Copenhagen, 1970.
22
JUNSHENG FANG, ROGER R. SMITH, AND STUART WHITE
Junsheng Fang, Department of Mathematics, Texas A&M University, College Station, Texas,
77843, U.S.A.
E-mail address: [email protected]
Roger R. Smith, Department of Mathematics, Texas A&M University, College Station,
Texas 77843, U.S.A.
E-mail address: [email protected]
Stuart White, Department of Mathematics, University of Glasgow, University Gardens,
Glasgow Q12 8QW, U.K.
E-mail address: [email protected]
|
1303.5545 | 1 | 1303 | 2013-03-22T08:47:33 | Stopping the CCR flow and its isometric cocycles | [
"math.OA",
"math.FA",
"math.PR"
] | It is shown how to use non-commutative stopping times in order to stop the CCR flow of arbitrary index and also its isometric cocycles, i.e., left operator Markovian cocycles on Boson Fock space. Stopping the CCR flow yields a homomorphism from the semigroup of stopping times, equipped with the convolution product, into the semigroup of unital endomorphisms of the von Neumann algebra of bounded operators on the ambient Fock space. The operators produced by stopping cocycles themselves satisfy a cocycle relation. | math.OA | math |
Stopping the CCR flow and its isometric cocycles
Alexander C. R. Belton
Kalyan B. Sinha
Department of Mathematics and Statistics
Lancaster University, United Kingdom
Jawaharlal Nehru Centre for Advanced
Scientific Research, Bangalore, India
[email protected]
[email protected]
21st March 2013
Abstract
It is shown how to use non-commutative stopping times in order to stop the CCR flow of
arbitrary index and also its isometric cocycles, i.e., left operator Markovian cocycles on
Boson Fock space. Stopping the CCR flow yields a homomorphism from the semigroup
of stopping times, equipped with the convolution product, into the semigroup of unital
endomorphisms of the von Neumann algebra of bounded operators on the ambient Fock
space. The operators produced by stopping cocycles themselves satisfy a cocycle relation.
Key words: quantum stopping time; quantum stop time; quantum Markov time; operator cocycle; Markov
cocycle; Markovian cocycle; quantum stochastic cocycle; CCR flow.
MSC 2010: 46L53 (primary); 46L55, 60G40 (secondary).
1
Introduction
Several authors have investigated the use of non-commutative stopping times to stop quantum
stochastic processes, beginning with the pioneering work of Hudson [9], Barnett and Lyons [4],
Parthasarathy and Sinha [13] and Sauvageot [14]; in the framework of Fock-space quantum
stochastic calculus, more recent developments have been produced by Attal and Sinha [3],
Hudson [10] and Coquio [7].
At its most general, a stopping time S is an increasing, time-indexed family (St)t∈[0,∞] of
orthogonal projections in a von Neumann algebra which are subordinate to some filtration.
Below, the ambient von Neumann algebra is B(F), the bounded operators on Boson Fock space
over L2(R+; k), with k an arbitrary complex Hilbert space; the filtration is that generated by
the increasing family of subspaces L2(cid:0)[0, t]; k(cid:1) ⊆ L2(R+; k).
The CCR flow σ is a semigroup (σt)t∈R+ of unital endomorphisms on B(F) which arise from the
isometric right shift on L2(R+; k); CCR flows are fundamental examples of E0 semigroups [2].
A stopped version σS of the CCR flow is constructed for any finite stopping time S, and the
composition of stopped flows is shown to be a homomorphism for the stopping-time convolution
introduced in [13, Section 7].
A cocycle V for the CCR flow σ is a family (Vt)t∈R+ of bounded operators on h ⊗ F, where h
is an arbitrary complex Hilbert space, such that
Vs+t = Vsσs(Vt)
for all s, t ∈ R+;
(1.1)
1
the endomorphism σs is extended to B(h ⊗ F) by ampliation. (Although such cocycles may be
obtained by solving quantum stochastic differential equations, we shall have no use for quantum
stochastic calculus in this work.) It is explained below how to stop such a cocycle V with a
stopping time S whenever the cocycle is isometric, strongly continuous and satisfies a locality
condition. Furthermore, the identity
is shown to hold for all t ∈ R+; this is a non-deterministic generalisation of the relation (1.1).
These results may be viewed as an extension of those obtained by Applebaum [1].
VS+t = VS σS(Vt)
2 Quantum stop times
Notation 2.1. Let FA denote the Boson Fock space over L2(A; k), where A is a subinterval
of R+ and k is a complex Hilbert space which is fixed henceforth. Let F := FR+, Ft) := F[0,t)
and F[t := F[t,∞) for all t ∈ R+, with similar abbreviations for the identity operators I, It)
and I[t on these spaces. Recall the tensor-product decompositions
F ∼= Fs) ⊗ F[s; ε(f ) ↔ ε(f [0,s)) ⊗ ε(f [s,∞))
and
F ∼= Fs) ⊗ F[s,t) ⊗ F[t; ε(f ) ↔ ε(f [0,s)) ⊗ ε(f [s,t)) ⊗ ε(f [t,∞)),
where s, t ∈ R+ are such that s < t and ε(g) is the exponential vector corresponding to g; these
isomorphisms will be used frequently without comment. Let E denote the linear span of the set
of exponential vectors. For further details, see [11] or [12].
Definition 2.2. A quantum stop time is a map
S : B[0, ∞] → B(F),
where B[0, ∞] is the Borel σ-algebra on the extended half-line [0, ∞] := R+ ∪ {∞}, such that
(i) S(A) is an orthogonal projection for all A ∈ B[0, ∞];
(ii) B[0, ∞] → C; A 7→ hx, S(A)yi is a complex measure for all x, y ∈ F;
(iii) S(cid:0)[0, ∞](cid:1) = I;
(iv) S(cid:0){0}(cid:1) ∈ CI and S(cid:0)[0, t](cid:1) ∈ B(Ft)) ⊗ I[t for all t ∈ (0, ∞).
In other words, a quantum stop time is a spectral measure on B[0, ∞] (conditions (i), (ii) and
(iii)) which is identity adapted (condition (iv)).
Remark 2.3. An equivalent way to view a quantum stop time is as an increasing, identity-
adapted family (St)t∈[0,∞] of orthogonal projections on F with S∞ = I. Given a quantum
theorem for self-adjoint operators may be used to move in the opposite direction. Consequently,
stop time in the sense of Definition 2.2, setting St := S(cid:0)[0, t](cid:1) yields such a family; the spectral
a quantum stop time S may be defined by specifying S(cid:0)[0, t](cid:1) for all t ∈ R+.
2
the latter possibility leads to a trivial theory, which corresponds to stopping immediately at 0,
Remark 2.4. Conditions (i) and (iv) in Definition 2.2 imply that S(cid:0){0}(cid:1) is either 0 or I; as
henceforth we shall require that any quantum stop time S is such that S(cid:0){0}(cid:1) = 0. The results
below take their most elegant form when S is finite, i.e., S(cid:0){∞}(cid:1) = 0, so this condition is
imposed as well.
3 Time projections and right shifts
Definition 3.1. Let E0 ∈ B(F) be the orthogonal projection onto Cε(0) and, for all t ∈ (0, ∞),
let
Et : F → F; ε(f ) 7→ ε(1[0,t)f )
be the orthogonal projection onto Ft) ⊗ ε(0[t,∞)), where 1[0,t) is the indicator function of the
interval [0, t). For all t ∈ R+, let θt be the isometric right shift, so that
θt : L2(R+; k) → L2(R+; k); (θtf )(s) = 1[t,∞)(s)f (s − t) =( f (s − t)
0
if s > t,
if 0 < s < t,
and let
be its second quantisation, which has adjoint
Γt : F → F; ε(f ) 7→ ε(θtf )
Γ∗
t : F → F; ε(f ) 7→ ε(θ∗
t f ),
where (θ∗
on F. Let E∞ := I, θ∞ := 0 and Γ∞ := Γ(θ∞) = E0.
t f )(s) = f (s+t) for all s ∈ R+. As is well known, (Γt)t∈R+ and (Γ∗
t )t∈R+ are semigroups
Note that EsΓt = E0 and EtΓs = ΓsEt−s for all s, t ∈ R+ such that s 6 t. Furthermore, the
maps t 7→ Γt and t 7→ Γ∗
t are
strongly continuous on [0, ∞].
t are strongly continuous on [0, ∞), whereas t 7→ Et and t 7→ ΓtΓ∗
Definition 3.2. Let J be a closed subinterval of R+ which contains 0, so that either J = [0, t]
for some t ∈ R+ or J = R+. A map F : J → F is future adapted if and only if
F (s) = ε(0[0,s)) ⊗ Fs ∈ ε(0[0,s)) ⊗ F[s
for all s ∈ J \ {0}.
If J = R+ then F (∞) := ε(0) and x ⊗ F∞ := x for all x ∈ F.
Notation 3.3. For all f , g ∈ L2(R+; k), let Sf,g be the finite Borel measure on [0, ∞] such that
Sf,g(A) =ZA
exp(cid:16)−Z ∞
s
hf (r), g(r)i dr(cid:17)hε(f ), S(ds)ε(g)i
for all A ∈ B[0, ∞].
3
Lemma 3.4. Let S be a finite quantum stop time, let J be a closed subinterval of R+ which
contains 0 and let F : J → F be continuous, bounded and future adapted. For all f ∈ L2(R+; k),
the integral
Z[0,t]
S(ds)ε(f [0,s)) ⊗ Fs = lim
π
n+1Xj=1
S(cid:0)(πj−1, πj](cid:1)ε(f[0,πj)) ⊗ Fπj
(3.1)
is well defined, where t := sup J and the limit is taken over the collection of all finite partitions
π = {0 = π0 < · · · < πn+1 = t}, ordered by refinement. If g ∈ L2(R+; k) and G : J → F is
continuous, bounded and future adapted then
hZ[0,t]
S(ds)ε(f [0,s)) ⊗ Fs,Z[0,t]
S(ds)ε(g[0,s)) ⊗ Gsi =Z[0,t]
hF (s), G(s)i Sf,g (ds),
(3.2)
and if r ∈ (0, ∞] then
S([0, r])Z[0,t]
S(ds)ε(f [0,s)) ⊗ Fs =Z[0,r∧t]
S(ds)ε(f [0,s)) ⊗ Fs,
(3.3)
where r ∧ t is the minimum of r and t.
Proof. Let IS,π(f, F ) denote the sum on the right-hand side of (3.1). Suppose n > 1 and let π′
be a refinement of π; for j = 0, . . . , n, let kj > j be such that π′
= πj and let lj > 1 be such
kj
that π′
= πj+1. Then
kj+lj
IS,π′(f, F ) − IS,π(f, F )
kj+l−1, π′
kj +l)) ⊗ Fπ′
kj +l
=
ljXl=1
n−1Xj=0
S(cid:0)(π′
ln−1Xl=1
S(cid:0)(π′
+
kj +l](cid:1)(cid:0)ε(f [0,π′
kn+l](cid:1)ε(f [0,π′
kn+l−1, π′
kn+l)) ⊗ Fπ′
kn+l
− ε(f [0,π′
kj +lj
)) ⊗ Fπ′
kj +lj(cid:1)
− S(cid:0)(πn, π′
kn+ln−1](cid:1)ε(f [0,t)) ⊗ Ft.
If r, s, u ∈ [0, t] are such that r < s 6 u then
kS(cid:0)(r, s](cid:1)ε(f [0,u)) ⊗ Fuk 6 kS(cid:0)(r, s](cid:1)ε(f )k kF k∞,
where kF k∞ := sup{kF (x)k : x ∈ [0, t]}; if, also, u < ∞ then
kS(cid:0)(r, s](cid:1)(cid:0)ε(f [0,s)) ⊗ Fs − ε(f [0,u)) ⊗ Fu(cid:1)k
6 kS(cid:0)(r, s](cid:1)ε(1[0,s)f )k(cid:0)kFs − ε(0[s,u)) ⊗ Fuk + kε(0[s,u)) − ε(f [s,u))k kFuk(cid:1)
6 kS(cid:0)(r, s](cid:1)ε(f )k(cid:0)kF (s) − F (u)k +(cid:0)kε(f [s,u))k2 − 1(cid:1)1/2kF k∞(cid:1).
4
Consequently,
kIS,π′(f, F ) − IS,π(f, F )k 6 kS(cid:0)[0, πn](cid:1)ε(f )k(cid:0)sup{kF (r) − F (πj)k : r ∈ [πj−1, πj], j = 1, . . . , n}
+ sup{(cid:0)kε(1[πj−1,πj)f )k − 1(cid:1)1/2 : j = 1, . . . , n} kF k∞(cid:1)
+ 2kS(cid:0)(πn, t)(cid:1)ε(f )k kF k∞
and the integral exists. The inner-product identity is readily verified.
For the final claim, note first that S(cid:0)[0, r](cid:1)IS,π(f, F ) = IS,π(f, F ) if r > t; otherwise, suppose
without loss of generality that r = πm+1 for some m > 0. Then S(cid:0)[0, r](cid:1)IS,π(f, F ) = IS,π′(f, F ),
where π′ := π ∩ [0, r]. As π is refined, so is π′ and the result follows.
Corollary 3.5. Let S be a finite quantum stop time. If J is a closed subinterval of R+ which
contains 0 and has supremum t, and F : J → F is Borel measurable, bounded and future adapted
then, for all f ∈ L2(R+; k), there exists a unique vector R[0,t] S(ds)ε(f [0,s)) ⊗ Fs ∈ F such that
(3.2) and (3.3) hold for any g ∈ L2(R+; k), any Borel-measurable, bounded and future-adapted
function G : J → F and any r ∈ (0, ∞].
Proof. For existence, note that F may be approximated in a suitable sense by a sequence of
continuous functions: see [13, Propositions 4.9 and 4.10] for details. Uniqueness holds because
the exponential vector ε(f ) equalsR[0,∞] S(ds)ε(f [0,s)) ⊗ ΓsΓ∗
Remark 3.6. Lemma 3.4 and Corollary 3.5 give the existence of a class of stop-time integrals
which is smaller than that introduced by Parthasarathy and Sinha [13, Section 4] but sufficient
for present purposes.
sε(f ), for any f ∈ L2(R+; k).
Theorem 3.7. Let S be a finite quantum stop time. For all t ∈ (0, ∞], there exists an orthogonal
projection ES,t ∈ B(F) such that
S(ds)ε(f [0,s)) ⊗ ε(0[s,∞))
for all f ∈ L2(R+; k).
S(ds)Esε(f ) :=Z[0,t]
ES,tε(f ) =Z[0,t]
Furthermore, ES,sES,t = ES,s∧t = ES,tES,s and S(cid:0)[0, s](cid:1)ES,∞ = ES,s, for all s ∈ (0, ∞].
j=1 S(cid:0)(πj−1, πj](cid:1)Eπj , where π = {0 = π0 < · · · < πn+1 = t} is a finite
Proof. Let ES,π := Pn+1
partition of [0, t]. Note that, as Es commutes with S(cid:0)(r, s](cid:1) whenever r < s, so ES,π is self
adjoint and idempotent, and therefore an orthogonal projection. Define ES,t on E by using
Lemma 3.4 with F ≡ ε(0) to let
ES,tε(f ) :=Z[0,t]
S(ds)ε(f [0,s)) ⊗ ε(0[s,∞)) = lim
π
ES,πε(f )
for all f ∈ L2(R+; k)
and extending by linearity. Since each ES,π is a contraction, so is ES,t, hence ES,t extends by
continuity to the whole of F; self-adjointness and idempotency hold weakly on E, so everywhere.
5
For the second claim, suppose without loss of generality that s = πm+1 for some m > 1 and
let π′ = {0 = π0 < · · · < πm+1 = s}. Then ES,sES,t = limπ ES,π′ES,π = limπ ES,π′ = ES,s
and S(cid:0)[0, s](cid:1)ES,∞ = limπ S(cid:0)[0, s](cid:1)ES,π = limπ ES,π′ = ES,s, in the weak and strong senses on E
respectively, which gives the result.
Theorem 3.8. Let S be a finite quantum stop time. There exists an isometry ΓS ∈ B(F) such
that
S(ds)ε(0[0,s)) ⊗ Γsx
for all x ∈ F ,
ΓSx =Z[0,∞]
S(ds)Γsx :=Z[0,∞]
where Γs is taken to have range F[s
∼= ε(0[0,s)) ⊗ F[s.
partition of [0, ∞]. For all x, y ∈ F,
Proof. Let ΓS,π :=Pn+1
j=1 S(cid:0)(πj−1, πj](cid:1)Γπj , where π = {0 = π0 < · · · < πn+1 = ∞} is a finite
n+1Xj=1
kS(cid:0)(πj−1, πj](cid:1)ε(0)k2hΓπj x, Γπj yi = hx, yi,
hΓπj x, S(cid:0)(πj−1, πj](cid:1)Γπj yi =
so ΓS,π is an isometry. Since ΓS,πx converges to R[0,∞] S(ds)ε(0[0,s)) ⊗ Γsx as π is refined,
applying Lemma 3.4 with J = R+, f = 0 and F = Γx : s 7→ Γsx gives the claim.
hΓS,πx, ΓS,πxi =
n+1Xj=1
Definition 3.9. Let S be a finite quantum stop time, let ES := ES,∞ and let im T denote the
range of the linear operator T . The pre S space
FS) := im ES = [t∈(0,∞]
im ES,t,
and the post S space F[S := im ΓS, with identity operators IS) and I[S respectively.
deterministic, i.e., S(cid:0){t}(cid:1) = I for some t ∈ (0, ∞), then F[S = ε(0[0,t)) ⊗ F[t, which, as noted
above, may naturally be identified with F[t.
If S is
Theorem 3.10. Let S be a finite quantum stop time. There exists an isometric isomorphism
jS : FS) ⊗ F[S → F such that
jS(cid:0)ES,tε(f ) ⊗ ΓSx(cid:1) =Z[0,t]
S(ds)ε(f [0,s)) ⊗ Γsx for all t ∈ (0, ∞], f ∈ L2(R+; k) and x ∈ F .
Proof. Let t ∈ (0, ∞], f , g ∈ L2(R+; k) and x, y ∈ F. Lemma 3.4 and Theorem 3.7 imply that
hZ[0,t]
S(ds)ε(1[0,s)f ) ⊗ Γsx,Z[0,t]
S(ds)ε(1[0,s)g) ⊗ Γsyi =Z[0,t]
hΓsx, Γsyi Sf,g(ds)
= hES,tε(f ), ES,tε(f ′)ihx, yi.
Hence jS is well defined and extends to an isometry from FS) ⊗ F[S to F.
6
To conclude, it suffices to show that ε(f ) ∈ im jS for any f ∈ L2(R+; k). Let π′ be a refinement
of the partition π = {0 = π0 < · · · < πn+1 = ∞} and, for j = 0, . . . , n, let kj > j be such
that π′
kj
= πj and let lj > 1 be such that π′
= πj+1. Then
kj+lj
ε(f ) =
nXj=0
ljXl=1
S(cid:0)(π′
kj+l−1, π′
kj +l](cid:1)ε(1[0,π′
kj +l)f + θπ′
kj +l
θ∗
π′
kj +l
f )
and, as π′ is refined with π fixed,
nXj=0
ljXl=1
S(cid:0)(π′
kj+l−1, π′
kj +l](cid:1)(cid:0)ε(1[0,π′
kj +l)f + θπ′
kj +l
θ∗
π′
kj +l
f ) − ε(1[0,π′
kj +l)f + θπ′
kj +l
θ∗
π′
kj +lj
→ Rπ := ε(f ) −
nXj=0
jS(cid:0)(ES,πj+1 − ES,πj )ε(f ) ⊗ ΓSε(θ∗
f )(cid:1)
πj+1 f )(cid:1);
f ) is continuous on [0, πj+1] for j = 0, . . . , n. If r, s, u ∈ R+ are such
note that s 7→ Γsε(θ∗
that r < s 6 u then
πj+1
kS(cid:0)(r, s](cid:1)(cid:0)ε(1[0,s)f + θsθ∗
s f ) − ε(1[0,s)f + θsθ∗
uf )(cid:1)k
and, for all h > 0,
6 kS(cid:0)(r, s](cid:1)ε(f )k kε(1[s,∞)f ) − ε(cid:0)1[s,∞)f (· + u − s)(cid:1)k
kε(1[s,∞)f ) − ε(cid:0)1[s,∞)f (· + h)(cid:1)k2 6 η(h) := 2 exp(cid:0)kf k2(cid:1)(cid:0)exp(cid:0)kf k kf − f (· + h)k(cid:1) − 1(cid:1).
Letting δπ := max{πj − πj−1 : j = 1, . . . , n} and refining π, it follows that
kRπk 6 kS(cid:0)[0, πn](cid:1)ε(f )k sup{η(h)1/2 : 0 6 h 6 δπ} + 2kS(cid:0)(πn, ∞)(cid:1)ε(f )k → 0.
Remark 3.11. Use of the time projection ES goes back at least as far as [5, Theorem 2.3],
and the existence of the stopped right shift ΓS was first established in [13, Theorem 5.1]. The
abstract strong Markov property, Theorem 3.10, is [13, Theorem 5.2].
4 Stop-time convolution
Definition 4.1. Let S and T be finite quantum stop times. There exists a unique spectral
measure S ⊗ T on B([0, ∞]2) such that
(S ⊗ T )(A × B) = jS(S(A) ⊗ ΓST (B)Γ∗
S)j∗
S
for all A, B ∈ B[0, ∞],
where S(A) is considered to act on FS), so setting
(S ⋆ T )(C) := (S ⊗ T )(cid:0){(s, t) ∈ [0, ∞] : s + t ∈ C}(cid:1)
gives a finite spectral measure S ⋆ T , their convolution.
for all C ∈ B[0, ∞]
The convolution of spectral measures was introduced by Fox [8] and first applied to quantum
stop times in [13, Proposition 7.1].
7
Theorem 4.2. The convolution of finite quantum stop times S and T is such that
for all u ∈ [0, ∞] and f ∈ L2(R+; k). Consequently, S ⋆ T is a finite quantum stop time such
that
S(ds)ε(f 1[0,s)) ⊗ ΓsT(cid:0)[0, u − s](cid:1)Γ∗
sε(f )
(4.1)
(S ⋆ T )(cid:0)[0, u](cid:1)ε(f ) =Z[0,u]
ES⋆T ε(f ) =Z[0,∞]
S(ds)ε(f [0,s)) ⊗ ΓsET Γ∗
sε(f )
and
for all f ∈ L2(R+; k) and y ∈ F .
S(ds)ε(f [0,s)) ⊗ ΓsZ[0,∞]
jS⋆T (ES⋆T ε(f ) ⊗ ΓS⋆T y) =Z[0,∞]
Remark 4.3. For any f ∈ L2(R+; k) and y ∈ F, the integral R[0,∞] T (dt)ε(f (· + s)[0,t)) ⊗ Γty
is the limit of a sequence of Riemann sums which depend continuously on s. Furthermore,
the norm of these integrals is uniformly bounded above, by kε(f ) ⊗ yk, so the iterated integral
in (4.2) is well defined.
T (dt)ε(f (· + s)[0,t)) ⊗ Γty
(4.2)
Proof of Theorem 4.2. Let f , g ∈ L2(R+; k) and let u ∈ (0, ∞); the cases with u = 0 and u = ∞
are clear. The proof of Theorem 3.10 gives that
ε(f ) = lim
π
n+1Xj=1
jS(cid:0)(ES,πj − ES,πj−1)ε(f ) ⊗ ΓSε(θ∗
πj f )(cid:1),
where the limit is taken over finite partitions π = {0 = π0 < · · · < πn+1 = ∞} ordered by
refinement; without loss of generality, u = πm+1 for some m which depends on π. Furthermore,
whenever 0 6 s < t 6 ∞
f ),(cid:0)S(ds) ⊗ T (dt)(cid:1)(ES,πk − ES,πk−1)ε(g) ⊗ ε(θ∗
πk
g)i
πj ε(f ), T (dt)Γ∗
πk
ε(g)i
πj ε(g)i
=
ES,t − ES,s = ESS(cid:0)(s, t](cid:1) = S(cid:0)(s, t](cid:1)ES
h(ES,πj − ES,πj−1)ε(f ) ⊗ ε(θ∗
πj
It follows that hε(f ), (S ⋆ T )(cid:0)[0, u](cid:1)ε(g)i is the limit, as π as refined, of
ZZ06s+t6u
n+1Xj,k=1
n+1Xj,k=1Z[0,u]
m+1Xj=1Z(πj−1,πj]
m+1Xj=1Z(πj−1,πj]
→Z[0,u]
hS(cid:0)(πj−1, πj](cid:1)ESε(f ), S(cid:0)[0, u − t](cid:1)S(cid:0)(πk−1, πk](cid:1)ESε(g)ihΓ∗
hε(f ), S(ds)S(cid:0)(πj−1, πj](cid:1)ESε(g)ihε(f ), Γπj T(cid:0)[0, u − s](cid:1)Γ∗
hε(f ), S(ds)ES ε(g)ihε(f ), Γπj T(cid:0)[0, u − s](cid:1)Γ∗
hε(f ), S(ds)ES ε(g)ihε(f ), ΓsT(cid:0)[0, u − s](cid:1)Γ∗
πj ε(g)i
sε(g)i.
=
=
8
If F : [0, u] → C is continuous then
Z[0,u]
hε(f ), S(ds)ES ε(g)iF (s) = lim
π
hε(f ), S(cid:0)(πj−1, πj](cid:1)Eπj ε(g)iF (πj ) =Z[0,u]
m+1Xj=1
Sf,g(ds)F (s),
so (4.1) now follows from Lemma 3.4 and the proof of Corollary 3.5. Furthermore, as S and T
are identity adapted, so is S ⋆ T : note that if s, u ∈ R+ are such that s 6 u then
sε(1[0,u)g)ihε(1[u,∞)f ), ε(1[u,∞)g)i.
hε(f ), ΓsT(cid:0)[0, u − s](cid:1)Γ∗
Next, note that
hε(f ), ES⋆T,πε(g)i =
sε(g)i = hε(1[0,u)f ), ΓsT(cid:0)[0, u − s](cid:1)Γ∗
n+1Xj=1Z[0,πj]
=Z[0,∞]
hε(f ), ΓsET,π−sΓ∗
sε(g)i Sf,g(ds),
hε(f ), ΓsT(cid:0)[(πj−1 − s)+, πj − s](cid:1)Eπj −sΓ∗
sε(g)i Sf,g(ds)
where x+ := max{x, 0} for all x ∈ R and π − s is the finite partition of [0, ∞] consisting of
intervals with end points {(π0 − s)+, · · · , (πn − s)+, ∞}; the integral form of ES⋆T ε(f ) now
follows after refining π. For the last identity, note that
hε(f ), jS⋆T (ES⋆T ε(g) ⊗ ΓS⋆T y)i
= lim
= lim
π
n+1Xj=1Z[0,πj]
π Z[0,∞]
π Z[0,∞]
= hε(f ),Z[0,∞]
= lim
hΓsΓ∗
sε(f ), ΓsT(cid:0)[(πj−1 − s)+, πj − s](cid:1)Γ∗
sε(g)ihε(f ), Γπj yiSf,g(ds)
hε(f ), Γs(cid:16)n+1Xj=1
1[0,πj](s)T(cid:0)[(πj−1 − s)+, πj − s](cid:1)ε(g(· + s)[0,πj −s)) ⊗ Γπj −sy(cid:17)i
hf (r + s), g(r + s)i dr(cid:17)Sf,g(ds)
× exp(cid:16)−Z ∞
πj −s
hε(f ), ΓsIT,π−s(g(· + s), Γy)iSf,g(ds)
S(ds)ε(g[0,s)) ⊗ ΓsZ[0,∞]
T (dt)ε(g(· + s)[0,t)) ⊗ Γtyi,
as required, where IT,π−s is as in the proof of Lemma 3.4 and (Γy)(t) := Γty for all t ∈ [0, ∞].
Remark 4.4. The identity (4.1) may also be used to show that
ΓS⋆T x =Z[0,∞]
S(ds)ε(0[0,s)) ⊗ ΓsET x = ΓS(cid:0)ΓT x(cid:1)
for all x ∈ F
and for any finite quantum stop times S and T , so that ΓS⋆T = ΓS ◦ ΓT . As we shall have no
use for this result [13, Proposition 7.6], its proof is left as an exercise.
9
(S + t)(cid:0)[0, u](cid:1) =(
Proposition 4.5. Let S and T be finite quantum stop times and suppose T is deterministic,
i.e., T(cid:0){t}(cid:1) = I for some t ∈ (0, ∞). Then S ⋆ T = S + t, where
0
if u ∈ [0, t),
S(cid:0)[0, u − t](cid:1) if u ∈ [t, ∞].
Proof. If u < t then T(cid:0)[0, u − s](cid:1) = 0 for all s ∈ [0, u] and therefore
(S ⋆ T )(cid:0)[0, u](cid:1)ε(f ) =Z[0,u]
If u > t then T(cid:0)[0, u − s](cid:1) equals I for all s ∈ [0, u − t] and equals 0 for all s ∈ (u − t, u], so
(S ⋆T )(cid:0)[0, u](cid:1)ε(f ) =Z[0,u−t]
S(ds)ε(f [0,s]) ⊗ ΓsT(cid:0)[0, u − s](cid:1)Γ∗
sε(f ) = S(cid:0)[0, u−t](cid:1)ε(f )
S(ds)ε(f [0,s])⊗ΓsΓ∗
sε(f ) = 0 for all f ∈ L2(R+; k).
for all f ∈ L2(R+; k),
where the second equality holds by (3.3).
Theorem 4.6. (Cf. [13, Theorem 7.5]) Let S and T be finite quantum stop times. Then
jS,T : FS) ⊗ ΓS(FT )) ⊗ F[S⋆T → F; ESx ⊗ ΓSET y ⊗ ΓS⋆T z 7→ jS(cid:0)ESx ⊗ ΓSjT (ET y ⊗ ΓT z)(cid:1)
is an isometric isomorphism such that
jS,T(cid:0)FS) ⊗ ΓS(FT )) ⊗ ε(0)(cid:1) = jS(cid:0)FS) ⊗ ΓS(FT ))(cid:1) = FS⋆T )
jS,T(cid:0)ε(0) ⊗ ΓS(FT )) ⊗ F[S⋆T(cid:1) = F[S.
(4.3)
(4.4)
jS,T = jS⋆T ◦ (jS(cid:12)(cid:12)FS)⊗ΓS (FT )) ⊗ I[S⋆T ).
and
Furthermore,
Proof. Note first that
jS,T = jS ◦ (IS) ⊗ ΓS) ◦ (IS) ⊗ jT ) ◦ (IS) ⊗ Γ∗
S ⊗ ΓT Γ∗
S⋆T );
as each of the maps on the right-hand side is an isometric isomorphism between the appropriate
spaces, so is jS,T . Next, let f , g ∈ L2(R+; k) and y ∈ F; it follows from Theorem 4.2 that
hES⋆T ε(f ), jS (ESε(g) ⊗ ΓSy)i =Z[0,∞]
=Z[0,∞]
hΓsET Γ∗
sε(f ), ΓsyiSf,g(ds)
hΓsΓ∗
sε(f ), ΓsET yiSf,g(ds)
= hε(f ), jS (ESε(g) ⊗ ΓSET y)i.
10
Thus
ES⋆T jS(ESx ⊗ ΓSy) = jS (ESx ⊗ ΓSET y) = jS,T(cid:0)ESx ⊗ ΓSET y ⊗ ε(0)(cid:1)
as jS is surjective, (4.3) follows. That (4.4) holds is an immediate consequence of the identity
for all x, y ∈ F;
jS,T (ε(0) ⊗ ΓSET y ⊗ ΓS⋆T z) = ΓSjT (ET y ⊗ ΓT z)
for all y, z ∈ F
and the surjectivity of jT .
Finally, if f , g, h ∈ L2(R+; k) and y, z ∈ F then
jS,T (ES ε(f ) ⊗ ΓSET ε(g) ⊗ ΓS⋆T y) = jS(cid:0)ESε(f ) ⊗ ΓSjT (ET ε(g) ⊗ ΓT y)(cid:1)
S(ds)ε(f [0,s)) ⊗ ΓsZ[0,∞]
=Z[0,∞]
T (dt)ε(g[0,t)) ⊗ Γty,
so Theorem 4.2 implies that
hjS,T (ES ε(f ) ⊗ ΓSET ε(g) ⊗ ΓS⋆T y), jS⋆T (ES⋆T ε(h) ⊗ ΓS⋆T z)i
T (dt)ε(h(· + s)[0,t)) ⊗ Γtz(cid:11) Sf,h(ds)
hy, zi T g,h(·+s)(dt) Sf,h(ds)
T (dt)ε(g[0,t)) ⊗ Γty,Z[0,∞]
=Z[0,∞](cid:10)Z[0,∞]
=Z[0,∞]Z[0,∞]
=Z[0,∞]
= hjS⋆T (jS(cid:0)ESε(f ) ⊗ ΓSET ε(g)(cid:1) ⊗ ΓS⋆T y), jS⋆T (ES⋆T ε(h) ⊗ ΓS⋆T z)i.
= hjS(ES ε(f ) ⊗ ΓSET ε(g)), ε(h)ihy, zi
hET ε(g), ET Γ∗
sε(h)i Sf,h(ds)hy, zi
5 The CCR flow
Definition 5.1. For all t ∈ R+, define a unital ∗-homomorphism
σt : B(F) → B(F); X 7→ It) ⊗ ΓtXΓ∗
t ,
where F[t is regarded as an isometric isomorphism from F onto F[t with inverse Γ∗
has range im σt = It) ⊗ B(F[t), that σs(Et) = Es+t for all s ∈ R+ and that
t . Note that σt
σt(X)Γt = ΓtX
for all X ∈ B(F).
(5.1)
Let σ∞ be the constant map on B(F) with value I.
As is well known, (σt)t∈R+ is a semigroup on B(F) called the CCR flow.
11
Theorem 5.2. Let S be a finite quantum stop time and let π = {0 = π0 < · · · < πn+1 = ∞} be
a finite partition of [0, ∞]. The map
σS,π : B(F) → B(F); X 7→
n+1Xj=1
σπj (X)S(cid:0)(πj−1, πj](cid:1)
is a unital ∗-homomorphism, the limit σS(X) := st.limπ σS,π(X) exists for all X ∈ B(F) and
the map
σS : B(F) → B(F); X 7→ σS(X)
is a unital ∗-homomorphism such that
σS(X)ΓS = ΓSX
for all X ∈ B(F).
(5.2)
Proof. Since
σs(X) ∈ Is) ⊗ B(F[s)
and
S(cid:0)(r, s](cid:1) ∈ B(Fs)) ⊗ I[s
for all r, s ∈ R+ with r < s, the first claim is immediate. For the next, suppose n > 1, let π′
be a refinement of π and, for j = 0, . . . , n, let kj > j be such that π′
= πj and let lj > 1 be
kj
such that π′
= πj+1. Then
kj+lj
(cid:0)σS,π′ − σS,π(cid:1)(X)ε(f ) =
ljXl=1
n−1Xj=0
S(cid:0)(π′
ln−1Xl=1
S(cid:0)(π′
+
kj+l](cid:1)(σπ′
kn+l](cid:1)σπ′
kj +l−1, π′
− σπ′
kj +lj
)(X)ε(f )
kj +l
If r, s, t ∈ R+ are such that r < s 6 t then
kn+l−1, π′
kn+ln−1](cid:1)ε(f ).
kS(cid:0)(r, s](cid:1)(cid:0)σt(X) − σs(X)(cid:1)ε(f )k = kS(cid:0)(r, s](cid:1)ε(1[0,s)f )k k(cid:0)σt(X) − σs(X)(cid:1)ε(1[s,∞)f )k
(X)ε(f ) − S(cid:0)(πn, π′
kn+l
6 kS(cid:0)(r, s](cid:1)ε(f )k kσs(σt−s(X) − X)ΓsΓ∗
= kS(cid:0)(r, s](cid:1)ε(f )k k(σt−s(X) − X)ε(cid:0)f (· + s)(cid:1)k,
sε(f )k
by (5.1); furthermore,
kS(cid:0)(r, s](cid:1)σs(X)ε(f )k 6 kS(cid:0)(r, s](cid:1)ε(f )k kXk.
Hence if f has support contained in [0, T ] ⊆ [0, πm] and δπ := max{πj − πj−1 : j = 1, . . . , m}
then
k(cid:0)σS,π′ − σS,π(cid:1)(X)ε(f )k
6 kS(cid:0)[0, πn](cid:1)ε(f )k sup{k(σr(X) − X)ε(cid:0)f (· + s)(cid:1)k : r ∈ [0, δπ], s ∈ [0, T ]}
+ kS(cid:0)(πn, ∞)(cid:1)ε(f )k(kXk + 1);
12
since t 7→ σt(X) is strongly continuous on F for all X ∈ B(F), this estimate gives the existence
of σS(X)ε(f ): note that s 7→ ε(cid:0)f (· + s)(cid:1) is uniformly continuous on [0, T ] and
sup{k(σr(X) − X)ε(cid:0)f (· + s)(cid:1)k : r ∈ [0, δ], s ∈ [0, T ]}
6 sup{k(σr(X) − X)ε(cid:0)f (· + sj)(cid:1)k : r ∈ [0, δ], j = 0, . . . , N }
+ 2kXk sup{kε(cid:0)f (· + s)(cid:1) − ε(cid:0)f (· + sj)(cid:1)k : s ∈ [sj, sj+1], j = 0, . . . , N },
where {0 = s0 < · · · < sN +1 = T } is any partition of [0, T ]. Thus σS(X)x := limπ σS,π(X)x
exists if x is any finite linear combination of exponential vectors corresponding to functions
with compact support; since kσS,π(X)k 6 kXk for all π, it follows that x 7→ σS(X)x extends
to a bounded operator σS(X) on the whole of F; a simple approximation argument now shows
that σS(X) is the limit of σS,π(X) in the strong operator topology.
The map σS is ∗-homomorphic and unital because each σS,π is. For the final identity, observe
that
σS,π(X)
n+1Xk=1
S(cid:0)(πk−1, πk](cid:1)Γπk =
n+1Xj=1
S(cid:0)(πj−1, πj](cid:1)σπj (X)Γπj =
n+1Xj=1
S(cid:0)(πj−1, πj](cid:1)Γπj X,
by (5.1), so refining π gives the claim.
Proposition 5.3. Let S be a finite quantum stop time. Then
σS(X) = jS(IS) ⊗ ΓSXΓ∗
S)j∗
S
for all X ∈ B(F),
where jS : FS) ⊗ F[S → F is the isometric isomorphism of Theorem 3.10.
Proof. Let π = {0 = π0 < · · · < πn+1 = ∞} be a finite partition containing t = πm ∈ (0, ∞)
and let f , f ′, g, g′ ∈ L2(R+; k). If r, s ∈ R+ are such that r < s then
With the notation used in the proofs of Lemma 3.4 and Theorems 3.7 and 3.8, it follows that
hε(1[0,s)f + θsg), S(cid:0)(r, s](cid:1)σs(X)ε(1[0,s)f ′ + θsg′)i = hε(1[0,s)f ), S(cid:0)(r, s](cid:1)ε(1[0,s)f ′)ihε(g), Xε(g′)i.
hIS,π∩[0,t](cid:0)f, Γε(g)(cid:1), σS,π(X)IS,π∩[0,t](cid:0)f ′, Γε(g′)(cid:1)i = hES,π∩[0,t]ε(f ), ES,π∩[0,t]ε(f ′)ihε(g), Xε(g′)i
hjS(cid:0)ES,tε(f ) ⊗ ΓSε(g)(cid:1), σS(X)jS(cid:0)ES,tε(f ′) ⊗ ΓSε(g′)(cid:1)i
= hES,tε(f ) ⊗ ΓSε(g), (IS) ⊗ ΓSXΓ∗
and therefore, by refining π,
S)(cid:0)ES,tε(f ′) ⊗ ΓSε(g′)(cid:1)i,
as required.
Theorem 5.4. The map S 7→ σS is a homomorphism from the convolution semigroup of finite
quantum stop times to the semigroup of unital endomorphisms of B(F) : for any finite quantum
stop times S and T ,
and, furthermore, σS(ET ) = ES⋆T .
σS⋆T = σS ◦ σT
13
Proof. Let X ∈ B(F). Proposition 5.3 and Theorem 4.6 imply that
σS(cid:0)σT (X)(cid:1) = jS (IS) ⊗ ΓSjT (IT ) ⊗ ΓT XΓ∗
T )j∗
T Γ∗
S)j∗
S = jS,T (IS) ⊗ IΓS(FT )) ⊗ ΓS⋆T XΓ∗
S⋆T )j∗
S,T
= jS⋆T (IS⋆T ) ⊗ ΓS⋆T XΓ∗
S⋆T )j∗
S⋆T
The second identity follows from this; with the notation used in the proof of Theorem 3.7,
= σS⋆T (X).
and therefore
σS(E0) = st.lim
π
σS,π(E0) = st.lim
π
ES,π = ES
σS(ET ) = σS(cid:0)σT (E0)(cid:1) = σS⋆T (E0) = ES⋆T .
6
Isometric cocycles
Notation 6.1. Let M ⊆ B(h) be a von Neumann algebra, where h is a complex Hilbert space,
and let ⊗ denote the ultraweak tensor product. Henceforth, S is a finite quantum stop time
extended by ampliation to act on h ⊗ F; the time projection ES extends by ampliation in the
same manner.
Definition 6.2. Let p be an orthogonal projection in B(k) which acts pointwise on L2(cid:0)[t, ∞); k(cid:1)
for all t ∈ R+, so that
(pf )(s) := pf (s)
for all s ∈ [t, ∞) and f ∈ L2(cid:0)[t, ∞); k(cid:1),
and let P[t ∈ B(F[t) be the second quantisation of p acting in this way, so that P[tε(f ) = ε(pf )
for all f ∈ L2(cid:0)[t, ∞); k(cid:1).
A family of operators V = (Vt)t∈R+ ⊆ M ⊗ B(F) is a p-adapted process if and only if
Vt = Vt) ⊗ P[t
for all t ∈ R+,
(6.1)
where Vt) ∈ M ⊗ B(Ft)).
If p = Ik then this is identity adaptedness, the usual choice for
Hudson -- Parthasarathy quantum stochastic calculus; if p = 0 then this is vacuum adaptedness
[6]. Note that V is vacuum adapted if and only if Vt = EtVtEt for all t ∈ R+.
Given a p-adapted process V , its identity-adapted projection is the process bV =(cid:0)bVt(cid:1)t∈R+
for all t ∈ R+.
, where
A p-adapted process V is a cocycle (more fully, a left operator Markovian cocycle) if and only if
where σ is the CCR flow of Definition 5.1, extended by ampliation.
for all s, t ∈ R+,
(6.2)
bVt := Vt) ⊗ I[t
Vs+t = bVs σs(Vt)
14
A p-adapted cocycle V is isometric if V ∗
t)Vt) = Ih ⊗ It) for all t ∈ R+, where Vt) is as in (6.1);
equivalently, bV ∗
and
t bVt = Ih⊗I for all t ∈ R+. For such an isometric cocycle V , it holds that kVtk = 1
V ∗
t Vt = Ih ⊗ It) ⊗ P[t ∈ Ih ⊗ It) ⊗ B(F[t)
for all t ∈ R+.
Example 6.3. Let W (f ) ∈ B(F) be the Weyl operator corresponding to f ∈ L2(R+; k), so
that
W (f )W (g) = exp(cid:0)−ihf, gi(cid:1)ε(f + g)
W (f )ε(g) = exp(cid:0)− 1
for all f , g ∈ L2(R+; k). Then(cid:0)Wt(c) := Ih ⊗ W (1[0,t)c)(cid:1)t>0 and(cid:0)Vt(c) := Ih ⊗ EtW (1[0,t)c)(cid:1)t>0
2 kf k2 − hf, gi(cid:1)ε(f + g)
are isometric cocycles for all c ∈ k, which are 1-adapted and 0-adapted respectively.
and
More examples of isometric cocycles may be constructed as the solutions of quantum stochastic
differential equations: see [11, Section 5] and references therein.
Lemma 6.4. Let V be an isometric p-adapted cocycle and let t ∈ (0, ∞). For any finite partition
π = {0 = π0 < · · · < πn+1 = t}, the Riemann sum
is such that
Vπj S(cid:0)(πj−1, πj](cid:1)
VS,π :=
n+1Xj=1
kVS,πzk 6 kS(cid:0)[0, t](cid:1)zk
for all z ∈ h ⊗ F .
Proof. If r, s ∈ R+ are such that r 6 s then, by the cocycle, p-adaptedness and isometry
properties,
V ∗
s Vr = σr(V ∗
s−r) (V ∗
r) ⊗ I[r)(Vr) ⊗ P[r) ∈ M ⊗ Ir) ⊗ B(F[r)
and
V ∗
r Vs = (V ∗
r) ⊗ P[r)(V ∗
r) ⊗ I[r) σr(Vs−r) ∈ M ⊗ Ir) ⊗ B(F[r).
(6.3)
(6.4)
Consequently, if j, k ∈ {1, . . . , n + 1} are such that j 6= k then, as S is identity adapted,
and thus, for all z ∈ h ⊗ F, we have that
S(cid:0)(πj, πj+1](cid:1)V ∗
πj+1
Vπk+1S(cid:0)(πk, πk+1](cid:1) = 0,
kVS,πzk2 =
n+1Xj=1
k(Ih ⊗ Iπj) ⊗ P[πj )S(cid:0)(πj−1, πj](cid:1)zk2 6
n+1Xj=1
kS(cid:0)(πj−1, πj](cid:1)zk2 = kS(cid:0)[0, t](cid:1)zk2.
Theorem 6.5. Let V be an isometric p-adapted cocycle and let t ∈ (0, ∞). If V is strongly
continuous then
is well defined on h ⊗ F . Furthermore, VS,t = VS,tS(cid:0)[0, t](cid:1) and
for all z ∈ h ⊗ F .
VS,t := st.lim
VS,π
π
kVS,tzk 6 kS(cid:0)[0, t](cid:1)zk
15
Proof. Let π′ be a refinement of π = {0 = π0 < · · · < πn+1 = t} and, for j = 0, . . . , m,
let kj > j be such that π′
kj
= πj and let lj > 1 be such that π′
= πj+1. Then
kj+lj
VS,π′ − VS,π =
nXj=0
ljXl=1
(Vπ′
kj +l
− Vπ′
kj +lj
)S(cid:0)(π′
kj +l−1, π′
kj+l](cid:1).
If a, b, c, p, q, r ∈ R+ are such that a < b 6 c and p < q 6 r then, by (6.3) and (6.4),
(Vc − Vb)∗(Vr − Vq) = V ∗
c Vr − V ∗
c Vq − V ∗
b Vr + V ∗
b Vq ∈ B(h) ⊗ Ib∧q) ⊗ B(F[b∧q)
and therefore
S(cid:0)(a, b](cid:1)(Vc − Vb)∗(Vr − Vq)S(cid:0)(p, q](cid:1) =( (Vc − Vb)∗(Vr − Vq)S(cid:0)(a, b](cid:1)S(cid:0)(p, q](cid:1)
S(cid:0)(a, b](cid:1)S(cid:0)(p, q](cid:1)(Vc − Vb)∗(Vr − Vq)
huε(f ), ZS(cid:0)(p, q](cid:1)uε(f )i = huε(1[q,∞)f ), (X ⊗ Y )uε(1[q,∞)f )i hε(1[0,q)f ), S(cid:0)(p, q](cid:1)ε(1[0,q)f )i
Furthermore, if Z = X ⊗ Iq) ⊗ Y , where X ∈ M and Y ∈ B(F[q), then
if b 6 q,
if q 6 b.
= huε(f ), Zuε(f )i hε(f ), S(cid:0)(p, q](cid:1)ε(f )i kε(f )k−2
This gives the first claim on the algebraic tensor product h ⊙ E. It follows from Lemma 6.4
that VS,t is a contraction on h ⊙ E, so it extends to a bounded operator on the whole of h ⊗ F,
and an approximation argument now gives strong convergence everywhere.
The second claim holds because VS,πS(cid:0)[0, t](cid:1) = VS,π; the third is an immediate consequence of
Lemma 6.4.
Corollary 6.6. If the isometric p-adapted cocycle V is strongly continuous then
is well defined.
VS = VS,∞ := st.lim
t→∞
VS,t
16
and therefore
Hence
k(VS,π′ − VS,π)uε(f )k2 =
k(Vr − Vq)S(cid:0)(p, q](cid:1)uε(f )k2 6 k(Vr − Vq)uε(f )k2kS(cid:0)(p, q](cid:1)ε(f )k2.
kj +l](cid:1)uε(f )(cid:13)(cid:13)2
nXj=0
nXj=0
6 sup{k(Vr − Vπj )uε(f )k2 : r ∈ [πj , πj+1], j = 0, . . . , n} kS(cid:0)[0, t](cid:1)ε(f )k2.
)S(cid:0)(π′
)uε(f )k2kS(cid:0)(π′
ljXl=1(cid:13)(cid:13)(Vπ′
ljXl=1
kj +l](cid:1)ε(f )k2
kj+l−1, π′
kj+l−1, π′
− Vπ′
− Vπ′
k(Vπ′
kj +lj
kj +lj
kj +l
kj +l
6
Proof. Suppose s, t ∈ R+ are such that s < t and let π = {0 = t0 < · · · < tn+1 = t} be a
partition of [0, t] with πm = s. If z ∈ h ⊗ F then
k(VS,π − VS,π′)zk2 =
n+1Xj=m+1
kVπj S(cid:0)(πj−1, πj](cid:1)zk2 6 kS(cid:0)(s, t](cid:1)zk2.
Hence k(VS,t −VS,s)zk 6 kS(cid:0)(s, t](cid:1)zk, which establishes the Cauchy nature of the net (VS,tz)t∈R+
for every z ∈ h ⊗ F, and from this follows the existence of the limit.
Proposition 6.7. Let V be an isometric vacuum-adapted cocycle which is strongly continuous
and let t ∈ (0, ∞]. Then
VS,tES,t = VS,t
and
kVS,tzk = kES,tzk
for all z ∈ h ⊗ F .
Proof. Suppose first that t ∈ (0, ∞) and let π = {0 = π0 < · · · < πn+1 = t}. Then, with the
notation used in the proof of Theorem 3.7,
VS,πES,π =
n+1Xj=1
Vπj S(cid:0)(πj−1, πj](cid:1) n+1Xk=1
S(cid:0)(πk−1, πk](cid:1)Eπk = VS,π,
since V is vacuum adapted and Eπj commutes with S(cid:0)(πj−1, πj](cid:1) for all j. Furthermore, the
working in the proof of Lemma 6.4 shows that
kVS,πzk2 =
n+1Xj=1
kS(cid:0)(πj−1, πj](cid:1)Eπj zk2 = kES,πzk2
for all z ∈ h ⊗ F .
Refining π now gives both identities as claimed; the remaining case follows by letting t → ∞.
Proposition 6.8. Let V be an isometric identity-adapted cocycle which is strongly continuous
and let t ∈ (0, ∞]. Then
kVS,tzk = kS(cid:0)[0, t](cid:1)zk
for all z ∈ h ⊗ F .
Proof. If π is a finite partition of [0, t], where t ∈ (0, ∞), and z ∈ h ⊗ F then the proof
of Lemma 6.4 gives that kVS,πzk = kS(cid:0)[0, t](cid:1)zk. The result follows by refining π and then
letting t → ∞.
Remark 6.9. Propositions 6.7 and 6.8 imply that VS,t is a partial isometry for all t ∈ [0, ∞]
if V is vacuum adapted or identity adapted, and VS is an isometry in the latter case.
17
7 A stopped cocycle relation
Notation 7.1. As in the preceding section, S is a finite quantum stop time extended by
ampliation to act on h ⊗ F; the maps ES, ΓS and σS extend similarly.
Theorem 7.2. Let S be a finite quantum stop time and let V be an isometric p-adapted cocycle.
Then
for all t ∈ R+.
(7.1)
Proof. Let t, T ∈ R+ be such that t < T , let π = {0 = π0 < · · · < πp+1 = ∞} be a finite
partition which contains t = πm and T = πn+1, and let π′ := π − t be the partition of [0, ∞]
such that π′
j = πj+m − t for j = 0, . . . , p + 1 − m. Then, in the notation of Lemma 6.4 and
Theorem 5.2,
VS+t = bVS σS(Vt)
VS+t,π∩[0,T ] =
n+1Xj=1
Vπj (S + t)(cid:0)(πj−1, πj](cid:1) =
=
(Vt)
k
n+1Xj=m+1bVπj −t σπj −t(Vt) S(cid:0)(πj−1 − t, πj − t](cid:1)
n+1−mXj=1 bVπ′
k](cid:1) σπ′
= bVS,π′∩[0,T −t] σS,π′(Vt).
j](cid:1) p+1−mXk=1
S(cid:0)(π′
S(cid:0)(π′
k−1, π′
j−1, π′
j
Refining π and then letting T → ∞ gives the result.
Remark 7.3. The identity (7.1) is a non-deterministic version of (6.2), the defining property
of a left cocycle.
Remark 7.4. In [1], Applebaum considers stopping a unitary identity-adapted process U which
satisfies the cocycle identity
Ut = Γ∗
sU ∗
s Us+tΓs
for all s, t ∈ R+
(7.2)
and the localisation property
U ∗
s Us+t ∈ B(h) ⊗ Is) ⊗ B(F[s,s+t)) ⊗ I[s+t
for all s, t ∈ (0, ∞);
(7.3)
for such processes, these conditions are equivalent to being a left operator Markovian cocycle.
The identity
US+t = USΓSUtΓ∗
S
for all t ∈ R+
(7.4)
is obtained [1, Theorem 4.3], where S is any finite quantum stop time. However, ΓSUtΓ∗
S is
taken to act on the range of the isometry ΓS [1, (4.1)], so it is clearer to write (7.4) in the
following manner:
US+tΓS = USΓSUtΓ∗
SΓS = USΓSUt.
18
From Theorem 7.2, if S is any finite quantum stop time and V is any isometric p-adapted
cocycle then, by (5.2),
which is the identity obtained by Applebaum. Furthermore, Proposition 6.8 gives that
VS+t = bVS σS(Vt) =⇒ VS+tΓS = bVS σS(Vt) ΓS = bVSΓSVt,
which generalises the localisation condition
bV ∗
S VS+t = bV ∗
SbVS σS(Vt) = σS(Vt) ∈ im σS,
U ∗
s Us+t ∈ B(h) ⊗ Is) ⊗ B(F[s) = im σs,
and
which is the stopped version of (7.2).
If V is vacuum adapted and t ∈ R+ then, by Theorem 5.4,
S VS+tΓS = Γ∗
SσS(Vt)ΓS = Γ∗
SΓSVt = Vt,
Γ∗
SbV ∗
so bV ∗
Theorem 5.4 gives that
S VS+t is vacuum adapted at S + t. Similarly, if V is identity adapted and t ∈ R+ then
bV ∗
S VS+t = σS(Vt) = σS(EtVtEt) = σS(Et) σS(Vt) σS(Et) = ES+tbV ∗
σS(Vt) σS+t(X) = σS(cid:0)Vt σt(X)(cid:1) = σS(cid:0)σt(X) Vt(cid:1) = σS+t(X) σS (Vt)
S VS+tES+t,
for all X ∈ M′ ⊗ B(F), so
S VS+t = σS(Vt) ∈ σS+t(cid:0)M′ ⊗ B(F)(cid:1)′
bV ∗
and bV ∗
S VS+t is identity adapted at S + t.
Acknowledgements
=(cid:0)M′ ⊗ IS+t) ⊗ B(F[S+t)(cid:1)′
= M ⊗ B(FS+t)) ⊗ I[S+t
The first author is grateful to the Jawaharlal Nehru Centre for Advanced Scientific Research,
Bangalore, for its hospitality, and to the laboratoire de math´ematiques de Besan¸con, Universit´e
de Franche-Comt´e, where helpful conversations with Professor Uwe Franz took place. Both
authors acknowledge support from the UKIERI research network Quantum Probability, Non-
commutative Geometry and Quantum Information.
References
[1] D. Applebaum, Stopping unitary processes in Fock space, Publ. RIMS Kyoto Univ. 24
(1988), 697 -- 705.
19
[2] W. Arveson, Noncommutative Dynamics and E-Semigroups, Monogr. Math., Springer-
Verlag, New York, 2003.
[3] S. Attal & K. B. Sinha, Stopping semimartingales on Fock space, in Quantum Probability
Communications X (R. L. Hudson & J. M. Lindsay, eds.), 171 -- 185, World Scientific,
Singapore, 1998.
[4] C. Barnett & T. Lyons, Stopping noncommutative processes, Math. Proc. Cambridge
Philos. Soc. 99 (1986), no. 1, 151 -- 161.
[5] C. Barnett & B. Thakrar, A noncommutative random stopping theorem, J. Funct.
Anal. 88 (1990), no. 2, 342 -- 350.
[6] A. C. R. Belton, Quantum Ω-semimartingales and stochastic evolutions, J. Funct.
Anal. 187 (2001), no. 1, 94 -- 109.
[7] A. Coquio, The optional stopping theorem for quantum martingales, J. Funct. Anal. 238
(2006), no. 1, 149 -- 180.
[8] D. W. Fox, Spectral measures and separation of variables, J. Res. Nat. Bur. Standards
Sect. B 80B (1976), no. 3, 347 -- 351.
[9] R. L. Hudson, The strong Markov property for canonical Wiener processes, J. Funct.
Anal. 34 (1979), no. 2, 266 -- 281.
[10] R. L. Hudson, Stop times in Fock space quantum probability, Stochastics 79 (2007),
no. 3 -- 4, 383 -- 391.
[11] J. M. Lindsay, Quantum stochastic analysis -- an introduction, in Quantum Independent
Increment Processes I (M. Schurmann and U. Franz, eds.), 181 -- 271, Lecture Notes in
Math. 1865, Springer-Verlag, Berlin, 2005.
[12] K. R. Parthasarathy, An Introduction to Quantum Stochastic Calculus, Monographs
in Mathematics 85, Birkhauser Verlag, Basel, 1992.
[13] K. R. Parthasarathy & K. B. Sinha, Stop times in Fock space stochastic calculus,
Probab. Theory Related Fields 75 (1987), no. 3, 317 -- 349.
[14] J.-L. Sauvageot, First exit time: a theory of stopping times in quantum processes, in
Quantum Probability and Applications III (L. Accardi & W. von Waldenfels, eds.), 285 -- 299,
Lecture Notes in Math. 1303, Springer-Verlag, Berlin, 1988.
20
|
1703.05190 | 3 | 1703 | 2018-11-06T18:44:42 | On some permanence properties of exact groupoids | [
"math.OA"
] | A locally compact groupoid is said to be exact if its associated reduced crossed product functor is exact. In this paper, we establish some permanence properties of exactness, including generalizations of some known results for exact groups. Our primary goal is to show that exactness descends to certain types of closed subgroupoids, which in turn gives conditions under which the isotropy groups of an exact groupoid are guaranteed to be exact. As an initial step toward these results, we establish the exactness of any transformation groupoid associated to an action of an exact groupoid on a locally compact Hausdorff space. We also obtain a partial converse to this result, which generalizes a theorem of Kirchberg and Wassermann. We end with some comments on the weak form of exactness known as inner exactness. | math.OA | math |
ON SOME PERMANENCE PROPERTIES OF EXACT
GROUPOIDS
SCOTT M. LALONDE
Abstract. A locally compact groupoid is said to be exact if its associated
reduced crossed product functor is exact.
In this paper, we establish some
permanence properties of exactness, including generalizations of some known
results for exact groups. Our primary goal is to show that exactness descends
to certain types of closed subgroupoids, which in turn gives conditions under
which the isotropy groups of an exact groupoid are guaranteed to be exact.
As an initial step toward these results, we establish the exactness of any trans-
formation groupoid associated to an action of an exact groupoid on a locally
compact Hausdorff space. We also obtain a partial converse to this result,
which generalizes a theorem of Kirchberg and Wassermann. We end with
some comments on the weak form of exactness known as inner exactness.
1. Introduction
The notion of an exact group was first introduced by Kirchberg and Wasser-
mann in [11] for the purpose of studying the continuity of C∗-bundles associated
to reduced crossed products. They defined a locally compact group G to be exact
if given any C∗-dynamical system (A, G, α) and any G-invariant ideal I ⊆ A, the
sequence
0 → I ⋊αI ,r G → A ⋊α,r G → A/I ⋊αI ,r G → 0
is exact. There are now several other characterizations of exactness, including
amenability at infinity [1, 6] and, in the case of discrete groups, the exactness of
the reduced C∗-algebra C∗
r (G) [11, Theorem 5.2].
Given the level of attention received by groupoids and their C∗-algebras in recent
years, it is natural and worthwhile to study exactness for locally compact groupoids.
Indeed, we can generalize the definition given by Kirchberg and Wassermann, and
declare a locally compact Hausdorff groupoid to be exact if the sequence
0 → I ⋊αI ,r G → A ⋊α,r G → A/I ⋊
αI ,r G → 0
is exact for any dynamical system (A, G, α) and any G-invariant ideal I. The study
of exact groupoids is fairly new, though they have been investigated in some detail
in [16, 15], and more recently in [2].
It is well-known that the full crossed product functor associated to any group
or groupoid is exact. At the time [11] was written, it was unknown whether there
were groups for which the exactness of the reduced crossed product failed. How-
ever, Gromov later produced examples of non-exact discrete groups. More recently,
Higson, Lafforgue, and Skandalis [9] have constructed reasonably simple examples
Date: November 7, 2018.
2010 Mathematics Subject Classification. Primary 46L55; Secondary 22A22, 43A07.
Key words and phrases. Exact groupoid, groupoid crossed product, inner exactness.
1
2
SCOTT M. LALONDE
of non-exact groupoids. These examples are particularly important, since they also
provide counterexamples to the Baum-Connes conjecture for groupoids.
The aim of this paper is to investigate some permanence properties for exact
groupoids. One such property was already observed in [16], where it was shown
that exactness is preserved under equivalence of groupoids. Most of the results that
we take up in this paper are inspired by ones established for groups in [12]. We
are focused primarily on showing that exactness descends to certain kinds of closed
subgroupoids. As a special case, such results have implications for the isotropy
bundle and isotropy groups of an exact groupoid. A crucial intermediate step is the
result that if G is an exact groupoid and X is any G-space, then the transformation
groupoid G ⋉ X is exact. We also obtain a partial converse to this result -- the
existence of a certain exact transformation groupoid guarantees the exactness of G.
This generalizes a result for groups from [12, §7].
The structure of this paper is as follows. In Section 2, we present some back-
ground information on locally compact groupoids, crossed products, and exactness.
Section 3 is where we establish our first results about exactness for transformation
groupoids, and in Section 4 we apply these results to determine when exactness
descends to subgroupoids. In Section 5 we establish a partial converse to the main
result of Section 3. Finally, in Section 6 we present a brief result together with
some examples, comments, and open questions pertaining to the related notion of
inner exactness (as defined by Anantharaman-Delaroche in [3]).
2. Preliminaries on Exact Groupoids
Let G be a locally compact Hausdorff groupoid. We denote the unit space of G
by G(0) and the set of composable pairs by G(2). The range and source maps are
denoted by r, s : G → G(0), respectively. Unless otherwise specified, it is assumed
that all groupoids are locally compact, Hausdorff, and second countable, and that
they admit continuous Haar systems.
Let us introduce a few bits of useful notation. Given a groupoid G and a set
A ⊆ G(0), we define
GA = s−1(A), GA = r−1(A).
The set GA = GA ∩ GA is a groupoid, called the reduction of G to A. Notice that
if A is invariant in G(0) (meaning it is invariant under the natural action of G on
G(0)), then GA = GA = GA. Of particular import is the special case where A is a
singleton: if u ∈ G(0),
G{u} = {γ ∈ G : r(γ) = s(γ) = u}
is a group, called the isotropy group at u. The union of all the isotropy groups is
called the isotropy bundle (or isotropy subgroupoid ), denoted by Iso(G).
Much of what we aim to do involves groupoid actions on fibered spaces. Let X
be a topological space. We say that X is a (left) G-space if there is a continuous,
open surjection p : X → G(0) and a continuous map Gs ∗p X → X, denoted by
(γ, x) 7→ γ · x, satisfying
(1) p(x) · x = x for all x ∈ X
(2) γ · (η · x) = γη · x whenever (γ, η) ∈ G(2) and s(η) = p(x).
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
3
Example 2.1 (Transformation groupoids). Suppose X is a locally compact Haus-
dorff space, and G is a groupoid acting on the left of X. Define
G ⋉ X = {(γ, x) ∈ G × X : r(γ) = p(x)} .
Then G ⋉ X is a locally compact Hausdorff space, which is second countable when-
ever G and X are. It becomes a groupoid under the operations
(γ, x)(η, η · x) = (γη, x),
(γ, x)−1 = (γ−1, γ−1 · x).
Notice that the range and source maps take the form
r(γ, x) = (r(γ), x),
s(γ, x) = (s(γ), γ−1 · x),
so we identify (G ⋉ X)(0) with X and write r(γ, x) = x and s(γ, x) = γ−1 · x. (See
[14, Proposition 2.3.11] for a more detailed discussion.) Furthermore, if G has a
Haar system {λu}u∈G(0), then {λp(x) × δx}x∈X is a Haar system for G ⋉ X, where
δx denotes the Dirac measure on X concentrated at x [14, Proposition 2.3.12].
It is worth noting that the definition of the transformation groupoid is occasion-
ally taken to be
G ∗ X = {(γ, x) ∈ G × X : s(γ) = p(x)}
equipped with the appropriate operations. In particular, this is the definition used
by Muhly [17, Remark 2.14(1)].
Let G be a groupoid with Haar system {λu}u∈G(0) , suppose A is an upper semi-
continuous C∗-bundle over G(0), and let A = Γ0(G(0), A) denote the associated
algebra of continuous sections that vanish at infinity. Then A is a C0(G(0))-algebra,
and the fiber of A over a point u is the quotient of A by the ideal of sections van-
ishing at u. We usually denote this fiber by Au, though we occasionally write A(u)
when it is more helpful (or notationally convenient) to think of Au in its role as a
quotient of A. Now suppose G acts on A via fiberwise isomorphisms -- that is, there
is a family α = {αγ}γ∈G, where αγ : As(γ) → Ar(γ) is an isomorphism for each
γ ∈ G, and the map (γ, a) 7→ αγ(a) defines a continuous action of G on A. Then
the triple (A, G, α) is called a groupoid dynamical system. We say the dynamical
system is separable if A is separable and G is second countable. All dynamical
systems in this paper are assumed to be separable.
Example 2.2. Suppose G acts on the left of a locally compact Hausdorff space
X. Then there is a continuous open surjection p : X → G(0), so C0(X) becomes a
C0(G(0))-algebra with fibers
C0(X)(u) = C0(p−1(u))
by [24, Example C.4]. Let C denote the associated upper semicontinuous C∗-bundle.
Then by [8, Proposition 4.38] or [19, Example 4.8], G acts on C via the family
lt = {ltγ}γ∈G, where ltγ : C0(p−1(s(γ))) → C0(p−1(r(γ))) is defined by
lt(f )(x) = f (γ−1 · x).
Thus (C, G, lt) is a groupoid dynamical system, which is separable whenever G and
X are second countable.
Given a groupoid dynamical system (A, G, α), the space Γc(G, r∗A) of continuous
compactly-supported sections becomes a ∗-algebra under the convolution product
f ∗ g(γ) =ZG
f (η)αη(cid:0)g(η−1γ)(cid:1) dλr(γ)(η)
4
SCOTT M. LALONDE
and involution
f ∗(γ) = αγ(cid:0)f (γ−1)∗(cid:1).
Moreover, Γc(G, r∗A) forms a topological ∗-algebra under the inductive limit topol-
ogy. (Recall that a net converges in the inductive limit topology if it converges
uniformly and the supports of its elements are eventually contained in a fixed
compact set.) However, there are many different ways to define a C∗-norm on
Γc(G, r∗A). One such example is the universal norm, which we define by setting
kf k = sup kπ(f )k, where π ranges over all representations of Γc(G, r∗A) on Hilbert
space that are continuous in the inductive limit topology. The resulting completion
is the full crossed product of A by G, denoted by A ⋊α G.
There is also a notion of reduced groupoid crossed product. As with other
reduced constructions, its norm is easier to handle, but it is more poorly behaved
then the full crossed product. In order to construct the reduced norm on Γc(G, r∗A),
it is necessary to delve into induced representations of groupoid crossed products.
Suppose (A, G, α) is a separable groupoid dynamical system. The space Γc(G, s∗A)
is a right pre-Hilbert A-module with respect to the operations
z · a(γ) = z(γ)a(s(γ))
and
hhz, wiiA(u) =ZG
z(ξ)∗w(ξ) dλu(ξ),
and its completion Z is a full right Hilbert A-module. Furthermore, A ⋊α G acts
on Z via adjointable operators, where the action is characterized by
f · z(γ) =ZG
α−1
γ (cid:0)f (η)(cid:1)z(η−1γ) dλr(γ)(η)
for f ∈ Γc(G, r∗A) and z ∈ Γc(G, s∗A). Thus we can use Rieffel induction to build
representations of A ⋊α G from those of A. If π : A → B(H) is a representation
of A, then we define Z ⊗A H to be the Hilbert space completion of the algebraic
tensor product Z ⊙ H with respect to the pre-inner product characterized by
( z ⊗ h w ⊗ k) = ( π(hhw, ziiA)h k)
for z, w ∈ Z and h, k ∈ H, as in [21, Equation 2.25]. The induced representation
Ind π of A ⋊α G then acts on Z ⊗A H by
Ind π(f )(z ⊗ h) = f · z ⊗ h.
We will refer to these sorts of induced representations as regular representations. If
we take π to be faithful, we can define the reduced norm by
kf kr = kInd π(f )k.
Notice that this norm is well-defined since induction respects weak containment.
The completion of Γc(G, r∗A) with respect to k·kr is called the reduced crossed
product of A by G, denoted by A ⋊α,r G.
As alluded to above, the full crossed product construction has good functorial
properties. On the other hand, the reduced crossed product can be quite poorly
behaved at times. One well-known instance pertains to short exact sequences. Let
(A, G, α) be a dynamical system, and let I ⊆ A be an ideal. Then [10, §3.3] shows
that I and A/I are both C0(G(0))-algebras, and we let I and A/I denote the
corresponding upper semicontinuous C∗-bundles. If I is G-invariant, meaning that
αγ(Is(γ)) = Ir(γ)
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
5
for all γ ∈ G, then we obtain dynamical systems (I, G, αI ) and (A/I, G, αI ).
Furthermore, the resulting sequence of full crossed products
0 → I ⋊αI G → A ⋊α G → A/I ⋊αI G → 0
is always exact. In contrast, the reduced crossed product functor is not exact in
general. That is, the sequence of reduced crossed products
(1)
0 → I ⋊αI ,r G → A ⋊α,r G → A/I ⋊
αI ,r G → 0
need not be exact.
Indeed, Higson, Lafforgue, and Skandalis [9] have produced
examples of groupoids for which sequences of the form (1) can fail to be exact.
These so-called HLS groupoids have since been studied in [23] and [3].
In [23],
Willett constructed an HLS groupoid G satisfying the weak containment property
(i.e., C∗(G) = C∗
r (G)) which nevertheless fails to be amenable. The properties
exhibited by the original HLS groupoids in [9] and by Willett's groupoid arise
in part due to the failure of the groupoids to be inner exact, a term coined by
Anantharaman-Delaroche in [3].
Finally, we will repeatedly appeal to the concept of groupoid equivalence and
its relationship to exactness. Recall that two locally compact Hausdorff groupoids
G and H are said to be equivalent if there is a locally compact Hausdorff space Z
such that
• G and H act freely and properly on the left and right of Z, respectively;
• the actions of G and H commute; and
• the structure maps Z → G(0) and Z → H (0) for the actions induce homeo-
morphisms Z/H ∼= G(0) and G\Z ∼= H (0).
Any such space Z is called a (G, H)-equivalence.
It is a well-known fact that if
G and H are equivalent groupoids, then C∗(G) and C∗(H) are Morita equivalent,
and the same is true for the reduced C∗-algebras [22, Theorem 4.1]. Furthermore,
equivalence preserves measurewise amenability [4, Theorem 3.2.16] and exactness
[16, Theorem 4.8]. We will need to invoke the latter result several times throughout
this paper.
3. Transformation Groupoids
Let G be a locally compact Hausdorff groupoid with Haar system, and suppose
X is a left G-space. This first part of this section is devoted to showing that any
(G ⋉ X)-dynamical system yields a G-dynamical system in a natural way, and the
resulting crossed products are isomorphic. As a result, we are then able to show
that if G is exact and X is any left G-space, then the transformation groupoid
G ⋉ X is also exact. We use this result in the next section to show that exactness
passes to closed subgroupoids.
We first need a fact regarding upper semicontinuous C∗-bundles. This result
is undoubtedly clear to experts, but we present the details here. The following
proposition is an extension of [24, Example C.4] to general C0(X)-algebras.
Proposition 3.1. Let X and Y be locally compact Hausdorff spaces, and suppose
σ : Y → X is continuous. Let A → Y be an upper semicontinuous C∗-bundle over
Y , and let A = Γ0(Y, A) denote the associated C0(Y )-algebra. Then A is also a
C0(X)-algebra, with the C0(X)-action given by
(ϕ · a)(y) = ϕ(σ(y))a(y)
6
SCOTT M. LALONDE
for a ∈ A and ϕ ∈ C0(X). The resulting upper semicontinuous C∗-bundle B → X
has fibers
Bx = Γ0(σ−1(x), A),
and the section algebra B = Γ0(X, B) is isomorphic to A via the map Θ : A → B
defined by
for a ∈ A, x ∈ X, and y ∈ σ−1(x).
Θ(a)(x)(y) = a(x)
Proof. Since A is a C0(Y )-algebra, there is a continuous map τ : Prim A → Y
by [24, Proposition C.5]. By composing with the map σ : Y → X, we obtain a
continuous map σ ◦ τ : Prim A → X, which makes A into a C0(X)-algebra.
By the discussion on page 355 of [24], the homomorphism ΦX
A : C0(X) → ZM (A)
associated to the C0(X)-action is defined to be
ΦX
A (ϕ) = ϕ ◦ (σ ◦ τ ) = (ϕ ◦ σ) ◦ τ = ΦY
A(ϕ ◦ σ),
where ΦY
on A. It is then clear that ϕ · a = (ϕ ◦ σ) · a for all ϕ ∈ C0(X) and all a ∈ A.
A : C0(Y ) → ZM (A) is the homomorphism implementing the C0(Y )-action
We will now identify the structure of the fibers of the associated upper semicon-
tinuous bundle B → X. Given x ∈ X, we have Bx = A/Ix, where
Notice that if σ(y) = x and ϕ(x) = 0, then
Ix = span(cid:8)ϕ · a : ϕ ∈ C0(X), ϕ(x) = 0, a ∈ A(cid:9).
(ϕ · a)(y) = ϕ(σ(y))a(y) = ϕ(x)a(y) = 0
It follows that Ix consists precisely of the sections in Γ0(X, B)
for all a ∈ A.
that vanish on σ−1(x), hence Bx = Γ0(σ−1(x), A). (See [8, Proposition 5.14], for
example.) That A and B are isomorphic follows immediately from [24, Theorem
C.26(c)], which guarantees the existence of a C0(X)-linear isomorphism Θ : A → B.
Given the manner in which we constructed the bundle B, the isomorphism is of
course given by restriction.
(cid:3)
We will apply Proposition 3.1 in a few different situations, but the first pertains
specifically to transformation groupoids. Suppose X is a left G-space, and let
p : X → G(0) denote the structure map for the G-action. Recall that we assume p
is continuous and open. Let G ⋉ X denote the associated transformation groupoid,
and suppose (A, G ⋉ X, α) is a separable groupoid dynamical system. Then A is
an upper semicontinuous C∗-bundle over X = (G ⋉ X)(0), and we let A = Γ0(X, A)
denote the associated C0(X)-algebra of sections. By Proposition 3.1, A is also a
C0(G(0))-algebra, and the resulting upper semicontinuous C∗-bundle B → G(0) has
fibers Bu = Γ0(p−1(u), A) for each u ∈ G(0). If we let B = Γ0(G(0), B) denote the
section algebra, then we also know that B is isomorphic to A. In fact, we can use
the (G ⋉ X)-action on A to construct a G-action on B.
Proposition 3.2. For each γ ∈ G, define βγ : Bs(γ) → Br(γ) by
Then the family {βγ}γ∈G defines a continuous action of G on B. Consequently,
(B, G, β) is a groupoid dynamical system.
βγ(b)(x) = α(γ,x)(cid:0)b(γ−1 · x)(cid:1).
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
7
To prove this proposition, we will proceed via a series of lemmas. First we need
to make a couple of observations. Let q : G × X → G denote the projection onto
the first coordinate. Since G ⋉ X ⊆ G × X, q restricts to a continuous surjection
G ⋉ X → G. As a result, the pullback algebras Γ0(G ⋉ X, s∗A) and Γ0(G ⋉ X, r∗A)
become C0(G)-algebras via Proposition 3.1. Furthermore, for each γ ∈ G the
corresponding fiber is
Γ0(G ⋉ X, r∗A)(γ) = Γ0(q−1(γ), r∗A).
Notice that
q−1
G⋉X (γ) = {(γ, x) ∈ G ⋉ X : p(x) = r(γ)} = {γ} × p−1(r(γ)),
and for each x ∈ p−1(r(γ)),
(r∗A)(γ,x) = Ar(γ,x) = Ax.
Thus
Γ0(G ⋉ X, r∗A)(γ) = Γ0(p−1(r(γ)), A) = Br(γ).
It will also be helpful to consider the set
G ∗ X = {(γ, x) ∈ G × X : s(γ) = p(x)}.
As we mentioned in Section 2, G ∗ X can be made into a groupoid in a natural
way. We will not need to consider the groupoid structure on G ∗ X, but it is worth
noting that the unit space can be identified with X, and the source map is given by
s(γ, x) = x. Thus we have a continuous surjection q : G ∗ X → G, so Γ0(G ∗ X, s∗A)
is a C0(G)-algebra with fibers given by
Γ0(G ∗ X, s∗A)(γ) = Γ0(q−1(γ), s∗A) = Γ0(p−1(s(γ)), A) = Bs(γ).
Now we can begin the proof. Compare the first lemma to [8, Lemma 4.37] and
the implementation of that result in the proof of [8, Proposition 4.38].
Lemma 3.3. Let s∗B = Γ0(G, s∗B) and r∗B = Γ0(G, r∗B) denote the pullback
C∗-algebras. There are C0(G)-linear isomorphisms
ιs : s∗B → Γ0(G ∗ X, s∗A),
ιr : r∗B → Γ0(G ⋉ X, r∗A)
given by
ιs(f )(γ, x) = f (γ)(x),
ιr(g)(γ, x) = g(γ)(x)
for f ∈ s∗B and g ∈ r∗B.
Proof. We will work out the proof for ιr, and things work similarly for ιs. First
notice that if f ∈ r∗B, then ιr(f ) is clearly a section of r∗A. Moreover, ιr is easily
seen to be a homomorphism. It is also straightforward to check that ιr is isometric:
if f ∈ r∗B, then
kιr(f )k∞ =
sup
kιr(f )(γ, x)k
(γ,x)∈G⋉X
=
sup
kf (γ)(x)k
(γ,x)∈G⋉X
= sup
γ∈G
= sup
γ∈G
sup
kf (γ)(x)k
x∈p−1(r(γ))
kf (γ)k∞
= kf k∞.
8
SCOTT M. LALONDE
However, it is not clear yet that ιr even maps into Γ0(G ⋉ X, r∗A). That is, we
need to check that ιr(f ) is continuous and vanishes at infinity for all f ∈ r∗B.
Recall that elementary tensors of the form ϕ ⊗ b for ϕ ∈ Cc(G) and b ∈ B, where
(ϕ ⊗ b)(γ) = ϕ(γ)b(r(γ)),
span a dense subalgebra of r∗B. (See [15, §3.2], for example.) On such elementary
tensors, we have
ιr(ϕ ⊗ b)(γ, x) = (ϕ ⊗ b)(γ)(x) = ϕ(γ)b(r(γ))(x).
Since the function x 7→ b(p(x))(x) vanishes at infinity by Proposition 3.1 (it is
precisely Θ−1(b) ∈ A), the set
{x ∈ X : kb(p(x))(x)k ≥ ε}
is compact for all ε > 0. Since ϕ is compactly supported on G, it is easy to see
that ιr(ϕ ⊗ b) vanishes at infinity on G ⋉ X.
It remains to see that ιr(ϕ ⊗ b)
is continuous. Suppose (γi, xi) → (γ, x) in G ⋉ X. Then ϕ(γi) → ϕ(γ) and
b(r(γi))(xi) → b(r(γ))(x) in A, so
in (G ⋉ X) × A. It follows that
(cid:0)(γi, xi), b(r(γi))(xi)(cid:1) →(cid:0)(γ, x), b(r(γ))(x)(cid:1)
(cid:0)(γi, xi), ϕ(γi)b(r(γi))(xi)(cid:1) →(cid:0)(γ, x), ϕ(γ)b(r(γ))(x)(cid:1)
in r∗A, so ιr(ϕ ⊗ b) ∈ Γ0(G ⋉ X, r∗A). Since Cc(G) ⊙ B is dense in r∗B and
ιr is isometric, it follows that ιr(f ) is continuous and vanishes at infinity for any
f ∈ r∗B.
Now suppose f ∈ r∗B with ιr(f ) = 0. Then ιr(f )(γ, x) = 0 for all (γ, x) ∈ G⋉X,
so f (γ)(x) = 0 for all γ ∈ G and x ∈ X. Thus f = 0, and ιr is injective. To see
that ιr is surjective, it suffices to show that the range of ιr is closed under the
C0(G)-action on Γ0(G ⋉ X, r∗A) and is fiberwise dense in r∗A. First observe that
for each f ∈ r∗B and ϕ ∈ C0(G),
(cid:0)ϕ · ιr(f )(cid:1)(γ, x) = ϕ(γ)f (γ)(x) =(cid:0)ϕ · f(cid:1)(γ)(x) = ιr(ϕ · f )(γ, x),
so the range of ιr is invariant under the C0(G)-action. Moreover, this computation
shows that ιr is C0(G)-linear. Now fix (γ, x) ∈ G ⋉ X and let a ∈ (r∗A)(γ,x) = Ax.
Choose b ∈ B with b(p(x))(x) = a, and let ϕ ∈ Cc(G) with ϕ(γ) = 1. Then
ιr(ϕ ⊗ b)(γ, x) = ϕ(γ)b(r(γ))(x) = a.
Thus the image of ιr is fiberwise dense, and it follows from [24, Proposition C.24]
that ιr is surjective.
(cid:3)
Lemma 3.4. The map lt : Γ0(G ∗ X, s∗A) → Γ0(G ⋉ X, s∗A) defined by
lt(f )(γ, x) = f (γ, γ−1 · x)
is a C0(G)-linear isomorphism, which restricts to a surjection from Γc(G ∗ X, s∗A)
onto Γc(G ⋉ X, s∗A). Moreover, lt is continuous with respect to the inductive limit
topology.
Proof. We first need to check that lt actually maps into Γ0(G ⋉ X, s∗A). To that
end, suppose first that f ∈ Γc(G∗X, s∗A). It is easy to see that lt(f ) is continuous --
if (γi, xi) → (γ, x) in G ⋉ X, then (γi, γ−1
· xi) → (γ, γ−1 · x) in G ∗ X, so
i
lt(f )(γi, xi) = f (γi, γ−1
i
· xi) → f (γ, γ−1 · x) = lt(f )(γ, x).
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
9
Now let K = supp(f ). Notice that supp(lt(f )) is precisely the image of K under
the continuous map (γ, x) 7→ (γ, γ · x) from G ∗ X → G ⋉ X, so it is compact. Thus
lt(f ) is compactly supported.
Now suppose fi → f in Γc(G ∗ X, s∗A) with respect to the inductive limit
topology. Then fi → f uniformly. A straightforward computation like the one
from Lemma 3.3 shows that lt is isometric with respect to the supremum norm,
so lt(fi) → lt(f ) uniformly as well. Furthermore, there is a compact set K such
that supp(fi) ⊆ K eventually. The same argument as above shows that the sets
supp(lt(fi)) are eventually contained in the image of K under the map (γ, x) 7→
(γ, γ · x), so lt(fi) → lt(f ) in the inductive limit topology.
Since lt is isometric, it extends to an injective map from Γ0(G ∗ X, s∗A) to
Γ0(G ⋉ X, s∗A). It is not hard to see that lt is an isomorphism, since it has an
inverse lt−1 : Γ0(G ⋉ X, s∗A) → Γ0(G ∗ X, s∗A) given by
lt−1(f )(γ, x) = f (γ, γ · x).
It is also worth noting that arguments like those above show that lt−1 takes com-
pactly supported sections to compactly supported sections, and it is continuous in
the inductive limit topology.
All that remains is to see that lt is C0(G)-linear. Let f ∈ Γ0(G ∗ X, s∗A) and
ϕ ∈ C0(G). Then for all (γ, x) ∈ G ⋉ X we have
lt(ϕ · f )(γ, x) = (ϕ · f )(γ, γ−1 · x)
= ϕ(γ)f (γ, γ−1 · x)
= ϕ(γ)lt(f )(γ, x)
= (ϕ · lt(f ))(γ, x).
(cid:3)
Since (A, G ⋉ X, α) is a groupoid dynamical system, there is a C0(G ⋉ X)-linear
isomorphism α : s∗A → r∗A given by α(f )(γ, x) = α(γ,x)(f (γ, x)) as a result of [19,
Lemma 4.3].
Lemma 3.5. The isomorphism α : s∗A → r∗A associated to the (G ⋉ X)-action
on A is C0(G)-linear.
Proof. Let f ∈ s∗A = Γ0(G ⋉ X, s∗A) and ϕ ∈ C0(G). Then
α(ϕ · f )(γ, x) = α(γ,x)(cid:0)(ϕ · f )(γ, x)(cid:1)
= α(γ,x)(cid:0)ϕ(γ)f (γ, x)(cid:1)
= ϕ(γ)α(γ,x)(cid:0)f (γ, x)(cid:1)
=(cid:0)ϕ · α(f )(cid:1)(γ, x)
= ϕ(γ)α(f )(γ, x)
for all (γ, x) ∈ G ⋉ X. Thus α(ϕ · f ) = ϕ · α(f ), so α is C0(G)-linear.
(cid:3)
Proof of Proposition 3.2. In light of [19, Lemma 4.3], it suffices to show that the
map β : s∗B → r∗B defined by β(f )(γ) = βγ(f (γ)) is a C0(G)-linear isomorphism.
Let f ∈ s∗B. Then ιs(f ) ∈ Γ0(G ∗ X, s∗A) is given by
and applying lt we get
ιs(f )(γ, x) = f (γ)(x),
(lt ◦ ιs)(f )(γ, x) = ιs(f )(γ, γ−1 · x) = f (γ)(γ−1 · x).
10
SCOTT M. LALONDE
Now apply α:
α(cid:0)(lt ◦ ιs)(f )(cid:1)(γ, x) = α(γ,x)(cid:0)(lt ◦ ιs)(f )(γ, x)(cid:1) = α(γ,x)(cid:0)f (γ)(γ−1 · x)(cid:1).
Equivalently, we have
Using the fact that the left side is (ι−1
α(cid:0)(lt ◦ ιs)(f )(cid:1)(γ, x) = βγ(f (γ))(x) = β(f )(γ)(x).
r ◦ α ◦ lt ◦ ιs)(f )(γ)(x), it follows that
β(f ) = (ι−1
r ◦ α ◦ lt ◦ ιs)(f ).
Since each of the functions in the composition is a C0(G)-linear isomorphism, it
follows that β is also a C0(G)-linear isomorphism.
It is also necessary to check that βγ ◦ βη = βγη whenever (γ, η) ∈ G(2). This is
straightforward:
βγ(βη(b))(x) = α(γ,x)(cid:0)βη(b)(γ−1 · x)(cid:1)
= α(γ,x)(cid:0)α(η,γ−1·x)(cid:0)b(η−1 · (γ−1 · x))(cid:1)(cid:1)
= α(γη,x)(cid:0)b((γη)−1 · x)(cid:1)
= βγη(b)(x)
for all b ∈ B. Therefore, (B, G, β) is a groupoid dynamical system.
(cid:3)
Now we have two closely related dynamical systems, (A, G ⋉ X, α) and (B, G, β).
We claim that they yield isomorphic crossed products. This result can be thought
of as simultaneously generalizing [8, Proposition 4.38] and [7, Remark 3.61].
Theorem 3.6. The map Φ : Γc(G ⋉ X, r∗A) → Γc(G, r∗B) given by
Φ(f )(γ)(x) = f (γ, x)
is a ∗-homomorphism, which extends to an isomorphism ¯Φ : A⋊α(G⋉X) → B⋊β G.
Proof. We will first check that Φ maps into Γc(G, r∗B). Let f ∈ Γc(G ⋉ X, r∗A).
Since Φ agrees with ι−1
r on Γc(G ⋉ X, r∗A), it is immediate that Φ(f ) is continuous.
Now let K denote the image of supp(f ) under the projection G ⋉ X → G. If γ 6∈ K,
then Φ(f )(γ) = 0, so supp(Φ(f )) is contained in the compact set K. Thus Φ(f ) is
compactly supported.
Since Φ is clearly linear, we just need to check that it preserves the convolution
and involution. If f, g ∈ Γc(G ⋉ X, r∗A), then
Φ(f ∗ g)(γ)(x) = (f ∗ g)(γ, x)
=ZG⋉X
=ZG
=ZG
=ZG
=ZG
f (η, y)α(η,y)(cid:0)g((η, y)−1(γ, x))(cid:1) d(λr(γ) × δx)(η, y)
f (η, x)α(η,x)(cid:0)g((η, x)−1(γ, x))(cid:1) dλr(γ)(η)
f (η, x)α(η,x)(cid:0)g(η−1γ, η−1 · x)(cid:1) dλr(γ)(η)
Φ(f )(η)(x)α(η,x)(cid:0)Φ(g)(η−1γ)(η−1 · x)(cid:1)dλr(γ)(η)
Φ(f )(η)(x)βη(cid:0)Φ(g)(η−1γ)(cid:1)(x) dλr(γ)(η)
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
11
since evaluation at x is a bounded linear map. However,
= ZG
Φ(f )(η)βη(cid:0)Φ(g)(η−1γ)(cid:1) dλr(γ)(η)!(x)
Φ(f )(η)βη(cid:0)Φ(g)(η−1γ)(cid:1) dλr(γ)(η) = Φ(f ) ∗ Φ(g)(γ),
ZG
and it follows that Φ is multiplicative. As for the involution, we have
Φ(f ∗)(γ)(x) = f ∗(γ, x)
= α(γ,x)(cid:0)f ((γ, x)−1)∗(cid:1)
= α(γ,x)(cid:0)f (γ−1, γ−1 · x)∗(cid:1)
= α(γ,x)(cid:0)Φ(f )(γ−1)(γ−1 · x)(cid:1)∗
= βγ(cid:0)Φ(f )(γ−1)∗(cid:1)(x)
= Φ(f )∗(γ)(x).
Thus Φ is a ∗-homomorphism.
To see that Φ extends to a homomorphism between the full crossed products, we
show it is continuous with respect to the inductive limit topology. Suppose fi → f in
Γc(G⋉X, r∗A) with respect to the inductive limit topology. Then fi → f uniformly,
and there is a fixed compact set K0 ⊆ G ⋉ X such that supp(fi) ⊆ K0 eventually.
Let K ⊆ G denote the image of K0 under projection onto the first coordinate.
Then K is compact. Moreover, supp(fi) ⊆ K0 implies Φ(fi)(γ) = f (γ, ·) = 0 when
γ 6∈ K. Therefore, supp(Φ(fi)) is eventually contained in K. Also,
kΦ(fi) − Φ(f )k∞ = sup
γ∈G
= sup
γ∈G
kΦ(fi)(γ) − Φ(f )(γ)k∞
sup
kΦ(fi)(γ)(x) − Φ(f )(γ)(x)k
x∈p−1(r(γ))
sup
kfi(γ, x) − f (γ, x)k
= sup
γ∈G
=
x∈p−1(r(γ))
sup
kfi(γ, x) − f (γ, x)k
(γ,x)∈G⋉X
= kfi − f k∞,
so Φ(fi) → Φ(f ) uniformly. Thus Φ(fi) → Φ(f ) in the inductive limit topology, so
Φ is continuous. It follows that Φ extends to a homomorphism ¯Φ : A ⋊α (G ⋉ X) →
B ⋊β G. It remains to see that ¯Φ is an isomorphism.
We begin by showing that Φ is injective on Γc(G⋉X, r∗A). This is fairly routine:
if Φ(f ) = 0 for some f ∈ Γc(G ⋉ X, r∗A), then Φ(f )(γ) = 0 for all γ ∈ G, which in
turn means that Φ(f )(γ)(x) = 0 for all γ ∈ G and all x ∈ p−1(r(γ)). But then
f (γ, x) = Φ(f )(γ)(x) = 0
for all (γ, x) ∈ G ⋉ X. Thus f = 0, and Φ is injective on Γc(G ⋉ X, r∗A). Thus ¯Φ
restricts to an isomorphism of Γc(G ⋉ X, r∗A) onto im Φ ⊆ Γc(G, r∗B).
Next we claim that im Φ is dense in Γc(G, r∗B) with respect to the inductive limit
topology, which will show that ¯Φ is surjective. We will use [24, Proposition C.24].
Notice first that im Φ is a C0(G)-module: if g = Φ(f ) for some f ∈ Γc(G ⋉ X, r∗A)
and ϕ ∈ C0(G), then
Φ(ϕ · f )(γ)(x) = (ϕ · f )(γ, x)
12
SCOTT M. LALONDE
= ϕ(γ)f (γ, x)
= ϕ(γ)Φ(f )(γ)(x)
= (ϕ · Φ(f ))(γ)(x)
= (ϕ · g)(γ)(x),
so ϕ · g ∈ im Φ. Now fix γ ∈ G. We need to show that
{g(γ) : g ∈ im Φ}
is dense in r∗Bγ. Let b ∈ Γc(p−1(r(γ)), A) ⊆ r∗Bγ, and use the vector-valued
Tietze extension theorem [18, Proposition A.5] to find f ∈ Γc(G ⋉ X, A) such that
f (γ, x) = b(x) for all x ∈ p−1(r(γ)). Then
Φ(f )(γ)(x) = f (γ, x) = b(x)
for all x ∈ p−1(r(γ)), so Φ(f )(γ) = b. It follows that {g(γ) : g ∈ im Φ} is dense
in (r∗B)γ, so [24, Proposition C.24] implies that im Φ is dense in Γ0(G, r∗B). Now
let g ∈ Γc(G, r∗B). Then there is a net fi ∈ Γc(G ⋉ X, r∗A) such that Φ(fi) → g
uniformly. Let K = supp(g), and choose ϕ ∈ Cc(G)+ such that ϕK ≡ 1 and
ϕ(x) < 1 for all x 6∈ K. Put gi = ϕ·Φ(fi) = Φ(ϕ·fi). Notice that ϕ·g = g, so gi → g
uniformly. Moreover, each gi is compactly supported, and supp(gi) ⊆ supp(ϕ) for
all i. Thus gi → g in the inductive limit topology. It follows that ¯Φ is surjective.
Recall that ¯Φ restricts to an isomorphism of Γc(G ⋉ X, r∗A) onto the dense
subalgebra im Φ, so we have an inverse Ψ : im Φ → Γc(G ⋉ X, r∗A) given by
Ψ(g)(γ, x) = g(γ)(x).
We claim that Ψ extends to a homomorphism at the level of crossed products -- to
show it, we will prove that Ψ is I-norm decreasing. Let g ∈ im Φ, and observe that
ZG⋉X
Similarly,
kΨ(g)(γ, y)k d(λr(x) × δx)(γ, y) =ZG
=ZG
≤ZG
kΨ(g)(γ, x)k dλr(x)(γ)
kg(γ)(x)k dλr(x)(γ)
kg(γ)k∞ dλr(x)(γ)
≤ kgkI .
ZG⋉X(cid:13)(cid:13)Ψ(g)(γ−1, γ−1 · y)(cid:13)(cid:13) d(λr(x) × δx)(γ, y) ≤ZG(cid:13)(cid:13)g(γ−1)(cid:13)(cid:13)∞ dλr(x)(γ) ≤ kgkI ,
and taking suprema leads to the conclusion that kΨ(g)kI ≤ kgkI . Since Ψ is I-norm
decreasing, it extends to a homomorphism ¯Ψ : B ⋊β G → A ⋊α (G ⋉ X). Moreover,
¯Ψ and ¯Φ are inverses on dense subalgebras, so they must be inverses everywhere.
That is, ¯Φ is an isomorphism.
(cid:3)
In order to prove results about exactness, we require an analogue of Theorem 3.6
for the reduced crossed product. Therefore, we need to show the homomorphism
Φ : Γc(G ⋉ X, r∗A) → Γc(G, r∗B) is isometric with respect to the reduced norms.
Let π : B → B(H) be a faithful representation, and let Y denote the right Hilbert
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
13
B-module obtained by completing Γc(G, s∗B) with respect to the B-valued inner
product defined by
hhw, ziiB (u) =ZG
w(γ)∗z(γ) dλu(γ)
for all u ∈ G(0). Then the induced representation Ind π acts on the Hilbert space
Y ⊗B H, which is equipped with the inner product
Moreover,
( z ⊗ h w ⊗ k) =(cid:0) π(cid:0)hhw, ziiB(cid:1)h(cid:12)(cid:12) k(cid:1) ,
kf kr = kInd π(f )k
for all f ∈ Γc(G, r∗B). Likewise, π ◦ Θ : A → B(H) is a faithful representation,
where Θ : A → B is the isomorphism afforded by Proposition 3.1.
If we let Z
denote the right Hilbert A-module completion of Γc(G ⋉ X, s∗A) with respect to
the inner product defined by
hhw, ziiA(x) =ZG
w(γ, γ · x)∗z(γ, γ · x) dλp(x)(γ)
for all x ∈ X, then the resulting induced representation Ind(π ◦ Θ) acts on Z ⊗A H,
which carries the inner product
Again,
( z ⊗ h w ⊗ k) =(cid:0) π ◦ Θ(cid:0)hhw, ziiA(cid:1)h(cid:12)(cid:12) k(cid:1) .
kf kr = kInd(π ◦ Θ)(f )k
for all f ∈ Γc(G ⋉ X, r∗A). Therefore, to show that Φ is isometric, it is enough to
show that (Ind π) ◦ Φ and Ind(π ◦ Θ) are unitarily equivalent. Notice that the map
lt−1 clearly takes compactly supported sections to compactly supported sections,
and an argument like the one from the proof of the previous theorem shows the same
is true for ι−1
s . Thus we have a natural map from Γc(G ⋉ X, s∗A) to Γc(G, s∗B),
namely (lt ◦ ιs)−1. We then define a map U : Γc(G ⋉ X, s∗A) ⊙ H → Γc(G, s∗B) ⊙ H
on elementary tensors by
which we claim extends to the completion Z ⊗A H.
U (z ⊗ h) = (lt ◦ ιs)−1(z) ⊗ h,
Proposition 3.7. The map U extends to a unitary U : Z ⊗A H → Y ⊗B H that
intertwines the representations (Ind π) ◦ Φ and Ind(π ◦ Θ).
Proof. To see that U extends to a unitary, we just need to check that U is isometric
and has dense range. Let z ⊗ h, w ⊗ k ∈ Γc(G ⋉ X, s∗A) ⊙ H and observe that
( U (z ⊗ h) U (w ⊗ k)) =(cid:0) (lt ◦ ιs)−1(z) ⊗ h(cid:12)(cid:12) (lt ◦ ιs)−1(w) ⊗ k(cid:1)
=(cid:0) π(cid:0)hh(lt ◦ ιs)−1(w), (lt ◦ ιs)−1(z)iiB(cid:1)h(cid:12)(cid:12) k(cid:1) ,
hh(lt ◦ ιs)−1(w), (lt ◦ ιs)−1(z)iiB(u)(x)
where
(lt ◦ ιs)−1(w)(γ)∗(lt ◦ ιs)−1(z)(γ) dλu(γ)(cid:19)(x)
(lt ◦ ιs)−1(w)(γ)(x)∗(lt ◦ ιs)−1(z)(γ)(x) dλu(γ)
=(cid:18)ZG
=ZG
14
SCOTT M. LALONDE
lt−1(w)(γ, x)∗lt−1(z)(γ, x) dλu(γ)
=ZG
=ZG
=ZG⋉X
= hhw, ziiA(x)
w(γ, γ · x)∗z(γ, γ · x) dλp(x)(γ)
w(γ, y)∗z(γ, y) dµx(γ, y)
for all u ∈ G(0) and all x ∈ p−1(u). Notice that
Θ(hhw, ziiA)(u, x) = hhw, ziiA(x),
so it follows that
Thus
Θ(hhw, ziiA) = hh(lt ◦ ιs)−1(w), (lt ◦ ιs)−1(z)iiB.
( U (z ⊗ h) U (w ⊗ k)) =(cid:0) π ◦ Θ(cid:0)hhw, ziiA(cid:1)h(cid:12)(cid:12) k(cid:1)
= ( z ⊗ h w ⊗ k)
in Γc(G ⋉ X, s∗A) ⊙ H, so U is isometric. To see that it has dense range, we first
claim that (lt ◦ ιs)−1 is continuous with respect to the inductive limit topologies on
Γc(G ⋉ X, s∗A) and Γc(G, s∗B). This is fairly straightforward -- an argument like
the one from the proof of Theorem 3.6 shows that ι−1
is continuous with respect to
the inductive limit topology, and we already established that lt−1 is continuous in
the inductive limit topology in the proof of Lemma 3.4. Thus the range of (lt◦ ιs)−1
is dense in Γc(G, s∗B) with respect to the inductive limit topology. An argument
identical to the one from [15, Lemma 5.5] then shows that the range of (lt ◦ ιs)−1
is norm-dense in the Hilbert B-module Y. It follows that the range of U is dense
in Y ⊗B H: if zi → z in Y, then for any h ∈ H we have
s
k(zi − z) ⊗ hk2 = ( (zi − z) ⊗ h (zi − z) ⊗ h)
=(cid:0) π(cid:0)hhzi − z, zi − ziiB(cid:1)h(cid:12)(cid:12) h(cid:1)
≤ khhzi − z, zi − ziiBkkhk2
= kzi − zk2khk2
→ 0,
so zi ⊗ h → z ⊗ h. Therefore, U extends to a unitary.
We are left with the verification that U intertwines (Ind π) ◦ Φ and Ind(π ◦ Θ).
Well, observe that if f ∈ Γc(G ⋉ X, r∗A) and z ⊗ h ∈ Γc(G ⋉ X, s∗A) ⊙ H, then
Ind π(Φ(f )) · U (z ⊗ h) = Ind π(Φ(f ))(cid:0)(lt ◦ ιs)−1(z) ⊗ h(cid:1)
= Φ(f ) · (lt ◦ ιs)−1(z) ⊗ h,
where
Φ(f ) · (lt ◦ ιs)−1(z)(γ)(x) =ZG
β−1
γ (cid:0)Φ(f )(η)(cid:1)(x)(lt ◦ ιs)−1(z)(η−1γ)(x) dλr(γ)(η)
=ZG
=ZG
α(γ−1,x)(cid:0)Φ(f )(η)(γ · x)(cid:1)lt−1(z)(η−1γ, x) dλr(γ)(η)
(γ,γ·x)(cid:0)f (η, γ · x)(cid:1)z(η−1γ, η−1γ · x) dλr(γ)(η)
α−1
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
15
α−1
(γ,γ·x)(cid:0)f (η, γ · x)(cid:1)z(cid:0)(η−1, η−1γ · x)(γ, γ · x)(cid:1) dλr(γ)(η)
(γ,γ·x)(cid:0)f (η, γ · x)(cid:1)z(cid:0)(η, γ · x)−1(γ, γ · x)(cid:1) dλr(γ)(η)
α−1
=ZG
=ZG
= f · z(γ, γ · x)
= lt−1(f · z)(γ, x)
= ι−1
s (cid:0)lt−1(f · z)(cid:1)(γ)(x).
That is,
so
Φ(f ) · (lt ◦ ιs)−1(z) = (lt ◦ ιs)−1(f · z),
Ind π(Φ(f )) · U (z ⊗ h) = Φ(f ) · (lt ◦ ιs)−1(z) ⊗ h
= (lt ◦ ιs)−1(f · z) ⊗ h
= U (f · z ⊗ h)
= U · Ind(π ◦ Θ)(f )(z ⊗ h).
Thus U indeed intertwines the given representations.
(cid:3)
Theorem 3.8. The map Φ : Γc(G ⋉ X, r∗A) → Γc(G, r∗B) from Theorem 3.6
extends to an isomorphism Φ : A ⋊α,r (G ⋉ X) → B ⋊β,r G.
Proof. We just need to check that Φ is isometric with respect to the reduced norms.
This follows easily from the last proposition: if π is a faithful representation of B
and f ∈ Γc(G ⋉ X, r∗A), then
kΦ(f )kr = kInd π(Φ(f ))k = kInd(π ◦ Θ)(f )k = kf kr.
(cid:3)
Before we can prove the main theorem of this section, we need one final result
about invariant ideals.
Proposition 3.9. Let Θ : A → B denote the isomorphism afforded by Proposition
3.1. Let I ⊆ A be an ideal, and let J = Θ(I). Then I is (G ⋉ X)-invariant if and
only if J is G-invariant.
Proof. Let I and J denote the upper semicontinuous C∗-bundles associated to I
and J. Suppose first that I is (G ⋉ X)-invariant, i.e., for all (γ, x) ∈ G ⋉ X,
Observe that for each γ ∈ G, Js(γ) = C0(p−1(s(γ)), I). Thus if b ∈ Js(γ) we have
α(γ,x)(cid:0)Iγ−1·x(cid:1) = Ix.
βγ(b)(x) = α(γ,x)(cid:0)b(γ−1 · x)(cid:1) ∈ Ix,
so βγ(b) ∈ C0(p−1(r(γ)), I) = Jr(γ). It is not hard to see that βγ(Js(γ)) = Jr(γ),
since the same argument shows that β−1
γ (Jr(γ)) ⊆ Js(γ). Thus J is a G-invariant
ideal.
Conversely, suppose that J is G-invariant. Let a ∈ Iγ−1·x, and choose b ∈
Js(γ) = C0(p−1(s(γ)), I) such that b(γ−1 · x) = a. Then we have
α(γ,x)(a) = α(γ,x)(cid:0)b(γ−1 · x)(cid:1) = βγ(b)(x).
Since J is G-invariant, βγ(b) ∈ Jr(γ) = C0(p−1(r(γ)), I). Thus βγ(b)(x) ∈ Ix, so
α(γ,x)(Iγ−1·x) ⊆ Ix. Again, it is not hard to show that α(γ,x) maps Iγ−1·x onto Ix,
so I is (G ⋉ X)-invariant.
(cid:3)
16
SCOTT M. LALONDE
Proposition 3.10. Let (A, G ⋉ X, α) and (B, G, β) be as above. Let I ⊆ A be a
(G ⋉ X)-invariant ideal, and let J = Θ(I) be the corresponding G-invariant ideal
in B. Then the sequence
0 → J ⋊β,r G → B ⋊β,r G → B/J ⋊β,r G → 0
is exact if and only if
0 → I ⋊α,r (G ⋉ X) → A ⋊α,r (G ⋉ X) → A/I ⋊α,r (G ⋉ X) → 0
is exact.
Proof. It is straightforward to check that the diagram
0
0
/ J ⋊β,r G
B ⋊β,r G
B/J ⋊β,r G
0
/ I ⋊α,r (G ⋉ X)
/ A ⋊α,r (G ⋉ X)
/ A/I ⋊α,r (G ⋉ X)
/ 0
commutes, where the vertical arrows are the isomorphisms coming from Theorem
3.8. Thus one sequence is exact if and only if the other is.
(cid:3)
Theorem 3.11. Let G be a locally compact Hausdorff groupoid, and let X be a left
G-space. If G is exact, then the transformation groupoid G ⋉ X is also exact.
Proof. Let (A, G⋉X, α) be a separable groupoid dynamical system, and let I ⊆ A =
Γ0(X, A) be a (G ⋉ X)-invariant ideal. Then there is a dynamical system (B, G, β)
with A ⋊α,r (G ⋉ X) ∼= B ⋊β,r G and an isomorphism Θ : A → B = Γ0(G(0), B). The
ideal J = Θ(I) of B is G-invariant, and Proposition 3.10 gives us an isomorphism
of short sequences. If G is exact, then the top sequence is exact, so the bottom one
must be exact as well. It follows that G ⋉ X is exact.
(cid:3)
4. Subgroupoids
Throughout this section, let G denote a locally compact Hausdorff groupoid. We
say that a subgroupoid H ⊆ G is wide if H (0) = G(0). As an interesting corollary of
Theorem 3.11, we show that if G is exact, then any wide subgroupoid of G is also
exact. We then move on to other types of subgroupoids, which require different
techniques.
Let H ⊆ G be a closed, wide subgroupoid with open range and source maps.
Then H acts freely and properly on G by right translation, so the orbit space G/H
is locally compact and Hausdorff. As we will see in the next result, G acts naturally
on G/H by left translation, so we can form the transformation groupoid G ⋉ G/H.
Moreover, there is a natural free and proper action of G ⋉ G/H on the left of G,
which commutes with the natural H-action, and we have (G ⋉ G/H)\G ∼= G(0) =
H (0) and G/H ∼= (G ⋉ G/H)(0).
In other words, G implements an equivalence
between the groupoids H and G ⋉ G/H. This is particularly noteworthy, since
groupoid equivalence is known to preserve exactness [16, Theorem 4.8].
Theorem 4.1. Let G be a locally compact Hausdorff groupoid with a Haar system,
and let H ⊆ G be a closed, wide subgroupoid with open range and source maps.
Then the space G is a (G ⋉ G/H, H)-equivalence.
/
/
/
/
/
/
/
/
/
/
/
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
17
Proof. We have already mentioned that H acts freely and properly on the right of
G. The structure map rG : G → G/H for the left G ⋉ G/H-action is given by
rG(x) = [x]. Notice that this map is open by [17, Proposition 5.27] and clearly
descends to a homeomorphism G/H → G/H given by the identity map. Observe
then that (γ, [η]) ∈ G ⋉G/H and x ∈ G are composable whenever s(γ, [η]) = [γ−1η]
equals rG(x) = [x], which implies that r(x) = s(γ). Thus left translation in G
induces an action of G ⋉ G/H on G:
(γ, [η]) · x = γx.
To see that the action is free, suppose (γ, [η]) · x = x for some x ∈ G and (γ, [η]) ∈
G ⋉ G/H. Then we have γx = x, which implies γ = r(x). Thus (γ, [η]) = (r(x), [η])
is a unit in G ⋉ G/H, so the action is free.
To prove the action is proper, we appeal to [8, Proposition 1.84]. Suppose {xi}
and {(γi, [ηi])} are nets in G and G ⋉ G/H, respectively, such that xi → x and
(γi, [ηi]) · xi → y for some x, y ∈ G. Note that the latter condition really says that
γixi → y, so we can conclude that {γi} has a convergent subnet by [8, Proposition
1.84] since G acts properly on itself by left translation. Pass to this subnet, relabel,
and observe that since [xi] = [γ−1
i ηi] → [x].
But then since γi → γ for some γ ∈ G, we have
i ηi] for all i and xi → x, we have [γ−1
γi · [γ−1
i ηi] → γ · [x] = [γx],
or [ηi] → [γx]. Thus the subnet {(γi, [ηi])} converges to (γ, γx), so G is a proper
left (G ⋉ G/H)-space.
It is straightforward to see that the actions of G ⋉ G/H and H on G commute,
so we just need to check that the structure map sG : G → H (0) = G(0) descends
to a homeomorphism of (G ⋉ G/H)\G with H (0).
Indeed, since sG is open it
suffices to see that it induces a bijection. Certainly if x, y ∈ G lie in the same
(G ⋉ G/H)-orbit, then sG(x) = sG(y), so sG does induce a well-defined, surjective
map sG : (G ⋉ G/H)\G → H (0). Now suppose sG(x) = sG(y) for some x, y ∈ G.
Then we define γ = yx−1, and observe that
(γ, [γx]) · x = γx = y,
so x and y belong to the same (G ⋉ G/H)-orbit. Thus sG is injective. It follows
that sG is a homeomorphism. Therefore, G is a (G ⋉ G/H, H)-equivalence.
(cid:3)
In order to discuss exactness for a subgroupoid H, we need to require that H
is equipped with a Haar system. As a consequence of Theorem 4.1, together with
[25, Theorem 2.1], it suffices to require that H has open range and source maps.
Theorem 4.2. Let G be a locally compact Hausdorff groupoid with a Haar system,
and let H ⊆ G be a closed, wide subgroupoid with open range and source maps.
Then H possesses a Haar system.
Proof. If G has a Haar system, then the transformation groupoid G ⋉ G/H also
has a Haar system. Since the property of having a Haar system is preserved under
equivalence of groupoids by [25, Theorem 2.1], it follows from Theorem 4.1 that H
has a Haar system.
(cid:3)
Theorem 4.3. Let G be a locally compact Hausdorff groupoid with a Haar system,
and let H be a closed, wide subgroupoid with open range and source maps. If G is
exact, then H is exact.
18
SCOTT M. LALONDE
Proof. If G is exact, then Theorem 3.11 guarantees the transformation groupoid
G ⋉ G/H is exact. But G ⋉ G/H is equivalent to H by Theorem 4.1, which implies
that H is also exact by [16, Theorem 4.8].
(cid:3)
Corollary 4.4. Let G be a locally compact Hausdorff groupoid with a Haar system,
and assume G has continuously varying stabilizers. Then the isotropy groupoid
Iso(G) is exact.
Proof. Recall that G has continuously varying stabilizers if and only if Iso(G) has
open range and source maps, or equivalently if Iso(G) has a Haar system. Thus
Iso(G) is a wide subgroupoid of G with open range and source maps, so it is exact
by Theorem 4.3.
(cid:3)
Under certain conditions, we can extend Theorem 4.3 slightly. Suppose H is
only orbit-wide, meaning H (0) meets every orbit in G(0). Put X = s−1(H (0)). As
observed in [8, Example 1.63], X is saturated with respect to the source map, so
the restriction s : X → H (0) is open. It is then easy to see that the reduction GH(0)
acts freely and properly on the right of X. If the restriction of the range map to X
also happens to be open, then X is also a free and proper left G-space. Indeed, X
becomes a (G, GH(0) )-equivalence, as observed in [17, Example 5.33(7)]. However,
Muhly also notes in [17, Example 5.29(2)] that the restriction r : X → G(0) need
not be open in general. In cases where it is open, we have the following result.
Theorem 4.5. Let G be a locally compact Hausdorff groupoid with a Haar system,
and suppose H ⊆ G is a closed, orbit-wide subgroupoid with open range and source
maps. Let X = s−1(H (0)), and assume the range map rX : X → G(0) is open. If
G is exact, then so is H.
Proof. As observed in [17, Example 5.33(7)], X is a (G, GH(0) )-equivalence, since
r : X → G(0) is open. Therefore, GH(0) has a Haar system by [25, Theorem
2.1], and it follows from [16, Theorem 4.3] that GH(0) is exact. Now H is a wide
subgroupoid of GH(0) with open range and source maps, so Theorem 4.2 guarantees
that H has a Haar system and Corollary 4.3 implies that H is exact.
(cid:3)
Now we turn to the question of whether exactness passes to general subgroupoids.
That is, if H is a subgroupoid of an exact groupoid G, can we still conclude that H
is exact without assuming it is wide or orbit-wide? Armed with our result for wide
subgroupoids, it suffices to know whether reductions of exact groupoids are exact.
Specifically, we consider reductions to closed invariant subsets of G(0).
Theorem 4.6. Let G be a locally compact Hausdorff groupoid with a Haar system,
and suppose F ⊆ G(0) is a closed invariant set. If G is exact, then the reduction
H = GF is exact.
Proof. Since F is invariant, H has a Haar system by [8, Proposition 5.17]. Let
(A, H, α) be a separable groupoid dynamical system, and let A = Γ0(F, A). Since
A is a C0(F )-algebra, there is a continuous map σ : Prim A → F . By composing
with the inclusion map F ֒→ G(0), we obtain a continuous map Prim A → G(0), so
A is also a C0(G(0))-algebra. The action of C0(G(0)) on A is characterized by
for u ∈ F . Thus if u ∈ F , we have
(ϕ · a)(u) = ϕ(u)a(u)
Iu = C0,u(G(0)) · A = C0,u(F ) · A,
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
19
since restriction clearly yields a surjection C0,u(G(0)) → C0,u(F ). Hence A/Iu = Au
for u ∈ F . If u 6∈ F , we claim that Iu = A. Given ϕ ∈ Cc(F ), we can use the Tietze
extension theorem to find a function ψ ∈ Cc(G(0)) with ψ(u) = 0 and ψ(v) = ϕ(v)
for all v ∈ supp(ϕ). Then given a ∈ A, for all v ∈ F we have
(ψ · a)(v) = ψ(v)a(v) = ϕ(v)a(v).
Thus ψ · a = ϕ · a, and it follows that C0,u(G(0)) · A = C0(F ) · A = A. Therefore,
the upper semicontinuous C∗-bundle B → G(0) associated to A has fibers
Bu =(Au
0
if u ∈ F
if u 6∈ F.
Since F is invariant, it is easy to see that G acts on B via the family {βγ}γ∈G,
where βγ = αγ if γ ∈ H, and βγ is trivial otherwise. Then
B ⋊β,r G = A ⋊α,r H.
Suppose I ⊆ A is an H-invariant ideal. Then I, viewed as an ideal of B, is G-
invariant. If G is assumed to be exact, then the sequence
0 → I ⋊β,r G → B ⋊β,r G → B/I ⋊β,r G → 0
is exact. It is straightforward to check that the diagram
0
0
/ I ⋊β,r G
B ⋊β,r G
B/I ⋊β,r G
/ I ⋊β,r H
/ A ⋊β,r H
/ A/I ⋊β,r H
0
/ 0
commutes, and the vertical arrows are isomorphisms, so the bottom row is exact.
Since (A, H, α) was an arbitrary dynamical system, it follows that H is exact. (cid:3)
Corollary 4.7. Let G be an exact groupoid, and H ⊆ G a closed subgroupoid with
a Haar system. If H (0) is an invariant subset of G(0), then H is exact.
Proof. By Theorem 4.6, the reduction GH(0) is exact since H (0) is a closed invariant
set. Then H is a wide subgroupoid of GH(0) , hence it is exact by Theorem 4.3. (cid:3)
Corollary 4.8. Let G be an exact group bundle. Then for each u ∈ G(0), the fiber
Gu is an exact group.
Proof. For each u ∈ G(0), the singleton {u} is a closed invariant set, and the reduc-
tion to {u} is precisely Gu. Thus the result is immediate from Theorem 4.6.
(cid:3)
In the event that G has continuously varying stabilizers, we know that Iso(G)
is an exact group bundle by Corollary 4.4. We can now conclude that all of the
isotropy groups must be exact as well.
Corollary 4.9. Let G be an exact groupoid, and assume G has continuously varying
stabilizers. Then for each u ∈ G(0), the isotropy group Gu is exact.
If G does not have continuously varying stabilizers, the situation is a little
murkier. However, if we assume the orbit space is sufficiently nice, then we can still
argue that the isotropy groups are exact.
Corollary 4.10. Let G be an exact groupoid, and assume that the orbit space
G(0)/G is T1. Then for each u ∈ G(0), the isotropy group Gu is exact.
/
/
/
/
/
/
/
/
/
/
/
20
SCOTT M. LALONDE
Proof. The requirement that G(0)/G is T1 guarantees that orbits are closed in G(0).
Let F = [u] be the orbit of u. Then F is closed and invariant, so Theorem 4.6
guarantees that the reduction GF is exact. But GF is a transitive groupoid,
which is equivalent to the isotropy group Gu. Theorem 4.3 of [16] then implies
that Gu is exact.
(cid:3)
We can even get a partial result in the other direction. That is, we show that
the existence of a certain exact subgroupoid of G guarantees that G is exact. This
sort of result appears to have been observed in the group case in [1].
Proposition 4.11. Let G be a locally compact Hausdorff groupoid, and suppose
H ⊆ G is a closed, wide, amenable subgroupoid such that the map p : G/H → G(0)
defined by p([γ]) = r(γ) is proper. Then G is exact.
Proof. Since p : G/H → G(0) is proper, G/H is a fiberwise compact G-space, in
the sense of [2]. Moreover, the transformation groupoid G ⋉ G/H is equivalent to
H, hence amenable. Thus G acts amenably on a fiberwise compact space, so it is
amenable at infinity. It follows from [2, Proposition 6.7] that G is exact.
(cid:3)
5. A Partial Converse for Transformation Groupoids
We can actually extend Proposition 4.11 considerably to obtain a partial converse
to Theorem 3.11. In the group case, Proposition 4.11 follows from the theorem in
Section 7 of [12]. This section is devoted to proving an analogue of that result for
groupoids.
Theorem 5.1. Let G be a locally compact Hausdorff groupoid, and suppose X
is a left G-space such that the structure map p : X → G(0) is proper and the
transformation groupoid G ⋉ X is exact. Then G is exact.
The proof relies on the following setup. The results are largely the same as
the content of [2, Lemma 5.6], though we present the details in a slightly different
way. Let (A, G, α) be a separable groupoid dynamical system with A = Γ0(G(0), A)
exact. Recall that C0(X) is a C0(G(0))-algebra with fibers given by
C0(X)(u) = C0(p−1(u)) = C(p−1(u)),
since p−1(u) is compact for all u ∈ G(0). For brevity, we will let C denote the
associated upper semicontinuous C∗-bundle over G(0). By [19, Example 4.8], G acts
on C via left translation: for each γ ∈ G, we have an isomorphism ltγ : Cs(γ) → Cr(γ)
given by
ltγ(f )(x) = f (γ−1 · x).
It now follows from the discussion on page 919 of [13] and [7, Proposition 6.9] that
G acts on the balanced tensor product A ⊗G(0) C0(X). More precisely, G acts on
the associated upper semicontinuous C∗-bundle A ⊗G(0) C, which has fibers
(A ⊗G(0) C)u = Au ⊗max Cu ∼= C(p−1(u), Au).
The action is just the tensor product of the original actions. That is, there are
isomorphisms αγ ⊗ ltγ : (A ⊗G(0) C)s(γ) → (A ⊗G(0) C)r(γ) for each γ ∈ G, which
take the form
for f ∈ C(p−1(s(γ)), As(γ)). Thus (A ⊗G(0) C, G, α ⊗ lt) is a groupoid dynamical
system.
(αγ ⊗ ltγ)(f )(x) = αγ(cid:0)f (γ−1 · x)(cid:1)
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
21
On the other hand, we can form the pullback bundle p∗A → X, which admits
an action of G ⋉ X via
σ(γ,x)(a) = αγ(a).
It is straightforward to see that this defines a continuous action. For one, if (γ, η) ∈
G(2), then
σ(γ,x)(η,γ−1·x)(a) = σ(γη,x)(a) = αγη(a) = αγ(cid:0)αη(a)(cid:1) = σ(γ,x)(cid:0)σ(η,γ−1·x)(a)(cid:1).
Now suppose (γi, xi) → (γ, x) in G ⋉ X and ai → a in p∗A with ai ∈ p∗As(γi,xi) for
all i. By viewing p∗A as a subset of X×A, we can write (γ−1
·xi, ai) and (γ−1·x, a) in
place of ai and a, respectively. It is then clear that ai → a in A, so αγi(ai) → αγ(a)
in A since α is a continuous action. It follows that (xi, αγi(ai)) → (x, αγ(a)) in
p∗A, so σ is continuous.
i
Proposition 5.2. There is an isomorphism
Φ : p∗A ⋊σ,r (G ⋉ X) → (A ⊗G(0) C) ⋊α⊗lt,r G,
which is characterized by
for f ∈ Γc(G ⋉ X, r∗(p∗A)).
Φ(f )(γ)(x) = f (γ, x)
Proof. We really just need to show that our dynamical systems fall under the
purview of Theorems 3.6 and 3.8. Notice first that the pullback algebra can be
written as p∗A = A⊗G(0) C0(X), which is precisely the section algebra of the bundle
A ⊗G(0) C → G(0). Viewing p∗A as a C0(G(0))-algebra, the fibers are precisely
p∗Au = Γ0(p−1(u), p∗A) = C(p−1(u), Au) = (A ⊗G(0) C)u.
Thus the bundle A ⊗G(0) C → G(0) is exactly the bundle B from Theorem 3.6. The
action α ⊗ lt is exactly what we expect, too: for f ∈ C(p−1(s(γ)), As(γ)),
(α ⊗ lt)γ(f )(x) = αγ(cid:0)f (γ−1 · x)(cid:1) = σ(γ,x)(cid:0)f (γ−1 · x)(cid:1).
Therefore, the existence of the desired isomorphism follows as a special case of
Theorem 3.8.
(cid:3)
Notice that if A is exact, then p∗A = A ⊗G(0) C0(X) is exact. Therefore, if
the transformation groupoid G ⋉ X is also assumed to be exact, then the reduced
crossed product p∗A ⋊σ,r (G ⋉ X) is exact by [15, Theorem 6.14]. It then follows
from the isomorphism of Proposition 5.2 that (A ⊗G(0) C) ⋊α⊗lt,r G is an exact
C∗-algebra.
The next step is to show that A ⋊α,r G embeds into (A ⊗G(0) C) ⋊α⊗lt,r G. Since
the fibers of p : X → G(0) are compact, each fiber Cu = C(p−1(u)) is unital. Thus
we have fiberwise embeddings Au ֒→ Au ⊗max Cu given by a 7→ a ⊗ 1.
Indeed,
we claim that these homomorphisms yield a continuous C∗-bundle homomorphism
ι : A ֒→ A ⊗G(0) C. To prove it, we will show that there is a C0(G(0))-linear
embedding ι : A → A ⊗G(0) C0(X).
Proposition 5.3. There is an injective C0(G(0))-linear homomorphism ι : A →
A ⊗G(0) C0(X) characterized by
for all u ∈ G(0). Furthermore, ι is G-equivariant.
ι(a)(u) = a(u) ⊗ 1
22
SCOTT M. LALONDE
Proof. To construct ι, we will first show that there is an embedding of A into the
pullback algebra p∗A = Γ0(X, p∗A) and then compose with the natural isomor-
phism p∗A ∼= A ⊗G(0) C0(X). We need to appeal to [14, Proposition 3.4.4]. Given
a ∈ A, define a map f : X → A by
f (x) = a(p(x)).
Then f is clearly continuous:
if xi → x in X, then p(xi) → p(x) in G(0), so
a(p(xi)) → a(p(x)) since a ∈ Γ0(G(0), A). Moreover, it is straightforward to show
that f vanishes at infinity. Let ε > 0 be given, and let K = {u ∈ G(0) : ka(u)k ≥ ε}.
By definition we have k f (x)k = ka(p(x))k for all x ∈ X, so
{x ∈ X : k f (x)k ≥ ε} = p−1(K)
is compact, since we have assumed that p : X → G(0) is a proper map. Therefore,
the section p∗a of p∗A → X defined by
(2)
p∗a(x) = (x, f (x)) = (x, a(p(x)))
belongs to Γ0(X, p∗A) by [14, Proposition 3.4.4].
Now we check that the map a 7→ p∗a is an injective homomorphism. Well, it is
immediate from (2) that the map is a homomorphism. If p∗a = 0 for some a ∈ A,
then we have a(p(x)) = 0 for all x ∈ X. Since p is surjective, this means a = 0,
and the map is injective. If we let Θ : p∗A → A ⊗G(0) C0(X) denote the natural
isomorphism afforded by Proposition 3.1, then we have
Θ(p∗a)(u)(x) = p∗a(x) = a(p(x)) = a(u)
for all u ∈ G(0) and x ∈ p−1(u). That is, Θ(p∗a)(u) (viewed as an element of
C(p−1(u), Au)) is the constant function x 7→ a(u), which we identify with the
elementary tensor a(u) ⊗ 1. Thus our embedding ι : A → A ⊗G(0) C0(X) takes the
desired form:
ι(a)(u) = Θ(p∗a)(u) = a(u) ⊗ 1
for all u ∈ G(0).
It remains to see that ι is C0(G(0))-linear. Let a ∈ A and ϕ ∈ C0(G(0)). Then
for all u ∈ G(0), we have
ι(ϕ · a)(u) = (ϕ · a)(u) ⊗ 1
= ϕ(u)a(u) ⊗ 1
= ϕ(u)ι(a)(u)
= ϕ(u)(cid:0)a(u) ⊗ 1(cid:1)
=(cid:0)ϕ · ι(a)(cid:1)(u),
so ι(ϕ · a) = ϕ · ι(a). Thus ι is C0(G(0))-linear. Consequently, it induces fiberwise
homomorphisms ιu : Au → Au ⊗max Cu, which are given by ιu(a) = a ⊗ 1.
Finally, we check that ι is G-equivariant. This amounts to verifying that for all
a ∈ A and all γ ∈ G,
On the left hand side we have
ιr(γ)(cid:0)αγ(a)(cid:1) = (α ⊗ lt)γ(cid:0)ιs(γ)(a)(cid:1).
ιr(γ)(cid:0)αγ(a)(cid:1) = αγ(a) ⊗ 1,
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
23
while on the right,
(α ⊗ lt)γ(cid:0)ιs(γ)(a)(cid:1) = (α ⊗ lt)γ(cid:0)a ⊗ 1(cid:1)
= αγ(a) ⊗ ltγ(1)
= αγ(a) ⊗ 1
Thus ι is G-equivariant.
= ιr(γ)(cid:0)αγ(a)(cid:1).
(cid:3)
Since ι : A → A ⊗G(0) C0(X) is G-equivariant, [15, Proposition 6.3] guaran-
tees that there is a homomorphism ι ⋊ id : Γc(G, r∗A) → Γc(G, r∗(A ⊗G(0) C))
characterized by
ι ⋊ id(f )(γ) = ιr(γ)(f (γ)) = f (γ) ⊗ 1.
Moreover, ι ⋊ id extends to a homomorphism between the associated full crossed
products. Proposition 6.10 of [15] then implies that there is also a homomorphism
ι⋊id : A⋊α,r G → (A⊗G(0) C)⋊α⊗id,r G at the level of the reduced crossed products.
All that remains is to prove that ι ⋊ id is injective.
Proposition 5.4. There is an injective homomorphism
ι ⋊ id : A ⋊α,r G → (A ⊗G(0) C) ⋊α⊗lt,r G,
which is characterized by
for f ∈ Γc(G, r∗A).
ι ⋊ id(f )(γ) = f (γ) ⊗ 1,
Proof. We first need to check that ι ⋊ id : Γc(G, r∗A) → Γc(G, r∗(A ⊗G(0) C)) is
injective. Suppose ι ⋊ id(f ) = 0 for some f ∈ Γc(G, r∗A). Then for all γ ∈ G,
ι ⋊ id(f )(γ) = ιr(γ)(f (γ)) = 0,
and since ιr(γ) is injective, f (γ) = 0. Thus f = 0, and ι ⋊ id is injective on
Γc(G, r∗A). Now the argument following the proof of [15, Proposition 6.10] shows
that
kι ⋊ id(f )kr = kf kr
for all f ∈ Γc(G, r∗A). Hence ι ⋊ id is isometric, so it extends to an injective
homomorphism ι ⋊ id : A ⋊α,r G → (A ⊗G(0) C) ⋊α⊗lt,r G.
(cid:3)
Proof of Theorem 5.1. Let (A, G, α) be a separable groupoid dynamical system and
let I ⊆ A be a G-invariant ideal.
If we form the pullback dynamical system
(p∗A, G ⋉ X, σ), then the pullback C∗-algebra p∗I is a (G ⋉ X)-invariant ideal
of p∗A by [16, Proposition 3.2]. Therefore, we obtain a short exact sequence
0 → p∗I ⋊σ,r (G ⋉ X) → p∗A ⋊σ,r (G ⋉ X) → p∗(A/I) ⋊σ,r (G ⋉ X) → 0
since G⋉X is exact. By Proposition 5.4, we have embeddings of I ⋊α,r G, A⋊α,r G,
and A/I ⋊α,rG into the respective pullback crossed products. This yields a diagram
0
0
/ I ⋊α,r G
A ⋊α,r G
A/I ⋊α,r G
0
/ p∗I ⋊σ,r (G ⋉ X)
/ p∗A ⋊σ,r (G ⋉ X)
/ p∗(A/I) ⋊σ,r (G ⋉ X)
/ 0
/
/
/
/
/
/
/
/
/
/
/
24
SCOTT M. LALONDE
which is easily seen to commute. If a belongs to the kernel of the map A ⋊α,r G →
A/I ⋊α,rG, then ι⋊id(a) is in the kernel of p∗A⋊σ,r(G⋉X) → p∗(A/I)⋊σ,r(G⋉X).
Hence
ι ⋊ id(a) ∈ ι ⋊ id(A ⋊α,r G) ∩ p∗I ⋊σ,r (G ⋉ X).
We claim that ι ⋊ id(a) ∈ ι ⋊ id(I ⋊α,r G). It will suffice to show that
ι ⋊ id(A ⋊α,r G) ∩ (I ⊗G(0) C) ⋊α⊗lt,r G = ι ⋊ id(I ⋊α,r G).
Suppose first that f ∈ Γc(G, r∗A) and ι ⋊ id(f ) ∈ (I ⊗G(0) C) ⋊α⊗lt,r G. Then we
have
ι ⋊ id(f )(γ) = f (γ) ⊗ 1 ∈ Ir(γ) ⊗max Cr(γ)
for all γ ∈ G, meaning that f (γ) ∈ Ir(γ) for all γ ∈ G. That is, f ∈ Γc(G, r∗I). It
follows that
ι ⋊ id(A ⋊α,r G) ∩ (I ⊗G(0) C) ⋊α⊗lt,r G ⊆ ι ⋊ id(I ⋊α,r G).
The other containment is clear, so we have the desired equality. Therefore, a ∈
I ⋊α,r G, and the top row of the diagram above is exact. Thus G is exact.
(cid:3)
Corollary 5.5. Let G be a locally compact Hausdorff groupoid, and suppose H ⊆ G
is a closed, wide, exact subgroupoid such that the map p : G/H → G(0) is proper.
Then G is exact.
Proof. Since H is exact, the transformation groupoid G ⋉ G/H is exact. Since the
map p : G/H → G(0) is proper, it follows from Theorem 5.1 that G is exact.
(cid:3)
Remark 5.6. By taking H = G, we can see that the converse of Corollary 5.5
holds as well. In this case we have G/H ∼= G(0), so the map p : G/H → G(0) is
certainly proper, and G ⋉ G/H ∼= G. Thus G is exact if and only if G ⋉ G/H is.
The setup of Theorem 5.1 is reminiscent of the notion of amenability at infinity
for groupoids. Recall that a groupoid G is amenable at infinity if there is a G-space
X such that p : X → G(0) is proper and G ⋉ X is an amenable groupoid. It is
known that if G is amenable at infinity, then G is exact [2, Proposition 6.7]. The
converse is known to hold if G is assumed to be weakly inner amenable, as defined
by Anantharaman-Delaroche in [2, Definition 4.2]. It is unknown whether exactness
implies amenability at infinity in general. (The two properties are now known to
be equivalent for groups by [6], however.) Thanks to Theorem 5.1, we have another
way of phrasing this open question.
Question 1. If a groupoid G acts on a fiberwise compact space X → G(0) such
that G ⋉ X is an exact groupoid, must G act amenably on some fiberwise compact
space?
6. Inner Exactness
We now conclude with a brief discussion inner exactness, as introduced by
Anantharaman-Delaroche in [3]. A locally compact groupoid G is said to be inner
exact if the sequence
0 → C∗
r (GG(0)\U ) → 0
r (GU ) → C∗
r (G) → C∗
is exact for any open invariant set U ⊆ G(0). It is clear that any exact groupoid
is inner exact. However, this condition is strictly weaker than exactness, since
any locally compact group is automatically inner exact. The importance of inner
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
25
exactness became clear long before it was formally defined in [3], since many of
the known examples of non-exact groupoids actually fail to be inner exact.
In
particular, the non-exact HLS groupoids constructed by Higson, Lafforgue, and
Skandalis in [9] as counterexamples to the Baum-Connes conjecture are not inner
exact.
It is natural to ask which of the known permanence properties for exact groupoids
translate over to inner exact groupoids. We begin our discussion with one such
result, which is a version of [16, Theorem 4.8] for inner exact groupoids.
Theorem 6.1. Let G and H be groupoids, and suppose Z is a (G, H)-equivalence.
If G is inner exact, then so is H.
Proof. Let sZ : Z → H (0) and rZ : Z → G(0) denote the structure maps for the G-
and H-actions on Z, respectively, and suppose U ⊆ H (0) is an open, H-invariant
set. Then s−1
Z (U ) is open in Z, and since the range map rZ : Z → G(0) is open,
V = rZ (s−1
Z (U )) is open in G(0). We claim that V is also G-invariant. Suppose
u ∈ V and γ ∈ G with s(γ) = u. Choose z ∈ s−1
Z (U ) with rZ(z) = u. Then
sZ (γ · z) = sZ(z) ∈ U,
so γ · z ∈ s−1
G-invariant.
Z (U ). This implies that rZ (γ · z) = r(γ) ∈ rZ (s−1
Z (U )) = V , so V is
Now let L = G ⊔ Z ⊔ Z op ⊔ H denote the linking groupoid for the (G, H)-
equivalence Z, as defined in [22, Lemma 2.1]. We claim that U ∪V is an L-invariant
subset of L(0). Clearly we just need to consider the elements of L coming from Z and
Z op. Suppose first that z ∈ Z with s(z) ∈ U ∪ V . Then sZ (z) ∈ U , so z ∈ s−1
Z (U ),
and we have rZ (z) ∈ V by construction. A similar argument works for elements of
Z op, so U ∪ V is invariant in L(0).
Since U ∪ V is an open, invariant subset of L(0), we can form the sequence of
reduced C∗-algebras
r (L(U∪V )c ) → 0.
0 → C∗
r (LU∪V ) → C∗
r (L) → C∗
We claim that this sequence is exact if and only if
r (G) → C∗
r (GV ) → C∗
0 → C∗
r (GV c) → 0
(3)
(4)
is exact, or similarly if and only if
(5)
0 → C∗
r (HU ) → C∗
r (H) → C∗
r (HU c ) → 0
is exact. To prove it, we consider the dynamical system (C0(L(0)), L, lt). The ideal
C0(U ∪ V ) of C0(L(0)) is L-invariant, and it is easy to see that
C0(U ∪ V ) ∩ C0(G(0)) = C0(V ),
which is a G-invariant ideal in C0(G(0)). Likewise,
C0(U ∪ V ) ∩ C0(H (0)) = C0(U )
is an H-invariant ideal in C0(H (0)). It then follows from [16, Corollary 4.6] that the
sequences (4) and (5) are exact if and only if (3) is exact. Thus if G is assumed to
be inner exact, then H must be as well.
(cid:3)
We can now use Theorem 6.1 to establish an analogue of Corollary 5.5 for inner
exactness.
26
SCOTT M. LALONDE
Theorem 6.2. Let G be a locally compact Hausdorff groupoid, and suppose H is a
closed, wide, inner exact subgroupoid such that the map p : G/H → G(0) is proper.
Then G is inner exact.
Proof. Since H is inner exact, the transformation groupoid G ⋉ G/H is inner exact
by Theorem 6.1. It is easily checked that Proposition 5.4 guarantees that there is
an embedding C∗
r (G ⋉ G/H) since p : G/H → G(0) is proper. Suppose
U ⊆ G(0) is open and invariant, and set V = p−1(U ). Then V is open in G/H, and
if [x] ∈ V and γ ∈ G with s(γ) = p([x]) = r(x), then
r (G) ֒→ C∗
p(γ · [x]) = p([γx]) = r(γx) = r(γ) ∈ U
since U is invariant. Thus γ · [x] ∈ V , so V is G-invariant. Consequently, the
sequence
0 → C∗
r ((G ⋉ G/H)V ) → C∗
r (G ⋉ G/H) → C∗
r ((G ⋉ G/H)V c ) → 0
is exact since G ⋉ G/H is inner exact. We then get a commuting diagram like the
one in the proof of Theorem 5.1,
0
0
/ C∗
r (GU )
C∗
r (G)
C∗
r (GU c )
0
/ C∗
r ((G ⋉ G/H)V )
/ C∗
r (G ⋉ G/H)
/ C∗
r ((G ⋉ G/H)V c)
/ 0
where the vertical arrows are injective. The result then follows from the same
diagram-chasing argument as in the proof of Theorem 5.1.
(cid:3)
Next we investigate a negative result on quotients of inner exact groupoids. As
Ozawa has observed in [20], quotients of exact groups need not be exact. Thus the
question of whether exactness descends to quotients of groupoids is already settled.
In the next example, we observe that the situation can actually be much worse: a
quotient of an exact groupoid by a closed, normal subgroupoid need not be inner
exact. Naturally enough, Willett's example of a non-amenable groupoid with weak
containment lies at the heart of the discussion.
Example 6.3. Let G = F2 × N, where F2 denotes the free group on two generators
and N is the one-point compactification of the natural numbers. We view G as
a group bundle over N. Notice that G is the transformation groupoid associated
to the trivial action of the exact group F2 on N, so it is exact by Theorem 3.11.
Now let {Kn}∞
n=1 be the sequence of subgroups of F2 described in [23, Lemma 2.8],
which are defined to be
Kn =\ ker ϕ,
Kn
H = [n∈N
where ϕ ranges over all group homomorphisms ϕ : F2 → Γ with Γ ≤ n. Then
each Kn is a finite index, normal subgroup of F2, the subgroups are nested, and
n=1 Kn is trivial. If we let K∞ be the trivial subgroup, then the bundle
T∞
is a clopen, wide, normal subgroupoid of G. Hence H is exact by Theorem 4.3.
However, it is easy to see that H encodes the equivalence relation on G described
in [9, Section 2], so the quotient G/H is precisely the HLS groupoid constructed
by Willett in [23]. But G/H is not inner exact by construction.
/
/
/
/
/
/
/
/
/
/
/
ON SOME PERMANENCE PROPERTIES OF EXACT GROUPOIDS
27
Note that it is not necessary to work exclusively with Willett's groupoid here.
Any non-inner exact HLS groupoid (such as the ones described in Section 2 of [9]
and Section 8.4 of [2]) will do.
The situation for transformation groupoids and subgroupoids is murkier. Thanks
to an example of Baum, Guentner, and Willett from [5], we can see that the trans-
formation groupoid associated to a group action might fail to be inner exact. Given
a Gromov monster group Γ, the authors construct an exact sequence of separable
commutative Γ-C∗-algebras
0 → C0(Γ) → C0(Z) → C0(∂Z) → 0
such that the sequence of reduced crossed products
(6)
0 → C0(Γ) ⋊r Γ → C0(Z) ⋊r Γ → C0(∂Z) ⋊r Γ → 0
is not exact [5, Theorem 7.4]. As with the HLS examples in [9], the failure of
exactness is realized at the level of K-theory. If we define G = Γ ⋉ Z, then (6) is
isomorphic to the sequence
0 → C∗
r (G∂Z ) → 0,
r (GΓ) → C∗
and it follows that G is not inner exact.
r (G) → C∗
The above discussion shows that the analogue of Theorem 3.11 for inner exact-
ness does not hold. This suggests that one would need a different approach to show
that exactness descends to wide subgroupoids. It may be possible to simply weaken
the hypotheses of Theorem 3.11 -- it would clearly suffice to show that if G is inner
exact and H ⊆ G is a wide subgroupoid, then G ⋉ G/H is inner exact. In any
event, the example described above seems to cast a pall over the whole situation,
and we simply pose the following open question.
Question 2. If G is inner exact and H ⊆ G is a wide subgroupoid, must H also
be inner exact?
We close with another question that may be of interest to others, which is inspired
in part by the aforementioned examples of HLS groupoids. Suppose G is a group
bundle. For G to be exact, it is necessary that each fiber is an exact group by
Corollary 4.8. This condition is not sufficient, as the non-exact HLS examples
show. However, those examples fail to be exact precisely because they are not
inner exact. Perhaps these two conditions together guarantee exactness.
Question 3. Let G be a group bundle. If G is inner exact and Gu is an exact
group for all u ∈ G(0), must G be exact?
Acknowledgements
The author would like to thank Jean Renault for useful suggestions, and Dana
Williams for providing helpful comments during the preparation of this paper.
References
1. Claire Anantharaman-Delaroche, Amenability and exactness for dynamical systems and their
C ∗-algebras, Trans. Amer. Math. Soc. 354 (2002), no. 10, 4153 -- 4178.
2.
3.
, Exact groupoids, arXiv:1605.05117v1 [math.OA], May 2016.
, Some remarks about the weak containment property for groupoids and semigroups,
arXiv:1604.01724v3 [math.OA], May 2016.
28
SCOTT M. LALONDE
4. Claire Anantharaman-Delaroche and Jean Renault, Amenable groupoids, Monographies de
l'Enseignement Math´ematique, vol. 36, L'Enseignement Math´ematique, Geneva, 2000.
5. Paul Baum, Erik Guentner, and Rufus Willett, Exactness and the Kadison-Kaplansky conjec-
ture, Operator algebras and their applications, Contemp. Math., vol. 671, Amer. Math. Soc.,
Providence, RI, 2016.
6. Jacek Brodzki, Chris Cave, and Kang Li, Exactness of locally compact groups, Adv. Math.
312 (2017), 209 -- 233.
7. Jonathan Henry Brown, Proper actions of groupoids on C ∗-algebras, Ph.D. thesis, Dartmouth
College, Hanover, NH, May 2009.
8. Geoff Goehle, Groupoid crossed products, Ph.D. thesis, Dartmouth College, Hanover, NH,
May 2009, arXiv:0905.4681v1 [math.OA].
9. N. Higson, V. Lafforgue, and G. Skandalis, Counterexamples to the Baum-Connes conjecture,
GAFA, Geom. funct. anal. 12 (2002), 330 -- 354.
10. Marius Ionescu and Dana P. Williams, Remarks on the ideal structure of Fell bundle C ∗-
algebras, Houston J. Math. 38 (2012), no. 4, 1241 -- 1260.
11. Eberhard Kirchberg and Simon Wassermann, Exact groups and continuous bundles of C ∗-
algebras, Math. Ann. 315 (1999), no. 2, 169 -- 203.
12.
13. Alexander Kumjian, Paul S. Muhly, Jean N. Renault, and Dana P. Williams, The Brauer
, Permanence properties of C ∗-exact groups, Doc. Math. 4 (1999), 513 -- 558.
group of a locally compact groupoid, Amer. J. Math. 120 (1998), 901 -- 954.
14. Scott M. LaLonde, Nuclearity and exactness for groupoid crossed product C ∗-algebras, Ph.D.
thesis, Dartmouth College, Hanover, NH, June 2014.
15.
, Nuclearity and exactness for groupoid crossed products, J. Operator Theory 74
(2015), no. 1, 213 -- 245.
, Equivalence and exact groupoids, Houston J. Math. 42 (2016), no. 4, 1267 -- 1290.
16.
17. Paul Muhly, Coordinates in operator algebras, In continuous preparation.
18. Paul Muhly and Dana P. Williams, Equivalence and disintegration theorems for Fell bundles
and their C ∗-algebras, Dissertationes Math. (Rozprawy Mat.) 456 (2008), 1 -- 57.
19. Paul S. Muhly and Dana P. Williams, Renault's equivalence theorem for groupoid crossed
products, NYJM Monographs, vol. 3, State University of New York, University at Albany,
Albany, NY, 2008.
20. Narutaka Ozawa, Amenable actions and applications, Proceedings of the International Con-
gress of Mathematicians (Zurich), vol. II, Eur. Math. Soc., 2006.
21. Iain Raeburn and Dana P. Williams, Morita equivalence and continuous-trace C ∗-algebras,
Math. Surveys Monogr., no. 60, American Mathematical Society, Providence, RI, 1998.
22. Aidan Sims and Dana P. Williams, Renault's equivalence theorem for reduced groupoid C ∗-
algebras, J. Operator Theory 68 (2012), no. 1, 223 -- 239.
23. Rufus Willett, A non-amenable groupoid whose maximal and reduced C ∗-algebras are the
same, Munster J. Math. 8 (2015), no. 1, 241 -- 252.
24. Dana P. Williams, Crossed products of C ∗-algebras, Math. Surveys Monogr., no. 134, Ameri-
can Mathematical Society, Providence, RI, 2007.
25.
, Haar systems on equivalent groupoids, Proc. Amer. Math. Soc. Ser. B 3 (2016), 1 -- 8.
Department of Mathematics, The University of Texas at Tyler, 3900 University
Boulevard, Tyler, TX 75799
E-mail address: [email protected]
|
0910.3299 | 2 | 0910 | 2010-06-07T05:05:01 | Crossed products by \alpha-simple automorphisms on C*-algebras C(X,A) | [
"math.OA"
] | Let $X$ be a Cantor set, and let $A$ be a unital separable simple amenable $C$*-algebra with tracial rank zero which satisfies the Universal Coefficient Theorem, we use $C(X,A)$ to denote the set of all continuous functions from $X$ to $A$, let $\alpha$ be an automorphism on $C(X,A)$. Suppose that $C(X,A)$ is $\alpha$-simple and $[\alpha]=[\mbox{id}_{1\otimes A}]$ in $KL(1\otimes A,1\otimes A)$, we show that $C(X,A)\rtimes_{\alpha}\mathbb{Z}$ has tracial rank zero. | math.OA | math |
CROSSED PRODUCTS BY α-SIMPLE AUTOMORPHISMS
ON C*-ALGEBRAS C(X,A)
JIAJIE HUA
Abstract. Let X be a Cantor set, and let A be a unital separable
simple amenable C*-algebra with tracial rank zero which satisfies the
Universal Coefficient Theorem, we use C(X, A) to denote the set of
all continuous functions from X to A, let α be an automorphism on
C(X, A). Suppose that C(X, A) is α-simple and [α] = [id1⊗A] in KL(1⊗
A, 1 ⊗ A), we show that C(X, A) ⋊α Z has tracial rank zero.
1. Introduction
Transformation group C*-algebras of minimal homeomorphisms of the
Cantor set are AT algebras (direct limits of circle algebras) with real rank
zero is implicit in Section 8 of [9], with the main step having been done in
[22]. Elliott and Evans proved in [5] that the irrational rotation algebras
are AT algebras with real rank zero, then Lin proved that irrational higher
dimensional noncommutative tori of the form C(T k) ⋊θ Z are in fact AT
algebras ([14]). Recently Phillips constructs an inductive proof that every
simple higher noncommutative torus is an AT algebra ([19]). More generally,
Lin and Phillips proved the following result ([17]) : Let X be an infinite
compact metric space with finite covering dimensional and let α: X → X
be a minimal homeomorphism, the associated crossed product C*-algebra
A = C(X) ⋊α Z has tracial rank zero whenever the image of K0(A) in
Aff(T (A)) is dense.
In particular, these algebras all belong to the class
known currently to be classifiable by K-theoretic invariants in the sense of
the Elliott classification program ([4]).
On the other hand, Lin proved in [16] that let A be a unital separable
simple C*-algebra with tracial rank zero and let α be an automorphism on
A. Suppose that α has certain Rokhlin property and there is an integer
J ≥ 1 such that [αJ ] = [idA] in KL(A, A). Then A ⋊α Z has tracial rank
zero.
In present, we don't know what happen when C*-algebras are nor commu-
tative neither simple. In this paper, let X be a Cantor set, let A be a unital
2000 Mathematics Subject Classification. Primary 46L55: Secondary 46L35, 46L40.
Key words and phrases. α-simple; crossed products; tracial rank zero.
The author was supported the National Natural Science Foundation of China (Nos.
10771069, 10671068,10771161).
1
2
JIAJIE HUA
separable simple amenable C*-algebra with tracial rank zero which satisfies
the Universal Coefficient Theorem, we consider the C*-algebra C(X, A), all
continuous functions from X to A. When A is isomorphic to C, it is just
the case in [22]. When C(X, A) is not isomorphic to C, C(X, A) is neither
commutative nor simple, so it is different from above cases and contains
Cantor set case. We prove the following result: let X be a Cantor set, let
A be a unital separable simple amenable C*-algebra with tracial rank zero
which satisfies the Universal Coefficient Theorem, and let α be an automor-
phism on C(X, A). Suppose that C(X, A) is α-simple and [α] = [id1⊗A] in
KL(1 ⊗ A, 1 ⊗ A), we present a proof that C(X, A) ⋊α Z has tracial rank
zero, therefore they are determined by their graded ordered K-theory.
The condition C(X, A) is α-simple is necessary to guarantee that C(X, A)⋊α
Z is simple, because in present classification theorems we have is mainly for
simple C*-algebras. The second condition [α] = [id1⊗A] in KL(1 ⊗ A, 1 ⊗ A)
means action of α take K-theory of each fiber of X invariant. An early ver-
sion of second condition is weaker and can not ensure approximate unitary
equivalence of the action on fiber which was pointed us by Hiroki Matui.
This paper is organized as follows. In Section 2 we introduce notation and
give some elementary properties of α-simple automorphisms on C*-algebras
C(X, A). In Section 3, we prove that, under our hypotheses, C(X, A) ⋊α Z
has tracial rank zero.
2. Notation and α-simple C*-algebras
We will use the following convention:
(1) Let A be a C*-algebra, let a ∈ A be a positive element and let p ∈ A
be a projection. We write [p] ≤ [a] if there is a projection q ∈ aAa and a
partial isometry v ∈ A such that v∗v = p and vv∗ = q.
(2) Let A be a C*-algebra. We denote by Aut(A) the automorphism
group of A. If A is unital and u ∈ A is a unitary, we denote by adu the
inner automorphism defined by adu(a) = u∗au for all a ∈ A.
(3) Let x ∈ A, ε > 0 and F ⊂ A. We write x ∈ε F, if dist(x, F) < ε, or
there is y ∈ F such that kx − yk < ε.
(4) Let A be a C*-algebra and α ∈ Aut(A). We say A is α-simple if A
does not have any non-trivial α-invariant closed two-sided ideals.
(5) Let A be a unital C*-algebra and T (A) the compact convex set of
tracial states of A. If α is an automorphism of A, we use T α(A) to denote
the α-invariant tracial states, which is again a compact convex set. We define
an affine mapping r of T (A ⋊α Z) into T α(A) by the restriction r(τ ) = τ A.
We recall the definition of tracial topological rank of C*-algebras.
Definition 2.1. [11] Let A be a unital simple C*-algebra. Then A is said
to have tracial (topological) rank zero if for any ε > 0, any finite set F ⊂ A
and any nonzero positive element a ∈ A, there exists a finite dimensional
C*-subalgebra B ⊂ A with idB = p such that:
(1) kpx − xpk < ε for all x ∈ F.
CROSSED PRODUCTS BY α-SIMPLE AUTOMORPHISMS ON C*-ALGEBRAS
3
(2) pap ∈ε B for all x ∈ F.
(3) [1 − p] ≤ [a].
If A has tracial rank zero, we write TR(A) = 0.
Definition 2.2. Let X be a compact metric space and let A be a C*-algebra,
we say a map β : X → Aut(A) is strongly continuous if for any {xn} with
d(xn, x) → 0 when n → ∞, we have kβxn(a) − βx(a)k → 0 for all a ∈ A.
Lemma 2.3. Let X be a compact metric space, let A be a unital simple
C*-algebra and α ∈Aut(C(X, A)). Then C(X, A) is α-simple if and only if
there is a minimal homeomorphism σ from X to X and there is a strongly
continuous map β from X to Aut(A), denote by x to βx, such that α(f )(x) =
βσ−1(x)(f (σ−1(x))).
Proof. We firstly prove that the set consists of all ideals of C(X, A) is same
as the one consists of {f ∈ C(X, A)f (x) = 0 for any x ∈ Y } for all closed
subset Y ⊂ X.
Clearly, {f ∈ C(X, A)f (x) = 0 for any x ∈ Y } is an ideal of C(X, A) for
any closed subset Y ⊂ X.
Conversely, let I be an ideal of C(X, A), and let XI = ∩f ∈I {x ∈ Xf (x) =
0} be a closed subset of X, We claim I = {f ∈ C(X, A)f (x) = 0 for any x ∈
XI }. Clearly, I ⊆ {f ∈ C(X, A)f (x) = 0 for any x ∈ XI }.
In the following we will prove {f ∈ C(X, A)f (x) = 0 for any x ∈ XI } ⊆
I.
1, . . . , k such that h(x0) = Pk
For any ε > 0 and any x0 /∈ XI , there is a f ∈ I such that f (x0) 6= 0,
thus for any h ∈ {f ∈ C(X, A)f (x) = 0 for any x ∈ XI }, since A is a
unital simple C*-algebra, by Lemma 3.3.6 of [12], there exists ai ∈ A, i =
i , since
f (x) ∈ I, we have g(x) ∈ I, then there is a neighborhood O(x0) of x0 such
that kg(y) − h(y)k < ε for all y ∈ O(x0). Since X is compact, there exist
gi(x) ∈ I and xi ∈ X, i = 1, · · · , n, such that {O(xi)}n
i=1 cover X and
kgi(x) − h(x)k < ε for all x ∈ O(xi).
i . Let g(x) = Pk
i=1 aif (x0)a∗
i=1 aif (x)a∗
Let {f1, · · · , fn} be a subset of nonnegative functions in C(X)(a partition
of the unit) satisfying the following conditions:
(1) fi(x) = 0 for all x /∈ O(xi), i = 1, 2, · · · , n and
i=1 fi(x) = 1 for all x ∈ X.
(2) Pn
We compute
k
n
Xi=1
gi(x)fi(x) − h(x)k ≤
n
Xi=1
kgi(x) − h(x)kfi(x) < ε,
then we have h ∈ I, so {f ∈ C(X, A)f (x) = 0 for any x ∈ XI } ⊆ I. The
claim follows.
From above, we know that the set of all the maximal ideals of C(X, A) is
{Ix}x∈X, where Ix = {f ∈ C(X, A)f (x) = 0}. Given x ∈ X, because α is an
automorphism, there is y ∈ X such that α(Ix) = Iy. We define σ : X → X
by x to y and βσ(x) : A → A by α(C(X, A)/Ix) = C(X, A)/Iσ(x), βx(a) =
4
JIAJIE HUA
α(a)(σ(x)). For any fixed y ∈ X and f ∈ C(X, A), then α(f )(y) = α((f −
f (σ−1(y))) + f (σ−1(y)))(y) = α(f (σ−1(y)))(y).
So α(f )(y) = βσ−1(y)(f (σ−1(y))) for all y ∈ X.
For any a ∈ A, if d(xn −x) → 0 when n → ∞, we have kβxn(a)−βx(a)k =
kα(a)(σ(xn)) − α(a)(σ(x))k → 0 when n → ∞.
For any a, b ∈ A and x ∈ X
βx(a + b) = α(a + b)(σ(x)) = α(a)(σ(x)) + α(b)(σ(x)) = βx(a) + βx(b),
βx(a · b) = α(a · b)(σ(x)) = α(a)(σ(x)) · α(b)(σ(x)) = βx(a) · βx(b),
βx(a∗) = α(a∗)(σ(x)) = α(a)∗(σ(x)) = βx(a)∗,
so βx is a homomorphism,
Since
βx(α−1(b)(x)) = βx((α−1(b)(x) − α−1(b)) + α−1(b)) = α(α−1(b))(σ(x)) = b
and A is simple, we have βx is an automorphism of A for any x ∈ X.
If there exist x1, x2 ∈ X, x1 6= x2 such that σ(x1) = σ(x2), it is easy to
find f ∈ C(X, A) such that f (x1) = 1, f (x2) = 0. So
α(f )(σ(x1)) = βx1(f (x1)) = 1
α(f )(σ(x2)) = βx2(f (x2)) = 0
this is impossible, so σ is injective.
By considering α−1, we can also get σ is surjective.
When n → ∞, if d(xn, x) → 0, then for any f ∈ C(X, A),
kf (σ(xn)) − f (σ(x))k
= kα(α−1(f ))(σ(xn)) − α(α−1(f ))(σ(x))k
= kβxn (α−1(f )(xn)) − βx(α−1(f )(x))k
= kβxn (α−1(f )(xn) − α−1(f )(x)) + βxn(α−1(f )(x)) − βx(α−1(f )(x))k
→ 0.
so σ is continuous, we can also get σ−1 is continuous by considering α−1, we
omit it. so σ is a homeomorphism.
Since C(X, A) is α-simple, it does not have any non-trivial α-invariant
closed two-sided ideals of C(X, A), then we have not any non-trivial σ-
invariant closed subset of X, so σ is a minimal homeomorphism of X.
Conversely, if σ is a minimal homeomorphism of X and β : X → Aut(A) is
a strongly continuous map, define α(f )(y) = βσ−1(y)f (σ−1(y)) for all y ∈ X,
it is easy to verify α ∈ Aut(C(X, A)) and C(X, A) is α-simple.
(cid:3)
Lemma 2.4. Let X be an infinite compact metric space, let A be a unital
simple C*-algebra and α ∈Aut(C(X, A)). Then C(X, A) is α-simple if and
only if the crossed product C(X, A) ⋊α Z is simple.
CROSSED PRODUCTS BY α-SIMPLE AUTOMORPHISMS ON C*-ALGEBRAS
5
Proof. Let I be an α-invariant norm closed two-sided ideal of C(X, A). Then
I ⋊α Z is a norm closed two-sided ideal of C(X, A) ⋊α Z by Lemma 1 of [7].
Conversely, for any positive element f of the C*-algebra C(X, A), any
finite set F = {fi; i = 1, 2, · · · .n} ⊂ C(X, A), any si ⊂ N, i = 1, 2, · · · , n,
and any ε > 0, we claim that there exists a positive element g ∈ C(X, A)
with kgk = 1 such that
kgf gk ≥ kf k − ε,
kgfiαsi(g)k ≤ ε,
i = 1, 2, · · · , n.
(∗)
Because X is a compact set, we can get a point x ∈ X such that kf (x)k =
kf k. By Lemma 2.3 we get a minimal homeomorphism σ of X . Since σ is the
minimal homeomorphism of X, there exists a neighborhood O(x) of x such
that σi(O(x)) are disjoint for i = 1, · · · , s, where s = max{si, i = 1, 2, · · · , n}
and kf (y)−f (x)k < ε for all y ∈ O(x). It is easy to find a continuous function
g of X to [0, 1] such that g(x) = 1 and g(z) = 0 for all z /∈ O(x). Then g
satisfies the conditions (∗). The claim follows.
Use the condition (∗) and C(X, A) is α-simple, we can complete the proof
(cid:3)
as same as Theorem 3.1 of [8], we omit it.
Let u be the unitary implementing the action of α in the transformation
group C*-algebra C(X, A) ⋊α Z, then uf u∗ = α(f ). For a nonempty closed
subset Y ⊂ X, we define the C*-subalgebra BY to be
BY = C ∗(C(X, A), uC0(X\Y, A)) ⊂ C(X, A) ⋊α Z.
We will often let B denote the transformation group C*-algebra C(X, A) ⋊α
Z. If Y1 ⊃ Y2 ⊃ · · ·
is a decreasing sequence of closed subsets of X with
∩∞
n=1Yn = {y}, then B{y} = lim BYn.
Let Y ⊂ X, and let x ∈ Y. If C(X, A) is α-simple, by Lemma 2.3 we have
a minimal homeomorphism σ of X. The first return time λY (x) (or λ(x) if Y
is understood) of x to Y is the smallest integer n ≥ 1 such that σn(x) ∈ Y .
The following result is well known in the area, and is easily proved:
Lemma 2.5. If Y is a nonempty clopen subset and σ is a minimal homeo-
morphism of X. Then supx∈Y {λY (x)} < ∞.
Let Y ⊂ X is a nonempty clopen subset. Let n(0) < n(1) < · · · < n(l) be
the distinct values of λ(x) for x ∈ Y . The Rokhlin tower based on a subset
Y ⊂ X with Y 6= ∅ consist of the partition
l
Y =
{x ∈ Y : λ(x) = n(k)}
ak=0
of Y (the sets here are the base sets), and the corresponding partition
X =
l
n(k)−1
ak=0
aj=0
σj({x ∈ Y : λ(x) = n(k)})
of X.
Actually, for our purposes it is more convenient to use the partition
6
JIAJIE HUA
Note that
X =
l
n(k)
ak=0
aj=1
σj({x ∈ Y : λ(x) = n(k)}).
Y =
l
ak=0
σn(k)({x ∈ Y : λ(x) = n(k)})
Since Y is both closed and open, the sets
Yk = {x ∈ Y : λ(x) = n(k)}
are all closed and open.
Proposition 2.6. Let X be a Cantor set, let A be a unital simple C*-
algebra and α ∈ Aut(C(X, A)). Suppose C(X, A) is α-simple, let Y ⊂ X be
a nonempty clopen subset. Then there exists a unique isomorphism
γY : BY →
l
Mk=0
C(Yk, Mn(k)(A))
such that if f ∈ C(X, A), then
γY (f )k = diag(α−1(f )Yk , α−2(f )Yk , · · · , α−n(k)(f )Yk )
and if f ∈ C0(X\Y, A), then
where sk ∈ Mn(k) ⊂ C(Yk, Mn(k)(A)) is defined by
(γY (uf ))k = skγY (f )k
sk =
0 0 0 · · ·
1 0 0 · · ·
0 1 0 · · ·
...
. . .
...
0 0 0 · · ·
0 0 0 · · ·
...
...
...
...
· · ·
· · ·
· · ·
. . .
· · ·
· · ·
0 0 1
0 0 0
0 0 0
...
...
...
...
0 0 0
0 1 0
...
...
Proof. By Lemma 2.3, there is a minimal homomorphism σ from X to X
and there is a strongly continuous map from X to Aut(A), denote by x to
βx, such that α(f )(x) = βσ−1(x)(f (σ−1(x))).
From above, we have
Y =
l
ak=0
σn(k)({x ∈ Y : λ(x) = n(k)}) =
l
ak=0
σn(k)(Yk).
It is easy to verify that the γY is a homomorphism.
Since α(f )(x) = βσ−1(x)(f (σ−1(x))), it is easy to see that γY is injective
(cid:3)
and surjective.
CROSSED PRODUCTS BY α-SIMPLE AUTOMORPHISMS ON C*-ALGEBRAS
7
3. Main result
Lemma 3.1. Let X be an infinite compact metric space, let A be a unital
simple C*-algebra and α ∈Aut(C(X, A)). If C(X, A) is α-simple, let y ∈ X,
then B{y} is simple.
Proof. Let I ⊂ B{y} be a nonzero ideal. Since X is an infinite compact
metric space, A is a unital simple C*-algebra and C(X, A) is α-simple, we
have a minimal homeomorphism σ of X by Lemma 2.3. Then I ∩ C(X, A)
is an ideal in C(X, A). So we can write I ∩ C(X, A) = C0(U, A) for some
open set U ⊂ X by the proof of Lemma 2.3, which is necessarily given by,
U = {x ∈ X : there is f ∈ I ∩ C(X, A) such that f (x) 6= 0}.
We first claim that U 6= ∅,
Write
B{y} = lim
→
BYm
for some decreasing sequence Y1 ⊃ Y2 ⊃ · · · of closed subsets of X with
∩∞
n=1Yn = {y}, and int(Ym) 6= ∅,
Then there exists m such that BYm ∩ I 6= {0} by Lemma 3.5.10 of [12].
Let a be a nonzero element of this intersection. Using Proposition 2.6, one
can fairly easily prove that there is N such that every element of BYm can
n=−N fnun, with fn ∈ C(X, A) for −N ≤ n ≤ N.
n=−N fnun, then f0 6= 0. Choose
x ∈ X such that f0(x) 6= 0, choose a neighborhood V of x such that the
sets hn(V ), for −N ≤ n ≤ N , are disjoint, and choose g ∈ C(X) such that
supp(g) ∈ V and g(x) 6= 0. Then g ∈ B{y}, and one checks that
be written in the form PN
Moreover, if a 6= 0, and one writes a∗a = PN
ga∗ag =
N
Xn=−N
gfnung =
N
Xn=−N
g(g ◦ σn)fnun = g2f0.
So g2f0 is a nonzero element of I ∩ C(X, A), proving the claim.
We next claim that σ−1(U \{σ(y)}) ⊂ U. So let x ∈ U \{σ(y)}. Choose
f ∈ I ∩ C(X, A) such that f (x) 6= 0, and choose g ∈ C0(X\{y}) such that
g(σ−1(x)) 6= 0. Then ug ∈ B{y}, and
(ug)∗f (ug) = gu∗f ug = g2(u∗f u).
Thus g2(u∗f u) ∈ I ∩ C(X, A) and is nonzero at σ−1(x), This proves the
claim.
We further claim that σ(U \{y}) ⊂ U. The proof is similar: let x ∈ U \{y},
let f ∈ I ∩ C(X, A) and g ∈ C0(X\{y}) be nonzero at x, and consider
ug ∈ B{y},
(ug)f (ug)∗ = ugf gu∗ = g ◦ σ−12(uf u∗).
Thus g ◦ σ−12(uf u∗) ∈ I ∩ C(X, A) and is nonzero at σ(x), So σ(U \{y}) ⊂
U.
8
JIAJIE HUA
Now set Z = X\U. The last two claims above imply that if x ∈ X and x
is not in the orbit of y, Then σk(x) ∈ Z for all k ∈ Z. Since σ is minimal,
Z is closed, and Z 6= X, this is impossible. If σn(y) ∈ Z for some n > 0,
then σ−1(U \{σ(y)}) ⊂ U \{σ(y)} implies hk(y) ∈ Z for all k ≥ n. Since σ
is dense by minimality, this is also a contradiction. Similarly, if σn(y) ∈ Z
for some n ≤ 0, then Z would contain the dense set {σk(y) : k ≤ n}, again
a contradiction.
So U = X and 1 ∈ I. Then B{y} is simple.
(cid:3)
Lemma 3.2. Let X be a Cantor set, let A be a unital separable simple
amenable C*-algebra with tracial rank zero which satisfies the UCT (Uni-
versal Coefficient Theorem), and let α ∈ Aut(C(X, A)). Suppose C(X, A)
is α-simple. It follows that for any y ∈ X, the C*-algebra B{y} has tracial
rank zero.
Proof. Let Y ⊂ X be a nonempty clopen subset, applying Proposition 2.6,
we have
BY ∼=
l
Mk=0
C(Yk, Mn(k)(A)) =
l
Mk=0
C(Yk) ⊗ Mn(k)(A).
Let y ∈ X, then there exists a decreasing sequence Z1 ⊃ Z2 ⊃ · · · of
clopen subsets of X with ∩∞
n=1Zn = {y}. Since the property of having lo-
cally finite decomposition rank passes to tensor products, inductive limits by
Proposition 1.3 of [26], then B{y} = lim BZn has locally finite decomposition
rank. Clearly, for any n ∈ N, BZn has real rank zero, so B{y} has also real
rank zero. Since A is a unital separable simple amenable C*-algebra with
tracial rank zero which satisfies the Universal Coefficient Theorem, A is an
AH-algebra of slow dimension growth with real rank zero by [13], then A
is an AH-algebra of bounded dimension growth by [3]. By Corollary 3.1 of
[24], A absorbs the Jiang-Su algebra Z tensorially. By Corollary 3.4 of [25],
B{y} absorbs the Jiang-Su algebra Z tensorially, so the C*-algebra B{y} has
tracial rank zero by Theorem 2.1 of [26].
(cid:3)
Lemma 3.3. Any trace on B{y} is restricted to C(X) is a σ-invariant mea-
sure on X.
Proof. Let τ be a normalized trace on B{y}, and let f ∈ C(X). Set
a = uf − f (y)1/2
and b = u(f − f (y) · 1)f − f (y)−1/2
Then a and b are both in B{y}. Moreover,
τ (f ◦σ−1 −f (y)·1) = τ (u(f −f (y)·1)u∗) = τ (ab∗) = τ (b∗a) = τ (f −f (y)·1).
Cancelling τ (f (y) · 1), we get τ (f ◦ σ−1) = τ (f ).
(cid:3)
CROSSED PRODUCTS BY α-SIMPLE AUTOMORPHISMS ON C*-ALGEBRAS
9
Lemma 3.4. Let X be a Cantor set, let A be a unital separable simple
amenable C*-algebra with tracial rank zero which satisfies the UCT, and
let α ∈ Aut(C(X, A)). Suppose C(X, A) is α-simple and [α] = [id1⊗A] in
KL(1 ⊗ A, 1 ⊗ A). Let B = C(X, A) ⋊α Z, and let y ∈ X. Then for any
ε > 0 and finite subset F ⊂ B, there is a projection p ∈ B{y} such that:
(1) kpa − apk < ε for all a ∈ F.
(2) pap ∈ pB{y}p for all a ∈ F.
(3) τ (1 − p) < ε for all τ ∈ T (B{y}).
Proof. We may assume that F = G ∪{u} for some finite subset G ⊂ C(X, A).
By Lemma 2.3, there is a minimal homomorphism σ from X to X and
there is a strongly continuous map from X to Aut(A), denote by x to βx,
such that α(f )(x) = βσ−1(x)(f (σ−1(x))).
and so small that d(x1, x2) < 4δ0 implies kf (x1) − f (x2)k < 1
Choose N0 ∈ N so large that 4π/N0 < ε. Choose δ0 > 0 with δ0 < 1
2 ε
4 ε for all f ∈
SN0
i=0 α−i(G). Choose δ > 0 with δ ≤ δ0 and such that whenever d(x1, x2) < δ
and 0 ≤ n ≤ N0, then d(σ−n(x1), σ−n(x2)) < δ0.
Since σ is minimal, there is N > N0 + 1 such that d(σN (y), y) < δ. Since
σ is free, there is a clopen neighborhood Y of y in X such that
σ−N0(Y ), σ−N0+1(Y ), . . . , Y, σ(Y ), . . . , σN (Y )
are disjoint and all have diameter less than δ, and furthermore µ(Y ) <
ε/(N + N0 + 1) for every σ-invariant Borel probability measure µ.
Define continuous functions projection q0(x) = (cid:26)1,
For −N0 ≤ n ≤ N set, qn = αn(q0) = (cid:26)1,
0,
0,
x ∈ σn(Y )
x ∈ X\σn(Y )
x ∈ Y
x ∈ X\Y
, so the qn
are mutually orthogonal projections in B{y}.
We now have a sequence of projections:
q−N0, . . . , q−1, q0, . . . , qN −N0, . . . , qN −1, qN .
The projections q0 and qN live over clopen sets which are disjoint but close
to each other, and similarly for the pairs q−1 and qN −1 down to q−N0 and
qN −N0. We are now going to use Berg's technique [1] to splice this sequence
along the pairs of indices (−N0, N − N0) through (0, N ), obtaining a loop
of length N on which conjugation by u is approximately the cyclic shift.
We claim that there is a partial isometry w ∈ B{y} such that w∗w = q0,
ww∗ = qN and kwf Y − f σN (Y )wk < ε
Let x ∈ Y. The first return time λY (x) (or λ(x) if Y is understood) of x
to Y is the smallest integer n ≥ 1 such that σn(x) ∈ Y . By Lemma 2.5, we
let n(0) < n(1) < · · · < n(l) be the distinct values of λ(x) for x ∈ Y .
4 for all f ∈ SN0
i=0 α−i(G).
We denote Y (k, j) = σj(λ−1(n(k))), So
X =
l
ak=0
n(k)
aj=1
σj({x ∈ Y : λ(x) = n(k)}) =
l
ak=0
n(k)
aj=1
Y (k, j).
10
JIAJIE HUA
Define continuous functions projection χY (k,j) = (cid:26)1,
Define w′ = Pl
k=0 χY (k,N )uN −n(k), then w′ ∈ B{y}.
0,
l
w′∗w′ = (
Xk=0
l
u−N +n(k)χY (k,N ))(
Xk=0
χY (k,N )uN −n(k)) =
l
l
χY (k,N )uN −n(k))(
u−N +n(k)χY (k,N )) =
Xk=0
x ∈ Y (k, j)
x ∈ X\Y (k, j)
l
Xk=0
χY (k,n(k)) = q0.
l
Xk=0
χY (k,N ) = qN .
w′w′∗ = (
Xk=0
Let aY = (cid:26)a,
0,
x ∈ Y
x ∈ X\Y
for all a ∈ A.
Since A is a unital separable simple amenable C*-algebra with tracial
rank zero which satisfies the UCT, by the classification theorem of [15],
[13] and [6], A is a unital separable simple AH-algebra. By the assumption
[α] = [id1⊗A] in KL(1 ⊗ A, 1 ⊗ A) and by applying Lemma 4.1 of [16], we
have τ (βx(a)) = τ (a) for all a ∈ A and for all τ ∈ A. Since TR(A) = 0 and
[α] = [id1⊗A] in KL(1 ⊗ A, 1 ⊗ A), τ (βx(a)) = τ (a) for all τ ∈ A and for all
τ ∈ A, by applying 3.6 of [18], there is a clopen neighborhood Yx of x, there
exists an unitary u′
aσj (Yx)
for all a ∈ {f (y)f ∈ ∪N0
i=0α−i(G)}. Because Y is compact subset, we can
get uj ∈ C(X, A) such that ujσj (Y )ujaY u∗ju∗
aσj (Y ) for all a ∈
{f (y)f ∈ ∪N0
j ∈ A such that u′
jσj (Yx)ujaYxu∗ju′∗
j σj (Yx) ≈ ε
j σj (Y ) ≈ ε
8
8
i=0α−i(G)}.
k=0 uN −n(k)χY (k,N ))w′, then w ∈ B{y}, w∗w = q0, ww∗ =
Define w = (Pl
qN and
l
waY w∗ =
waY ≈ ε
claim follows.
8
uN −n(k)χY (k,N )uN −n(k)aY u−N +n(k)χY (k,N )u∗
Xk=0
aσN (Y )w, so kwf Y − f σN (Y )wk < ε
N −n(k) ≈ ε
8
aσN (Y ),
4 for all f ∈ SN0
i=0 α−i(G). The
For t ∈ R define v(t) = cos(πt/2)(q0 + qN ) + sin(πt/2)(w − w∗). Then v(t)
is a unitary in the corner (q0 + qN )B{y}(q0 + qN ) whose matrix with respect
to the obvious block decomposition is
v(t) = (cid:18)cos(πt/2) − sin(πt/2)
cos(πt/2) (cid:19)
sin(πt/2)
So kv(t)(f Y + f σN (Y )) − (f Y + f σN (Y ))v(t)k = k((cos(πt/2)(q0 + qN ) +
sin(πt/2)(w − w∗))(f Y + f σN (Y )) − (f Y + f σN (Y ))(cos(πt/2)(q0 + qN ) +
sin(πt/2)(w−w∗))k = k sin(πt/2)(wf Y −w∗f σN (Y ))−sin(πt/2)(f σN (Y )w−
f Y w∗)k < ε
(∗∗)
For 0 ≤ k ≤ N0 define wk = u−kv(k/N0)uk, so wk ∈ (q−k+qN −k)B{y}(q−k+
2 for all f ∈ SN0
i=0 α−i(G).
qN −k).
CROSSED PRODUCTS BY α-SIMPLE AUTOMORPHISMS ON C*-ALGEBRAS
11
kuwk+1u∗ − wkk = kv(k/N0) − v((k − 1))/N0k ≤ 2π/N0 < 1
Now define en = qn for 0 ≤ n ≤ N − N0, and for N − N0 ≤ n ≤ N write
k = N − n and set en = wkq−kw∗
k. The two definitions for n = N − N0
agree because wN0q−N0w∗
N0 = qN −N0, and moreover eN = e0. Therefore
kuen−1u∗ − enk = 0 for 1 ≤ n ≤ N − N0, and also ueN u∗ = e1, while for
N − N0 < n ≤ N we have
2 ε.
kuen−1u∗ − enk ≤ 2kuwN −n+1u∗ − wN −nk < ε.
Also, clearly en ∈ B{y} for all n.
Set e = PN
First,
n=1 en and p = 1 − e. We verify that p satisfies (1) through (3).
p − upu∗ = ueu∗ − e =
N
(uen−1u∗ − en).
Xn=N0+1
The terms in the sum are orthogonal and have norm less than ε, so kupu∗ −
pk < ε.
Furthermore, pup ∈ B{y}.
Next, let g ∈ G. The sets U0, U1, . . . , UN all have diameter less than δ. We
have d(σN (y), y) < δ, so the choice of δ implies that d(σn(y), σn−N (y)) < δ0
for N − N0 ≤ n ≤ N. Also, Un−N = σn−N (U0) has diameter less than δ.
Therefore Un−N ∪ Un has diameter less than 2δ + δ0 ≤ 3δ0. Since g varies
by at most 1
4 ε on any set with diameter less than 4δ0, and since the sets
σ(Y ), σ2(Y ), σN −N0−1(Y ), σN −N0 (Y )∪σ−N0 (Y ), σN −N0+1(Y )∪σ−N0+1(Y ),
. . . , σN (Y ) ∪ Y are disjoint.
For 0 ≤ n ≤ N − N0, we have gen = gqn = qng = eng.
For N − N0 < n ≤ N and any g ∈ G, we use en ∈ (qn−N + qn)B{y}(qn−N +
qn) and (∗∗) to get kgen − engk = kgwkq−kw∗
kgwkq−k −
q−kw∗
kgwkk < ε. It follows that kpg − gpk = kge − egk < ε. That pgp ∈ B{y}
follows from the fact that g and p are in this subalgebra. So we also have
(2) for g.
k − wkq−kw∗
kgk = kw∗
It remains only to verify (3). Let τ ∈ T (B{y}), and let µ be the corre-
sponding σ-invariant probability measure on X by Lemma 3.3. We have
1 − p = e ≤
N
Xn=−N0
qn,
so
τ (1 − p) ≤
N
Xn=−N0
µ(σn(Y )) = (N + N0 + 1)µ(Y ) < ε.
This completes the proof.
(cid:3)
We recall two results from Lemma 4.3 and Lemma 4.4 of [17].
12
JIAJIE HUA
Lemma 3.5. Let A be a unital simple C*-algebra. Suppose that for every
ε > 0 and every finite subset F ⊂ A, there exists a unital C*-subalgebra
B ⊂ A which has tracial rank zero and a projection p ∈ B such that
kpa − apk < ε and
dist(pap, pBp) < ε
for all a ∈ F. Then A has the local approximation property of Popa [21],
that is for every ε > 0 and every finite subset F ⊂ A, there exists a nonzero
projection q ∈ A and a finite dimensional unital C*-subalgebra D ⊂ qAq
such that
kqa − aqk < ε
and
dist(qaq, D) < ε
for all a ∈ F.
Lemma 3.6. Let A be a unital simple C*-algebra. Suppose that for every
finite subset F ⊂ A, every ε > 0, and every nonzero positive element c ∈ A,
there exists a projection p ∈ A and a unital simple subalgebra B ⊂ pAp with
tracial rank zero such that:
(1) k[a, p]k < ε for all a ∈ F.
(2) dist(pap, B) < ε for all a ∈ F.
(3) 1 − p is Murray-von Neumann equivalent to a projection in cAc.
Then A has tracial rank zero.
Theorem 3.7. Let X be a Cantor set, and let A be a unital separable
simple amenable C*-algebra with tracial rank zero which satisfies the UCT.
Let C(X, A) denote all continuous functions from X to A and α be an
automorphism of C(X, A). Suppose that C(X, A) is α-simple and [α] =
[id1⊗A] in KL(1 ⊗ A, 1 ⊗ A). Then C(X, A) ⋊α Z is a unital simple C*-
algebra with tracial rank zero.
Proof. Let B = C(X, A) ⋊α Z, then B is unital simple by Lemma 2.4.
We verify the conditions of Lemma 3.6.
Let B{y} = C ∗(C(X, A), uC0(X\{y}, A)) for any given point y ∈ X. So
B{y} has tracial rank zero by Lemma 3.2. Thus, let F ⊂ B be a finite subset,
let ε > 0, and let c ∈ B be a nonzero positive element. Beyond Lemma 3.4,
the main step of the proof is to find a nonzero projection in B{y} which is
Murray-von Neumann equivalent to a projection in cBc.
The algebra B is simple, so Lemma 3.4, Lemma3.5 and Lemma 2.12 of
[10] imply that B has property (SP). Therefore there is a nonzero projection
e ∈ cBc.
Set
δ0 =
1
18
inf
τ ∈T (B)
τ (e) ≤
1
18
.
By Lemma 3.4, there is a projection q ∈ B{y} and an element b0 ∈ qB{y}q
such that
kqe − eqk < δ0,
kqeq − b0k < δ0,
and
sup
τ (1 − q) ≤ δ0.
τ ∈T (B{y})
CROSSED PRODUCTS BY α-SIMPLE AUTOMORPHISMS ON C*-ALGEBRAS
13
Then supτ ∈T (B) τ (1 − q) ≤ supτ ∈T (B{y}) τ (1 − q) ≤ δ0.
Replacing b0 by 1
0), we may assume that b0 is self-adjoint. We have
−δ0 ≤ b0 ≤ 1 + δ0, so applying continuous functional calculus we may find
b ∈ qB{y}q such that 0 ≤ b ≤ 1 and kqeq − bk < 2δ0. Using kqe − eqk < δ0
on the last term in the second expression, we get
2 (b0 + b∗
kb2 − bk ≤ 3kb − qeqk + k(qeq)2 − qeqk < 3 · 2δ0 + δ0 = 7δ0 <
1
4
.
Therefore there is a projection e1 ∈ qB{y}q such that ke1 − bk < 14δ0, giving
ke1 − qeqk < 16δ0. Similarly ( actually, one gets a better estimate ) there is
a projection e2 ∈ (1 − q)B(1 − q) such that ke2 − (1 − q)e(1 − q)k < 16δ0.
Therefore ke1 + e2 − [qeq + (1 − q)e(1 − q)]k < 16δ0 and, using kqe − eqk < δ0
again, we have ke1 + e2 − ek < 18δ0 ≤ 1. It follows that e1 - e. Also, for
τ ∈ T (B), we have
τ (e1) > τ (qeq) − 16δ0 = τ (e) − τ ((1 − q)e(1 − q)) − 16δ0
≥ τ (e) − τ (1 − q) − 16δ0 > 0,
so e1 6= 0.
Now set ε0 = inf({τ (e1) : τ ∈ T (B{y})}). By Lemma 3.4, there is a
projection p ∈ B{y} such that:
(1) kpa − apk < ε for all a ∈ F.
(2) pap ∈ pB{y}p for all a ∈ F.
(3) τ (1 − p) < ε0 for all τ ∈ T (B{y}).
Since B{y} has tracial rank zero which implies that the order on projections
is determined by traces, it follows that 1 − p - e1 - e. Since pB{y}p also
has tracial rank zero, we have verified the hypotheses of Lemma 3.6. Thus
B has tracial rank zero.
(cid:3)
Corollary 3.8. For j=1,2, let X be a Cantor set, and let Aj be a unital
separable simple amenable C*-algebra with tracial rank zero which satisfy
the UCT. Let C(X, Aj) denote all continuous functions from X to Aj, and
let αj be an automorphism of C(X, Aj). Suppose that C(X, Aj) is αj-simple
and [αj] = [id1⊗Aj ] in KL(1 ⊗ Aj, 1 ⊗ Aj). Let Bj = C(X, Aj) ⋊αj
Z for
j = 1, 2. Then B1 ∼= B2 if and only if
(K0(B1), K0(B1)+, [1B1 ], K1(B1)) ∼= (K0(B2), K0(B2)+, [1B2 ], K1(B2)).
Proof. This is immediate from Theorem 5.2 of [15] and Theorem 3.7.
(cid:3)
References
[1] I. D. Berg, On approximation of normal operators by weighted shifts, Michigan Math.
J. 21(1974), 377 -- 383.
[2] B. Blackadar, K-Theory for Operator Algebras, MSRI Publication Series 5, Springer-
Verlag, New York, Heidelberg, Berlin, Tokyo, 1986.
[3] M. Dadarlat and G. Gong, A classification result for approximately homogeneous C*-
algebras of real rank zero, Geom. Funct. Anal. 7(1997), 646-711.
14
JIAJIE HUA
[4] G. A. Elliott, The classification problem for amenable C*-algebras, pages 922 -- 932
in: Proceedings of the International Congress of Mathematications, Zurich, 1994,
S.D.Chatterji, ed., Birkhauser, Basel, 1995.
[5] G. A. Elliott and D. E. Evans, The structure of the irrational rotation algebra, Ann.
of Math. (2)138(1993), 477 -- 501.
[6] G. A. Elliott and G. Gong, On classification of C*-algebras of real rank zero. II, Ann.
Math. 144 (1996), 497-610.
[7] S. Jang and S. Lee, Simplicity of crossed products of C*-algebras, Proc. Amer. Math.
Soc. 118(1993), no. 3, 823 -- 826.
[8] A. Kishimoto, Outer automorphisms and reduced crossed products of simple C*-algebra,
Comm. Math. Phys. 81(1981), 429 -- 435.
[9] R. H. Herman, I. F. Putnam, and C. F. Skau, Ordered Bratteli diagrams, dimension
groups, and topological dynamics, International J. Math. 3(1992), 827 -- 864.
[10] H. Lin, Tracially AF C*-algebras, Trans. Amer. Math. Soc. 353(2001), 693 -- 722.
[11] H. Lin, Tracial topological ranks of C*-algebra, Proc. London Math. Soc. 83(2001),
199 -- 234.
[12] H. Lin, An Introduction to the Classification of Amenable C*-algebra, World Scien-
tific, New Jersey/London/Singapore/Hong Kong/Bangalore, 2001.
[13] H. Lin, Simple AH-algebras of real rank zero, Proc. Amer. Math. Soc. 131(2003), no.
12, 3813 -- 3819 (electronic).
[14] H. Lin, Classification of simple C*-algebras and higher dimensional tori, Ann. of
Math. 157(2003), 521 -- 544.
[15] H. Lin, Classification of simple C*-algebra with tracial topological rank zero, Duke
Math. J. 125(2004), no. 1, 91 -- 119.
[16] H. Lin, The Rokhlin property for automorphisms on simple C ∗-algebras,
arXiv:math/0602513v2[math.OA] 27 Jan 2010.
[17] H. Lin and N. C. Phillips, Crossed products by minimal homeomorphisms, J. Reine
angew. Math. to appear.
[18] H. Lin, AF-embedding of crossed products of AH-algebras by Z and asympototic AF-
embedding, Indiana Univ. Math. J.57 (2008),891-944.
[19] N. C. Phillips, Every simple higher dimensional noncommutative torus is an AT
algebra, preprint (arXiv: math.OA/0609783).
[20] N. C. Phillips, Cancellation and stable rank for direct limits of recersive subhomoge-
neous algebras, Trans. Amer. Math. Soc. 359 (2007), 4625-4652.
[21] S. Popa, On local finite dimensional approximation of C*-algebras, Pacific J. Math.
181(1997), 141 -- 158.
[22] I. F. Putnam, On the topological stable rank of certain transformation group C*-
algebras, Ergod. Th. Dynam. Sys. 10(1990), 197 -- 207.
[23] M. Rφrdam, Classification of certain infinte simple C*-algebras, J. Funct. Anal.
131(1995), 415 -- 458.
[24] A. S. Toms and W. Winter, Z-stable ASH C*-algebras, Preprint, Math. Archive
math.OA/0508218, to appear in Can. J. Math., 2005.
[25] A. S. Toms and W. Winter, Strongly self-absorbing C*-algebras, Trans. Amer. Math.
Soc. 359 (2007), 3999C4029.
[26] W. Winter, Simple C*-algebras with locally finite decomposition rank, J. Funct. Anal.
243(2007), 394-425.
Department of Mathematics, Tongji University, Shanghai 200092,
P.R.CHINA.
E-mail address: [email protected]
|
1611.01633 | 1 | 1611 | 2016-11-05T11:28:09 | Characterizations of centralizers and derivations on some algebras | [
"math.OA"
] | A linear mapping $\phi$ on an algebra $\mathcal{A}$ is called a centralizable mapping at $G\in\mathcal{A}$ if $\phi(AB)=\phi(A)B=A\phi(B)$ for each $A$ and $B$ in $\mathcal{A}$ with $AB=G$, and $\phi$ is called a derivable mapping at $G\in\mathcal{A}$ if $\phi(AB)=\phi(A)B+A\phi(B)$ for each $A$ and $B$ in $\mathcal{A}$ with $AB=G$. A point $G$ in $\mathcal{A}$ is called a full-centralizable point (resp. full-derivable point) if every centralizable (resp. derivable) mapping at $G$ is a centralizer (resp. derivation). We prove that every point in a von Neumann algebra or a triangular algebra is a full-centralizable point. We also prove that a point in a von Neumann algebra is a full-derivable point if and only if its central carrier is the unit. | math.OA | math |
Characterizations of centralizers and derivations on some
algebras
Jun He, Jiankui Li∗, and Wenhua Qian
Department of Mathematics, East China University of Science and Technology
Shanghai 200237, China
Abstract
A linear mapping φ on an algebra A is called a centralizable mapping at G ∈ A
if φ(AB) = φ(A)B = Aφ(B) for each A and B in A with AB = G, and φ is called a
derivable mapping at G ∈ A if φ(AB) = φ(A)B + Aφ(B) for each A and B in A with
AB = G. A point G in A is called a full-centralizable point (resp. full-derivable point)
if every centralizable (resp. derivable) mapping at G is a centralizer (resp. derivation).
We prove that every point in a von Neumann algebra or a triangular algebra is a
full-centralizable point. We also prove that a point in a von Neumann algebra is a
full-derivable point if and only if its central carrier is the unit.
Keywords: Centralizer, derivation, full-centralizable point, full-derivable point,
von Neumann algebra, triangular algebra
Mathematics Subject Classification(2010): 47B47; 47L35
1
Introduction
Let A be an associative algebra over the complex field C, and φ be a linear mapping
from A into itself. φ is called a centralizer if φ(AB) = φ(A)B = Aφ(B) for each
A and B in A. Obviously, if A is an algebra with unit I, then φ is a centralizer
if and only if φ(A) = φ(I)A = Aφ(I) for every A in A. φ is called a derivation if
φ(AB) = φ(A)B + Aφ(B) for each A and B in A.
A linear mapping φ : A → A is called a centralizable mapping at G ∈ A if φ(AB) =
φ(A)B = Aφ(B) for each A and B in A with AB = G, and φ is called a derivable
∗Corresponding author. E-mail address: [email protected]
1
mapping at G ∈ A if φ(AB) = φ(A)B + Aφ(B) for each A and B in A with AB = G.
An element G in A is called a full-centralizable point (resp. full-derivable point) if every
centralizable (resp. derivable) mapping at G is a centralizer (resp. derivation).
In [3], Bresar proves that if R is a prime ring with a nontrival idempotent, then
0 is a full-centralizable point.
In [16], X. Qi and J. Hou characterize centralizable
and derivable mappings at 0 in triangular algebras. In [17], X. Qi proves that every
nontrival idempotent in a prime ring is a full-centralizable point. In [19], W. Xu, R.
An and J. Hou prove that every element in B(H) is a full-centralizable point, where H
is a Hilbert space. For more information on centralizable and derivable mappings, we
refer to [2, 10, 11, 13, 18, 20].
For a von Neumann algebra A, the central carrier C(A) of an element A in A is
the projection I − P , where P is the union of all central projections Pα in A such that
PαA = 0.
This paper is organized as follows. In Section 2, by using the techniques about cen-
tral carriers, we show that every element in a von Neumann algebra is a full-centralizable
point.
Let A and B be two unital algebras over the complex field C, and M be a unital
(A, B)-bimodule which is faithful both as a left A-module and a right B-module. The
algebra
T ri(A, M, B) =(" A M
0 B # : A ∈ A, B ∈ B, M ∈ M)
under the usual matrix addition and matrix multiplication is called a triangular algebra.
In Section 3, we show that if A and B are two unital Banach algebras, then every
element in T ri(A, M, B) is a full-centralizable point.
In Section 4, we show that for every point G in a von Neumann algebra A, if ∆
is a derivable mapping at G, then ∆ = D + φ, where D : A → A is a derivation and
φ : A → A is a centralizer. Moreover, we prove that G is a full-derivable point if and
only if C(G) = I.
2 Centralizers on von Neumann algebras
In this section, A denotes a unital algebra and φ : A → A is a centralizable mapping
at a given point G ∈ A. The main result is the following theorem.
Theorem 2.1. Let A be a von Neumann algebra acting on a Hilbert space H. Then
every element G in A is a full-centralizable point.
Before proving Theorem 2.1, we need the following several lemmas.
2
Lemma 2.2. Let A be a unital Banach algebra with the form A = Pi∈ΛL Ai. Then
φ(Ai) ⊆ Ai. Moreover, suppose G = Pi∈Λ
point in Ai for every i ∈ Λ, then G is a full-centralizable point in A.
Gi, where Gi ∈ Ai. If Gi is a full-centralizable
Proof. Let Ii be the unit in Ai. Suppose that Ai is an invertible element in Ai, and t
is an arbitrary nonzero element in C. It is easy to check that
(I − Ii + t−1GA−1
i
)((I − Ii)G + tAi) = G.
So we have
(I − Ii + t−1GA−1
i
)φ((I − Ii)G + tAi) = φ(G).
Considering the coefficient of t, since t is arbitrarily chosen, we have (I − Ii)φ(Ai) = 0.
It follows that φ(Ai) = Iiφ(Ai) ∈ Ai for all invertible elements. Since Ai is a Banach
algebra, every element can be written into the sum of two invertible elements. So the
above equation holds for all elements in Ai. That is to say φ(Ai) ⊆ Ai.
Ai. Assume AB = G. Since
AiBi = Gi and φ(Ai) ⊆ Ai, we have
Let φi = φ Ai . For every A in A, we write A = Pi∈Λ
Bi =Xi∈Λ
φ(Gi) =Xi∈Λ
Xi∈Λ
φ(Ai)Xi∈Λ
φ(Ai)Bi.
It implies that φi(Gi) = φi(Ai)Bi. Similarly, we can obtain φi(Gi) = Aiφi(Bi). By
assumption, Gi is a full-centralizable point, so φi is a centralizer. Hence
φ(A) =Xi∈Λ
φi(Ai) =Xi∈Λ
φi(Ii)Ai =Xi∈Λ
φi(Ii)Xi∈Λ
Ai = φ(I)A.
Similarly, we can prove φ(A) = Aφ(I). Hence G is a full-centralizable point.
Lemma 2.3. Let A be a C ∗-algebra. If G∗ is a full-centralizable point in A, then G is
a full-centralizable point in A.
For each A and B in A with AB = G∗, we have B∗A∗ = G. It follows that φ(G) =
Proof. Define a linear mapping eφ : A → A by: eφ(A) = (φ(A∗))∗ for every A in A.
φ(B∗)A∗ = B∗φ(A∗). By the definition of eφ, we obtain eφ(G∗) = eφ(A)B = Aeφ(B).
Since G∗ is a full-centralizable point in A, we have that eφ is a centralizer. Thus φ is
also a centralizer. Hence G is a full-centralizable point in A.
For a unital algebra A and a unital A-bimodule M, an element A ∈ A is called
a left separating point (resp. right separating point) of M if AM = 0 implies M = 0
(M A = 0 implies M = 0) for every M ∈ M.
Lemma 2.4. Let A be a unital Banach algebra and G be a left and right separating
point in A. Then G is a full-centralizable point.
3
Proof. For every invertible element X in A, we have
φ(I)G = φ(G) = φ(XX −1G) = φ(X)X −1G.
Since G is a right separating point, we obtain φ(I) = φ(X)X −1. It follows that φ(X) =
φ(I)X for each invertible element X and so for all elements in A. Similarly, we have
that φ(X) = Xφ(I). Hence G is a full-centralizable point.
Lemma 2.5. Let A be a von Neumann algebra. Then G = 0 is a full-centralizable
point.
Proof. For any projection P in A, since P (I − P ) = (I − P )P = 0, we have
φ(P )(I − P ) = P φ(I − P ) = φ(I − P )P = (I − P )φ(P ) = 0.
It follows that φ(P ) = φ(I)P = P φ(I). By [6, Proposition 2.4] and [4, Corollary 1.2],
we know that φ is continuous. Since A = span{P ∈ A : P = P ∗ = P 2}, it follows that
φ(A) = φ(I)A = Aφ(I) for every A ∈ A. Hence G is a full-centralizable point.
Lemma 2.6. Let A be a von Neumann algebra acting on a Hilbert space H and P be
the range projection of G . If C(P ) = C(I − P ) = I, then G is a full-centralizable point.
Proof. Set P1 = P, P2 = I − P , and denote PiAPj by Aij, i, j = 1, 2. For every A in
A, denote PiAPj by Aij.
Firstly, we claim that the condition AAij = 0 implies APi = 0, and similarly,
Aij A = 0 implies Pj A = 0. Indeed, since C(Pj ) = I, by [8, Proposition 5.5.2], the range
of APj is dense in H. So APiAPj = 0 implies APi = 0. On the other hand, if AijA = 0,
then A∗Aji = 0. Hence A∗Pj = 0 and Pj A = 0.
Besides, since P1 = P is the range projection of G, we have P1G = G. Moreover, if
AG = 0, then AP1 = 0.
In the following, we assume that Aij is an arbitrary element in Aij, i, j = 1, 2, and
t is an arbitrary nonzero element in C. Without loss of generality, we may assume that
A11 is invertible in A11.
Claim 1 φ(A12) ⊆ A12.
Since (P1 + tA12)G = G, we have φ(G) = φ(P1 + tA12)G. It implies that φ(A12)G =
0. Hence φ(A12)P1 = 0.
By (P1 + tA12)G = G, we also have φ(G) = (P1 + tA12)φ(G).
It follows that
A12φ(G) = A12φ(P1)G = 0. So A12φ(P1)P1 = 0. Hence P2φ(P1)P1 = 0.
Since (A11 + tA11A12)(A−1
11 G − A12A22 + t−1A22) = G, we have
φ(A11 + tA11A12)(A−1
11 G − A12A22 + t−1A22) = φ(G).
(2.1)
Since t is arbitrarily chosen in (2.1), we obtain
φ(A11)(A−1
11 G − A12A22) + φ(A11A12)A22 = φ(G).
4
Since A12 is also arbitrarily chosen, we can obtain φ(A11)A12A22 = φ(A11A12)A22.
Taking A22 = P2, since φ(A12)P1 = 0, we have
φ(A11A12) = φ(A11)A12.
Taking A11 = P1, since P2φ(P1)P1 = 0, we have
P2φ(A12) = P2φ(P1)A12 = 0.
(2.2)
(2.3)
So
φ(A12) = φ(A12)P1 + P1φ(A12)P2 + P2φ(A12)P2 = P1φ(A12)P2 ⊆ A12.
Claim 2 φ(A11) ⊆ A11.
Considering the coefficient of t−1 in (2.1), we have φ(A11)A22 = 0. Thus φ(A11)P2 =
0. By (2.2), we obtain P2φ(A11)A12 = P2φ(A11A12) = 0. It follows that P2φ(A11)P1 = 0.
Therefore, φ(A11) = P1φ(A11)P1 ⊆ A11.
Claim 3 φ(A22) ⊆ A22.
By (A11 + tA11A12)(A−1
11 G − A12A22 + t−1A22) = G, we also have
(A11 + tA11A12)φ(A−1
11 G − A12A22 + t−1A22) = φ(G).
Through a similar discussion to equation (2.1), we can prove P1φ(A22) = 0 and
φ(A12A22) = A12φ(A22).
(2.4)
Thus A12φ(A22)P1 = φ(A12A22)P1 = 0. It follows that P2φ(A22)P1 = 0. Therefore,
φ(A22) = P2φ(A22)P2 ⊆ A22.
Claim 4 φ(A21) ⊆ A21.
Since (A11 + tA11A12)(A−1
11 G − A12A21 + t−1A21) = G, we have
(A11 + tA11A12)φ(A−1
11 G − A12A21 + t−1A21) = φ(G).
According to this equation, we can similarly obtain that P1φ(A21) = 0 and
A12φ(A21) = φ(A12A21).
(2.5)
Hence A12φ(A21)P2 = φ(A12A21)P2 = 0. It follows that P2φ(A21)P2 = 0. Therefore,
φ(A21) = P2φ(A21)P1 ⊆ A21.
Claim 5 φ(Aij) = φ(Pi)Aij = Aij φ(Pj) for each i, j ∈ {1, 2}.
By taking A11 = P1 in (2.2), we have φ(A12) = φ(P1)A12. By taking A22 = P2 in
(2.4), we have φ(A12) = A12φ(P2).
By (2.2), we have φ(A11)A12 = φ(A11A12) = φ(P1)A11A12. It follows that φ(A11) =
φ(P1)A11. On the other hand, φ(A11)A12 = φ(A11A12) = A11A12φ(P2) = A11φ(A12) =
A11φ(P1)A12. It follows that φ(A11) = A11φ(P1).
5
By (2.4) and (2.5), through a similar discussion as above, we can obtain that
φ(A22) = A22φ(P2) = φ(P2)A22 and φ(A21) = A21φ(P1) = φ(P2)A21.
Now we have proved that φ(Aij) ⊆ Aij and φ(Aij ) = φ(Pi)Aij = Aij φ(Pj ). It follows
that
φ(A) = φ(A11 + A12 + A21 + A22)
= φ(P1)(A11 + A12 + A21 + A22) + φ(P2)(A11 + A12 + A21 + A22)
= φ(P1 + P2)(A11 + A12 + A21 + A22)
= φ(I)A.
Similarly, we can prove that φ(A) = Aφ(I). Hence G is a full-centralizable point.
Proof of Theorem 2.1. Suppose the range projection of G is P . Set Q1 = I − C(I −
P ), Q2 = I − C(P ), and Q3 = I − Q1 − Q2. Since Q1 ≤ P and Q2 ≤ I − P , {Qi}i=1,2,3
are mutually orthogonal central projections. Therefore A =
Obviously, Ai is also a von Neumann algebra acting on QiH. For each element A in A,
3Pi=1L Ai =
3Pi=1L(QiA).
we write A =
Ai =
QiA.
3Pi=1
3Pi=1
We divide our proof into two cases.
Case 1 ker(G) = {0}
Since Q1 ≤ P , we have ranG1 = ranQ1G = Q1H. Since G is injective on H,
G1 = Q1G is also injective on Q1H. Hence G1 is a separating point(both right and
left) in A1. By Lemma 2.4, G1 is a full-centralizable point in A1.
Since Q2 ≤ I − P , we have G2 = Q2G = 0. By Lemma 2.5, G2 is a full-centralizable
point in A2.
Note that ranG3 = ranQ3G = Q3P = P3. Denote the central carrier of P3 in
A3 by CA3(P3). We have Q3 − CA3(P3) ≤ Q3 − P3 = Q3(I − P ) ≤ I − P . Obviously,
Q3−CA3(P3) is a central projection orthogonal to Q2, so Q3−CA3(P3)+I −C(P ) ≤ I −P .
That is Q3 − CA3(P3) + P ≤ C(P ). It implies that Q3 − CA3(P3) = 0, i.e. CA3(P3) = Q3.
Similarly, we can prove CA3(Q3 − P3) = Q3. By Lemma 2.6, G3 is a full-centralizable
point in A3.
By Lemma 2.2, G is a full-centralizable point.
Case 2 ker(G) 6= {0}
In this case, G2 and G3 are still full-centralizable points. Since ranG1 = Q1H , we
1 is a full-centralizable point in A1. By Lemma 2.3,
have ker(G∗
G1 is also a full-centralizable point in A1.
1) = {0}. By Case 1, G∗
By Lemma 2.2, G is a full-centralizable point.
6
3 Centralizers on triangular algebras
In this section, we characterize the full-centralizable points on triangular algebras.
The following theorem is our main result.
Theorem 3.1. Let J =" A M
B # be a triangular algebra, where A and B are two
0
unital Banach algebras. Then every G in J is a full-centralizable point.
Proof. Let φ : J → J be a centralizable mapping at G.
0 Z # in J , we write
Since φ is linear, for every" X Y
φ" X Y
0
0 Z # =" f11(X) + g11(Y ) + h11(Z) f12(X) + g12(Y ) + h12(Z)
f22(X) + g22(Y ) + h22(Z) # ,
where f11 : A → A, f12 : A → M, f22 : A → B, g11 : M → A, g12 : M → M,
g22 : M → B, h11 : B → A, h12 : B → M, h22 : B → B, are all linear mappings.
In the following, we denote the units of A and B by I1 and I2, respectively. We
0 B # and
write G =" A M
φ" A M
f22(A) + g22(M ) + h22(B) # .
0 B # =" f11(A) + g11(M ) + h11(B) f12(A) + g12(M ) + h12(B)
0
(3.1)
We divide our proof into several steps.
Claim 1 f12 = f22 = 0.
Let S =" X M
0 B # and T =" X −1A 0
0
I2 #, where X is an invertible element in
A. Since ST = G, we have
φ(G) = φ(S)T
=" f11(X) + g11(M ) + h11(B) f12(X) + g12(M ) + h12(B)
0 f22(X) + g22(M ) + h22(B) # .
=" ∗ f12(X) + g12(M ) + h12(B)
I2 #
f22(X) + g22(M ) + h22(B) #" X −1A 0
0
0
(3.2)
By comparing (3.1) with (3.2), we obtain f12(X) = f12(A) and f22(X) = f22(A) for
each invertible element X in A. Noting that A is a fixed element, for any nonzero
element λ in C, we have f12(λX) = f12(A) = λf12(X) = λf12(A).
It follows that
f12(X) = 0 for each invertible element X. Thus f12(X) = 0 for all X in A. Similarly,
we can obtain f22(X) = 0.
7
Claim 2 h12 = h11 = 0.
Let S =" I1
0 BZ −1 # and T =" A M
0 Z #, where Z is an invertible element in
0
B. Since ST = G, we have
φ(G) = Sφ(T )
0
0 BZ −1 #" f11(A) + g11(M ) + h11(Z) f12(A) + g12(M ) + h12(Z)
f22(A) + g22(M ) + h22(Z) #
=" I1
=" f11(A) + g11(M ) + h11(Z) f12(A) + g12(M ) + h12(Z)
# .
0
0
∗
(3.3)
By comparing (3.1) with (3.3), we obtain h12(Z) = h12(B) and h11(Z) = h11(B) for
each invertible element Z in B. Similarly as the previous discussion, we can obtain
h12(Z) = h11(Z) = 0 for all Z in B.
Claim 3 g22 = g11 = 0.
For every Y in M, we set S =" I1 M − Y
B
0
#, T =" A Y
I2 #. Obviously, ST =
0
G. Thus we have
φ(G) = φ(S)T
∗
=" ∗
=" ∗
I2 #
0 f22(I1) + g22(M − Y ) + h22(B) #" A Y
0 f22(I1) + g22(M − Y ) + h22(B) # .
∗
0
(3.4)
By comparing (3.1) with (3.4), we obtain f22(I1) + g22(M − Y ) + h22(B) = f22(A) +
g22(M ) + h22(B). Hence g22(Y ) = f22(I1 − A). It means g22(Y ) = 0 immediately.
On the other hand,
φ(G) = Sφ(T )
0
#" f11(A) + g11(Y ) + h11(I2) ∗
∗ #
=" I1 M − Y
∗ # .
=" f11(A) + g11(Y ) + h11(I2) ∗
B
0
0
(3.5)
By comparing (3.1) with (3.5), we obtain g11(Y ) = g11(M ) + h11(B − I2). Hence
g11(Y ) = 0.
According to the above three claims, we obtain that
h22(Z)#
0 Z# ="f11(X) g12(Y )
φ"X Y
0
8
0 Z # in J .
for every" X Y
Let S = " X M − XY
B
0
Claim 4 f11(X) = f11(I1)X for all X in A, and g12(Y ) = f11(I1)Y for all Y in M.
# and T = " X −1A Y
I2 #, where X is an invertible
0
element in A, and Y is an arbitrary element in M. Since ST = G, we have
φ(G) = φ(S)T
I2 #
#" X −1A Y
0
0
h22(B)
=" f11(X) g12(M − XY )
=" ∗ f11(X)Y + g12(M − XY )
h22(B) # .
=" f11(A) g12(M )
0
0
∗
#
So we have f11(X)Y = g12(XY ). It follows that
g12(Y ) = f11(I1)Y
(3.6)
(3.7)
by taking X = I1. Replacing Y in (3.7) with XY , we can obtain g12(XY ) = f11(I1)XY =
f11(X)Y for each invertible element X in A and Y in M. Since M is faithful, we have
f11(X) = f11(I1)X
(3.8)
for all invertible elements X and so for all elements in A.
Claim 5 h22(Z) = Zh22(I2) for all Z in B, and g12(Y ) = Y h22(I2) for all Y in M.
Let S =" I1
0 BZ −1 # and T =" A M − Y Z
Z
Y
0
#, where Z is an invertible element
in B, and Y is an arbitrary element in M. Since ST = G, we have
φ(G) = Sφ(T )
0
Y
0 BZ −1 #" f11(A) g12(M − Y Z)
=" I1
=" ∗ g12(M − Y Z) + Y h22(Z)
h22(B) # .
=" f11(A) g12(M )
h22(Z)
#
0
0
∗
#
(3.9)
So we have g12(Y Z) = Y h22(Z). Through a similar discussion as the proof of Claim 4,
we obtain h22(Z) = Zh22(I2) for all Z in B and g12(Y ) = Y h22(I2) for all Y in M.
Thus we have that
0 Z# ="f11(I1)X f11(I1)Y
φ"X Y
Zh22(I2)#
Zh22(I2)# ="f11(I1)X Y h22(I2)
0
0
9
for every" X Y
0 Z # in J . So it is sufficient to show that f11(I1)X = Xf11(I1) for all X
in A, and h22(I2)Z = Zh22(I2) for all Z in B. Since f11(I1)Y = Y h22(I2) for all Y in M,
we have f11(I1)XY = XY h22(I2) = Xf11(I1)Y . It implies that f11(I1)X = Xf11(I1).
Similarly, h22(I2)Z = Zh22(I2). Now we can obtain that φ(J) = φ(I)J = J φ(I) for
all J in J , where I = " I1
0
0
I2 # is the unit of J . Hence, G is a full-centralizable
point.
As applications of Theorem 3.1, we have the following corollaries.
Corollary 3.2. Let A be a nest algebra on a Hilbert space H. Then every element in
A is a full-centralizable point.
Proof. If A = B(H), then the result follows from Theorem 2.1. Otherwise, A is iso-
morphic to a triangular algebra. By Theorem 3.1, the result follows.
Corollary 3.3. Let A be a CDCSL(completely distributive commutative subspace lat-
tice) algebra on a Hilbert space H. Then every element in A is a full-centralizable
point.
Proof. It is known that A ∼= Pi∈ΛL Ai, where each Ai is either B(Hi) for some Hilbert
space Hi or a triangular algebra T ri(B, M, C) such that the conditions of Theorem 3.1
hold(see in [7] and [14]). By Lemma 2.2, the result follows.
Remark For the definition of a CDCSL algebra, we refer to [5].
4 Derivations on von Neumann algebras
In this section, we characterize the derivable mappings at a given point in a von
Neumann algebra.
Lemma 4.1. Let A be a von Neumann algebra. Suppose ∆ : A → A is a linear
mapping such that ∆(A)B + A∆(B) = 0 for each A and B in A with AB = 0. Then
∆ = D + φ, where D : A → A is a derivation, and φ : A → A is a centralizer. In
particular, ∆ is bounded.
Proof. Case 1 A is an abelian von Neumann algebra. In this case, A ∼= C(X ) for some
compact Hausdorff space X . If AB = 0, then the supports of A and B are disjoint. So
the equation ∆(A)B + A∆(B) = 0 implies that ∆(A)B = A∆(B) = 0. By Lemma 2.5,
∆ is a centralizer.
10
Case 2 A ∼= Mn(B)(n ≥ 2), where B is also a von Neumann algebra. By [1,
Theorem 2.3], ∆ is a generalized derivation with ∆(I) in the center. That is to say, ∆
is a sum of a derivation and a centralizer.
For general cases, we know A ∼=
1 or Case 2. We write A =
nPi=1L Ai, where each Ai coincides with either Case
Ai with Ai ∈ Ai and denote the restriction of ∆ in Ai
nPi=1
by ∆i. It is not difficult to check that ∆(Ai) ∈ Ai. Moreover, setting AiBi = 0, we
have ∆(Ai)Bi + Ai∆(Bi) = ∆i(Ai)Bi + Ai∆i(Bi) = 0. By Case 1 and Case 2, each ∆i
is a sum of a derivation and a centralizer. Hence, ∆ =
and a centralizer.
nPi=1
∆i is a sum of a derivation
Remark In [9], the authors prove that for a prime semisimple Banach algebra A
with nontrival idempotents and a linear mapping ∆ from A into itself, the condition
∆(A)B + A∆(B) = 0 for each A and B in A with AB = 0 implies that ∆ is bounded.
By Lemma 4.1, we have that for a von Neumann algebra A, the result holds still even
if A is not prime.
Now we prove our main result in this section.
Theorem 4.2. Let A be a von Neumann algebra acting on a Hilbert space H, and G be
a given point in A. If ∆ : A → A is a linear mapping derivable at G, then ∆ = D + φ,
where D is a derivation, and φ is a centralizer. Moreover, G is a full-derivable point if
and only if C(G) = I.
Proof. Suppose the range projection of G is P . We note that C(G) = C(P ).
Set Q1 = I − C(I − P ), Q2 = I − C(P ), and Q3 = I − Q1 − Q2. Then we have
A =
3Pi=1L(QiA). For every A in A, we write A =
For any central projection Q, setting Q⊥ = I − Q, we have
3Pi=1L Ai =
Ai =
3Pi=1
3Pi=1
QiA.
(Q⊥ + t−1QGA−1)(Q⊥G + tQA) = G,
where A is an arbitrary invertible element in A, and t is an arbitrary nonzero element
in C. So we obtain
∆(G) = (Q⊥ + t−1QGA−1)∆(Q⊥G + tQA) + ∆(Q⊥ + t−1QGA−1)(Q⊥G + tQA).
Considering the coefficient of t, we obtain Q⊥∆(QA) + ∆(Q⊥)(QA) = 0. Since the
ranges of Q and Q⊥ are disjoint, it follows that Q⊥∆(QA) = 0 and so ∆(QA) ∈ QA.
Since Qi are central projections, we have ∆(Ai) ⊆ Ai.
Denote the restriction of ∆ to Ai by ∆i. Setting AiBi = Gi, it is not difficult to
check that ∆i(Gi) = ∆(Ai)Bi + Ai∆(Bi).
Since Q1 ≤ P , we have ranG1 = ranQ1G = Q1H. So G1 is a right separating point
in A1. By [12, Corallary 2.5], ∆1 is a Jordan derivation and so is a derivation on A1.
11
Since Q2 ≤ I − P , we have G2 = Q2G = 0. By Lemma 4.1, ∆2 is a sum of a
derivation and a centralizer on A2.
Note that ranG3 = ranQ3G = Q3P = P3. As we proved before, CA3(P3) =
CA3(Q3 − P3) = Q3. So by [15, Theorem 3.1], ∆3 is a derivation on A3.
Hence, ∆ =
∆i is a sum of a derivation and a centralizer.
3Pi=1
If C(G) = I, then Q2 = 0, A = A1L A3 and G = G1 + G3 is a full-derivable point.
If C(G) 6= I, then Q2 6= 0. Define a linear mapping δ : A → A by δ(A) = A2 for all
A ∈ A. One can check that δ is not a derivation but derivable at G. Thus G is not a
full-derivable point.
As an application, we obtain the following corollary.
Corollary 4.3. Let A be a von Neumann algebra. Then A is a factor if and only if
every nonzero element G in A is a full-derivable point.
Proof. If A is a factor, for each nonzero element G in A, we know that C(G) = I. By
Theorem 4.2, G is a full-derivable point.
If A is not a factor, then there exists a nontrival central projection P . Define a
linear mapping δ : A → A by δ(A) = (I − P )A for all A ∈ A. One can check that δ is
not a derivation but derivable at P . Thus P is not a full-derivable point.
Acknowledgements. This paper was partially supported by National Natural Science
Foundation of China(Grant No. 11371136).
References
[1] G. An, J.Li, Characterizations of linear mappings through zero products or zero
Jordan products, Electron. J. Linear Algebra 31(2016), 408-424
[2] R. An, J. Hou, Characterizations of derivations on triangular rings: Additive
maps derivable at idempotents, Linear Algebra Appl. 431(2009), 1070-1080
[3] M. Bresar, Characterizing homomorphisms, derivations and multipliers in rings
with idempotents, Proc. Roy. Soc. Edinburgh, Sect. A 137(2007), 9-21.
[4] J. Cuntz, On the continuity of semi-norms on operator algebras, Math. Ann. 220
(1976), 171-183.
[5] K. Davidson, Nest algebras, Pitman Research Notes in Mathematics Series 191,
1988.
[6] A. Essaleh, M. Peralta, M. Ramirez, Weak-local derivations and homomor-
phisms on C*-algebras, arXiv:1411.4795
12
[7] F. Gilfeather, R. Moore, Isomorphisms of certain CSL algebras, J. Func.
Anal. 67(1986), 264-291.
[8] R. Kadison, J. Ringrose, Fundamentals of the Theory of Operator Algebras,
Academic Press Inc. 1983.
[9] T. Lee, C. Liu, Partially defined derivations on semisimple Banach algebras,
Studia Math. 190(2009), 193-202.
[10] J. Li, Z. Pan, Annihilator-preserving maps, multipliers and local derivations,
Linear Algebra Appl. 432(2010), 5-13.
[11] J. Li, Z. Pan, H. Xu, Characterizations of isomorphisms and derivations of some
algebras, J. Math. Anal. Appl. 332(2007), 1314-1322.
[12] J. Li, J. Zhou, Characterizations of Jordan derivations and Jordan homomor-
phisms, Linear Multilinear Algebra 59(2)(2011), 193-204.
[13] F. Lu, Characterizations of derivations and Jordan derivations on Banach alge-
bras, Linear Algebra Appl. 430(2009), 2233-2239.
[14] F. Lu, Lie derivations of certain CSL algebras, Israel Journal of Mathematics
155(2006), 149-156
[15] Z. Pan, Derivable maps and derivational points, Linear Algebra Appl. 436(2012),
4251-4260.
[16] X. Qi, J. Hou, Characterizing centralizers and generalized derivations on trian-
gular algebras by acting on zero product, Acta Math. Sinica (Engl. Ser.) 29(2013),
1245-1256.
[17] X. Qi, Characterization of centralizers on rings and operator algebras, Acta Math.
Sinica, (Chin. Ser.) 56(2013), 459-468.
[18] X. Qi, J. Hou, Additive maps derivable at some points on J -subspace lattice
algebras, Linear Algebra Appl. 429(2008), 1851-1863.
[19] W. Xu, R. An, J. Hou, Equivalent characterization of centralizers on B(H),
preprint.
[20] J. Zhu, All-derivable points of operator algebras, Linear Algebra Appl. 427(2007),
1-5.
13
|
1009.1722 | 3 | 1009 | 2011-09-07T20:34:42 | A note on trace scaling actions and fundamental groups of $C^*$-algebras | [
"math.OA"
] | Using Effros-Handelman-Shen theorem and Elliott's classification theorem of AF algebras, we show that there exists a unital simple AF algebra A with unique trace such that $A\otimes \mathbb{K}$ admits no trace scaling action of the fundamental group of A. | math.OA | math |
A NOTE ON TRACE SCALING ACTIONS AND
FUNDAMENTAL GROUPS OF C ∗-ALGEBRAS
NORIO NAWATA
Abstract. Using Effros-Handelman-Shen theorem and Elliott's clas-
sification theorem of AF algebras, we show that there exists a unital
simple AF algebra A with unique trace such that A ⊗ K admits no trace
scaling action of the fundamental group of A.
1. Introduction
Let M be a factor of type II1 with a normalized trace τ . Murray and
von Neumann introduced the fundamental group F(M ) of M in [13]. They
showed that if M is hyperfinite, then F(M ) = R×
+. Since then there has
been many works on the computation of the fundamental groups. Voiculescu
[23] showed that F(L(F∞)) of the group factor of the free group F∞ con-
tains the positive rationals and Radulescu proved that F(L(F∞)) = R×
+ in
[20]. Connes [3] showed that if G is an ICC group with property (T), then
F(L(G)) is a countable group. Popa showed that any countable subgroup
of R×
+ can be realized as the fundamental group of some factor of type II1
in [17]. Furthermore Popa and Vaes [18] exhibited a large family S of sub-
groups of R×
+ itself, all of its countable subgroups, as well
as uncountable subgroups with any Hausdorff dimension in (0, 1), such that
for each G ∈ S there exist many free ergodic measure preserving actions of
F∞ for which the associated II1 factor M has the fundamental group equal
to G. In our previous paper [15] (see also [14]), we introduced the funda-
mental group F(A) of a simple unital C ∗-algebra A with a normalized trace
τ based on the computation of Picard groups by Kodaka [10], [11] and [12].
The fundamental group F(A) is defined as the set of the numbers τ ⊗ T r(p)
for some projection p ∈ Mn(A) such that pMn(A)p is isomorphic to A. We
computed the fundamental groups of several C ∗-algebras and showed that
any countable subgroup of R×
+ can be realized as the fundamental group of
a separable simple unital C ∗-algebra with unique trace [16].
+, containing R×
The fundamental group of a II1 factor M is equal to the set of trace-
scaling constants for automorphisms of M ⊗ B(H). We have a similar fact,
that is, the fundamental group of a C ∗-algebra A is equal to the set of trace-
scaling constants for automorphisms of A ⊗ K [15] (see also [14]). It is of
interest to know whether A ⊗ K admits a trace scaling action of F(A). In
the case where M is a factor of type II1, the existence of a trace scaling
(continuous) R×
+-action on M ⊗ B(H) is equivalent to the existence of a
2000 Mathematics Subject Classification. Primary 46L40, Secondary 06F20.
Key words and phrases. Fundamental group; Trace scaling action; Dimension group:
Effros-Handelman-Shen theorem.
1
2
NORIO NAWATA
type III1 factor having a core isomorphic to M ⊗ B(H) by the continuous
decomposition of type III1 factors. (See [22] and [4].) Hence this question is
important in the theory of von Neumann algebras. Radulescu showed that
L(F∞) ⊗ B(H) admits a trace scaling action of R×
+ in [21]. Therefore there
exists a type III1 factor having a core isomorphic to L(F∞) ⊗ B(H). Popa
and Vaes [19] showed that there exists a II1 factor M such that F(M ) = R×
and M ⊗ B(H) admits no trace scaling (continuous) action of R×
+.
In this paper we consider trace scaling actions on certain AF algebras.
If A is a UHF algebra, then A ⊗ K admits a trace scaling action of F(A).
Using Effros-Handelman-Shen theorem and Elliott's classification theorem
of AF algebras, we show that there exists a unital simple AF algebra A
with unique trace such that A ⊗ K admits no trace scaling action of F(A).
Note that there exist remarkable works of the classification of trace scaling
automorphisms in [1], [7] and [8]. But we do not consider the classification
of trace scaling actions in this paper.
+
2. Examples
We recall some definitions in [15]. Let A be a unital simple C ∗-algebra
with a unique normalized trace τ and T r the usual unnormalized trace on
Mn(C). Put
F(A) := {τ⊗T r(p) ∈ R×
Then F(A) is a multiplicative subgroup of R×
+ by Theorem 3.1 in [15]. For
an additive subgroup E of R containing 1, we define the positive inner
multiplier group IM+(E) of E by
+ p is a projection in Mn(A) such that pMn(A)p ∼= A}.
IM+(E) = {t ∈ R×
+ t ∈ E, t−1 ∈ E, and tE = E}.
Then we have F(A) ⊂ IM+(τ∗(K0(A))) by Proposition 3.7 in [15]. This
obstruction enables us to compute fundamental groups easily. For x ∈ (A ⊗
K)+, set τ (x) = sup{τ ⊗ T r(y) : y ∈ ∪nMn(A), y ≤ x}. Define M+
τ = {x ≥
0 : τ (x) < ∞} and Mτ = spanM+
τ . Then τ is a densely defined (with the
domain Mτ ) lower semicontinuous trace on A⊗K. Since the normalize trace
on a unital C ∗-algebra A is unique, the lower semicontinuous densely defined
trace on A ⊗ K is unique up to constant multiple. It is clear that for any
α ∈ Aut(A⊗ K), τ ◦α is a densely defined (with the domain α−1(Mτ )) lower
semicontinuous trace on A ⊗ K. Therefore there exists a positive number
λ such that τ ◦ α = λτ , and hence α−1(Mτ ) = Mτ . We define the set of
trace-scaling constants for automorphisms:
S(A) := {λ ∈ R×
+ τ ◦ α = λτ for some α ∈ Aut(A ⊗ K) }.
Then F(A) = S(A) by Proposition 3.28 in [15]. Therefore it is of interest
to know whether A ⊗ K admits a trace scaling action of F(A).
It is clear that if the fundamental group of A is singly generated, A ⊗ K
admits a trace scaling action of F(A). See [15] and [16] for such examples.
We shall show some examples of AF algebras A such that A ⊗ K admits a
trace scaling action of F(A).
Example 2.1. Consider a UHF algebra M2∞3∞ . Then the fundamental
group of M2∞3∞ is a multiplicative subgroup generated by 2 and 3. Hence
TRACE SCALING ACTIONS AND FUNDAMENTAL GROUPS
3
F(M2∞3∞ ) is isomorphic to Z2 as a group. Since M2∞3∞ ⊗ K is isomorphic
to M2∞ ⊗ K⊗M3∞ ⊗ K, there exists a trace scaling Z2-action on M2∞3∞ ⊗ K.
In general, if A is a UHF algebra, then F(A) is a free abelian group (see
[15]) and A ⊗ K admits a trace scaling action of F(A).
Example 2.2. Let A be a unital simple AF algebra such that K0(A) =
Z + Z√3, K0(A)+ = (Z + Z√3) ∩ R+ and [1]0 = 1. Then F(A) = {(2 +
√3)n : n ∈ Z} (see Proposition 3.17 and Corollary 3.18 in [15]). Consider
5 ]√3 and τ∗
B = M5∞ ⊗ A. Then it is easily seen that τ∗(K0(B)) = Z[ 1
is an order isomorphism. We shall show that IM+(τ∗(K0(B))) is generated
by 5 and 2 + √3. Since τ∗(K0(B)) is a subring of R, IM+(τ∗(K0(B))) is
a group of positive invertible elements. Define a multiplicative map N of
5 ] + Z[ 1
Z[ 1
If
a + b√3 is an invertible element in Z[ 1
5 ]√3, then there exists an
integer n such that N (a + b√3) = ±5n. Elementary computations shows
that x2 − 3y2 ≡ 0 mod 25 implies x ≡ 0 mod 5 and y ≡ 0 mod 5. It is
easy to see that no integers x and y satisfy equations x2 − 3y2 = ±5 or
x2 − 3y2 = −1. There exist integers x and y satisfy the equation x2 − 3y2 =
(See, for example, [9] and [15].) Therefore it can be easily checked
−1.
that IM+(τ∗(K0(B))) is generated by 5 and 2 + √3. Hence we see that
F(B) = {5n(2 + √3)m : n, m ∈ Z} by Proposition 3.18 in [15] and B ⊗ K
5 ] by N (a + b√3) = a2 − 3b2 for any a, b ∈ Z[ 1
5 ]√3 to Z[ 1
5 ] + Z[ 1
5 ]+ Z[ 1
5 ].
admits a trace scaling action.
We shall show that there exists a unital simple AF algebra A with unique
j + k√3
trace such that A ⊗ K admits no trace scaling action of F(A). Define
E = {(
E+ = {(r,(cid:18) x
,(cid:18) x
y (cid:19)) ∈ E : r > 0} ∪ {(0,(cid:18) 0
y (cid:19)) ∈ R×Z2 i, j, k, x, y ∈ Z, x ≡ j mod 9, y ≡ k mod 3}
0 (cid:19))} and [u]0 = (1,(cid:18) 1
0 (cid:19)).
56i
Then there exists a simple AF -algebra A with a unique normalized trace τ
such that (K0(A), K0(A)+, [1A]0) = (E, E+, u) by Effros-Handelman-Shen
theorem [5].
Lemma 2.3. With notation as above the fundamental group of A is equal
to the multiplicative group generated by 5 and 2 + √3.
Proof. Since τ∗(K0(A)) is equal to Z[ 1
5 ]√3, F(A) is a subgroup of
{5n(2 + √3)m : n, m ∈ Z} by an argument in Example 2.2. Define an
additive homomorphism φ : E → E by
5 ] + Z[ 1
9
φ((r,(cid:18) x
y (cid:19))) = (5r,(cid:18) 5
6 11 (cid:19)(cid:18) x
y (cid:19)).
Computations show that φ is a well-defined order isomorphism of E with
φ(u) = (5,(cid:18) 5
in Mn(A) such that [p]0 = (5,(cid:18) 5
(K0(pMn(A)p), K0(pMn(A)p)+, [p]0) = (E, E+, (5,(cid:18) 5
6 (cid:19)). There exist a natural number n and a projection p
6 (cid:19)) and τ ⊗ T r(p) = 5. Since we have
6 (cid:19))), there exists an
4
NORIO NAWATA
isomorphism f : A → pMn(A)p with f∗ = φ by Elliott's classification theo-
rem of AF algebra [6]. Therefore 5 ∈ F(A). Define an additive homomor-
phism ψ : E → E by
ψ((r,(cid:18) x
y (cid:19))) = ((2 + √3)r,(cid:18) 2 3
1 2 (cid:19)(cid:18) x
y (cid:19)).
Then we see that 2 + √3 ∈ F(A). Consequently F(A) is the multiplicative
group generated by 5 and 2 + √3.
(cid:3)
We shall consider the order automorphisms of (E, E+).
Lemma 2.4. Let φ be an order automorphism of (E, E+). Then there exist
integers a, b, c, d and a positive invertible element λ in Z[ 1
that ad − bc = ±1 and
5 ]√3 such
5 ] + Z[ 1
φ((r,(cid:18) x
y (cid:19))) = (λr,(cid:18) a b
c d (cid:19)(cid:18) x
y (cid:19)).
Moreover if λ = 5, then
(cid:18) a b
c d (cid:19) ≡ (cid:18) 5 0
0 2 (cid:19) ,(cid:18) 5 0
3 2 (cid:19) ,(cid:18) 5 0
6 2 (cid:19) mod 9
and if λ = 2 + √3, then
y (cid:19)))
(cid:18) a b
y (cid:19))), φ2((r,(cid:18) x
4 2 (cid:19) ,(cid:18) 2 3
7 2 (cid:19) mod 9.
der isomorphism. Hence there exist integers m1, m2, m3 and m4 such
1 2 (cid:19) ,(cid:18) 2 3
c d (cid:19) ≡ (cid:18) 2 3
Proof. We denote by (φ1((r,(cid:18) x
for any (r,(cid:18) x
and (0,(cid:18) 0
that m1m4 − m2m3 = ±1 and φ2((0,(cid:18) x
for any (0,(cid:18) x
y (cid:19)) ∈ E. Consider a subgroup F generated by (0,(cid:18) 9
y (cid:19)))) the element φ((r,(cid:18) x
0 (cid:19))
3 (cid:19)). Then F is an φ-invariant subgroup because φ is an or-
3 m4 (cid:19)(cid:18) x
y (cid:19)
y (cid:19)) ∈ F . Furthermore we see that there exists a pos-
5 ]√3 such that φ1((r,(cid:18) x
y (cid:19))) =
5 ] + Z[ 1
0 (cid:19))) = φ((9,(cid:18) 0
0 (cid:19))) for any i ∈ Z, we see that
0 (cid:19)). This observation and easy computations show
3 ∈ Z. In a similar
1 (cid:19)). It is easily
1 (cid:19))),
φ((9,(cid:18) 0
3 m4 (cid:19)(cid:18) 1
that φ((1,(cid:18) 1
way, we see that φ((√3,(cid:18) 0
1 (cid:19))) = (λ,(cid:18) m1 3m2
seen that φ is determined by the values of φ((1,(cid:18) 1
56i ,(cid:18) 0
0 (cid:19))) = (9λ,(cid:18) 0
0 (cid:19)) and m3
3 m4 (cid:19)(cid:18) 0
0 (cid:19))) = (λ,(cid:18) m1 3m2
y (cid:19))) = (cid:18) m1 3m2
0 (cid:19))), φ((√3,(cid:18) 0
itive invertible element λ in Z[ 1
λr. Since 56iφ(( 9
m3
m3
m3
TRACE SCALING ACTIONS AND FUNDAMENTAL GROUPS
5
φ((0,(cid:18) 9
0 (cid:19))) and φ((0,(cid:18) 0
a positive invertible element λ in Z[ 1
3 (cid:19))). Therefore there exist integers a, b, c, d and
5 ]√3 such that ad− bc = ±1 and
c d (cid:19)(cid:18) x
5 ] + Z[ 1
y (cid:19))) = (λr,(cid:18) a b
y (cid:19)).
φ((r,(cid:18) x
Let λ = 5, then a ≡ 5 mod 9, b ≡ 0 mod 9, c ≡ 0 mod 3 and d ≡ 5 mod 3
by the definition of E. If ad − bc = 1, then d ≡ 55 mod 9, −b ≡ 0 mod 9,
−c ≡ 0 mod 3 and a ≡ 55 mod 3 because φ is an isomorphism. Therefore
computations show
(cid:18) a b
c d (cid:19) ≡ (cid:18) 5 0
0 2 (cid:19) ,(cid:18) 5 0
3 2 (cid:19) ,(cid:18) 5 0
6 2 (cid:19) mod 9.
If ad − bc = −1, then then −d ≡ 55 mod 9, b ≡ 0 mod 9, c ≡ 0 mod 3 and
−a ≡ 55 mod 3. There does not exist a integer a such that a ≡ 5 mod 9 and
−a ≡ 55 mod 3. Therefore we reach a conclusion in the case λ = 5. In the
case λ = 2 + √3, the similar argument as above proves the lemma.
Theorem 2.5. There exists a unital simple AF algebra A with unique trace
such that A ⊗ K admits no trace scaling action of F(A).
Proof. Let
(cid:3)
j + k√3
56i
E = {(
E+ = {(r,(cid:18) x
,(cid:18) x
y (cid:19)) ∈ E : r > 0} ∪ {(0,(cid:18) 0
y (cid:19)) ∈ R×Z2 i, j, k, x, y ∈ Z, x ≡ j mod 9, y ≡ k mod 3}
0 (cid:19))} and [u]0 = (1,(cid:18) 1
0 (cid:19)).
Then there exists a simple AF algebra A with a unique normalized trace τ
such that (K0(A), K0(A)+, [1A]0) = (E, E+, u) by Effros-Handelman-Shen
theorem [5]. By Lemma 2.3, F(A) = {5n(2 + √3)m : n, m ∈ Z}. Let α be
an automorphism of A ⊗ K such that τ ◦ α = 5τ and β an automorphism of
A⊗K such that τ◦β = (2+√3)τ . Then α∗ and β∗ are order isomorphisms of
(K0(A), K0(A)+). Lemma 2.4 and computations show that α∗◦β∗ 6= β∗◦α∗.
Therefore A ⊗ K admits no trace scaling action of F(A).
Remark 2.6. Let A be a unital simple C ∗-algebra with a unique normalized
trace τ . We denote by Pic(A) the Picard group of A (see [2]). Assume that
the normalized trace on A separates equivalence classes of projections. Then
we have the following exact sequence [15] (see also [10]).
(cid:3)
1 −−−−→ Out(A)
ρA−−−−→ Pic(A)
T−−−−→ F(A) −−−−→ 1.
If A ⊗ K admits a trace scaling action of F(A), then Pic(A) is isomorphic
to a semidirect product of Out(A) with F(A). Example 2.1 and Example
2.2 are such examples. We do not know whether there exists a simple C ∗-
algebra A with a unique normalized trace τ such that the normalized trace
on A separates equivalence classes of projections and A⊗ K admits no trace
scaling action of F(A).
Remark 2.7. If A is a C ∗-algebra in the proof of Theorem 2.5, then it can
be checked that Out(A) is not a normal subgroup of Pic(A) by Lemma 2.4,
Proposition 1.5 in [10] and Elliott's classification theorem of AF algebras.
6
NORIO NAWATA
References
[1] O. Bratteli and A. Kishimoto, Trace scaling automorphisms of certain stable AF
algebras. II, Q. J. Math. 51 (2000), 131 -- 154.
[2] L. G. Brown, P.Green and M. A. Rieffel, Stable isomorphism and strong Morita
equivalence of C ∗-algebras, Pacific J. Math. 71 (1977), 349-363.
[3] A. Connes, A factor of type I I1 with countable fundamental group, J. Operator Theory
4 (1980), 151 -- 153.
[4] A. Connes and M. Takesaki, The flow of weights on factors of type III, Tohoku Math.
J. (2) 29 (1977), 473 -- 575.
[5] E. Effros, D. Handelman and C. L. Shen, Dimension groups and their affine repre-
sentations, Amer. J. Math. 102 (1980), 385 -- 407.
[6] G. A. Elliott, On the classification of inductive limits of sequences of semisimple
finite-dimensional algebras, J. Algebra 38 (1976), 29 -- 44.
[7] G. A. Elliott, D. E. Evans and A. Kishimoto, Outer conjugacy classes of trace scaling
automorphisms of stable UHF algebras, Math. Scand. 83 (1998), 74 -- 86.
[8] D. E. Evans and A. Kishimoto, Trace scaling automorphisms of certain stable AF
algebras, Hokkaido Math. J. 26 (1997), 211 -- 224.
[9] M. Jacobson and H. Williams, Solving the Pell Equation, Springer, 2009.
[10] K. Kodaka, Full projections, equivalence bimodules and automorphisms of stable al-
gebras of unital C ∗-algebras, J. Operator Theory, 37 (1997), 357 -- 369.
[11] K. Kodaka, Picard groups of irrational rotation C ∗-algebras, J. London Math. Soc.
(2) 56 (1997), 179 -- 188.
[12] K. Kodaka, Projections inducing automorphisms of stable UHF-algebras, Glasg. Math.
J. 41 (1999), 345 -- 354.
[13] F. Murray and J. von Neumann, On rings of operators IV, Ann. Math. 44, (1943),
716 -- 808.
[14] N. Nawata, Fundamental group of simple C ∗-algebras with unique trace III, to appear
in Canad. J. Math. doi:10.4153/CJM-2011-052-0.
[15] N. Nawata and Y. Watatani, Fundamental group of simple C ∗-algebras with unique
trace, Adv. Math. 225 (2010), 307 -- 318.
[16] N. Nawata and Y. Watatani, Fundamental group of simple C ∗-algebras with unique
trace II, J. Funct. Anal. 260 (2011), no. 2, 428 -- 435.
[17] S. Popa, Strong rigidity of I I1 factors arising from malleable actions of w-rigid groups,
I, Invent. Math. 165 (2006), 369 -- 408.
[18] S. Popa and S. Vaes, Actions of F∞ whose I I1 factors and orbit equivalence relations
have prescribed fundamental group, J. Amer. Math. Soc. 23 (2010), 383 -- 403.
[19] S. Popa and S. Vaes, On the fundamental group of II1 factors and equivalence relations
arising from group actions, Quanta of maths, 519 -- 541, Clay Math. Proc., 11, Amer.
Math. Soc., Providence, RI, 2010.
[20] F. Radulescu, The fundamental group of the von Neumann algebra of a free group
with infinitely many generators is R∗
+, J. Amer. Math. Soc. 5 (1992), 517 -- 532.
[21] F. Radulescu, A one-parameter group of automorphisms of L(F∞) ⊗ B(H) scaling the
trace, C. R. Acad. Sci. Paris Ser. I Math. 314 (1992), 1027 -- 1032.
[22] M. Takesaki, Duality for crossed products and the structure of von Neumann algebras
of type III, Acta Math. 131 (1973), 249 -- 310.
[23] D. Voiculescu, Circular and semicircular systems and free product factors, in Operator
algebras, unitary representations, enveloping algebras, and invariant theory, Progr.
Math. 92, Birkhauser, Boston, 1990, 45 -- 60.
(Norio Nawata) Graduate School of Mathematics, Kyushu University, Mo-
tooka, Fukuoka, 819-0395, Japan
E-mail address: [email protected]
|
1805.07236 | 5 | 1805 | 2019-11-18T15:05:50 | Approximate equivalence of representations of AH algebras into semifinite von Neumann factors | [
"math.OA"
] | In this paper, we prove a non-commutative version of the Weyl-von Neumann theorem for representations of unital, separable AH algebras into countably decomposable, semifinite, properly infinite, von Neumann factors, where an AH algebra means an approximately homogeneous ${\rm C}^{\ast}$-algebra. We also prove a result for approximate summands of representations of unital, separable AH algebras into finite von Neumann factors. | math.OA | math |
Approximate equivalence of representations of AH
algebras into semifinite von Neumann factors
Junhao Shen and Rui Shi
Abstract. In this paper, we prove a non-commutative version of the Weyl-von Neumann theo-
rem for representations of unital, separable AH algebras into countably decomposable, semifinite,
properly infinite, von Neumann factors, where an AH algebra means an approximately homo-
geneous C∗-algebra. We also prove a result for approximate summands of representations of
unital, separable AH algebras into finite von Neumann factors.
1. Introduction
The classical Weyl-von Neumann theorem states that, for each bounded linear self-adjoint
operator a on a separable Hilbert space H, there is a diagonal self-adjoint operator d such that
a − d is a Hilbert-Schmidt operator of arbitrarily small norm, which was first proved by Weyl
[30] in 1909 and was improved by von Neumann [22] in 1935. This theorem provides important
techniques in the perturbation theory for bounded linear operators on H.
In 1971, a version of Weyl-von Neumann theorem for normal operators was proved in [2].
It states that a normal operator a is diagonalizable up to an arbitrarily small compact pertur-
bation. As a corollary of the main theorem in [29], Voiculescu proved that a normal operator
is diagonalizable up to an arbitrarily small Hilbert-Schimidt perturbation. Later, another proof
was provided by Davidson in [6].
As one important ingredient of the famous Brown-Douglas-Fillmore theory, the Weyl-von
Neumann theorem attracted much attention during the past decades. While Voiculescu [28]
proved the striking non-commutative Weyl-von Neumann theorem in B(H), several interesting
commutative versions of the Weyl-von Neumann theorem are proved in the setting of semifinite
von Neumann algebras ([31, 15, 16, 11]).
One goal of the current paper is to set up a non-commmutative version of the Weyl-von
Neumann theorem in semifinite factor von Neumann algebras. For this purpose, we have to
face two main obstacles. One is the fact that a semifinite von Neumann algebra might contain
no minimal projections, since minimal projections play an important role in the proof of the
non-commmutative Weyl-von Neumann theorem in the case of B(H). The other obstacle is that
the non-commutative AH algebras are much more complicated than commutative C∗-algebras.
To deal with these two obstructions, we develop new techniques in semifinite von Neumann
algebras.
We recall several terms in von Neumann algebras. A von Neumann algebra is a ∗-algebra of
bounded linear operators on a Hilbert space which is closed in the weak operator topology and
contains the identity. A factor (or von Neumann factor ) is a von Neumann algebra with trivial
center. Factors are classified by Murray and von Neumann [21] into three types, i.e., type I,
type II, and type III factors. A factor is called semifinite if it is of type I or II. This is equivalent
2010 Mathematics Subject Classification. Primary 47C15.
Key words and phrases. Weyl-von Neumann theorem, AH algebras, semifinite von Neumann factors.
1
2
JUNHAO SHEN AND RUI SHI
to say that a factor equipped with a faithful, normal, semifinite, tracial weight, is semifinite (see
Definition 7.5.1 of [14] for a weight on a C∗-algebra). A factor is called (properly) infinite if the
identity is an infinite projection. The reader is referred to [13, 14, 26, 27, 3] for the theory
of von Neumann algebras. Throughout this paper, let H be a complex, separable Hilbert space
and B(H) be the set of all the bounded linear operators on H. By definition, B(H) is a factor
of type I.
In the setting of semifinite von Neumann factors, Kaftal [15] and Zsid´o [31] proved the
extended Weyl-von Neumann theorem for self-adjoint operators. Inspired by these perturbation
results in semifinite von Neumann factors, the authors of [16] proved a Weyl-von Neumann
theorem for normal operators with respect to an arbitrarily small max{k · k, k · k2}-perturbation
in semifinite von Neumann factors (M, τ ), where by k · k we denote the operator norm and the
norm k · k2 is defined as kxk2 := τ (x2)1/2 for every x ∈ M (see Theorem 6.1.2 of [16]).
In a different point of view, the Weyl-von Neumann-Berg theorem states that, for a unital,
separable, commutative C∗-subalgebra A of B(H), every ∗-representation of A is approximately
unitarily equivalent to a diagonal representation relative to K(H), where K(H) denotes the
two-sided closed ideal of all the compact operators in B(H) (see Theorem II.4.6 of [5] for more
details).
It is remarkable that Voiculescu proved the non-commutative Weyl-von Neumann theorem
in [28] for every unital separable C∗-subalgebra of B(H) instead of commutative ones.
In
the current paper, we will refer to the theorem in [28] as Voiculescu's theorem for simplicity.
Precisely, Voiculescu's theorem is cited as follows:
Voiculescu's Theorem ([28]). Suppose A is a separable unital C∗-algebra and H is a
complex, separable Hilbert space. Let φ and ψ be unital ∗-representations of A into B(H). The
following statements are equivalent:
(1) φ ∼a ψ.
(2) φ ∼A ψ mod K(H).
(3) ker φ = ker ψ, φ−1 (K(H)) = ψ−1 (K(H)), and the nonzero parts of the restrictions
φφ−1(K(H)) and ψψ−1(K(H)) are unitarily equivalent.
In this theorem, by φ ∼a ψ, we mean the approximately unitary equivalence of φ and ψ, i.e.,
there exists a sequence of unitary operators {un}∞
n=1 in B(H) such that
lim
n→∞
ku∗
nφ(a)un − ψ(a)k = 0,
∀ a ∈ A.
By φ ∼A ψ mod K(H), we mean the approximately unitary equivalence of φ and ψ relative to
K(H), i.e., there exists a sequence of unitary operators {un}∞
n=1 in B(H) such that
(1) for each a in A and every n ∈ N,
(2) for each a in A,
u∗
nφ(a)un − ψ(a) ∈ K(H), and
lim
n→∞
ku∗
nφ(a)un − ψ(a)k = 0.
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
3
This theorem is important in both operator theory and operator algebras. Its applications
can be found in the diagonalization problem of normal operators in [29], the eighth problem
proposed by Halmos in [12], and the well-known Brown-Douglas-Fillmore theory (see Chapter
IX of [5]).
By introducing quasi-central approximate units of C∗-algebras, Arveson provided another
beautiful proof of Voiculescu's theorem in [1]. Later, Hadwin [10] provided an algebraic char-
acterization of approximate equivalence of representation.
In [10], Hadwin also proved the
following result:
Hadwin's Theorem (Lemma 2.3 of [10]). Suppose that A is a separable unital C∗-algebra,
H0 and H1 are Hilbert spaces, Let φ : A →B(H0) and ψ : A →B(H1) be unital representations.
The following are equivalent:
(1) There is a representation γ : A → B(H2) for some Hilbert space H2 such that
(2) For every A ∈ A,
ψ ⊕ γ ∼a φ.
rank (ψ (A)) ≤ rank (φ (A)) .
In [7], Ding and Hadwin extended some results of [10] to the case where B(H0) is replaced
with a von Neumann algebra.
As another important application of [28], the authors of [4] characterized properly infinite
injective von Neumann algebras and nuclear C∗-algebras by using a uniqueness theorem.
Inspired by the preceding interesting results, we focus on analogues of Voiculescu's theorem
and Hadwin's theorem in the setting of semifinite von Neumann factors.
This paper is organized as follows. Since factors of type II contain no minimal projections,
they are quite different from factors of type I. Thus, to prove the main theorems in the cur-
rent paper, we need to prepare related notation and definitions in Section 2. In particular, we
introduce the strongly-approximately-unitarily-equivalent ∗-homomorphisms which was first de-
fined in [16] to extend the concept of approximately unitarily equivalence of ∗-homomorphisms
relative to K(H) in the setting of B(H).
Note that, at the end of [7], Hadwin pointed out that there exist unital representations π
and ρ of C∗(F2) into a hyperfinite type II1 factor (R, τ ) with τ ◦ π = τ ◦ ρ such that π and ρ are
not weakly approximately equivalent in R. In this sense, we can't expect to extend Voiculescu's
theorem for every C∗-subalgebra in semifinite von Neumann factors.
In terms of [11], it is reasonable to choose the family of AH-algebras, a classical collection
of inductive limit C∗-algebras, to build up a non-commutative version of Voiculescu's theorem
in semifinite von Neumann factors. Meanwhile, the choice makes sense, since the family of AH
algebras plays an important role in the study of C∗-algebras (see [8, 9, 18, 19, 20, 23]). Hence,
we prepare the definition of AH algebras and certain properties of AH algebras from [24, 3] in
Section 2.
In Section 3, we prove the following theorem for AH algebras in II1 factors:
4
JUNHAO SHEN AND RUI SHI
THEOREM 3.7. Let A be a unital separable AH subalgebra in a type II1 factor (N , τ ) with
separable predual. Let p be a projection in (N , τ ). Suppose that π : A → N is a unital ∗-
homomorphism and ρ : A → pN p is a unital ∗-homomorphism such that:
Then, there exists a unital ∗-homomorphism γ : A → p⊥N p⊥ such that
τ (R (ρ (a))) ≤ τ (R (π (a))),
∀ a ∈ A.
ρ ⊕ γ ∼a π
in N .
In Theorem 3.7, by ρ ⊕ γ ∼a π in N , we mean the approximately unitary equivalence of π
and ρ ⊕ γ in N , i.e., there exists a sequence of unitary operators {un}∞
n=1 in N such that
ku∗
lim
n→∞
n(cid:0)(ρ ⊕ γ)(a)(cid:1)un − π(a)k = 0,
∀ a ∈ A.
In Section 4, we prove an extended Voiculescu's theorem for AH algebras in semifinite,
(properly) infinite von Neumann factors:
THEOREM 4.11. Let M be a countably decomposable, properly infinite, semifinite factor with
a faithful, normal, semifinite, tracial weight τ . Suppose that A is a separable AH subalgebra of
M with an identity IA.
If φ and ψ are unital ∗-homomorphisms of A into M, then the following statements are
equivalent:
(i) φ ∼a ψ in M;
(ii) φ ∼A ψ mod K(M, τ ).
The reader is referred to Definition 2.5 in Section 2 for the notation φ ∼A ψ mod K(M, τ ).
2. Preliminary
In the following definition, the definitions of finite rank operators and compact operators in
B(H) are extended to analogues in von Neumann algebras with a faithful, normal, semifinite,
tracial weight τ . These definitions will be frequently mentioned in this paper.
Definition 2.1. Let (M, τ ) be a von Neumann algebra with a faithful, normal, semifinite,
tracial weight τ , where M acts on a Hilbert space H. Let k · k denote the operator norm.
Define
(2.1)
to be the set of finite trace projections in (M, τ ). In terms of PF (M, τ ), define F (M, τ ) to be
the set in the form
PF (M, τ ) = {p : p = p∗ = p2 ∈ M and τ (p) < ∞}
F (M, τ ) = {xpy : p ∈ PF (M, τ ) and x, y ∈ M}.
(2.2)
Each element in F (M, τ ) is said to be of (M, τ )-finite-rank. When no confusion can arise,
elements in F (M, τ ) are called finite-rank operators. If M is of type I, then F (M, τ ) coincides
with the set of finite rank operators. In this point of view, the concept of finite-rank operator in
(M, τ ) is a natural analogue of the concept of finite rank operator in B(H).
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
5
Define K(M, τ ) to be the k · k-norm closure of F (M, τ ) in M. Each element in K(M, τ ) is
said to be compact in M.
For an element x ∈ M, denote by R(x) the range projection of x. From Proposition 6.1.6 of
[14], an operator a in M is of (M, τ )-finite-rank if and only if τ (R(a)) < ∞.
Remark 2.2. By virtue of Theorem 6.8.3 of [14], K(M, τ ) is a k · k-norm closed two-sided
ideal in M. We denote by K(M) the k · k-norm closed ideal generated by finite projections in
M. In general, K(M, τ ) is a subset of K(M). That is because a finite projection might not be a
finite-rank projection with respect to τ . However, if M is a countably decomposable, semifinite
factor, then Proposition 8.5.2 entails that
for a faithful, normal, semifinite tracial weight τ .
K(M, τ ) = K(M)
To introduce the definition of approximate equivalence of two unital ∗-homomorphisms of
a separable C∗-algebra A into M (relative to K(M, τ )), we need to develop the following
notation and definitions.
Suppose that {ei,j}∞
i,j=1 is a system of matrix units for B(l2). For a countably decomposable,
properly infinite von Neumann algebra M with a faithful normal semifinite tracial weight τ ,
there exists a sequence {vi}∞
i=1 of partial isometries in M such that
∞
viv∗
i = IM,
v∗
i vi = IM,
and vjv∗
i = 0 when i 6= j.
(2.3)
Xi=1
Definition 2.3. For all x ∈ M and all P∞
φ : M → M ⊗ B(l2)
i,j=1 xi,j ⊗ ei,j ∈ M ⊗ B(l2), define
and ψ : M ⊗ B(l2) → M
by
φ(x) =
Xi,j=1
∞
∞
∞
(vixv∗
j ) ⊗ ei,j
and
ψ(
xi,j ⊗ ei,j) =
v∗
i xi,jvj.
Xi,j=1
Xi,j=1
where {vi}∞
i=1 is a sequence of partial isometries in M as in (2.3) and {ei,j}∞
i,j=1 is a system of
matrix units for B(l2) such that P∞
We further define a mapping τ : (M ⊗ B(l2))+ → [0, ∞] to be
i=1 ei,i equals the identity of B(l2).
τ (y) = τ (ψ(y)),
∀ y ∈ (M ⊗ B(l2))+.
By Lemma 2.2.2 of [16], both φ and ψ are normal ∗-homomorphisms satisfying
ψ ◦ φ = idM
and
φ ◦ ψ = idM⊗B(l2).
The following statements are proved in Lemma 2.2.4 of [16]:
(i) τ is a faithful, normal, semifinite tracial weight of M ⊗ B(l2).
∞
∞
∞
(ii) τ (
Xi,j=1
xi,j ⊗ ei,j) =
Xi=1
τ (xi,i) for all
Xi,j=1
xi,j ⊗ ei,j ∈ (M ⊗ B(l2))+.
6
(iii)
JUNHAO SHEN AND RUI SHI
PF (M ⊗ B(l2), τ ) = φ(PF (M, τ )),
F (M ⊗ B(l2), τ ) = φ(F (M, τ )),
K(M ⊗ B(l2), τ ) = φ(K(M, τ )).
Remark 2.4. Note that τ is a natural extension of τ from M to M ⊗ B(l2). If no confusion
arises, τ will be also denoted by τ . By Proposition 2.2.9 of [16], the ideal K(M ⊗ B(l2), τ ) is
independent of the choice of the system of matrix units {ei,j}∞
i,j=1 of B(l2) and the choice of the
family {vi}∞
i=1 of partial isometries in M.
Now we are ready to introduce the definition of approximate equivalence of ∗-homomorphisms
of a separable C∗-algebra into M relative to K(M, τ ).
Let A be a separable C∗-subalgebra of M with an identity IA. Suppose that ρ is a positive
mapping from A into M such that ρ(IA) is a projection in M. Then for all 0 ≤ x ∈ A, we have
0 ≤ ρ(x) ≤ kxkρ(IA). Therefore, it follows that
ρ(x)ρ(IA) = ρ(IA)ρ(x) = ρ(x)
for all positive x ∈ A.
ψ(A) ⊆ ψ(IA)Mψ(IA).
In other words, ψ(IA) can be viewed as an identity of ψ(A). Or,
The following definition is a special case of Definition 2.3.1 of [16] when the norm is fixed
to be the operator norm k · k.
Definition 2.5. Let A be a separable C∗-subalgebra of M with an identity IA and B a ∗-
subalgebra of A such that IA ∈ B. Suppose that {ei,j}i,j≥1 is a system of matrix units for B(l2).
Let M, N ∈ N ∪ {∞}. Suppose that ψ1, . . . , ψM and φ1, . . . , φN are positive mappings from A
into M such that ψ1(IA), . . . , ψM (IA), φ1(IA), . . . , φN (IA) are projections in M.
(a) Let F ⊆ A be a finite subset and ǫ > 0. Then we say that
ψ1 ⊕ · · · ⊕ ψM is (F , ǫ)-strongly-approximately-unitarily-equivalent to φ1 ⊕ · · · ⊕ φN over B,
denoted by
ψ1 ⊕ ψ2 ⊕ · · · ⊕ ψM ∼(F ,ǫ)
B
φ1 ⊕ φ2 ⊕ · · · ⊕ φN ,
mod K(M, τ )
if there exists a partial isometry v in M ⊗ B(l2) such that
M
N
(i) v∗v =
Xi=1
φi(IA) ⊗ ei,i;
ψi(IA) ⊗ ei,i and vv∗ =
Xi=1
ψi(x) ⊗ ei,i − v∗ N
φi(x) ⊗ ei,i! v ∈ K(M ⊗ B(l2), τ ) for all x ∈ B;
Xi=1
Xi=1
ψi(x) ⊗ ei,i − v∗ N
φi(x) ⊗ ei,i! vk < ǫ for all x ∈ F .
Xi=1
Xi=1
M
(ii)
M
(iii) k
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
7
(b) We say that
ψ1 ⊕ · · · ⊕ ψM is strongly-approximately-unitarily-equivalent to φ1 ⊕ · · · ⊕ φN over B,
denoted by
ψ1 ⊕ ψ2 ⊕ · · · ⊕ ψM ∼B φ1 ⊕ φ2 ⊕ · · · ⊕ φN ,
mod K(M, τ )
if, for any finite subset F ⊆ B and ǫ > 0,
ψ1 ⊕ ψ2 ⊕ · · · ⊕ ψM ∼(F ,ǫ)
B
φ1 ⊕ φ2 ⊕ · · · ⊕ φN ,
mod K(M, τ ).
By virtue of the preceding definitions, assume that M = B(H) for a complex, separable,
infinite dimensional, Hilbert space H, φ and ψ are unital ∗-homomorphisms of A into M. It
follows that φ and ψ are strongly-approximately-unitarily-equivalent over A if and only if φ and
ψ are approximately unitarily equivalent relative to K(H).
In the following, we recall the definitions of inductive limit C∗-algebras, AH algebras and
certain useful properties.
Remark 2.6. By Proposition 6.2.4 of [24], every inductive sequence of C∗-algebras
A1
φ1
/ A2
φ2
/ A3
φ3
/ · · ·
has an inductive limit (A, {ϕn}n≥1) which is also a C∗-algebra such that
(1) the diagram
φn
An+1
①①①①①①①①①
ϕn+1
An
ϕn
A
commutes for each n in N, where φn's and ϕn's are ∗-homomorphisms;
(2) the C∗-algebra A equals the norm-closure of the union of ϕn(An) i.e.,
A = ∪n≥1ϕn(An)
k·k
.
(2.4)
(2.5)
(2.6)
Note that the diagram in (2.5) implies that {ϕn(An)}n≥1 forms a monotone increasing sequence
of C∗-algebras.
If this inductive limit C∗-algebra A is a subalgebra of (M, τ ) and K(M, τ ) is as in (2.1),
then Lemma 3.4.1 of [5] entails that:
A ∩ K(M, τ ) = ∪n≥1(An ∩ K(M, τ ))
k·k
.
(2.7)
By Definition V.2.1.9 of [3], a separable C∗-algebra A is AH if it is ∗-isomorphic to an
inductive limit of locally homogeneous C∗-algebras (in the sense of IV.1.4.1 of [3]). In addition,
the following are useful:
(i) A C∗-algebra A is n-subhomogeneous, if and only if A∗∗ is a direct sum of Type Im von
Neumann algebras for m ≤ n. In particular, A is n-homogeneous, if and only if A∗∗ is
a Type In von Neumann algebra (IV.1.4.6 of [3]);
/
/
/
/
/
8
JUNHAO SHEN AND RUI SHI
(ii) If ϕ : A → B is a bounded linear mapping between C∗-algebra, then by general consider-
ations ϕ∗∗ : A∗∗ → B∗∗ is a normal linear mapping of the same norm as ϕ. In addition,
ϕ∗∗ is a ∗-homomorphism if and only if ϕ is a ∗-homomorphism. (III.5.2.10 of [3]).
As a quick application, if φ is a unital ∗-homomorphism of a unital locally homogeneous
C∗-algebra A into another unital C∗-algebra B, then φ(A) is also locally homogeneous.
For more about inductive limit, see Chapter XIV of [27] and Chapter 6 of [24]. It is conve-
nient to assume that ϕn's are injective ∗-homomorphisms and {An}n≥1 is an increasing sequence
of C∗-subalgebras of A whose union is norm-dense in A.
3. Representations of AH algebras to type II1 factors
In this section, we always assume that (N , τ ) is a type II1 factor with separable predual,
where τ is the faithful, normal, tracial state. For two ∗-homomorphisms ρ and π of a unital
C∗-algebra A into N , if there is a unitary operator u in N such that the equality u∗ρ(a)u = π(a)
holds for every a in A, then ρ and π are unitarily equivalent (denoted by ρ ≃ π in N ). Let A+
denote the set of positive elements of A.
The following Lemma 3.1 and Lemma 3.2 are prepared for Lemma 3.3.
Lemma 3.1. Let C(X) be a unital, separable, abelian C∗-algebra with X a compact metric
space. Suppose that p is a projection in a type II1 factor (N , τ ).
If π : C(X) → N is a unital ∗-homomorphism and ρ : C(X) → pN p is a unital ∗-
homomorphism such that:
then, for every positive function h in C(X),
∀ f ∈ C(X),
τ(cid:16)R (ρ (f ))(cid:17) ≤ τ(cid:16)R (π (f ))(cid:17),
τ(cid:0)ρ (h)(cid:1) ≤ τ(cid:0)π (h)(cid:1).
(3.1)
(3.2)
Proof. By applying Theorem II.2.5 of [5], there are regular Borel measures µρ and µπ on
X, such that ρ (resp. π) extends to a weak∗-WOT continuous ∗-isomorphism ρ (resp. π) of
L∞(µρ) (resp. L∞(µπ)) onto ρ(C(X))′′ (resp. π(C(X))′′).
Let ∆ be a Borel subset of X and χ∆ be the characteristic function on ∆. Note that, for
each regular Borel measure µ on X, every µ-measurable set is a disjoint union of a Borel set
and a set of µ-measure 0. Thus we only need to concentrate on χ∆ for every Borel subset ∆ of
X instead of considering measurable subsets.
If ∆ is a non-empty open subset of X, then there exists a positive function f in C(X)+ such
that f (λ) 6= 0 for λ ∈ ∆ and f (λ) = 0 for λ ∈ X\∆. The weak∗-WOT continuity of ρ entails
that
R(ρ(f )) = WOT- limn→∞ ρ(f )
1
n = WOT- limn→∞ ρ(f
1
n ) = ρ(χ∆).
Thus, the hypothesis in (3.1) implies that the inequality
holds for each open subset ∆ of X.
τ(cid:0)ρ(χ∆)(cid:1) ≤ τ(cid:0)π(χ∆)(cid:1)
(3.3)
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
9
If ∆ is a Borel subset of X, then all the open subsets Oα's of X with ∆ ⊆ Oα, form a net with
respect to each regular Borel measure µ, i.e. µ(∆) = limα µ(Oα). Let χα be the characteristic
function on Oα. It follows that χα converges to χ∆ in the weak∗ topology. Thus, the inequality
in (3.3) holds for every Borel subset ∆ of X.
Given ǫ > 0 and a positive function h in C(X), there exist positive numbers λ1, . . . , λm and
a Borel partition ∆1, . . . , ∆m of X such that
m
It follows the inequality in (3.2). This completes the proof.
kh −
λkχ∆mk ≤ ǫ.
Xk=1
(cid:3)
A lemma from [26] is prepared as follows.
Lemma 3.2 (Lemma 2.2 of [26]). Let A be a C∗-algebra and {π, H} be a representation of
A. Then there is a unique linear mapping π of the second conjugate space A∗∗ of A onto π(A)′′
such that
(1) the diagram
A∗∗
i
A
π
#●●●●●●●●●
/ π(A)′′
π
(3.4)
is commutative, where i is the canonical imbedding of A into A∗∗.
(2) the mapping π is continuous with respect to the σ(A∗∗, A∗)-topology and the weak oper-
ator topology of π(A)′′.
By virtue of Lemma 3.1 and Lemma 3.2, we are ready for the following lemma.
Lemma 3.3. Let A be a separable C∗-subalgebra in a type II1 factor (N , τ ). Let p be a
projection in (N , τ ). Suppose that π : A → N is a unital ∗-homomorphism and ρ : A → pN p
is a unital ∗-homomorphism such that:
Let π : A∗∗ → π(A)′′ and ρ : A∗∗ → ρ(A)′′ be the weak∗-WOT continuous ∗-homomorphisms
extended by π and ρ, respectively. Then, for every projection e in A∗∗,
τ(cid:16)R (ρ (a))(cid:17) ≤ τ(cid:16)R (π (a))(cid:17),
∀ a ∈ A.
Proof. As an application of Lemma 3.2, we have the following commutative diagram:
τ (ρ (e)) ≤ τ (π (e)) .
A∗∗
ρ
{✇✇✇✇✇✇✇✇✇
π
#●●●●●●●●●
i
Aρ
where i is the canonical imbedding of A into A∗∗.
ρ(A)′′
/ π(A)′′
π
(3.5)
#
O
O
/
{
#
o
o
O
O
/
10
JUNHAO SHEN AND RUI SHI
Given a projection e in A∗∗, by virtue of [13, Theorem 1.6.5] and Kaplansky's Density
Theorem (Corollary 5.3.6 of [13]), there exists a sequence {an}n≥1 of positive operators in the
unit ball of A such that i(an) is SOT-convergent to e. Then, Lemma 7.1.14 of [14] entails that
ρ(an) = ρ◦i(an) is SOT-convergent to ρ(e) in ρ(A)′′. Likewise, π(an) = π◦i(an) is SOT-convergent
to π(e) in π(A)′′.
In terms of Lemma 3.1, we have that
Since τ is a normal mapping, it follows that the inequality
τ (ρ (an)) ≤ τ (π (an)) ,
∀ n ≥ 1.
τ (ρ (e)) ≤ τ (π (e))
holds for every projection e in A∗∗. This completes the proof.
(cid:3)
Lemma 3.3 and the following Lemma 3.4 are prepared for Lemma 3.6. Note that the following
Lemma 3.4 is a routine calculation. For completeness, we sketch its proof.
Lemma 3.4. Let (N , τ ) be a type II1 factor with tracial state τ . Suppose that A is a unital
separable C∗-subalgebra of (N , τ ), ∗-isomorphic to Mn(C), with an identity IA. Let φ and ψ be
∗-homomorphisms of A into N such that
τ (φ(a)) = τ (ψ(a)),
∀ a ∈ A.
Then, there exists a partial isometry v in N such that
φ(a) = v∗ψ(a)v,
∀a ∈ A.
(3.6)
(3.7)
Proof. Let {eij}1≤i,j≤n be a system of matrix units for A satisfying
(1) e∗
(2) eijekl = δjkeil for each i, j, k, l ∈ N;
ij = eji for each i, j ∈ N;
i=1 eii = IA.
(3) Pn
Then {φ(eij)}1≤i,j≤n (resp. {ψ(eij)}1≤i,j≤n) is a system of matrix units for φ(A) (resp. ψ(A)).
Note that if a certain ei0j0 = 0, then each eij = 0 for 1 ≤ i, j ≤ n. Thus, since τ (φ(a)) = τ (ψ(a))
for each a ∈ A, we obtain ker φ = ker ψ. This is because each element a of A can be expressed
as a matrix in terms of {eij}1≤i,j≤n.
Note that
∀ 1 ≤ i, j ≤ n.
Since N is a factor, there exists a partial isometry vi in N such that
τ (φ(eij)) = τ (ψ(eij)) < ∞,
φ(eii) = v∗
i vi
and
ψ(eii) = viv∗
i .
Let v be defined as
Then, it is routine to verify that
v :=X1≤i≤n
φ(ei1)v∗
1ψ(e1i).
(1)
(2)
v∗v = Iψ(A)
φ(eij)v = vψ(eij) for 1 ≤ i, j ≤ n.
and vv∗ = Iφ(A);
This completes the proof.
(cid:3)
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
11
Recall that a locally homogeneous C∗-algebra is a (finite) direct sum of homogeneous C∗-
algebra. A C∗-algebra is homogeneous if it is n-homogeneous for some n. A C∗-algebra A is
n-homogeneous if every irreducible representation of A is of dimension n. The reader is referred
to [3, IV.1.4.1] for the definition.
Remark 3.5. If follows from [3, IV.1.4.6] that if a C∗-algebra is n-homogeneous, then its
double dual is a type In von Neumann algebra.
In the following lemmas, we will frequently
mention type In von Neumann algebras. Thus the following facts about type In von Neumann
algebras are useful.
Let A be a type In von Neumann algebra on a Hilbert space H. Then there exists a system of
matrix units {eij}1≤i,j≤n for A. Let Rn be the von Neumann algebra generated by {eij}1≤i,j≤n.
Then Rn is ∗-isomorphic to Mn(C). Define P = R′
n ∩ A. Since A is a type In von Neumann
algebra, it follows that P is abelian and A = (P ∪ Rn)′′. Moreover, A is ∗-isomorphic to the
von Neumann tensor product P ⊗ Mn(C).
The following observation is useful in the sequel. For each element a in A and ǫ > 0, there
are projections p1, . . . , pm in P with 1P =P1≤i≤m pi and matrices a1, . . . , am in Rn such that
piaik < ǫ.
ka − X1≤i≤m
(3.8)
For Theorem 3.7, we prepare the following lemma.
Lemma 3.6. Let A be a unital, separable, locally homogeneous C∗-subalgebra in a type II1 fac-
tor (N , τ ). Let p be a projection in (N , τ ). Suppose that π : A → N is a unital ∗-homomorphism
and ρ : A → pN p is a unital ∗-homomorphism such that:
τ (R (ρ (a))) ≤ τ (R (π (a))),
∀ a ∈ A.
Let π (resp. ρ) be the weak∗-WOT continuous ∗-homomorphism of A∗∗ onto π(A)′′ (resp. ρ(A)′′)
extended by the ∗-homomorphism π (resp. ρ).
For a finite subset F of A∗∗ and ǫ > 0, there exists a finite dimensional von Neumann
subalgebra B in A∗∗ such that
(1) for each a in F , there is an element b in B satisfying
ka − bk < ǫ;
(2) there is a unital ∗-homomorphism γ : B → p⊥N p⊥ satisfying
ρB ⊕ γ ≃ πB
in N .
Moreover, if γ′ : B → p⊥N p⊥ is another unital ∗-homomorphism satisfying
ρB ⊕ γ′ ≃ πB
in N ,
then γ′ ≃ γ in p⊥N p⊥.
(3.9)
(3.10)
(3.11)
Proof. We first assume that A is n-homogeneous.
It follows from Remark 3.5 that the
double dual A∗∗ of A can be expressed as A∗∗ = (P ∪ Rn)′′ on some Hilbert space H, where
n ∩ A∗∗ is an abelian von Neumann subalgebra. Let
Rn is ∗-isomorphic to Mn(C) and P = R′
12
JUNHAO SHEN AND RUI SHI
F = {a1, . . . , ak} be a finite subset of A. Since A is isometrically imbedded into A∗∗, we can
also view ai as an element in A∗∗ for each 1 ≤ i ≤ k.
For each ǫ > 0, by (3.8) in Remark 3.5, there are finitely many projections p1, . . . , pm in P
with IP =P1≤j≤m pj and there are matrices ai1, . . . , aim in Rn for 1 ≤ i ≤ k such that
pjaijk < ǫ.
(3.12)
kai − X1≤j≤m
Note that each pj is in the center of A∗∗ and IP is the identity of A∗∗. Define a finite
dimensional von Neumann subalgebra B in A∗∗ as follows
B :=
m
Xj=1
pjRn.
(3.13)
Thus, by virtue of Lemma 3.3, there is a finite dimensional von Neumann subalgebra M of
p⊥N p⊥ in the form
such that
M :=
Mj
m
Xj=1
(3.14)
(1) for each 1 ≤ j ≤ m, Mj is ∗-isomorphic to Mn(C);
(2) the identity qj of Mj satisfies
p⊥ = X1≤j≤m
qj
and
τ (qj) = τ(cid:0)π(pj)(cid:1) − τ(cid:0)ρ(pj)(cid:1),
for 1 ≤ j ≤ m.
Since each Mj is ∗-isomorphic to Mn(C), we obtain that B is ∗-isomorphic to M. In terms of
Lemma 3.4, we can define a unital ∗-isomorphism γ of B into M satisfying γ(pj) = qj for each
1 ≤ j ≤ m, and
ρB ⊕ γ ≃ πB
Moreover, γ is unique up to unitary equivalence.
in N .
If A is a unital, separable, locally homogeneous C∗-subalgebra of N , then A can be expressed
k=1 Ak such that
as A =Pm
(1) for each 1 ≤ k ≤ m, Ak is nk-homogeneous for some nk ∈ N,
(2) the identities IAk of Ak are mutually orthogonal.
By composing the preceding arguments for each Ak, we can complete the proof.
(cid:3)
We are ready for the main theorem of this section. For a unital, separable C∗-subalgebra A
of N and two unital ∗-homomorphisms φ and ψ of A into N , recall that φ ∼a ψ in N means
the approximately unitary equivalence of φ and ψ in N , i.e., there exists a sequence of unitary
operators {un}∞
n=1 in N such that
lim
n→∞
ku∗
nφ(a)un − π(a)k = 0,
∀ a ∈ A.
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
13
Theorem 3.7. Let A be a unital, separable AH subalgebra in a type II1 factor (N , τ ). Let p be
a projection in (N , τ ). Suppose that π : A → N is a unital ∗-homomorphism and ρ : A → pN p
is a unital ∗-homomorphism such that:
Then, there exists a unital ∗-homomorphism γ : A → p⊥N p⊥ such that
τ (R (ρ (a))) ≤ τ (R (π (a))),
∀a ∈ A.
ρ ⊕ γ ∼a π
in N .
(3.15)
Proof. As in Remark 2.6, we can assume that
(3.16)
where {An}n≥1 is a monotone increasing sequence of unital, locally homogeneous C∗-algebras.
Since A is separable, let F1 ⊆ F2 ⊆ · · · be a monotone increasing sequence of finite subsets
in ∪n≥1An such that ∪i≥1Fi is k · k-dense in A, where k · k is the operator norm. By dropping to
a subsequence, we can assume that Fi ⊂ Ai for each i in N. We further require that 1A ∈ F1.
A = ∪n≥1An
,
k·k
In terms of [3, III.5.2.10], for each n ∈ N, we have that
An ⊆ An+1 ⊆ A =⇒ A∗∗
n ⊆ A∗∗
n+1 ⊆ A∗∗.
In the following, we identify A as a unital, separable C∗-subalgebra of A∗∗. Note that each A∗∗
i
is a direct sum of finitely many type Im von Neumann algeba for m less than a certain n.
As an application of Lemma 3.6, there is a finite dimensional von Neumann subalgebra B1
1 corresponding to F1 such that for each a in F1, there is an a1 in B1 satisfying
of A∗∗
Applying Lemma 3.6 once more for F2 and the system of matrix units of B1, there exists a finite
dimensional von Neumann subalgebra B2 of A∗∗
2 such that
ka − a1k <
1
22
.
(1) B2 ⊇ B1;
(2) for each a in F2, there is an a2 in B2 satisfying
ka − a2k <
1
23
.
By induction, there is a finite dimensional von Neumann subalgebra Bi of A∗∗
each Fi such that
i corresponding to
(1) Bi ⊇ Bi−1 for each i ≥ 2;
(2) for each a in Fi, there is an ai in Bi satisfying
1
ka − aik <
.
2i+1
Moreover, there is a unital ∗-homomorphism φi : Bi → p⊥N p⊥ such that
ρBi ⊕ φi ≃ πBi
in N .
Note that Bi+1 ⊇ Bi implies that
ρBi ⊕ φi+1Bi ≃ ρBi ⊕ φi ≃ πBi
in N .
(3.17)
(3.18)
14
JUNHAO SHEN AND RUI SHI
Thus we have φi+1Bi ≃ φi
in p⊥N p⊥.
Define γ1 := φ1. Let u2 be the unitary operator in p⊥N p⊥ such that u∗
2(φ2B1)u2 = γ1 and
define
γ2(·) := u∗
2φ2(·)u2.
Likewise, let ui+1 be the unitary operator in p⊥N p⊥ such that u∗
i+1(φi+1Bi)ui+1 = γi and define
(3.19)
γi+1(·) := u∗
With respect to the choice of the family {Bi}∞
we construct a sequence of ∗-homomorphisms {γi}i≥1 such that the equality
i+1φi+1(·)ui+1.
i=1 of finite dimensional von Neumann algebras,
γi+k(b) = γi(b)
(3.20)
holds for every b in Bi and each i ≥ 1, k ≥ 1.
By virtue of (3.17), for every a in Fi, there is a sequence {ak : ak ∈ Bk}k≥1 such that
ak = 0
for k < i,
and
ka − akk <
1
2k+1
for k ≥ i.
(3.21)
It follows that {ak}∞
that {γk(ak)}∞
and each p ≥ 1, (3.20) and (3.21) imply that
k=1 is a Cauchy sequence in the operator norm topology. Moreover, we assert
k=1 is a Cauchy sequence in the operator norm topology. Notice that, for k ≥ i
kγk+p(ak+p) − γk(ak)k = kγk+p(ak+p) − γk+p(ak)k ≤ kak+p − akk <
1
2k
.
This guarantees that {γk(ak)}k≥1 is also a Cauchy sequence in the operator norm topology.
Note that, for each fixed a in Fi, if there is another sequence {a′
k : a′
k ∈ Bk} satisfying
a′
k = 0
for k < i,
and
ka − a′
kk <
1
2k+1
for k ≥ i,
then both {a′
thermore, the limit limk→∞ kak − a′
k=1 and {γk(a′
k)}∞
k}∞
k=1 are Cauchy sequences in the operator norm topology. Fur-
kk = 0 entails that the equality
limk→∞ γk(a′
k) = limk→∞ γk(ak)
holds in the operator norm topology. Since kγkk ≤ 1 for each k ∈ N, the mapping
γ : ∪i≥1Fi → p⊥N p⊥,
γ(a) := limk→∞ γk(ak)
extends to a well-defined unital ∗-homomorphism of A into p⊥N p⊥.
Note that, for each i ≥ 1, (3.18) and (3.19) entail that there is a unitary operator vi in N
i (ρi ⊕ γi)vi = πi, where ρi (resp. πi) is the restriction of ρ (resp. π) on Bi. Thus, for
such that v∗
each a ∈ Fi, it follows that
kπ(a) − v∗
i (ρ(a) ⊕ γ(a))vik ≤ kπ(a) − πi(ai)k + kρi(ai) ⊕ γi(ai) − ρ(a) ⊕ γ(a)k <
Since ∪i≥1Fi is k · k-dense in A, we obtain that
limi→∞ kπ(a) − v∗
i (ρ(a) ⊕ γ(a))vik = 0,
∀ a ∈ A.
This completes the proof.
1
2i
.
(cid:3)
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
15
4. Representations of AH algebras to semifinite, properly infinite factors
Let (M, τ ) be a countably decomposable von Neumann factor with a faithful, normal, semifi-
nite, tracial weight τ . Recall that F (M, τ ) is the set of all (M, τ )-finite-rank operators in M
and K(M, τ ) is the k · k-norm closure of F (M, τ ), where k · k denotes the operator norm. See
Definition 2.1 in Section 2 for details.
Let A be a separable AH subalgebra of M with an identity IA. Let φ and ψ be unital
∗-homomorphisms of A into M. The main goal of this section is to prove the equivalence of the
following statements:
φ ∼a ψ in M, i.e., φ and ψ are approximately unitarily equivalent in M;
φ ∼A ψ mod K(M, τ ) (see Definition 2.5).
(1)
(2)
Note that the analogue in B(H) is proved in three steps. First, a separable C∗-algebra is cut
into two parts in terms of all compact operators in the C∗-algebra, i.e., the part containing all
compact operators in the C∗-algebra and the part containing no compact operators. Then, the
equivalence of (1) and (2) is proved with respect to the two parts, respectively.
The proof with respect to the 'non-compact' part of a separable C∗-algebra in B(H) is due to
Voiculescu [28], sometimes called the absorption theorem. His proof works for every separable
C∗-algebra in B(H). Recently, the authors in [16] extends the preceding absorption theorem for
separable nuclear C∗-algebras in (M, τ ).
Notice that the proof associated with the 'compact' part of a separable C∗-algebra in B(H)
is based on minimal projections. This leads to the first obstruction that (M, τ ) may contain
no minimal projections. Meanwhile, for a separable AH algebra in (M, τ ), there might not be
sufficiently many projections in the AH algebra, generally. This is the second obstruction. Thus,
it is necessary to develop new techniques to overcome these two obstructions. In this section,
Lemma 4.4 is devoted to cut each C∗-algebra A according to A ∩ K(M, τ ). By virtue of a series
of lemmas, the equivalence of (1) and (2) mentioned above with respect to the 'compact' part is
proved in Theorem 4.9. Combining with Theorem 4.10 ([16, Theorem 5.3.1]), the equivalence
of (1) and (2) is proved in Theorem 4.11.
Lemma 4.1. For i = 1, 2, let Mi be a von Neumann algebra with a faithful, normal, semifi-
nite, tracial weight τi. Let F (Mi, τi) be the set of all (Mi, τi)-finite-rank operators in (Mi, τi).
Assume that Ai is a ∗-subalgebra of F (Mi, τi) such that Ai is weak∗-dense in Mi, for i = 1, 2.
If ρ : A1 → A2 is a ∗-isomorphism such that
τ2(ρ(x)) = τ1(x),
∀ x ∈ A1,
(4.1)
then ρ extends uniquely to a normal ∗-isomorphism ρ′ : M1 → M2 satisfying that
τ2(ρ′(x)) = τ1(x),
∀ 0 < x ∈ M1.
Proof. Since τi is a faithful, normal, semifinite tracial weight on Mi for i = 1, 2, then
(a, b) := τi(b∗a),
(4.2)
defines a definite inner product on Ai. Let Hi be the completion of Ai relative to the norm
associated with the inner product defined in (4.2). Denote by L2(Mi, τi) the completion of
16
JUNHAO SHEN AND RUI SHI
{x : x ∈ Mi, τi(x∗x) < ∞} relative to the norm associated with the inner product in (4.2). From
the fact that each Ai is weak∗-dense in Mi, it follows that Hi = L2(Mi, τi). By applying Theorem
7.5.3 of [14], the faithful, normal tracial weight τi induces a faithful, normal, representation πi
of Mi on Hi for i = 1, 2.
Let {aλ}λ∈Λ be a bounded net in A1 such that aλ converges to a ∈ M1 in the weak∗ topology.
Note that, for each b in A1, the equality
(π1(aλ)b, b) = τ1(b∗aλb) = τ2(ρ(b∗aλb)) = (π2(ρ(aλ))ρ(b), ρ(b))
entails that π2(ρ(aλ)) converges to an operator x in π2(M2) in the weak operator topology. Note
that π2 is a normal ∗-isomorphism between von Neumann algebras M2 and π2(M2). It follows
2 (x) in the weak operator topology. For a, the weak∗ limit of aλ, in
that ρ(aλ) converges to π−1
M1, define ρ′(a) := π−1
2 (x). It is easily verified that ρ′ is well-defined. In this way, ρ extends
uniquely to a normal ∗-isomorphism ρ′ : M1 → M2. Combining with the fact that each τi is
normal, we can further conclude that
τ2(ρ′(x)) = τ1(x),
∀ 0 < x ∈ M1.
This completes the proof.
(cid:3)
Lemma 4.2. Let M be a von Neumann algebra with a faithful, normal, semifinite, tracial
weight τ . Suppose that A is a ∗-subalgebra of F (M, τ ) and A is a linear span of A+.
If ρ is a ∗-isomorphism of A into M such that ρ and the identity mapping id of A are
approximately equivalent in M, written as
id ∼a ρ,
in M,
then ρ extends uniquely to a normal ∗-isomorphism ρ′ of the WOT-closure of A into M such
that
τ (ρ′(a)) = τ (a),
for each postive operator a in the WOT-closure of A.
Proof. We will apply Lemma 4.1 to extend ρ uniquely to a von Neumann algebra isomor-
It is
phism of the WOT-closure of A to the WOT-closure of ρ(A) with the desired property.
sufficient to prove the following equality:
Recall that PF (M, τ ) is the set of projections p in M with τ (p) < ∞. Let a > 0 be a
(M, τ )-finite-rank operator in A. Note that
τ (a) = τ (ρ(a)),
∀ a ∈ A+.
(4.3)
τ (a) = sup{τ (ap) : p ∈ PF (M, τ )}.
We claim that τ (ρ(a)p) ≤ τ (a) for each finite trace projection p in M.
Since ρ and the identity mapping id are approximately equivalent in M, there exists a
sequence of unitary operators {uk}k≥1 in M such that
Let p be a finite trace projection in M. In terms of the Holder inequality, we have
lim
k→∞
kρ(a) − u∗
kaukk = 0.
τ(cid:0)(ρ(a) − u∗
kauk)p(cid:1) ≤ kρ(a) − u∗
kaukkkpk1 → 0, as k → ∞.
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
17
This implies that
τ (ρ(a)p) = lim
k→∞
This completes the proof of the claim.
τ ((u∗
kauk)p) ≤ ku∗
kaukk1kpk = τ (a).
Since
it follows that
Similarly, we have
Thus, we have
τ (ρ(a)) = sup{τ (ρ(a)p) : p ∈ PF (M, τ )},
τ (ρ(a)) ≤ τ (a).
τ (a) ≤ τ (ρ(a)).
τ (a) = τ (ρ(a)).
This completes the proof of (4.3). Thus, by virtue of Lemma 4.1, ρ can be extended uniquely
to a von Neumann algebra isomorphism ρ′ of the WOT-closure of A to the WOT-closure of ρ(A)
with the desired property.
(cid:3)
Remark 4.3. Let M be a von Neumann algebra with a faithful, normal, semifinite, tracial
weight τ . Let A be a (separable) AH subalgebra of (M, τ ). It is convenient to assume that there
exists an increasing sequence {An}n≥1 of locally homogeneous C∗-algebras, as in Remark 2.6,
such that
By applying Lemma 3.4.1 of [5], we have
A = ∪n≥1An
k·k
.
(4.4)
(4.5)
A ∩ K(M, τ ) = ∪n≥1(An ∩ K(M, τ ))
k·k
.
Let M be a von Neumann algebra with a faithful, normal, semifinite, tracial weight τ . For
a separable, unital C∗-subalgebra A in (M, τ ), the reader is referred to [11, Lemma 3.1] and
[11, Lemma 3.2] for several useful properties for operators in K(M, τ ). By a routine continuous
function calculus and [11, Lemma 3.1], we construct a sequence of (M, τ )-finite-rank operators
k · k-norm dense in A+ ∩ K(M, τ ). Define
In the following lemma, we prove that the projection pK(A,τ ) reduces A.
pK(A,τ ) := _x∈A∩K(M,τ )
R(x).
(4.6)
Lemma 4.4. Let (M, τ ) be a von Neumann algebra with a faithful, normal, semifinite, tracial
weight τ . Suppose that A is a separable C∗-subalgebra of M. Let {xn}∞
n=1 be a sequence of
positive, (M, τ )-finite-rank operators in the unit ball of A+ ∩ F (M, τ ) such that {xn}∞
n=1 is
k · k-dense in the unit ball of A+ ∩ K(M, τ ), where k · k is the operator norm. Then the following
statements are true:
(1) pK(A,τ ) = ∨n≥1R(xn), where pK(A,τ ) is defined as in (4.6);
(2) pK(A,τ )x = xpK(A,τ ), ∀ x ∈ A.
18
JUNHAO SHEN AND RUI SHI
Proof. Assume that A ∩ K(M, τ ) 6= 0 and the von Neumann algebra M acts on a Hilbert
space H.
Let p be the union of the range projections R(xn) of all these positive, finite-rank operators
xn's. For each element x in the unit ball of A∩K(M, τ ), let x = x∗v be the polar decomposition
of x (see Theorem 6.1.2 of [14]). For every ǫ > 0, there is a positive, finite-rank operator xn
such that kx∗ − xnk ≤ ǫ. Thus, for each unit vector ξ in H, it follows that
kxξ − xnvξk ≤ kx∗ − xnk ≤ ǫ.
Since pxnvξ = xnvξ, it follows, eventually, that px = x holds for every x in A ∩ K(M, τ ). Thus,
we have that
p = pK(A,τ ).
In the following, assume contrarily that there exists an operator a in A such that pap⊥ 6= 0.
Then there exists a positive, (M, τ )-finite-rank operator a1 such that R(a1)ap⊥ 6= 0. Since the
restriction of each bounded linear positive operator on the closure of its range is injective, the
equality
ker(a1) = ker(a
1
2
1 )
entails that
1
2
R(a
1 )ap⊥ = R(a1)ap⊥ 6= 0
and
1
2
1 ap⊥ = a
a
1 R(a
1 )ap⊥ 6= 0.
1
2
1
2
Thus p⊥a∗a1ap⊥ 6= 0 implies pa∗a1a 6= a∗a1a. Note that the inequality τ (R(a∗a1a)) < ∞
ensures that a∗a1a is a positive, (M, τ )-finite-rank operator. Then the fact that pa∗a1a 6= a∗a1a
contradicts the definition of p. It follows that p reduces A. This completes the proof.
(cid:3)
Let M be a von Neumann algebra and A a C∗-subalgebra of M. Recall that by W ∗(A) we
denote the WOT-closure of A, which is also the von Neumann algebra generated by A.
As mentioned in Remark 4.3, a separable AH algebra is a inductive limit of locally homoge-
neous C∗-algebras. The following three lemmas are developed with respect to locally homoge-
neous C∗-algebras in (M, τ ), which are prepared for Theorem 4.9.
Lemma 4.5. Let M be a von Neumann algebra with a faithful, normal, semifinite, tracial
weight τ . Suppose that A is a (separable) locally homogeneous C∗-subalgebra of M.
Let a be a non-zero positive finite M-rank operator in A ∩ F (M, τ ). Then there exists a
central projection e of W ∗(A), the WOT-closure of A in M, such that
e ∈ F (M, τ );
(1)
(2) R(a) ≤ e;
(3) W ∗(A)e ⊆ W ∗(A ∩ F (M, τ )).
Proof. By Definition IV.1.4.1 of [3], a C∗-algebra A is locally homogeneous, if it is a
finite direct sum of homogeneous C∗-algebras. For the sake of simplicity, we assume A to be
n-homogeneous.
Claim 4.5.1. If A is n-homogeneous, then W ∗(A) is a type In von Neumann algebra.
Let id : A → A be the identity mapping of A. In terms of Proposition IV.1.4.6 of [3], the
double dual A∗∗ of A is a type In von Neumann algebra. By Proposition 1.21.13 of [25], the
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
19
identity mapping id extends uniquely to a σ(A∗∗, A∗)-WOT continuous ∗-homomorphism id of
A∗∗ onto W ∗(A). It follows that W ∗(A) is a type In von Neumann algebra. This completes the
proof of this claim.
(End of the proof of Lemma 4.5.) Without loss of generality, we assume that {eij}n
system of matrix units for W ∗(A). Let
i,j=1 is a
n
e :=
R(eija).
(4.7)
_i,j=1
We can verify directly that eije = eeije for all 1 ≤ i, j ≤ n. Thus, e is a central projection of
i=1 eiia,
we have that R(a) ≤ e. From the fact that each eija belongs to W ∗(A ∩ F (M, τ )), we further
conclude that W ∗(A)e ⊆ W ∗(A ∩ F (M, τ )).
(cid:3)
W ∗(A). Since a ∈ F (M, τ ), the definition of e in (4.7) entails e ∈ F (M, τ ). As a =Pn
Lemma 4.6. Let M be a von Neumann factor with a faithful, normal, semifinite, tracial
weight τ . Suppose that A is a C∗-subalgebra of (M, τ ), ∗-isomorphic to Mn(C). Let φ and ψ be
∗-homomorphisms of A into M such that
Then, there exists a partial isometry v in M such that
τ(cid:0)φ(a)(cid:1) = τ(cid:0)ψ(a)(cid:1),
∀ a ∈ A+.
φ(a) = v∗ψ(a)v,
∀ a ∈ A.
(4.8)
(4.9)
Moreover, if τ(cid:0)φ(a)(cid:1) = τ(cid:0)ψ(a)(cid:1) < ∞, for each a ∈ A+, then there is a unitary operator u in
M such that φ(a) = u∗ψ(a)u, for every a ∈ A.
Proof. Let {eij}1≤i,j≤n be a system of matrix units for A. Then {φ(eij)}1≤i,j≤n is a system
of matrix units for φ(A). So is {ψ(eij)}1≤i,j≤n for ψ(A).
Note that
Since M is a factor, there exists a partial isometry vi in M such that
τ (φ(eii)) = τ (ψ(eii)),
∀ 1 ≤ i ≤ n.
φ(eii) = v∗
i vi
and
ψ(eii) = viv∗
i .
Let v be defined as
Then, it is routine to verify that
v :=X1≤i≤n
φ(ei1)v∗
1ψ(e1i).
(1)
(2)
v∗v = ψ(IA)
φ(eij)v = vψ(eij) for all 1 ≤ i, j ≤ n.
and vv∗ = φ(IA);
This completes the proof of (4.9).
Furthermore, if τ(cid:0)φ(a)(cid:1) = τ(cid:0)ψ(a)(cid:1) < ∞, for each a ∈ A+, then there exists a partial
isometry w in M such that w∗w = I − ψ(IA) and ww∗ = I − φ(IA). Define u = v + w. It follows
that u is a unitary operator in M as desired.
(cid:3)
20
JUNHAO SHEN AND RUI SHI
Lemma 4.7. Let M be a von Neumann factor with a faithful, normal, semifinite, tracial
weight τ . Suppose that A is a (separable) locally homogeneous C∗-subalgebra of (M, τ ). Let F
be a finite subset of A and e be a projection in A∗∗ satisfying ae = ea for each a in F , where A
is identified with its imbedded image in A∗∗.
For each ǫ > 0, there is a finite dimensional von Neumann subalgebra B of A∗∗ with e being
the identity such that, for each a in F , there exists an element b in B satisfying
kae − bk < ǫ.
Proof. By applying IV.1.4.6 of [3], A C∗-algebra A is n-homogeneous if and only if the
double dual A∗∗ of A is a type In von Neumann algebra. By taking a finite partition in the
center of A∗∗ fine enough, we can complete the proof.
(cid:3)
Definition 4.8. Let M be a von Neumann algebra with a faithful, normal, semifinite, tracial
weight τ . Suppose that A is a separable C∗-subalgebra of M.
By virtue of Lemma 4.4, the projection pK(A,τ ) reduces A. Define
id0(a) := apK(A,τ )
and
ide(a) := ap⊥
K(A,τ ) ∀ a ∈ A.
(4.10)
Then id0 and ide are well-defined ∗-homomorphisms of A into ApK(A,τ ) and Ap⊥
tively.
K(A,τ ), respec-
Let ρ be a unital ∗-isomorphism of A into M. Define
ρ0(a) := id0(ρ(a))
and
ρe(a) := ide(ρ(a)) ∀ a ∈ A.
(4.11)
Then ρ0 and ρe are well-defined ∗-homomorphisms of A into ρ(A)pK(ρ(A),τ ) and ρ(A)p⊥
respectively.
K(ρ(A),τ ),
Theorem 4.9. Let M be a countably decomposable, properly infinite, semifinite von Neu-
mann factor with a faithful normal semifinite tracial weight τ . Suppose that A is a separable
AH subalgebra of (M, τ ). Let id and ρ be unital ∗-homomorphisms of A into M such that id
and ρ are approximately unitarily equivalent.
Then id0 and ρ0, as in Definition 4.8, are strongly-approximately-unitarily-equivalent over
A, as in Definition 2.5, i.e.
id0 ∼A ρ0 mod K(M, τ ).
(4.12)
Proof. Since A is AH, as in Remark 4.3, it is convenient to assume that there is a monotone
increasing sequence of locally homogeneous C∗-subalgebras {An}n≥1 of (M, τ ) such that
A = ∪n≥1An
k·k
.
We assume that A ∩ K(M, τ ) 6= 0.
Let {xn}n≥1 be a sequence of positive, finite M-rank operators in the unit ball of ∪n≥1An,
which is k · k-norm dense in the unit ball of A+ ∩ K(M, τ ). Define two projections p and q as
follows
p := ∨n≥1R(xn)
and
q := ∨n≥1R(ρ(xn)).
(4.13)
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
21
Then, by virtue of Lemma 4.2 and Lemma 4.4, we obtain that
p = pK(A,τ )
and
q = pK(ρ(A),τ ).
Let F1 ⊆ F2 ⊆ · · · be a monotone increasing sequence of finite subsets of the unit ball of
∪n≥1An such that ∪n≥1Fn is k · k-norm dense in the unit ball of A.
In the following, we construct at most countably many, mutually orthogonal, finite M-rank
projections in W ∗(A) with nice properties.
Choose n1 ≥ 1 such that F1 ∪ {x1} ⊂ An1. Since An1 is locally homogeneous, by virtue of
Lemma 4.5, there is a central projection p1 in W ∗(An1) such that
(1) p1 belongs to F (M, τ );
(2) R(x1) ≤ p1 in M;
(3) W ∗(An1)p1 ⊆ W ∗(An1 ∩ F (M, τ )).
Define y1 := x1. If p1 = p, then we complete the construction. Otherwise, suppose k ≥ 2
and we obtain y1, . . . , yk in {xn}n≥1 and p1, . . . , pk in F (M, τ ) satisfying
(1) yi+1 is the first element after yi in {xn}n≥1 such that (p − pi)yi+1 6= 0, for 1 ≤ i ≤ k − 1;
(2) Fi+1 ∪ {y1, . . . , yi+1} ⊂ Ani+1, for all 1 ≤ i ≤ k − 1;
(3) Fi+1 ∪ {y1, . . . , yi+1, p1, . . . , pi} ⊂ W ∗(Ani+1) and ni ≤ ni+1, for all 1 ≤ i ≤ k − 1;
(4) pi+1 is a central projection in W ∗(Ani+1) such that
pi ∨ R(yi+1) ≤ pi+1,
1 ≤ i ≤ k − 1.
If pk = p, then we complete the construction. Otherwise, let yk+1 be the first element after yk
in {xn}n≥1 such that (p − pk)yk+1 6= 0. Choose nk+1 ≥ nk such that
Fk+1 ∪ {y1, . . . , yk+1} ⊂ Ank+1.
Note that the projections p1, . . . , pk are also in W ∗(Ank+1). In terms of Lemma 4.5, there is a
central projection pk+1 of W ∗(Ank+1) such that:
(1) pk+1 belongs to F (M, τ );
(2) pk ∨ R(yk+1) ≤ pk+1 in W ∗(Ank+1);
(3) W ∗(Ank+1)pk+1 ⊆ W ∗(Ank+1 ∩ F (M, τ )).
Define e1 := p1 and ek+1 := pk+1 − pk for each k ≥ 1. Let F0 := F1. It follows that aei = eia for
each a in Fj and i = j + 1, . . . , k + 1, where j ≥ 0.
Recursively, we obtain a sequence of at most countably many, mutually orthogonal, (M, τ )-
finite-rank projections {ei}1≤i≤N in F (M, τ ) such that
where N ∈ N ∪ {∞}.
SOT-X1≤i≤N
ei = p,
By applying Lemma 4.2, the unital ∗-homomorphism ρ extends uniquely to a normal ∗-
isomorphism ρ′ of the WOT-closure of A ∩ F (M, τ ) into M such that
τ (ρ′(a)) = τ (a),
∀ a ∈ W ∗(cid:0)A ∩ F (M, τ )(cid:1)+
.
22
JUNHAO SHEN AND RUI SHI
Moreover, by the preceding arguments and Lemma 4.4, we have
and, for each projection e in W ∗(Ani)pi, we have
SOT-X1≤i≤N
ρ′(ei) = q.
τ (e) = τ (ρ′(e)) < ∞.
Fix ǫ > 0. For each i ≥ 1, in terms of Lemma 4.7, there exists a finite dimensional von
Neumann algebra Bi containing ei as its identity, in W ∗(Ani)pi, such that, for each a ∈ Fi−1,
there is an operator ai ∈ Bi satisfying
kaei − aik <
ǫ
2i+1
.
Then, by virtue of Lemma 4.6, we obtain a partial isometry ui in M, for 1 ≤ i ≤ N, such that
ρ′(ei) = uiu∗
i ,
It follows that, for each a ∈ Fi−1 and i ≥ 1,
ei = u∗
i ui,
and
uibu∗
i = ρ′(b),
∀ b ∈ Bi.
kui(aei)u∗
i − ρ′(aei)k ≤ kui(aei − ai)u∗
i k + kρ′(ai − aei)k <
Define u =P1≤i≤N ui in M. It follows that
u∗u = p
and
q = uu∗.
ǫ
2i
.
(4.14)
(4.15)
Note that for each a in ∪i≥1F ,
id0(a) = SOT-X1≤i≤N
By (4.14) and (4.15), we have that:
aei
and
ρ0(a) = SOT-X1≤i≤N
ρ′(aei).
(1) for every a in F0,
kuid0(a)u∗ − ρ0(a)k = k(X1≤i≤N
=kX1≤i,k,l≤N
≤X1≤i≤N
ui)id0(a)(X1≤i≤N
k − ρ0(a)k = kX1≤i≤N(cid:16)uieiaeiu∗
i − ρ′(aei)k < ǫ.
kuieiaeiu∗
uleiaeiu∗
(2) for every a in ∪i≥1Fi, a similar computation implies that
ui)∗ − ρ0(a)k;
i − ρ′(aei)(cid:17)k
kuid0(a)u∗ − ρ0(a)k < ∞ and
uid0(a)u∗ − ρ0(a) ∈ K(M, τ ).
For each j ≥ 1, define {Ei = Fi+j−1}i≥1 and E0 := E1. We can iterate the preceding arguments
to construct a partial isometry vj in M with respect to {Ei}i≥0 such that
(1) for every a in ∪i≥1Fi and j ≥ 1,
kvjid0(a)v∗
(2) for each a in Fj,
j − ρ0(a)k < ∞ and
vjid0(a)v∗
j − ρ0(a) ∈ K(M, τ );
kvjid0(a)v∗
j − ρ0(a)k <
1
2j
.
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
23
Note that ∪i≥1Fi is k · k-norm dense in the unit ball of A. Thus, for each a in A, we obtain that
limj→∞ kv∗
j id0(a)vj − ρ0(a)k = 0.
This completes the proof of (4.12).
(cid:3)
We cite Theorem 5.3.1 of [16] as another important tool. Note that, in the remainder, the
symbol "∼A" follows from Definition 2.5.
Theorem 4.10. Let M be a countably decomposable, properly infinite, semifinite factor with
a faithful, normal, semifinite, tracial weight τ . Let K(M, τ ) be the set of compact operators
in (M, τ ). Suppose that A is a separable nuclear C ∗-subalgebra of M with an identity IA. If
ρ : A → M is a ∗-homomorphism satisfying ρ(A ∩ K(M, τ )) = 0, then
We are ready for the main theorem.
idA ∼A idA ⊕ ρ
mod K(M, τ ).
Theorem 4.11. Let M be a countably decomposable, infinite, semifinite factor with a faithful
normal semifinite tracial weight τ . Suppose that A is a separable AH subalgebra of M with an
identity IA.
If φ and ψ are unital ∗-homomorphisms of A into M, then the following statements are
equivalent:
(i) φ ∼a ψ in M;
(ii) φ ∼A ψ mod K(M, τ ).
Proof. Note that the implication (ii) ⇒ (i) is easy by Definition 2.5. Thus, we only need
to prove the implication (i) ⇒ (ii).
The assumption φ ∼a ψ in M entails that φ and ψ have the same kernel. It follows that
the mapping
ρ : φ(A) → ψ(A),
defined by
ρ(b) := ψ(φ−1(b)), ∀ b ∈ φ(A)
is a well-defined ∗-isomorphism of φ(A) onto ψ(A). Moreover, the following are equivalent:
φ ∼A ψ mod K(M, τ );
idφ(A) ∼φ(A) ρ mod K(M, τ ).
(1)
(2)
In terms of Lemma 4.2, the restriction of ρ on φ(A) ∩ F (M, τ ) extends uniquely to a
normal ∗-isomorphism of the WOT-closure of φ(A) ∩ F (M, τ ) into M. Furthermore, Lemma 4.2
and Lemma 4.4 guarantee that there exists a sequence {xn}n≥1 of positive, (M, τ )-finite-rank
operators in the unit ball of φ(A)+ ∩ F (M, τ ) such that the projections
p := ∨n≥1R(xn)
and
q := ∨n≥1R(ρ(xn))
(4.16)
reduce φ(A) and ψ(A), respectively. Moreover, we have that
Thus, the identity mapping id on φ(A) can be expressed in the form
px = x,
qρ(x) = ρ(x),
∀ x ∈ φ(A) ∩ K(M, τ ).
id = id0 ⊕ ide,
(4.17)
24
JUNHAO SHEN AND RUI SHI
where id0 is the compression of id(·)p on ran p, and ide is the compression of id(·)p⊥ on ran p⊥.
We also write that
id0(φ(A)) = φ0(A)
and
ide(φ(A)) = φe(A).
It follows that ide(φ(A) ∩ K(M, τ )) = 0.
Likewise, the ∗-isomorphism ρ of φ(A) can be expressed in the form
ρ = ρ0 ⊕ ρe,
(4.18)
where ρ0(A) = ρ(A)qranq and ρe(A) = ρ(A)q⊥ranq⊥ for every a in φ(A). We also write that
ρ0(φ(A)) = ψ0(A)
and
ρe(φ(A)) = ψe(A).
It follows that ρe(φ(A) ∩ K(M, τ )) = 0.
By virtue of Theorem 4.9, there exists a partial isometry w in M such that
p = w∗w and
q = ww∗.
It is worth noting that, in general, many operators in φ0(A) don't belong to φ(A)∩K(M, τ ).
This is the motivation to develop Theorem 4.9.
By virtue of (4.4), there exists a monotone increasing sequence F1 ⊆ F2 ⊆ · · · of finite
subsets of the unit ball of ∪k≥1Ak such that ∪k≥1Fk is k · k-norm dense in the unit ball of A.
Likewise, the union ∪k≥1φ(Fk) (resp. ∪k≥1ψ(Fk)) is k · k-norm dense in the unit ball of φ(A)
(resp. ψ(A)). Similarly, ∪k≥1φ0(Fk) (resp. ∪k≥1ψ0(Fk)) is k · k-norm dense in the unit ball of
φ0(A) (resp. ψ0(A)).
By applying Theorem 4.9, for every k ≥ 1, there exists a partial isometry vk in (M, τ ) such
that the inequality
kvkφ0(a)v∗
k − ψ0(a)k <
1
2k
holds for every a in Fk.
Furthermore, for every a in A, we have that vkφ0(a)v∗
k − ψ0(a) belongs to the ideal K(M, τ ).
Therefore, there exists a sequence {vk}k≥1 of partial isometries in M such that
(1) limk→∞ kvkφ0(a)v∗
(2) vkφ0(a)v∗
k − ψ0(a)k = 0, for every a in A;
k − ψ0(a) belongs to K(M, τ ) for every a in A and k ≥ 1.
Notice that
ide(φ(A) ∩ K(M, τ )) = ρe(φ(A) ∩ K(M, τ )) = 0.
Thus, by applying Theorem 4.10, Theorem 4.9, and the decompositions in (4.17) and (4.18), it
follows that
φ = (id0 ◦ φ) ⊕ (ide ◦ φ) ∼A (id0 ◦ φ) ⊕ (ide ◦ φ) ⊕ (ρe ◦ φ)
mod K(M, τ )
= φ0 ⊕ φe ⊕ ψe
∼A ψ0 ⊕ ψe ⊕ φe
= (ρ0 ◦ φ) ⊕ (ρe ◦ φ) ⊕ (ide ◦ φ)
= (ρ ◦ φ) ⊕ (ide ◦ φ) ∼A (ρ ◦ φ) = ψ
mod K(M, τ )
mod K(M, τ )
This completes the proof.
(cid:3)
APPROXIMATE EQUIVALENCE OF REPRESENTATIONS OF AH ALGEBRAS
25
References
[1] William Arveson. Notes on extensions of C∗-algebras. Duke Math. J. 44 (1977), no. 2, 329 -- 355.
[2] David Berg. An extension of the Weyl-von Neumann theorem to normal operators. Trans. Amer. Math. Soc.
160 (1971), 365 -- 371.
[3] Bruce Blackadar. Operator algebras. Theory of C∗-algebras and von Neumann algebras. Encyclopaedia
of Mathematical Sciences, 122. Operator Algebras and Non-commutative Geometry, III. Springer-Verlag,
Berlin, 2006.
[4] Alin Ciuperca, Thierry Giordano, Ping Wong Ng and Zhuang Niu. Amenability and uniqueness. Adv. Math.
240 (2013), 325 -- 345.
[5] Kenneth Davidson. C∗-algebras by example. Fields Institute Monographs, 6. American Mathematical Society,
Providence, RI, 1996.
[6] Kenneth Davidson. Normal operators are diagonal plus Hilbert-Schmidt. J. Operator Theory 20 (1988), no.
2, 241 -- 249.
[7] Huiru Ding and Don Hadwin. Approximate equivalence in von Neumann algebras. Sci. China Ser. A 48
(2005), no. 2, 239 -- 247.
[8] George Arthur Elliott, Guihua Gong, Liangqing Li. On the classification of simple inductive limit C∗-
algebras. II. The isomorphism theorem. Invent. Math. 168 (2007), no. 2, 249 -- 320.
[9] Guihua Gong, Chunlan Jiang, Liangqing Li, Cornel Pasnicu. A reduction theorem for AH algebras with the
ideal property. Int. Math. Res. Not. (2018), no. 24, 7606 -- 7641.
[10] Donald Hadwin. Nonseparable approximate equivalence. Trans. Amer. Math. Soc. 266 (1981), no. 1, 203 --
231.
[11] Donald Hadwin and Rui Shi. A note on the Voiculescu's theorem for normal operators in semifinite von
Neumann algebras. Oper. Matrices, 12 (2018), no. 4, 1129 -- 1144.
[12] Paul Halmos. Ten problems in Hilbert space. Bull. Amer. Math. Soc. 76 (1970), 887 -- 933.
[13] Richard Kadison and John Ringrose. Fundamentals of the theory of operator algebras. Vol. I. Elementary
theory. Reprint of the 1983 original. Graduate Studies in Mathematics, 15. American Mathematical Society,
Providence, RI, 1997.
[14] Richard Kadison and John Ringrose. Fundamentals of the theory of operator algebras. Vol. II. Advanced
theory. Corrected reprint of the 1986 original. Graduate Studies in Mathematics, 16. American Mathematical
Society, Providence, RI, 1997.
[15] Victor Kaftal. On the theory of compact operators in von Neumann algebras. II. Pacific J. Math. 79 (1978),
no. 1, 129 -- 137.
[16] Qihui Li, Junhao Shen, Rui Shi. A generalization of the Voiculescu theorem for normal operators in semifinite
von Neumann algebras. arXiv:1706.09522 [math.OA].
[17] Qihui Li, Junhao Shen, Rui Shi, Liguang Wang. Perturbations of self-adjoint operators in semifinite von
Neumann algebras: Kato-Rosenblum theorem. J. Funct. Anal. 275 (2018), no. 2, 259 -- 287.
[18] Huaxin Lin, Homomorphisms from AH-algebras. J. Topol. Anal. 9 (2017), no. 1, 67 -- 125.
[19] Huaxin Lin, Locally AH algebras. Mem. Amer. Math. Soc. 235 (2015), no. 1107.
[20] Huaxin Lin, The range of approximate unitary equivalence classes of homomorphisms from AH-algebras.
Math. Z. 263 (2009), no. 4, 903 -- 922.
[21] Francis Joseph Murray and John Von Neumann. On rings of operators. Ann. of Math. (2) 37 (1936), no. 1,
116 -- 229.
[22] John von Neumann. Charakterisierung des Spektrums eines Integraloperators. Actualits Sci. Indust. 229,
Hermann, Paris, 1935.
[23] Zhuang Niu, Mean dimension and AH-algebras with diagonal maps. J. Funct. Anal. 266 (2014), no. 8,
4938 -- 4994.
[24] Mikael Rørdam, Flemming Larsen, and Niels Jakob Laustsen, An introduction to K-theory for C∗-algebras.
London Mathematical Society Student Texts, 49. Cambridge University Press, Cambridge, 2000.
26
JUNHAO SHEN AND RUI SHI
[25] Shoichiro Sakai, C∗-algebras and W∗-algebras. Reprint of the 1971 edition. Classics in Mathematics. Springer-
Verlag, Berlin, 1998.
[26] Masamichi Takesaki. Theory of operator algebras. I. Reprint of the first (1979) edition. Encyclopaedia of
Mathematical Sciences, 124. Operator Algebras and Non-commutative Geometry, 5. Springer-Verlag, Berlin,
2002.
[27] Masamichi Takesaki. Theory of operator algebras. III. Encyclopaedia of Mathematical Sciences, 127. Oper-
ator Algebras and Non-commutative Geometry, 8. Springer-Verlag, Berlin, 2003.
[28] Dan Voiculescu. A non-commutative Weyl-von Neumann theorem. Rev. Roumaine Math. Pures Appl. 21
(1976), no. 1, 97 -- 113.
[29] Dan Voiculescu. Some results on norm-ideal perturbations of Hilbert space operators. J. Operator Theory
2 (1979), no. 1, 3 -- 37.
[30] Hermann Weyl. Uber beschrankte quadratische formen, deren differenz vollstetig ist. Rend. Circ. Mat.
Palermo 27 (1) (1909), 373 -- 392.
[31] L´aszl´o Zsid´o. The Weyl-von Neumann theorem in semifinite factors. J. Funct. Anal. 18 (1975), 60 -- 72.
Department of Mathematics & Statistics, University of New Hampshire, Durham, 03824, US
E-mail address: [email protected]
School of Mathematical Sciences, Dalian University of Technology, Dalian, 116024, China
E-mail address: [email protected], [email protected]
|
1610.05828 | 2 | 1610 | 2017-12-10T15:28:19 | Boundary representations of operator spaces, and compact rectangular matrix convex sets | [
"math.OA",
"math.FA"
] | We initiate the study of matrix convexity for operator spaces. We define the notion of compact rectangular matrix convex set, and prove the natural analogs of the Krein-Milman and the bipolar theorems in this context. We deduce a canonical correspondence between compact rectangular matrix convex sets and operator spaces. We also introduce the notion of boundary representation for an operator space, and prove the natural analog of Arveson's conjecture: every operator space is completely normed by its boundary representations. This yields a canonical construction of the triple envelope of an operator space. | math.OA | math |
BOUNDARY REPRESENTATIONS OF OPERATOR SPACES, AND COMPACT
RECTANGULAR MATRIX CONVEX SETS
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
Abstract. We initiate the study of matrix convexity for operator spaces. We define the notion of compact rect-
angular matrix convex set, and prove the natural analogs of the Krein-Milman and the bipolar theorems in this
context. We deduce a canonical correspondence between compact rectangular matrix convex sets and operator
spaces. We also introduce the notion of boundary representation for an operator space, and prove the natural
analog of Arveson's conjecture: every operator space is completely normed by its boundary representations.
This yields a canonical construction of the triple envelope of an operator space.
1. Introduction
It is hard to overstate the importance of the theory of convexity in analysis. This is all the more true in the
study of operator systems, which can be seen as the noncommutative analog of compact convex sets. Indeed,
given any operator system S, the space of matrix-valued unital completely positive maps on S is endowed with
a natural notion of convex combinations with matrix coefficients (matrix convex combination), and a topology
which is compact as long as one restricts the target to a fixed matrix algebra. The compact matrix convex sets
that arise in this way have been initially studied by Effros and Wittstock in [49, 20]. The program of developing
the theory of compact matrix convex sets as the noncommutative analog of compact convex sets has been
proposed by Effros in [20]. This program has been pursued in [22, 47], where matricial generalizations of the
classical Krein-Milman and bipolar theorems are proved. Compact matrix convex sets and the corresponding
notion of matrix extreme points have been subsequently studied in a number of papers. This line of research
has recently found outstanding applications. These include the matrix convexity proof of Arveson's conjecture
on boundary representations due to Davidson and Kennedy [17] building on previous work of Farenick [24, 25],
and the work of Helton, Klep, and McCullogh in free real algebraic geometry [31, 32].
The main goal of this paper is to provide the nonselfadjoint analog of the results above in the setting
of operator spaces. Precisely, we introduce the notion of compact rectangular matrix convex set, which is
the natural analog of the notion of compact matrix convex set where convex combinations with rectangular
matrices are considered. We then prove generalizations of the Krein-Milman and bipolar theorems in the
setting of compact rectangular matrix convex sets. We then deduce that compact rectangular matrix convex
sets are in canonical functorial one-to-one correspondence with operator spaces. It follows from this that any
operator space is completely normed by the matrix-valued completely contractive maps that are rectangular
matrix extreme points.
We also introduce the notion of boundary representation for operator spaces. The natural operator space
analog of Arveson's conjecture is then established: any operator space is completely normed by its boundary
representations. This gives an explicit description of the triple envelope of an operator space in terms of
boundary representations. We also obtain in this setting an analog of Arveson's boundary theorem. As an
application, we compute boundary representations for multiplier spaces associated with pairs of reproducing
kernel Hilbert spaces.
The results of this paper can be seen as the beginning of a convexity theory approach to the study of operator
spaces. Convexity theory has played a crucial role in the setting of Banach spaces, such as in the groundbreaking
work of Alfsen and Effros on M-ideals in Banach spaces [2, 3] or the work of Lazar and Lindenstrauss on L1-
predual spaces [36, 37]. This work can be seen as a first step towards establishing noncommutative analogs of
Date: August 13, 2018.
2000 Mathematics Subject Classification. Primary 46L07, 47L25; Secondary 46E22, 47L07.
Key words and phrases. Operator space, operator system, boundary representation, compact matrix convex set, matrix-gauged
space.
M.H. was partially supported by an Ontario Trillium Scholarship and a Feodor Lynen Fellowship. M.L. was partially supported
by the NSF Grant DMS-1600186. This work was initiated during a visit of M.H. at the California Institute of Technology in the
Spring 2016, and continued during a visit of M.H. and M.L. at the Oberwolfach Mathematics Institute supported by an Oberwolfach
Leibnitz Fellowship. The authors gratefully acknowledge the hospitality and the financial support of both institutions.
1
2
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
the results of Alfsen-Effros and Lazar-Lindenstrauss mentioned above. We will see below that the crucial notion
of collinearity for bounded linear functionals on a Banach space, which is of key importance for the work of
Alfsen and Effros on facial cones and M-ideals, has a natural interpretation in the setting of rectangular matrix
convexity.
The results of the present paper have already found application in [38]. The fact established here that an
operator space is completely normed by its matrix-valued completely contractive maps that are rectangular
matrix extreme is used there to prove that the noncommutative Gurarij space introduced by Oikhberg in [39] is
the unique separable nuclear operator space with the property that the canonical map from the maximal TRO
to the triple envelope is injective.
This paper is divided into three sections, besides the introduction.
In Section 2 we introduce the notion
of boundary representation for operator spaces, and prove that any operator space is completely normed by
its boundary representations. The boundary theorem for operator spaces and applications to multiplier spaces
for pairs of reproducing kernel Hilbert spaces are also considered in this section. In Section 3 we introduce
the notion of compact rectangular matrix convex set and rectangular matrix extreme point. We prove the
Krein-Milman and bipolar theorem for compact rectangular matrix convex sets, and deduce the correspondence
between compact rectangular matrix convex sets and operator spaces. Finally in Section 4 we consider the
notion of matrix-gauged space. Such a concept has recently been introduced by Russell in [42] in order to
provide an abstract characterization of selfadjoint subspaces of C*-algebras (selfadjoint operator spaces), and
to capture the injectivity property of B(H) in such a category. We prove in Section 4 that the construction of
the injective envelope and the C*-envelope of an operator system can be naturally generalized to the setting
of matrix-gauged spaces. A geometric approach to the study of matrix-gauged spaces and selfadjoint operator
spaces is also possible, as we show that matrix-gauged spaces are in functorial one-to-one correspondence with
compact matrix convex sets with a distinguished matrix extreme point.
2. Boundary representations and the Shilov boundary of an operator space
2.1. Notation and preliminaries. Recall that a ternary ring of operators (TRO) T is a subspace of the
C*-algebra B(H) of bounded linear operators on a Hilbert space H that is closed under the triple product
(x, y, z) 7→ xy∗z. An important example of a TRO is the space B(H, K), where H, K are Hilbert spaces. A
TRO has a canonical operator space structure coming from the inclusion T ⊂ B(H), which does not depend
on the concrete representation of T as a ternary ring of operators on H. A triple morphism between TROs
is a linear map that preserves the triple product. Any TRO can be seen as the 1-2 corner of a canonical C*-
algebra L (T ) called its linking algebra. A triple morphism between TROs can be seen as the 1-2 corner of a
*-homomorphism between the corresponding linking algebras [30]; see also [10, Corollary 8.3.5].
The notions of (nondegenerate, irreducible, faithful) representations admit natural generalizations from C*-
algebras to TROs. A representation of a TRO T is a triple morphism θ : T → B(H, K) for some Hilbert
spaces H, K. A linear map ψ : T → B(H, K) is nondegenerate if, whenever p, q are projections in B(H) and
B(K), respectively, such that qθ(x) = θ(x)p = 0 for every x ∈ T , one has p = 0 and q = 0. Similarly a
representation θ of T is irreducible if , whenever p, q are projections in B(H) and B(K), respectively, such that
qθ(x)p + (1 − q) θ(x) (1 − p) = θ(x) for every x ∈ T (equivalently, qθ(x) = θ(x)p for every x ∈ T ), one has p = 1
and q = 1, or p = 0 and q = 0. Finally, θ is called faithful if it is injective or, equivalently, completely isometric.
Various characterizations of nondegenerate and irreducible representations are obtained in [11, Lemma 3.1.4 and
Lemma 3.1.5]. A concrete TRO T ⊂ B(H, K) is said to act nondegenerately or irreducibly if the corresponding
inclusion representation is nondegenerate or irreducible, respectively.
In the following we will use frequently without mention the Haagerup-Paulsen-Wittstock extension theorem
[40, Theorem 8.2], asserting that, if H, K are Hilbert spaces, then the space B(H, K) of bounded linear operators
from H to K is injective in the category of operator spaces and completely contractive maps. We will also often
use the canonical way, due to Paulsen, to assign to an operator space X ⊂ B(H, K) an operator system
S(X) ⊂ B(K ⊕ H). This operator system, called the Paulsen system, is defined to be the space of operators
where IH and IK denote the identity operator on H and K, respectively. Any completely contractive map φ :
X → Y between operator spaces extends canonically to a unital completely positive map S(φ) : S(X) → S(Y )
defined by
(cid:26)(cid:20)λIK
y∗
x
µIH(cid:21) : x, y ∈ X, λ, µ ∈ C(cid:27)
(cid:20)λIK
y∗
x
µIH(cid:21) 7→(cid:20) λIK
φ(y)∗ µIH(cid:21) ,
φ(x)
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
3
see [40, Lemma 8.1]. (The Paulsen system is defined in [40, Chapter 8] and [10, Section 1.3] only in the case
when H = K. The same proofs from [40, Chapter 8] and [10, Section 1.3] apply with no change to this more
general situation.)
2.2. Dilations of rectangular operator states. Suppose that X is an operator space. A rectangular operator
state on X is a nondegenerate linear map φ : X → B(H, K) such that kφkcb = 1. We say that a rectangular
a dilation θ a minimal dilation of φ. In the sequel, it will often be convenient to identify H with a subspace of
such that w∗ψ(x)v = φ(x) for every x ∈ X. The same proof as [40, Theorem 8.4] gives the following.
Proposition 2.1. Any rectangular operator state φ : T → B(H, K) on a TRO T ⊂ B (H0, K0) can be dilated
In order to prove Proposition 2.1 one can proceed as in [40, Theorem 8.4], by replacing M2 (A) for a given
C*-algebra A with the C*-algebra generated by S (T ) inside B(K0 ⊕ H0). It is clear that in Proposition 2.1
operator state ψ : X → B(eH, eK) is a dilation of φ if there exist linear isometries v : H → eH and w : K → eK
to a nondegenerate triple morphism θ : T → B(eH, eK). If H0, K0, H, K are finite-dimensional, then one can
take eH and eK to be finite-dimensional.
one can choose θ and the linear isometries v : H → eH and w : K → eK in such a way that eK is the linear span
of θ(T )θ(T )∗wK ∪ θ (T ) vH, and eH is the linear span of θ(T )∗θ(T )vH ∪ θ (T )∗ wK. In this case, we call such
eH and K with a subspace of eK.
Definition 2.2. Let φ : X → B(H, K) be a rectangular operator state and let ψ : X → B(eH, eK) be a dilation
of φ. We can assume that H ⊂ eH and K ⊂ eK. Let p be the orthogonal projection from eH onto H and let q be
the orthogonal projection from eK onto K. The dilation ψ is trivial if
It is clear that, when X is an operator system, eH = eK, q = p, and ψ is a unital completely positive map,
state φ on X has the unique extension property if any rectangular operator state eφ of T whose restriction to X
Definition 2.3. Suppose that X is a subspace of a TRO T such that T is generated as a TRO by X. The operator
for every x ∈ X. The operator state φ on an operator space X is maximal if it has no nontrivial dilation.
the notion of trivial dilation as above recovers the usual notion of trivial dilation.
coincides with φ is automatically a triple morphism.
We now observe, that a rectangular operator state on an operator space is maximal if and only if it has the
unique extension property. The analogous fact for unital completely positive maps on operator systems is well
known; see [6].
Lemma 2.4. For a TRO T , the set of positive elements of the C∗-algebra T T ∗ is the closed convex cone
generated by {xx∗ : x ∈ T }.
Proof. Let C ⊂ T T ∗ denote the closed convex cone generated by {xx∗ : x ∈ T }. It is clear that C is contained
in the set of positive elements of T T ∗. Conversely, suppose that a ∈ T T ∗ is positive. By the remarks at the
beginning of Section 2.2 of [23], the C∗-algebra T T ∗ admits a contractive approximate identity (ei) of elements
of the form
ψ (x) = qψ (x) p + (1 − q)ψ (x) (1 − p)
as T is a left T T ∗-module. Thus, a1/2eia1/2 ∈ C for every i. It follows that a = limi a1/2eia1/2 ∈ C.
(cid:3)
Lemma 2.5. Let T be a TRO, let ψ : T → B(eH, eK) be a completely contractive linear map and suppose that
H ⊂ eH and K ⊂ eK are closed subspaces with corresponding orthogonal projections p ∈ B(eH) and q ∈ B(eK). If
θ : T → B(H, K),
x 7→ qψ(x)p,
the map
is a non-degenerate triple morphism, then ψ is a trivial dilation of θ.
where xj ∈ T . For such an element of T T ∗ one has that
xj x∗
j
Xj
j a1/2 =Xj
a1/2Xj
xjx∗
(a1/2xj)(a1/2xj)∗ ∈ C,
4
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
Proof. By dilating ψ if necessary, we may assume without loss of generality that ψ is a triple morphism. Thus,
there exists a unique ∗-homomorphism
The assumption further implies that there exists a ∗-homomorphism
π : T T ∗ → B(K)
such that π(xy∗) = qψ(x)pψ(y)∗q.
σ : T T ∗ → B(eK)
such that
σ(xy∗) = ψ(x)ψ(y)∗
(x, y ∈ T ).
Since θ is non-degenerate, π is non-degenerate as well. Consider now the map
ϕ : T T ∗ → B(K),
a 7→ qσ(a)q − π(a).
We claim that ϕ = 0. To this end, observe that if x ∈ T , then
ϕ(xx∗) = qψ(x)(1 − p)ψ(x)∗q ≥ 0,
so ϕ is a positive map by Lemma 2.4. Let (ei) be an approximate identity for T T ∗. Since π is non-degenerate,
π(ei) tends to q in the strong operator topology. Since qσ(ei)q ≤ q, it follows that ϕ(ei) = qσ(ei)q − π(ei) tends
to zero in the strong operator topology. Combining this with positivity of ϕ, it follows that ϕ = 0 (see, e.g. [16,
Lemma I.9.5]).
In particular, we see that for x ∈ T ,
0 = ϕ(xx∗) = qψ(x)(1 − p)ψ(x)∗q = [qψ(x)(1 − p)][qψ(x)(1 − p)]∗,
so that qψ(x)(1 − p) = 0. A similar argument, replacing T T ∗ with T ∗T , shows that pψ(x)∗(1 − q) = 0. Thus,
ψ is a trivial dilation of θ.
(cid:3)
Proposition 2.6. Suppose that φ : X → B(H, K) is a rectangular operator state of X, and T is a TRO
containing X as a generating subspace. Then φ is maximal if and only if it has the unique extension property.
for every x ∈ T . The restriction of θ to X is a rectangular operator state that dilates φ. By maximality of φ,
we can conclude that θ(x) = qθ(x)p + (1 − q) θ(x) (1 − p) for every x ∈ X. Since X generates T as a TRO and
Proof. Suppose initially that φ is maximal. Let eφ : T → B(H, K) be an extension of φ. Let θ : T → B(eH, eK) be
a dilation of eφ to a triple morphism. We can identify H with a subspace of eH and K with a subspace of eK. Let
p and q be the orthogonal projections of eH and eK onto H and K, respectively. We have that eφ(x) = qθ(x)H
θ is a triple morphism, it follows that this identity holds for every x ∈ T . It follows that eφ is a triple morphism
Suppose now that φ has the unique extension property. Let ψ : X → B(eH, eK) be a dilation of φ. As above, we
will identify H and K as subspaces of eH and eK, respectively, and denote by p and q the corresponding orthogonal
projections. We can extend ψ to a rectangular operator state ψ : T → B(eH, eK). Observe that x 7→ qψ(x)H is
a rectangular operator state extending φ. Since φ has the unique extension property, x 7→ qψ(x)H is a triple
morphism. Hence from Lemma 2.5 we can conclude that ψ is a trivial dilation of φ. Since ψ was arbitrary, we
can conclude that φ is maximal.
(cid:3)
as well.
Simple examples show that the implication "unique extension property implies maximal" of Proposititon 2.6
may fail if φ is a degenerate completely contractive map. Indeed, there are degenerate representations of TROs
which have non-trivial dilations.
2.3. Boundary representations. Suppose that X is an operator space, and T is a TRO containing X as a
generating subspace.
Definition 2.7. A boundary representation for X is a rectangular operator state φ : X → B(H, K) with the
property that any rectangular operator state on T extending X is an irreducible representation of T .
In other words, a rectangular operator state φ : X → B(H, K) is a boundary representation for X if and only
if it has the unique extension property, and the unique extension of φ to T is an irreducible representation of
T . In the following we will identify a boundary representation of X with its unique extension to an irreducible
representation of T . It follows from Proposition 2.6 that the notion of boundary representation does not depend
on the concrete realization of X as a space of operators. We remark that our terminology differs slightly from
Arveson's original use of the term boundary representation in the context of operator systems [7]. Indeed, for
Arveson, a boundary representation is a representation of the C∗-algebra generated by the operator system.
More precisely, if S is an operator system that generates the C∗-algebra A, then according to Arveson, a
boundary representation for S is an irreducible representation π of A such that π is the unique completely
positive extension of πS. We follow the convention, which is for example used in [17], that a boundary
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
5
representation of S is a unital completely positive map φ : S → B(H) such that every extension of φ to a
completely positive map on A is an irreducible representation of A. Since this notion does not depend on the
concrete representation of S, these two points of view are equivalent.
In the rest of this section, we will observe that the boundary representations of X completely norm X. This
will be deduced from the corresponding fact about operator systems, proved in [6] in the separable case and in
[17] in full generality.
Proposition 2.8. Suppose ω : S(X) → B(Lω) is a boundary representation of the Paulsen system S(X)
associated with X. Then one can decompose Lω as an orthogonal direct sum Kω ⊕ Hω in such a way that
ω = S(ψ) for some boundary representation ψ : X → B(Hω, Kω) of X.
Proof. Suppose that T ⊂ B(H, K) is a TRO containing X as a generating subspace. Let A be the C*-algebra
generated by S(X) inside B(K ⊕ H). Observe that
Since ω is a boundary representation of S(X), it extends to an irreducible representation ω : A → B(Lω). Set
A =(cid:26)(cid:20)x11 + λIK
x21
x12
x22 + µIH(cid:21) : x11 ∈ T T ∗, x12 ∈ T , x21 ∈ T ∗, x22 ∈ T ∗T , λ, µ ∈ C(cid:27) .
qω := ω(cid:18)(cid:20)IK 0
0(cid:21)(cid:19) , and pω := ω(cid:18)(cid:20)0
IH(cid:21)(cid:19) .
0
0
0
Observe that pω, qω ∈ B(Lω) are orthogonal projections such that pω + qω = ILω . Denote by Kω the range of
qω and by Hω the range of pω. The fact that ω is a unital *-homomorphism implies that, with respect to the
decomposition Lω = Kω ⊕ Hω one has that
ω =(cid:20) σ
θ∗ π(cid:21)
θ
where θ : T → B(Hω, Kω) is a triple morphism.
We claim that θ is irreducible. Suppose that p ∈ B(Hω) and q ∈ B(Kω) are projections such that qθ(x) =
θ(x)p for every x ∈ T . Since σ (ab∗) = θ(a)θ(b)∗ and π (a∗b) = θ(a)∗θ(b) for every a, b ∈ T , we can conclude
that (q ⊕ p) ω(x) = ω(x)(q ⊕ p) for every x ∈ A. Since ω is an irreducible representation of A, it follows that
q ⊕ p = 1 or q ⊕ p = 0. This concludes the proof that θ is irreducible.
Denote by ψ the restriction of θ to X. Observe that ω = S(ψ). We claim that ψ is maximal. Indeed, let
φ : X → B(eH, eK) be a dilation of ψ. We can identify H and K as subspaces of eH and eK, with corresponding
orthogonal projections p and q. Then S(φ) : S(X) → B(eK ⊕ eH) is a dilation of ω. By maximality of ω, we
have that
0
0
(cid:20)q
0 p(cid:21) ω(x) = ω(x)(cid:20)q
0 p(cid:21)
for every x ∈ S(X). It follows that qφ(x) = φ(x)p for every x ∈ X. This shows that φ is a trivial dilation of ψ,
concluding the proof that ψ is maximal.
(cid:3)
The following result is now an immediate consequence of Proposition 2.8 and [17, Theorem 3.4].
Theorem 2.9. Suppose that X is an operator space. Then X is completely normed by its boundary represen-
tations.
2.4. Rectangular extreme points and pure unital completely positive maps. Suppose that X is an
operator space, and φ : X → B(H, K) is a completely contractive linear map. A rectangular operator convex
combination is an expression φ = α∗
nφnβn, where βi : H → Hi and αi : K → Ki are linear maps,
and φi : X → B(Hi, Ki) are completely contractive linear maps for i = 1, 2, . . . , ℓ such that α∗
nαn = 1,
and β∗
nβn = 1. Such a rectangular convex combination is proper if αi, βi are surjective, and trivial
if α∗
1φ1β1 + · · · + α∗
1 β1 + · · · + β∗
i φiβi = λiφ for some λi ∈ [0, 1].
i βi = λi1, and α∗
i αi = λi1, β∗
1α1+· · ·+α∗
Definition 2.10. A completely contractive map φ : X → B(H, K) is a rectangular operator extreme point if
any proper rectangular operator convex combination φ = α∗
1φ1β1 + · · · + α∗
nφnβn is trivial.
Suppose now that X is an operator system. An operator state on X is a unital completely positive map
nφnαn, where αi : H → Hi
nαn = 1.
i αi = λi1
φ : X → B(H). An operator convex combination is an expression φ = α∗
are linear maps, and φi : X → B(Hi) are operator states for i = 1, 2, . . . , ℓ such that α∗
1α1 + · · · + α∗
Such an operator convex combination is proper if αi is right invertible for i = 1, 2, . . . , ℓ, and trivial if α∗
and α∗
1φ1α1+· · ·+α∗
i φiαi = λiφ for some λi ∈ [0, 1].
and
Ψ0(cid:18)(cid:20)IL1
0
It follows that the two projections (cid:20)IL1
0
In particular, w = Ψ(1) is a diagonal element, and the unital completely positive map Ψ0 = w−1/2Ψw−1/2
satisfies
0
0
0
0
0
Ψ(cid:18)(cid:20)x 0
0(cid:21)(cid:19) =(cid:20)ϕ1(x)
Ψ(cid:18)(cid:20)0
0 y(cid:21)(cid:19) =(cid:20)0
0(cid:21)(cid:19) =(cid:20)IK 0
0(cid:21)
0(cid:21) and (cid:20)0
0(cid:21) .
0 ϕ0(y)(cid:21) .
and Ψ0(cid:18)(cid:20)0
0 IL0(cid:21) belong to the multiplicative domain of Ψ0 [10,
IL0(cid:21)(cid:19) =(cid:20)0
IH(cid:21) .
0
0
0
0
0
0
0
0
Ψ0(cid:18)(cid:20)λIL1
y∗
x
µIL0(cid:21)(cid:19) =(cid:20) λIK
ψ(y)∗ µIH(cid:21) ,
ψ(x)
6
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
We say that φ is an operator extreme point if any proper operator convex combination φ = α∗
1φ1α1 + · · · +
nφnαn is trivial. The proof of [24, Theorem B] shows that an operator state is an operator extreme point if
α∗
and only if it is a pure element in the cone of completely positive maps.
When H is finite-dimensional, the notion of proper operator convex combination coincides with the notion of
proper matrix convex combination from [47]. In this case, the notion of operator extreme point coincides with
the notion of matrix extreme point from [47, Definition 2.1].
Lemma 2.11. Suppose that φ : X → B(H, K) is a completely contractive linear map, Ψ : S(X) → B(K ⊕ H) is
a completely positive map such that S(φ)−Ψ is completely positive. Suppose that Ψ(1) is an invertible element of
B(K ⊕ H). Then there exist positive invertible elements a ∈ B(K) and b ∈ B(H), and a completely contractive
map ψ : X → B(H, K) such that
Ψ(cid:18)(cid:20) λ
y∗ µ(cid:21)(cid:19) =(cid:20)a 0
ψ(y)∗ µIH(cid:21)(cid:20)a 0
b(cid:21) .
b(cid:21)(cid:20) λIK
ψ(x)
x
0
0
Proof. Fix a concrete representation X ⊂ B(L0, L1) of X. In this case S(X) ⊂ B(L1 ⊕ L0). Set Φ := S(φ)
and let T ⊂ B(L0, L1) denote the TRO generated by X. By Arveson's extension theorem, we may extend Φ
and Ψ to the C*-algebra A generated by S(X) inside B(L1 ⊕ L0) in such a way that Φ − Ψ is still completely
positive. In the following we regard Φ, Ψ as maps from A to B(K ⊕ H). Since Φ − Ψ is completely positive, the
argument in the proof of [40, Theorem 8.3] shows that there exist linear maps ϕ1 : T T ∗ + CIL1 → B(K) and
ϕ0 : T ∗T + CIL0 → B(H) such that
Proposition 1.3.11], so that there exists a completely contractive map ψ : X → B(H, K) such that
which finishes the proof.
(cid:3)
Proposition 2.12. Suppose that φ : X → B(H, K) is a completely contractive map and S(φ) : S(X) →
B(K ⊕ H) is the associated unital completely positive map defined on the Paulsen system. The following
assertions are equivalent:
(1) S(φ) is a pure completely positive map;
(2) S(φ) is an operator extreme point;
(3) φ is a rectangular operator extreme point.
Proof. We have already observed that the equivalence of (1) and (2) holds, as the argument in the proof of [24,
Theorem B] shows.
(2) =⇒ (3) Suppose that φ = α∗
ℓ φℓβℓ is a proper rectangular matrix convex combination.
Define γi = αi ⊕ βi for i = 1, 2, . . . , ℓ. Then we have that S(φ) = γ∗
ℓ S (φℓ) γℓ is a proper
matrix convex combination. Since by assumption S(φ) is an operator extreme point in the state space of S(X),
we can conclude that the proper matrix convex combination γ∗
ℓ S (φℓ) γℓ is trivial. This
implies that the proper rectangular matrix convex combination α∗
1 S (φ1) γ1 + · · · + γ∗
1φ1β1 + · · · + α∗
1 S (φ1) γ1 + · · · + γ∗
ℓ φℓβℓ is trivial as well.
1φ1β1 + · · · + α∗
(3) =⇒ (1) Suppose that S(φ) = Ψ1 + Ψ2 for some completely positive maps Ψ1, Ψ2 : S(X) → B(K ⊕ H).
Fix ε > 0 and define Ξi = (1 − ε) Ψi + (ε/2) S(φ) for i = 1, 2. Then Ξ1, Ξ2 : S(X) → B(K ⊕ H) are completely
positive maps such that Ξ1 + Ξ2 = S(φ) and Ξi(1) is invertible for i = 1, 2; cf. the proof of [17, Lemma 2.3]. By
Lemma 2.11 we have that, for i = 1, 2,
Ξi(cid:18)(cid:20) λ
y∗ µ(cid:21)(cid:19) =(cid:20)ai
x
0
0
bi(cid:21)(cid:20) λ1
ψi(y)∗
ψi(x)
µ1 (cid:21)(cid:20)ai
0
0
bi(cid:21)
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
7
for some positive invertible elements ai ∈ B(K), bi ∈ B(H), and completely contractive ψi : X → B(H, K).
Thus we have that φ = a1ψ1b1 + a2ψ2b2 is a proper rectangular operator convex combination. By assumption,
we have that a2
i = ti1, and aiψibi = tiφ for some ti ∈ [0, 1] and i = 1, 2. It follows that Ξi = tiS(φ) for
i = 1, 2. Since this is true for every ε, it follows that the Ψi are also scalar multiples of S(φ). This concludes
the proof that S(φ) is pure.
(cid:3)
i = ti1, b2
The following corollary is an immediate consequence of Proposition 2.12, Proposition 2.8, and [17, Theorem
2.4].
Corollary 2.13. Suppose that φ : X → B(H, K) is a rectangular operator state. If φ is rectangular operator
extreme, then φ admits a dilation to a boundary representation of X.
Suppose that X is a Banach space. We regard X as an operator space endowed with its canonical minimal
operator space structure obtained from the canonical inclusion of X in the C*-algebra C (Ball (X ′)), where X ′
denotes the dual space of X. In [2] Alfsen and Effros considered the following notion. Suppose that φ0, φ1 are
nonzero contractive linear functionals on X. Then φ0, φ1 are codirectional if kφ0 + φ1k = kφ0k + kφ1k. This
is equivalent to the assertion that φ0 + φ1S( φ0+φ1
kφ1k ). The relation ≺ on
nonzero contractive linear functionals is defined by setting φ ≺ ψ if and only if either φ = ψ or φ and ψ − φ are
codirectional. This is equivalent to the assertion that kφk S( φ
kφ0+φ1k ) = kφ0k S( φ0
kφ0k ) + kφ1k S( φ1
kφk ) ≤ kψk S( ψ
kψk ).
Proposition 2.14. Suppose that X is a Banach space and φ is a bounded linear functional on X of norm 1.
The following statements are equivalent:
(1) φ is an extreme point of Ball(X ′);
(2) if ψ ∈ Ball(X ′) is a nonzero linear functional such that ψ ≺ φ, then ψ is a scalar multiple of φ;
(3) φ is a rectangular extreme point.
Proof. The implications (3)⇒(1)⇒(2) are straightforward.
(2)⇒(3) Suppose that, for every nonzero ψ ∈ Ball(X ′), ψ ≺ φ implies that ψ is a scalar multiple of φ.
The proof of Proposition 2.12 shows that it suffices to show that every proper rectangular convex combination
of two elements is trivial. Thus, consider a rectangular convex combination φ = s0t0φ0 + s1t1φ1 for some
φ0, φ1 ∈ Ball(X ′) and non-zero s0, s1, t0, t1 ∈ C such that s02 + s12 = 1 and t02 + t12 = 1. Observe that
1 =(cid:13)(cid:13)s0t0φ0 + s1t1φ1(cid:13)(cid:13) ≤(cid:13)(cid:13)s0t0φ0(cid:13)(cid:13) +(cid:13)(cid:13)s1t1φ1(cid:13)(cid:13) ≤(cid:12)(cid:12)s0t0(cid:12)(cid:12) kφ0k +(cid:12)(cid:12)s1t1(cid:12)(cid:12) kφ1k ≤(cid:12)(cid:12)s0t0(cid:12)(cid:12) +(cid:12)(cid:12)s1t1(cid:12)(cid:12) ≤ 1.
Hence (cid:13)(cid:13)s0t0φ0(cid:13)(cid:13) +(cid:13)(cid:13)s1t1φ1(cid:13)(cid:13) = kφ0k = kφ1k = 1 and s0 = t0 and s1 = t1. By hypothesis we have that
In particular, ρi = si2 = ti2 for i = 0, 1. Since
s0t0φ0 = ρ0φ and s1t1φ1 = ρ1φ for some ρ0, ρ1 ∈ C.
φ = s0t0φ0 + s1t1φ1, it follows that ρ0 + ρ1 = 1. Combined with ρ0 + ρ1 = 1, this implies that ρ0, ρ1 ∈ [0, 1],
so that the rectangular convex combination was trivial.
(cid:3)
2.5. TRO-extreme points. Suppose that X is an operator space and ϕ : X → B(H, K) is a completely
contractive map. We say that ϕ is a TRO-extreme point if whenever ϕ = α∗
ℓ ϕℓβℓ is a proper
rectangular matrix convex combination such that ϕi : X → B(H, K) for i = 1, 2, . . . , ℓ, then α∗
i αi = ti1,
β∗
i βi = ti1, and α∗
i ϕiβi = tiϕ for some ti ∈ [0, 1]. When H, K are finite-dimensional, this is equivalent to
requiring that there exist unitaries ui ∈ Mn (C) and wi ∈ Mm (C) such that ϕi = u∗
i ϕwi. This can be seen
arguing as in the proof of Lemma 3.7 below. The notion of TRO-extreme point can be seen as the operator
space analog of the notion of C*-extreme point considered in [33, 27, 26].
1ϕ1β1 + · · · + α∗
A similar proof as [24, Theorem B] gives the following lemma.
Lemma 2.15. Let X be an operator space, and A be the C*-algebra generated by S(X). If ϕ : X → B(H, K)
is a TRO-extreme point such that the range of ϕ is an irreducible subspace of B(H, K), then there exists a pure
unital completely positive map Φ : A → B(K ⊕ H) that extends S(ϕ).
Using this lemma, one can prove similarly as [24, Theorem C] the following fact:
Proposition 2.16. Suppose that T is a TRO, and ϕ : T → B(H, K) is a TRO-extreme point. Then there exist
pairwise orthogonal projections (pi)i∈I in B(H) and (qi)i∈I in B(K) such that piϕqi is rectangular operator
extreme for every i, and piϕqj = 0 for every i 6= j.
8
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
2.6. The Shilov boundary of an operator space. Suppose that X is an operator space. A triple cover of X
is a pair (ι, T ) where T is a TRO and ι : X → T is a completely isometric linear map whose range is a subspace
of T that generates T as a TRO. Among the triple covers there exists a canonical one: the triple envelope. This
is the (unique) triple cover (ιe, Te(X)) with the property that for any other triple cover (ι, T ) of X, there exists
a triple morphism θ : T → Te(X) such that θ ◦ ι = ιe. The existence of the triple envelope was established by
Hamana in [29] using the construction of the injective envelope of an operator space; see also [10, Section 4.4].
The triple envelope of an operator space is referred to as the (noncommutative) Shilov boundary in [9]. It is
remarked in [9] at the beginning of Section 4, referring to the Shilov boundary of an operator space, that "the
spaces above are not at the present time defined canonically", and "this lack of canonicity is always a potential
source of blunders in this area, if one is not careful about various identifications." We remark here that the
theory of boundary representations provides a canonical construction of the Shilov boundary of an operator
space X. Indeed one can consider ιe : X → B(H, K) to be the direct sum of all the boundary representations
for X, and then let (ιe, Te(X)) be the subTRO of B(H, K) generated by the image of ιe. Proposition 2.6 implies
that ιe is maximal, so we may argue as in the proof of [18, Theorem 4.1] that Te(X) is indeed the triple envelope
of X. It also follows that if X has a completely isometric boundary representation θ : X → B(H), then the
triple envelope of X is the TRO generated by the range of θ inside B(H).
2.7. The Shilov boundary of a Banach space. A TRO T is commutative if xy∗z = zy∗x for every x, y, z ∈ T .
Several equivalent characterizations of commutative TROs are provided in [10, Proposition 8.6.5]. Suppose that
E is a locally trivial line bundle over a locally compact Hausdorff space U . Then the space Γ0(E) of continuous
sections of E that vanish at infinity is a commutative TRO such that Γ0(E)∗Γ0(E) = C0(E). Conversely, it is
observed in [9, Section 4] -- see also [19] -- that any commutative TRO is of this form. One can also describe the
commutative TROs as the Cσ-spaces from the Banach space literature.
Suppose that E is a locally trivial line bundle over a locally compact Hausdorff space U with point at infinity
∞ and X ⊂ Γ0(E) be a closed subspace. Assume that the set of elements {hx, yi : x, y ∈ X} of C0 (U ) separates
the points of U and does not identitically vanish at any point of U . This is equivalent to the assertion that X
generates Γ0(E) as a TRO, as proved in [9, Theorem 4.20]. An irreducible representation of Γ0(E) is of the form
x 7→ x (ω0) for some ω0 ∈ U . A linear map from Γ0(E) to C of norm 1 has the form x 7→R x (ω) dµ (ω) for some
Borel probability measure µ on U . We say that µ is a representing measure for ω0 ∈ U ifR x (ω) dµ (ω) = x (ω0)
for every x ∈ X. A point ω0 ∈ U is a Choquet boundary point if the point mass at ω0 is the unique representing
measure for ω0. It follows from the observations above that ω0 is a Choquet boundary point for X if and only
if the map x 7→ x (ω0) is a boundary representation for X. The Choquet boundary Ch(X) of X is the set of
Choquet boundary points of X.
Suppose that ∂SX ∪ {∞} is the closure of Ch(X) ∪ {∞} inside U ∪ {∞}. Then it follows from Theorem
2.9 that E∂S X is the Shilov boundary of X in the sense of [9, Theorem 4.25]. This means that the linear map
X → Γ0(E∂S X ), x 7→ x∂S X is isometric, and for any locally trivial line bundle over a locally compact Hausdorff
space V and linear isometry J : X → Γ0 (V ) with the property that the set {J(x)∗J(y) : x, y ∈ X} separates
the points of V and does not identically vanish at any point of V , there exists a proper continuous injection
ϕ : ∂SX → V with the property that J(x) ◦ ϕ = x∂S X for every x ∈ X. This gives a canonical construction
of the Shilov boundary of a Banach space, analogous to the canonical construction of a Shilov boundary of a
unital function space; see [10, Section 4.1].
2.8. The rectangular boundary theorem. Arveson's boundary theorem [5, Theorem 2.1.1] asserts that if
S ⊂ B(H) is an operator system which acts irreducibly on H such that the C*-algebra C∗(S) contains the
algebra of compact operators K(H), then the identity representation of C∗(S) is a boundary representation for
S if and only if the quotient map B(H) → B(H)/K(H) is not completely isometric on S.
The following result is a rectangular generalization of Arveson's boundary theorem.
Theorem 2.17. Let X ⊂ B(H, K) be an operator space such that the TRO T generated by X acts irreducibly
and such that T ∩ K(H, K) 6= {0}. Then the identity representation of T is a boundary representation for X if
and only if the quotient map B(H, K) → B(H, K)/K(H, K) is not completely isometric on X.
Proof. Suppose first that the quotient map π is completely isometric on X. Then π, regarded as a map from X →
π(X), admits a completely isometric inverse, which extends to a complete contraction ψ : B(H, K)/K(H, K) →
B(H, K). Clearly, ψ ◦ π is a completely contractive map which extends the inclusion of X into B(H, K), but it
does not extend the inclusion of T into B(H, K), since it annihilates the compact operators.
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
9
Conversely, suppose that the quotient map is not completely isometric on X. Let S(X) ⊂ B(K ⊕ H) denote
the Paulsen system associated with X. We will verify that S(X) satisfies the assumptions of Arveson's boundary
theorem.
To see that S(X) acts irreducibly, suppose that p is an orthogonal projection on K ⊕ H which commutes
with S(X). In particular, p commutes with IK ⊕ 0 and 0 ⊕ IH , from which we deduce that p = p1 ⊕ p2, where
p1 ∈ B(K) and p2 ∈ B(H) are orthogonal projections. For x ∈ X, we therefore have
(cid:20)0 p1x
0 (cid:21) =(cid:20)p1
0
0
0
p2(cid:21)(cid:20)0 x
0(cid:21) =(cid:20)0 x
0(cid:21)(cid:20)p1
0
0
0
0 (cid:21) ,
p2(cid:21) =(cid:20)0 xp2
0
0
hence p1x = xp2 for all x ∈ X. Since X acts irreducibly, it follows that either p1 ⊕ p2 = 0 or p1 ⊕ p2 = IK⊕H ,
so that S(X) acts irreducibly. The assumption that T contains a non-zero compact operator implies that S(X)
contains a non-zero compact operator, hence by irreducibility of S(X), we see that K(K ⊕ H) ⊂ C∗(S(X)).
Since the quotient map B(H, K) → B(H, K)/K(H, K) is not completely isometric on X, there exists x ∈
Mn(X) and k ∈ Mn(K(H, K)) such that x − k < x. Regarding x as an element of Mn(S(X)) in the
canonical way and correspondingly k as an element of Mn(K(K ⊕H)), we see that the quotient map B(K ⊕H) →
B(K ⊕ H)/K(K ⊕ H) is not completely isometric on S(X). Thus, Arveson's boundary theorem implies that
the identity representation is a boundary representation of S(X). According to Proposition 2.8, there exists
a boundary representation ψ : X → B(L1, L2) of X such that S(ψ) is the inclusion of S(X) into B(K ⊕ H).
It easily follows now that L1 = H, L2 = K and that ψ is the inclusion of X into B(H, K), which finishes the
proof.
(cid:3)
2.9. Rectangular multipliers. A reproducing kernel Hilbert space H on a set X is a Hilbert space of functions
on X such that for every x ∈ X, the functional
is bounded. The unique function k : X × X → C which satisfies k(·, x) ∈ H for all x ∈ X and
H → C,
f 7→ f (x),
hf, k(·, x)i = f (x)
for all x ∈ X and f ∈ H is called the reproducing kernel of H. We will always assume that H has no common
zeros, meaning that there does not exist x ∈ X such that f (x) = 0 for all x ∈ X. Equivalently, k(x, x) 6= 0 for
all x ∈ X. We refer the reader to the books [41] and [1] for background material on reproducing kernel Hilbert
spaces.
If H and K are reproducing kernel Hilbert spaces on the same set X, we define the multiplier space
Mult(H, K) = {ϕ : X → C : ϕ · f ∈ K for all f ∈ H},
where (ϕ · f ) (x) = ϕ (x) f (x) for x ∈ X; see [41, Section 5.7]. By [41, Theorem 5.21], every ϕ ∈ Mult(H, K)
induces a bounded multiplication operator Mϕ : H → K. Moreover, since K has no common zeros, every mul-
tiplier ϕ is uniquely determined by its associated multiplication operator Mϕ. We may thus regard Mult(H, K)
as a subspace of B(H, K).
The best studied case occurs when H = K, in which case Mult(H) = Mult(H, H) is an algebra, called
the multiplier algebra of H; [1, Section 2.3]. Nevertheless, the rectangular case of two different reproducing
kernel Hilbert spaces has been studied as well, see for example [46] and [45], where multipliers between weighted
Dirichlet spaces are investigated.
We say that a reproducing kernel Hilbert space K on X with reproducing kernel k is irreducible if X cannot be
partitioned into two non-empty sets X1 and X2 such that k(x, y) = 0 for all x ∈ X1 and y ∈ X2. This definition
is more general than the definition of irreducibility in [1, Definition 7.1], but it suffices for our purposes.
For the next Lemma, we observe that if H contains the constant function 1, then Mult(H, K) is contained
in K.
Lemma 2.18. Let H and K be reproducing kernel Hilbert spaces on the same set. Suppose that
• H contains the constant function 1,
• Mult(H, K) is dense in K, and
• K is irreducible.
Then Mult(H, K) ⊂ B(H, K) acts irreducibly.
Proof. Suppose that p ∈ B(H) and q ∈ B(K) are orthogonal projections which satisfy qMϕ = Mϕp for all
ϕ ∈ Mult(H, K). Define ψ = p1 ∈ H. Then for all ϕ ∈ Mult(H, K), the identity
qϕ = qMϕ1 = Mϕp1 = ψϕ
10
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
holds. Since Mult(H, K) is dense in K, we deduce that ψ ∈ Mult(K) and that q = Mψ. We claim that ψ
is necessarily constant. To this end, let k denote the reproducing kernel of K. Note that Mψ is in particular
selfadjoint, so that
ψ(x)k(x, y) = hMψk(·, y), k(·, x)i = hk(·, y), Mψk(·, x)i = ψ(y)k(x, y)
for all x, y ∈ X. Hence ψ(x) is real for all x ∈ X and ψ(x) = ψ(y) if k(x, y) 6= 0. Fix x0 ∈ X, and suppose for
a contradiction that
X1 = {x ∈ X : ψ(x) = ψ(x0)}
is a proper subset of X and let X2 = X \ X1.
If x ∈ X1 and y ∈ X2, then ψ(y) 6= ψ(x0) = ψ(x), hence
k(x, y) = 0. This contradicts irreducibility of K, so that ψ is constant. Moreover, since Mψ is a projection, we
necessarily have ψ = 1 or ψ = 0. If ψ = 1, then q = Mψ = IK and Mϕ(IH − p) = 0 for all ϕ ∈ Mult(H, K).
Similarly, if ψ = 0, then q = 0 and Mϕp = 0 for all ϕ ∈ Mult(H, K). We may thus finish the proof by showing
that
\ϕ∈Mult(H,K)
ker(Mϕ) = {0}.
To this end, note that if x ∈ X, then {f ∈ K : f (x) = 0} is a proper closed subspace of K, as K has no common
zeros. Since Mult(H, K) is dense in K, it cannot be contained in such a subspace, thus for every x ∈ X, there
exists ϕ ∈ Mult(H, K) such that ϕ(x) 6= 0. If f ∈ H satisfies Mϕf = 0 for all ϕ ∈ Mult(H, K), it therefore
follows that f (x) = 0 for all x ∈ X, that is, f = 0, as desired.
(cid:3)
We can now use the rectangular boundary theorem to show that for many multiplier spaces, the identity
representation is always a boundary representation.
Proposition 2.19. Let H and K be reproducing kernel Hilbert spaces on the same set and let M = Mult(H, K).
Suppose that
• H contains the constant function 1,
• M is dense in K,
• K is irreducible, and
• M contains a non-zero compact operator.
Then the identity representation is a boundary representation of M . In particular, the triple envelope of M
is the TRO generated by M .
Proof. Lemma 2.18 shows that M acts irreducibly. Moreover, the quotient map by the compacts is not isometric
on M since M contains a non-zero compact operator. An application of the rectangular boundary theorem
(Theorem 2.17) now finishes the proof.
(cid:3)
For s ∈ R, let
Hs =nf (z) =
∞Xn=0
anzn : f H2
s =
∞Xn=0
an2(n + 1)−s < ∞o.
This is a reproducing kernel Hilbert space on the open unit disc D with reproducing kernel
ks(z, w) =
(n + 1)s(zw)n.
∞Xn=0
This scale of spaces is a frequent object of study in the theory of reproducing kernel Hilbert spaces. The space
H0 is the classical Hardy space H 2, the space H−1 is the Dirichlet space, and the space H1 is the Bergman
space.
The elements of Mult(Hs, Ht) were characterized in [46] and [45]. We remark that the spaces Dα of [46] are
related to the spaces above via the formula Dα = H−α. In [45], a slightly different convention is used. There,
Dα = H−2α, at least with equivalent norms. Theorem 4 of [46] shows that Mult(Hs, Ht) = {0} if s > t. On
the other hand, if s ≤ t, then Hs ⊂ Ht, hence Mult(Hs) ⊂ Mult(Hs, Ht). Since Mult(Hs) at least contains the
polynomials, the same is true for Mult(Hs, Ht).
In the square case s = t, boundary representations of operator spaces related to the algebras Mult(Hs), and
their analogs on higher dimensional domains, were studied in [28, 34, 14]; see in particular [28, Section 2] and [34,
Section 5.2]. It is well known that if s ≥ 0, then Mult(Hs) = H ∞, the algebra of all bounded analytic functions
on the unit disc, endowed with the supremum norm. This can be deduced, for example, from [44, Proposition
26 (ii)].
In particular, the C*-envelope of Mult(Hs) is commutative, so that the identity representation of
Mult(Hs) on Hs is not a boundary representation. On the other hand, if s < 0, then the identity representation
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
11
of Mult(Hs) on Hs is a boundary representation. This follows, for instance, from Corollary 2 in Section 2 of [5]
and its proof.
We use the results above to prove that, in the rectangular case, the identity representation is always a
boundary representation.
Corollary 2.20. If s < t, then the identity representation is a boundary representation of Mult(Hs, Ht).
Proof. We verify that the pair (Hs, Ht) satisfies the assumptions of Proposition 2.19. It is clear that Hs contains
the constant function 1. By the remark above, Mult(Hs, Ht) contains the polynomials, and is therefore dense
in K. Moreover, since kt(0, w) = 1 for all w ∈ D, the space Ht is irreducible. Finally, since
zn2
Ht
zn2
Hs
= (n + 1)s−t,
which tends to zero as n → ∞, the inclusion Hs ⊂ Ht is compact, so that M1 ∈ Mult(Hs, Ht) is a compact
operator. Therefore, the result follows from Proposition 2.19.
(cid:3)
3. Operator spaces and rectangular matrix convex sets
In the following we will use notation from [22] and [47].
In particular, if V and V ′ are vector spaces in
duality via a bilinear map h·, ·i, x = [xij ] ∈ Mn,m (V ), and ψ = [ψαβ] ∈ Mr,s (V ′), then we let hhx, ψii be the
element [hxij , ψαβi] of Mnr,ms (C), where the rows of hhx, ψii are indexed by (i, α) and the columns of hhx, ψii
are indexed by (j, β). We also let ψ(n,m) be the map hh·, ψii : Mn,m (V ) → Mnr,ms (C).
3.1. Rectangular matrix convex sets.
Definition 3.1. A rectangular matrix convex set in a vector space V is a collection K = (Kn,m) of subsets
of Mn,m (V ) with the property that for any αi ∈ Mni,n (C) and βi ∈ Mmi,m (C) and vi ∈ Kni,mi for 1 ≤ i ≤ ℓ
such that
kα∗
1α1 + · · · + α∗
ℓ αℓk kβ∗
1 β1 + · · · + β∗
ℓ βℓk ≤ 1
one has that α∗
1v1β1 + · · · + α∗
ℓ vℓβℓ ∈ Kn,m.
When V is a topological vector space, we say that K is compact if Kn,m is compact for every n, m. The
following characterization of rectangular matrix convex sets can be easily verified using Proposition 2.1 and the
fact that any finite-dimensional representation of Mn,m (C) as a TRO is unitarily conjugate to a finite direct
sum of copies of the identity representation [11, Lemma 3.2.3].
Lemma 3.2. Suppose that K = (Kn,m) where Kn,m ⊂ Mn,m (V ). The following assertions are equivalent:
(1) K is a rectangular convex set;
(2) x ⊕ y ∈ Kn+m,r+s for any x ∈ Kn,r and y ∈ Km,s, and α∗xβ ∈ Kr,s for any x ∈ Kn,m, α ∈ Mn,r (C)
and β ∈ Mm,s (C) with kα∗αk kβ∗βk ≤ 1;
(3) x ⊕ y ∈ Kn+m,r+s for any x ∈ Kn,r and y ∈ Km,s, and (σ ⊗ idV ) [Kn,m] ⊂ Kr,s for any completely
contractive map σ : Mn,m (C) → Mr,s (C).
It is clear that, if K is a rectangular matrix convex set, then (Kn,n) is a matrix convex set in the sense of
[49]. Furthermore if K and T are rectangular matrix convex sets such that Tn = Kn for every n ∈ N then
Tn,m = Kn,m for every n, m ∈ N. If S = (Sn,m) is a collection of subsets of a (topological) vector space V , the
(closed) rectangular matrix convex hull of S is the smallest (closed) rectangular matrix convex set containing
S.
Example 3.3. Suppose that X is an operator space. Set Kn,m to be space of completely contractive maps from
X to Mn,m (C). Then CBall(X) = (Kn,m) is a rectangular matrix convex set.
3.2. The rectangular polar theorem. Suppose that V and V ′ are vector spaces in duality. We endow both
V and V ′ with the weak topology induced from such a duality. Let S = (Sn,m) be a collection of subsets
Sn,m ⊂ Mn,m (V ). We define the rectangular matrix polar Sρ to be the closed rectangular matrix convex subset
of V ′ such that f ∈ Sρ
n,m if and only if khhv, f iik ≤ 1 for every r, s ∈ N and every v ∈ Sr,s. The same proof as
[22, Lemma 5.1] shows that f ∈ Sρ
n,m if and only if khhv, f iik ≤ 1 for every v ∈ Sn,m.
If A ⊂ V , then its absolute polar A◦ is the set of f ∈ V ′ such that hv, f i ≤ 1 for every v ∈ A. The classical
bipolar theorem asserts that the absolute bipolar A◦◦ is the closed absolutely convex hull of A [15, Theorem
8.1.12]. We will prove below the rectangular analog of this fact. The proof is analogous to the one of [22,
Theorem 5.4].
12
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
Theorem 3.4. If S = (Sn,m) is a collection of subsets Sn,m ⊂ Mn,m (V ), then the rectangular matrix bipolar
Sρρ is the closed rectangular matrix convex hull of S.
The proof of [21, Theorem C] shows that if K is a rectangular matrix convex set in a vector space V , and F
is a linear functional on Mn,m (V ) satisfying(cid:13)(cid:13)F Kn,m(cid:13)(cid:13) ≤ 1, then
r, s ∈ N, α ∈ Mn,r (C), β ∈ Mm,s (C), and v ∈ Mr,s (V ), and
(1) there exist states p on Mn (C) and q on Mm (C) such that F (α∗vβ)2 ≤ p (α∗α) q (β∗β) for every
(2) there exist matrices γ ∈ Mn2,1 (C), δ ∈ Mm2,1 (C), and a map ϕ : V → Mn,m (C), such that F (w) =
γ∗ hhw, ϕii δ for every w ∈ Mn,m (W ) and khhw, ϕiik ≤ 1 for every r, s ∈ N and w ∈ Kr,s.
From this one can easily deduce the following proposition, which gives the rectangular matrix bipolar theorem
as an easy consequence.
Proposition 3.5. Suppose that V and V ′ are vector spaces in duality, and K is a compact rectangular convex
space in V . If v0 ∈ Mn,m (V ) \Kn,m, then there exists ϕ ∈ K ρ
n,m such that khhv0, ϕiik > 1.
Proof. By the classical bipolar theorem there exists a continuous linear functional F on Mn,m (V ) such that
n,m and contractive γ ∈ Mn2×1 (C) and
δ ∈ Mm2×1 (C) such that F (v) = γ∗ hhv, ϕii δ. Thus we have khhv0, ϕiik ≥ kγ∗ hhv0, ϕii δk = kF (v0)k > 1. (cid:3)
(cid:13)(cid:13)F Kn,m(cid:13)(cid:13) ≤ 1 and F (v0) > 1. By the remarks above there exists ϕ ∈ K ρ
3.3. Representation of rectangular convex sets. Suppose that K is a rectangular matrix convex set in a
vector space V . A rectangular matrix convex combination in a rectangular convex set K is an expression of the
form α∗
1v1β1 +· · ·+α∗
ℓ αℓ = 1,
1 β1 + · · ·+ β∗
and β∗
ℓ βℓ = 1. A proper rectangular matrix convex combination is a rectangular convex combination
1v1β1 + · · · + α∗
α∗
ℓ vℓβℓ where furthermore α1, . . . , αℓ and β1, . . . , βℓ are right invertible. Observe that these
notions are a particular instance of the notions of (proper) rectangular operator convex combination introduced
in Subsection 2.4.
ℓ vℓβℓ for vi ∈ Kni,mi, αi ∈ Mni,n (C), and βi ∈ Mmi,m (C) such that α∗
1α1 +· · ·+α∗
Definition 3.6. A rectangular matrix affine mapping from a rectangular convex set K to a rectangular convex
set T is a sequence θ of maps θn,m : Kn,m → Tn,m that preserves rectangular matrix convex combinations.
When K and T are compact rectangular convex sets, we say that θ is continuous (respectively, a homeo-
morphism) when θn,m is continuous (respectively, a homeomorphism) for every n, m ∈ N.
Given a compact rectangular matrix convex set K we let Aρ(K) be the complex vector space of continuous
rectangular matrix affine mappings from K to CBall (C). Here CBall (C) is the compact rectangular matrix
convex set defined as in Example 3.3, where C is endowed with its canonical operator space structure. The space
Aρ(K) has a natural operator space structure where Mn,m (Aρ(K)) is identified isometrically with a subspace
of C (Kn,m, Mn,m (C)) endowed with the supremum norm. More generally if Y is any operator space, then
we define Aρ(K, Y ) to be the operator space of continuous rectangular affine mappings from K to CBall(Y ).
Observe that Mn,m (Aρ(K)) is completely isometric to Aρ(K).
Starting from the operator space Aρ(K) one can consider the compact rectangular matrix convex set
CBall(Aρ(K)′) as in Example 3.3. Here Aρ(K)′ denotes the dual space of the operator space Aρ(K), en-
dowed with its canonical operator space structure. There is a canonical rectangular matrix affine mapping θ
from K to CBall(Aρ(K)′) given by point evaluations. It is clear that this map is injective. It is furthermore
surjective in view of the rectangular bipolar theorem. The argument is similar to the one of the proof of [47,
Proposition 3.5]. This shows that the map θ is indeed a rectangular matrix affine homeomorphism from K
onto CBall(Aρ(K)′). This implies that the assignment X 7→ CBall(X ′) is a 1:1 correspondence between op-
erator spaces and rectangular convex sets. It is also not difficult to verify that this correspondence is in fact
an equivalence of categories, where morphisms between operator spaces are completely contractive linear maps,
and morphisms between rectangular convex sets are continuous rectangular matrix affine mappings.
3.4. The rectangular Krein-Milman theorem. The notion of (proper) rectangular convex combination
yields a natural notion of extreme point in a rectangular convex set. An element v of a rectangular convex set K
1v1β1 +· · ·+α∗
is a rectangular matrix extreme point if for any proper rectangular convex combination α∗
ℓ vℓβℓ = v
for vi ∈ Kni,mi one has that, for every 1 ≤ i ≤ ℓ, ni = n, mi = m, and vi = u∗
i vwi for some unitaries
ui ∈ Mn (C) and wi ∈ Km. We now observe that the notion of rectangular extreme point coincides with the
notion of rectangular operator extreme operator state from Definition 2.10. The argument is borrowed from
the proof of [24, Theorem B].
Lemma 3.7. Suppose that X is an operator space, K = CBall(X), and φ ∈ Kn,m. Then φ is a rectangular
matrix extreme point of K if and only if it is a rectangular operator extreme operator state of X.
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
13
Proof. It is clear that a rectangular operator extreme point is a rectangular matrix extreme point. We prove
the converse implication. Suppose that φ is a rectangular matrix extreme point. Let φ = α∗
ℓ φℓβℓ
be a proper rectangular matrix convex combination, where φi ∈ Kni,mi for i = 1, 2, . . . , ℓ. By assumption, we
have that ni = n and mi = m for i = 1, 2, . . . , m, and there exist unitaries ui ∈ Mn (C) and wi ∈ Mm (C) such
that φi = u∗
i φwi for i = 1, 2, . . . , ℓ. Therefore we have that
1φ1β1 + · · · + α∗
φ = (u1α1)∗ φ (w1β1) + · · · + (uℓαℓ)∗ φℓ (wℓβℓ) .
(1)
Define R ⊂ Mm+m (C) to be the range of S(φ). Observe that it follows from the fact that φ is a rectangular
extreme point that the commutant of R is one-dimensional. Set
Ai =(cid:20)uiαi
0
0
wiβi(cid:21)
for i = 1, 2, . . . , n. Define the unital completely positive map Ψ : Mn+m (C) → Mn+m (C), z 7→ A∗
1zA1 + · · · +
A∗
ℓ zAℓ. By Equation (1) we have that Ψ(z) = z for every z ∈ R. It follows from this and [4, Theorem 2.11]
that Ψ(z) = z for every z ∈ Mn+m (C). By the uniqueness statement in the Choi's representation of a unital
completely positive map [13], we deduce that there exist λi ∈ C such that Ai = λi1 for i = 1, 2, . . . , ℓ. Therefore
i φiβi = (uiαi)∗ φ (wiβi) = λi2 φ. This
i αi = (uiαi)∗ (uiαi) = λi2 1, β∗
α∗
concludes the proof that φ is a rectangular operator extreme point.
i βi = (wiβi)∗ (wiβi) = λi2 1, and α∗
(cid:3)
We denote by ∂ρK = (∂ρKn,m) set of rectangular matrix extreme points of K. Recall that the Krein-Milman
theorem asserts that, if K ⊂ V is a compact convex subset of a topological vector space V , then K is the closed
convex hull of the set of its extreme points. The following is the natural analog of the Krein-Milman theorem
for compact rectangular matrix convex sets. The proof is analogous to the proof of the Krein-Milan theorem
for compact matrix convex sets [47, Theorem 4.3].
Theorem 3.8. Suppose that K is a compact rectangular convex set. Then K is the closed rectangular matrix
convex hull of ∂ρK.
Proof. Suppose that K is a compact rectangular convex set.
In view of the representation theorem from
Subsection 3.3, we can assume without loss of generality that K = CBall(X ′) for some operator space X. We
will assume that X is concretely represented as a subspace of B(H) for some Hilbert space H. We will also
canonically identify Mn,r(X ′) with the space of bounded linear functionals on Mn,r(X).
Fix n, m ∈ N. Let X be the space of operators of the form
(cid:20)λI ⊕n
y∗
x
µI ⊕m(cid:21)
λIrn
ϕ(n,m)(y)∗
ϕ(n,m)(x)
µIms (cid:21)
(cid:20)λI ⊕n
y∗
x
µI ⊕n(cid:21) 7→(cid:20)
ξ ⊙ η :=(cid:20)In ⊗ ξ
0
0
Im ⊗ η(cid:21) .
for λ, µ ∈ C and x, y ∈ Mn,m(X), where I ⊕n and I ⊕m are the identity operator on, respectively, the n-fold and
m-fold Hilbertian sum of H by itself. If ϕ ∈ Mr,s(X ′), then we denote by ϕ ∈ Mnr+ms( X ′) the element defined
by
where Irn and Ims denote the identity rn × rn and ms × ms matrices. If ξ ∈ Mr,n (C) and η ∈ Ms,m (C) we
also set
We let ∆n,m be the set of elements of Mn2+m2( X ′) of the form (ξ ⊙ η)∗ ϕ (ξ ⊙ η) for r, s ∈ N, ϕ ∈ Kr,s,
ξ ∈ Mr,n (C) and η ∈ Ms,m (C) such that kξk2 = kηk2 = 1. It is not difficult to verify as in [47, §4] that one
can assume without loss of generality that r ≤ n, s ≤ m, and ξ, η are right invertible. The computation below
shows that ∆n,m is convex. If t1, t2 ∈ [0, 1] are such that t1 + t2 = 1 then
t1 (ξ1 ⊙ η1)∗ ϕ1 (ξ1 ⊙ η1) + t2 (ξ1 ⊙ η1)∗ ϕ2 (ξ2 ⊙ η2) = (ξ ⊙ η)∗ ϕ (ξ ⊙ η)
where
ξ =(cid:20)t1ξ1
t2ξ2(cid:21) , η =(cid:20)t1η1
t2η2(cid:21) , and ϕ =(cid:20)ϕ1
0 ϕ2(cid:21) .
0
Thus ∆n,m is a compact convex subset of the space of unital completely positive maps from X to Mn2+m2 (C).
Consider now an element (ξ ⊙ η)∗ ϕ (ξ ⊙ η) of ∆n,m, where ξ ∈ Mr,n (C) and η ∈ Ms,m (C) are right invertible
and ϕ ∈ Kn,m. Assume that (ξ ⊙ η)∗ ϕ (ξ ⊙ η) is an extreme point of ∆n,m. We claim that this implies that ϕ
is a rectangular extreme point of K. Indeed suppose that, for some sk, rk ∈ N, ϕk ∈ Krk,sk , δk ∈ Msk,s (C), and
14
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
γk ∈ Mrk,r (C), γ∗
we have that
1 ϕ1δ1 + · · · + γ∗
ℓ ϕℓδℓ is a proper rectangular convex combination in K that equals ϕ. Then
(ξ ⊙ η)∗ ϕ (ξ ⊙ η) = (γ1ξ ⊙ δ1η)∗ ϕ1 (γ1ξ ⊙ δ1η) + · · · + (γ1ξ ⊙ δ1η)∗ ϕℓ (γ1ξ ⊙ δ1η) .
Let
for k = 1, 2, . . . , ℓ. Observe that
tk = k(γkξ ⊙ δkη)k2
Therefore, if we set
t2
k =
1
r + s
ℓXk=1
ℓXk=1
(rTr (ξ∗γ∗
kγkξ) + sTr (η∗δ∗
kδkη)) = 1.
ψk := t−2
k (ξγ1 ⊙ ηδ1)∗ ϕk (ξγ1 ⊙ ηδ1)
ℓ ψℓ = (ξ ⊙ η)∗ ϕ (ξ ⊙ η). Since
for k = 1, 2, . . . , ℓ, we obtain elements ψ1, . . . , ψℓ of ∆n,m such that t2
by assumption (ξ ⊙ η)∗ ϕ (ξ ⊙ η) is an extreme point of ∆n,m, we can conclude that ψk = (ξ ⊙ η)∗ ϕ (ξ ⊙ η) for
k = 1, 2, . . . , ℓ. The fact that ξ and η are right invertible now easily implies that rk = r, sk = s, γ∗
i 1,
δ∗
i δi = t2
i ϕ. This conclude the proof that ϕ is a rectangular extreme point of K.
1ψ1 + · · · + t2
i 1, and γ∗
i ϕiδi = t2
i γi = t2
We are now ready to conclude the proof that K is the rectangular convex hull of ∂ρK.
rectangular bipolar theorem, it is enough to prove that if n, m ∈ N and z ∈ Mn,m(X) are such that(cid:13)(cid:13)ϕ(n,m)(z)(cid:13)(cid:13) ≤
1 for every r ≤ n, s ≤ m, and ϕ ∈ ∂ρKs,t, then(cid:13)(cid:13)ψ(n,m)(z)(cid:13)(cid:13) ≤ 1 for every ψ ∈ Kn,m. If x ∈ Mn,m(X) then we
let x be the element
In view of the
x
(cid:20)I ⊕n
x∗
I ⊕m(cid:21)
of X. Observe that if ϕ ∈ Mr,s(X ′) and x ∈ Mn,m(X), then
If furthermore ξ ∈ Mr,n (C) and η ∈ Ms,m (C) then
ϕ(n,m)(x)∗
Inr
ϕ(n,m)(x) =(cid:20)
(ξ ⊙ η)∗ ϕ(x) (ξ ⊙ η) ="
ϕ(n,m)(x)
Ims
(cid:21) ∈ Mnr+ms (C) .
In ⊗ ξ∗ξ
(ξ∗ϕη)(n,m) (x)
(ξ∗ϕη)(n,m) (x)∗
Im ⊗ η∗η
# .
Let now (ξ ⊙ η)∗ ϕ (ξ ⊙ η) be an extreme point of ∆n,m, where ξ ∈ Mr,n (C) and η ∈ Ms,m (C) are right invertible
and such that kξk2 = kηk2 = 1, and ϕ ∈ ∂Kr,s. By assumption we have that(cid:13)(cid:13)(cid:0)idMn,m(C) ⊗ ϕ(cid:1) (z)(cid:13)(cid:13) ≤ 1. Thus
by [40, Lemma 3.1] we have
ϕ(n,m)(z)
Inr
(cid:20)
ϕ(n,m)(z)∗
Ims
(cid:21) ≥ 0
and hence
(ξ ⊙ η)∗ ϕ(z) (ξ ⊙ η) ="
In ⊗ ξ∗ξ
(ξ∗ϕη)(n,m) (z)
(ξ∗ϕη)(n,m) (z)∗
Im ⊗ η∗η
# ≥ 0.
It follows from this and the classical Krein-Milman theorem that ψ(z) ≥ 0 for any ψ ∈ ∆n,m. Let us fix
ϕ ∈ Kn,m. If ξ = In and η = Im then (ξ ⊙ η)∗ ϕ (ξ ⊙ η) ∈ ∆n,m and
(ξ ⊙ η)∗ ϕ(z) (ξ ⊙ η) ="
In ⊗ ξ∗ξ
(ξ∗ϕη)(n,m) (z)
(ξ∗ϕη)(n,m) (z)∗
Im ⊗ η∗η
# =(cid:20)
Inr
ϕ(n,m)(z)∗
(cid:0)ϕ(n,m)(cid:1) (z)
Ims
(cid:21) ≥ 0.
Kn,m, the proof is concluded.
This implies again by [40, Lemma 3.1] that (cid:13)(cid:13)ϕ(n,m)(z)(cid:13)(cid:13) ≤ 1. Since this is valid for an arbitrary element of
Remark 3.9. The proof of Theorem 3.8 shows something more precise: if K is a rectangular convex set, then
for every n, m ∈ N, Kn,m is equal to the closed rectangular convex hull of Kr,s for r ≤ n and s ≤ m.
(cid:3)
The following is an immediate corollary of the rectangular Krein-Milman theorem as formulated in Remark
3.9.
Corollary 3.10. Suppose that X is an operator space, and K = CBall(X ′) is the corresponding compact
rectangular matrix convex set. If n, m ∈ N and x ∈ Mn,m(X), then kxk is the supremum of(cid:13)(cid:13)ϕ(n,m)(x)(cid:13)(cid:13) where
ϕ ranges among all the rectangular extreme points of Kr,s for r ≤ n and s ≤ m.
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
15
4. Boundary representations and the C*-envelope of a matrix-gauged space
4.1. Selfadjoint operator spaces. By a concrete selfadjoint operator space we mean a closed selfadjoint
subspace X of B(H). Any selfadjoint operator space is endowed with a canonical involution, matrix norms,
and matrix positive cones inherited from B(H).
An (abstract) matrix-ordered matrix-normed ∗-vector space -- see [42, Subsection 3.1] -- is a vector space V
endowed with
• a conjugate-linear involution v 7→ v∗,
• a complete norm in Mn (V ) for every n ∈ N,
• a distinguished positive cone Mn (V )+ ⊂ Mn (V )
such that, for every n, k ∈ N, x ∈ Mn(X), and a, b ∈ Mn,k (C),
(1) Mn (V )+ is proper, i.e. Mn (V )+ ∩ (−Mn (V )+) = {0},
(2) Mn (V )+ is closed in the topology induced by the norm,
(3) ka∗xbk ≤ kak kxk kbk, and
(4) when a∗xa ∈ Mk (V )+, when x ∈ Mn (V )+.
A matrix-ordered matrix-normed ∗-vector space V is normal
if, for every n ∈ N and x, y, z ∈ Mn (V ),
x ≤ y ≤ z implies that kyk ≤ max {kxk , kzk}. It is essentially proved in [48] -- see also [42, Theorem 3.2] and
[43, Theorem 5.6] -- that a matrix-ordered matrix-normed ∗-vector space V is normal if and only if there exists a
completely positive completely isometric selfadjoint linear map φ : V → B(H), where H is a Hilbert space and
the space B(H) of bounded linear operators on H is endowed with its canonical matrix-ordered matrix-normed
∗-vector space structure.
4.2. Matrix-gauged spaces. Suppose that V is a real vector space. A gauge over V is a subadditive and
positively-homogeneous function ν : V → [0, +∞). The conjugate gauge ν is defined by ν(x) = ν(−x). The
seminorm kxkν corresponding to a gauge is given by kxkν = max {ν(x), ν(x)}. A gauge is proper if the seminorm
k·kν is a norm. The positive cone associated with a gauge ν is the set V+,ν = {x ∈ V : ν(x) = 0}.
The following notion is considered in [42] under the name of L∞-matricially ordered vector space.
Definition 4.1. A matrix-gauged space is a ∗-vector space V endowed with a sequence of proper gauges νn :
Mn (V )sa → [0, +∞) for n ∈ N with the property that, for every n, k ∈ N, x ∈ Mn (V ), y ∈ Mk (V ), and
a ∈ Mn,k (C), one has that
and
νk (a∗xa) ≤ kak2 νn(x)
νn+k (x ⊕ y) = max {νn(x), νk(y)} .
A linear map φ : V → W between matrix-gauged spaces is completely gauge-contractive if it is selfadjoint
and ν(cid:0)φ(n)(x)(cid:1) ≤ ν(x) for every n ∈ N and x ∈ Mn(X)sa, and completely gauge-isometric if it is selfadjoint
and ν(cid:0)φ(n)(x)(cid:1) = ν(x) for every n ∈ N and x ∈ Mn(X)sa; see [42, Definition 3.11]. Matrix-gauged spaces
naturally form a category, where the morphisms are the completely gauge-contractive maps, and isomorphism
are completely gauge-isometric surjective maps.
In the following we will consider matrix-gauged spaces as
objects in this category. By [42, Corollary 3.10], every matrix-gauged space is completely gauge-isometrically
isomorphic to a concrete selfadjoint operator space.
Any matrix-gauged space V has a canonical normal matrix-ordered matrix-normed ∗-vector space struc-
ture, obtained by considering the gauge norms and the gauge cones associated with the given matrix-gauges.
Conversely, suppose that V is a normal matrix-ordered matrix-normed ∗-vector space. Letting νn(x) be the
distance of x from −Mn (V )+ for every x ∈ Mn (V )sa defines a canonical matrix-gauged structure on X. This
matrix-gauge structure induces the original matrix-order and matrix-norms on V that one started from; see
[42, Proposition 3.5]. Furthermore a selfadjoint linear map φ : V → B(H) is completely positive and com-
pletely contractive if and only if it is completely gauge-contractive with respect to these specific matrix-gauges.
However, there might be different matrix-gauges on V that induce the same matrix-order and matrix-norms on
V .
Suppose now that S is an operator system with order unit 1. Then, in particular, S is a normal matrix-
ordered matrix-normed ∗-vector space. Furthermore, it admits a unique matrix-gauge structure compatible
with its matrix-order and matrix-norms. These matrix-gauges are defined by νn(x) = inf {t > 0 : x ≤ t1} for a
selfadjoint x ∈ S. Uniqueness can be deduced from Arveson's extension theorem [12, Theorem 1.6.1], as proved
in [43, Theorem 6.9]. In the following we will regard an operator system as a matrix-gauge space with such
16
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
canonical matrix-gauges. A unital selfadjoint linear map between operator systems is completely positive if and
only if it is completely gauge-contractive, and completely isometric if and only if it is completely gauge-isometric.
It is proved in [42, Subsection 3.3] that any matrix-gauged space W admits a completely gauge isometric
embedding as a subspace of codimension 1 into an operator system W †, called the unitization of W , that
satisfies the following universal property: any completely gauge-contractive map from W to an operator system
V admits a unique extension to a unital completely positive map from W † to V . The unitization W † of W is
uniquely characterized by the above. If W is a normal matrix-ordered matrix-normed space, then we define the
unitization of W to be the unitization of W endowed with the canonical matrix-gauges described above.
Suppose that A is a (not necessarily unital) C*-algebra. Then A is endowed with canonical matrix-gauges,
obtained by setting νn(x) = kx+k for a selfadjoint x ∈ A, where x+ denotes the positive part of x. In the
following we will consider a C*-algebra as a matrix-gauged space with these canonical matrix-gauges. It follows
from the unitization construction that any matrix-gauged space admits a completely gauge-isometric embedding
into B(H). The following result can also be found in [12, Proposition 2.2.1] with a different proof.
Lemma 4.2. Suppose that A ⊂ B(H) is a C*-algebra such that the identity 1 of B(H) does not belong to A.
Let Y be an operator system. Then for any completely positive completely contractive map φ : A → Y there
exists a unital completely positive map ψ : span {A, 1} → Y extending φ.
Proof. Let Z := span {A, 1} ⊂ B(H) and assume that Y ⊂ B(L) for some Hilbert space L. We have to prove
that if z = x + α1 ∈ Mn (Z) for α ∈ Mn (C) is positive, then φ(n)(x) + α1 ∈ Mn(Y ) is positive. Since 1 /∈ A, we
have that α is positive. Without loss of generality, we can assume that α is invertible. After replacing z with
α− 1
2 we can assume that α = 1. By Stinespring's theorem [8, Theorem II.6.9.7], there exist a Hilbert space
K, a *-homomorphism π : A → B(K), and a linear map v : L → K such that kvk = 1 and φ(x) = v∗π(x)v for
every x ∈ A. Observe that π extends to a unital ∗-homomorphism from Z into B(L), which we still denote by
π. Let v(n) : H (n) → K (n) be the map v ⊕ · · · ⊕ v. Then we have that
2 zα− 1
φ(n)(x) + 1 = v(n)∗π(n)(x)v(n) + 1
≥ v(n)∗π(n)(x)v(n) + v(n)∗π(n)(1)v(n) = v(n)∗π(n) (x + 1) v(n) ≥ 0.
This concludes the proof.
(cid:3)
It follows from the previous lemma that the unitization of a C*-algebra A as a matrix-gauged space coincides
with the unitization of A as a C*-algebra; see also [48, Corollary 4.17]. Furthermore, it follows from Lemma 4.2,
[43, Theorem 6.9], and Arveson's extension theorem that a C*-algebra admits unique compatible matrix-gauges.
One can then deduce from [42, Theorem 3.16] that a linear map between C*-algebras or operator systems is
completely gauge-contractive if and only if it is completely positive contractive.
4.3. The injective envelope of a matrix-gauged space. We say that a matrix-gauged space is injective if
it is injective in the category of matrix-gauged spaces and completely gauge-contractive maps. Theorem 3.14
of [42] shows that B(H) is an injective matrix-gauged space when endowed with its canonical matrix-gauges.
It follows from this that the unitization functor W 7→ W † is an injective functor from the category of matrix-
gauged spaces and gauge-contractive maps to the category of operator systems and unital completely positive
maps.
Our goal now is to show that any injective matrix-gauged space is (completely gauge-isometrically isomorphic
to) a unital C*-algebra. This is a generalization of a theorem of Choi and Effros see [40, Theorem 15.2].
Proposition 4.3. Let X be an injective matrix-gauged space. Then X is completely gauge-isometrically iso-
morphic to a unital C*-algebra.
Proof. We may assume that X ⊂ B(H) is concretely represented as a selfadjoint operator space. Since X is
injective, there exists a gauge-contractive and hence completely contractive and completely positive projection
Φ : B(H) → X. We define the Choi-Effros product on X by
As in the proof of [40, Theorem 15.2], one shows that
x ·Φ y = Φ(xy).
Φ(Φ(a)x) = Φ(ax)
and Φ(xΦ(a)) = Φ(xa)
holds for all x ∈ X and all a ∈ B(H). Indeed, the proof only requires the Schwarz inequality for unital completely
positive maps, which remains valid for completely positive completely contractive maps. In particular, we see
that e := Φ(IH ) ∈ X is a unit for the Choi-Effros product. Moreover, the proof of [40, Theorem 15.2] shows
that (X, ·Φ), endowed with the norm and involution of B(H), is a C*-algebra.
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
17
It is clear from the above that the identity map from X onto (X, ·Φ) is an isometry. To see that it is an order
isomorphism, suppose that x ∈ X is positive with respect to the order on B(H). Then cIH − x ≤ c for all
c ≥ x, so since Φ is contractive ce − x ≤ c for all c ≥ x, thus x is a positive element of the C*-algebra
(X, ·Φ). Conversely, if x is positive in the C*-algebra (X, ·Φ), then there exists y ∈ X such that x = y∗ ·Φ y,
hence x = Φ(y∗y) is positive in B(H). Moreover, the argument at the end of the proof of [40, Theorem 15.2]
shows that Mn(X), endowed with the Choi-Effros product ·Φ(n) , is the C*-tensor product of (X, ·Φ) with Mn(C).
By the above, the identity map is an isometry and an order isomorphism between Mn(X) ⊂ Mn(B(H)) and
(Mn(X), ·Φ). Therefore, the identity map from X onto (X, ·Φ(n) ) is a selfadjoint complete isometry and complete
order isomorphism.
To see that the identity map from X onto (X, ·Φ) is in fact a complete gauge isometry, observe that Φ is a
unital completely positive map from B(H) onto (X, ·Φ) by the preceding paragraph, so it is completely gauge
contractive. Conversely, if x ∈ Mn(X) is self-adjoint and satisfies x+ ≤ 1, where the positive part is taken in
the C*-algebra (Mn(X), ·Φ), then x ≤ e(n) in (Mn(X), ·Φ), hence x ≤ ICn ⊗ IH in Mn(B(H)) by the preceding
paragraph, so that the identity map from (X, ·Φ) to X is completely gauge contractive as well.
(cid:3)
In particular, we see that every injective matrix-gauged space is (completely gauge-isometrically isomorphic
to) an injective operator system. Conversely, since the unitization functor is injective, every operator system
that is injective in the category of operator systems and unital completely positive maps is also injective as a
matrix-gauged space, when endowed with the unique compatible matrix-gauge structure.
The usual proof of the existence of the injective envelope of an operator system yields the existence of a gauge
analog of Hamana's injective envelope of operator spaces. Let us say that a gauge-extension of a matrix-gauged
space X is a pair (Y, i) where Y is a matrix-gauged space and i : X → Y is a completely gauge-isometric map.
As in the case of operator systems, we say that such a gauge-extension is:
(1) rigid if the identity map of Y is the unique gauge-contractive map φ : Y → Y such that φ ◦ i = i;
(2) essential if whenever u : Y → Z is a gauge-contractive map to a matrix-gauged space Z such that u ◦ i
is a completely gauge-isometric, then u is a completely gauge-isometric;
(3) an injective envelope if Y is injective, and there is no proper injective subspace of Y that contains X.
The same proof as [10, Lemma 4.2.4] shows that if X is a matrix-gauged space, and (Y, i) is a gauge-extension
of X such that Y is injective, then the following assertions are equivalent: 1) (Y, i) is an injective envelope of
X; 2) (Y, i) is essential; 3) (Y, i) is rigid. To this purpose one can consider the gauge analog of the notion of
projections and seminorms from [10, Subsection 4.2.1].
Suppose that W is a matrix-gauged space, and X is a selfadjoint subspace of W . A completely gauge-
contractive X-projection on W is an idempotent completely gauge-contractive map u : W → W that restricts
to the identity on X. A gauge X-seminorm on W is a seminorm of the form p(x) = ku(x)k for some completely
gauge-contractive X-projection u on W . One can define an order on completely gauge-contractive X-projections
by u ≤ v if and only if u ◦ v = v ◦ u = u, while gauge X-seminorms are ordered by pointwise comparison. The
same proof as [10, Lemma 4.2.2] shows that any gauge X-seminorm majorizes a minimal gauge X-seminorm,
and if p is a minimal gauge X-seminorm and u : W → W is a completely gauge-contractive map that restricts
to the identity on X, then u is a minimal gauge X-projection. To this purpose, it is enough to observe that the
set of completely gauge-contractive selfadjoint maps from W to B(H) is closed in the weak* topology of the
space of CB (W, B(H)) of completely bounded maps from W to B(H). Indeed φ : W → B(H) is completely
gauge-contractive if and only if it is selfadjoint and (cid:10)φ(n)(x)ξ, ξ(cid:11) ≤ ν(x) for every n ∈ N, ξ ∈ H ⊕n, and
x ∈ Mn(W )sa.
The proof of [10, Lemma 4.2.4] can now be easily adapted to prove the claim above, by replacing X-projections
with gauge X-projections and X-seminorms with gauge X-seminorms. Similarly the same proof as [10, Theorem
4.2.6] shows that if a matrix-gauged space X is contained in an injective matrix-gauged space W , then there
exists an injective envelope X ⊂ Z ⊂ W . Furthermore the injective envelope of X is essentially unique.
We denote by I(X) the injective envelope of a matrix-gauged space X, and we identify X with a selfadjoint
subspace of I(X). It is clear that, when X is an operator system endowed with its canonical matrix-gauges,
the injective envelope of X as a matrix-gauged space coincides with the injective envelope of X as an operator
system (endowed with the canonical matrix-gauges). Furthermore, it is a consequence of Proposition 4.3 that
the unitization of a matrix-gauged space X is span {X, 1} ⊂ I(X), where 1 denotes the identity of the unital
C*-algebra I(X).
4.4. Boundary representations. Most fundamental notions in dilation theory admit straightforward versions
in the setting of matrix-gauged spaces. Suppose that X is a matrix-gauged space. An operator state on X
18
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
is a completely gauge-contractive map φ : X → B(H). We say that an operator state ψ : X → B(eH) is a
dilation of φ if there exists a linear isometry v : H → eH such that v∗ψ(x)v = φ(x) for every x ∈ X. It follows
from Stinespring's dilation theorem [8, Theorem II.6.9.7] that if A is a C*-algebra, then an operator state on A
admits a dilation which is a *-homomorphism. A dilation ψ of an operator state φ : x 7→ v∗ψ(x)v on X is trivial
if ψ(x) = vv∗ψ(x)vv∗ + (1 − vv∗)ψ(x) (1 − vv∗). We say that φ is maximal if it has no nontrivial dilation. As
in the case of operator systems, one can prove that an operator state φ : X → B(H) is maximal if and only if
for any dilation ψ of φ one has that kψ(x)ξk = kφ(x)ξk for every x ∈ X, and ξ ∈ H.
Suppose that X is a selfadjoint subspace of a C*-algebra A such that A is generated as a C*-algebra by
X. An operator state φ on X has the unique extension property if any completely positive contractive map
eφ : A → B(H) whose restriction to X coincides with φ is automatically a *-homomorphism. The same
argument as in the operator systems setting shows that an operator state is maximal if and only if it has the
unique extension property; see [6].
Definition 4.4. A boundary representation for a matrix-gauged space X ⊂ B(H) is an operator state φ : X →
B(H) with the property that any completely positive contractive map ψ : C∗(X) → B(H) extending X is an
irreducible representation of C∗(X).
In the following we will identify a boundary representation of X with its unique extension to an irreducible
representation of C∗(X). It follows from the remarks above that the notion of boundary representation does
not depend on the concrete realization of X as a selfadjoint space of operators. In the following we will assume
that A is a C*-algebra, and X ⊂ A is a selfadjoint subspace that generates A as a C*-algebra. We regard X as
a matrix-gauged space endowed with the matrix-gauges induced by A.
Proposition 4.5. Suppose that φ : X → B(H) is an operator state of X, and φ† : X † → B(H) is its canonical
unital completely positive extension to the unitization of X. If φ† is a boundary representation for X †, then φ
is a boundary representation for X.
Proof. Let Φ : A → B(H) be a completely positive contractive map extending φ. Extend Φ to a unital
completely positive Φ† : A† → B(H). Then since by assumption φ† is a boundary representation, we conclude
that Φ† is an irreducible representation for A†. Therefore ΦA is an irreducible representation of A. This
concludes the proof.
(cid:3)
The following result is then a consequence of Proposition 4.5 and [17, Theorem 3.1].
Theorem 4.6. Suppose that X is a matrix-gauged space. Then the matrix-gauges of X are completely de-
termined by the boundary representations of X. Precisely, if x ∈ Mn(X), then νn(x) is the supremum of
(cid:13)(cid:13)φ(n)(x)+(cid:13)(cid:13) where φ ranges among all the boundary representations of X.
Suppose that X is a matrix-gauged space, and φ : X → B(H) is an operator state. An operator convex
combination is an expression φ = α∗
nφnαn, where αi : H → Hi are linear maps, and φi : X →
B(Hi) are operator states for i = 1, 2, . . . , ℓ. Such a rectangular convex combination is proper if the αi's are
right invertible and α∗
1φ1α1 + · · · + α∗
i φiαi = λiφ for some λi ∈ [0, 1].
i αi = λi1 and α∗
1α1 + · · · + α∗
nαn = 1 and trivial if α∗
Definition 4.7. An operator state φ : X → B(H) is an operator extreme point if for any proper operator
convex combination φ = α∗
1φ1α1 + · · · + α∗
nφnαn is trivial.
Observe that the map φ 7→ φ† establishes a 1:1 correspondence between operator states on X and operator
states on the operator system X †. Furthermore this correspondence is operator affine in the sense that it
preserves operator convex combinations. The following proposition is then an immediate consequence of this
observation and (the proof of) [24, Theorem B].
Proposition 4.8. Suppose that φ : X → B(H) is an operator state, and let φ† : X † → B(H) be its unital
extension to the unitization of X. The following assertions are equivalent:
(1) φ is a pure element in the cone of completely gauge-contractive maps from X to B(H);
(2) φ is an operator extreme point;
(3) φ† is an operator extreme point.
The following corollary is an immediate consequence of Proposition 4.8, Proposition 4.5, and [17].
Corollary 4.9. Suppose that X is a matrix-gauged space, and φ is an operator state on X. If φ is operator
extreme, then φ admits a dilation to a boundary representation of X.
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
19
4.5. The C*-envelope of a matrix-gauged space . Suppose that X is a matrix-gauged space. A pair (A, i)
is a C*-cover if A is a C*-algebra and i : X → A is a completely gauge-isometric map whose range generates A
as a C*-algebra.
Definition 4.10. A C*-envelope (C∗
for any C*-cover (B, j) of X, there exists a *-homomorphism θ : B → C∗
e (X), i) of X is a C*-cover of X if it has the following universal property:
e (X) such that θ ◦ j = i.
It is clear that the C*-envelope of a matrix-gauged space, if it exists, it is essentially unique. We will prove
below that any matrix-gauged space has a C*-envelope. The proof is essentially the same as the one for the
existence of the C*-envelope of an operator system.
Suppose that X is an matrix-gauged space. Let X ⊂ I(X) be the injective envelope of X. By Proposition
4.3, I(X) is a unital C*-algebra. Let A be the C*-subalgebra of I(X) generated by X. As in the proof of [40,
Theorem 15.16], one sees that (i, A) where i : X → A is the inclusion map, is the C*-envelope of X.
It follows from the construction that the C*-envelope (as defined above) of an operator system regarded as
matrix-gauged space with its unique compatible matrix-gauges coincides with the usual notion of C*-envelope
of an operator system.
Alternatively, one can construct the C*-envelope of a matrix-gauged space using boundary representations,
as for the C*-envelope of an operator system. Indeed let X be a matrix-gauged space. Define ιe : X → B(H)
to be the direct sum of all the boundary representations for X, and then let A be the C*-subalgebra of B(H)
generated by the image of ιe. It follows from the unique extension property of boundary representations that
(ιe, A) is indeed the C*-envelope of X.
In particular, this construction shows that, for any C*-algebra B,
C∗
e (B) = B.
As in the case of operator systems and operator spaces, one can define a maximal or universal C*-algebra that
contains a given ordered operator space as a generating subset. Explicitly, the maximal C*-algebra C∗
max(X)
of an ordered operator space is a C*-cover (i, A) of X that has the following universal property: given any
other completely gauge-contractive map f : X → B, where B is a C*-algebra, there exists a *-homomorphism
θ : A → B such that θ ◦ i = f . In order to see that such a maximal C*-algebra exists, one can consider the
collection F of all completely gauge-contractive maps from X to Mn (C) for n ∈ N. Then let i be the direct
s∈F Mns (C) generated by
the image of i. The same proof as [35, Proposition 8] shows that such a C*-cover satisfies the required universal
property.
sum of the elements s : X → Mns (C) of F , and then A to be the C*-subalgebra ofL∞
4.6. Selfadjoint ordered operator spaces and compact matrix convex sets. We want to conclude by
observing that selfadjoint operator spaces are in canonical 1:1 correspondence with compact matrix convex sets
with a distinguished extreme point.
Suppose that K = (Kn) is a compact matrix convex set, and e ∈ K1 is a matrix extreme point. Define
A0(K, e) to be the set of continuous matrix-affine functions from K to (Mn(C))n∈N that vanish at e. Then
A0(K, e) is a selfadjoint subspace of codimension 1 of the operator system A(K).
Conversely suppose that X ⊂ B(H) is a selfadjoint operator space. Consider X as a normal matrix-ordered
and matrix-normed space with respect to the induced matrix-cones and matrix-norms, and let X † be the
unitization of X. Let, for n ∈ N, Kn be the space of completely positive completely contractive selfadjoint
maps from X to Mn (C), endowed with the topology of pointwise convergence. Observe that K = (Kn) is a
compact matrix-convex set, and Kn can be identified with the space of unital completely positive maps from
X † to Mn (C). Let e ∈ K1 be the zero functional on X. We have a canonical unital complete order isomorphism
X † ∼= A(K). Under this isomorphism X is mapped into A0(K, e). Since X has codimension 1 in X †, such an
isomorphism in fact maps X onto A0(K, e).
The above construction shows that one can identify the unitization of A0(K, e) with the operator system
A(K). Furthermore it is easy to see that the correspondence (K, e) 7→ A0(K, e) is a contravariant equivalence
of categories from the category of compact matrix convex sets K with a distinguished matrix extreme point
e ∈ K1, where morphisms are continuous matrix-affine maps that preserve the distinguished point, to the
category of selfadjoint operator spaces and completely positive completely contractive selfadjoint maps.
The commutative analog of the argument above establishes a correspondence between compact convex sets
with a distinguished extreme point and selfadjoint operator spaces that can be represented inside an abelian
C*-algebra (selfadjoint function spaces).
1. Jim Agler and John E. McCarthy, Pick interpolation and Hilbert function spaces, Graduate Studies in Mathematics, vol. 44,
American Mathematical Society, Providence, RI, 2002. (page 9)
References
20
ADAM H. FULLER, MICHAEL HARTZ, AND MARTINO LUPINI
2. Erik M. Alfsen and Edward G. Effros, Structure in real Banach spaces. I, Annals of Mathematics. Second Series 96 (1972),
98 -- 128. (page 1, 7)
, Structure in real Banach spaces. II, Annals of Mathematics. Second Series 96 (1972), 129 -- 173. (page 1)
3.
4. William Arveson, Subalgebras of C*-algebras II, Acta Mathematica 128 (1972), no. 1, 271 -- 308. (page 13)
5.
6.
, Subalgebras of C ∗-algebras. II, Acta Math. 128 (1972), no. 3-4, 271 -- 308. (page 8, 11)
, The noncommutative Choquet boundary, Journal of the American Mathematical Society 21 (2008), no. 4, 1065 -- 1084.
(page 3, 5, 18)
7. William B. Arveson, Subalgebras of C*-algebras, Acta Mathematica 123 (1969), no. 1, 141 -- 224. (page 4)
8. Bruce Blackadar, Operator algebras, Encyclopaedia of Mathematical Sciences, vol. 122, Springer-Verlag, Berlin, 2006. (page 16,
18)
9. David P. Blecher, The Shilov boundary of an operator space and the characterization theorems, Journal of Functional Analysis
182 (2001), no. 2, 280 -- 343. (page 8)
10. David P. Blecher and Christian Le Merdy, Operator algebras and their modules -- an operator space approach, London Mathe-
matical Society Monographs. New Series, vol. 30, Oxford University Press, Oxford, 2004. (page 2, 3, 6, 8, 17)
11. Dennis Bohle, K-theory for ternary structures, Ph.D. thesis, Westfalische Wilhelms-Universitat Munster, 2011. (page 2, 11)
12. Nathanial P. Brown and Narutaka Ozawa, C*-algebras and finite-dimensional approximations, Graduate Studies in Mathemat-
ics, vol. 88, American Mathematical Society, Providence, RI, 2008. (page 15, 16)
13. Man Duen Choi, Completely positive linear maps on complex matrices, Linear Algebra and its Applications 10 (1975), 285 -- 290.
(page 13)
14. Raphael Clouatre and Michael Hartz, Multiplier algebras of complete Nevanlinna-Pick spaces: dilations, boundary representa-
tions and hyperrigidity, in preparation. (page 10)
15. John B. Conway, A course in functional analysis, second ed., Graduate Texts in Mathematics, vol. 96, Springer-Verlag, New
York, 1990. (page 11)
16. Kenneth R. Davidson, C*-algebras by example, Fields Institute Monographs, vol. 6, American Mathematical Society, Provi-
dence, RI, 1996. (page 4)
17. Kenneth R. Davidson and Matthew Kennedy, The Choquet boundary of an operator system, Duke Mathematical Journal 164
(2015), no. 15, 2989 -- 3004. (page 1, 4, 5, 6, 7, 18)
18. Michael A. Dritschel and Scott A. McCullough, Boundary representations for families of representations of operator algebras
and spaces, Journal of Operator Theory 53 (2005), no. 1, 159 -- 167. (page 8)
19. Maurice J. Dupr´e and Richard M. Gillette, Banach bundles, Banach modules and automorphisms of C*-algebras, Research
Notes in Mathematics, vol. 92, Pitman (Advanced Publishing Program), Boston, MA, 1983. (page 8)
20. Edward G. Effros, Aspects of noncommutative order, C*-algebras and applications to physics (Proc. Second Japan-USA Sem.,
Los Angeles, Calif., 1977), Lecture Notes in Mathematics, vol. 650, Springer, Berlin, 1978, pp. 1 -- 40. (page 1)
21. Edward G. Effros and Zhong-Jin Ruan, On the abstract characterization of operator spaces, Proceedings of the American
Mathematical Society 119 (1993), no. 2, 579 -- 584. (page 12)
22. Edward G. Effros and Soren Winkler, Matrix convexity: operator analogues of the bipolar and Hahn-Banach theorems, Journal
of Functional Analysis 144 (1997), no. 1, 117 -- 152. (page 1, 11)
23. George K. Eleftherakis and Evgenios T. A. Kakariadis, Strong morita equivalence of operator spaces, 446, no. 2, 1632 -- 1653.
(page 3)
24. Douglas Farenick, Extremal matrix states on operator systems, Journal of the London Mathematical Society 61 (2000), no. 3,
885 -- 892. (page 1, 6, 7, 12, 18)
25.
26. Douglas Farenick and Phillip Morenz, C*-extreme points in the generalized state spaces of a C*-algebra, Transactions of the
, Pure matrix states on operator systems, Linear Algebra and its Applications 393 (2004), 149 -- 173. (page 1)
American Mathematical Society 349 (1997), no. 5, 1725 -- 1748. (page 7)
27. Douglas Farenick and Phillip B. Morenz, C*-extreme points of some compact C*-convex sets, Proceedings of the American
Mathematical Society 118 (1993), no. 3, 765 -- 775. (page 7)
28. Kunyu Guo, Junyun Hu, and Xianmin Xu, Toeplitz algebras, subnormal tuples and rigidity on reproducing C[z1, . . . , zd]-
modules, Journal of Functional Analysis 210 (2004), no. 1, 214 -- 247. (page 10)
29. Masamichi Hamana, Injective envelopes of dynamical systems, Operator algebras and operator theory, Pitman Research Notes
in Mathematics Series, vol. 271, Longman Sci. Tech., Harlow, 1992. (page 8)
30.
, Triple envelopes and Shilov boundaries of operator spaces, Mathematics Journal of Toyama University 22 (1999),
77 -- 93. (page 2)
31. J. William Helton and Scott McCullough, Every convex free basic semi-algebraic set has an LMI representation, Annals of
Mathematics. Second Series 176 (2012), no. 2, 979 -- 1013. (page 1)
32. William Helton, Igor Klep, and Scott McCullough, Free convex algebraic geometry, Semidefinite optimization and convex
algebraic geometry, MOS-SIAM Ser. Optim., vol. 13, SIAM, Philadelphia, PA, 2013, pp. 341 -- 405. (page 1)
33. Alan Hopenwasser, Robert L. Moore, and Vern I. Paulsen, C*-extreme points, Transactions of the American Mathematical
Society 266 (1981), no. 1, 291 -- 307. (page 7)
34. Matthew Kennedy and Orr Moshe Shalit, Essential normality, essential norms and hyperrigidity, Journal of Functional Analysis
268 (2015), no. 10, 2990 -- 3016. (page 10)
35. Eberhard Kirchberg and Simon Wassermann, C*-algebras generated by operator systems, Journal of Functional Analysis 155
(1998), no. 2, 324 -- 351. (page 19)
36. Aldo J. Lazar and Joram Lindenstrauss, On Banach spaces whose duals are L1 spaces, Israel Journal of Mathematics 4 (1966),
no. 3, 205 -- 207. (page 1)
37.
, Banach spaces whose duals are L1 spaces and their representing matrices, Acta Mathematica 126 (1971), no. 1,
165 -- 193. (page 1)
BOUNDARY REPRESENTATIONS AND COMPACT RECTANGULAR MATRIX CONVEX SETS
21
38. Martino Lupini, Uniqueness, universality, and homogeneity of the noncommutative Gurarij space, Advances in Mathematics
298 (2016), 286 -- 324. (page 2)
39. Timur Oikhberg, The non-commutative Gurarii space, Archiv der Mathematik 86 (2006), no. 4, 356 -- 364. (page 2)
40. Vern I. Paulsen, Completely bounded maps and operator algebras, Cambridge Studies in Advanced Mathematics, vol. 78,
Cambridge University Press, Cambridge, 2002. (page 2, 3, 6, 14, 16, 17, 19)
41. Vern I. Paulsen and Mrinal Raghupathi, An introduction to the theory of reproducing kernel Hilbert spaces, Cambridge Studies
in Advanced Mathematics, Cambridge University Press, 2016. (page 9)
42. Travis B. Russell, Characterizations of ordered self-adjoint operator spaces, arXiv:1508.06272 (2015). (page 2, 15, 16)
43.
44. Allen L. Shields, Weighted shift operators and analytic function theory, Topics in operator theory, Amer. Math. Soc., Providence,
, Characterizations of ordered operator spaces, arXiv:1608.00939 (2016). (page 15, 16)
R.I., 1974, pp. 49 -- 128. Math. Surveys, No. 13. (page 10)
45. David A. Stegenga, Multipliers of the Dirichlet space, Illinois J. Math. 24 (1980), no. 1, 113 -- 139. (page 9, 10)
46. Gerald D. Taylor, Multipliers on Dα, Trans. Amer. Math. Soc. 123 (1966), 229 -- 240. (page 9, 10)
47. Corran Webster and Soren Winkler, The Krein-Milman theorem in operator convexity, Transactions of the American Mathe-
matical Society 351 (1999), no. 1, 307 -- 322. (page 1, 6, 11, 12, 13)
48. Wend Werner, Subspaces of L(H) that are *-invariant, Journal of Functional Analysis 193 (2002), no. 2, 207 -- 223. (page 15,
16)
49. Gerd Wittstock, On matrix order and convexity, Functional analysis: surveys and recent results, III (Paderborn, 1983), North-
Holland Math. Stud., vol. 90, North-Holland, Amsterdam, 1984, pp. 175 -- 188. (page 1, 11)
Department of Mathematics, Ohio University, Athens, OH 45701
E-mail address: [email protected]
URL: https://sites.google.com/site/afullermath/
Department of Mathematics, Washington University in St Louis, One Brookings Drive, St. Louis, MO 63130
E-mail address: [email protected]
Mathematics Department, California Institute of Technology, 1200 E. California Blvd, MC 253-37, Pasadena, CA
91125
E-mail address: [email protected]
URL: http://www.lupini.org/
|
1202.6413 | 2 | 1202 | 2012-04-07T19:41:11 | Structure for Regular Inclusions | [
"math.OA"
] | We study pairs (C,D) of unital C*-algebras where D is a regular abelian C*-subalgebra of C. When D is a MASA in C, we prove the existence and uniqueness of a completely positive unital map E of C into the injective envelope I(D) of D whose restriction to D is the identity on D. We show that the left kernel of E, L(C,D), is the unique closed two-sided ideal of C maximal with respect to having trivial intersection with D. When L(C,D)=0, we show the MASA D norms C. We apply these results to extend existing results in the literature on isometric isomorphisms of norm-closed subalgebras which lie between D and C.
The map E can be used as a substitute for a conditional expectation in the construction of coordinates for C relative to D. Coordinate constructions of Kumjian and Renault may partially be extended to settings where no conditional expectation exists.
As an example, we consider the situation in which C is the reduced crossed product of a unital abelian C*-algebra D by an arbitrary discrete group acting as automorphisms of D. We characterize when the relative commutant, D', of D in C is abelian in terms of the dynamics of the action of the group and show that when D' is abelian, L(C,D')=0. This setting produces examples where no conditional expectation of C onto D' exists.
When C is separable, and D is a regular MASA in C, we show the set of pure states on D with unique state extensions to C is dense in D. We introduce a new class of well behaved state extensions, the compatible states; we identify compatible states when D is a MASA in C in terms of groups constructed from local dynamics near a pure state on D.
A particularly nice class of regular inclusions is the class of C*-diagonals. We show that the pair (C,D) regularly embeds into a C*-diagonal precisely when the intersection of the left kernels of the compatible states is trivial. | math.OA | math |
STRUCTURE FOR REGULAR INCLUSIONS
DAVID R. PITTS
Dedicated to the memory of William B. Arveson
Abstract. We study pairs (C, D) of unital C ∗-algebras where D is an abelian C ∗-subalgebra of C
which is regular in C in the sense that the span of {v ∈ C : vDv∗ ∪ v∗Dv ⊆ D} is dense in C. When
D is a MASA in C, we prove the existence and uniqueness of a completely positive unital map E
of C into the injective envelope I(D) of D whose restriction to D is the identity on D. We show
that the left kernel of E, L(C, D), is the unique closed two-sided ideal of C maximal with respect
to having trivial intersection with D. When L(C, D) = 0, we show the MASA D norms C in the
sense of Pop-Sinclair-Smith. We apply these results to significantly extend existing results in the
literature on isometric isomorphisms of norm-closed subalgebras which lie between D and C.
The map E can be used as a substitute for a conditional expectation in the construction of
coordinates for C relative to D. We show that coordinate constructions of Kumjian and Renault
which relied upon the existence of a faithful conditional expectation may partially be extended to
settings where no conditional expectation exists.
As an example, we consider the situation in which C is the reduced crossed product of a unital
abelian C ∗-algebra D by an arbitrary discrete group Γ acting as automorphisms of D. We charac-
terize when the relative commutant Dc of D in C is abelian in terms of the dynamics of the action
of Γ and show that when Dc is abelian, L(C, Dc) = (0). This setting produces examples where no
conditional expectation of C onto Dc exists.
In general, pure states of D do not extend uniquely to states on C. However, when C is separable,
and D is a regular MASA in C, we show the set of pure states on D with unique state extensions to
C is dense in D. We introduce a new class of well behaved state extensions, the compatible states;
we identify compatible states when D is a MASA in C in terms of groups constructed from local
dynamics near an element ρ ∈ D.
A particularly nice class of regular inclusions is the class of C ∗-diagonals; each pair in this
class has the extension property, and Kumjian has shown that coordinate systems for C ∗-diagonals
are particularly well behaved. We show that the pair (C, D) regularly embeds into a C ∗-diagonal
precisely when the intersection of the left kernels of the compatible states is trivial.
Contents
Introduction, Background and Preliminaries
1.
1.1. Preliminaries
2. Dynamics of Regular Inclusions
2.1. Normalizers and Intertwiners
2.2. Quasi-Freeness and the Extension Property
2.3. D-modular states and ideals of (C, D)
3. Pseudo-Conditional Expectations for Regular MASA Inclusions
4. Compatible States
5. The D-Radical and Embedding Theorems
6. An Example: Reduced Crossed Products by Discrete Groups
2
6
8
8
9
12
13
24
31
33
2000 Mathematics Subject Classification. 46L05, 46L30, 47L30.
Key words and phrases. C ∗-diagonal, inclusions of C ∗-algebras, groupoid C ∗-algebra.
The author is grateful for the support of the University of Nebraska's NSF ADVANCE grant #0811250 in the
completion of this paper.
1
7. A Description of S(C, D) for a Regular MASA Inclusion
8. The Twist of a Regular Inclusion
8.1. Twists and their C ∗-algebras
8.2. Compatible Eigenfunctionals and the Twist for (C, D)
9. Applications
References
38
49
50
52
59
61
1. Introduction, Background and Preliminaries
In this paper, we investigate the structure of the class of regular inclusions.
Definition 1.1. An inclusion is a pair (C, D) of C ∗-algebras with D abelian, and D ⊆ C. When C
has a unit I, we always assume that I ∈ D. For any inclusion, let
N(C, D) := {v ∈ C : v∗Dv ∪ vDv∗ ⊆ D};
elements of N(C, D) are called normalizers. If (Ci, Di) (i = 1, 2) are inclusions, and θ : C1 → C2 is
a ∗-homomorphism, we will say that θ is a regular ∗-homomorphism if
When θ is regular and one-to-one, we will say that (C1, D1) regularly embeds into (C2, D2).
θ(N(C1, D1)) ⊆ N(C2, D2).
The inclusion (C, D) is a
regular inclusion if span N(C, D) is norm-dense in C;
MASA inclusion if D is a MASA in C;
EP inclusion if D has the extension property relative to C, that is, every pure state σ on D
has a unique extension to a state on C and no pure state of C annihilates D;
Cartan inclusion if (C, D) is a regular MASA inclusion and there exists a faithful conditional
C ∗-diagonal
expectation E : C → D;
if (C, D) is a Cartan inclusion and also an EP-inclusion.
Examples of regular inclusions are commonplace in the theory of C ∗-algebras: any MASA D in
a finite dimensional C ∗-algebra C yields a C ∗-diagonal (C, D); the categogy of C ∗-diagonals and
regular ∗-monomorphisms is closed under inductive limits [10, Theorem 4.23]; when a discrete group
Γ acts topologically freely on a compact Hausdorff space X, the reduced crossed product C(X)⋊r Γ
together with the canonical embedding of C(X) yields a Cartan pair (C(X) ⋊r Γ, C(X)) [31]; when
C is the C ∗-algebra of a directed graph and D is the C ∗-subalgebra generated by the range and
source projections of the partial isometries corresponding to the edges of the graph, the pair (C, D)
is a regular inclusion. Other examples arise from certain constructions in the theory of groupoid
C ∗-algebras or from C ∗-algebras constructed from combinatoral data.
In this paper, we present a number of structural results for regular inclusions. Our main results
include: Theorem 3.10, which establishes the existence and uniqueness of a psuedo-expectation
for a regular MASA inclusion; Theorem 3.22, which shows that the left kernel L(C, D)) of the
pseudo-expectation is a two-sided ideal in C which is maximal with respect to being diagaonal
disjoint; Theorem 5.9, which characterizes when a regular inclusion can be regularly embedded
into a C ∗-diagonal; Theorem 8.14 which shows how constructions of Kumjian and Renault may
be used to produce a twist associated to a regular inclusion; Theorem 9.2, which shows that for a
regular MASA inclusion with L(C, D) = (0), D norms C in the sense of Pop-Sinclair-Smith; and
Theorem 9.4, which gives conditions on which an isometric isomorphism of a subalgebra A of C
containing D can be extended to a ∗ isomorphism of the C ∗-algebra generated by A. We turn now
to some background.
2
In a landmark paper, J. Feldman and C. Moore [13] considered pairs (M, D) consisting of a
(separably acting) von Neumann algebra M containing a MASA D ≃ L∞(X, µ) such that the set
of normalizing unitaries, {u ∈ M : u is unitary and uDu∗ = D}, has σ-weakly dense span in M
and there exists a faithful normal conditional expectation E : M → D. In this context, Feldman
and Moore showed that there is a Borel equivalence relation R on X and a cocycle c such that
the pair (M, D) can be identified with a von-Neumann algebra arising from certain Borel functions
on R. In a heuristic sense, their construction may be viewed as the left regular representation of
the equivalence relation where the multiplication is twisted by the cocycle c. This result may be
viewed as a means of coordinatizing the von Neumann algebra along D.
Work on a C ∗-algebraic version of the Feldman-Moore result was studied by Kumjian in [22]. In
that article, Kumjian introduced the notion of a C ∗-diagonals as well as the notion of regularity.
(We should mention that the axioms for a C ∗-diagonal given in [22], while equivalent to the axioms
given in Definition 1.1, do not explicitly mention the extension property. In the sequel, we will
have considerable interest in the extension property or its failure, which is why we use the axioms
given in Definition 1.1.) Kumjian showed that if (C, D) is a C ∗-diagonal, with C separable and D
second countable, then it can be coordinatized via a twisted groupoid over a topological equivalence
relation. This provided a very satisfying parallel to the von Neumann algebraic context.
The requirement of the extension property in the axioms for a C ∗-diagonal is at times too
stringent, which is one of the advantages to Cartan inclusions, which need not have the extension
property. For example, let H = ℓ2(N), with the usual orthonormal basis {en}, and let S be the
unilateral shift, Sen = en+1. Let C := C ∗(S) be the Toeplitz algebra, and let D = C ∗({SnS∗n : n ≥
0}). Routine arguments show this is a Cartan inclusion, but the state ρ∞(T ) = limn→∞ hT en, eni
on D fails to have a unique extension to a state on C,
Cartan inclusions were introduced by Renault in [31], where he showed that if (C, D) is a Cartan
inclusion (again with the separability and second countability hypotheses), then there is a satis-
factory coordinatization of the pair (C, D) via a twisted groupoid. In this paper, Renault makes
a very convincing case that Cartan inclusions are the appropriate analog of the Feldman-Moore
setting in the C ∗-context.
Let (C, D) be an inclusion. A conditional expectation E : C → D gives a preferred class,
{ρ ◦ E : ρ ∈ D}, of extensions of pure states on D to states on C, and when the expectation E is
unique, this class may be used for construction of coordinates. Indeed, when (C, D) is a Cartan
inclusion (or C ∗-diagonal), elements of the twisted groupoid arise from ordered pairs [v, ρ], where
v ∈ N(C, D) and ρ ∈ D with ρ(v∗v) 6= 0. Such a pair determines a linear functional of norm one on
C by the rule,
(1)
[v, ρ](x) =
ρ(E(v∗x))
ρ(v∗v)1/2
(x ∈ C).
The work of Feldman-Moore also makes essential use of conditional expectations.
By [2, Theorem 3.4], any EP inclusion (C, D) has a unique conditional expectation E : C → D.
Thus for a C ∗-diagonal, the extension property guarantees the the uniqueness of the expectation ex-
pectation E : C → D. Interestingly, the extension property is not necessary to guarantee uniqueness
of expectations. Indeed, Renault showed that when (C, D) is a Cartan inclusion, with expectation
E : C → D, then E is the unique conditional expectation of C onto D.
What other regular inclusions (C, D) have unique expectations of C onto D? While we do
not know the answer to this question in general, we give a positive result along these lines in
Theorem 2.10 below: this result shows that when (C, D) is a regular MASA inclusion with D
an injective C ∗-algebra, then (C, D) is an EP inclusion, and therefore has a unique expectation.
Section 2 also contains results on dynamics of regular inclusions and introduces the notion of a
quasi-free action. Theorem 2.9 characterizes the extension property for regular MASA inclusions in
3
terms of the dynamics of regular inclusions: the regular MASA inclusion (C, D) is an EP-inclusion
if and only if the ∗-semigroup N(C, D) acts quasi-freely on D.
Unfortunately, conditional expectations do not always exist, even when (C, D) is a regular MASA
inclusion, and the C ∗-algebras involved are well-behaved. Here is a simple example, which is a
special case of the far more general setting considered in Section 6.
Example 1.2. Let X be a connected, compact Hausdorff space, and let α : X → X be a home-
omorphism such that α2 is the identity map on X. Let F ◦ be the interior of the set of fixed
points for α; we assume that F ◦ is neither empty nor all of X. (For a concrete example, take
X = {z ∈ C : z ≤ 1 and ℜ(z) ℑ(z) = 0} and let α(z) = z.) Define θ : C(X) → C(X) by
θ(f ) = f ◦ α−1, and set
C :=(cid:26)(cid:18) f0
θ(f1) θ(f0)(cid:19) : f0, f1 ∈ C(X)(cid:27) and D :=(cid:26)(cid:18)f0
f1
0
0
θ(f0)(cid:19) : f0 ∈ C(X)(cid:27) .
Then C is a C ∗-subalgebra of M2(C(X)), and (C, D) is a regular inclusion. (C may be regarded as
C(X) ⋊ (Z/2Z).)
As F ◦ /∈ {∅, X}, we have D ( Dc ( C, and another calculation shows Dc is abelian. Since
Suppose E : C → Dc is a conditional expectation. Then for some f0, f1 ∈ C(X) with supp(f1) ⊆
f1
A calculation shows that the relative commutant Dc of D in C is
I 0(cid:1) =(cid:0) f0
f1θ(f0)
Dc =(cid:26)(cid:18) f0
θ(f1) θ(f0)(cid:19) ∈ C : supp(f1) ⊆ F ◦(cid:27) .
(cid:0) 0 1
1 0(cid:1) ∈ N(C, Dc), it follows that (C, Dc) is a regular MASA inclusion.
f1 0(cid:1) ∈ Dc. We have
F ◦, we have E(cid:0) 0 I
0(cid:19)(cid:18)0 I
0(cid:19)(cid:19) =(cid:18)f1
θ(f1) θ(f0)(cid:1). Notice θ(f1) = f1 and(cid:0) 0 f1
(cid:18) f 2
1 (cid:19) =(cid:18) 0
0 θ(f0)(cid:1).
I 0(cid:1)) =(cid:0) f0
Thus E((cid:0) 0 I
Now if g1 ∈ D is such that supp(g1) ⊆ F ◦, then θ(g1) = g1, so(cid:0) 0 g1
g1 0(cid:1) ∈ Dc. Thus,
(cid:18) 0
0(cid:19) = E(cid:18)(cid:18)g1
0(cid:19) =(cid:18)g1
g1(cid:19)(cid:18)f0
=(cid:18)g1f0
g1(cid:19)(cid:18)0 I
θ(g1f0)(cid:19) .
0(cid:19) = E(cid:18)(cid:18) 0
0(cid:19) E(cid:18)0 I
0
g1(cid:19) E(cid:18)0 I
0(cid:19)(cid:19) =(cid:18)g1
0
1
f1f0
f 2
f1
0
0
f1
f1
0
I
f1
I
0
0
g1
g1
0
0
f1
I
0
f1(cid:19) ,
0
0
I
0
θ(f0)(cid:19)
so f 2
1 = f1. As X is connected, this yields f1 = 0 or f1 = I. But supp(f1) ⊆ F ◦ 6= X, so f1 = 0.
Hence g1 = 0 for every such g1. This implies that F ◦ = ∅, contrary to hypothesis. Hence no
conditional expectation of C onto Dc exists.
One of the goals of this paper is to show that even though conditional expectations may fail to
exist for a regular MASA inclusion, there is a map which which may be used as a replacement.
Here is the relevant definition.
Definition 1.3. Let (C, D) be an inclusion and let (I(D), ι) be an injective envelope for D. A
pseudo-conditional expectation for ι, or more simply, a pseudo-expectation for ι, is a unital com-
pletely positive map E : C → I(D) such that ED = ι. When the context is clear, we sometimes
drop the reference to ι and simply call E a pseudo-expectation.
The existence of pseudo-expectations follows immediately from the injectivity of I(D). In general,
the pseudo-expectation need not be unique.
However, in Section 3 below, we show that for any regular MASA inclusion (C, D), there is always
a unique pseudo-expectation E : C → I(D), see Theorem 3.10. Let Mod(C, D) be the family of all
4
states on C which restrict to elements of D. The family of states, Ss(C, D) := {ρ ◦ E : ρ ∈ [I(D)}
covers D in the sense that the restriction map, Ss(C, D) ∋ ρ 7→ ρD ∈ D, is onto. Interestingly,
Ss(C, D) is the unique minimal closed subset of Mod(C, D) which covers D, see Theorem 3.13. We
also show that Ss(C, D) is closely related to the extension property. When (C, D) is "countably
generated," Theorem 3.13 also shows that Ss(C, D) is the closure of all states in Mod(C, D) whose
restrictions to D extend uniquely to C.
For a regular MASA inclusion, the intersection of the left kernels of the states in Ss(C, D) is the
left kernel of the pseudo-expectation E. Let L(C, D) be the left kernel of E. Theorem 3.22 shows
that L(C, D) is an ideal of C, and moreover, is the unique ideal of C which is maximal with respect
to the property of having trivial intersection with D. When the pseudo-expectation takes values in
D (rather than I(D)), the ideal L(C, D) may be viewed as a measure of the failure of the inclusion
to be Cartan in Renault's sense.
We define a regular MASA inclusion to be a virtual Cartan inclusion when the pseudo-expectation
is faithful, or equivalently, when L(C, D) = 0. The purpose of Section 6 is to give a large class of
virtual Cartan inclusions. Theorem 6.9 shows that when C is the reduced crossed product of the
abelian C ∗-algebra D by a discrete group Γ, then, provided the relative commutant Dc of D in C
is abelian, (C, Dc) is a virtual Cartan inclusion. We characterize when Dc is abelian in terms of
the dynamics of the action of Γ on D in Theorem 6.6; this result shows that Dc is abelian precisely
when the germ isotropy subgroup H x of Γ is abelian for every x ∈ D. The results of Section 6 are
summarized in Theorem 6.10.
One of the motivations for our study of inclusions was to provide a context for the study of
certain nonselfadjoint subalgebras. If (C, D) is a C ∗-diagonal, there are numerous papers devoted
to the study of various (usually nonselfadjoint) closed algebras A with D ⊆ A ⊆ C, see [7, 9, 10, 21,
23, 24, 25] to name just a few. An often successful strategy for the analysis of the subalgebras of
C ∗-diagonals is to use the coordinatization of (C, D) (via the twist) to impose coordinates on the
subalgebras; properties of the coordinate system then reflect properties of the subalgebra.
Since many classes of regular MASA inclusions are neither C ∗-diagonals nor Cartan inclusions, it
is natural to wonder whether coordinate methods may be used to analyze nonselfadjoint subalgebras
of regular MASA inclusions. A strategy for doing so is to try to regularly embed a given regular
MASA inclusion (C, D) into a C ∗-diagonal (C1, D1) and then to restrict the coordinates obtained
from (C1, D1) to the (C, D) or to the given subalgebra. This leads to the following problem.
Problem 1.4. Characterize when a given regular inclusion (C, D) can be regularly embedded into
a C ∗-diagonal.
We give a solution to Problem 1.4 in Section 5. To do this, we introduce a new family S(C, D) ⊆
Mod(C, D), which we call compatible states. When (C, D) is a regular MASA inclusion, Ss(C, D) ⊆
S(C, D). The intersection of the left kernels of the states in S(C, D) is an ideal of C, Rad(C, D).
The regular inclusion (C, D) regularly embeds in a C ∗-diagonal if and only if Rad(C, D) = (0), see
Theorem 5.9. In particular, any virtual Cartan inclusion regularly embeds into a C ∗-diagonal.
The needed properties of compatible states are developed in Section 4. Compatible states can be
defined for any inclusion, and we expect that they may be useful in other contexts as well. While
compatible states exist in abundance for any regular MASA inclusion, Theorem 4.8 implies that
compatible states need not exist for a general regular inclusion.
For a regular MASA inclusion (C, D), it is always the case that Rad(C, D) ⊆ L(C, D). We have
been unable to resolve the question of whether equality holds. We provide some insight into this
question in Section 7. Given σ ∈ D, there is an equivalence relation R1 on Hσ := {v ∈ N(C, D) :
ρ(v∗dv) = ρ(d) for all d ∈ D}, and the set Hσ/R1 of equivalence classes of this equivalence relation
may be made into a T-group. The main result of this section, Theorem 7.13, shows that there
5
is a bijective correspondence between {ρ ∈ S(C, D) : ρD = σ} and a certain family of pre-
homomorphisms on Hσ/R1. Theorem 7.13 thus gives a description of a certain class of state
extensions of σ. Our description leads us to suspect that it is possible for Rad(C, D) to be a proper
subset of L(C, D).
The purpose of Section 8 is to discuss certain twisted groupoids arising from a regular MASA
inclusion. The methods of this section are suitable modifications to our context of the methods used
by Kumjian and Renault when coordinatizing C ∗-diagonals and Cartan inclusions. We show that
given any regular MASA inclusion (C, D), there is a twist associated to (C, D). We use Ss(C, D)
or another suitable subset F of S(C, D) as unit space of the twist, and define functionals [v, ρ] on
C much as in equation (1), except that the functional ρ ◦ E appearing in that formula is replaced
with an element of F . The main result of this section, Theorem 8.14, shows that when Rad(C, D)
is trivial, there is a regular ∗-monomorphism of (C, D) into the Cartan inclusion arising from the
twist associated to (C, D). This result gives perspective to the embedding results of Section 5, and
suggests that it is indeed possible to coordinatize subalgebras using this twist, as indicated prior
to Problem 1.4.
In [28], we gave a method for extending an isometric isomorphism between subalgebras Ai of
C ∗-diagonals (Ci, Di) with Di ⊆ Ai ⊆ Ci to a ∗-isomorphism of the C ∗-subalgebra C ∗(A1) of C1
generated by A1 onto the corresponding subalgebra C ∗(A2). The two main ingredients of this
method were: a) show that Di norms Ci in the sense of Pop-Sinclair-Smith ([29]), and b) show
that the C ∗-envelope of Ai is isometrically isomorphic to C ∗(Ai). In Section 9, we show that if
the hypothesis that (Ci, Di) is weakened from C ∗-diagonal to virtual Cartan inclusion, then both
these ingredients still hold. Theorem 9.4 generalizes [28, Theorem 2.16] to the context of virtual
Cartan inclusions. This is a considerable generalization, and allows for the simplification of some
arguments in the literature.
The author is indebted to William Arveson, who greatly influenced the author and the field of
operator algebras. His passing saddens us all.
We thank Ken Davidson, Allan Donsig, William Grilliette, Vern Paulsen, and Vrej Zarikian for
several very helpful conversations.
1.1. Preliminaries. Given a Banach space X, we will use X# instead of the traditional X∗ to
denote the Banach space dual. Likewise if α : X → Y is a bounded linear map between Banach
spaces, we use α# to denote the adjoint map, f ∈ Y# 7→ f ◦ α ∈ X#.
If X is a topological space and E ⊆ X, E◦ denotes the interior of E. Also, for f : X → C, we
write supp f for the set {x ∈ X : f (x) 6= 0}.
Standing Assumption 1.5. For the remainder of this paper, all C ∗-algebras will be unital, and
if D is a sub-C ∗-algebra of the C ∗-algebra C, we assume that the unit for D is the same as the unit
for C.
Let C be a C ∗-algebra, and let S(C) be the state space of C. For ρ ∈ S(C) let Lρ = {x ∈ C :
ρ(x∗x) = 0} be the left kernel of ρ, and let (πρ, Hρ, ξ) be the GNS representation corresponding to
ρ. We regard C/Lρ as a dense subset of Hρ, and for x ∈ C will often write x + Lρ to denote the
vector πρ(x)ξ. Denote the inner product on Hρ by h·, ·iρ.
We now recall some facts about projective topological spaces, projective covers, and injective
envelopes of abelian C ∗-algebras. Following [17], given a compact Hausdorff space X, a pair (P, f )
consisting of a compact Hausdorff space P and a continuous map f : P → X is called a cover
for X (or simply a cover ) if f is surjective. A cover (P, f ) is rigid if the only continuous map
h : P → P which satisfies f ◦ h = f is h = idP ; the cover (P, f ) is essential if whenever Y is a
compact Hausdorff space and h : Y → P is continuous and satisfies f ◦ h is onto, then h is onto.
A compact Hausdorff space P is projective if whenever X and Y are compact Hausdorff spaces
and h : Y → X and f : P → X are continuous maps with h surjective, there exists a continuous
6
map g : P → Y with g ◦ h = f . A Hausdorff space which is extremally disconnected (i.e. the closure
of every open set is open) and compact is Stonean. In [15, Theorem 2.5], Gleason proved that a
compact Hausdorff space P is projective if and only P is Stonean.
By [17, Proposition 2.13], if (P, f ) is a cover for X with P a projective space, then (P, f ) is rigid
if and only if (P, f ) is essential. A projective cover for X is a rigid cover (P, f ) for X such that P
is projective. A projective cover for X always exists [17, Theorem 2.16] and is unique in the sense
that if (P1, f1) and (P2, f2) are projective covers for X, then there is a unique homeomorphism
h : P1 → P2 such that f1 = f2 ◦ h.
The concept of an injective envelope for an abelian unital C ∗-algebras is dual to the concept of
a projective cover of a compact Hausdorff space: if (P, f ) is a cover for X, let ι : C(X) → C(P ) be
the map d 7→ d ◦ f ; then (P, f ) is a projective cover if and only if (C(P ), ι) is an injective, envelope
of C(X) [17, Corollary 2.18]. (Injective envelopes can be also be defined for general unital C ∗-
algebras, not just abelian, unital C ∗-algebras. Like projective covers, injective envelopes of unital
C ∗-algebras have a uniqueness property.
If A is a unital C ∗-algebra, and (B1, σ1) and (B2, σ2)
are injective envelopes for A, then there exists a unique ∗-isomorphism θ : B1 → B2 such that
θ ◦ σ1 = σ2 [18, Theorem 4.1].)
Recall that a unital C ∗-algebra C is monotone complete if every bounded increasing net in the
self-adjoint part, Cs.a., of C has a least upper bound in Cs.a.. Hamana [20] shows that every injective
C ∗-algebra is monotone complete. When (xλ) is a bounded increasing net in Cs.a., supC xλ means
the least upper bound of (xλ) in C.
Theorem 6.6 of [20] implies that if A is a unital, abelian C ∗-algebra then any injective envelope
(B, σ) for A is Hamana-regular in the sense that whenever x ∈ B is self-adjoint, x = supB{σ(a) :
a ∈ A, a = a∗ and a ≤ x}. (Since A is abelian, we regard {σ(a) : a ∈ A, a = a∗ and a ≤ x}
as a net indexed by itself.) As Hamana observes in [20], when x ∈ A is positive, then also,
x = supB{σ(a) : a ∈ A and 0 ≤ σ(a) ≤ b}.
Here is a description of an injective envelope of an abelian C ∗-algebra. For details, see [16,
Theorem 1]. Let X be a compact Hausdorff space. Define an equivalence relation on the algebra
B(X) of all bounded Borel complex-valued functions on X by f ∼ g if and only if {x ∈ X :
f (x) − g(x) 6= 0} is a set of first category. The equivalence class J of the zero function is an ideal in
B(X) and the quotient D(X) := B(X)/J is called the Dixmier algebra. Define j : C(X) → D(X)
by j(f ) = f + J. Then (D(X), j) is an injective envelope for D.
We conclude this section with a few comments regarding categories. Let C be the category of
unital abelian C ∗-algebras with ∗-homomorphisms, and let O be the category of operator systems
with completely positive (unital) maps. Let D be a unital abelian C ∗-algebra. Then D is an injective
object in C if and only if D is an injective object in O, see [17, Theorem 2.4]. (The statement of [17,
Theorem 2.4], mentions the category of operator systems without explicitly giving the morphisms,
but the proof makes it clear that the authors mean the category of operator systems and unital,
completely positive maps.)
Hamana shows that in the category O, there is an injective object I(D) and a one-to-one com-
pletely positive ι : D → I(D) such that the extension (I(D), ι) is rigid and essential. Hamana and
Hadwin-Paulsen (see [19] and [17, Corollary 2.18]) observe that I(D) is endowed with a product
which makes it into an abelian C ∗-algebra (and ι a ∗-monomorphism). Set X = D, and let (P, f )
be a projective cover for X, so that the map τ : D → C(P ) given by τ (x) = x ◦ f is a one-to-one
∗-homomorphism of D into C(P ). Corollary 2.18 of [17] also shows the existence of a ∗-isomorphism
θ : C(P ) → I(D) such that θ ◦ τ = ι.
Thus, for us an injective envelope (I(D), ι) for D will be a rigid and essential extension of D in
the category C. The comments above show that this is equivalent to saying that (I(D), ι) is a rigid
and essential extension for D in O.
7
2. Dynamics of Regular Inclusions
Given a regular inclusion (C, D), the ∗-semigroup N(C, D) of normalizers acts via partial home-
omorphisms on the maximal ideal space D of D. The purpose of this section is to discuss some
of the features of this action. The first subsection is devoted to notation and some background
facts regarding normalizers and intertwiners. The second subsection gives a characterization of the
extension property in terms of the dynamics associated with the action of N(C, D) on D (Theo-
rem 2.9). An interesting consequence is that for any regular MASA inclusion (C, D) with D injective
is an EP inclusion, see Theorem 2.10.
2.1. Normalizers and Intertwiners. Let (C, D) be an inclusion. Closely related to normalizers
are intertwiners.
Definition 2.1. An intertwiner for D is an element v ∈ C such that vD = Dv. We denote the set
of all intertwiners by I(C, D).
Proposition 3.3 of [10] shows that for any intertwiner v, the elements v∗v and vv∗ belong to the
relative commutant Dc of D in C. When v is a normalizer, the fact that D is unital shows that
both v∗v and vv∗ belong to D, and it is easy to see that {v ∈ I(C, D) : v∗v, vv∗ ∈ D} ⊆ N(C, D).
In general however, there exists v ∈ I(C, D) with {v∗v, vv∗} 6⊆ D. (For a simple example, observe
that every operator in M2(C) is an intertwiner for the inclusion, (M2(C), CI2).) Also, by [10,
Proposition 3.4], N(C, D) ⊆ I(C, D). However, the final paragraph of the proof of [10, Proposi-
tion 3.4], shows somewhat more than this, namely that {v ∈ I(C, D) : v∗v, vv∗ ∈ D} = N(C, D).
Summarizing, we have the following result.
Proposition 2.2 ([10, Propositions 3.3 and 3.4]). Let (C, D) be an inclusion. Then
Furthermore, when D is a MASA in C, N(C, D) = I(C, D).
{v ∈ I(C, D) : v∗v, vv∗ ∈ D} = N(C, D) ⊆ I(C, D).
The following example of a regular homomorphism will be useful in the sequel.
Lemma 2.3. Suppose (C, D) is an inclusion such that the relative commutant, Dc, of D in C is
abelian. Then (C, Dc) is a MASA inclusion and N(C, D) ⊆ N(C, Dc); in particular the identity map
id : (C, D) → (C, Dc) is a regular ∗-homomorphism.
Proof. Since D and Dc are abelian, (C, Dc) is a MASA inclusion.
To show that N(C, D) ⊆ N(C, Dc), suppose v ∈ I(C, D). Fix h ∈ Dc, and let d ∈ D be self-
adjoint. Since v is a D-intertwiner, we may find a self-adjoint d′ ∈ D so that dv = vd′. Then
d(vhv∗) = vd′hv∗ = vhd′v∗ = (vhv∗)d, from which it follows that vhv∗ ∈ Dc. Similarly, v∗hv ∈ Dc.
We conclude that I(C, D) ⊆ N(C, Dc). Since N(C, Dc) is norm-closed, Proposition 2.2 yields,
N(C, D) ⊆ I(C, D) ⊆ N(C, Dc).
Thus, (C, Dc) is a MASA inclusion and the identity mapping id : (C, D) → (C, Dc) is a regular
∗-homomorphism.
(cid:3)
For any topological space X, a partial homeomorphism is a homeomorphism h : S → R, where
S and R are open subsets of X. As usual, dom(h) and ran(h) will denote the domain and range
of the partial homeomorphism h. We use InvO(X) to denote the inverse semigroup of all partial
homeomorphisms of X. When S is a ∗-semigroup, a semigroup homomorphism α : S → InvO(X)
is a ∗-homomorphism if for every s ∈ S, α(s∗) = α(s)−1. A subset G of InvO(X) which is closed
under composition and inverses (i.e. a sub inverse semigroup) is called a pseudo-group on X.
Associated to any pseudo-group G on X is the groupoid of germs which is the set of equivalence
classes, {[x, φ, y] : φ ∈ G, y ∈ dom(φ), x = φ(y)}, where [x, φ, y] = [x1, φ1, y1] if and only y = y1
8
and there exists a neighborhood N of y such that φN = φ1N . Elements [w, φ, x] and [y, ψ, z] are
composable if x = y and then [w, φ, x] [y, ψ, z] = [w, φψ, z] and [x, φ, y]−1 = [y, φ−1, x]. The range
and source maps are r([x, φ, y]) = x and s([x, φ, y]) = y. Thus X may be identified with the unit
space of the groupoid of germs.
Recall (see [22, Proposition 6]) that a normalizer v determines a partial homeomorphism
βv : {ρ ∈ D : ρ(v∗v) > 0} → {ρ ∈ D : ρ(vv∗) > 0}
given by βv(ρ)(d) =
ρ(v∗dv)
ρ(v∗v)
(d ∈ D).
Clearly N(C, D) and I(C, D) are ∗-semigroups under multiplication. Routine, but tedious, calcula-
tions show that the map N(C, D) ∋ v 7→ βv is a ∗-semigroup homomorphism β : N(C, D) → InvO( D).
We record this fact as a proposition.
Proposition 2.4 ([31, Lemma 4.10]). Suppose (C, D) is an inclusion. Then the following state-
ments hold.
(1) Suppose that v, w ∈ N(C, D) and ρ ∈ D satisfies ρ(w∗v∗vw) 6= 0. Then ρ(w∗w) 6= 0, and
βvw(ρ) = βv(βw(ρ)).
(2) For every v ∈ N(C, D), βv∗ = (βv)−1.
Following Renault [31] we will call the collection,
PG(C, D) := {βv : v ∈ N(C, D)}
the Weyl pseudo-group of the inclusion. Also, the groupoid of germs of PG(C, D) is the Weyl
groupoid of the inclusion, which we denote by WG(C, D).
Remarks 2.5.
(1) Observe that if θ is a regular homomorphism, then θ(D1) ⊆ D2:
indeed, for d ∈ D1
with d ≥ 0, d1/2 ∈ N(C1, D1) and so θ(d) = θ(d1/2)1θ(d1/2) ∈ D2. It follows that the
dynamics of inclusions under regular homomorphisms are well-behaved in the sense that if
θ : (C1, D1) → (C2, D2) is a regular homomorphism of the inclusions (Ci, Di), then whenever
v ∈ N(C1, D1) \ ker θ, the following diagram commutes:
D2 ⊇ dom(βθ(v))
βθ(v)
range(βθ(v)) ⊆ D2
θ#
θ#
D1 ⊇ dom(βv)
βv
/ range(βv) ⊆ D1.
(2) For some purposes, it is easier to work with intertwiners than normalizers. Thus, one might
be tempted to define a regular homomorphism using intertwiners instead of normalizers,
that is, by mandating θ(I(C1, D1)) ⊆ I(C2, D2). However, a disadvantage of doing so is that
such a θ need not carry D1 into D2, which is why we use N(C, D) rather than I(C, D) in
Definition 1.1. For an example of this, let C = C([0, 1]) and let D = {f ∈ C([0, 1]) : f (0) =
f (1/2)}. Then C = I(C, D), and since every unitary in C normalizes D, we see that (C, D)
is regular (as in Definition 1.1). Taking (Ci, Di) = (C, D), then any automorphism θ of C
satisfies θ(I(C1, D1)) ⊆ I(C2, D2), yet clearly one may choose θ so that θ(D1) 6⊆ θ(D2).
2.2. Quasi-Freeness and the Extension Property. By Proposition 2.4, S := {βv : v ∈
N(C, D)} is an inverse semigroup of partial homeomorphisms of D. Recall that a group G of
homeomorphisms of a space X acts freely if whenever g ∈ G has a fixed point, then g is the
identity. Paralleling the notion for groups, we make the following definition.
9
/
/
/
Definition 2.6. Suppose that S is a ∗-semigroup and X is a compact Hausdorff space, and that
α : S → InvO(X) is a ∗-semigroup homomorphism. We say that S acts quasi-freely on X if whenever
s ∈ S, {x ∈ dom(α(s)) : α(s)(x) = x} is an open set in X.
When Γ is a group acting quasi-freely on X, this says that for each s ∈ Γ, the set of fixed points
of α(s) is a clopen set; in particular, when X is a connected set, the notions of free and quasi-free
actions for a group (acting as homeomorphisms) on X coincide.
In some circumstance, quasi-freeness is automatic. Recall that a Hausdorff topological space X
is extremally disconnected if the closure of every open subset of X is open and that X is a Stonean
space if X is compact, Hausdorff, and extremally disconnected. We shall show that if a ∗-semigroup
acts on a Stonean space, then it acts quasi-freely. To do this we require the following topological
proposition. The proof is a straightforward adaptation of the elegant proof by Arhangel′skii of
Frol´ık's Theorem [14, Theorem 3.1] on fixed points of homeomorphisms of extremally disconnected
spaces. We provide a sketch of the proof for the convenience of the reader.
Proposition 2.7 (Frol´ık's Theorem). Let X be an extremally disconnected space, let V, W be clopen
subsets of X, and suppose h : V → W is a homeomorphism of V onto W . Then the set of fixed
points F := {x ∈ V : h(x) = x} is a clopen subset of X. Moreover, there are three disjoint clopen
subsets C1, C2, C3 of X such that for i = 1, 2, 3, h(Ci) ∩ Ci = ∅ = Ci ∩ F and V = F ∪ C1 ∪ C2 ∪ C3.
Proof (see [4, Theorem 1]). Call an open subset A ⊆ V h-simple if h(A) ∩ A = ∅. By the Hausdorff
maximality theorem, there exists a maximal chain G of h-simple sets. Put U = S G. Then U is
also a h-simple subset of V , and since U is open, maximality shows that U is in fact clopen.
Next observe that h(U ) ∩ V and h−1(U ∩ W ) are clopen h-simple sets, and put
M = U ∪ (h(U ) ∩ V ) ∪ h−1(U ∩ W ).
Since the intersection of F with any h-simple subset of V is empty, we have M ∩ F = ∅. We shall
show that F = V \ M .
Suppose to the contrary, that x ∈ V \ M satisfies h(x) 6= x. Let H be an open subset of V such
that x ∈ H and H ∩ M and h(H) ∩ H are both empty. Then H is h-simple and
(2)
H ∩ U = H ∩ (h(U ) ∩ V ) = H ∩ h−1(U ∩ W ) = ∅.
But (2) implies that H ∪U is a h-simple set which properly contains U , contradicting the maximality
of U . So F = V \ M .
Since both V and M are clopen, so is F . Finally, to complete the proof, take C1 := U , C2 :=
(cid:3)
h(U ) ∩ V, and C3 := h−1(U ∩ W ) \ (h(U ) ∩ V ).
With this preparation, we now show that any action of a ∗-semigroup on a Stonean space is
quasi-free.
Theorem 2.8. Suppose that X is a Stonean space, S is a ∗-semigroup, and α : S → InvO(X) is a
∗-semigroup homomorphism. Then S acts quasi-freely on X.
Proof. Fix s ∈ S, and consider the open sets G := dom(α(s)) and H := ran(α(s)). Since X is com-
pact and extremally disconnected, the Stone- Cech compactifications βG and βH are homeomorphic
to G and H respectively ([34, Exercises 15G(1) and 19G(2)]). Since α(s) is a homeomorphism of
G onto H, general properties of the Stone- Cech compactification show that α(s) extends to a
homeomorphism h of G onto H.
Let F ⊆ G be the set of fixed points for h; Proposition 2.7 shows that F is clopen in X. Therefore,
{x ∈ dom(α(s)) : α(s)(x) = x} = F ∩ dom(α(s))
is open in X. Thus S acts quasi-freely on X.
(cid:3)
10
Quasi-freeness is intimately related to the extension property. The following result is known in the
case when (C, D) is a regular MASA inclusion with faithful expectation, see [31, Proposition 5.11].
Parts of the proof below follow the proof of [10, Proposition 3.12], but we reproduce it here for
convenience of the reader.
Theorem 2.9. Let (C, D) be a regular inclusion. Then the following statements are equivalent:
a) D has the extension property in C;
b) D is a MASA in C and the action v 7→ βv of the semigroup N(C, D) is a quasi-free action
on D;
c) D is a MASA in C and for each σ ∈ D, the isotropy group of WG(C, D) at σ is the trivial
group.
Proof. (a)⇒(b). Suppose that D has the extension property. Then [3, Corollary 2.7] shows that
D is a MASA in C, and that there exists a conditional expectation E : C → D. Suppose that
v ∈ N(C, D), and σ ∈ D satisfies σ(v∗v) > 0 and βv(σ) = σ. By [10, Proposition 3.12], we have
v∗E(v) ∈ D. Also, if G is the unitary group of D, we have for g ∈ G,
σ(v∗gvg−1) = σ(v∗gv)σ(g−1) = βv(σ)(g)σ(v∗v)σ(g−1) = σ(g)σ(v∗v)σ(g−1) = σ(v∗v).
The extension property and [3, Theorem 3.7] show that E(v) ∈ co{gvg−1 : g ∈ G}, so that
σ(v∗E(v)) = σ(v∗v),
whence σ(v∗E(v)) = σ(v∗v) 6= 0.
Hence there exists an open set U ⊆ D so that σ ∈ U and τ (v∗E(v)) 6= 0 for every τ ∈ U . Since
v∗E(v) ∈ D, we have τ = βv∗E(v)(τ ) = βv∗ (βE(v)(τ )) = βv∗ (τ ) for every τ ∈ U . But β−1
v = βv∗ , so
βv(τ ) = τ for τ ∈ U . Thus {σ ∈ D : σ(v∗v) > 0 and βv(σ) = σ} is open in D, so the semigroup
{βv : v ∈ N(C, D)} acts quasi-freely on D.
Before proving (b)⇒(a), we establish some notation. We use [C, D] for the set {cd − dc : d ∈
D, c ∈ C}. Also, a normalizer v ∈ N(C, D) is a free normalizer if v2 = 0. Kumjian notes in [22],
that any free normalizer belongs to span[C, D], because v(v∗v)1/n − (v∗v)1/nv = v(v∗v)1/n → v.
We turn now to the proof of (b)⇒(a). So assume that (b) holds. We shall prove that
(3)
N(C, D) ⊆ D + span[C, D].
Once this inclusion is established, an application of [3, Theorem 2.4] will show that whenever ρ1
and ρ2 are states of C with ρ1D = ρ2D ∈ D, then for every v ∈ N(C, D), ρ1(v) = ρ2(v). Regularity
of (C, D) then implies that ρ1 = ρ2, so that D has the extension property.
To show (3), fix v ∈ N(C, D). Let F = {σ ∈ D : σ(v∗v) > 0 and βv(σ) = σ}. By hypothesis, F
is an open subset of D. Let ε > 0, and let
Xε := F ∩ {σ ∈ D : σ(v∗v) ≥ ε2}.
Then Xε is a closed subset of D. Let fε ∈ D be such that 0 ≤ fε ≤ I, fεXε ≡ 1 and supp( fε) ⊆ F .
Next let
Yε := F c ∩ {σ ∈ D : σ(v∗v) ≥ ε2}.
Clearly Yε ∩ supp( fε) = ∅.
For σ ∈ Yε, we have βv(σ) 6= σ, so we may find d ∈ D with σ(d) = 1, 0 ≤ d ≤ 1, supp( d) ∩
supp( fε) = ∅ and (vd)2 = 0. Compactness of Yε and a partition of unity argument show that there
exists n ∈ N and a collection of functions {gj}n
j=1 gj, we have:
j=1 ⊆ D such that, with gε =Pn
0 ≤ gε ≤ I,
and gεYε ≡ 1.
(vgj )2 = 0,
0 ≤ gj ≤ I,
supp(gj) ∩ supp( fε) = ∅,
11
Then gεfε = 0, and kv(I − (fε + gε))k < ε. So v = limε→0(vfε + vgε). From Kumjian's observa-
tion, vgε ∈ span[C, D]. Moreover, since the closed support of fε is contained in F , we find that vfε
commutes with D. Hence vfε ∈ D, as D is a MASA in C.
Let εn be a sequence of positive numbers decreasing to 0. We may choose elements fεn ∈ D as
above but with the additional condition that fεm ≤ fεn whenever m < n. For n > m and σ ∈ D,
we have
σ(v∗v)σ(fεn − fεm)2
if σ(fεn − fεm) = 0
if σ(fεn − fεm) 6= 0.
(4)
σ(v∗v(fεn − fεm)2) =(0
Notice that if σ ∈ F ∩ F c, then σ(v∗v) = 0. By continuity ofdv∗v, given δ > 0, there exists a open
set U ⊆ D with U ⊇ F ∩ F c and σ(v∗v) < δ if σ ∈ U . Since supp(fεn − fεm) ⊆ U when n, m are
sufficiently large, it follows from (4) that vfεn is a Cauchy sequence in D, and hence converges to
k ∈ D. Then
v − k = lim
n→∞
(vgεn) ∈ span[C, D],
whence v = k + (v − k) ∈ D + span[C, D]. As noted above, this is sufficient to complete the proof
that D has the extension property. Thus (a) holds.
(b)⇒(c). Fix σ ∈ D, and suppose that [σ, φ, σ] belongs to the isotropy group of WG(C, D) at σ.
Then there is an open set N and v ∈ N(C, D) such that with σ ∈ N ⊆ dom φ ∩ dom βv such that
βvN = φN . By quasi-freeness of the action, there exists a neighborhood N1 of σ contained in N
such that βvN1 = idN1. Thus, [σ, φ, σ] = [σ, id, σ] so that the isotropy group of WG(C, D) at σ is
trivial.
(c)⇒(b). Let v ∈ N(C, D) and suppose that σ ∈ dom(βv) is a fixed point for βv. Then [σ, βv, σ] is
in the isotropy group for WG(C, D) at σ, so that [σ, βv, σ] = [σ, id, σ]. By the definition of the Weyl
groupoid, σ belongs to the interior of {x ∈ dom(βv) : βv(x) = x}. Hence {x ∈ dom(βv) : βv(x) = x}
is an open set in X. As this holds for every v ∈ N(C, D), the action v 7→ βv is quasi-free.
(cid:3)
As an immediate corollary of our work, we have the following.
Theorem 2.10. Suppose (C, D) is a regular MASA inclusion, with D an injective C ∗-algebra.
Then (C, D) has the extension property.
Proof. Results of Dixmier and Gonshor [8, 16] show that D is injective if and only if D is a compact
extremally disconnected space. Now combine Theorems 2.8 with the equivalence of (a) and (b) in
Theorem 2.9.
(cid:3)
Example 2.11. Suppose that M is a von Neumann algebra and D ⊆ M is a MASA. Let C be the
norm-closure of N(M, D). Then (C, D) has the extension property. Note that in particular, when
D is a Cartan MASA in M in the sense of Feldman and Moore [13], then (C, D) is a C ∗-diagonal.
Remark 2.12. The regularity hypothesis in Theorem 2.10 cannot be removed. Indeed, [1, Corol-
lary 4.7] shows that when C is the hyperfinite II1 factor, and D ⊆ C is any MASA, then (C, D) fails
to have the extension property.
2.3. D-modular states and ideals of (C, D). In this final subsection, we make a simple obser-
vation: the action N(C, D) ∋ v 7→ βv ∈ InvO D may be regarded as an action on certain states of
C.
Definition 2.13. Let (C, D) be an inclusion. A state ρ on C is D-modular if for every x ∈ C and
d ∈ D,
ρ(dx) = ρ(d)ρ(x) = ρ(xd).
12
We let Mod(C, D) be the collection of all D-modular states on C; equip Mod(C, D) with the relative
weak-∗ topology. Then Mod(C, D) is closed and hence is compact. Using the Cauchy-Schwartz
inequality, it is easy to see that
Mod(C, D) = {ρ ∈ S(C) : ρD ∈ D}.
For σ ∈ D, let Mod(C, D, σ) be the set of all state extensions of σ, so
Mod(C, D, σ) := {ρ ∈ S(C) : ρD = σ}.
The following simple observation will be useful during the sequel.
Lemma 2.14. Let (C, D) be an inclusion, and v ∈ N(C, D). If σ ∈ dom(βv) and βv(σ) 6= σ, then
ρ(v) = 0 for every ρ ∈ Mod(C, D, σ).
Proof. Let d ∈ D satisfy βv(σ)(d) = 0 and σ(d) = 1. Then for ρ ∈ Mod(C, D, σ), we have
ρ(v) =
ρ(vv∗v)
ρ(v∗v)
=
ρ(dvv∗v)
σ(v∗v)
=
ρ(vv∗dv)
σ(v∗v)
= ρ(v)βv(σ)(d) = 0.
(cid:3)
When ρ ∈ Mod(C, D) and v ∈ N(C, D) satisfies ρ(v∗v) 6= 0, the state βv(ρ) on C given by
βv(ρ)(x) :=
ρ(v∗xv)
ρ(v∗v)
again belongs to Mod(C, D). When there is no danger of confusion, we sometimes simplify no-
tation and write βv(ρ) instead of βv(ρ). Thus N(C, D) also acts on Mod(C, D), and for every
ρ ∈ Mod(C, D), we have βv(ρ)D = βv(ρD).
Definition 2.15. A subset F ⊆ Mod(C, D) is N(C, D)-invariant if for every v ∈ N(C, D) and ρ ∈ F
with ρ(v∗v) 6= 0, we have βv(ρ) ∈ F .
We record the following fact for use in the sequel.
Proposition 2.16. Let (C, D) be a regular inclusion and suppose that F ⊆ Mod(C, D) is N(C, D)-
invariant. Then the set
KF := {x ∈ C : ρ(x∗x) = 0 for all ρ ∈ F }
is a closed, two-sided ideal in C. Moreover, if {ρD : ρ ∈ F } is weak-∗ dense in D, then KF ∩D = (0).
Proof. As KF is the intersection of closed left-ideals, it remains only to prove that KF is a right
ideal. By regularity, it suffices to prove that if x ∈ KF and v ∈ N(C, D), then xv ∈ KF . Let ρ ∈ F .
If ρ(v∗v) 6= 0, then by hypothesis, we obtain ρ(v∗x∗xv) = βv(ρ)(x∗x)ρ(v∗v) = 0. On the other
hand, if ρ(v∗v) = 0, then ρ(v∗x∗xv) ≤ kx∗xk ρ(v∗v) = 0. In either case, we find ρ(v∗x∗xv) = 0. As
this holds for every ρ ∈ F, we find xv ∈ KF , as desired. The final statement is obvious.
(cid:3)
3. Pseudo-Conditional Expectations for Regular MASA Inclusions
As noted in Example 1.2, there exist regular MASA inclusions (C, D) for which no conditional
expectation of C onto D exists. The purpose of this section is to show that nevertheless, there is
always a unique pseudo-expectation for a regular MASA inclusion.
Given a normalizer v, our first task is to connect the dynamics of βv with the ideal structure of
D.
Lemma 3.1. Let (C, D) be an inclusion and suppose v ∈ N(C, D). If d ∈ D and supp( d) ⊆ (fix βv)◦,
then vd = dv. Moreover, if (C, D) is a MASA inclusion, then vd = dv ∈ D.
13
Proof. Note that v∗dv and v∗vd both belong to D. We first show that for every ρ ∈ D,
(5)
Let ρ ∈ D. There are three cases. First suppose ρ(v∗v) = 0. Then as ρ(v∗v) = kvk2
Schwartz inequality gives,
ρ(v∗dv) = ρ(v∗vd).
ρ, the Cauchy-
so (5) holds when ρ(v∗v) = 0.
ρ(v∗dv) = hdv, viρ ≤ kdvkρ kvkρ = 0,
Suppose next that ρ(v∗v) > 0 and βv(ρ)(d) 6= 0. Then βv(ρ) ∈ supp( d), so βv(ρ) ∈ fix βv =
fix βv∗ . Thus, we get βv(ρ) = βv∗ (βv(ρ)) = ρ, and hence ρ(v∗dv) = ρ(v∗v)ρ(d) = ρ(v∗vd).
Finally suppose that ρ(v∗v) > 0 and βv(ρ)(d) = 0. Then ρ(v∗dv) = 0. We shall show that ρ(d) =
0. If not, the hypothesis on d shows that ρ ∈ fix βv. Hence, 0 6= ρ(d) = βv(ρ)(d) = ρ(v∗dv)
ρ(v∗v) = 0,
which is absurd. So ρ(d) = 0, and (5) holds in this case also. Thus we have established (5) in all
cases.
Thus v∗dv = v∗vd. So for every n ∈ N,
0 = v∗dv − v∗vd = v∗(dv − vd) = vv∗(dv − vd) = (vv∗)n(dv − vd).
It follows that for every polynomial p with p(0) = 0, we have, p(vv∗)(dv − vd) = 0. Therefore, for
every n ∈ N,
0 = (vv∗)1/n(dv − vd) = d(vv∗)1/nv − (vv∗)1/nvd.
Since limn→∞(vv∗)1/nv = v, we have vd = dv.
Now suppose that (C, D) is a MASA inclusion. For a ∈ D, we have supp(cda) ⊆ supp( d) ⊆
(fix βv)◦, so we have v(da) = (da)v. Since vd = dv, we get (vd)a = a(vd). Since D is a MASA,
vd ∈ D and the proof is complete.
(cid:3)
Let v ∈ N(C, D). Observe that if d ∈ D and supp( d) ⊆ fix βv, then we actually have supp( d) ⊆
(fix βv)◦. Thus
{d ∈ D : supp( d) ⊆ fix βv} = {d ∈ D : supp( d) ⊆ (fix βv)◦}
is a closed ideal of D isomorphic to C0((fix βv)◦). By the Fuglede-Putnam-Rosenblum commutation
theorem,
(6)
{d ∈ D : dv = vd} = {d ∈ D : dv∗ = v∗d},
and it follows that {d ∈ D : dv = vd ∈ D} is a closed ideal of D. The next proposition shows how
the set (fix βv)◦ may be described algebraically.
Proposition 3.2. Let (C, D) be a MASA inclusion. If v ∈ N(C, D), then
{d ∈ D : supp( d) ⊆ (fix βv)◦} = Dv∗v ∩ {d ∈ D : dv = vd ∈ D}.
Proof. Notice that Dv∗v = {d ∈ D : supp( d) ⊆ supp(dv∗v)}. Since (fix βv)◦ ⊆ supp(dv∗v), Lemma 3.1
{d ∈ D : supp( d) ⊆ (fix βv)◦} ⊆ Dv∗v ∩ {d ∈ D : dv = vd ∈ D}.
shows that
Now suppose that d ∈ Dv∗v ∩ {h ∈ D : hv = vh ∈ D} and d 6= 0. Let ρ0 ∈ supp( d) and set
r := ρ0(d). Then r > 0. Put G = {ρ ∈ D : ρ(d) > r/2}. We show that G ⊆ fix βv. Fix
ρ ∈ G. Since d ∈ Dv∗v, we have supp( d) ⊆ supp((dv∗v)), so ρ(v∗v) 6= 0. Since d belongs to the ideal
{f ∈ D : f v = vf ∈ D}, we find (using (6)) that for every a ∈ D,
ρ(v∗av)
ρ(v∗v)
=
ρ(v∗avd)
ρ(v∗v)ρ(d)
ρ(v∗(ad)v)
ρ(v∗v)ρ(d)
=
ρ(adv∗v)
ρ(v∗v)ρ(d)
= ρ(a).
βv(ρ)(a) =
=
14
It follows that G ⊆ fix βv. Since G is an open subset of D with ρ0 ∈ G, we have ρ0 ∈ (fix βv)◦. So
supp( d) ⊆ (fix βv)◦, as desired.
(cid:3)
We need some notation.
Notation. Let (C, D) be an inclusion.
a) For a closed ideal J in D, we let
J⊥ := {d ∈ D : dg = 0 for all g ∈ J}
denote the complement of J in the lattice of closed ideals of D.
b) Given any two closed ideals J1 and J2 in D, J1 ∨ J2 denotes the closed ideal generated by
J1 and J2.
c) For S ⊆ D, hSiD denotes the closed two-sided ideal of D generated by the set S. When the
context is clear, we drop the subscript and simply use hSi.
d) For v ∈ N(C, D), let
(7)
Jv := {d ∈ D : vd = dv ∈ D} ∩ hv∗vi
and Kv := hv∗vi⊥ ∨ h{v∗hv − hv∗v : h ∈ D}i .
Recall that an ideal J in a C ∗-algebra C is an essential ideal if J ∩ L 6= (0) for every closed
two-sided ideal (0) 6= L ⊆ C.
We will show Jv ∨ Kv is an essential ideal in D. It is easy to see that Jv ⊆ K ⊥
v . If equality holds,
then the fact that Jv ∨ Kv is an essential ideal follows readily. However, we have not found a simple
proof of this fact, so we proceed along different lines.
Proposition 3.3. Let (C, D) be a regular MASA inclusion and let v ∈ N(C, D). Then Jv ∨ Kv is
an essential ideal in D.
Proof. We shall show that Jv ∨ Kv is essential by showing that (Jv ∨ Kv)⊥ = (0). Suppose that
d ∈ D and d(Jv ∨ Kv) = 0.
First we show
(8)
Indeed, if ρ ∈ D and ρ /∈ suppdv∗v, then we may find h ∈ D such that h(ρ) = 1 and h(σ) = 0 for
every σ ∈ suppdv∗v. Then h ∈ hv∗vi⊥ ⊆ Kv, so dh = 0. As ρ(h) = 1, this shows that ρ /∈ supp d.
Next, we claim that
Thus (8) holds.
supp d ⊆ suppdv∗v.
(9)
d ∈ J ⊥
supp d ∩ suppdv∗v ⊆ (fix βv)◦.
d ∈ K ⊥
v , this gives ρ(v∗hv) = ρ(v∗v)ρ(h) for every h ∈ D. Since ρ(v∗v) 6= 0 by hypothesis, we have
Let ρ ∈ supp d ∩ suppdv∗v. For every h ∈ D, we have v∗hv − v∗vh ∈ Kv. Since ρ(d) 6= 0 and
ρ ∈ fix βv. As supp d ∩ suppdv∗v is an open subset of D, we obtain (9).
Suppose that ρ ∈ (fix βv)◦. By Proposition 3.2 there exists h ∈ Jv so that ρ(h) = 1. Since
v , we obtain ρ(d) = 0. Hence
(10)
(fix βv)◦ ∩ supp d = ∅.
Combining (8), (9), and (10) we obtain,
But suppdv∗v \ suppdv∗v has empty interior, so supp d = ∅. Therefore, d = 0 as desired.
15
supp d ⊆ suppdv∗v \ suppdv∗v.
(cid:3)
The following is probably well known, but we include it for completeness. Recall that if A is a
unital injective C ∗-algebra, then any bounded increasing net xλ of self-adjoint elements of A has a
least upper bound in A, which we denote by supA xλ.
Lemma 3.4. Let D be a unital, abelian C ∗-algebra, let (I(D), ι) be an injective envelope for D and
suppose that J ⊆ D is a closed ideal. Let J +
1 be the positive part of the unit ball of J and regard
J +
1 as a net indexed by itself. Put P := supI(D) ι(J +
If in addition, J is an essential ideal of D, the following hold:
1 ). Then P is an projection in I(D).
i) if a, b ∈ I(D) and aι(h) = bι(h) for every h ∈ J, then a = b;
ii) P = I.
Proof. The mapping x ∈ J +
x ∈ J +
P 2 ≤ P . So P is a projection.
1 }. So P 2 is also an upper bound for ι(J +
1 7→ x1/2 ∈ J +
1 is an order isomorphism of J +
1 , so P = supI(D){ι(x1/2) :
1 ). Hence P ≤ P 2. But as kP k ≤ 1, we have
Now assume J is an essential ideal, and suppose a, b ∈ I(D) satisfy (a − b)ι(h) = 0 for every
h ∈ J. By Hamana-regularity of ι(D) in I(D), we have a− b = supI(D){d ∈ D : 0 ≤ ι(d) ≤ a− b}.
But K := {d ∈ D : ι(d) ∈ ha − biI(D)} is a closed ideal of D with K ⊆ J ⊥. Hence K = 0, so
a − b = 0.
Notice that if d ∈ J +
part (i), P = I when J is an essential ideal.
1 , then ι(d)P = ι(d). It follows that for every d ∈ J, ι(d)P = ι(d). By
(cid:3)
The following extension of Definition 2.13 will be useful.
Definition 3.5. Let (C, D) be an inclusion and B an algebra.
(1) A linear map ∆ : C → B is D-modular (or more simply modular ) if for every x ∈ C and
d ∈ D,
∆(xd) = ∆(x)∆(d)
and ∆(dx) = ∆(d)∆(x).
(2) A homomorphism θ : D → B is D-thick in B if for every non-zero element b ∈ B, the ideal
{d ∈ D : bθ(d) = 0}⊥ is a non-zero ideal of D.
When B is abelian, notice that the restriction of a D-modular map to D is a homomorphism.
The next lemma gives an example which will be used in the proof of Theorem 3.10.
Lemma 3.6. Let D be an abelian C ∗-algebra and let (I(D), ι) be an injective envelope for D. Then
ι is D-thick in I(D).
Proof. Suppose b ∈ I(D) is non-zero. The Hamana regularity of I(D) ensures that there exists a
non-zero h ∈ D such that 0 ≤ ι(h) ≤ b∗b. If d ∈ D satisfies ι(d)b = 0, then supp(dι(d)) ∩ supp(b) = ∅.
Since supp(dι(h)) ⊆ supp(b), we get ι(dh) = 0, whence h ∈ {d ∈ D : ι(d)b = 0}⊥.
Our interest in D-thick homomorphisms will be with the restrictions of D-modular maps to D.
(cid:3)
The following lemma will be useful.
Lemma 3.7. Let (C, D) be a regular MASA inclusion, let B be a unital abelian Banach algebra
and let v ∈ N(C, D). For i = 1, 2, suppose ∆i : C → B are bounded D-modular maps such that
∆1D = ∆2D and set ι := ∆iD. Then for every h ∈ Jv ∨ Kv,
(11)
∆1(vh) = ∆2(vh).
In fact,
a) for every h ∈ Kv, ∆1(vh) = 0 = ∆2(vh);
b) for every h ∈ Jv, ∆1(vh) = ι(vh) = ∆2(vh).
Moreover, if ι is also D-thick in B, then ∆1 = ∆2.
16
Proof. For (a), we consider two cases. First, if h ∈ hv∗vi⊥, then for i = 1, 2,
∆i(vh) = lim
n→∞
∆i(v(v∗v)1/nh) = 0.
Second, suppose that h ∈ {v∗dv − v∗vd : d ∈ D}. Then for some d ∈ D,
∆i(vh) = ∆i(v(v∗dv − v∗vd)) = ∆i(vv∗dv) − ∆i(vv∗vd)
= ι(dv∗v)∆i(v) − ∆i(vv∗v)ι(d) = ι(d)∆i(v)ι(v∗v) − ι(d)∆i(vv∗v) = 0.
Thus ∆i(vh) = 0 for all h in a generating set for Kv. Since ∆i are D-modular and bounded, we
obtain (a).
Next, suppose that h ∈ Jv. Then since vh ∈ D,
(12)
This gives (b).
∆1(vh) = ι(vh) = ∆2(vh).
Parts (a) and (b) imply that ∆1(vh) = ∆2(vh) for all h in a generating set for Jv ∨ Kv, so
boundedness and D-modularity of ∆i yields (11).
Now suppose that ι is D-thick in B and v ∈ N(C, D). Let b = ∆1(v) − ∆2(v) and choose any
h ∈ {d ∈ D : bι(d) = 0}⊥ ∩ (Jv ∨ Kv). Since h ∈ Jv ∨ Kv, we have bι(h) = ∆1(v)ι(h) − ∆2(v)ι(h) =
∆1(vh) − ∆2(vh) = 0. Since h ∈ {d ∈ D : bι(d) = 0}⊥, we get h2 = 0. As D is abelian, h = 0.
This shows that {d ∈ D : bι(d) = 0}⊥ ∩ (Jv ∨ Kv) = (0). Since Jv ∨ Kv is an essential ideal,
{d ∈ D : bι(d) = 0}⊥ = (0). Since ι is D-thick in B, we see that b = 0. Hence ∆1(v) = ∆2(v).
Since this holds for every v ∈ N(C, D), regularity of (C, D) yields ∆1 = ∆2.
(cid:3)
Lemma 3.7 has an interesting consequence for uniqueness of extensions of pure states on D to C,
which we now present. This result generalizes a result found in [31], however, the proof is rather
different. Notice that Theorem 3.8 holds when C is separable or when there is a countable subset
X ⊆ N(C, D) such that C is the C ∗-algebra generated by D and X. We shall use Theorem 3.8 in
the proof of Theorem 9.2.
Theorem 3.8. Suppose (C, D) is a regular MASA inclusion and that N ⊆ N(C, D) is a countable
set such that the norm-closed D-bimodule generated by N is C. Let
U := {σ ∈ D : σ has a unique state extension to C}.
Then U is dense in D.
Proof. For each v ∈ N , let Gv := {σ ∈ D : σJv∨Kv 6= 0}. Clearly Gv is open in D and since Jv ∨ Kv
is an essential ideal in D, Gv is dense in D. Baire's theorem shows that
P := \v∈N
Gv
is dense in D.
Let σ ∈ P and suppose for i = 1, 2, ρi are states on C such that ρiD = σ. The Cauchy-Schwartz
inequality shows that ρi : C → C are D-modular maps.
Fix v ∈ N . Since σ ∈ Gv, we may find h ∈ Jv ∨ Kv such that σ(h) = 1. By Lemma 3.7 we have
ρ1(v) = ρ1(v)σ(h) = ρ1(vh) = ρ2(vh) = ρ2(v)σ(h) = ρ2(v).
Since N generates C as a D-bimodule and ρi are D-modular, we see that ρ1 = ρ2. Hence P ⊆ U,
and the proof is complete.
(cid:3)
We now show that any regular MASA inclusion has a unique completely positive mapping E :
C → I(D) which extends the inclusion mapping of D into I(D).
17
Lemma 3.9. Let (C, D) be an inclusion and (I(D), ι) an injective envelope of D. Let E : C → I(D)
be a pseudo-expectation for ι. Then E is D-modular.
Proof. Let ρ ∈ [I(D) and put σ = ρ ◦ E. Then σD ∈ D, so the Cauchy-Schwartz inequality implies
that for every x ∈ C and d ∈ D, σ(xd) = σ(x)σ(d). Hence,
ρ(E(xd)) = σ(xd) = σ(x)σ(d) = ρ(E(x))ρ(E(d)) = ρ(E(x)E(d)).
As this holds for every ρ ∈ [I(D), we obtain E(xd) = E(x)E(d). The proof that E(dx) = E(d)E(x)
is similar.
(cid:3)
Theorem 3.10. Let (C, D) be a regular MASA inclusion, and let (I(D), ι) be an injective envelope
of D. Then there exists a unique pseudo-expectation E : C → I(D) for ι. Furthermore, suppose
v ∈ N(C, D). Then
a) E(vh) = ι(vh) for every h ∈ Jv;
b) E(vh) = 0 for every h ∈ Kv;
c) E(v)2 = ι(v∗v)P , where P := supI(D)(ι((Jv)+
1 )).
Proof. The injectivity of I(D) guarantees the existence of a completely positive unital map E :
C → I(D) such that for every d ∈ D,
(13)
E(d) = ι(d).
Lemma 3.7 and Lemma 3.6 imply that E is unique.
Now suppose that v ∈ N(C, D). Since ED = ι, parts (a) and (b) follow from Lemma 3.7.
We turn now to (c). Observe that Kv ⊆ J ⊥
v . Indeed Jv ⊆ hv∗vi, so if h ∈ hv∗vi⊥ then hJv = 0.
On the other hand, if d ∈ D and h = v∗dv − v∗vd, then for g ∈ Jv, we have hg = (v∗dv − v∗vd)g =
v∗dgv − v∗vdg = 0 because dg ∈ Jv. Thus hJv = 0 for every h in a generating set for Kv, so
Kv ⊆ J ⊥
v .
Now let h ∈ Kv. By [20, Corollary 4.10], we have
ι(h)∗P ι(h) = ι(h)∗"sup
I(D)
ι((Jv)+
1 )# ι(h) = sup
I(D)
ι(h∗(Jv)+
1 h) = 0.
Since P is a projection, it follows that ι(h)P = 0. Hence ι(v∗v)P ι(h∗h) = 0. Next, by part (b), we
have for every h ∈ Kv, E(v)2ι(h∗h) = ι(h∗)E(v)∗E(v)ι(h) = ι(h∗)E(v)∗E(vh) = 0. Therefore, for
h ∈ Kv,
(14)
E(v)2ι(h∗h) = ι(v∗v)P ι(h∗h) = 0.
Since E(vh) = ι(vh) for every h ∈ Jv, we see that for h ∈ Jv,
(15)
E(v∗)E(v)ι(h∗h) = E((vh)∗)E(vh) = ι(h∗v∗)ι(vh) = ι(v∗v)ι(h∗h) = ι(v∗v)P ι(h∗h).
Combining (14) and (15), we see that for every h ∈ Jv ∨ Kv,
E(v)∗E(v)ι(h∗h) = ι(v∗v)P ι(h∗h).
Then E(v)∗E(v) = ι(v∗v)P by Lemma 3.4, so we have (c).
(cid:3)
The following "dual" to Theorem 3.10 is now easily established. Notice that in the context of
Theorem 3.10, when ρ ∈ [I(D), E#(ρ) = ρ ◦ E ∈ Mod(C, D).
Theorem 3.11. Let (C, D) be a regular MASA inclusion, let (I(D), ι) be an injective envelope for
D, and let E be the pseudo-expectation for ι. The map E# : [I(D) → Mod(C, D) is the unique
continuous map of [I(D) into Mod(C, D) such that for every ρ ∈ [I(D), E#(ρ)D = ρ ◦ ι.
18
Proof. Clearly E# has the stated property, so we need only prove uniqueness.
Suppose that ℓ is a continuous map of [I(D) into Mod(C, D) such that for every ρ ∈ [I(D),
ℓ(ρ)D = ρ ◦ ι. For x ∈ C, define a function φx : [I(D) → C by φx(ρ) = ℓ(ρ)(x). Since ℓ is
continuous, φx is continuous. Hence there exists a unique element E1(x) ∈ I(D) such that φx
is the Gelfand transform of E1(x). Using the fact that Mod(C, D) ⊆ S(C), we find E1 is linear,
bounded, unital and positive. Since I(D) is abelian, E1 is completely positive. For d ∈ D we
have ρ(E1(d)) = ℓ(ρ)(d) = ρ(ι(d)). Therefore, E1D = ι, so E1 is a pseudo-expectation for ι. By
Theorem 3.10, E1 = E, hence ℓ = E#.
(cid:3)
Definition 3.12. Let (C, D) be a regular MASA inclusion, let (I(D), ι) be a C ∗-envelope for D,
and let E be the (unique) pseudo-expectation for ι. Define
Ss(C, D) := {ρ ◦ E : ρ ∈ [I(D)}.
We shall call states belonging to Ss(C, D) strongly compatible states. Clearly, Ss(C, D) is a closed
subset of Mod(C, D). Observe that D = {τ D : τ ∈ Ss(C, D)}; this is because D = {ρ◦ι : ρ ∈ [I(D)}.
Let r : Mod(C, D) → D be the restriction map, r(ρ) = ρD. We now show that Ss(C, D) is the
unique minimal closed subset of Mod(C, D) for which r is onto. In a certain sense, this allows us
to determine Ss(C, D) without the use of the pseudo-expectation.
Theorem 3.13. Let (C, D) be a regular MASA inclusion and suppose F ⊆ Mod(C, D) is closed and
r(F ) = D. Then Ss(C, D) ⊆ F .
Suppose further that there exists a countable subset N ⊆ N(C, D) such that the norm-closed
D-bimodule generated by N is C and set
U(C, D) := {ρ ∈ Mod(C, D) : ρD has a unique state extension to C}.
Then
Ss(C, D) = U(C, D)
w-∗
.
Proof. Since F is closed and Mod(C, D) is compact, F is compact and Hausdorff. As [I(D) is
projective (in the category of compact Hausdorff spaces and continuous maps) and r maps F onto
D, there exists a continuous map ℓ : [I(D) → F such that ι# = r ◦ ℓ. Let ǫ : F → Mod(C, D)
be the inclusion map. Then ℓ′ := ǫ ◦ ℓ is a continuous map of [I(D) into Mod(C, D) such that
r ◦ ℓ′ = ι#. Theorem 3.11 shows that ℓ′ = E#. Therefore, the range of E# is contained in F , that
is, Ss(C, D) ⊆ F .
Suppose now that there is a countable subset N ⊆ N(C, D) which generates C as a D-bimodule.
Theorem 3.8 implies that D = r(U(C, D)), so we have Ss(C, D) ⊆ U(C, D). To complete the proof,
observe that Ss(C, D) is closed and U(C, D) ⊆ Ss(C, D).
(cid:3)
The following result shows that, in the terminology of Section 4, each element of Ss(C, D) is a
compatible state.
Proposition 3.14. Let (C, D) be a regular MASA inclusion and let σ ∈ Ss(C, D). Then for every
v ∈ N(C, D),
σ(v)2 ∈ {0, σ(v∗v)}.
Proof. Let ρ ∈ [I(D) be such that σ = ρ ◦ E and suppose v ∈ N(C, D) is such that σ(v) 6= 0. Then
0 6= ρ(E(v))2, so by part (c) of Theorem 3.10, ρ(P ) 6= 0. By Lemma 3.4, P is a projection, so
ρ(P ) = 1. Thus, ρ(E(v))2 = ρ(ι(v∗v)) = ρ(E(v∗v)) = σ(v∗v).
(cid:3)
19
Our next goal is Theorem 3.21 below, which shows that Ss(C, D) is an N(C, D)-invariant subset of
Mod(C, D). In the case that C is countably generated as a D-bimodule, this follows from the second
part of Theorem 3.13, but we have not found a proof in the general case using Theorem 3.13. The
fact that the left kernel of the pseudo-expectation is an ideal will follow easily from Theorem 3.21,
see Theorem 3.22.
Our route to Theorem 3.21 involves a study of the Gelfand support of E(v) for v ∈ N(C, D); we
also obtain a formula for the Gelfand transform of E(v). We begin with three lemmas on properties
of projective covers.
Lemma 3.15. Let X be a compact Hausdorff space and let (P, f ) be a projective cover for X.
a) If G ⊆ X is an open set, then (f −1(G), f f −1(G)) is a projective cover for G.
b) If Q ⊆ P is clopen, then f (Q)◦ is dense in f (Q) and Q = f −1(f (Q)◦).
Proof. Before beginning the proof, observe that if ι : C(X) → C(P ) is given by ι(d) = d ◦ f , then
(C(P ), ι) is an injective envelope for C(X).
a) Let Y := f −1(G). Then Y is compact, so f (Y ) is a closed subset of X which contains
G. Hence G ⊆ f (Y ). On the other hand, f −1(G) is a closed subset of P containing f −1(G), so
Y ⊆ f −1(G). Hence f (Y ) ⊆ G. Therefore (Y, f ) is a cover for G.
Since P is projective, it is Stonean, and hence Y is clopen in P . Since clopen subsets of projective
spaces are projective, Y is projective. The proof of part (a) will be complete once we show that
(Y, f Y ) is a rigid cover for G (see [17, Theorem 2.16]).
So suppose that h : Y → Y is continuous and f ◦ h = f Y . Define h : P → P by
h(t) =(h(t)
t
if t ∈ Y
if t /∈ Y .
Since Y is clopen in P , h is continuous. Moreover, f ◦ h = f , so by the rigidity of (P, f ), we see
that h is the identity map on P . Therefore h is the identity map on Y , which shows that (Y, f Y )
is a rigid cover.
b) The case when Q = ∅ is trivial, so we assume Q is non-empty. Let
M := {d ∈ C(X) : 0 ≤ ι(d) ≤ χQ}
and set G := [d∈M
supp(d).
Then G is non-empty because χQ 6= 0 and (C(P ), ι) is Hamana regular. Notice that f −1(G) ⊆ Q:
indeed, if p ∈ f −1(G), then there exists d ∈ M such that f (p) ∈ supp(d), so 0 < d(f (p)) ≤ χQ(p),
whence p ∈ Q.
As P is extremally disconnected, W := f −1(G) is a clopen subset of P . We will show that
W = Q. Clearly W ⊆ Q. If Q \ W 6= ∅, then Q \ W is a clopen subset of P , so we may find a non-
zero d1 ∈ C(X) with 0 ≤ ι(d1) ≤ χQ\W ≤ χQ. But then d1 ∈ M, so supp(ι(d1)) ⊆ f −1(G) ⊆ W,
contradicting 0 ≤ ι(d1) ≤ χQ\W . Hence W = Q.
By part (a), (W, f W ) is a cover for G. Thus, f (W ) = f (Q) = G. As f −1(G) ⊆ Q, we have
G ⊆ f (Q)◦, so that f (Q)◦ is dense in f (Q).
Finally, put W1 := f −1(f (Q)◦). Since G ⊆ f (Q)◦, we have Q = W ⊆ W1. Part (a) again shows
that (W1, f W1) is a projective cover for f (Q)◦ = f (Q), so in particular, this cover is essential. The
inclusion map α of W into W1 satisfies f (α(W )) = f (Q). Because (W1, f W1) is an essential cover
of f (Q), we conclude α is onto. Thus W = W1, and the proof of (b) is complete.
(cid:3)
We leave the proof of the following lemma to the reader.
20
[G∈T
G = H1 = [G∈T
G we have [G∈T
f −1(G) = f −1(H1) = [G∈T
f −1(G).
Lemma 3.16. Suppose that for i ∈ {1, 2}, Xi is a compact Hausdorff space and that φ : X1 →
X2 is a homeomorphism. Let (Ci, fi) be a projective cover for Xi. Then there exists a unique
homeomorphism Φ : C1 → C2 such that f2 ◦ Φ = φ ◦ f1.
Our final lemma on projective covers is a strengthening of Lemma 3.16: rather than extending
a homeomorphism to the injective envelope, partial homeomorphisms are extended.
Lemma 3.17. Let (P, f ) be a projective cover for the compact Hausdorff space X and suppose
h ∈ InvO(X) is a partial homeomorphism. Then there exists a unique partial homeomorphism
I(h) ∈ InvO(P ) such that:
a) dom(I(h)) = f −1(dom(h)), range(I(h)) = f −1(range(h)) and
b) h ◦ f = f ◦ I(h).
Proof. Let H1 = dom(h) and H2 = range(h). Set
T := {G ⊆ H : G is open in X and G ⊆ H1}.
Let G ∈ T. Lemma 3.15 shows that (f −1(G), f ) and (f −1(h(G)), f ) are projective covers for G
and h(G). By Lemma 3.16, there exists a unique homeomorphism hG : f −1(G) → f −1(h(G)) such
that
We now let I(h) be the inductive limit of {hG}G∈T. Here is an outline.
h ◦ f f −1(G) = f ◦ hG.
Since
Let G1, G2 ∈ T and suppose G1 ∩ G2 6= ∅. Put Q := f −1(G1) ∩ f −1(G2). Then Q is a clopen set
in P , so by Lemma 3.15, Q = f −1(f (Q)◦). Note that f (Q)◦ ∈ T. Hence
Thus, for i = 1, 2,
h ◦ f Q = h ◦ f f −1(f (Q)◦) = f ◦ hf (Q)◦ .
f ◦ hGiQ = h ◦ f Q.
This means that given p ∈ P , we may define I(h)(p) = hG(p) where G is any element of T containing
p.
Clearly I(h) satisfies (a) and (b). If H is another such map, then for every G ∈ T, the restrictions
of H and I(h) to f −1(G) are both equal to hG; this gives uniqueness of I(h). The continuity and
bijectivity of I(h) are left to the reader.
(cid:3)
When (I(D), ι) is an injective envelope for D, ([I(D), ι#[I(D)
) is a projective cover for D. In the
following technical result, we will simply write ι# instead of ι#[I(D)
Proposition 3.18. Suppose that (C, D) is a regular MASA inclusion and v ∈ N(C, D). Let (I(D), ι)
be an injective envelope for D, and let E be the pseudo-expectation for ι. Then
.
(16)
(ι#)−1((fix βv)◦) ⊆ supp([E(v)) ⊆ (ι#)−1(fix βv) and
(ι#)−1((fix βv)◦) = supp([E(v)).
Moreover, if ρ ∈ (ι#)−1((fix βv)◦), then for any d ∈ Jv with ρ(ι(d)) 6= 0,
(17)
ρ(E(v)) =
ρ(ι(vd))
ρ(ι(d))
.
21
Proof. Suppose that ρ ∈ [I(D) and ρ ◦ ι ∈ (fix βv)◦. Then ρ(ι(v∗v)) 6= 0 (as fix βv ⊆ dom(βv) =
supp(dv∗v)). By Proposition 3.2 there exists d ∈ Jv such that ρ(ι(d)) 6= 0. Since vd ∈ D,
0 6= ρ(ι(d∗d))ρ(ι(v∗v)) = ρ(ι(d∗v∗vd)) = ρ(ι(vd))2.
Hence
ρ(E(v)) =
ρ(E(v))ρ(ι(d))
ρ(ι(d))
=
ρ(E(vd))
ρ(ι(d))
=
ρ(ι(vd))
ρ(ι(d))
6= 0.
Thus we obtain (17) and also (ι#)−1((fix βv)◦) ⊆ supp([E(v)).
We next show
(18)
Suppose that ρ ∈ supp([E(v)). By Proposition 3.14, 0 6= ρ(E(v))2 = ρ(E(v∗v)) = ρ(ι(v∗v)).
Hence ρ ◦ ι ∈ dom(βv). Let d ∈ D be such that d ≥ 0 and ρ(ι(d)) 6= 0. Then ρ(E(d1/2v)) =
ρ(ι(d))1/2ρ(E(v)) 6= 0. Then
supp([E(v)) ⊆ (ι#)−1(fix βv).
(19)
βv(ρ ◦ ι)(d) =
ρ(ι(v∗dv))
ρ(ι(v∗v))
=
ρ(E(v∗dv))
ρ(ι(v∗v))
=
ρ(E((d1/2v)∗(d1/2v)))
ρ(ι(v∗v))
=
ρ(E(d1/2v))2
ρ(ι(v∗v))
6= 0.
(The last equality in (19) follows from Proposition 3.14.) As (19) holds for every d ∈ D+ with
ρ(ι(d)) 6= 0, we conclude that βv(ρ ◦ ι) = ρ ◦ ι. This gives (18).
The first paragraph of the proof gives (ι#)−1((fix βv)◦) ⊆ supp([E(v)). Let
Then Q is a clopen set, and I is a closed ideal in D.
Q := supp([E(v)) \ (ι#)−1((fix βv)◦) and I := {d ∈ D : suppdι(d) ⊆ Q}.
We claim that I ⊆ (Jv ∨ Kv)⊥. To see this, fix d ∈ I. Proposition 3.2 shows that for any h ∈ Jv,
supp(dι(h)) ⊆ (ι#)−1((fix(βv)◦)); thus dh = 0 for any h ∈ Jv. Now we show dKv = 0. Suppose
h ∈ Kv. By Theorem 3.10(b), E(v)ι(h) = 0, so dι(h) vanishes on supp([E(v)). Continuity implies
that dι(h) vanishes on Q as well. Therefore, ι(d)ι(h) = 0, so dh = 0. As dh = 0 for all h belonging
As Jv ∨ Kv is an essential ideal, (Jv ∨ Kv)⊥ = (0), whence I = (0). The Hamana regularity of
(cid:3)
to a generating set for Jv ∨ Kv, d ∈ (Jv ∨ Kv)⊥. The claim follows.
(I(D), ι) now implies that Q = ∅, which completes the proof.
Notation 3.19. Let (C, D) be a regular MASA inclusion, and let v ∈ N(C, D). Define v :
(fix βv)◦ → C as follows. Given σ ∈ (fix βv)◦, choose d ∈ Jv so that σ(d) 6= 0 and set
v(σ) =
σ(vd)
σ(d)
.
Proposition 3.18 shows this is well-defined, and it is easy to show that v is a bounded continuous
function on (fix βv)◦. Extend v to a bounded Borel function on D by defining it to be zero off
(fix βv)◦; we denote this extension by v as well.
Remark 3.20. Take (I(D), ι) to be the Dixmier algebra of D and ι to be the map which takes
d ∈ D to the equivalence class of d in I(D). Then E(v) is the equivalence class of the bounded
Borel function v in the Dixmier algebra.
Theorem 3.21. Suppose that (C, D) is a regular MASA inclusion. Then Ss(C, D) is a compact
N(C, D)-invariant subset of Mod(C, D) and the restriction mapping Ss(C, D) ∋ ρ 7→ ρD is a con-
tinuous surjection.
22
In fact, given v ∈ N(C, D) and ρ ∈ Ss(C, D) such that ρ(v∗v) 6= 0, let τ ∈ [I(D) satisfy ρ = τ ◦ E.
Then βv(ρ) = I(βv)(τ ) ◦ E.
Proof. As noted following Definition 3.12, Ss(C, D) is a closed subset of Mod(C, D) and r(Ss(C, D)) =
D. So Ss(C, D) is compact, and the weak-∗-weak-∗ continuity of r is clear.
Now let ρ ∈ Ss(C, D) and fix τ ∈ [I(D) so that ρ = τ ◦ E. Let v ∈ N(C, D) be such that
ρ(v∗v) 6= 0. Then τ (E(v∗v)) = τ (ι(v∗v)) 6= 0, so Lemma 3.17 shows that τ ∈ dom(I(βv)).
For λ ∈ dom(I(βv)), define states on C by
µλ(x) =
λ(E(v∗xv))
λ(ι(v∗v))
and µ′
λ(x) = I(βv)(λ)(E(x)).
(Observe that µτ = βv(ρ).) Note that
µλD = βv(λ ◦ ι) = µ′
λD.
Hence for d ∈ D and x ∈ C, we have
µλ(x)βv(λ ◦ ι)(d) = µλ(xd)
and µ′
λ(x)βv(λ ◦ ι)(d) = µ′
λ(xd).
In particular, if w ∈ N(C, D) and d ∈ Jw we have
(20)
µλ(w)βv(λ ◦ ι)(d) = µλ(wd) = βv(λ ◦ ι)(wd) = µ′
λ(w)βv(λ ◦ ι)(d).
To complete the proof, it suffices to show that for every w ∈ N(C, D), µτ (w) = µ′
τ (w) by proving the following two statements:
w ∈ N(C, D). We show that µτ (w) = µ′
λ(wd) = µ′
τ (w). So fix
τ (w) 6= 0, then µτ (w) = µ′
τ (w); and
τ (w) = µτ (w).
(1) if µ′
(2) if µτ (w) 6= 0, then µ′
If µ′
τ (w) 6= 0, then I(βv)(τ ) ∈ supp( [E(w)), so Proposition 3.18 implies that there exists a net
ρα ∈ (ι#)−1((fix βw)◦) such that ρα → I(βv)(τ ). As range(I(βv)) is an open subset of [I(D), we may
assume that ρα ∈ range(I(βv)) for every α. Put τα = I(βv)−1(ρα), so (τα)α is a net in dom(I(βv))
such that τα → τ . Since ρα ◦ι ∈ (fix βw)◦, given α we may find dα ∈ Jw such that 0 6= ρα(ι(dα)). But
ρα(ι(dα)) = I(βv)(τα)(ι(dα)) = βv(τα ◦ ι)(dα). Taking λ = τα in (20) shows that µτα(w) = µ′
τα(w).
Continuity of the maps λ 7→ µλ and λ 7→ µ′
λ gives µτ (w) = µ′
Turning to (2), suppose µτ (w) 6= 0. Then τ ∈ supp( \E(v∗wv)), so there exists a net τα ∈
(ι#)−1((fix βv∗wv)◦) with τα → τ . Then τα ◦ ι ∈ dom(βv∗wv) ⊆ dom βv. Thus, for a given α, we may
find a neighborhood N of τα ◦ ι with N ⊆ (fix βv∗wv)◦ ∩ dom(βv). Now for each y ∈ N , we have
βv∗wv(y) = y, so βw((βv)(y)) = βv(y). Hence βv(N ) ⊆ fix(βw). Therefore βv(τα ◦ ι) ∈ fix(βw)◦.
So if d ∈ Jw satisfies βv(τα ◦ ι)(d) 6= 0, then (20) gives µ′
τα(w) = µτα(w). Continuity again gives
µ′
τ (w) = µτ (w).
Thus both (1) and (2) hold, and the proof is complete.
τ (w).
(cid:3)
We now show the left kernel of the pseudo-expectation on a regular MASA inclusion is an ideal
which is the unique ideal which is maximal with respect to being disjoint from D.
Theorem 3.22. Let (C, D) be a regular MASA inclusion. Then the left kernel of the pseudo-
expectation E,
L(C, D) := {x ∈ C : E(x∗x) = 0}
is an ideal of C such that L(C, D) ∩ D = (0).
Moreover, if K ⊆ C is an ideal such that K ∩ D = (0), then K ⊆ L(C, D).
In the
Proof. Theorem 3.21 shows that Ss(C, D) satisfies the hypotheses of Proposition 2.16.
notation of Proposition 2.16, we have L(C, D) = KSs(C,D), so L(C, D) is a norm-closed two-sided
ideal of C. If x ∈ L(C, D) ∩ D, then 0 = E(x∗x) = ι(x∗x). As ι is one-to-one, x = 0.
23
Suppose now that K ⊆ C is an ideal with K ∩ D = (0). Let C1 = C/K, and let q : C → C1 be the
quotient map. Since K ∩ D = (0), qD is faithful, so we may regard D as a subalgebra of C1. Let
F := {ρ ◦ q : ρ ∈ Mod(C1, D)}. Then F is a closed subset of Mod(C, D), and the restriction map,
f ∈ F 7→ f D maps F onto D. By Theorem 3.13, Ss(C, D) ⊆ F . Hence every element of Ss(C, D)
annihilates K, so K ⊆ L(C, D).
(cid:3)
The ideal L(C, D) behaves reasonably well with respect to certain regular ∗-homomorphisms, as
the next result shows.
Corollary 3.23. Suppose for i = 1, 2 that (Ci, Di) are regular MASA inclusions, and that α :
(C1, D1) → (C2, D2) is a regular ∗-homomorphism such that αD is one-to-one. Then
(21)
{x ∈ C1 : α(a) ∈ L(C2, D2)} ⊆ L(C1, D1).
Proof. The set {x ∈ C1 : α(x) ∈ L(C2, D2)} is an ideal of C1 whose intersection with D1 is
trivial.
(cid:3)
Remark 3.24. We expect that equality holds in (21) if for every 0 ≤ h ∈ D2, h = supD2{α(d) :
d ∈ D1 and 0 ≤ α(d) ≤ h}. Also, it would not be surprising if this condition characterized equality
in (21).
Since every Cartan inclusion (C, D) satisfies L(C, D) = (0), we make the following definition.
Definition 3.25. A virtual Cartan inclusion is a regular MASA inclusion such that L(C, D) = (0).
4. Compatible States
Since the extension property does not always hold for an inclusion (C, D), we identify a useful
class of states in Mod(C, D), which we call D-compatible states.
To motivate the definition, observe that when (C, D) is a regular EP inclusion, the only way to
extend a pure state σ ∈ D to C is via composition with the expectation: ρ := σ ◦ E. Then the GNS
representation (πρ, Hρ) arising from ρ has the property that for any v ∈ N(C, D) either I + Lρ and
v + Lρ are orthogonal in the Hilbert space Hρ, or one is a scalar multiple of the other, according
to whether or not the Gelfand transform of E(v) is zero in a neighborhood of σ. Furthermore,
the techniques used in the proof of [10, Proposition 5.4] show that πρ(D)′′ is an atomic MASA in
B(Hρ) and also that for every v ∈ N(C, D), v + Lρ is an eigenvector for πρ(D). The intersection J
of the kernels of such representations is the left kernel of the expectation E, D ∩ J = (0), and the
quotient of (C, D) by J yields a C ∗-diagonal with the same coordinate system as (C, D), see [10,
Theorem 4.8].
We shall define the set of compatible states to be those states ρ on C for which the vectors
{v + Lρ : v ∈ N(C, D)} form an orthogonal set of vectors. These states have many of the properties
listed in the previous paragraph, but have the advantage of not needing the extension property or
a conditional expectation for their definition. Here is the formal definition.
Definition 4.1. Let (C, D) be an inclusion.
(1) A state ρ on C is called D-compatible if for every v ∈ N(C, D),
ρ(v)2 ∈ {0, ρ(v∗v)}.
When the context is clear, we will simply use the term compatible state instead of D-
compatible state.
(2) We will use S(C, D) to denote the set of all D-compatible states on C. Topologize S(C, D)
with the relative weak-∗ topology.
24
(3) For ρ ∈ S(C, D), let ∆ρ := {v ∈ N(C, D) : ρ(v) 6= 0}, and Λρ := {v ∈ N(C, D) : ρ(v∗v) > 0}.
Define a relation ∼ρ on Λρ by (v, w) ∈∼ρ if and only if v∗w ∈ ∆ρ. (We shall prove that ∼ρ
is an equivalence relation momentarily, and then will simply write v ∼ρ w.)
Remarks 4.2.
(1) When (C, D) is a regular MASA inclusion, Proposition 3.14 shows that Ss(C, D) ⊆ S(C, D).
(2) As ρ(x)2 ≤ ρ(x∗x) for any state ρ ∈ C# and any x ∈ C, we see that D-compatible states
satisfy an extremal property relative to the normalizers for D, and one might expect an
inclusion relationship between compatible states and pure states. However, there is not.
Example 7.17 gives an example of a Cartan inclusion (C, D) and element of S(C, D) which is
not a pure state on C, while Example 7.16 gives an example of a Cartan inclusion (C, D) and
a pure state ρ on C such that ρ ∈ Mod(C, D), yet ρ /∈ S(C, D). As we shall see momentarily,
S(C, D) ⊆ Mod(C, D). Thus no such inclusion relationship exists.
(3) For general inclusions, it is possible that S(C, D) = ∅ (see Theorem 4.8).
(4) The following simple observation will be useful during the sequel: for i = 1, 2, let (Ci, Di)
be inclusions and suppose that α : C1 → C2 is a regular and unital ∗-homomorphism. Then
α#(S(C2, D2)) ⊆ S(C1, D1).
Here are some properties of elements of S(C, D) which hold for any inclusion.
Proposition 4.3. Let (C, D) be an inclusion and let ρ ∈ S(C, D). The following statements hold.
(1) Suppose v ∈ ∆ρ. Then for every x ∈ C,
ρ(vx) = ρ(v)ρ(x) = ρ(xv).
(2) The restriction of ρ to D is a multiplicative linear functional on D.
(3) Suppose v ∈ ∆ρ. Then for every x ∈ C,
ρ(v∗xv) = ρ(v∗v)ρ(x).
(4) If v1, v2 ∈ Λρ and (v1, v2) ∈∼ρ, then
ρ(v∗
1v2)2 = ρ(v∗
1v1)ρ(v∗
2v2).
Moreover, ∼ρ is an equivalence relation on Λρ.
(5) S(C, D) is an N(C, D)-invariant subset of Mod(C, D).
(6) If v ∈ Λρ, then v + Lρ is an eigenvector for πρ(D); in particular, for every d ∈ D,
πρ(d)(v + Lρ) =
ρ(v∗dv)
ρ(v∗v)
v + Lρ.
(7) The set S(C, D) is weak-∗ closed in C# and the restriction mapping, ρ ∈ S(C, D) 7→ ρD,
is a weak-∗ -- weak-∗ continuous mapping of S(C, D) into D.
Proof. Statement (1). Since ρ ∈ S(C, D), an easy calculation yields v − ρ(v)I ∈ Lρ. But Lρ is a
left ideal and Lρ ⊆ ker ρ. So for x ∈ C, we have ρ(x(v − ρ(v)I)) = 0. So ρ(xv) = ρ(x)ρ(v). As
ρ(v∗) = ρ(v) 6= 0, a similar argument shows that 0 = ρ((v − ρ(v)I)x). So part (1) holds.
Statement (2). Since D ⊆ N(C, D), this follows from part (1) and continuity of ρ.
Statement (3). This follows from part (1) and the fact that ∆ρ is closed under the adjoint
operation.
Statement (4). Let σ = ρD and for i = 1, 2 put σi = βvi (σ). Then σ1 = σ2 by statement (3) and
Proposition 2.4. Therefore, since ρ(v∗
1v2) 6= 0, we have
ρ(v∗
1v2)2 = ρ(v∗
2v1v∗
1v2) = σ2(v1v∗
2v2) = σ1(v1v∗
1)σ(v∗
25
1)σ(v∗
2v2) = ρ(v∗
1v1)ρ(v∗
2v2).
Clearly the relation ∼ρ is reflexive and symmetric on Λρ. For i = 1, 2, 3, suppose vi ∈ Λρ,
(v1, v2) ∈∼ρ and (v2, v3) ∈∼ρ. The equality verified in the previous paragraph shows that in Hρ,
hv1 + Lρ, v2 + Lρiρ 2 = kv1 + Lρk2
ρ. Hence there exists a non-zero scalar t such that
tv1 + Lρ = v2 + Lρ. Similarly, there exists a non-zero scalar s such that v2 + Lρ = sv3 + Lρ. So
{v1 + Lρ, v3 + Lρ} is a linearly dependent set of non-zero linearly vectors in Hρ. Thus ρ(v∗
1v3) 6= 0,
whence (v1, v3) ∼ρ .
ρ kv2 + Lρk2
Statement (5). Let v ∈ Λρ. For w ∈ N(C, D), we claim that ρ(w∗v)2 ∈ {0, ρ(w∗w)ρ(v∗v)}. If
ρ(w∗v) 6= 0, then as ρ(w∗v)2 ≤ ρ(w∗w)ρ(v∗v), we find that w ∈ Λρ and w ∼ρ v, so the claim holds
by statement (4). Hence
βv(ρ)(w)2 =
ρ(v∗(wv))2
ρ(v∗v)2
∈(cid:26)0,
ρ(v∗v)ρ(v∗w∗wv)
ρ(v∗v)2
(cid:27) = {0, βv(ρ)(w∗w)},
so βv(ρ) ∈ S(C, D).
Statement (6). Suppose v ∈ N(C, D) and that ρ(v∗v) 6= 0. For d ∈ D, let σ1(d) =
ρ(v∗dv)
ρ(v∗v)
. Then
σ1 ∈ D, and for d ∈ D, we have
k(πρ(d) − σ1(d)I)v + Lρk2
ρ = ρ(v∗(d − σ1(d)I)∗(d − σ1(d)I)v)
= σ1((d − σ1(d)I)∗(d − σ1(d)I))ρ(v∗v) = 0.
We conclude that πρ(d)v +Lρ = σ1(d)v +Lρ, so v +Lρ is an eigenvector for πρ(D) and statement (6)
holds.
Statement (7). Suppose (ρλ)λ∈Λ is a net in S(C, D) and ρλ converges weak-∗ to ρ ∈ C#. Let
v ∈ N(C, D). If ρ(v) 6= 0, then for large enough λ, ρλ(v) 6= 0. Hence ρ(v)2 = limλ ρλ(v)2 =
limλ ρλ(v∗v) = ρ(v∗v). It follows that ρ ∈ S(C, D), so S(C, D) is weak-∗ closed. The continuity of
the restriction mapping is obvious.
(cid:3)
Remark 4.4. Statement (1) says that if v ∈ ∆ρ, then v ∈ Mρ, where Mρ = {x ∈ C : ρ(xy) =
ρ(yx) = ρ(x)ρ(y) ∀ y ∈ C}, see [2]. Also, if B is the closed linear span of ∆ρ, then B is a C ∗-algebra
because ∆ρ is closed under multiplication. Clearly D ⊆ B, so that (B, D) is an inclusion enjoying
the properties of regularity or MASA inclusion when (C, D) has the same properties.
We turn now to the issue of existence of compatible states. When (C, D) is a regular MASA
inclusion, Theorem 3.21, shows that every σ ∈ D extends to an element of Ss(C, D). Applying
Proposition 3.14, we see that compatible states exist in abundance for regular MASA inclusions.
We record this fact as a theorem.
Theorem 4.5. Let (C, D) be a regular MASA inclusion. If σ ∈ D, there exists ρ ∈ S(C, D) such
that ρD = σ. Moreover, ρ may be chosen so that ρ ∈ Ss(C, D).
The following result summarizes what we know regarding the existence of compatible states when
the hypothesis of regularity in Theorem 4.5 is weakened. Notice that in both parts of the following
result, a conditional expectation is present.
Theorem 4.6. Suppose (C, D) is an inclusion.
a) If (C, D) is a MASA inclusion and there exists a conditional expectation E : C → D, then
E# D is a continuous one-to-one map of D into S(C, D).
b) When (C, D) has the extension property (but is not necessarily regular), then E# D is a
homeomorphism of D onto S(C, D).
26
Proof. a) Since E is onto, E# is injective and continuous. We must show that E# carries D into
S(C, D).
Let σ ∈ D and set ρ = σ ◦ E. Suppose v ∈ N(C, D), and ρ(v) 6= 0. By the Cauchy-Schwartz
inequality, ρ(vv∗) 6= 0. The definition of ρ shows E(v) 6= 0. Let x :=
v∗E(v)
ρ(vv∗)
.
We claim that x commutes with D. This is easy to see when v ∈ I(C, D). Since D is a MASA,
Proposition 2.2 gives N(C, D) = I(C, D). A continuity argument now establishes the claim.
Therefore, x ∈ D. Hence ρ(v)ρ(x) = ρ(vx) = ρ(vv∗E(v)ρ(vv∗)−1) = ρ(v), so that ρ(x) = 1.
Since vx ∈ D we obtain,
ρ(v)2 = ρ(vx)2 = ρ(x∗v∗vx) = ρ(x∗)ρ(v∗v)ρ(x) = ρ(v∗v).
We conclude that ρ ∈ S(C, D), as desired.
b) Now suppose that (C, D) has the extension property.
If ρ ∈ S(C, D), put σ = ρD. Then
σ ∈ D. By the extension property, we have ρ = σ ◦ E, so ρ = E#(σ), whence E# D is onto. If
E#(σ1) = E#(σ2), then the extension property yields σ1 = σ2. So E# is a continuous bijection of
D onto S(C, D). Since D and S(C, D) are both compact and Hausdorff, E# D is a homeomorphism.
(cid:3)
Theorem 4.7. Let (C, D) be a regular inclusion (we do not assume D is a MASA in C). The
following statements hold.
i) Suppose (C1, D1) is a regular MASA inclusion and α : (C, D) → (C1, D1) is a regular and
unital ∗-homomorphism. Then α# maps Ss(C1, D1) into S(C, D).
ii) If the relative commutant Dc of D in C is abelian, then Ss(C, Dc) ⊆ S(C, D) and the
restriction map ρ ∈ Ss(C, Dc) 7→ ρD, is a weak-∗ -- weak-∗ continuous mapping of Ss(C, Dc)
onto D.
Proof. We have already observed in Remark 4.2(4) that α# carries S(C1, D1) into S(C, D). As
Ss(C1, D1) ⊆ S(C1, D1), the first statement holds.
Now suppose that Dc is abelian. Lemma 2.3 shows that (C, Dc) is a regular MASA inclusion
and that the identity mapping of C onto itself is regular. By part (i), Id# carries Ss(C, Dc) into
S(C, D); thus Ss(C, Dc) ⊆ S(C, D). As any element of D can be extended to an element of cDc, we
see that the restriction map is onto. Part (7) of Proposition 4.3 gives the weak-∗ continuity.
(cid:3)
We turn now to a result which shows that there are inclusions with few compatible states. In
fact, some inclusions have no compatible states. This result applies when the relative commutant
of D in C is all of C, e.g. (C, CI). The result shows that when N(C, D) is too large, it may happen
that S(C, D) is empty. For example, when C is a unital simple C ∗-algebra with dim(C) > 1, then
S(C, CI) = ∅.
Theorem 4.8. Let (C, D) be an inclusion and let U(C) be the unitary group of C. Assume that
U(C) ⊆ N(C, D). Then (C, D) is regular and S(C, D) is the set of all multiplicative linear functionals
on C.
Proof. Since span(U(C)) = C, (C, D) is a regular inclusion. As every multiplicative linear functional
on C is a compatible state, we need only prove that every element of S(C, D) is a multiplicative
linear functional.
Fix ρ ∈ S(C, D). Then for every unitary U ∈ C we have ρ(U ) ∈ {0} ∪ T. Let π be a universal
representation of C, and identify C## with the von Neumann algebra π(C)′′. Also, regard C as a
subalgebra of C##. Let ρ## denote the normal state on C## obtained from ρ. By [33, II.4.11], every
27
unitary in C## is the strong-∗ limit of a net of unitaries in C. Since ρ## is normal, ρ##(W ) ∈ {0}∪T
for every unitary W ∈ C##.
Let P be a projection in C##. We shall show that ρ##(P ) ∈ {0, 1}. We argue by contradiction.
Suppose that 0 < ρ##(P ) < 1. Then 0 < ρ##(P ) + iρ##(I − P ) < 1. Put W = P + i(I − P ).
Then W is a unitary belonging to C##, and therefore we may find a net Uα of unitaries in C so that
Uα converges strong-∗ to W . But then ρ(Uα) → ρ##(W ) ∈ (0, 1). This implies that there exists
a unitary U ∈ C such that ρ(U ) ∈ (0, 1), which is a contradiction. Therefore ρ##(P ) ∈ {0, 1} for
every projection P ∈ C##.
Now let P, Q ∈ C## be projections. We claim that ρ##(P Q) = ρ##(P )ρ##(Q). By the Cauchy-
Schwartz inequality, ρ##(P Q) ≤ ρ##(P )ρ##(Q), so that ρ##(P Q) = 0 if 0 ∈ {ρ##(P ), ρ##(Q)}.
Suppose then that ρ##(P ) = ρ##(Q) = 1. Since 2P − I and 2Q − I are unitaries in C##, we may
find nets of unitaries uα and vα in C so that uα and vα converge ∗-strongly to 2P − I and 2Q − I
respectively. Both ρ(uα) and ρ(vα) are eventually non-zero because
lim ρ(uα) = ρ##(2P − I) = 1 = ρ##(2Q − I) = lim ρ(vα).
As multiplication on bounded subsets of C## is jointly continuous in the strong-∗ topology, uαvα
converges strongly to (2P − I)(2Q − I). By Proposition 4.3(1),
ρ##((2P − I)(2Q − I)) = lim ρ(uαvα) = lim ρ(uα)ρ(vα) = ρ##(2P − I)ρ##(2Q − I) = 1.
On the other hand, a calculation shows that
ρ##((2P − I)(2Q − I)) = 4ρ##(P Q) − 3.
Combining these equalities gives ρ##(P Q) = 1, as desired. The claim follows.
Let X = Pn
j=1 λjPj and Y = Pn
j=1 and
j=1 in C##. It follows from the previous paragraph that ρ##(XY ) = ρ##(X)ρ##(Y ). Since
{Qj}n
any von Neumann algebra is the norm closure of the span of its projections, ρ## is multiplicative
on C##. It then follows that ρ is multiplicative on C.
j=1 µjQj be linear combinations of projections {Pj}n
(cid:3)
We now turn to the representations arising from states in S(C, D). We begin with a simple
lemma concerning states on regular inclusions, whose proof we leave to the reader.
Lemma 4.9. Let (C, D) be a regular inclusion and suppose that ρ is a state on C. Then
span{v + Lρ : v ∈ N(C, D), ρ(v∗v) > 0}
is norm-dense in Hρ.
Proposition 4.10. Let (C, D) be a regular inclusion, let ρ ∈ S(C, D), and let T ⊆ Λρ be chosen
so that for every v ∈ T , ρ(v∗v) = 1 and T contains exactly one element from each ∼ρ equivalence
class. Then the following statements hold.
(1) {v + Lρ : v ∈ T } is an orthonormal basis for Hρ.
(2) For v ∈ T , let Kv := {ξ ∈ Hρ : πρ(d)ξ = ρ(v∗dv)ξ for all d ∈ D} and let σ = ρD. Then
Kv = span{w + Lρ : w ∈ T and βw(σ) = βv(σ)}.
(3) For v ∈ T , let Pv be the orthogonal projection of Hρ onto Kv. Then Pv is a minimal
projection in πρ(D)′′ andWv∈T Pv = I.
(4) πρ(D)′′ is an abelian and atomic von Neumann algebra.
If v, w ∈ T are distinct, then ρ(v∗w) = 0, so that {v + Lρ : v ∈ T } is
Proof. Statement (1).
an orthonormal set. Part (4) of Proposition 4.3 and the Cauchy-Schwartz inequality show that
28
if v ∈ T , and w ∈ Λρ is such that v ∼ρ w, then w + Lρ ∈ span{v + Lρ}. This, together with
Lemma 4.9, shows that
span{v + Lρ : v ∈ T } = span{v + Lρ : v ∈ Λρ} = span{v + Lρ : v ∈ N(C, D)} = Hρ.
Thus {v + Lρ : v ∈ T } is an orthonormal basis for Hρ.
Statement (2). If ξ ∈ span{w + Lρ : βw(σ) = βv(σ)}, Part (6) of Proposition 4.3 implies that
ξ ∈ Kv. For the opposite inclusion, suppose ξ ∈ Kv. Then for w ∈ T and d ∈ D we have
βv(σ)(d) hξ, w + Lρi = hπρ(d)ξ, w + Lρi = hξ, πρ(d∗)(w + Lρ)i = βw(ρ)(d) hξ, w + Lρi .
Hence if hξ, w + Lρi 6= 0, then βv(σ) = βw(σ). This yields ξ ∈ span{w + Lρ : w ∈ T and βw(σ) =
βv(σ)}.
Statement (3). First note that for v ∈ T , v + Lρ ∈ Kv; thus, since {v + Lρ : v ∈ T } is an
orthonormal basis for Hρ, we obtainWv∈T Pv = I.
Let X ∈ πρ(D)′ and ξ ∈ Kv. Then for d ∈ D,
πρ(d)Xξ = Xπρ(d)ξ = ρ(v∗dv)Xξ.
Therefore Xξ ∈ Kv, showing that Kv is an invariant subspace for X. As this holds for every
X ∈ πρ(D)′, we conclude that Pv ∈ πρ(D)′′.
Let v ∈ T and suppose that Q ∈ πρ(D)′′ is a projection with 0 ≤ Q ≤ Pv. For all d ∈ D we have
πρ(d)Pv = βv(σ)(d)Pv = hπρ(d)(v + Lρ), v + Lρi Pv. The Kaplansky Density Theorem shows that
for every X ∈ πρ(D)′′ we have XPv = hX(v + Lρ), v + Lρi Pv ∈ CPv. Since Q commutes with Pv,
QPv is a projection; hence QPv ∈ {0, Pv}, so Pv is a minimal projection in πρ(D)′′.
Statement (4). This follows from statement (3).
(cid:3)
The following result shows that elements of S(C, D) arise from regular representations π of
(C, D), which can be taken so that π(D)′′ is atomic. For vectors h1, h2 in a Hilbert space H we use
the notation h1h∗
2 for the rank-one operator h 7→ hh, h2i h1.
Theorem 4.11. Let (C, D) be a regular inclusion. The following statements hold.
i) Let ρ ∈ S(C, D). and let
Aρ := {(v + Lρ)(v + Lρ)∗ : v ∈ N(C, D)}′′ ⊆ B(Hρ).
Then Aρ is an atomic MASA in B(Hρ) and πρ : (C, D) → (B(Hρ), Aρ) is a regular ∗-
homomorphism.
ii) Conversely, suppose π : C → B(H) is a regular ∗-homomorphism with π(D)′′ a (not neces-
sarily atomic) MASA in B(H), and let E : B(H) → π(D)′′ be any conditional expectation.
Then for any pure state σ of π(D)′′, σ ◦ E ◦ π ∈ S(C, D).
Proof. For the first statement, choose T as in the statement of Proposition 4.10. For v ∈ N(C, D),
we have v + Lρ = 0 if v /∈ Λρ; and if v + Lρ 6= 0, then there exists w ∈ T such that v ∼ρ w, so
(v + Lρ)(v + Lρ)∗ ∈ C(w + Lρ)(w + Lρ)∗. Since B := {w + Lρ : w ∈ T } is an orthonormal basis for
Hρ, we see that Aρ is an atomic MASA in B(Hρ).
We now show that πρ is a regular homomorphism. Let v ∈ N(C, D) and let w ∈ T . Then
πρ(v)(w + Lρ)(w + Lρ)∗πρ(v)∗ = (vw + Lρ)(vw + Lρ)∗ ∈ Aρ. As span{(w + Lρ)(w + Lρ)∗ : w ∈ T }
is weakly dense in Aρ, we conclude that πρ(v)Aρπρ(v)∗ ⊆ Aρ. Similarly πρ(v)∗Aρπρ(v) ⊆ Aρ. Thus
πρ is a regular ∗-homomorphism.
For the second statement, Theorem 4.6 shows that if σ ∈ \πρ(D)′′, then σ ◦ E ∈ S(B(H), π(D)′′).
(cid:3)
Remark 4.2(4) completes the proof.
29
Remark 4.12. We have πρ(D)′′ ⊆ Aρ always, but in general they can be very different. Consider
the state ρ = ρ∞ from Example 7.17. Then, using the notation from that example, {Sn+Lρ : n ∈ Z}
is an orthonormal basis for Hρ (where Sn = S∗n when n < 0). Note that πρ(D)′′ = CI, while Aρ
is a MASA.
The following proposition characterizes when πρ(D)′′ and Aρ coincide. We first make a definition.
Definition 4.13. Let (C, D) be an inclusion and let f ∈ Mod(C, D). The D-stabilizer of f is the
set,
D-stab(f ) := {v ∈ N(C, D) : for all d ∈ D, f (v∗dv) = f (d)}.
If for every v ∈ D-stab(f ) and x ∈ C, we have f (x) = f (v∗xv), then we call f a D-rigid state.
Proposition 4.14. Let (C, D) be a regular inclusion, and suppose that ρ ∈ S(C, D). The following
statements are equivalent.
(1) πρ(D)′′ is a MASA in B(Hρ).
(2) If v ∈ D-stab(ρ), then ρ(v) 6= 0.
(3) ρ is a pure and D-rigid state.
Proof. Throughout the proof, we let σ = ρD, which by Proposition 4.3(2), belongs to D.
Suppose πρ(D)′′ is a MASA in B(Hρ) and let v ∈ D-stab(ρ), so v ∈ Λρ and βv(σ) = σ. Then,
using the notation of Proposition 4.10, we find that PI (v + Lρ) = v + Lρ. Since πρ(D)′′ is a MASA,
PI is the orthogonal projection onto C(I + Lρ). We conclude that v + Lρ is a non-zero scalar
multiple of I + Lρ. Hence 0 6= hv + Lρ, I + Lρi = ρ(v). Thus v ∈ ∆ρ, so statement (1) implies
statement (2).
Before proving the next implication, we pause for some generalities. Suppose that f is any state
on C with the property that f D = σ. If v ∈ N(C, D) and f (v∗v) = 0, then the Cauchy-Schwartz
inequality yields f (v) = 0. Also note that if v ∈ Λρ satisfies βv(σ) 6= σ, then f (v) = 0. Indeed, for
such v ∈ Λρ, choose d ∈ D so that βv(σ)(d) = 0 and σ(d) = 1. Then as v∗dv ∈ D,
0 = f (v∗dv) = f (v)f (v∗dv) = f (vv∗dv) = f (dvv∗v) = f (d)f (v)f (v∗v).
As f (d) and f (v∗v) are both non-zero, we conclude that f (v) = 0.
Now suppose statement (2) holds. We first prove that ρ is pure. So suppose that t ∈ [0, 1] and
that for i = 1, 2, ρi are states on C and ρ = tρ1 + (1 − t)ρ2. As ρD is a pure state on D, we have
ρiD = σ. We claim that for every v ∈ N(C, D), ρ1(v) = ρ2(v) = ρ(v). By the previous paragraph,
it remains only to prove this for v ∈ Λρ such that βv(σ) = σ. So suppose v has this property. By
the hypothesis in statement (2), ρ(v) 6= 0. Clearly ρi(v) ≤ ρi(v∗v)1/2 = ρ(v∗v)1/2 = ρ(v). Thus
we have
t
ρ1(v)
ρ(v)
+ (1 − t)
= 1,
ρ2(v)
ρ(v)
which expresses 1 as a convex combination of elements of the closed unit disk. Hence ρi(v) = ρ(v),
establishing the claim. By regularity, we conclude that ρ1 = ρ2 = ρ, so ρ is a pure state.
Next, if v ∈ Λρ and βv(σ) = σ, then by hypothesis, ρ(v) 6= 0. So the final part of statement (3)
follows from part (1) of Proposition 4.3. Thus statement (2) implies statement (3).
Finally, suppose that statement (3) holds. Let v, w ∈ Λρ be such that βv(σ) = βw(σ). We shall
show that {v +Lρ, w+Lρ} is a linearly dependent set, showing that Kv is one-dimensional. We have
ρ(v∗wxw∗v)
βw∗v(σ) = σ = βv∗w(σ), so ρ(v∗ww∗v)−1/2w∗v ∈ D-stab(ρ). By hypothesis, ρ(x) =
ρ(v∗ww∗v)
for every x ∈ C. Thus if η = ρ(v∗ww∗v)−1/2w∗v + Lρ, we have hπρ(x)η, ηi = hπρ(x)(I + Lρ), I + Lρi
for every x ∈ C. Since ρ is pure, πρ(C)′′ = B(Hρ), so that for every T ∈ B(Hρ) we obtain
hT η, ηi = hT (I + Lρ), I + Lρi .
30
Hence {η, I + Lρ} is a linearly dependent set. Thus, {w∗v + Lρ, I + Lρ} is linearly dependent.
Since both vectors in this set are non-zero, we find 0 6= hw∗v + Lρ, I + Lρi = ρ(w∗v). Applying
part (4) of Proposition 4.3 and the Cauchy-Schwartz inequality, we obtain {v + Lρ, w + Lρ} is
linearly dependent, as desired.
As Kv is one-dimensional, Proposition 4.10 implies that πρ(D)′′ is a MASA.
(cid:3)
5. The D-Radical and Embedding Theorems
Our purpose in this section is to prove two embedding theorems. The first characterizes when
a regular inclusion can be regularly embedded into a regular MASA inclusion, while the second
characterizes when a regular inclusion may be regularly embedded into a C ∗-diagonal.
The first of these theorems shows that the obvious necessary condition suffices.
Theorem 5.1. Let (C, D) be a regular inclusion. The following statements are equivalent.
a) There exists a regular MASA inclusion (C1, D1) and a regular ∗-monomorphism α : (C, D) →
(C1, D1).
b) The relative commutant Dc of D in C is abelian.
Proof. Suppose that (C1, D1) is a regular MASA inclusion and α : (C, D) → (C1, D1) is a regular
∗-monomorphism. Let (I(D1), ι1) be an injective envelope for D1 and let E1 : C1 → I(C1) be the
pseudo-expectation for ι1.
Observe that (Dc, D) is a regular inclusion, and N(Dc, D) ⊆ N(C, D). Let ρ ∈ Ss(C1, D1).
Part (i) of Theorem 4.7 shows that ρ ◦ α ∈ S(C, D), and hence ρ ◦ αDc ∈ S(Dc, D). Theorem 4.8
implies ρ ◦ αDc is a multiplicative linear functional on Dc. By the definition of Ss(C1, D1), we see
that for every τ ∈ \I(D1), τ ◦ E1 ◦ αDc is a multiplicative linear functional on Dc. We conclude
E1 ◦ αDc is a ∗-homomorphism of Dc into I(D1).
Let u ∈ Dc be a unitary element. Clearly, u ∈ N(C, D), so regularity of α implies that α(u) ∈
N(C1, D1). Let Jα(u) be the ideal of D1 as defined in (7). Since E1(α(u)) is unitary, Theorem 3.10(c)
shows supI(D1)(ι(Jα(u))+
1 ) is the identity of I(D1). Hence Jα(u) is an essential ideal in D1. Then
(fix βα(u))◦ is dense in D1 by Proposition 3.2. As dom(βα(u)) = D1, we get D1 = (fix βα(u))◦ =
fix βα(u). Therefore, D1 = (fix βα(u))◦. Another application of Proposition 3.2 gives Jα(u) = D1,
and hence α(u) ∈ D1. This shows that the image of the unitary group of Dc under α is abelian.
Since α is faithful, we see that the unitary group of Dc is abelian, and hence Dc is abelian.
For the converse, take α to be the identity map on C. Lemma 2.3 shows that α : (C, D) → (C, Dc)
is regular, so statement (a) follows from statement (b).
(cid:3)
Question 5.2. When does a regular inclusion regularly embed into a regular EP-inclusion?
We conjecture that the conditions of Theorem 5.1 also characterize when a regular inclusion may
be regularly embedded into a regular EP-inclusion. Here is an approach to this problem. Suppose
(C, D) is a regular inclusion with Dc abelian. Let π : C → B(H) be a faithful representation of C,
let D1 = π(Dc)′′ and let C1 be the (concrete) C ∗-algebra generated by π(C) and D1. Then (C1, D1)
is regular, and π : (C, D) → (C1, D1) is a regular ∗-monomorphism. If D1 is a MASA in C1, then
Theorem 2.10 shows that (C1, D1) is an EP-inclusion. Unfortunately, we have not been able to
decide whether the faithful represetation π can be chosen so that (C1, D1) is a MASA inclusion.
We now define a certain ideal, the D-radical of an inclusion, and show its relevance to embedding
regular inclusions into C ∗-diagonals.
Definition 5.3. For an inclusion (C, D), the D-radical of (C, D) is the set
Rad(C, D) := {x ∈ C : kπρ(x)k = 0 for all ρ ∈ S(C, D)},
31
provided S(C, D) 6= ∅; otherwise define Rad(C, D) = C. (Note that Rad(C, D) 6= C whenever (C, D)
is a regular MASA inclusion.)
When (C, D) is a regular inclusion, we have the following description of Rad(C, D).
Proposition 5.4. Suppose that (C, D) is a regular inclusion. Then
Rad(C, D) = {x ∈ C : ρ(x∗x) = 0 for all ρ ∈ S(C, D)}.
Proof. Let J := {x ∈ C : ρ(x∗x) = 0 for all ρ ∈ S(C, D)}. If x ∈ Rad(C, D) and ρ ∈ S(C, D), then
ρ(x∗x) = kπρ(x)(I + Lρ)k = 0, and we find that Rad(C, D) ⊆ J. For the opposite inclusion, let
x ∈ J. Part (5) of Proposition 4.3 and Corollary 2.16 show that J is a closed, two-sided ideal of
C. Hence for every c ∈ C and ρ ∈ S(C, D) we have ρ(c∗x∗xc) = 0, which means that πρ(x) = 0 for
every ρ. So x ∈ Rad(C, D), showing Rad(C, D) = J.
(cid:3)
Examples 5.5. Here are some examples of the D-radical.
(1) By Proposition 5.7, Rad(C, D) = (0) for any Cartan inclusion.
(2) Suppose that (C, D) is a regular EP inclusion. Then Rad(C, D) is the left kernel of the
associated conditional expectation E, that is, Rad(C, D) = {x ∈ C : E(x∗x) = 0}. This
follows from Propositions 5.4 and 4.6.
(3) Suppose that (C, D) is an inclusion such that U(C) ⊆ N(C, D). Then Theorem 4.8 shows
that S(C, D) is the set of characters on C. Since the intersection of the kernels of all
characters is the commutator ideal, it follows from Proposition 5.4 that Rad(C, D) is the
commutator ideal of C.
Question 5.6. Observe that when (C, D) is a regular MASA inclusion, Rad(C, D) ⊆ L(C, D),
because Ss(C, D) ⊆ S(C, D). Is it possible for the inclusion to be proper?
Proposition 5.7. Let (C, D) be a regular inclusion and suppose (C1, D1) is a regular MASA in-
If α : (C, D) → (C1, D1) is a regular and unital ∗-homomorphism, then Rad(C, D) ⊆
clusion.
α−1(L(C1, D1)).
Proof. By Theorem 4.7, α(Rad(C, D)) ⊆ L(C1, D1).
(cid:3)
Notice that when L(C1, D1) = (0), Proposition 5.7 implies that Rad(C, D) ⊆ ker α. We have
been unable to decide whether equality holds in general. However, the following lemma shows that
one can construct a C ∗-diagonal and a regular ∗-homomorphism such that equality holds.
Lemma 5.8. Suppose that (C, D) is a regular inclusion. Then there exists a C ∗-diagonal (C1, D1)
and a regular ∗-homomorphism α : (C, D) → (C, D1) with ker α = Rad(C, D).
Proof. For each ρ ∈ S(C, D), let (πρ, Hρ) be the GNS representation of C arising from ρ. Let
As Aρ is an atomic MASA in B(Hρ), we see that D1 is an atomic MASA in B(H). Let C1 =
spanN(B(H), D1). By Theorem 2.10 and the fact that the expectation onto an atomic MASA in
B(H) is faithful, (C1, D1) is a C ∗-diagonal.
H :=Lρ∈S(C,D) Hρ and let D1 =Lρ∈S(C,D) Aρ, where Aρ is as in the statement of Theorem 4.11.
For each v ∈ N(C, D), the regularity of πρ (Theorem 4.11) shows that Lρ∈S(C,D) πρ(v) ∈
N(B(H), D1). Hence for each x ∈ C,Lρ∈S(C,D) πρ(x) ∈ C1. Thus if α : C → C1 is given by α(x) =
Lρ∈S(C,D) πρ(x), then α is a regular ∗-homomorphism. By construction, ker α = Rad(C, D).
The following is our main embedding result.
Theorem 5.9. Let (C, D) be a regular inclusion. Then there exists a C ∗-diagonal (C1, D1) and a
regular ∗-monomorphism α : (C, D) → (C1, D1) if and only if Rad(C, D) = 0.
32
(cid:3)
Proof. Suppose that (C1, D1) is a C ∗-diagonal and α : (C, D) → (C1, D1) is a regular ∗-monomorphism.
Since L(C1, D1) = (0), Proposition 5.7 gives Rad(C, D) ⊆ ker α = (0).
The converse follows from Lemma 5.8.
(cid:3)
Corollary 5.10. Suppose that (C, D) is an inclusion such that U(C) ⊆ N(C, D). Then there is a
regular ∗-monomorphism of (C, D) into a C ∗-diagonal if and only if C is abelian.
Proof. Since the commutator ideal is the intersection of the kernels of all multiplicative linear
functionals, the result follows directly from Theorem 4.8 and Example 5.5(3).
(cid:3)
6. An Example: Reduced Crossed Products by Discrete Groups
In this section we consider the regular inclusion (C, D), where C = D ⋊r Γ is the reduced crossed
product of the unital abelian C ∗-algebra D = C(X) by a discrete group Γ of homeomorphisms of
X.
The main results of this section are: Theorem 6.6, which characterizes when the relative commu-
tant Dc of D in D ⋊r Γ is abelian in terms of the associated dynamical system; Theorem 6.9, which
shows that when Dc is abelian, L(D ⋊r Γ, Dc) = (0); and a summary result, Theorem 6.10 which
gives a number of characterizations for when (D ⋊r Γ, D) regularly embeds into a C ∗-diagonal. By
choosing the space X and group Γ appropriately, the methods in this section can be used to produce
an example of a virtual Cartan inclusion (C, D) where C is not nuclear, see [6, Theorem 4.4.3 or
Theorem 5.1.6].
Some of the results in this section complement results from [32].
We begin by establishing our notation. This is standard material, but we include it because
there are a number of variations in the literature.
Throughout, let X be a compact Hausdorff space, let Γ be a discrete group with unit element
e acting on X as homeomorphisms of X. Thus there is a homomorphism Ξ of Γ into the group
of homeomorphisms of X, and for (s, x) ∈ Γ × X, we will write sx instead of Ξ(s)(x). We will
sometimes refer to the pair (X, Γ) as a discrete dynamical system. For s ∈ Γ, let αs ∈ Aut(C(X))
be given by
(αs(f ))(x) = f (s−1x),
f ∈ C(X), x ∈ X.
If Y is any set, and z ∈ Y , we use δz to denote the characteristic function of the singleton set
{z}.
Let D = C(X), and let Cc(Γ, D) be the set of all functions a : Γ → D such that {s ∈ Γ : a(s) 6= 0}
is a finite set. We will sometimes write a(s, x) for the value of a(s) at x ∈ X instead of a(s)(x).
Then Cc(Γ, D) is a ∗-algebra under the usual twisted convolution product and adjoint operation:
for a, b ∈ Cc(Γ, D),
(ab)(t) =Xr∈Γ
a(r)αr(b(r−1t)) and (a∗)(t) = αt(a(t−1))∗.
Let C = C(X) ⋊r Γ be the reduced crossed product of C(X) by Γ.
The group Γ is naturally embedded into C via s 7→ ws, where ws is the element of Cc(Γ, D) given
by ws(t) =(0
I
if t 6= s
if t = s.
Also, D is embedded into Cc(Γ, D) via the map d 7→ dwe and we identify
D with its image under this map. Now wsdws−1 = αs(d) and span{dws : d ∈ D, s ∈ Γ} is norm
dense in C, so {ws : s ∈ Γ} ⊆ N(C, D). Thus (C, D) is a regular inclusion.
It is well known (see for example, the discussion of crossed products in [6]) that the map E :
Cc(Γ, D) → D given by E(a) = a(e) extends to a faithful conditional expectation E of C onto D.
33
Likewise, the maps Es : Cc(Γ, D) → D given by Es(a) = a(s) extend to norm-one linear mappings
Es of C onto D. Notice that for a ∈ C and s ∈ Γ,
The maps Es allow a useful "Fourier series" viewpoint for elements of C: a ∼Ps∈Γ Es(a)ws.
The following is well-known. We sketch a proof for convenience of the reader.
Es(a) = E(aws−1).
Proposition 6.1. If a ∈ C and Es(a) = 0 for every s ∈ Γ, then a = 0.
Proof. For a ∈ Cc(Γ, D) and s, t ∈ Γ, a calculation shows that
(22)
Es(wtawt−1) = αt(Et−1st(a));
a continuity argument then shows that (22) actually holds for every a ∈ C.
Let J = {a ∈ C : Et(a) = 0 ∀t ∈ Γ}. Clearly J is closed. Then (22) shows that if a ∈ J and
s ∈ Γ, then wsaws−1 ∈ J. Easy calculations now show that if d ∈ D, s ∈ Γ and a ∈ J, then
{da, ad, wsa, aws} ⊆ J, and by taking linear combinations and closures, we find that J is a closed
two-sided ideal of C. Thus, if a ∈ J, a∗a ∈ J, so that Ee(a∗a) = E(a∗a) = 0. Hence a = 0 by
faithfulness of E. This shows that J = (0), completing the proof.
(cid:3)
Definition 6.2. We make the following definitions.
(1) For s ∈ Γ, let Fs = {x ∈ X : sx = x} be the set of fixed points of s.
(2) For s ∈ Γ, let Fs = {f ∈ D : supp(f )) ⊆ F ◦
s }. Thus {Fs : s ∈ Γ} is a family of closed ideals
in D.
(3) For x ∈ X, let Γx := {s ∈ Γ : sx = x} be the isotropy group at x.
(4) For x ∈ X, let H x := {s ∈ Γ : x ∈ (Fs)◦}. We will call H x the germ isotropy group at x.
Remarks. We chose the terminology 'germ isotropy' because s ∈ H x if and only if the homeomor-
phisms s and idX agree in a neighborhood of x, that is, they have the same germ. It is easy to see
that H x is a group; in fact, H x is a normal subgroup of Γx. To see that H x is a normal subgroup
of Γx, fix x ∈ X and let s ∈ H x. Then there exists an open neighborhood V of x such that V ⊆ Fs.
Let t ∈ Γx and put W = t−1V . Since tx = x, x ∈ W . For y ∈ W , ty ∈ V , so sty = ty. Hence
t−1sty = y. Therefore, W ⊆ Ft−1st. As W is open and x ∈ W , we see that x belongs to the interior
of Ft−1st, so t−1st ∈ H x as desired.
Simple examples show the inclusion of H x in Γx can be proper.
We record a description of the relative commutant of D in C.
Proposition 6.3. We have
Dc = {a ∈ C : αs(d)Es(a) = dEs(a) for all d ∈ D and all s ∈ Γ}
= {a ∈ C : Es(a) ∈ Fs for all s ∈ Γ}.
Proof. A computation shows that for a ∈ C, d ∈ D and s ∈ Γ,
(23)
Es(da − ad) = (d − αs(d))Es(a).
Thus if a ∈ Dc, we obtain αs(d)Es(a) = dEs(a) for every d ∈ D and s ∈ Γ. Conversely, if
Es(a)αs(d) = dEs(a) for every d ∈ D and s ∈ Γ, Proposition 6.1 gives a ∈ Dc.
For the second equality, suppose that a ∈ C and Es(a) ∈ Fs for every s ∈ Γ. Since Es(a)
is supported in F ◦
s , an examination of (23) shows that Es(da − ad) = 0 for every d ∈ D. By
Proposition 6.1 again, a ∈ Dc. For the reverse inclusion, suppose that a ∈ Dc. Then for d ∈ D
and s ∈ Γ, 0 = (d − αs(d))Es(a). Thus if x ∈ X and Es(a)(x) 6= 0, we have d(x) − d(s−1x) = 0 for
every d ∈ D. It follows that the support of Es(a) is contained in Fs. But supp(Es(a)) is open, so
the reverse inclusion holds.
(cid:3)
34
We now describe a representation useful for establishing certain formulae.
The very discrete representation. Let H = ℓ2(Γ × X). Then {δ(t,y) : (t, y) ∈ Γ × X} is an
orthonormal basis for H. For f ∈ C(X), s ∈ Γ, and ξ ∈ H, define representations π of C(X) and
U of Γ on H by
(π(f )ξ)(t, y) = f (ty)ξ(t, y) and (Usξ)(t, y) = ξ(s−1t, y).
In particular,
π(f )δ(t,y) = f (ty)δ(t,y)
and Usδ(t,y) = δ(st,y).
The C ∗-algebra generated by the images of π and U is isometrically isomorphic to the reduced
crossed product of C(X) by Γ (see [6, pages 117-118]), and hence determines a faithful representa-
tion θ : C → B(H).
A computation shows that for a ∈ C, t, r ∈ Γ and x, y ∈ X,
Also for a ∈ C, t ∈ Γ and y ∈ X, we have
(24)
Es(a)(sty)δ(st,y).
Ert−1(a)(ry)
if x 6= y;
if x = y.
(cid:10)θ(a)δ(t,y), δ(r,x)(cid:11) =(0
θ(a)δ(t,y) =Xs∈Γ
We now define some notation. Let λ : Γ → B(ℓ2(Γ)) be the left regular representation, and
r (H x) is the C ∗-algebra generated by
for x ∈ X, regard ℓ2(H x) as a subspace of ℓ2(Γ). Then C ∗
{λsℓ2(H x) : s ∈ H x}. Define Vx : ℓ2(H x) → H by
(Vxη)(s, y) =(0
η(s)
if (s, y) /∈ H x × {x}
if (s, y) ∈ H x × {x}.
Then for r ∈ H x, we have Vxδr = δ(r,x), so Vx is an isometry.
Proposition 6.4. For x ∈ X and a ∈ C, define Φx(a) := V ∗
positive unital mapping of C onto C ∗
r (H x).
Proof. Clearly Φx is completely positive and unital. For d ∈ D, r ∈ Γ and s, t ∈ H x we have
(25)
r (H x) and ΦxDc is a ∗-epimorphism of Dc onto C ∗
hΦx(dwr)δs, δti = hV ∗
x θ(a)Vx. Then Φx is a completely
= d(rx) hλrδs, δti .
x θ(dwr)Vxδs, δti =(cid:10)π(d)Urδ(s,x), δ(t,x)(cid:11)
=(cid:10)π(d)δ(rs,x), δ(t,x)(cid:11) = d(rsx)(cid:10)δ(rs,x), δ(t,x)(cid:11) = d(rx) hδrs, δti
Φx(dwr) =(0
if r /∈ H x,
if r ∈ H x.
d(x)λrℓ2(H x)
Hence for every d ∈ D and r ∈ Γ,
(26)
Therefore Φx maps a set of generators for C into C ∗
r (H x), giving Φx(C) ⊆ C ∗
r (H x).
To show that ΦxDc is a ∗-homomorphism, it suffices to prove that the range of Vx is an invariant
subspace for θ(Dc). Note that range(Vx) = span{δ(t,x) : t ∈ H x}. Let a ∈ Dc and fix t ∈ H x. We
claim that if s ∈ Γ, d ∈ Fs and stx ∈ supp(d), then s ∈ H x. Indeed, suppose that stx ∈ supp(d).
As t ∈ H x, stx = sx. So sx ∈ F ◦
s . Thus s ∈ H x, so the claim holds.
s−1, which yields x ∈ F ◦
s = F ◦
Next by (24) and Proposition 6.3, for t ∈ H x, we have
θ(a)δ(t,x) =Xs∈Γ
Es(a)(stx)δ(st,x) = Xs∈H x
as desired. It follows that ΦxDc is a ∗-homomorphism.
35
Es(a)(stx)δ(st,x) ∈ range(Vx),
It remains to show Φx(Dc) = C ∗
If s ∈ H x, let d ∈ Fs be such that d(x) = 1, and
put a = dws. Then (26) shows that Φx(a) = λsℓ2(H x). By Proposition 6.3, a ∈ Dc, and hence
Φx(Dc) is dense in C ∗
r (H x). Since ΦxDc is a homomorphism, it has closed range. Therefore
Φx(Dc) = C ∗
r (H x).
r (H x).
(cid:3)
kf (x)k < ∞)
Let
C ∗
r (H x) :=(f ∈ Yx∈X
Mx∈X
and for f ∈ Lx∈X C ∗
multiplication and involution defined point-wise,Lx∈X C ∗
Corollary 6.5. The map Φ : C →Lx∈X C ∗
positive unital mapping such that ΦDc is a ∗-monomorphism.
r (H x) : sup
x∈X
C ∗
r (H x), define kf k = supx∈X kf (x)k. Then with product, addition, scalar
r (H x) is a C ∗-algebra.
r (H x) given by Φ(a)(x) = Φx(a) is a faithful completely
Proof. It follows from the definition of Φx that Φ is unital and completely positive. Proposition 6.4
shows that ΦDc is a ∗-homomorphism; it remains to check that Φ is faithful.
For x ∈ X, let trx be the the trace on C ∗
r (H x). For d ∈ D and s ∈ Γ equation (26) gives,
trx(Φx(dws)) = (0
d(x)
if s 6= e
if s = e) = E(dws)(x).
This formula extends by linearity and continuity, so that for a ∈ C, trx(Φx(a)) = E(a)(x). So if
a ≥ 0 belongs to C and Φ(a) = 0, then E(a) = 0, so a = 0. Thus, Φ is faithful.
(cid:3)
Theorem 6.6. The relative commutant, Dc, of D in C is abelian if and only if H x is an abelian
group for every x ∈ X.
Proof. Corollary 6.5 shows that if H x is abelian for every x ∈ X, then Dc is abelian.
For the converse, we prove the contrapositive. Suppose that H x is non-abelian for some x ∈ X.
Fix s, t ∈ H x so that st 6= ts. Then x ∈ (Fs)◦ ∩ (Ft)◦, so we may find d ∈ D so that d(x) = 1 and
supp(d) ⊆ (Fs)◦ ∩ (Ft)◦. Then for h ∈ D and z ∈ X we have (by examining the cases z ∈ Fs and
z /∈ Fs),
(αs(h)(z) − h(z))d(z) = (h(s−1z) − h(z))d(z) = 0.
Proposition 6.3 shows that dws ∈ Dc. Likewise, dwt ∈ Dc.
Then dwsdwt = dαs(d)wst. Note that by choice of d, s supp(d) = supp(d). For z ∈ X,
αs(d)(z) = d(s−1z) =(0
d(z)
if z /∈ supp(d)
if z ∈ supp(d).
Thus, αs(d) = d, and likewise, αt(d) = d. Therefore,
(dws)(dwt) = d2wst 6= d2wts = (dwt)(dws),
so Dc is not abelian.
Proposition 6.7. Let (X, Γ) be a discrete dynamical system such that for each x ∈ X, the germ
isotropy group H x is abelian. Let Γ1 ⊆ Γ be a subgroup of Γ, set
C1 := D ⋊r Γ1,
and let D1 = {x ∈ C1 : xd = dx for all d ∈ D}.
Then (C1, D1) is a regular MASA inclusion and C1 ∩ L(C, Dc) ⊆ L(C1, D1).
36
(cid:3)
P = Yx∈X cH x
be the Cartesian product of the dual groups. Denote by p(x) the "x-th component" of p ∈ P . For
(x, p) ∈ X × P , define a state ρ(x,p) on C by
ρ(x,p)(a) = p(x) (Φx(a))
(here a ∈ C),
and let A := {ρ(x,p) : (x, p) ∈ X × P }.
Corollary 6.5 shows that the restriction of ρ(x,p) to Dc is a multiplicative linear functional, so in
particular, A ⊆ Mod(C, Dc).
For each s ∈ Γ, let
Xs := (X \ Fs) ∪ F ◦
s .
Then Xs is a dense, open subset of X. Set
Xs
and B := {ρ(y,p) : (y, p) ∈ Y × P }.
Proof. As D1 ⊆ Dc, D1 is abelian, and as D1 is the relative commutant of D in C1, (C1, D1) is
a regular MASA inclusion. Let ǫ : C1 → C be the inclusion map. Notice that each map in the
diagram,
(C1, D1)
ǫ
/ (C, D1)
id /
/ (C, Dc)
is a regular map. The first is clearly regular, while the regularity of the second follows from the
fact that the relative commutant of D1 in C is Dc and an application of Lemma 2.3. Therefore,
ǫ : (C1, D1) → (C, Dc) is a regular ∗-monomorphism. An application of Corollary 3.23 completes
the proof.
(cid:3)
Notation 6.8. When G is an abelian group with dual group G, we use the notation hg, γi to
denote the value of γ ∈ G at g ∈ G. Also, we will identify C ∗(G) with C( G); lastly, for γ ∈ G and
a ∈ C ∗(G), we will write γ(a) instead of a(γ).
Theorem 6.9. Suppose that (X, Γ) is a discrete dynamical system such that for each x ∈ X, the
germ isotropy group H x is abelian. Then L(C, Dc) = (0).
Proof. First assume that Γ is a countable discrete group. Let
Y := \s∈Γ
Our goal is to show that
(27)
B ⊆ Ss(C, Dc).
Fix (y, p) ∈ Y × P , and suppose that τ ∈ Mod(C, Dc) satisfies ρ(y,p)Dc = τ Dc. We claim that
ρ(y,p) = τ . To see this, it suffices to show that for each s ∈ Γ, ρ(y,p)(ws) = τ (ws). Given s ∈ Γ, if
sy 6= y, we may choose d ∈ D so that d(sy) = 1 and d(y) = 0. Using (26),
ρ(y,p)(d) = p(y)(Φy(d)) = 0
and ρ(y,p)(w∗
s dws) = ρ(y,p)(αs−1(d)) = p(y)(d(sy)I) = 1.
Then
ρ(y,p)(ws) = ρ(y,p)(ws)ρ(y,p)(w∗
s dws) = ρ(y,p)(ws(w∗
Likewise, τ (ws) = 0, so τ (ws) = ρ(y,p)(ws) = 0 when y /∈ Fs.
s dws)) = ρ(y,p)(d)ρ(y,p)(ws) = 0.
On the other hand, if sy = y, then as y ∈ Xs, we have y ∈ F ◦
s , so s ∈ H y. Choose d ∈ D so that
d(y) = 1 and supp d ⊆ F ◦
s . Then dws ∈ Dc, so that
ρ(p,y)(ws) = ρ(y,p)(dws) = τ (dws) = τ (ws).
Therefore, ρ(p,y) = τ .
Let U(C, Dc) = {τ ∈ Mod(C, Dc) : τ Dc extends uniquely to C}. The previous paragraph shows
that B ⊆ U(C, Dc). By Theorem 3.13, B ⊆ U(C, Dc) = Ss(C, Dc), so (27) holds.
37
/
Suppose now that a ∈ L(C, Dc). Then for every ρ ∈ Ss(C, Dc), we have ρ(a∗a) = 0. In particular,
for each (y, p) ∈ Y × P ,
0 = ρ(y,p)(a∗a) = p(y) (Φy(a∗a)) .
Now cH y = {p(y) : p ∈ P }, so holding y fixed and varying p, yields Φy(a∗a) = 0. Hence, we
have Ee(a∗a)(y) = E(a∗a)(y) = 0 for every y ∈ Y . By Baire's theorem, Y is dense in X, so that
E(a∗a) = 0. Since E is faithful, a = 0. This gives the theorem in the case when Γ is countable.
We turn now to the general case. Let Γ be any discrete group and suppose a ∈ L(C, Dc). Then
there exists a countable subgroup Γ1 ⊆ Γ such that a ∈ D ⋊r Γ1. Put C1 = D ⋊r Γ1 and let
D1 = {x ∈ C1 : dx = xd for all d ∈ D} be the relative commutant of D in C1. By Proposition 6.7,
we have a ∈ L(C1, D1) = (0). This completes the proof.
(cid:3)
We collect the main results of this section into a main theorem.
Theorem 6.10. Let X be a compact Hausdorff space and let Γ be a discrete group acting as
homeomorphisms on X. Let C = C(X) ⋊r Γ and D = C(X). The following statements are
equivalent.
a) For every x ∈ X, the germ isotropy group H x is abelian;
b) The relative commutant, Dc, of D in C is abelian;
c) L(C, Dc) = (0);
d) (C, D) regularly embeds into a C ∗-diagonal.
Proof. Theorem 6.6 gives the equivalence of (a) and (b) and Theorem 6.9 shows that (a) implies
(c).
Suppose (c) holds. Since Rad(C, Dc) ⊆ L(C, Dc), Theorem 5.9 shows that (C, Dc) regularly
embeds into a C ∗-diagonal. Lemma 2.3 shows that the inclusion map of (C, D) into (C, Dc) is
a regular ∗-monomorphism. Composing the embedding of (C, Dc) into a C ∗-diagonal with the
inclusion map shows that (c) implies (d).
Finally, if (d) holds, Theorem 5.1 shows that Dc is abelian, so (d) implies (b).
(cid:3)
7. A Description of S(C, D) for a Regular MASA Inclusion
For a regular inclusion, (C, D), the D-radical, Rad(C, D) is the intersection of the left kernels of
compatible states, and when (C, D) is a regular MASA inclusion, L(C, D) is the intersection of the
left kernels of strongly compatible states. Question 5.6 asks whether it is possible for these ideals
to be distinct. In order to make progress on this question, it seems likely that a description of
S(C, D) will be useful. The purpose of this section is to provide this description.
The description is in terms of groups which are determined locally by the action of N(C, D) and
certain positive definite forms on these groups.
We begin with some generalities on T-groups, and describe a class of positive-definite functions
on T-groups which behave like compatible states. The following is more-or-less standard.
Definition 7.1. Let G be a locally compact group with identity element 1, and let U be the
connected component of the identity. We say that G is a T-group if U is clopen, isomorphic and
homeomorphic to T, and contained in the center of G. A subgroup H of G is a T-subgroup of G
if H contains U . When G is a T-group, we will always identify U with T (and so will write G/T
instead of G/U ).
Equivalently, a T-group is a central extension of T by a discrete group K,
1 → T ֒→ G
q
։ K → 1.
If f : G → T is a continuous homomorphism, we define the index of f to be the unique integer n
for which f (λ) = λn for every λ ∈ T.
38
As a set, G may be identified with T × K, and the topology on G is the product of the usual
topology on T with the discrete topology on K. Also, the Haar measure on G is the product of
Haar measure on T with the counting measure on K.
The T-subgroups of G are in one-to-one correspondence with the subgroups of K: if H is a T-
subgroup of G, then q(H) is a subgroup of K and for any subgroup Γ of K, q−1(Γ) is a T-subgroup
of G.
We also recall that a function f : G → C is positive definite if f is continuous, and if for every
n ∈ N and g1, . . . , gn ∈ G, the n × n complex matrix, A := (f (g−1
Proposition 7.2. Let G be a T-group with identity 1.
i gj))i,j satisfies A ≥ 0.
(1) Let f : G → C be a positive-definite function such that f (1) = 1 and which satisfies
f (g) ∈ {0, 1} for every g ∈ G. Set
H := {g ∈ G : f (g) 6= 0}.
Then
a) f (g1g2) = f (g1)f (g2) for any g1, g2 ∈ G such that H ∩ {g1, g2} 6= ∅; and
b) H is a T-subgroup of G and f H is a continuous homomorphism of H onto T.
(2) Let H ⊆ G be a T-subgroup and suppose φ : H → T is a continuous homomorphism. Define
f : G → C by
f (g) =(φ(h)
0
if h ∈ H
if h /∈ H.
Then f is a positive definite function on G such that f (1) = 1 and for every g ∈ G,
f (g) ∈ {0, 1}.
Proof. Suppose f satisfies the hypotheses in (1). Since f (1) = 1 and f is positive definite, we have
f (g) = f (g−1) for every g ∈ G. Continuity of f and connectedness of T yield T ⊆ H.
Recall that if H1 and H2 are Hilbert spaces, A ∈ B(H1), C ∈ B(H2), and B ∈ B(H2, H1) with
A invertible, then
(cid:18) A B
B∗ C(cid:19) ≥ 0
and h3 = g2) implies that
(28)
if and only if A ≥ 0, C ≥ 0 and C − B∗A−1B ≥ 0.
Let g1, g2 ∈ G. Then the positive definiteness of f (using the group elements h1 = 1, h2 = g−1
1 ,
Applying (28) with A = (1), B =(cid:0)f (g−1
1
f (g1)
f (g−1
1
f (g2)
f (g1g2)
f (g−1
1 )
1
2 g−1
2 ) f (g−1
1 )
≥ 0.
1 ) f (g2)(cid:1), and C =(cid:18)
2 )f (g2) (cid:19) =: M.
(cid:19) .
1
2 g−1
f (g−1
1 )
f (g1g2) − f (g1)f (g2)
f (g1g2) − f (g1)f (g2)
2 ) = 1, so that
1 − f (g−1
0
1 − f (g−1
1 ) − f (g−1
1 )f (g1)
2 g−1
2 )f (g−1
1 )
f (g−1
1 − f (g1)f (g−1
1 )
2 )f (g−1
1 ) − f (g−1
1 )
2 g−1
f (g−1
0 ≤(cid:18)
M =(cid:18)
f (g1g2)
1 (cid:19) gives
Suppose now that g2 ∈ H, that is, f (g2) 6= 0. Then f (g2)f (g−1
But then 0 ≤ det(M ) = −f (g1g2) − f (g1)f (g2)2, so f (g1g2) = f (g1)f (g2).
The case when f (g1) 6= 0 is the same. Thus when H ∩ {g1, g2} 6= ∅ we obtain,
f (g1g2) = f (g1)f (g2).
39
The facts that H is a T-subgroup of G and f H : H → T is a continuous homomorphism are now
apparent.
Turning now to statement (2), let H be a T-subgroup of G, and φ a continuous homomorphism
of H into T. Let f : G → T be given as in the statement. The continuity of f is clear, as is
the fact that f (1) = 1 and f (g) ∈ {0, 1} for every g ∈ G. To show f is positive definite, let
g1, . . . , gn ∈ G. Since φ(h) = φ(h−1), it follows that f (g−1
j gi). Put X = {1, . . . , n}.
Define an equivalence relation R on X by (i, j) ∈ R if and only if g−1
i gj ∈ H, and let X/R be the
set of equivalence classes. Let q : X → X/R be the map which sends j ∈ X to its equivalence class,
and let u : X/R → X be a section for q. Let δx,y be the Kronecker delta function on X/R and for
x ∈ X/R, define
i gj) = f (g−1
cx :=(cid:16)f (g−1
u(q(1))g1)δq(1),x
. . .
f (g−1
u(q(n))gn)δq(n),x(cid:17) .
Then c∗
xcx is an n × n matrix whose i, j-th entry is
f (g−1
i gu(q(i)))f (g−1
u(q(j))gj)δq(i),xδq(j),x =(f (g−1
0
i gj)
if q(i) = q(j) = x
otherwise.
Hence the i, j-th entry ofPx∈X/R c∗
(f (g−1
xcx is f (g−1
i gj) if (i, j) ∈ R and 0 otherwise. Therefore
i gj))i,j∈X = Xx∈X/R
c∗
xcx ≥ 0,
as desired.
(cid:3)
Corollary 7.3. Let f be a continuous positive definite function on the T-group G such that f (g) ∈
{0, 1} for every g ∈ G. Then there exists p ∈ Z such that for every λ ∈ T and g ∈ G,
f (λg) = λpf (g).
Proof. The set {g ∈ G : f (g) 6= 0} contains T, and Proposition 7.2 shows the restriction of f to T
is a character on T. So there exists p ∈ Z such that f (λ) = λp for every λ ∈ T. The corollary now
follows from another application of Proposition 7.2.
(cid:3)
Definition 7.4. Given a T-group G, call a positive-definite function f on G satisfying f (1) = 1
and f (g) ∈ {0, 1} a pre-homomorphism.
We will call the number p appearing in Corollary 7.3 the index of f , and will denote it by ind(f ).
Finally, the group H := {g ∈ G : f (g) 6= 0} will be called the supporting subgroup for f , and will
be denoted by supp(f ).
Notation. Some notation will be useful.
(1) Let (C, D) be an inclusion. For σ ∈ D, let
S(C, D, σ) := {ρ ∈ S(C, D) : ρD = σ},
Ss(C, D, σ) := {ρ ∈ Ss(C, D) : ρD = σ}.
and
(2) For any T-group G, let pHom1(G) denote the set of all pre-homomorphisms f : G → T∪{0}
with ind(f ) = 1.
(3) Finally, recall the seminorms, Bρ,σ on C from [10, Definition 2.4]: for each ρ, σ ∈ D, the
seminorm Bρ,σ is defined on C by
Bρ,σ(x) := inf{kdxek : d, e ∈ D, ρ(d) = σ(e) = 1}
(x ∈ C).
We shall require these seminorms for ρ = σ; however, instead of writing Bσ,σ we shall write
Bσ.
40
Associated to each regular inclusion (C, D) and σ ∈ D is a certain T-group, denoted Hσ/R1,
which we now construct. We produce a distinguished unitary representation T of Hσ/R1. Our goal
is to exhibit a bijection between elements of
{f ∈ pHom1(Hσ/R1) : f determines a state on the C ∗-algebra generated by T (Hσ/R1)}
and S(C, D, σ).
Definition 7.5. Let (C, D) be an inclusion, and let σ ∈ D.
(1) Define
Hσ := {v ∈ N(C, D) : σ(v∗dv) = σ(d) for every d ∈ D}.
We remark that Hσ is the set which arises when considering the D-stabilizer of ρ ∈
Mod(C, D), where σ = ρD (see Definition 4.13). However, there our interest was in a
particular extension of σ, while here we do not wish to specify the extension.
Notice that for v ∈ Hσ, we have σ(v∗v) = 1, and that Hσ is closed under the adjoint
operation: replace d by v∗dv in the definition. Furthermore, it is easy to see that Hσ is a
∗-semigroup.
(2) Let Λ ⊆ T be a subgroup, (we write the product multiplicatively). Define
RΛ := {(v, w) ∈ Hσ × Hσ : Bσ(λI − w∗v) = 0 for some λ ∈ Λ}.
When Λ = {1}, we write R1 instead of R{1}.
Lemma 7.6. Let (C, D) be an inclusion, let σ ∈ D, and let v, w ∈ Hσ. Then
Bσ(v) = 1 and Bσ(v − w) = Bσ(I − v∗w) = Bσ(I − vw∗).
Proof. Note that for any h ∈ D, σ(h) = inf{kd∗hdk : d ∈ D, σ(d) = 1}. Hence given ε > 0, and
e ∈ D with σ(e) = 1, we may find d ∈ D with σ(d) = 1 and
kd∗v∗e∗evdk − σ(v∗e∗ev) < ε.
Since σ(v∗e∗ev) = σ(e∗e) = 1, we obtain 1 − ε < kevdk2 < 1 + ε. Hence 1 − ε < Bσ(v)2 < 1 + ε,
and the fact that Bσ(v) = 1 follows.
Next, let x, d, e ∈ D with σ(x) = σ(d) = σ(e) = 1. Since σ(xvdv∗) = 1, we have
Bσ(v − w) ≤ k(xvdv∗)(v − w)ek = kxv(dv∗ve − dv∗we)k
≤ kxvk kdv∗ve − dv∗wek
≤ kxvk [kdv∗ve − dek + kd(I − v∗w)ek] .
It follows that
Bσ(v − w) ≤ Bσ(I − v∗w).
A similar argument using multiplication on the right by w∗ewx gives Bσ(I − v∗w) ≤ Bσ(w∗ − v∗).
But Bσ(v∗ − w∗) = Bσ(v − w), so we obtain Bσ(v − w) = Bσ(I − v∗w). Finally, Bσ(v − w) =
Bσ(v∗ − w∗) = Bσ(I − vw∗).
(cid:3)
Proposition 7.7. Let (C, D) be a regular MASA inclusion, σ ∈ D and suppose that v ∈ Hσ. Let
Λ ⊆ T be a subgroup. The following statements are equivalent:
(1) for some λ ∈ Λ, Bσ(λI − v) = 0;
(2) there exists λ ∈ Λ such that f (v) = λ whenever f ∈ Mod(C, D, σ);
(3) there exists λ ∈ Λ such that ρ(v) = λ whenever ρ ∈ Ss(C, D, σ);
(4) σ ∈ (fix βv)◦ and v(σ) ∈ Λ (where v is as in Remark 3.19);
(5) there exists λ ∈ Λ and h, k ∈ D such that σ(h) = 1 = σ(k) and vh = λk.
41
Proof. (1) ⇒ (2). If f ∈ Mod(C, D, σ) and x ∈ C, then f (x) = f (dxe) ≤ kdxek whenever d, e ∈ D
and σ(d) = σ(e) = 1. Thus, f (x) ≤ Bσ(x) for every x ∈ C. The implication (1) ⇒ (2) follows.
(2) ⇒ (3) is trivial.
(3) ⇒ (4). Suppose λ ∈ Λ and that ρ(v) = λ for every ρ ∈ {f ∈ Ss(C, D) : f D = σ}. By
Lemma 2.14, σ ∈ fix βv. To show that σ ∈ (fix βv)◦, we argue by contradiction. So suppose that
σ ∈ fix βv \(fix βv)◦. Then every neighborhood of σ contains an element in D \fix βv. Hence we may
find a net (σs) in D such that σs → σ and such that σs /∈ fix βv. By Theorem 3.21, the restriction
map f 7→ f D from Ss(C, D) to D is onto. Thus we may choose fs ∈ Ss(C, D) such that fsD = σs.
By passing to a subnet if necessary, we may assume that fs converges to a state f . Theorem 3.21
shows that Ss(C, D) is closed, so f ∈ Ss(C, D). Clearly f D = σ. Lemma 2.14 gives fs(v) = 0 for
every s, so 0 6= λ = f (v) = lims fs(v) = 0. This is absurd, so we conclude that σ ∈ (fix βv)◦.
Let ρ ∈ Ss(C, D, σ). Then for h ∈ Jv, we have
(29)
v(σ)σ(h) = σ(vh) = ρ(vh) = ρ(v)σ(h).
By Proposition 3.2, σJv 6= 0. Statement (4) now follows from equation (29).
(4) ⇒ (5). Let λ = v(σ) and choose h ∈ Jv such that σ(h) = 1. Then σ(vh) = v(σ). Put
k = v(σ)vh.
(5) ⇒ (1). Let h, k ∈ D be chosen so that σ(h) = σ(k) = 1 and vh = λk. Then
Bσ(λI − v) ≤
inf
{d∈D:σ(d)=1}
kd(λI − v)hdk = σ(λh − vh) = 0.
Thus (1) holds.
(cid:3)
We next observe that Bσ gives the quotient norm on the quotient of spanHσ by the ideal generated
by ker σ.
Proposition 7.8. Let (C, D) be a regular inclusion and suppose σ ∈ D. Let Cσ = spanHσ. Then
(Cσ, D) is a regular inclusion. If Iσ is the closed, two-sided ideal of Cσ generated by ker σ, then Bσ
vanishes on Iσ, and for x ∈ Cσ,
Bσ(x) = inf{kx + jk : j ∈ Iσ}.
Moreover, the following statements hold.
(1) If ρ is a state on Cσ which annihilates Iσ, then ρ extends uniquely to a state ρ on C. When
ρ ∈ S(Cσ, D) annihilates Iσ, ρ ∈ S(C, D).
(2) The map Hσ ∋ u 7→ u + Iσ ∈ Cσ/Iσ is a ∗-homomorphism of the ∗-semigroup Hσ into the
unitary group of Cσ/Iσ.
Proof. Since Hσ is closed under multiplication and the adjoint map, we see that Cσ is a C ∗-algebra.
For d ∈ D, d = limt→0 tI + d. But for all sufficiently small t 6= 0, (t + σ(d))−1(tI + d) ∈ Hσ,
so d is a limit of scalar multiples of elements of Hσ. So D ⊆ Cσ. Therefore (Cσ, D) is a regular
inclusion.
Next, suppose that x ∈ C and d ∈ ker σ. Then Bσ(xd) = Bσ(dx) = 0. When x = v ∈ Hσ and
y ∈ Cσ,
Bσ(vdy) = Bσ(vv∗vdy) = Bσ(vdv∗vy) = σ(vdv∗)Bσ(vy) = 0.
It follows that when x ∈ span Hσ, we have Bσ(xdy) = 0. Taking closures we obtain Bσ(xdy) = 0
when x, y ∈ Cσ and d ∈ ker σ. Therefore, Bσ(z) = 0 for every z ∈ Iσ.
This gives Bσ(x) ≤ Bσ(x + j) + Bσ(j) = Bσ(x + j) ≤ kx + jk for every x ∈ Cσ and j ∈ Iσ. Thus,
Bσ(x) ≤ dist(x, Iσ).
42
Notice that if ρ is a state on Cσ which annihilates Iσ, then ρD annihilates ker σ, so ρD = σ.
Any extension of σ to a state g on C belongs to Mod(C, D). Hence g(x) ≤ Bσ(x) for every x ∈ C.
In particular, ρ(x) ≤ Bσ(x) for every x ∈ Cσ. So if x ∈ Cσ, we obtain,
kx + Iσk = sup{ρ(x) : ρ is a state on Cσ and ρIσ = 0} ≤ Bσ(x).
Hence Bσ gives the quotient norm on Cσ/Iσ.
Turning now to statement (1), let ρ be a state on Cσ which annihilates Iσ, and suppose for
i = 1, 2, that τi are states on C with τiCσ = ρ. Let v ∈ N(C, D) \ {λu : u ∈ Hσ and λ ∈ C}.
We claim that τi(v) = 0. Since ρ annihilates Iσ, we have σ = ρD = τiD. Suppose that
σ(v∗v) 6= 0. By multiplying v by a suitable scalar, we may assume that σ(v∗v) = 1. Since v is not
a scalar multiple of an element of Hσ, we see that βv(σ) 6= σ. Lemma 2.14 shows that τi(v) = 0.
On the other hand, if σ(v∗v) = 0, then v∗v ∈ Iσ, so τi(v) = limn→∞ τi(v(v∗v)1/n) = 0. Thus the
claim holds.
Since τ1(v) = τ2(v) for every v ∈ Hσ, we have τ1(v) = τ2(v) for every v ∈ N(C, D). Hence
τ1 = τ2. Thus ρ extends uniquely to a state on C. Notice also that if ρ ∈ S(Cσ, D), then this
argument shows that ρ ∈ S(C, D).
To prove statement (2), use the fact that Bσ gives the quotient norm on Cσ/Iσ and apply
(cid:3)
Lemma 7.6 to u ∈ Hσ.
The following is a corollary of Proposition 7.8.
Proposition 7.9. Let (C, D) be a regular inclusion, σ ∈ D and let Λ be a subgroup of T. Then
RΛ is an equivalence relation on Hσ. Denote the equivalence class of v ∈ Hσ by [v]Λ. The product
[v]Λ[w]Λ := [vw]Λ is a well-defined product on Hσ/RΛ. With this product, [I]Λ is the unit and for
each v ∈ Hσ, [v]−1
Λ = [v∗]Λ. Thus Hσ/RΛ is a group.
Furthermore, the map Tσ : Hσ/R1 → Cσ/Iσ given by Tσ([u]1) = u + Iσ is a one-to-one group
homomorphism of Hσ/R1 into the unitary group of Cσ/Iσ, and Tσ(Hσ/Iσ) generates Cσ/Iσ.
Proof. Let u, v ∈ Hσ. By Proposition 7.8 and Lemma 7.6, (u, v) ∈ RΛ if and only if there exists
λ ∈ Λ such that u + Iσ = λv + Iσ. Routine arguments now show that Hσ/RΛ is a group under the
indicated operations. The final statement follows from Proposition 7.8(2).
(cid:3)
Lemma 7.10. Let (C, D) be a regular MASA inclusion, σ ∈ D, and suppose u, v ∈ Hσ are such
that (u, v) ∈ R1. Then the following statements hold.
(1) If ρ ∈ S(C, D, σ), then
ρ(v) = ρ(u).
If in addition, 0 6= ρ(v) then ρ(v) ∈ T.
(2) σ ∈ (fix βv)◦ if and only if σ ∈ (fix βu)◦, and when this occurs, v(σ) = u(σ) ∈ T.
Proof. Suppose ρ(v) 6= 0. Since ρ(x) ≤ Bσ(x) for every x ∈ C and Bσ(I − u∗v) = 0, we have
ρ(u∗v) = 1. Therefore, by part 1 of Proposition 4.3,
ρ(u) = ρ(u)ρ(u∗v) = ρ(uu∗v) = σ(uu∗)ρ(v) = ρ(v).
Likewise, if ρ(u) 6= 0, then ρ(u) = ρ(v). Thus, we have ρ(u) = ρ(v) whenever ρ ∈ S(C, D, σ).
Next, when ρ(v) 6= 0, the fact that ρ ∈ S(C, D) gives
ρ(v)2 = ρ(v∗v) = σ(v∗v) = 1,
so ρ(v) ∈ T. This completes the proof of the first statement.
We now turn to the second statement. Since (u, v) ∈ R1, Proposition 7.7 implies σ ∈ (fix βv∗u)◦.
Thus σ ∈ (fix βv)◦ if and only if σ ∈ (fix βu)◦.
43
Next suppose that σ ∈ (fix(βv))◦. Let (I(D), ι) be an injective envelope for D and let E be the
pseudo-expectation for ι. Let h ∈ Jv satisfy σ(h) = 1 and let ρ ∈ [I(D) be such that ρ ◦ ι = σ. Then
ρ ◦ E ∈ Ss(C, D), and Proposition 3.18 shows that ρ ∈ supp([E(v)). Thus
v(σ) = σ(vh) = ρ(E(v)) = ρ(E(w)) = σ(wh) = w(σ).
Since ρ(E(v)) 6= 0, we have v(σ) ∈ T by part (1).
(cid:3)
Remark. Lemma 7.10 shows that for ρ ∈ S(C, D) with ρD = σ, we have a well-defined map
ρ : Hσ/R1 → T ∪ {0} given by ρ([v]) = ρ(v).
Theorem 7.11. Let (C, D) be a regular MASA inclusion, and let σ ∈ D. The function
is a well defined metric on Hσ/R1 and makes Hσ/R1 into a T-group. More specifically, the following
statements hold.
d([v], [w]) := Bσ(v − w)
(1) Let
U = {[v] ∈ Hσ/R1 : σ ∈ (fix βv)◦}.
Then U is clopen and is the connected component of the identity in Hσ/R1.
(2) The subgroup U is contained in the center of Hσ/R1.
(3) The map [v] ∈ U 7→ v(σ) is an isomorphism of U onto T.
(4) The quotient of Hσ/R1 by U is Hσ/RT.
Proof. Let Tσ be the isomorphism of Hσ/R1 onto a subgroup of the unitary group of Cσ/Iσ defined
in Proposition 7.9. Then
d([v], [w]) = kTσ([v]) − Tσ([w])kCσ/Iσ .
It follows that d is a well-defined metric which makes Hσ/R1 into a topological group.
We now show that U is an open set. Let u ∈ Hσ be such that [u] ∈ U and suppose that v ∈ Hσ
satisfies d([u], [v]) < 1/2. We will show that σ ∈ (fix βv)◦. To do this we modify the proof of the
implication (3) ⇒ (4) in Proposition 7.7 slightly. Since Bσ(u − v) < 1/2, for every ρ ∈ Ss(C, D),
we have ρ(u) − ρ(v) < 1/2.
Suppose, to obtain a contradiction, that σ /∈ (fix βv)◦. Then we may find find a directed set S
and a net (σs)s∈S such that σs /∈ fix βv for every s and such that σs → σ. As usual, let (I(D), ι) be
an injective envelope for D and let E be the pseudo-expectation for ι. For each s, choose τs ∈ [I(D)
such that τs ◦ι = σs. Passing to a subnet if necessary, we may assume that τs converges to τ ∈ [I(D).
Then τ ◦E ∈ Ss(C, D) and τ ◦ED = σ. Notice that τ (E(v)) 6= 0 because τ (E(v))−τ (E(u)) < 1/2
and τ (E(u)) ∈ T. Since τs ◦ ED = σs /∈ fix βv, Lemma 2.14 shows that τs(E(v)) = 0 for every
s ∈ S. Then
τ (E(v)) = lim
s
τs(E(v)) = 0,
contradicting the fact that τ (E(v)) 6= 0. Thus σ ∈ (fix βv)◦. Therefore [v] ∈ U , so U is an open
subset of Hσ/R1.
Similarly, the complement of U is open in Hσ/R1, so U is also closed.
Let γ : U → T be the map γ([u]) = u(σ). Suppose that [u], [v] ∈ U and u(σ) = v(σ). Then
d([u], [v]) = Bσ(u − v) ≤ Bσ(u − u(σ)I) + Bσ(v(σ)I − v) = 0,
so γ is one-to-one. Since γ([λu]) = λγ([u]) for any λ ∈ T, γ is onto.
Let ρ ∈ S(C, D, σ). Since
ρ([v]) − ρ([w]) = ρ(v − w) ≤ Bσ(v − w) = d([v], [w]),
44
we see that ρ is a continuous map on Hσ/R1. By Lemma 7.10, for [v] ∈ U, ρ([v]) = γ([v]). So γ is
also continuous. The map λ ∈ T 7→ [λI] is the inverse of γ, and we see that γ is a homeomorphism.
In particular, U is connected, and hence U is the connected component of the identity in Hσ/R1.
To see that U is contained in the center of Hσ/R1, observe that for [u] ∈ U , we have [u] = [γ([u])I],
which evidently belongs to the center of Hσ/R1.
Since the connected component of the identity is compact, Hσ/R1 is a locally compact group.
Finally, [v] = [w] mod U if and only if [v∗w] ∈ U . Therefore, [v] = [w] mod U if and only if
Bσ(γ([v∗w])I − v∗w) = 0. Hence the quotient of Hσ/R1 by U is Hσ/RT.
(cid:3)
We now are prepared to exhibit a bijection between S(C, D, σ) and a class of pre-homomorphisms
on Hσ/R1. We pause for some notation.
Let q : Cσ → Cσ/Iσ be the quotient map. Define a ∗-homomorphism θ : Cc(Hσ/R1) → Cσ/Iσ by
θ(φ) =ZHσ/R1
φ(s)Tσ(s) ds,
where ds is Haar measure on Hσ/R1. Then the image of θ is dense in Cσ/Iσ.
Definition 7.12. We will say that a positive definite function f on Hσ/R1 is dominated by Bσ if
for every φ ∈ Cc(G),
ZHσ/R1
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
φ(t)f (t) dt(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ZHσ/R1
φ(t)Tσ(t) dt(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Cσ/Iσ
.
Theorem 7.13. Let (C, D) be a regular MASA inclusion and let σ ∈ D. Let
Mσ := {f ∈ pHom1(Hσ/R1) : f is dominated by Bσ}.
For τ ∈ S(C, D, σ) the map τ : Hσ/R1 → C given by
τ ([v]1) = τ (v)
is well-defined, and τ ∈ Mσ. Moreover, the map τ 7→ τ is a bijection between S(C, D, σ) and Mσ.
Proof. If τ ∈ S(C, D, σ), then τ (x) ≤ Bσ(x). Hence τ annihilates Iσ, so that τ determines a state
τ ′ on Cσ/Iσ such that
Then τ = τ ′ ◦ Tσ, so τ is well-defined.
Set
τ Cσ = τ ′ ◦ q.
G := {[v] ∈ Hσ/R1 : τ ([v]) 6= 0}.
Proposition 4.3(1) implies that G is closed under products. For v ∈ Hσ with [v] ∈ G, we have
τ (v) ∈ T, and as τ is a state, τ (v∗) = τ (v). Therefore, [v]−1 = [v∗] ∈ G, so G is closed under
inverses. It follows that G is a T-subgroup of Hσ/R1. Since τ ∈ S(C, D), τ ([v]) ∈ {0, 1} for every
[v] ∈ Hσ/R1. Proposition 7.2 implies that τ is a pre-homomorphism on Hσ/R1. Since τ is linear,
for λ ∈ T and v ∈ Hσ, we have τ ([λv]1) = τ (λv) = λτ ([v]1). So ind(τ ) = 1.
For φ ∈ Cc(Hσ/R1) we have
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
ZHσ/R1
φ(s)τ (s) ds(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
=(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
τ ′ ZHσ/R1
φ(s)Tσ(s) ds!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
ZHσ/R1
φ(s)Tσ(s) ds(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Cσ/Iσ
.
Thus, τ is dominated by Bσ. Therefore, τ ∈ Mσ.
45
Next we show that the map τ 7→ τ is surjective. So suppose that f : Hσ/R1 → {0} ∪ T is a
pre-homomorphism dominated by Bσ and ind(f ) = 1. Then the map F0 : θ(Cc(Hσ/R1)) → C
given by
F0 ZHσ/R1
φ(t)Tσ(t) dt! =ZHσ/R1
φ(t)f (t) dt
(φ ∈ Cc(Hσ/R1))
extends by continuity to a bounded linear functional F on Cσ/Iσ. Since θ is a ∗-homomorphism
and f is a positive definite function, F is a positive linear functional. Clearly kF k ≤ 1.
We next show that F (v +Iσ) = f ([v]) for every v ∈ Hσ. Since Hσ/R1 is a T-group, the connected
component of the identity is T. Since f is continuous, ind(f ) = 1 and f (1) = 1, we have f (λ) = λ
for every λ ∈ T. Given v ∈ Hσ, let φ ∈ Cc(Hσ/R1) be the function given by
φ(t) =(λ if t = [λv]1 for some λ ∈ T
otherwise.
0
Now for t ∈ {[λv]1 : λ ∈ T} ⊆ Hσ/R1, we have Tσ(t) = λq(v), so
Thus,
θ(φ) =ZHσ/R1
F (q(v)) =ZHσ/R1
φ(t)Tσ(t) dt =ZT
φ(t)f (t) dt =ZT
λq(λv)dt = q(v).
λf ([λv]1) dt = f ([v]1).
It follows that kF k = 1, so F is a state on Cσ/Iσ.
As N(Cσ, D) ⊆ N(C, D), we see that if w ∈ N(Cσ, D), then w is a scalar multiple of an element of
Hσ. Since F ◦ q(v) = f ([v]1) ∈ {0, 1} for each v ∈ Hσ, we find F ◦ q ∈ S(Cσ, D). Proposition 7.8
shows that F ◦ q extends uniquely to an element τ ∈ S(C, D, σ). As τ = f , we find that the map
τ 7→ τ is onto.
To show that τ 7→ τ is one-to-one, suppose τ and τ1 belong to S(C, D, σ) and τ = τ1. Then
τ (v) = τ1(v) for every v ∈ Hσ, so that τ Cσ = τ1Cσ . Proposition 7.8 then shows τ = τ1.
Thus the mapping τ 7→ τ is indeed a bijection.
(cid:3)
We conclude this section with a pair of very closely related questions and two examples.
Notation 7.14. Let (C, D) be a regular inclusion and let
B(C, D) := {x ∈ C : Bσ(x) = 0 for all σ ∈ D}.
By Proposition 7.8, B(C, D) = {x ∈ C : ρ(x∗x) = 0 for all ρ ∈ Mod(C, D)}, and by Proposi-
tion 2.16, B(C, D) is an ideal of C which satisfies
(30)
B(C, D) ⊆ Rad(C, D).
Question 7.15. Let (C, D) be a regular MASA inclusion.
(1) Let σ ∈ D. For x ∈ C, is Bσ(x) = supρ∈S(C,D) ρ(x∗x)1/2?
(2) Is B(C, D) = Rad(C, D)?
When Rad(C, D) = (0), the answer to both questions is yes.
Example 7.16. This example applies the previous results to a special case of reduced crossed
products to produce a Cartan inclusion (C, D) and a pure state τ on C such that τ D ∈ D, yet
τ /∈ S(C, D).
46
Let Γ be an infinite discrete group, let X := Γ ∪ {∞} be the one-point compactification of Γ and
let Γ act on X by extending the left regular representation to X: for s ∈ Γ and x ∈ X, let
sx :=(sx if x ∈ Γ;
if x = ∞.
x
We now use the notation from Section 6: let C = C(X) ⋊r Γ and D be the cannonical image of
C(X) in C. We identify D with X. The action of Γ on X is topologically free, so (C, D) is a regular
MASA inclusion. Moreover, the conditional expectation E : C → D is the pseudo-expectation.
Let σ ∈ D be the map σ(f ) = f (∞). We claim that Hσ/RT = Γ and Hσ/R1 = T × Γ. Observe
first that S(C, D, σ) = {σ ◦ E} and the map θ : Γ → Hσ/RT given by θ(s) = [ws]T is a group
homomorphism. Next, suppose v ∈ Hσ. Then there exists t ∈ Γ, so that Et(v) = E(vwt−1 ) 6= 0.
Since both v and wt−1 ∈ Hσ we have vwt−1 ∈ Hσ, so σ(E(vwt−1 )) 6= 0. By Proposition 7.7,
(v, wt) ∈ RT. Thus θ is surjective. Proposition 7.7 also implies θ is one-to-one:
if [ws]T = [wt]T
then σ(E(wst−1 )) 6= 0, so that E(wst−1) 6= 0. Thus s = t. Therefore θ is an isomorphism of Hσ/RT
onto Γ, and we use θ to identify Γ with Hσ/RT. The map Γ ∋ s 7→ [ws]1 ∈ Hσ/R1 is a group
homomorphism and also a section for the quotient map of Hσ/R1 onto Hσ/RT. It follows that
Hσ/R1 is isomorphic to T × Γ.
Let ρ = σ ◦ E and let (πρ, Hρ) be the GNS representation of C associated to ρ. Proposition 7.8
implies that Iσ ⊆ ker πρ, so πρ induces a representation, again denoted πρ, of Cσ/Iσ on Hσ.
We shall show that the image of Cσ/Iσ under πρ is isomorphic to C ∗
r (Γ). To do this, first
observe that for v, w ∈ Hσ, v + Lρ = w + Lρ if and only if (v, w) ∈ R1. Indeed, since v, w ∈ Hσ,
ρ(w∗w) = ρ(v∗v) = 1, so ρ((v − w)∗(v − w)) = 2 − 2ℜ(ρ(v∗w)). Since ρ(v∗w) ∈ {0, 1}, we get
v + Lρ = w + Lρ if and only if ρ(v∗w) = 1, which by Proposition 7.7 gives the observation. This
observation and regularity of Cσ implies that
{ws + Lρ : s ∈ Γ}
is an orthonormal basis for Hρ. Thus there is a unitary operator U : ℓ2(Γ) onto Hρ which carries
the basis element δs ∈ ℓ2(Γ) to ws + Lρ ∈ Hρ.
Now let λ : Γ → B(ℓ2(Γ)) be the left regular representation. For s, t ∈ Γ we have U λ(s)δt =
wst + Lρ = πρ(ws)U δt, so
(31)
U λ(s) = πρ(ws)U.
r (Γ).
It follows from Proposition 7.9 that the set {ws + Iσ : s ∈ Γ} generates Cσ/Iσ, so (31) shows that
πρ(Cσ/Iσ) is isomorphic to C ∗
Thus there is a surjective ∗-homomorphism Ψ : Cσ → C ∗
r (Γ) which annihilates Iσ. The composi-
tion of Ψ with any pure state f on C ∗
r (Γ) yields a pure state on Cσ, which in turn may be extended
to a pure state τ ∈ Mod(C, D, σ). Apply this process when Γ = F2 is the free group on 2-generators
u1 and u2. By [26, Theorem 2.6 and Remark 3.4], there exists a pure state f on C ∗
r (F2) such that
f (u1) /∈ {0, 1}. It follows that there exists a pure state τ on C such that τ D = σ, yet τ /∈ S(C, D).
Example 7.17. Denote by {en}n∈N the standard orthonormal basis for H := ℓ2(N). Consider the
inclusion (C, D), where C is the Toeplitz algebra (the C ∗-algebra generated by the unilateral shift
S acting on H) and D ⊆ B(H) is the C ∗-algebra generated by {SkS∗k : k ≥ 0}. Then (C, D) is a
regular MASA inclusion and D is homeomorphic to the one-point compactification of N, N ∪ {∞}.
We identify D with this space.
Here the pseudo-expectation is the conditional expectation E : C → D which takes T ∈ C to the
operator E(T ) which acts on basis elements via E(T )en = hT en, eni en.
We shall do the following:
(1) give a description of S(C, D);
47
(2) show that not every element of S(C, D) is a pure state of C and identify the pure states in
S(C, D).
The strongly compatible states are easy to identify. Let ρn and ρ∞ be the states on C given by
ρn(X) = hXen, eni and ρ∞(X) = limn→∞ ρn(X). Then
Ss(C, D) = {ρn : n ∈ N ∪ {∞}} = {σ ◦ E : σ ∈ D}.
For each n ∈ N, the set {n} is clopen in D, so ρn is the unique extension of ρnD to a state on C.
(This can be proved directly or viewed as a consequence of Theorem 3.8.)
Thus, to complete a description of S(C, D), we need only describe S(C, D, σ∞), where σ∞ =
ρ∞D.
To do this, let K = K(H) be the compact operators and let q : C → C/K = C(T) be the quotient
map. Given z ∈ T, we write τz for the state on C given by τz(T ) = q(T )(z). Also, for z ∈ T,
we let αz be the gauge automorphism on C determined by αz(S) = zS. For each N ∈ N, let
λ(N ) = exp(2πi/N ) and define ΦN : C → C by
ΦN (T ) :=
1
N
αk
λ(N )(T ).
N −1Xk=0
(Note that if T =Pp
We claim that
k=−p akSk is a "trigonometric polynomial," then ΦN (T ) =Pk∈NZ akSk.)
(32)
S(C, D, σ∞) = {ρ∞} ∪ {τz ◦ ΦN : z ∈ T, N ∈ N}.
Each state of the form τz ◦ Φ1 = τz is multiplicative on C. Therefore, {τz ◦ Φ1 : z ∈ T} is a set
of pure states and is a subset of S(C, D). Also, we have
ρ∞ =ZT
τz dz
and for N ≥ 1,
τz ◦ ΦN =
1
N
N −1Xk=0
τλ(N )kz,
so the only pure states on the right hand side of (32) are those of the form τz. We will show that
states of the form τz ◦ ΦN are compatible states. We proceed by first identifying the elements of
N(C, D) with σ∞(v∗v) 6= 0.
Suppose v ∈ N(C, D) satisfies σ∞(v∗v) > 0. Put Av = {n ∈ N : hv∗ven, eni 6= 0} and Bv = {n ∈
N : hvv∗en, eni 6= 0}. Then βv induces a bijection f : Av → Bv, and there exist scalars cj so that
vij := hvej, eii =(cj
0
if i = f (j);
otherwise.
Moreover, note that q(v) = cq(S)m for some c ∈ C and m ∈ Z. Since σ∞(v∗v) 6= 0, c 6= 0, so v
is a Fredholm operator. Let m be the Fredholm index of v. Then q(v) = ρ∞(vS−m)q(S)m. Thus
v = Smd for some d ∈ D with σ∞(d) 6= 0. The fact that each τz ◦ ΦN ∈ S(C, D), now follows. It
remains to show that we have found all elements of S(C, D, σ∞).
We will write H∞, B∞, C∞, and J∞ rather than the more cumbersome Hσ∞, Bσ∞, Cσ∞ , and
Jσ∞ . Then
H∞ = {Smd : m ∈ Z and d ∈ D, σ∞(d) ∈ T}.
Next, for v, w ∈ H∞, we have B∞(v − w) = 0 if and only if ρ∞(w∗v) = 1.
Thus, J∞ = K and C∞ = C. By Proposition 7.8, B∞ gives the quotient norm on C∞/J∞ = C(T).
In other words,
(v, w) ∈ R1 if and only if q(v) = q(w). Therefore, H∞/R1 is isomorphic to the direct product
T × Z. Observe that H∞/RT is isomorphic to Z, and we obtain the trivial T-group extension,
1 → T → T × Z → Z → 1.
48
The generators of the subgroups of Z are the non-negative integers, so the T-subgroups of T × Z
are
{T × nZ : n ≥ 0}.
Let f be a pre-homomorphism of index 1 on H∞/R1. Then there exists a non-negative integer
N such that f is a character on T × N Z. Since f has index 1, there exists λ ∈ Z such that for
(z, n) ∈ T × Z = H∞/R1,
f (z, n) =(zλn
0
if n ∈ N Z;
otherwise.
As B∞ is the quotient norm on C∞, f is dominated by B∞. Let τ = τz ◦ ΦN . With the notation
of Theorem 7.13, we get τ = f , so by Theorem 7.13, the compatible state corresponding to f is
τz ◦ ΦN . This completes the proof of (32).
We now identify the topology on S(C, D). Notice that if M, N are positive integers with N /∈ MZ,
then for any z ∈ T, τz(ΦN (SN )) = τz(SN ) = zN 6= 0 = τz(ΦM (SN )). Given distinct positive
integers N, M , either N /∈ MZ or M /∈ N Z. Thus for N > 0, {φz ◦ ΦN : z ∈ T} is a connected
component of S(C, D) and is homeomorphic to T.
We next show that if G is a weak-∗ open neighborhood of ρ∞, then there exists N ∈ N such that
Tn := {τz ◦ Φn : z ∈ Z} ⊆ G for every n ≥ N . To do this, it suffices to show that for every a ∈ C
and ε > 0, the set Ga,ε := {φ ∈ S(C, D) : φ(a) − ρ∞(a) < ε} contains Tn for all sufficiently large
n, and this is what we shall do. First observe that for every b ∈ C,
(33)
lim
n→∞
kΦn(b) − E(b)k = 0.
(This is clear for "trigonometric polynomials" in S, approximate b in norm with a trigonometric
polynomial to obtain (33).) Fix N ∈ N so that kΦn(a) − E(a)k < ε for every n ≥ N . For any
z ∈ T, we have τz(E(a)) = ρ∞(a), so we find
τz(Φn(a)) − ρ∞(a) < ε.
So Tn ⊆ Ga,ε for all n ≥ N .
We conclude that
S(C, D) = Ss(C, D) ∪ {τz ◦ ΦN : z ∈ T, N ∈ N},
which may be viewed as the one-point compactification of the space N ∪(cid:0)SN ∈N T × {N }(cid:1), with
ρ∞ corresponding to the point at infinity.
8. The Twist of a Regular Inclusion
Throughout this section, we fix, once and for all, a regular inclusion (C, D) and a closed N(C, D)-
invariant subset F ⊆ S(C, D) such that the restriction map, f ∈ F 7→ f D, is a surjection of F onto
D. When (C, D) is a regular MASA inclusion, Theorem 3.13 shows Ss(C, D) ⊆ F . For this reason,
we have in mind taking F = Ss(C, D), though other choices (e.g. F = S(C, D)) may be useful for
some purposes.
In this section, we show that associated to this data, there is a twist (Σ, G), which, when (C, D)
is a C ∗-diagonal (in which case F is necessarily Ss(C, D)) or a Cartan pair (with F = Ss(C, D))
gives the twist of the pair as defined by Kumjian [22] for C ∗-diagonals, or for Cartan pairs given
by Renault in [31]. The set F will be used as the unit space for the ´etale groupoid G associated to
the twist (Σ, G).
Our construction parallels the constructions by Kumjian and Renault, but with several differ-
ences. First, in the Renault and Kumjian contexts, a conditional expectation E : C → D is present,
and since {ρ ◦ E : ρ ∈ D} is homeomorphic to D, Renault and Kumjian use D as the unit space for
the twists they construct. In our context, we need not have a conditional expectation, so we use the
set F as a replacement for D. Next, as in the constructions of Kumjian and Renault, we construct a
49
regular ∗-homomorphism θ : (C, D) → (C ∗
r (Σ, G),
and the kernel of θ is not trivial unless the ideal KF = {x ∈ C : f (x∗x) = 0 for all f ∈ F } = (0).
We note however, that this ideal is trivial in the cases considered by Kumjian and Renault.
r (Σ, G), C(G(◦))), however, θ(C) need not equal C ∗
8.1. Twists and their C ∗-algebras. Before proceeding, it is helpful to recall some generalities
on twists and the (reduced) C ∗-algebras associated to them.
Definition 8.1. The pair (Σ, G) is a twist if Σ and G are Hausdorff locally compact topological
groupoids, G is an ´etale groupoid and the following hold:
(1) there is a free action of T by homeomorphisms of Σ such that whenever (σ1, σ2) ∈ Σ(2) and
z1, z2 ∈ T, we have (z1σ1, z2σ2) ∈ Σ(2) and (z1σ1)(z2σ2) = (z1z2)(σ1σ2);
(2) there is a continuous surjective groupoid homomorphism γ : Σ ։ G such that for every
σ ∈ Σ, γ−1(γ(σ)) = {zσ : z ∈ T};
(3) the bundle (Σ, G, γ) is locally trivial.
Notice that γΣ(◦) : Σ(◦) → G(◦) is a homeomorphism of the unit space of Σ onto the unit space
of G. We will usually use this map to identify Σ(◦) and G(◦).
Recall that given a twist Σ over the ´etale topological groupoid G, one can form the twisted
groupoid C ∗-algebra of the pair (Σ, G). We summarize the construction in our context, for details,
see [31, Section 4] and [22, Section 2]. We note that in both [31] and [22], there is a blanket
assumption that the ´etale groupoid G is second countable. However, for what we require here, this
hypothesis is not used. The reader may also wish to consult Section 3 of [12].
Let Cc(Σ, G) be the family of all compactly supported continuous complex valued functions f
on Σ which are equivariant, that is, which satisfy f (zσ) = zf (σ) for all σ ∈ Σ, z ∈ T. Given
f, g ∈ Cc(Σ, G), notice whenever τ, σ ∈ Σ with s(τ ) = s(σ) and z ∈ T, we have f (στ −1)g(τ ) =
f (σ(zτ )−1)g(zτ ). For x ∈ G and σ ∈ Σ with s(x) = s(σ), let τ ∈ γ−1(x). Then
does not depend on the choice of τ ∈ γ−1(x). The product of f with g is defined by
(f ⊛ g)(σ, x) := f (στ −1)g(τ )
(f ⋆ g)(σ) = Xx∈G
s(x)=s(σ)
(f ⊛ g)(σ, x),
and the adjoint operation is defined by
f ∗(σ) = f (σ−1).
These operations make Cc(Σ, G) into a ∗-algebra.
Similarly, notice that for f ∈ Cc(Σ, G), z ∈ T and σ ∈ Σ, f (σ) = f (zσ), so for x ∈ G, we
In particular, f may be viewed as a
denote by f (x) the number f (σ), where σ ∈ γ−1(x).
function on G.
One may norm Cc(Σ, G) as in [31] or [22]. For convenience, we provide a sketch of an equivalent,
but slightly different method. Given x ∈ G(◦), let ηx : Cc(Σ, G) → C by ηx(f ) = f (x). Then
ηx is a positive linear functional in the sense that for each f ∈ Cc(Σ, G), ηx(f ∗ ⋆ f ) ≥ 0. Let
Nx := {f ∈ Cc(Σ, G) : ηx(f ∗ ⋆ f ) = 0} and let Hx be the completion of Cc(Σ, G)/Nx with respect
to the inner product hf + Nx, g + Nxi = ηx(g∗ ⋆ f ).
Recall that a slice of G is an open set U ⊆ G so that rU and sU are one-to-one. By [12,
Proposition 3.10], given f ∈ Cc(Σ, G), there exist n ∈ N and slices U1, . . . , Un of G such that the
k=1 Uk. Let n(f ) ≥ 0 be the smallest integer such that there exist
support of f is contained inSn
slices U1, . . . Un(f ) with the support of f is contained inSn(f )
k=1 Uk.
50
For f, g ∈ Cc(Σ, G), a calculation shows that
k(f ⋆ g) + Nxk ≤ n(f ) kf k∞ kg + NxkHx .
Therefore, the map g + Nx 7→ (f ⋆ g) + Nx extends to a bounded linear operator πx(f ) on Hx. It
is easy to see that πx is a ∗-representation of Cc(Σ, G). Also, if πx(f ) = 0 for every x ∈ G(◦), then
kf + NxkHx = 0 for each x ∈ G(◦). A calculation then gives f = 0. Thus,
kf k := sup
x
kπx(f )k .
defines a norm on Cc(Σ, G). The (reduced) twisted C ∗-algebra, C ∗(Σ, G), is the completion of
Cc(Σ, G) relative to this norm. Clearly, the representation πx extend by continuity to a represen-
tation, again called πx, of C ∗(Σ, G).
As observed in the remarks following [31, Proposition 4.1], elements of C ∗(Σ, G) may be regarded
as equivariant continuous functions on Σ, and the formulas defining the product and involution on
Cc(Σ, G) remain valid for elements of C ∗(Σ, G). Also, as in [31, Proposition 4.1], for σ ∈ Σ, and
f ∈ C ∗(Σ, G),
f (σ) ≤ kf k .
Definition 8.2. We shall call the smallest topology on C ∗(Σ, G) such that for every σ ∈ Σ, the
point evaluation functional, C ∗(Σ, G) ∋ f 7→ f (σ) is continuous, the G(◦)-compatible topology on
C ∗(Σ, G). Clearly this topology is Hausdorff.
The open support of f ∈ C ∗(Σ, G) is supp(f ) = {σ ∈ Σ : f (σ) 6= 0}. Then C0(G(◦)) may be
identified with
{f ∈ C ∗(Σ, G) : supp(f ) ⊆ G(◦)}.
In order to remain within the unital context, we now assume that the unit space of G is compact.
In this case C ∗(Σ, G) is unital, and C(G(◦)) ⊆ C ∗(Σ, G), so that (C ∗(Σ, G), C(G(◦))) is an inclusion.
We wish to show that it is a regular inclusion.
Recall (see [12, Section 3]) that a slice (or G-set) of G is an open subset S ⊆ G such that
the restrictions of the range and source maps to S are one-to-one. We will say that an element
f ∈ C ∗(Σ, G) is supported in the slice S if γ(supp(f )) ⊆ S.
If f ∈ C ∗(Σ, G) is supported in a slice U , then a computation (see [31, Proposition 4.8]) shows
that f ∈ N(C ∗(Σ, G), C(G(◦))), and because the collection of slices forms a basis for the topology of
G ([12, Proposition 3.5]), it follows (as in [31, Corollary 4.9]) that (C ∗(Σ, G), C(G(◦))) is a regular
inclusion.
Proposition 8.3. Let Σ be a twist over the Hausdorff ´etale groupoid G. Assume that the unit
space X of G is compact. Then there is a faithful conditional expectation E : C ∗(Σ, G) →
C(G(◦)), the inclusion (C ∗(Σ, G), C(G(◦))) is regular. If in addition, is C(G(◦)) is a MASA, then
Rad(C ∗(Σ, G), C(G(◦))) = (0).
Remark 8.4. The condition that C(G(◦)) is a MASA is satisfied when G(◦) is second countable
and G is essentially principal, that is, when the interior of the isotropy bundle for G is G(◦), see [31,
Proposition 4.2]. We expect that it is possible to remove the hypothesis of second countability here,
but we have not verified this.
Proof of Proposition 8.3. The existence of the conditional expectation is proved as in [31, Propo-
sition 4.3] or [30, Proposition II.4.8], and we have already observed that the inclusion is regular.
When C(G(◦)) is a MASA in C ∗(Σ, G), the triviality of the radical follows from Proposition 5.7.
(cid:3)
51
8.2. Compatible Eigenfunctionals and the Twist for (C, D). We turn next to a discussion
of eigenfunctionals, for a certain class of eigenfunctionals will yield our twist. Recall (see [10])
that an eigenfunctional is a non-zero element φ ∈ C# which is an eigenvector for both the left and
right actions of D on C#; when this occurs, there exist unique elements ρ, σ ∈ D so that whenever
d1, d2 ∈ D and x ∈ C, we have φ(d1xd2) = ρ(d1)φ(x)σ(d2). We write
s(φ) := σ and r(φ) := ρ.
Definition 8.5. A compatible eigenfunctional
N(C, D),
is a eigenfunctional φ such that for every v ∈
(34)
φ(v)2 ∈ {0, s(φ)(v∗v)}.
Let Ec(C, D) denote the set consisting of the zero functional together with the set of all compatible
eigenfunctionals, and let E1
c (C, D) be the set of compatible eigenfunctionals which have unit norm.
c(C, D) with the relative σ(C#, C) topology.
Equip both Ec(C, D) and E1
Remark. Notice that when φ is an eigenfunctional and v ∈ N(C, D) is such that φ(v) 6= 0, then
for every d ∈ D,
(35)
s(φ)(v∗dv)
s(φ)(v∗v)
= r(φ)(d).
Indeed, φ(v)s(φ)(v∗dv) = φ(vv∗dv) = φ(dvv∗v) = r(φ)(d)φ(v)s(φ)(v∗v). Thus, taking d = 1, the
condition in (34) is equivalent to
(36)
φ(v)2 ∈ {0, r(φ)(vv∗)}.
We now show that associated with each φ ∈ E1
c (C, D) is a pair f, g ∈ S(C, D) which extend r(φ)
and s(φ). Note that regularity of the inclusion (C, D) ensures the existence of v ∈ N(C, D) such
that φ(v) > 0.
Proposition 8.6. Let φ ∈ E1
f, g ∈ C# by
c (C, D), and let v ∈ N(C, D) satisfy φ(v) > 0. Define elements
f (x) =
φ(xv)
φ(v)
and
g(x) =
φ(vx)
φ(v)
.
Then the following statements hold.
i) f, g ∈ S(C, D), r(φ) = f D and s(φ) = gD.
ii) For every x ∈ C,
φ(x) =
g(v∗x)
g(v∗v)1/2 =
f (xv∗)
f (vv∗)1/2 .
iii) For every x ∈ C, g(v∗xv) = g(v∗v)f (x) and f (vxv∗) = f (vv∗)g(x).
Proof. The definitions show f D = r(φ) and gD = s(φ). We next claim that kf k = kgk = 1. For
any d ∈ D with s(φ)(d) = 1, replacing v by vd in the definition of f does not change f . Thus, if
x ∈ C and kxk ≤ 1, we have f (x) ≤ inf{ kvdk
φ(v) : d ∈ D, s(φ)(d) = 1} = 1 (because d may be chosen
so that kvdk = kd∗v∗vdk1/2 is as close to s(φ)(v∗v)1/2 as desired). This shows kf k = 1. Likewise
kgk = 1. As f (1) = g(1) = 1, both f and g are states on C.
If w ∈ N(C, D) and f (w) 6= 0, we have (using (35))
2
=
s(φ)(v∗w∗wv)
s(φ)(v∗v)
= r(φ)(w∗w) = f (w∗w),
and it follows that f ∈ S(C, D). Likewise, g ∈ S(C, D).
52
f (w)2 =(cid:12)(cid:12)(cid:12)(cid:12)
φ(wv)2
φ(v) (cid:12)(cid:12)(cid:12)(cid:12)
Statements (ii) and (iii) are calculations using (34) and (36) whose verification is left to the
(cid:3)
reader.
Notation. For v ∈ N(C, D) and f ∈ S(C, D) such that f (v∗v) > 0, let [v, f ] ∈ C# be defined by
[v, f ](x) :=
f (v∗x)
f (v∗v)1/2
=*x + Lf ,
v + Lf
kv + Lf kHf+Hf
.
(This notation is borrowed from Kumjian [22]. There, Kumjian works in the context of C ∗-diagonals
and uses states on C of the form σ ◦ E with σ ∈ D. As we assume no conditional expectation here,
we replace functionals of the form σ ◦ E, with elements from S(C, D). See also [31].)
We have the following.
Lemma 8.7. If v ∈ N(C, D) and f ∈ S(C, D) with f (v∗v) > 0, then [v, f ] ∈ E1
following statements hold.
c (C, D) and the
i) s([v, f ]) = f D and r([v, f ]) = βv(f D).
ii) [v, f ] = [w, g] if and only if f = g and f (v∗w) > 0.
Proof. Suppose that f ∈ S(C, D), v ∈ N(C, D) and f (v∗v) 6= 0. Let φ = [v, f ]. A calculation shows
that φ is a norm-one eigenfunctional and that statement (i) holds.
If w ∈ N(C, D) and φ(w) 6= 0, then
φ(w)2 =
f (v∗w)2
f (v∗v)
=
f (v∗ww∗v)
f (v∗v)
= βv(s(φ))(ww∗) = r(φ)(ww∗),
so φ belongs to E1
c (C, D) by (36).
Turning now to part (ii), suppose that φ = [v, f ] = [w, g]. Then we have s(φ) = f D = gD. For
every x ∈ C, Proposition 8.6 gives
f (x) =
φ(vx)
φ(v)
and g(x) =
φ(wx)
φ(w)
.
Since
g(w∗v)
g(w∗w)1/2 = φ(v) = f (v∗v)1/2, we obtain
g(w∗v) = f (v∗v)1/2g(w∗w)1/2 > 0.
Likewise, f (v∗w) > 0. Also,
f (x) =
φ(vx)
φ(v)
=
[w, g](vx)
[v, f ](v)
=
g(w∗vx)
f (v∗v)1/2g(w∗w)1/2
=
g(w∗v)g(x)
g(w∗v)
= g(x),
where the fourth equality follows from Proposition 4.3.
Conversely, if f ∈ S(C, D) and v, w ∈ N(C, D) with f (v∗w) > 0, Proposition 4.3 shows that
f (v∗w)2 = f (w∗w)f (v∗v), so that in the GNS Hilbert space Hf , we have hv + Lf , w + Lf i =
kv + Lf k kw + Lf k. By the Cauchy-Schwartz inequality, there exists a positive real number t so
that v + Lf = tw + Lf . But then for any x ∈ C,
[v, f ](x) =
hx + Lf , v + Lf i
kv + Lf k
=
hx + Lf , tw + Lf i
ktw + Lf k
= [w, f ](x).
(cid:3)
Combining Proposition 8.6 and Lemma 8.7 we obtain the following.
53
Theorem 8.8. If φ ∈ E1
whenever v ∈ N(C, D) satisfies φ(v) 6= 0 and x ∈ C,
c(C, D), then there exist unique elements s(φ), r(φ) ∈ S(C, D) such that
φ(vx) = φ(v) s(φ)(x)
and φ(xv) = r(φ)(x) φ(v).
If v ∈ N(C, D) satisfies φ(v) > 0, then φ = [v, s(φ)]. Moreover,
E1
c(C, D) = {[v, f ] : v ∈ N(C, D), f ∈ S(C, D) and f (v∗v) 6= 0}.
Proof. Suppose v, w ∈ N(C, D) are such that φ(v) > 0 and φ(w) > 0. For x ∈ C, set
f (x) :=
φ(vx)
φ(v)
and g(x) :=
φ(wx)
φ(w)
.
Proposition 8.6 shows that φ = [v, f ] = [w, g]. Lemma 8.7 yields f = g. Another application of
Proposition 8.6 shows that for any x ∈ C,
φ(xv)
φ(v)
=
φ(xw)
φ(w)
.
Then taking s(φ) = f , and r(φ) = φ(xv)
φ(v) , we obtain the result.
(cid:3)
Notice that for φ ∈ E1
Definition 8.9. Let E1
F -compatible eigenfunctional. Notice that
c (C, D), we have s(φ) ∈ F if and only if r(φ) ∈ F .
F (C, D) := {φ ∈ E1
c (C, D) : s(φ) ∈ F }. We shall call φ ∈ E1
F (C, D) an
E1
F (C, D) = {[v, f ] : f ∈ F and f (v∗v) 6= 0}.
With these preparations in hand, we can show that E1
F (C, D) forms a topological groupoid. The
topology has already been defined, so we need to define the source and range maps, composition
and inverses.
Definition 8.10. Given φ ∈ E1
following definitions.
F (C, D), let v ∈ N(C, D) be such that φ(v) > 0. We make the
(1) We say that s(φ) and r(φ) are the source and range of φ respectively.
(2) Define the inverse, φ−1 by the formula,
φ−1(x) := φ(x∗).
If φ ∈ E1
c (C, D) and v ∈ N(C, D) is such that φ(v) > 0, (so that φ = [v, s(φ)]), then a
calculation shows that φ−1 = [v∗, r(φ)]. The fact that F is N(C, D)-invariant ensures that
φ−1 ∈ E1
F (C, D). Thus, our definition of φ−1 is consistent with the definition of inverse in
the definition of the twist of a C ∗-diagonal arising in [22] and the twist of a Cartan MASA
from [31].
(3) For i = 1, 2, let φi ∈ E1
F (C, D). We say that the pair (φ1, φ2) is a composable pair if
s(φ2) = r(φ1). As is customary, we write E1
F (C, D)(2) for the set of composable pairs. To
define the composition, choose vi ∈ N(C, D) with φi(vi) > 0, so that φi = [vi, s(φi)]. By
Proposition 8.6(iii), we have
s(φ2)(v∗
2v∗
1v1v2) = r(φ2)(v∗
1v1)s(φ2)(v∗
2v2) = s(φ1)(v∗
1v1)s(φ2)(v∗
2v2) > 0,
so that [v1v2, s(φ2)] is defined. The product is then defined to be φ1φ2 := [v1v2, s(φ2)].
We show now that this product is well defined. Suppose that (φ1, φ2) ∈ E1
F (C, D)(2),
f = s(φ2), r(φ2) = g = s(φ1), and that for i = 1, 2, vi, wi ∈ N(C, D) are such that
φ1 = [v1, g] = [w1, g] and [v2, f ] = [w2, f ]. Then using Lemma 8.7, we have g(w∗
1v1) > 0 and
54
2w2) > 0, so, as f ∈ S(C, D), there exists a positive scalar t such that v2+Lf = tw2+Lf .
f (v∗
Hence,
f ((w1w2)∗(v1v2)) = hπf (v1)(v2 + Lf ), πf (w1)(w2 + Lf )i
1v1)w2)
= t hπf (v1)(w2 + Lf ), πf (w1)(w2 + Lf )i
= tf (w∗
= tf (w∗
= tf (w∗
= tf (w∗
2(w∗
2w2)r(φ2)(w∗
2w2)s(φ1)(w∗
2w2)g(w∗
1v1)
1v1)
1v1) > 0.
By Lemma 8.7, [v1v2, f ] = [w1w2, f ], so that the product is well defined.
(4) For φ ∈ E1
F (C, D), denote the map C ∋ x 7→ φ(x) by φ. Observe that for φ, ψ ∈ E1
F (C, D),
φ = ψ if and only if there exists z ∈ T such that φ = zψ. Let RF (C, D) := {φ : φ ∈
E1
F (C, D)}. We now define source and range maps, along with inverse and product maps on
RF (C, D).
Since a state on C is determined by its values on the positive elements of C, we identify
f ∈ S(C, D) with f ∈ RF (C, D). Define s(φ) = s(φ) and r(φ) = r(φ). Next we
define inversion in RF (C, D) by φ−1 = φ−1, and composable pairs by RF (C, D)(2) :=
{(φ, ψ) : (φ, ψ) ∈ E1
c (C, D)(2)}, and the product by RF (C, D)(2) ∋ (φ, ψ) 7→ φψ.
Topologize RF (C, D) with the topology of point-wise convergence: φλ → φ if and only if
φλ(x) → φ(x) for every x ∈ C. We call RF (C, D) the spectral groupoid over F of (C, D).
F (C, D) ∋ (z, φ) 7→ zφ, where (zφ)(x) = φ(zx).
F (C, D) by T × E1
Notice that if φ is written as φ = [v, f ], where v ∈ N(C, D) and f ∈ F , then zφ = [zv, f ].
(5) Define an action of T on E1
We have the following fact, whose proof is essentially the same as that of [10, Proposition 2.3]
(the continuity of the range and source maps follows from their definition).
Proposition 8.11. The set E1
s, r : E1
F (C, D) → S(C, D) are weak-∗ -- weak-∗ continuous.
F (C, D) ∪ {0} is a weak-∗ compact subset of C#, and the maps
F (C, D) and RF (C, D) be as above. Then E1
Theorem 8.12. Let E1
F (C, D) and RF (C, D) are locally
compact Hausdorff topological groupoids and RF (C, D) is an ´etale groupoid. Their unit spaces
are E1
F (C, D) is a locally trivial topological twist over
RF (C, D).
F (C, D)(◦) = RF (C, D)(◦) = F . Moreover, E1
Proof. That inversion on E1
F (C, D) is continuous follows readily from the definition of inverse map
and the weak-∗ topology. Suppose (φλ)λ∈Λ and (ψλ)λ∈Λ are nets in E1
F (C, D) converging to φ, ψ ∈
F (C, D)(2) for all λ. Since the source and range
E1
F (C, D) respectively, and such that (φλ, ψλ) ∈ E1
maps are continuous, we find that s(φ) = limλ s(φλ) = limλ r(ψλ) = r(ψ), so (φ, ψ) ∈ E1
F (C, D)(2).
Let v, w ∈ N(C, D) be such that φ(v) > 0 and ψ(w) > 0. There exists λ0, so that λ ≥ λ0
implies φλ(v) and ψλ(w) are non-zero. For each λ ≥ λ0, there exists scalars ξλ, ηλ ∈ T such that
φλ(v) = ξλ[v, s(φλ)] and ψλ = ηλ[v, s(ψλ)]. Since
lim
λ
φλ(v) = φ(v) = lim
λ
[v, s(φλ)](v)
and
lim
λ
ψλ(v) = ψ(v) = lim
λ
[v, s(ψλ)](v),
we conclude that lim ηλ = 1 = lim ξλ. So for any x ∈ C,
(φψ)(x) =
s(ψ)((vw)∗x)
(s(ψ)((vw)∗(vw)))1/2 = lim
λ
s(ψλ)((vw)∗x)
(s(ψλ)((vw)∗(vw)))1/2 = lim
λ
[v, s(φλ)][w, s(ψλ)]
= lim
λ
(φλψλ)(x),
55
giving continuity of multiplication. Notice that for φ ∈ E1
and F ⊆ E1
space F .
F (C, D), s(φ) = φ−1φ and r(φ) = φφ−1,
F (C, D) is a locally compact Hausdorff topological groupoid with unit
F (C, D). Thus, E1
The definitions show that RF (C, D) is a groupoid. By construction, the map q defined by φ 7→ φ
is continuous and is a surjective groupoid homomorphism. The topology on RF (C, D) is clearly
Hausdorff. If φ ∈ E1
F (C, D), and v ∈ N(C, D) is such that φ(v) 6= 0, then W := {α ∈ RF (C, D) :
α(v) > φ(v)/2} has compact closure so RF (C, D) is locally compact. Also, if α1, α2 ∈ W and
r(α1) = r(α2) = f , then writing αi = ψi for ψi ∈ E1
F (C, D), we see that ψi(v) 6= 0, so there exist
z1, z2 ∈ T so that for i = 1, 2 and every x ∈ C, ψi(x) = zif (xv∗)f (v)−1. Hence α1 = α2 showing
that the range map is locally injective. We already know that the range map is continuous, so by
local compactness, the range map is a local homeomorphism.
Note that convergent nets in RF (C, D) can be lifted to convergent nets in E1
F (C, D). Indeed, if
q(φλ) → q(φ) for some net (φλ) and φ in E1
F (C, D), choose v ∈ N(C, D) so that φ(v) > 0. Then for
large enough λ, φλ(v) 6= 0, and we have [φλ, s(φλ)] → φ. Also, [φλ, s(φλ)] → φ. The fact that
the groupoid operations on RF (C, D) are continuous now follows easily from the continuity of the
groupoid operations on E1
F (C, D). Thus RF (C, D) is a locally compact Hausdorff ´etale groupoid.
Finally, q(φ1) = q(φ2) if and only if there exist z ∈ T so that φ1 = zφ2. Moreover, for each
v ∈ N(C, D), the map f 7→ [v, f ], where f ∈ {g ∈ S(C, D) : g(v∗v) > 0} is a continuous section for
q. Also, the action of T on E1
F (C, D)
is a twist over RF (C, D).
F (C, D) into a T-groupoid. So E1
F (C, D). given above makes E1
Notation 8.13. We now let
Σ = E1
F (C, D) and G = RF (C, D),
so that G(◦) = F.
(cid:3)
For a ∈ C, define a : E1
a is a continuous equivariant function on E1
F (C, D) → C to be the 'Gelfand' map: for φ ∈ E1
F (C, D).
F (C, D), a(φ) = φ(a). Then
Note that if w ∈ N(C, D), then w is compactly supported. Indeed, for φ = [v, f ] ∈ E1
F (C, D),
φ ∈ supp( w) if and only if [v, f ](w) 6= 0, which occurs exactly when f (v∗w) 6= 0. Proposition 4.3
shows this occurs precisely when f (w∗w) 6= 0. Hence,
supp w = {φ ∈ E1
F (C, D) : s(φ)(w∗w) 6= 0},
and it follows that w has compact support. Moreover, w is supported on a slice, so that we find
w ∈ N(C ∗(Σ, G), C(G(◦))).
Before stating the main result of this section, recall that Proposition 2.16 shows that KF = {x ∈
C : f (x∗x) = 0 for all f ∈ F } is an ideal of C whose intersection with D is trivial.
Theorem 8.14. Let (C, D) be a regular inclusion, and let G := RF (C, D) and Σ := E1
F (C, D).
The map sending w ∈ N(C, D) to w ∈ C ∗(Σ, G) extends uniquely to a regular ∗-homomorphism
θ : (C, D) → (C ∗(Σ, G), C(G(◦))) with ker θ = KF . Furthermore, θ(C) is dense in C ∗(Σ, G) in the
G(◦)-compatible topology.
Remark. In general, θ(C) and θ(D) may be proper subsets of C ∗(Σ, G) and C(G(◦)) respectively.
Proof. The point is that the norms on C/KF and C ∗(Σ, G) both arise from the left regular repre-
sentation on appropriate spaces. Here are the details.
We have already observed that the map w 7→ w sends normalizers to normalizers. Let C0 =
span N(C, D). Then for any a ∈ C0, a ∈ Cc(Σ, G).
56
Let f ∈ F = G(◦). Then f can be regarded as either a state on C or as determining a state on
C ∗(Σ, G) via evaluation at f . We write fC when viewing f as a state on C, and fΣ when viewing
f as a state on C ∗(Σ, G).
Let (πC,f , HC,f ) be the GNS representation of C arising from fC, and let (πΣ,f , HΣ,f ) be the GNS
representation of C ∗(Σ, G) determined by fΣ. (Writing x = f , the restriction of πΣ,f to Cc(Σ, G)
is the representation πx discussed above when defining the norm on Cc(Σ, G).)
For typographical reasons, when the particular f is understood, we will drop the extra f in the
notation: write πC or πΣ instead of πC,f or πΣ,f .
Now fix f ∈ G(◦). For a ∈ C0, we claim that
ka + Nf kHΣ
(37)
= ka + Lf kHC
.
Let T ⊆ Λf be chosen so that f (w∗w) = 1 for every w ∈ T and so that T contains exactly one
element from each ∼f equivalence class. Proposition 8.7 shows that when w1, w2 ∈ Λf , we have
[w1, f ] = [w2, f ] if and only if w1 ∼f w2. Proposition 4.10 shows that {w + Lf : w ∈ T } is an
orthonormal basis for HC. Writing φ = [w, f ], we have,
ka + Nf k2
HΣ
a(φ)2 = Xw∈T
[w, f ](a)2 = Xw∈T
f (w∗a)2
ha + Lf , w + Lf i 2 = ka + Lf k2
.
HC
= Xφ∈G
= Xw∈T
s(φ)=f
It follows that the map a + Lf 7→ a + Nf extends to an isometry Wf : HC → HΣ.
To see that Wf is a unitary operator, fix ξ ∈ Cc(Σ, G). Since ξ is compactly supported, the set
Sξ := {φ ∈ G : s(φ) = f and ξ(φ) 6= 0}
is a finite set. Let φ1, . . . , φn be the elements of Sξ. For 1 ≤ j ≤ n, we may find vj ∈ N(C, D)
such that φj = [vj, f ]. By Proposition 8.7, we may assume each vj belongs to the set T . Let
j=1 zjvj. Clearly a ∈ C0. Using the fact that f (w∗w) = 1 for each w ∈ T
1w2) = 0 for distinct elements w1, w2 ∈ T , we find that for 1 ≤ k ≤ n,
zj = ξ(φj) and set a =Pn
and the fact that f (w∗
[vk, f ](a) = zk.
Then
k(a − ξ) + Nf k2
HΣ
= fΣ ((a − ξ)∗ ⋆ (a − ξ)) =
(a − ξ)(φj )2
nXj=1
(a(φj) − ξ(φj))(a(φj) − ξ(φj)) =
([vj, f ](a) − ξ(φj))2
nXj=1
(zj − ξ(φj))2 = 0.
=
=
nXj=1
nXj=1
Therefore, {a + Nf : a ∈ C0} = {ξ + Nf : ξ ∈ Cc(Σ, G)}. As this set is dense in HΣ, Wf is a unitary
operator.
Next, we show that for a ∈ C0, we have
(38)
πΣ(a)Wf = Wf πC(a).
To do this, it suffices to show that for each v ∈ T ,
πΣ(a)Wf (v + Lf ) = Wf πC(a)(v + Lf ).
57
Letting φ = [w1, f ] and y = [w2, f ] we find φy−1 = [w1w2, βw2(f )], and a computation yields,
a(φy−1)v(y) =
f (w∗
1aw2)f (w∗
f (w1w1)1/2f (w∗
Therefore, when s(φ) = f , we have
0
2v)
2w2)
=
(a ⋆ v)(φ) = Xy∈G
s(y)=f
f (w∗
1av)
1w1)1/2
f (w∗
if w2 6∼f v
if w2 ∼f v
=(0
cav(φ)
if w2 6∼f v
if w2 ∼f v.
(a ⊛ v)(φ, y) =cav(φ).
So for y ∈ G with s(y) = f , (a ⋆ v −cav)(y) = 0. Then,
a ⋆ v + Nf =cav + Nf
because
k((a ⋆ v) −cav) + Nf k2
Hσ
= Xy∈G
s(y)=f
(a ⋆ v −cav)(y)2 = 0.
Hence,
which gives (38).
πΣ(a)Wf (v + Lf ) = πΣ(a)(v + Nf ) =cav + Nf = Wf (πC(a))(v + Lf ),
The definition of the norm on C ∗(Σ, G) and (38) imply that for a, b ∈ C0,(cid:13)(cid:13)(cid:13)bab − ab(cid:13)(cid:13)(cid:13)C ∗(Σ,G)
Therefore, the map a ∈ C0 7→ a is multiplicative and
kakC ∗(Σ,G) = sup
= 0.
The existence of θ now follows from continuity, and the fact that ker θ = KF follows from Propo-
sition 5.4. For v ∈ N(C, D), θ(v) = v belongs to N(C ∗(Σ, G), C(G(◦))), so θ is a regular ∗-
homomorphism.
Finally, we turn to showing the G(◦)-compatible density of θ(C) in C ∗(Σ, G). Let M ⊆ C ∗(Σ, G)#
be the linear span of the evaluation functionals ξ 7→ ξ(σ) where ξ ∈ C ∗(Σ, G) and σ ∈ Σ. Suppose
µ ∈ M annihilates θ(C). Then there exists n ∈ N, scalars λ1, . . . , λn, elements v1, . . . , vn ∈ N(C, D)
and f1, . . . , fn ∈ F such that for any ξ ∈ C ∗(Σ, G),
f ∈S(C,D)(cid:13)(cid:13)πC,f (a)(cid:13)(cid:13) .
µ(ξ) =
λkξ([vk, fk]).
0 = µ(a) =
λk[vk, fk](a).
nXk=1
nXk=1
Without loss of generality we may assume that [vi, fi] 6= [vj, fj] if i 6= j. Since µ annilhlates θ(C),
for every a ∈ C0,
Fix 1 ≤ j ≤ n, and let d, e ∈ D be such that βvj (fj)(d) = fj(e) = 1. For i 6= j, since [vi, fi] 6= [vj, fj],
either fj 6= fi or βvi(fi) 6= βvj (fj). Hence we assume that d and e have been chosen so that if i 6= j,
then [vi, fi](dvje) = 0. Then
µ( vj) = λjfj(v∗
j vj)1/2 = 0.
As fj(v∗
j vj) 6= 0, we obtain λj = 0. It follows that µ = 0. Since the dual of C ∗(Σ, G) equipped with
the G(◦)-compatible topology is M, we conclude that θ(C) is dense in the G(◦)-compatible topology
on C ∗(Σ, G). This completes the proof.
58
(cid:3)
9. Applications
In this section we give some applications which apply to regular MASA inclusions with L(C, D) =
(0). Theorem 6.10 gives a very large class of such inclusions.
Here is an application of our work to norming algebras. We begin with a definition. Recall that
N(C, D) is a closed ∗-semigroup containing D.
Definition 9.1. A ∗-subsemigroup F ⊆ N(C, D) with D ⊆ F is countably generated over D if there
exists a countable set F ⊆ F so that the smallest ∗-subsemigroup of N(C, D) containing F ∪ D is
F. The set F will be called a generating set for F.
We will say that the inclusion (C, D) is countably regular if there exists a ∗-subsemigroup F ⊆
N(C, D) such that F is countably generated over D and C = span(F).
The following result generalizes [28, Lemma 2.15] and gives a large class of norming algebras. In
particular, notice that the result holds for Cartan inclusions.
Theorem 9.2. Suppose (C, D) is a regular MASA inclusion such that L(C, D) = (0). Then D
norms C.
Proof. Let F ⊆ N(C, D) be a ∗-subsemigroup which is countably generated over D by the (count-
able) set F . Let CF ⊆ C be the C ∗-subalgebra generated by F. (Notice that CF is simply the closed
linear span of F.) Then (CF, D) is a countably regular MASA inclusion.
We will show that D norms CF. Let
Y := {σ ∈ D : σ has a unique state extention to CF}.
Theorem 3.8 shows that Y is dense in D. For each element σ ∈ Y , let σ′ denote the unique extension
of σ to all of CF. Notice that if ρ ∈ [I(D) and ρ ◦ ι = σ, then σ′ = ρ ◦ E because σ′D = σ = ρ ◦ ED.
For σ ∈ Y , let πσ be the GNS representation for σ′. Proposition 4.14 shows that πσ(D)′′ is a
MASA in B(Hσ).
Define an equivalence relation R on Y by σ1 ∼ σ2 if and only if there exists v ∈ F such that
σ2 = βv(σ1). (Since F is a ∗-semigroup, this is an equivalence relation.)
We claim that if πσ1 is unitarily equivalent to πσ2, then σ1 ∼ σ2. To see this, we use a modification
of the argument in [10, Lemma 5.8]. Let U ∈ B(Hσ2 , Hσ1 ) be a unitary operator such that
U ∗πσ1U = πσ2.
Let Lσi be the left kernel of σ′
such that U (I + Lσ2) = X + Lσ1. Then for every x ∈ C,
i. Since πσi is irreducible, C/Lσi = Hσi . Hence we may find X ∈ C
σ′
2(x) = hπσ2(x)(I + Lσ2), (I + Lσ2)i = σ′
1(X ∗xX).
Fix ρi ∈ [I(D) such that ρi ◦ ι = σi.
The map C ∋ x 7→ σ′
dense in C, there exists v ∈ F so that σ′
positive maps shows that for any d ∈ D,
1(X ∗x) is a non-zero linear bounded linear functional on C. Since span(F) is
1(X ∗v) 6= 0. The Cauchy-Schwartz inequality for completely
σ′
1(X ∗vd)2 = ρ1(E(X ∗vd)E(d∗v∗X)) ≤ ρ1(E(X ∗vdd∗v∗X)) = σ′
1(X ∗vdd∗v∗X) = σ2(vdd∗v∗).
When d ∈ D and σ1(d) 6= 0, we have σ′
satisfies σ1(d) 6= 0,
(39)
In particular, σ2(vv∗) 6= 0. For any d ∈ D, we have
0 < σ2(vdd∗v∗).
1(X ∗vd) = σ′
1(X ∗v)σ1(d) 6= 0. Therefore, when d ∈ D
βv∗ (σ2)(d) =
59
σ2(vdv∗)
σ2(vv∗)
.
If βv∗ (σ2) 6= σ1, then there exists d ∈ D with σ1(dd∗) 6= 0 and βv∗ (dd∗) = 0. But this is impossible
by (39). So βv∗ (σ2) = σ1. Hence σ1 ∼ σ2 as claimed.
Thus, if σ1 6∼ σ2, then πσ1 and πσ2 are disjoint representations (as they are both irreducible).
Let Y ⊆ Y be chosen so that Y contains exactly one element from each equivalence class of Y .
Put
Then
(40)
ker π = \σ∈Y
ker πσ = \σ∈Y
π =Mσ∈Y
πσ.
ker πσ = {x ∈ CF : σ′(z∗x∗xz) = 0 for all σ ∈ Y and all z ∈ CF}.
We next prove that
(41)
L(CF, D) ⊇ ker π.
Suppose to obtain a contradiction, that x ∈ ker π and that E(x∗x) is a non-zero element of I(D).
Let
L := {ρ ∈ [I(D) : ρ(E(x∗x)) > kE(x∗x)k /2}.
Then L is a clopen set. By Lemma 3.15, ι∗(L) = {σ ∈ D : σ = ρ ◦ ι for some ρ ∈ L} has non-
empty interior. Since Y is dense in D, we may find σ ∈ Y and ρ ∈ L such that ρ ◦ ι = σ. Then
σ′(x∗x) = ρ(E(x∗x)) 6= 0, so x /∈ ker πρ, contradicting (40). Hence (41) holds.
Since L(C, D) ⊇ L(CF, D) ⊇ ker π, we see that π is a faithful representation of CF.
Since the representations in the definition of π are disjoint and each πσ(D)′′ is a MASA in B(Hσ),
π(D)′′ is an atomic MASA in B(Hπ). Therefore, π(D)′′ is locally cyclic (see [29, p. 173]) for B(Hπ).
By [29, Theorem 2.7 and Lemma 2.3] π(D) norms B(Hπ). But then π(D) norms π(CF). Since π is
faithful, D norms CF.
Finally, suppose that k ∈ N and that x = (xij) ∈ Mn(C). For each n ∈ N and i, j ∈ {1, . . . , k},
we may find a finite set Fn,i,j ⊆ N(C, D) so that(cid:13)(cid:13)(cid:13)xij −Pv∈Fn,i,j
F = ∪{Fn,i,j : n ∈ N, i, j ∈ {1, . . . , k}}.
v(cid:13)(cid:13)(cid:13) < 1/n. Let
Then F is countable. Let F be the closed ∗-subsemigroup of N(C, D) generated by F and D. Then
for i, j ∈ {1, . . . , k}, xij ∈ CF. Since D norms CF, we conclude that
kxkMk(C) = kxkMk(CF) = sup{kRxCk : R ∈ M1,n(D), C ∈ Mn,1(D), kRk ≤ 1, kCk ≤ 1}.
Hence D norms C.
(cid:3)
For any norm closed subalgebra A of the C ∗-algebra C, let C ∗(A) be the C ∗-subalgebra of C
e (A) be the C ∗-envelope of A. (There are a number of references which
generated by A, and let C ∗
discuss C ∗-envelopes; see [5, 11, 27].)
The following result is a significant generalization of [10, Theorem 4.21]. Theorem 9.3 was
observed by Vrej Zarikian, who has kindly consented to its inclusion here.
Theorem 9.3. Let C and D be C ∗-algebras, with D ⊆ C (D is not assumed abelian). Let (I(D), ι)
be an injective envelope for D. Suppose there exists a unique unital completely positive map Φ : C →
I(D) such that ΦD = ι, and assume also that Φ is faithful. Let A be a norm-closed (not necessarily
self-adjoint) subalgebra of C such that D ⊆ A ⊆ C. Then the C ∗-subalgebra of C generated by A is
the C ∗-envelope of A.
60
Proof. Let θ : A → C ∗
the C ∗-algebra generated by the image of θ is C ∗
q : C ∗(A) → C ∗
e (A) be a unital completely isometric (unital) homomorphism such that
e (A). Then there exists a unique ∗-epimorphism
e (A) such that qA = θ. Our task is to show that q is one-to-one.
Since I(D) is injective in the category of operator systems and completely contractive maps,
there exists a unital completely contractive map Φe : C ∗
e (A) → I(D) such that Φe ◦ θ = ι. Also,
there exists a unital completely contractive map ∆ : C → I(D) so that ∆C ∗(A) = Φe ◦ q. Then for
d ∈ D, we have θ(d) = q(d), so ι(d) = Φe(θ(d)) = Φe(q(d)) = ∆(d). The uniqueness of Φ gives
∆ = Φ. Then if x ∈ C ∗(A) and q(x) = 0, we have Φ(q(x∗x)) = 0, so q(x∗x) = 0 by the faithfulness
of Φ. Thus q is one-to-one, and the proof is complete.
(cid:3)
We now obtain the following generalization of [28, Theorem 2.16]. While the outline of the proof
is the same as the proof of [28, Theorem 2.16], the details in obtaining norming subalgebras are
different.
Theorem 9.4. For i = 1, 2, suppose that (Ci, Di) are regular MASA inclusions such that L(Ci, Di) =
(0) and that Ai ⊆ Ci are norm closed subalgebras such that Di ⊆ Ai ⊆ Ci. Let C ∗(Ai) be the C ∗-
subalgebra of Ci generated by Ai. If u : A1 → A2 is an isometric isomorphism, then u extends
uniquely to a ∗-isomorphism of C ∗(A1) onto C ∗(A2).
Proof. Theorem 9.2 implies that Di norms Ci. Taken together, Theorem 3.10 and Theorem 9.3
imply that C ∗(Ai) is the C ∗-envelope of Ai. Finally, an application of [28, Corollary 1.5] completes
the proof.
(cid:3)
References
1. Charles Akemann and David Sherman, Conditional expectations onto maximal abelian ∗-subalgebras,
arXiv:0906.1831v2 [math.OA].
2. Joel Anderson, Extensions, restrictions, and representations of states on C ∗-algebras, Trans. Amer. Math. Soc.
249 (1979), no. 2, 303 -- 329. MR 525675 (80k:46069)
3. R. J. Archbold, J. W. Bunce, and K. D. Gregson, Extensions of states of C ∗-algebras. II, Proc. Roy. Soc.
Edinburgh Sect. A 92 (1982), no. 1-2, 113 -- 122. MR 84a:46134
4. A. V. Arhangel′skii, On topological and algebraic structure of extremally disconnected semitopological groups,
Comment. Math. Univ. Carolin. 41 (2000), no. 4, 803 -- 810. MR 1800164 (2001k:54063)
5. David P. Blecher and Christian Le Merdy, Operator algebras and their modules -- an operator space approach,
London Mathematical Society Monographs. New Series, vol. 30, The Clarendon Press Oxford University Press,
Oxford, 2004, Oxford Science Publications. MR 2111973
6. Nathanial P. Brown and Narutaka Ozawa, C ∗-algebras and finite-dimensional approximations, Graduate Studies
in Mathematics, vol. 88, American Mathematical Society, Providence, RI, 2008. MR 2391387
7. K. R. Davidson and S. C. Power, Isometric automorphisms and homology for nonselfadjoint operator algebras,
Quart. J. Math. Oxford Ser. (2) 42 (1991), no. 167, 271 -- 292. MR 92h:47058
8. J. Dixmier, Sur certains espaces consid´er´es par M. H. Stone, Summa Brasil. Math. 2 (1951), 151 -- 182. MR 0048787
(14,69e)
9. A. P. Donsig, T. D. Hudson, and E. G. Katsoulis, Algebraic isomorphisms of limit algebras, Trans. Amer. Math.
Soc. 353 (2001), no. 3, 1169 -- 1182 (electronic). MR 1804417 (2001k:47103)
10. Allan P. Donsig and David R. Pitts, Coordinate systems and bounded isomorphisms, J. Operator Theory 59
(2008), no. 2, 359 -- 416.
11. Edward G. Effros and Zhong-Jin Ruan, Operator spaces, London Mathematical Society Monographs. New Series,
vol. 23, The Clarendon Press Oxford University Press, New York, 2000. MR 1793753 (2002a:46082)
12. Ruy Exel, Inverse semigroups and combinatorial C ∗-algebras, Bull. Braz. Math. Soc. (N.S.) 39 (2008), no. 2,
191 -- 313. MR 2419901 (2009b:46115)
13. Jacob Feldman and Calvin C. Moore, Ergodic equivalence relations, cohomology, and von Neumann algebras. II,
Trans. Amer. Math. Soc. 234 (1977), no. 2, 325 -- 359. MR 58 #28261b
14. Z. Frol´ık, Maps of extremally disconnected spaces, theory of types, and applications, General topology and its
relations to modern analysis and algebra, III (Proc. Conf., Kanpur, 1968), Academia, Prague, 1971, pp. 131 -- 142.
MR 0295305 (45 #4373)
15. Andrew M. Gleason, Projective topological spaces, Illinois J. Math. 2 (1958), 482 -- 489. MR 0121775 (22 #12509)
61
16. Harry Gonshor, Injective hulls of C ∗ algebras. II, Proc. Amer. Math. Soc. 24 (1970), 486 -- 491. MR 0287318 (44
#4525)
17. Don Hadwin and Vern I. Paulsen, Injectivity and projectivity in analysis and topology, Sci. China Math. 54 (2011),
no. 11, 2347 -- 2359.
18. Masamichi Hamana, Injective envelopes of C ∗-algebras, J. Math. Soc. Japan 31 (1979), no. 1, 181 -- 197.
MR 80g:46048
19.
20.
, Injective envelopes of operator systems, Publ. Res. Inst. Math. Sci. 15 (1979), no. 3, 773 -- 785. MR 566081
(81h:46071)
, Regular embeddings of C ∗-algebras in monotone complete C ∗-algebras, J. Math. Soc. Japan 33 (1981),
no. 1, 159 -- 183. MR 597486 (82i:46089)
21. Alan Hopenwasser, Justin R. Peters, and Stephen C. Power, Subalgebras of graph C ∗-algebras, New York J. Math.
11 (2005), 351 -- 386 (electronic). MR 2188247 (2007b:47189)
22. Alexander Kumjian, On C ∗-diagonals, Canad. J. Math. 38 (1986), no. 4, 969 -- 1008. MR 88a:46060
23. Paul S. Muhly, Chao Xin Qiu, and Baruch Solel, Coordinates, nuclearity and spectral subspaces in operator
algebras, J. Operator Theory 26 (1991), no. 2, 313 -- 332. MR 94i:46075
24. Paul S. Muhly and Baruch Solel, On triangular subalgebras of groupoid C ∗-algebras, Israel J. Math. 71 (1990),
no. 3, 257 -- 273. MR 92m:46089
25. P. S. Muhly, K. Saito and B. Solel, Coordinates for triangular operator algebras, Ann. Math. 127 (1988), 245 -- 278.
MR 0932297 (89h:46088)
26. William L. Paschke, Pure eigenstates for the sum of generators of the free group, Pacific J. Math. 197 (2001),
no. 1, 151 -- 171. MR 1810213 (2001m:46122)
27. Vern Paulsen, Completely bounded maps and operator algebras, Cambridge Studies in Advanced Mathematics,
vol. 78, Cambridge University Press, Cambridge, 2002. MR 1976867 (2004c:46118)
28. David R. Pitts, Norming algebras and automatic complete boundedness of isomorphisms of operator algebras,
Proc. Amer. Math. Soc. 136 (2008), no. 5, 1757 -- 1768. MR 2373606
29. Florin Pop, Allan M. Sinclair, and Roger R. Smith, Norming C ∗-algebras by C ∗-subalgebras, J. Funct. Anal. 175
(2000), no. 1, 168 -- 196. MR 2001h:46105
30. J. Renault, A groupoid approach to C ∗-algebras, Lecture Notes in Mathematics, no. 793, Springer-Verlag, New
York, 1980.
31. Jean Renault, Cartan subalgebras in C ∗-algebras, Irish Math. Soc. Bull. (2008), no. 61, 29 -- 63. MR 2460017
32. Christian Svensson and Jun Tomiyama, On the commutant of C(X) in C ∗-crossed products by Z and their
representations, J. Funct. Anal. 256 (2009), no. 7, 2367 -- 2386. MR 2498769 (2010f:46105)
33. Masamichi Takesaki, Theory of operator algebras. I, Springer-Verlag, New York, 1979. MR 81e:46038
34. Stephen Willard, General topology, Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont.,
1970. MR 0264581 (41 #9173)
Dept. of Mathematics, University of Nebraska-Lincoln, Lincoln, NE, 68588-0130
E-mail address: [email protected]
62
|
1310.6585 | 1 | 1310 | 2013-10-24T12:45:24 | Pseudodifferential calculus on manifolds with fibred corners : the groupoid of phi-calculus | [
"math.OA",
"math.DG"
] | This paper is concerned with pseudodifferential calculus on manifolds with fibred corners. Following work of Connes, Monthubert, Skandalis and Androulidakis, we associate to every manifold with fibred corners a longitudinally smooth groupoid which algebraic and differential structure is explicitely described. This groupoid has a natural geometric meaning as a holonomy groupoid of singular foliation, it is a singular leaf space in the sense of Androulidakis and Skandalis. We then show that the associated compactly supported pseudodifferential calculus coincides with Mazzeo and Melrose's phi-calculus and we introduce an extended algebra of smoothing operators that is shown to be stable under holomorphic functional calculus. This result allows the interpretation of phi-calculus as the pseudodifferential calculus associated with the holonomy groupoid of the singular foliation defined by the manifold with fibred corners. It is a key step to set the index theory of those singular manifolds in the noncommutative geometry framework. | math.OA | math |
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS
WITH FIBRED CORNERS :
THE GROUPOID OF PHI-CALCULUS
LAURENT GUILLAUME
Abstract. This paper is concerned with pseudodifferential calculus on mani-
folds with fibred corners. Following work of Connes, Monthubert, Skandalis
and Androulidakis, we associate to every manifold with fibred corners a lon-
gitudinally smooth groupoid which algebraic and differential structure is ex-
plicitely described. This groupoid has a natural geometric meaning as a ho-
lonomy groupoid of singular foliation, it is a singular leaf space in the sense
of Androulidakis and Skandalis. We then show that the associated compactly
supported pseudodifferential calculus coincides with Mazzeo and Melrose's φ-
calculus and we introduce an extended algebra of smoothing operators that is
shown to be stable under holomorphic functional calculus. This result allows
the interpretation of φ-calculus as the pseudodifferential calculus associated
with the holonomy groupoid of the singular foliation defined by the manifold
with fibred corners. It is a key step to set the index theory of those singular
manifolds in the noncommutative geometry framework.
Introduction
In order to generalize the Atiyah-Patodi-Singer index theorem, Melrose intro-
duced in [Mel93] a pseudodifferential calculus on manifolds with boundary and
manifolds with corners : the b-calculus. Melrose and Mazzeo then analyzed the case
of fibred boundaries and introduced φ-calculus to extend the theorem to families
of operators [MM98]. Moreover the study initiated by Thom [Tho69] and Mather
[Mat70] of pseudo-stratified manifolds and the existence of a desingularization pro-
cess ([Ver84],[BHS91]) show that any pseudo-stratified manifold is a quotient space
of a manifold with fibred corners ([DL09],[DLR11]). Those results motivate the
need to understand pseudodifferential calculus on manifolds with fibred corners.
The aim of this paper is to set the index theory of those singular manifolds in
In particular we want to obtain the
the noncommutative geometry framework.
analogue of the leaf space of a foliation for those spaces. Ehresmann, Haefliger
[Hae84] and Wilkelnkemper [Win83] introduced groupoids to model the leaf space of
a regular foliation. Pradines and Bigonnet [BP85], Debord [Deb01], Androulidakis
and Skandalis [AS06] studied the case of singular foliations. Among others they
have contributed to realizing the idea that groupoids are natural substitutes to
singular spaces.
Following fundamental work by Connes [Con79, Con82, Con94], it also appears
that groupoids are key objects to understand pseudodifferential calculus and K-
theory groups which are the receptacles of the index. Pseudodifferential calculus
on Lie groupoids has been defined independently by Monthubert-Pierrot [MP97]
1
2
LAURENT GUILLAUME
and by Nistor, Weinstein and Xu [NWX99]. Androulidakis and Skandalis pro-
posed a general framework to deal with singular foliations [AS09]. Some particular
groupoid models for manifolds with corners and conic manifolds were proposed by
Monthubert [Mon03] and Debord-Lescure-Nistor [DLN06]. Monthubert showed in
[Mon03] that to every manifold with corners X could be associated a longitudinally
smooth groupoid Γ(X) whose pseudodifferential calculus Ψ∞(Γ(X)) coincides with
Melrose's b-calculus. This original result gave new proofs for the study of b-calculus.
In this article we show that such a construction exists for φ-calculus by associ-
ating to every manifold with fibred boundary, then to every manifold with fibred
corners a longitudinally smooth groupoid Γφ(X). We then show that the associ-
ated compactly supported pseudodifferential calculus coincides with Mazzeo and
Melrose's φ-calculus and we introduce an extended algebra of smoothing opera-
tors that is shown to be stable under holomorphic functional calculus. Finally we
show that the groupoid we built has a natural geometric meaning as a holonomy
groupoid of singular foliation, it is an explicit example of a singular leaf space in
the sense of Androulidakis and Skandalis [AS06]. This result allows the interpreta-
tion of φ-calculus as the pseudodifferential calculus associated with the holonomy
groupoid of the singular foliation defined by the manifold with fibred corners. The
reward of this conceptual approach is a simplified exposition of φ-calculus definition
and properties, as well as a new geometric interpretation as a holonomy groupoid.
Those results are based on the PhD thesis of the author (see [Gui12]).
The paper is organized in 5 sections.
In the first section we recall classical
material on groupoids, their associated pseudodifferential calculus and normal cone
deformations. These are crucial notions to set the index theory of manifolds with
fibred corners in the noncommutative geometry framework.
In the second section we introduce some definitions related to manifolds with
corners and manifolds with embedded fibred corners :
following Monthubert the
latter are embedded in smooth manifolds endowed with a transverse family of fibred
submanifolds of codimension 1 : we call such a structure a fibred decoupage.
In the third section we define the groupoid associated to a fibred decoupage in
two steps:
• we construct a "puff groupoid" G VF for any foliated codimension 1 sub-
manifold (V, F ) of the manifold M. This groupoid is described as a gluing
between the holonomy groupoid of M \ V and a deformation groupoid
Dϕ = Dϕ ⋊ R∗
+:
G VF = Dϕ[Ψ
Hol(M \ V ).
Dϕ is the normal cone deformation of the groupoid immersion H ol(F ) →
V × V .
• the groupoid of a fibred decoupage is given by the fibered product of the
puff groupoids G VF i for each face.
We can then define the groupoid Γφ(X) of a manifold with fibred corners X as the
restriction of the groupoid of a decoupage in which X is embedded. One recovers
the groupoid of b-calculus by Monthubert in the special case of a trivial fibration.
The section ends with the proof of amenability and longitudinal smoothness of the
groupoid Γφ(X).
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
3
Pseudodifferential calculus on manifolds with fibred corners is studied in the
fourth section. To any manifold with fibred corners X we associate a pseudo-
differential calculus with compact support Ψ∞
c (Γφ(X)) and we show it coincides
with Melrose's φ-calculus. We then introduce an extended calculus Ψ∞(Γφ(X)) =
Ψ∞
c (Γφ(X))+Sψ(Γφ(X)) derived from a polynomial length function ψ and we show
this last algebra is stable under holomorphic calculus and includes the operators of
the extended φ-calculus of Mazzeo and Melrose.
Finally in the fifth section the groupoid of a fibred decoupage is shown to be
the holonomy groupoid of the singular foliation defined by the manifold with fibred
corners. The groupoid described in this article is therefore an explicit example of
a singular leaf space in the sense of Androulidakis and Skandalis.
Contents
Index theory on singular spaces
Introduction
1. Preliminaries on Lie groupoids
1.1. Lie groupoids - Algebroids
1.2. G -pseudodifferential calculus
1.3.
1.4. Normal cone deformation
2. Manifolds with fibred corners
2.1. Manifolds with corners
2.2. Manifolds with embedded corners
2.3. Manifolds with iterated fibred corners
2.4. Decoupages
2.5. Manifolds with fibred corners
3. The groupoid of manifolds with fibred corners
3.1. The groupoid of a codimension 1 foliated submanifold
3.2. The groupoid of manifolds with fibred corners
4. Pseudodifferential calculus on manifolds with fibred corners
b-stretched product and b-calculus
4.1.
4.2. φ-stretched product and φ-calculus
4.3. φ-calculus and the groupoid of manifolds with fibred boundary
4.4.
4.5. Extended pseudodifferential calculus
4.6. Total ellipticity and Fredholm index
5. The holonomy groupoid of manifolds with fibred corners
5.1. The singular foliation of a fibred boundary
5.2. The groupoid of a fibred decoupage as a holonomy groupoid
References
Identification of Ψ∞
c (Γφ(X)) with φ-calculus
1
3
3
4
5
6
8
8
8
9
10
10
11
11
15
17
17
17
19
20
21
23
24
24
25
27
1. Preliminaries on Lie groupoids
The elements of index theory recalled hereafter mainly come from work by
Connes [Con79], Monthubert-Pierrot [MP97] et Nistor-Weinstein-Xu [NWX99].
The nice exposition in [CR07] is used as a guideline.
1.1. Lie groupoids - Algebroids.
4
LAURENT GUILLAUME
1.1.1. Groupoids.
Definition 1.1. A groupoid is a small category in which all morphisms are invert-
ible. It is composed of a set of objects (or units) G (0) and a set of arrows G with two
source and target applications s, r : G → G (0) and a composition law m : G (2) → G
associative on the set of composable arrows G (2) = {(γ, η) ∈ G × G : s(γ) = r(η)}.
It is also assumed there exists a unit map u : G (0) → G such that r◦u = s◦u = Id
and an inverse involutive map i : G → G such that s ◦ i = r.
Finally by denoting m(γ, η) = γ · η, every element γ of G satisfy the relations
γ · γ−1 = u(r(γ)), γ−1 · γ = u(s(γ)) and r(γ) · γ = γ · s(γ) = γ.
A Lie groupoid is a groupoid for which G and G (0) are smooth manifolds, all
maps above are smooth and s, r are submersions.
It is classicaly denoted GA = s−1(A), G B = r−1(B) and G B
A = GA ∩ G B.
1.1.2. Algebroids.
Definition 1.2. Let M be a smooth manifold. A Lie algebroid on M is given by
a vector bundle A → M , a bracket [. , .] : Γ(A) × Γ(A) → Γ(A) over the module
Γ(A) of sections of A and by a bundle morphism p : A → T M called anchor map,
such that:
(1) [. , .] is R-bilinear, antisymetric et satisfies Jacobi identity,
(2) [V, f W ] = f [V, W ] + p(V )(f )W for all V, W ∈ Γ(A) and f ∈ C∞(M )
(3) p([V, W ]) = [p(V ), p(W )] for all V, W ∈ Γ(A).
Every Lie groupoid G defines a Lie algebroid A G by the normal bundle of the
inclusion G (0) ⊂ G . Sections of A G are in a bijective correspondance with vector
fields on G which are s-vertical and right-invariant, it therefore induces a Lie algebra
structure on Γ(A G ).
Remark 1.3. Every Lie algebroid defines a foliation by the image of its anchor
map p(C∞
c (M, A)). In particular every Lie groupoid defines a foliation.
1.2. G -pseudodifferential calculus. A G -pseudodifferential operator is a differ-
entiable family of pseudodifferential operators {Px}x∈G (0) acting on C∞
c (Gx) such
that for γ ∈ G and Uγ : C∞
c (Gr(γ)) the induced operator, the follow-
ing G -invariance condition is verified:
c (Gs(γ)) → C∞
Pr(γ) ◦ Uγ = Uγ ◦ Ps(γ).
Operators can act more generally on sections of a vector bundle E → G (0). The
exact condition of differentiability is defined in [NWX99].
Only uniformly supported operators will be considered here, let us briefly recall
this notion. Let P = (Px, x ∈ G (0)) be a G -operator and denote kx the Schwartz
kernel of Px. The support and reduced support of P are defined by
supp P := ∪xsupp kx
suppµP := µ1(supp P )
where µ1(g′, g) = g′g−1. We say that P is uniformly supported if suppµP is compact
in G .
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
5
In the following we denote Ψm
c (G , E) the space of G -pseudodifferential operators
uniformly supported. We denote also
Ψm
c (G , E) and Ψ−∞
c
Ψm
c (G , E).
(G , E) =\m
Ψ∞
c (G , E) =[m
Remark 1.4. The reduced support condition is justified by the fact that Ψ−∞
is identified with C∞
c (G , End(E)) using the Schwartz kernel theorem ([NWX99]).
(G , E)
c
Remark 1.5. Ψ∞
In particular, Ψ−∞
c (G , E) is a filtered algebra, ie Ψm
c
(G , E) is a two-sided ideal.
c (G , E)Ψm′
c (G , E) ⊂ Ψm+m′
c
(G , E).
The definition above is equivalent to that of [Mon03], definition 1.1, where the
space of pseudodifferential kernels on a longitudinally smooth groupoid G is defined
as the space I∞(G , G (0), Ω
2 ) of distributional sections K on G with values in half-
2 which are smooth outside G (0) and given by oscillatory integral in a
densities Ω
neighborhood of G (0) :
1
1
K(γ) = (2π)−nZA∗Gr(γ)
exp (i < φ(γ), ξ >)a(γ, ξ)dξ,
where a is a polyhomogeneous symbol of any order with values in Ω
1
2 .
If P = (Px, x ∈ G (0)) ∈ Ψm
The notion of principal symbol extends easily to Ψ∞
c (G , E). Denote by π :
A∗G → G (0) the projection.
c (G , E, F ) is a pseu-
dodifferential operator of order m on G , the principal symbol σm(Px) of Px is a
C∞ section of the vector bundle End(π∗
xr∗F ) over T ∗Gx, such that the
morphism defined over each fiber is homogeneous of degree m. The existence of
invariant connections allows the definition of exponentiation on the Lie algebroid
A∗G and provides a section σm(P ) of End(π∗E, π∗F ) over A∗G which satisfies
xr∗E, π∗
(1)
σm(P )(ξ) = σm(Px)(ξ) ∈ End(Ex, Fx) if ξ ∈ A∗
x
G
modulo the space of symbols of order m − 1. Terms of order m of σm(P ) are
invariant under a different choice of connection and the equation above induces a
unique surjective linear map
σm : Ψm
c (G , E) → S m(A∗G , End(E, F )),
(2)
with kernel Ψm−1
(G , E) ([NWX99], proposition 2) where S m(A∗G , End(E, F ))
denotes the sections of the fiber End(π∗E, π∗F ) over A∗G homogeneous of degree
m at each fiber.
c
1.3. Index theory on singular spaces. In the classical case of a compact mani-
fold M without boundary, recall that an elliptic pseudodifferential operator D has
a kernel and cokernel of finite dimension. The Fredholm index of D is the integer:
ind D = dim ker D − dim Coker D
This index is stable for any compact perturbation of D and its value only de-
pends on topological data. In the 60's Atiyah and Singer introduced fundamen-
tal constructions in K-theory which allowed the understanding of the application
D → ind D through the group morphism
called the analytical index of M.
inda : K 0(T ∗M ) → Z,
6
LAURENT GUILLAUME
More precisely, if Ell(M ) denotes the set of pseudodifferential operators on M,
we have the following commutative diagram:
Ell(M ) ind
Z
symb
inda
K 0(T ∗M )
symb
→ K 0(T ∗M ) is the surjective application which maps an operator
where Ell(M )
to the class of its principal symbol in K 0(T ∗M ). Atiyah and Singer then defined a
topological index indt : K 0(T ∗M ) → Z with characteristic K-theoretical properties,
showed that a unique morphism can check those properties and that the analytical
index do satisfy them. The identity of the analytic and topological index is the
celebrated Atiyah-Singer theorem [AS63, AS68a, AS68b].
In the case of singular spaces such as the space of leaves of a regular foliation,
orbifolds or manifolds with corners, similar constructions can be obtained by in-
troducing groupoids [Con79, CS84]. A groupoid is a small category in which all
morphisms are invertible. Groupoids generalize the concepts of spaces, groups and
equivalence relations. A groupoid is said to be Lie if the sets involved are smooth
manifolds and morphisms are differential maps. A pseudodifferential calculus was
developed for Lie groupoids [Con79, MP97, NWX99] and more generally for longi-
tudinally smooth groupoids such as continuous family groupoids [LMN00].
The analytic index of elliptic operators on these groupoids is a morphism:
inda : K 0(A∗G ) → K0(C∗
r (G ))
r (G ) is the reduced C∗-algebra
where A∗G is the algebroid of the groupoid G and C∗
of G which plays the role of the algebra of continuous functions on the singular space
represented by G [Ren80]. Indices are thus generally not integers but elements of
the K-theory group K0(C∗
r (G )). However in the case of a compact manifold M
without boundary they are classical indices as G = M × M , A∗G = T ∗M and
K0(C∗
r (G )) = K0(K ) = Z.
A fundamental property of the analytic index is that it can be factorized through
the principal symbol map, there is a commutative diagram :
Ell(G )
ind
K0(C∞
c (G ))
.
σ
j
K 0(A∗G )
K0(C∗
r (G )).
inda
Indeed inda is the boundary morphism associated with the short exact sequence of
C∗-algebras ([Con79, CS84, MP97, NWX99])
(3)
0 → C∗
r (G ) −→ Ψ0(G ) σ−→ C0(S∗G ) → 0
where Ψ0(G ) is a C∗−completion of Ψ0
the extension of the principal symbol.
c(G ), S∗G the sphere bundle of A∗G and σ
1.4. Normal cone deformation. We recall in this subsection some basic prop-
erties of normal cone deformation groupoids, introduced by Hilsum and Skandalis
[HS87]. These important objects in noncommutative geometry especially allow
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
7
the construction of elements in KK-theory and the formulation of index problems
(tangent groupoid of Connes [Con94]). They are central in the description of the
groupoid of manifolds with fibred corners.
1.4.1. Definition. Let M be a smooth manifold and N be a smooth submanifold
of M . Denote N M
N → M the normal vector bundle to the inclusion N ⊂ M :
N M
N → M = TN M/T N . The normal cone deformation of the inclusion N ⊂ M is
defined as a set by :
DM
N = N M
N × 0G M × R∗.
More generally, if G1 and G2 are two Lie groupoids with respective Lie alge-
broids AG1 and AG2 satisfying AG1 ⊂ AG2, the normal cone deformation groupoid
([HS87]) can be defined for the induced immersion ϕ : G1 → G2 by :
Dϕ = G1 ×s1
N × 0G G2 × R∗.
The groupoid G1 acts on the normal N = AG2/AG1 by its holonomy component
hxy : Nx → Ny :
(γ, v) = ((x, y, h), (y, ξ)) 7→ γ · v = (x, h−1(ξ))
Thus G1 ×sV
N = {(γ, v) ∈ G1 × N , v ∈ Ns1(γ)} is given a composition law by
:
(4)
(γ1, v) · (γ2, w) = (γ1γ2, v + γ1 · w)
when t(γ2) = s(γ1), with inverse (γ, v)−1 = (γ−1, −γ−1 · v).
G2×R∗ is naturally a Lie groupoid with point composition (γ1, t)·(γ2, t) = (γ1·γ2, t).
Dϕ is then the union of the Lie groupoids G1 ×sV
N and G2 × R∗.
1.4.2. Differential structure. In the simple case of the canonical inclusion Rp×{0} ⊂
p is the set Rp×Rq×R
Rp×Rq = Rn, where q = n−p, the normal cone deformation Dn
with the C∞ structure induced by the bijection Θ : Rp × Rq × R → Dn
p :
Θ(x, ξ, t) =(cid:26) (x, ξ, 0)
(x, tξ, t)
if t = 0
if t 6= 0
In the local case of an open set U ⊂ Rn and a submanifold V = U ∩ (Rp × {0}),
p provided with the above
DU
structure as Θ−1(DU
V = V × Rq × {0}F U × R∗ is an open subset of Dn
V ) is the set
V = {(x, ξ, t) ∈ Rp × Rq × R , (x, tξ) ∈ U }
ΩU
(5)
which is an open set of Rp × Rq × R and therefore a smooth manifold.
In the general case of manifolds suppose M of dimension n and N of dimension
N can be described locally from the open sets
p. The differential structure on DM
Dn
p with adapted charts.
Definition 1.6 ([CR07]). A local chart (U, φ) of M is said to be a N -slice if
(1) φ : U ≃→ U ⊂ Rp × Rq
(2) if U ∩ N = V, V = φ−1(V ) (where V = U ∩ (Rp × {0}) as above)
8
LAURENT GUILLAUME
Let (U, φ) be a N -slice. When x ∈ V we have φ(x) ∈ Rp × {0} : we denote by
φ1 the component of φ on Rp such that φ(x) = (φ1(x), 0) and dnφv : Nv → Rq the
normal component of the differential dφv for x ∈ V. Let φ : DV
U be the map
defined by :
U → DV
φ :(cid:26) φ(v, ξ, 0) = (φ1(v), dnφv(ξ), 0)
φ(u, t) =
(φ(u), t)
if t 6= 0
Then the map ϕ = Θ−1 ◦ φ obtained as the composition
is a diffeomorphism of DU
way as (5).
V on the open set ΩU
DU
V
φ
−→ DU
V
Θ−1
−→ ΩU
V
V of Rp × Rq × R defined in the same
The global differential structure of DM
N is then described by the following propo-
sition :
Proposition 1.7 ([CR07]). Let {(Uα, φα)} be a C∞ atlas of M composed of N-
slices. Then {(DUα
Vα
, ϕα)} is a C∞ atlas of DM
N .
A change of atlas is described by the following result (see [CR07]) : Let U ⊂
Rp × Rq and U ′ ⊂ Rp × Rq be open sets and F : U → U ′ a C∞ diffeomorphism
decomposed as a Rp × {0}-slice F = (F1, F2) such that F2(x, 0) = 0. Then the map
F : ΩU
V ′ defined by :
V → ΩU ′
F (x, ξ, t) =(cid:26) ∂F
∂ξ (x, 0) · ξ
1
t F (x, tξ)
is a C∞ diffeomorphism from ΩU
V to ΩU ′
V ′.
if t = 0
if t 6= 0
2. Manifolds with fibred corners
2.1. Manifolds with corners. A manifold with corners X is a topological space
where any point p has a neighbourhood diffeomorphic to Rk
+ × Rn−k, with 0 as
image of p. k is called the codimension of p. The connected components of the set
of points with codimension k are called open faces of codimension k. Their closure
are the faces of X. A face of codimension 1 is called an hyperface of X. The
boundary of X is denoted by ∂X, it is the union of faces with codimension k > 0.
2.2. Manifolds with embedded corners. Manifolds with embedded corners
have been studied by Melrose to understand and extend Atiyah-Patodi-Singer in-
dex theorem (see [Mel93]). An equivalent description in terms of decoupages has
been proposed by Monthubert in [Mon03]. The two definitions are recalled below
(section 2.6 for the definition with decoupages).
Definition 2.1 (Manifold with embedded corners, [Mel93]). A manifold with em-
bedded corners X is a manifold with corners endowed with a subalgebra C∞(X)
satisfying the following conditions:
· there exists a manifold X and a map j : X → X such that
C∞(X) = j∗C∞( X),
· there exists a finite family of functions ρi ∈ C∞( X) such that
j(X) = {y ∈ X, ∀i ∈ I, ρi(y) ≥ 0},
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
9
· for any J ⊂ I and any point of X where the (ρj)j∈J simultaneously vanish,
the differentials dρj are independent.
The functions ρi are called defining functions of the hyperfaces.
Remark 2.2. Manifolds with embedded corners are manifolds with corners : the
number of functions ρj which vanish at a point p determine the codimension of p
since the ρj, which are independent, can be used as local coordinates and induce a
diffeomorphism from a neighbourhood of p to some Rk
+ × Rn−k.
Remark 2.3. A definition function ρ of an hyperface F induces a trivialization of
the normal bundle on its interior F ◦.
Examples 2.4. The edge, the square, cubes of dimension n are manifolds with
embedded corners. The simplex Σn = {(t0, . . . , tn) ∈ Rn+1
+ , t0 + · · · + tn = 1} is a
manifold with embedded corners. The drop with one corner is not a manifold with
embedded corners, contrary to the drop with two corners.
Figure 1. Drops with one and two corners.
2.3. Manifolds with iterated fibred corners. The study of pseudo-stratified
manifolds motivates the introduction of a fibred structure on manifolds with embed-
ded corners ([DLR11]). Indeed the existence of a desingularization process shows
that any pseudo-stratified manifold is a quotient space of a manifold with fibred
corners ([DL09],[DLR11],[ALMP09]). For such a desingularization each hyperface
Fi is the total space of a fibration φi : Fi → Yi where Yi is also a manifold with
corners such that the family of fibrations φ = (φ1, . . . , φk) satisfy the properties of
an iterated fibred structure ([DLR11]), ie :
· ∀I ⊂ {1, . . . , k} such thatTi∈I Fi 6= ∅, the set {Fi, i ∈ I} is totally ordered,
· if Fi < Fj, then Fi ∩ Fj 6= ∅, φi : Fi ∩ Fj → Yi is a surjective submersion
.
= φj(Fi ∩ Fj ) ⊂ Yj is an hyperface of the manifold with corners Yj.
and Yji
Moreover there exists a surjective submersion φji : Yji → Yi which satisfies
φji ◦ φj = φi on Fi ∩ Fj .
· the hyperfaces of Yj are exactly the Yji with Fi < Fj. In particular if Fi is
minimal Yi is a manifold without boundary.
We call manifold with iterated fibred corners a manifold with embedded corners
endowed with an iterated fibred structure.
10
LAURENT GUILLAUME
2.4. Decoupages. We recall the definition of a decoupage which is a useful notion
to associate to any manifold with embedded corner a "puff" Lie groupoid [Mon03].
Generalizing this approach we then introduce an original definition of fibred de-
coupage suited to the case of fibrations and allowing the direct construction of a
puff groupoid for manifolds with fibred corners [Gui12]. Before let us recall a few
facts about transversality.
Definition 2.5. Let X1, . . . , Xn be a family of smooth manifolds and for each i
let φi : Xi → Y be a smooth map. If (xi)1≤i≤n ∈Q1≤i≤n Xi is a family such that
φ1(x1) = · · · = φn(xn), the family (φi)1≤i≤n, φi : Xi → Y is said to be transverse
if the orthogonals (in T ∗Y ) of the spaces dφi(T Xi) are in direct sum.
Under this condition the fibered product of the Xi over Y is a smooth subman-
ifold of Q1≤i≤n Xi.
A family of submanifolds will be called transverse if the inclusions of these man-
ifolds are transverse; a smooth map is said to be transverse if it is transverse to the
inclusion of this submanifold.
Definition 2.6 (Decoupage, [Mon03] 2.4). A decoupage is given by a manifold M
and a finite family (Vi)i∈I of submanifolds of codimension 1 such that ∀J ⊂ I the
family of inclusions of the (Vj)j∈J is transverse.
The decoupage is said to be oriented if every submanifold Vi is transversally
the intersection
oriented. Moreover if each Vi splits M in two parts M +
of the M +
i
is called positive part.
i and M −
i
The equivalent definition of a manifold with embedded corners as a decoupage
is then:
Definition 2.7 (Manifold with embedded corners). A manifold with embedded
corners X is the positive part of an oriented decoupage (M, (Vi)i∈I ). X is the
positive part of this decoupage and (M, (Vi)i∈I ) is said to be an extension of X.
Definition 2.8 (Fibred decoupage, [Gui12] 2.3). A fibred decoupage is given by
a decoupage (M, (Vi)i∈I ) and a family of fibrations φi : Vi → Yi such that ∀J ⊂ I
the family of inclusions of the (Vj ×Yj Vj)j∈J in M 2 is transverse.
2.5. Manifolds with fibred corners.
Definition 2.9 ([Gui12] 2.4). Let X be a manifold with embedded corners and
(M, (Vi)i∈I ) an extension of X. We call X a manifolds with fibred corners if there
: Vi → Yi such that (M, (Vi, φi)i∈I ) is a fibred
exists a family of fibrations φi
decoupage.
Example 2.10. A family of disjoint fibred submanifolds of codimension 1 is always
transverse. In particular manifolds with fibred boundaries, the objects of Melrose's
φ-calculus, are manifolds with fibred corners.
Example 2.11. X = R3
+. Let M = R3, πx, πy and πz be the projections on the
canonical basis, Vx = ker πx, Vy = ker πy, Vz = ker πz. Let φx : Vx → {0}, φy :
Vy → ker πx ∩ker πy, φz : Vz → ker πy ∩ker πz. Then (M, (Vx, φx), (Vy, φy), (Vz, φz))
is a fibred decoupage.
Example 2.12. Same notations as previous example but with φ′
Vy → ker πy ∩ker πz, φ′
is not a fibred decoupage.
z : Vz → ker πx ∩ker πz. Then (M, (Vx, φ′
x : Vx → {0}, φ′
y), (Vz, φ′
x), (Vy, φ′
y :
z))
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
11
Example 2.13. The example 2.11 is a fibred decoupage but is not a iterated fibred
structure. Indeed φz(Vy ∩ Vz) = Vz ∩ ker πy, Yyz = φy(Vy ∩ Vz) = {0} and there is
no surjection of Yyz on Yz = ker πy ∩ ker πz as dim Yyz = 0 < 1 = dim Yz.
Remark 2.14. The definition used for manifolds with fibred corners ensures longi-
tudinal smoothness of the puff groupoid (see Proposition 3.10). The results of this
chapter can therefore be applied to fibrations coming from other structures than
iterated fibred structures, since any fibred decoupage does not necessarily come
from an iterated fibred structures (eg 2.13).
However, it is perfectly possible to limit the study to the category of iterated
fibred decoupages, ie the decoupages which fibrations satisfy the properties of an
iterated structure.
3. The groupoid of manifolds with fibred corners
3.1. The groupoid of a codimension 1 foliated submanifold. In this subsec-
tion we construct a "puff groupoid" G VF for any foliated codimension 1 submanifold
(V, F ) of a manifold M. This groupoid is described as a gluing between the holo-
nomy groupoid of M \ V and a deformation groupoid Dϕ = Dϕ ⋊ R∗
+ built from
the normal cone deformation Dϕ of the groupoid immersion H ol(F ) → V × V :
G VF = Dϕ[Ψ
Hol(M \ V ).
Gluing is preformed through a map Ψ and one has the following commutative
diagram
V × V × R∗
+ × R∗ Ψ
Hol(M \ V )
iA
Dϕ
iM \V
G VF
iD
G VF encodes the space of leaves of the singular foliation FM defined by vector
fields on M tangent to F on V . This property will be precised in the section devoted
to singular foliations : we will show G VF is nothing but the holonomy groupoid
Hol(M, FM ) of FM (proposition 5.7).
3.1.1. Notations. Let M be a smooth manifold and V ⊂ M a connected submani-
fold of codimension 1 transversally oriented. V is supposed to be foliated by a regu-
lar foliation F defined by an integrable vector subspace T F ⊂ T V , i.e. C∞(V, T F )
is a Lie subalgebra of C∞(V, T V ). Let denote N = T V /T F and suppose M is
given a connexion w which restriction to V provides a decomposition of the tangent
bundle T V = T F ⊕ N .
Let (N, π, V ) be a tubular neighbourhood of V in M : N is an open set of M
including V, π : N → V a vector bundle with R type fibre. Let {fi
: Ni =
π−1(Vi) → Vi × R}i∈I be a local trivialization of N, which components are denoted
fi = (πi, ρi)i∈I .
3.1.2. The deformation groupoid Dϕ = Dϕ ⋊ R∗
+.
12
LAURENT GUILLAUME
Definition. Let G1 be the holonomy groupoid [Con82, Con94] associated with F and
ϕ : G1 → V × V the immersion induced by the inclusion T F ⊂ T V . The normal
cone deformation groupoid of ϕ is denoted Dϕ.
Dϕ induces on its space of units V ×R a singular foliation Fadp defined by F ×{0}
and T V × {t} for t 6= 0 resulting from the integration of the partial adiabatic Lie
algebroid Aadp described in ([Deb01], Example 4.5):
padp : Aadp = T F ⊕ N × R →
T V × T R
(v1, v2, t)
7→ (v1 + tv2, (t, 0))
In particular the foliation of R transverse to V induced by Dϕ is trivial in points
as each t 6= 0 defines a leaf V × {t} of V × R.
But the foliation FM defined by the transversally oriented faces Vi of a manifold
with fibred corners has a singular transverse structure of the form Vi × {0},Vi × R∗
+
and Vi × R∗
−. For that purpose we introduce the deformation groupoid Dϕ obtained
by the transverse action of R∗
+ on Dϕ explicitely given by :
(λ, h) = (λ, (γ, t)) 7→ λ · γ = (γ, λ · t)
The orbits of Dϕ = Dϕ ⋊ R∗
An expression of the manifold Dϕ as a set is :
+ then coincide with the foliation FM .
Dϕ = (G1 ×s1
N × R∗
+) × {0}G V × V × R∗
+ × R∗ ⇒ V × R.
Composition law. Let p : V × R∗
+ × R → V be the canonical projection, s = p ◦ sϕ
and r = p ◦ rϕ. For a given element g = (γ, λ, t) of Dϕ, s(g) and r(g) only depend
on γ ∈ G1 ×s1
corresponding elements.
N F V × V . To keep notations simple we denote s(γ) and r(γ) the
The composition law on Dϕ is then the product of the law composition on Dϕ
as a normal cone groupoid and the law on H = R ⋊ R∗
+ acts on R by
multiplication. If g1 = (γ1, λ1, t1) and g2 = (γ2, λ2, t2) are two elements of Dϕ such
that s(γ2) = r(γ1) and sH(λ2, t2) = rH (λ1, t1), we have :
+ where R∗
(6)
g1 · g2 = (γ1 · γ2, λ1 · λ2, t1)
3.1.3. The puff groupoid G VF . We are ready to define the gluing function Ψ. Let
+ × R∗ and set
A ⊂ Dϕ be the restriction of Dϕ to the product component V ×V × R∗
Aj
i = {(γ, λ, t) ∈ Dϕ / s(γ) ∈ Vi, r(γ) ∈
Vj}.
+ × R∗ where D j
i = Vi × Vj × R∗
i = A ∩ D j
Now consider the maps Ψij : Aj
i → M ◦ × M ◦ defined by
Ψij(γ, λ, t) = (f −1
i
(s(γ), t), f −1
j
(r(γ), λ · t))
The Aj
i being disjoint we have A = ⊔i,jAj
i and one defines Ψ : A → M ◦ × M ◦ by
Ψ(a) = Ψij(a) for a ∈ Aj
i . It is immediate to check that Ψ preserves the transverse
orientation of V in M : Ψ(Aj+
i ) ⊂ (M − \ V )2. V being
connected, the image of Ψ is thus in the union of the connected components of
(M \ V )2, that is to say in Hol(M \ V ).
i ) ⊂ (M + \ V )2, Ψ(Aj−
G VF is then defined as a topological space by the gluing of Dϕ and Hol(M \ V )
by the map Ψ:
G VF = Dϕ[Ψ
Hol(M \ V )
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
13
and one has the following commutative diagram :
V × V × R∗
+ × R∗ Ψ
Hol(M \ V )
iA
Dϕ
iM \V
G VF .
iD
3.1.4. Differential structure of the puff groupoid G VF . Let (ϕij )i,j∈I 2 be an atlas on
Dϕ which charts are slices of D j
+). Set p = dim G1 and q = dim N =
2 · dim V − p.
i ∩ (G1 × R∗
Let U j
i , W and Z be respective open sets of Rp+q, Rp and R satisfying W = U j
−1(U j
i × R∗) ⊂ Vi × Vj × R∗.
Topology on Dϕ is generated (see section 1.4.2) by the open sets Θ ◦ ϕij (ΩU j
i
i ∩ (Rp × {0})
W ) × exp Z
and ϕij
where:
i
ΩU j
W = {(x, ξ, t) ∈ Rp × Rq × R , (x, tξ) ∈ U j
is an open set of Dϕ generated by the image of ΩVi×Vj
and
i is an open set of Dϕ generated by the image of Z = R and
i }.
W
In particular D j
i
Z = R. Similary Aj
ΩVi×Vj
= ΩVi×Vj
∗
W
∩ (Rp × Rq × R∗).
One then checks that sϕ × rϕ is an open map on A2 as the map (t, λ) 7→ (t, λ · t)
i → M ◦ × M ◦ is thus open as a
+. The map Ψij : Aj
is a diffeomorphism of R∗ × R∗
composition of the open maps fi × fj and sϕ × rϕ.
Therefore iD : Dϕ → G VF defined by :
iD (g) =(cid:26) Ψ(g)
g
if g ∈ A
otherwise
is an open map and an atlas A = {(Ωj
Ωj
i = iD (D j
Let B be an atlas of the smooth manifold M ◦ × M ◦ and let B = {(Ωβ, ϕβ)} be
i , φij )i,j∈I 2 } on G VF is obtained by setting
i ) and φij = ϕij ◦ i−1
D .
the induced atlas on the manifold Hol(M \ V ).
The differential structure on G VF is obtained as the product structure of M ◦ ×
M ◦ and the differential structure induced by the family (φij ). More precisely we
have the following proposition :
Proposition 3.1. A ∪ B is a smooth atlas on G VF .
Demonstration. Let (Ω, ϕ) be a chart of B. We must prove that any chart of A
meeting Ω is compatible with ϕ. Let i and j be such that Ωj
i ∩ Ω 6= ∅ and define
Φij : ϕ(Ωj
ij ◦ ϕ−1. It is immediate to see that
i ∩ Ω ⊂ Ψij(Aj
Ωj
i \ Vi × N −
i ) by Φij = φij ◦ Ψ−1
i ∩ Ω) → φij (D j
j \ Vj and that
i \ Vi × N +
i ) = N +
Ωj
i ∩ Ω →
(x, y) → (πi(x), πj (y), ρj(y)/ρi(x), ρi(x))
j \ VjF N −
Aj
i
:
Ψ−1
ij
is a diffeomorphism on its image. It implies that Φij is a C∞ clutching function
and the compatibility of A and B follows.
(cid:3)
14
LAURENT GUILLAUME
Remark 3.2. In the general case above Dϕ only describes the differential structure
of G VF in a neighbourhood of V. The description gets simpler if a trivialisation
(π, ρ) is given with π and ρ globally defined on M. The map Ψ is then bijective,
the inverse map Ψ−1 : Hol(M \ V ) → V × V × R∗
+ being defined by :
Ψ−1(x, y) = (π(x), π(y), ρ(y)/ρ(x), ρ(x)) .
Thus iA ◦ Ψ−1 is injective and iD is a diffeomorphism of Dϕ on Im iD = G VF .
3.1.5. Composition law. The groupoid G1 = H ol(FM ) acts on the normal bundle
N by its holonomy component hxy : Nx → Ny :
(γ, v) = ((x, y, h), (y, ξ)) 7→ γ · v = (x, h−1(ξ))
Therefore G1 ×sV
N × R∗
+ = {(γ, v, λ) ∈ G1 × N × R∗
+, v ∈ Ns1(γ)} is endowed
with a composition law :
(γ1, v, λ1) · (γ2, w, λ2) = (γ1γ2, v + γ1 · w, λ1 · λ2)
(7)
when t(γ2) = s(γ1), with inverse (γ, v, λ)−1 = (γ−1, −γ−1 · v, λ−1).
The composition law on G VF is defined as the law (7) on NF = G1 ×s1
N × R∗
+
and the canonical product law on M ◦
− × M ◦
−.
Source and target maps of G VF are defined on NFV and its complementary by:
and
+ × M ◦
+F M ◦
s :(cid:26) s(γ, ξ, λ) = s1(γ)
r :(cid:26) r(γ, ξ, λ) = r1(γ)
s(x, y)
r(x, y)
= x
= y
The groupoid structure of G VF can be easily reformulated on the deformation
groupoid Dϕ. In fact Dϕ endowed with its composition law 6 is a Lie groupoid iso-
morphic to a neighbourhood of NF in G VF . More precisely, let iD be the inclusion
of Dϕ in G VF given by the gluing Ψ. Then :
Proposition 3.3. The Lie groupoids Dϕ and iD (Dϕ) ⊂ G VF are isomorphic.
Demonstration. iD is by construction a diffeomorphism from Dϕ to iD (Dϕ). We
observe that iD can be written under the form :
iD :(cid:26) g = (γ, λ, t) ∈ D j
i 7→ (f −1
i
◦ sϕ(g), f −1
j
◦ rϕ(g))
(γ, λ, 0)
if t 6= 0
if t = 0
Let g1 = (γ1, λ1, t1) and g2 = (γ2, λ2, t2) be two elements of Dϕ.
The relation iD (g1) · iD (g2) = iD (g1 · g2) is trivial if t1 = 0 since then iD = Id.
For t1 6= 0, g1 · g2 = (γ1 · γ2, λ1 · λ2, t1). One computes :
iD (g1) · iD (g2) = (f −1
= (f −1
= (f −1
= iD (g1 · g2)
◦ rϕ(g1)) · (f −1
◦ rϕ(g2)) with t2 = λ1 · t1
◦ (r(γ2), λ2 · λ1 · t1))
◦ sϕ(g1), f −1
◦ sϕ(g1), f −1
◦ sϕ(g1), f −1
◦ sϕ(g2), f −1
◦ rϕ(g2))
j
j
j
j
i
i
i
i
Thus iD (g1) · iD (g2) = iD (g1 · g2) for any composable elements of Dϕ and iD is
indeed a Lie groupoid morphism.
(cid:3)
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
15
3.1.6. The case of fibrations. Assume the foliation F of V is a fibration φ : V → Y .
The immersion ϕ : G1 → GV is then an embedding with G1 = V ×Y V and
GV = V × V . The holonomy of a fibration is trivial and G1 trivially acts on N by:
(γ, v) = ((x, x′), (x′, ξ)) 7→ γ · v = (x, ξ).
Moreover for a local trivialization also trivializing the fibration φ : V → Y , N
can be identified as the pushout of T Y . The puff groupoid is then locally described
as a set by:
G VφNi = (Vi ×Y Vi ×Y T Y × R∗
i \ Vi) × (N −
i \ Vi).
+)G(N +
i \ Vi) × (N +
i \ Vi)G(N −
The two limit cases of a coarse fibration Y = {pt} and a ponctual fibration Y = V
correspond to the geometric situations of b-calculus and 0-calculus. The normal
cone deformations Dϕ are then respectively isomorphic to the product groupoid
V × V and to the adiabatic groupoid of V. The puff of the coarse fibration is the
groupoid described in [Mon03] and [NWX99]. The definition of G VF as a holonomy
groupoid Hol(M, FM ) of the foliation FM (proposition 5.7) gives those groupoids
a geometric meaning independent of any particular construction.
Remark 3.4. A similar construction allows the description of a manifold which
interior is endowed with a coarser fibration Φ : M → B: if GΦ = V ×ΦV denotes the
holonomy groupoid of the fibration Φ restricted to the boundary, G Vφ is obtained
as the gluing of Dϕ and Hol(M, Φ) = M ×B M , where Dϕ = Dϕ ⋊ R∗
+ is built from
the normal cone deformation ϕ : Gφ → GΦ.
3.2. The groupoid of manifolds with fibred corners. The groupoid of a fibred
decoupage is given by the fibered product of the puff groupoids G VF i for each face.
The groupoid Γφ(X) of a manifold with fibred corners X is then defined as the
restriction of the groupoid of a decoupage in which X is embedded. One recovers
the groupoid of b-calculus by Monthubert in the special case of a trivial fibration.
The section ends with the proof of amenability and longitudinal smoothness of the
groupoid Γφ(X).
3.2.1. Definition.
Definition 3.5. Let Eφ = (M, (Vi, φi)i∈I ) be a fibred decoupage. The groupoid
of Eφ is the fibered product of the puff groupoids G Viφ through the maps s ⊕ r :
G Viφ → M 2.
Definition 3.6. Let X be a manifold with fibred corners and Eφ a fibred decoupage
associated with X. The groupoid of X is the restriction to X of the groupoid G Eφ,
Γφ(X) = (G Eφ)X
X .
Remark 3.7. Each Vi splits M in two parts according to definition 2.6.
Remark 3.8. In the case of a manifold X with a connected fibred boundary φ :
∂X → Y , an expression of the groupoid of X as a set is:
Γφ(X) = (∂X ×Y ∂X ×Y T Y × R∗
+) ⊔ X ◦ × X ◦.
Monthubert-Pierrot [MP97] and Nistor-Weinstein-Xu [NWX99] showed how to
define a pseudodifferential calculus Ψ∞
c (G ) for any smooth differentiable longitudi-
nally smooth groupoid G . We now prove that G Eφ is longitudinally smooth, which
directly implies the longitudinal smoothness of Γφ(X). We can therefore associate
16
LAURENT GUILLAUME
to any manifold with fibred corners X a compact support calculus Ψ∞
c (Γφ(X)). The
properties of this calculus and the link with Melrose's φ-calculus will be studied in
the next section.
Proposition 3.9. The groupoid of a fibred decoupage is a Lie groupoid.
Demonstration. Denote ϕi = s ⊕ r : G Viφ → M 2. Let γ = (γi)i∈I be an element
of the fibered product G Eφ of the G Viφ. Then there exists a subset J of I such that
(γi)i∈I ∈ Yi∈J
G ViφViYi6∈J
G Viφ(M\Vi ).
i : G ViφVi → M 2, i ∈
It is thus enough to show the transversality of the morphisms ϕ′
J, which are the canonical projections, and φ′
i : G Viφ(M\Vi) → M 2 which are the
inclusions. The normal bundles of the latter being trivial, it suffices to consider
that i ∈ J. But the canonical projections factorize through the submersion ψi :
G ViφVi → Vi ×Yi Vi: if ii is the inclusion of Vi ×Yi Vi in M 2, then ϕ′
i = ii ◦ ψi. The
transversality of the morphisms ii is implied by the definition of a fibred decoupage
and the surjectivity of the differentials dψi then shows that the morphisms ϕ′
i, i ∈ J,
and consequently ϕi, are transverse.
Thus G Eφ is a submanifold ofQi∈I
law induced by the inclusion, it is a Lie groupoid.
G Viφ naturally endowed with the composition
3.2.2. Longitudinal smoothness of the groupoid.
Proposition 3.10. The groupoid of a fibred decoupage is longitudinally smooth.
Demonstration. Let x ∈ M and F be the open face including x. Let J ⊂ I such
φ of G Eφ in x is composed as the
that F is the interior of Ti∈J Vj. The fibre G E x
fibered product :
Yi∈J
G V x
iφViYi6∈J
G V x
iφ(M\Vi ).
(cid:3)
(cid:3)
G V x
iφ(M\Vi) = ∅ when i 6∈ J, as x is an element of Ti∈J Vj. Besides when i ∈ J,
. The transversality of the maps
iφ then implies the smoothness of the fibered
iφVi
G V x
is the smooth vector bundle N
s ⊕ r : G Viφ → M 2 restricted to G V x
Vi×Vi×R∗
+
Vi×Yi {x}
and therefore that of the fibre G E x
φ .
product Qi∈J
G V x
iφVi
3.2.3. Amenability of the groupoid.
Proposition 3.11. The groupoid of a fibred decoupage is amenable.
Demonstration. Recall (see [Ren80]) that if G is a groupoid and U an open set of
G , then G is amenable if and only if GU and GX\U are amenable. Also a groupoid
such that ϕ = s ⊕ r : G → (G (0))2 is surjective and open is amenable if and only if
its isotropy subgroups are amenable.
Since then G Vφ is amenable:
the isotropy of G VφM\V = (M \ V )2 is trivial
and G VφV has isotropy groups isomorphic to TyY × R, which is abelian and thus
amenable.
G Eφ is therefore also amenable as the fibered product of amenable groupoids.
(cid:3)
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
17
4. Pseudodifferential calculus on manifolds with fibred corners
We briefly recall the definition of φ-calculus, a particular pseudodifferential calcu-
lus introduced by Mazzeo and Melrose on manifolds with fibred boundaries [MM98].
It is a generalization of the b-calculus, built from a geometric desingularization of
the manifold with corner X 2 and the fibrations of its boundaries.
We then associate to any manifold with fibred corners X a pseudodifferential
calculus with compact support Ψ∞
c (Γφ(X)) from the groupoid Γφ(X) built in sect-
ion 2 and we show that φ-calculus identifies with the pseudodifferential calculus on
this groupoid in the case of manifolds with fibred boundary.
Finally we introduce a Schwartz type algebra Sψ(Γφ(X)) which allows to define
an extended calculus Ψ∞(Γφ(X)) = Ψ∞
c (Γφ(X)) + Sψ(Γφ(X)). We show that
Sψ(Γφ(X)) is stable under holomorphic functionnal calculus and that the operators
of extended φ-calculus are elements of this algebra.
4.1. b-stretched product and b-calculus. Following the work of Hormander who
showed that pseudodifferential calculus on a manifold M is given by Schwartz kernel
on M 2, Melrose defined b-calculus on manifolds with boundary using Schwartz
kernels on a space playing the role of M 2, the "b-stretched product".
Let X be a compact smooth manifold with boundary. The b-stretched product
of X ([Mel93]) is a compact smooth manifold with corners defined by the gluing
X 2
b = X 2 \ (∂X)2 ∪ S+N,
where S+N = (N X 2
+ is a fibred space which can be made trivial under
the form (∂X)2 × [−1, 1] by choosing a definition function ρ of ∂X. The embedding
of X 2 \ (∂X)2 in X 2 × [−1, 1] by
(∂X)2 − {0})/R∗
(x, y) → (x, y,
ρ(x) − ρ(y)
ρ(x) + ρ(y)
)
allows the identification of X 2
embedding in X × X × R is given by :
b with a submanifold of X 2 × [−1, 1]. Another possible
X 2
b = {(x, y, t) ∈ X × X × R, (1 − t)ρ(x) = (1 + t)ρ(y)} .
The b-stretched product has three boundary components, lb = {(x, y, −1) ∈ ∂X × X × R} ≃
∂X × X, rb = {(x, y, 1) ∈ X × ∂X × R} ≃ X × ∂X and S+N .
The diagonal ∆b = {(x, x, 0) ∈ X × X × R} only intersects the boundary com-
ponent S+N and this intersection is transverse, which allows a direct definition of
operators with Schwartz kernels on X 2
b , the operators of b-calculus (see [Mel93]).
The kernel K of a b-pseudodifferential operator is an element of I∞(X 2
2 )
with Taylor series development vanishing on lb ∪ rb (K is a C∞ function in the
neighbourhood of lb ∪ rb as the intersection of lb ∪ rb with ∆b is empty). The
situation for X = R+ is illustrated on figure 4.1.
b , ∆b, Ω
1
4.2. φ-stretched product and φ-calculus. When the boundary of X is the total
space of a fibration φ : ∂X → Y , the construction of φ-calculus not only needs
transverse intersection of the diagonal ∆b with the S+N component but also with
the submanifold Φ of S+N defined by
Φ = {(x, y, 0) ∈ S+N ≃ (∂X)2 × [−1, 1], φ(x) = φ(y)}.
18
LAURENT GUILLAUME
X 2
b
lb
∆b
S+N
rb
Figure 2. The b-stretched product of R+.
This condition is satisfied by introducing the φ-stretched product [MM98], which
is obtained as a gluing
X 2
φ = X 2
b \ Φ ∪ S+Nφ
where S+Nφ = (N X 2
+ is a fibred space which can be made locally trivial
under the choice of a definition function ρ of ∂X and a local trivialization φ of φ
over ∂X.
Φ − {0})/R∗
b
Boundary components of the φ-stretched product are lb, rb, S+Nφ and sb, the
closure of S+N \ Φ in X 2
φ \ lb ∪ rb ∪ sb is shown to be
diffeomorphic to the subset of elements (x, x′, S, Y ) ∈ X × X × R × Rq such that:
φ. When dim Y = q, X 2
ρ(x′) = ρ(x)(1 + Sρ(x)) and φ(x) = φ(x′) − ρ(x)Y
The diagonal ∆φ = {(x, x) ∈ X 2
φ} only intersects the boundary component S+Nφ
and this intersection is transverse which allows a direct definition of operators with
Schwartz kernels on X 2
φ, the operators of φ-calculus (see [MM98]). The definition
is similar to the one of previous paragraph. The kernel K of a φ-pseudodifferential
1
operator is an element of I∞(X 2
2 ) with Taylor series development vanishing
on lb ∪ rb ∪ sb (K is a C∞ function in the neighbourhood of lb ∪ rb ∪ sb as the
intersection of lb∪rb∪sb with ∆φ is empty). The situation for X = R+ is illustrated
on figure 4.2.
φ, ∆φ, Ω
X 2
φ
lb
∆φ
S+Nφ
sb
rb
Figure 3. The φ-stretched product of R+ with the trivial fibration.
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
19
4.3. φ-calculus and the groupoid of manifolds with fibred boundary. From
the previous discussions it appears that two geometric objects model calculus on
manifolds with fibred boundary : on the one hand φ-stretched product X 2
φ and the
associated φ-calculus, on the other hand the groupoid Γφ(X) = (G Eφ)X
X and its
pseudodifferential calculus Ψ∞
c (Γφ(X)).
We show here that the groupoid Γφ(X) of a manifold with fibred boundary can
be embedded in the φ-stretched product X 2
φ \lb∪rb∪sb
and that through this identification φ-calculus coincides with the pseudodifferential
calculus Ψ∞
φ as the open submanifold X 2
c (Γφ(X)).
4.3.1. Notations. Let X be a manifold with fibred boundary φ : ∂X → Y , p =
dim ∂X −dim Y and q = dim Y . Denote ∂Xj the connected components of ∂X with
induced fibrations φj : ∂Xj → Yj, and Eφ = (M, (∂Xj, φj )j∈J ) the fibred decoupage
which positive part defines X and its fibration. As the connected components of
∂Xj are disjoint, the puff G Eφ of Eφ coincides with the union ∪j∈J G ∂Xjφ of the
puff groupoids of ∂Xj in M . It is given below an explicit C∞ atlas of G Eφ on which
the embedding is defined.
Let {(Uα = Wα × R∗
+ which charts are slices of each Hol(φj) × R∗
+, ϕα = ψα × log)} be a C∞ atlas of Sj∈J ∂Xj × ∂Xj ×
R∗
+. For a
given U = Uα, U = ϕα(Uα) is an open set of R2p+q × Rq × R, a slice of V =
U ∩ (R2p+q × {0} × {0}) and the composition
+ = ∂Xj ×Yj ∂Xj × R∗
D Uα
Vα
ϕα−→ D U
V
Θ−1
−→ ΩU
V
is a diffeomorphism of D U
(see also section 1.4.2):
V on the open set ΩU
V of R2p+q × Rq × R × R defined by
ΩU
V = {(x, ξ, λ, t) ∈ R2p+q × Rq × R × R , (x, tξ, t · eλ) ∈ U }.
4.3.2. Embedding of Γφ(X) in X 2
φ. Let F : U → R2p+q × Rq × R be the C∞ map
defined by (x, ξ, λ) 7→ (x, ξ, eλ − 1). Let denote U ′ = F (U ) and V ′ = F (V ). The
relation F (x, 0, 0) = (x, 0, 0) implies the map F : ΩU
V ′ (see 1.4.2) to be a
diffeomorphism.
V → ΩU ′
Then consider the map Θα : D Uα
Vα
→ R2p+q × Rq × R defined by the composition
D Uα
Vα
Θ−1◦ ϕα−→ ΩUα
Vα
F−→ ΩU ′
α
V ′
α
and its projection πΘα : D Uα
→ Rq × R over the last two factors. If s and r denote
Vα
the source and target maps of Γφ(X) we define is : Γφ(X) → X × X × Rq × R for
γ ∈ iD (D Uα
Vα
) by :
Proposition 4.1. The map is is an embedding of Γφ(X) in X 2
φ.
is(γ) =(cid:0)s(γ), r(γ), πΘα ◦ i−1
D (γ)(cid:1)
Demonstration. Let denote φ : Wα → Rq the projection of ψα on the Rq factor
of ψα(Wα) ⊂ R2p+q × Rq. The expression 3.1.4 of i−1
D for an element (x, y) of
X ◦ × X ◦ shows that
ϕα ◦ i−1
D (x, y) = (ψα(π(x), π(y)), log
ρ(y)
ρ(x)
, ρ(x)).
20
LAURENT GUILLAUME
The relation
F (ψα(π(x), π(y)), log
then implies
ρ(y)
ρ(x)
) =(cid:18)ψα(π(x), π(y)),
ρ(y) − ρ(x)
ρ(x)
(cid:19)
πΘα ◦ i−1
D (x, y) = φ(π(x), π(y))
ρ(x)
,
ρ(y) − ρ(x)
ρ2(x) ! .
Now X 2
φ \ lb ∪ rb ∪ sb is the submanifold of X × X × Rq × R with elements
φ(x) = φ(y) − ρ(x)Y. The
thus belongs
(x, y, Y, S) such that ρ(y) = ρ(x)(1 + Sρ(x)) and
element is(x, y) = (x, y, Y, S) where Y =
to X 2
and S = ρ(y)−ρ(x)
φ(π(x),π(y))
φ \ lb ∪ rb ∪ sb as :
ρ2(x)
ρ(x)
and
ρ(x)(1 + Sρ(x)) = ρ(x)(1 +
ρ(y) − ρ(x)
ρ(x)
) = ρ(y)
φ(y) − ρ(x)Y = φ(y) − φ(π(x), π(y)) = φ(x).
The trivial holonomy of fibrations implies the injectivity of source and target
maps of Γφ(X) and the injectivity of is restricted to the interior of Γφ(X). Moreover
the fiber in 0 of a deformation D Uα
V 0 = {(x, ξ, λ, 0) ∈
Vα
R2p+q × Rq × R∗
D when
+ × {0} , x ∈ V }, which proves the surjectivity of πΘα ◦ i−1
is diffeomorphic to the set ΩU
restricted to the boundary Sα
Therefore is(Γφ(X)) ⊂ X 2 is in bijection with X 2
φ \ lb ∪ rb ∪ sb, the definition of
is as a composition of smooth maps allows to conclude that is is an embedding of
Γφ(X) in X 2
φ, with image X 2
φ \ lb ∪ rb ∪ sb.
D Uα
Vα0 of Γφ(X).
(cid:3)
4.3.3. Identification of Γφ(X) in X 2
φ.
Proposition 4.2. Γφ(X) is an open submanifold of X 2
φ and
X 2
φ \ Γφ(X) = lb ∪ rb ∪ sb.
Demonstration. The (non-disjoint) boundary components of X 2
φ are
∂X 2
φ = lb ∪ rb ∪ sb ∪ S+Nφ.
The relation is(Γφ(X)) = X 2
then proves that the closed submanifold lb ∪ rb ∪ sb is the complementary in X 2
the image of the embedding is.
φ \ lb ∪ rb ∪ sb established in the previous demonstration
φ of
(cid:3)
c (Γφ(X)) with φ-calculus. Let denote Ψ∞
4.4. Identification of Ψ∞
c (Γφ(X)) the
algebra of operators with compact support on Γφ(X) and Ψ∞
φ,c(X) the algebra of
operators with compact support of φ-calculus. The submanifold lb ∪ rb ∪ sb is
disjoint from a neighbourhood of ∆φ. Pseudodifferential calculus with compact
support on Γφ(X) thus coincides with distributional sections on X 2
φ which vanish
in a neighbourhood of lb ∪ rb ∪ sb. One recovers the definition of φ-calculus :
Theorem 4.3. The pseudodifferential calculus on Ψ∞
small φ-calculus with compact support Ψ∞
φ,c(X) of Melrose.
c (Γφ(X)) coincides with the
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
21
4.5. Extended pseudodifferential calculus. We define an extended pseudodif-
ferential calculus Ψ∞(Γφ(X)) over the groupoid Γφ(X) of a manifold with fibred
corners :
Ψ∞(Γφ(X)) = Ψ∞
c (Γφ(X)) + Sψ(Γφ(X)).
The algebra Sψ(Γφ(X)) of functions with rapid decay built from a length func-
tion ψ with polynomial growth (lemma 4.5) is naturally stable under holomorphic
functionnal calculus. It includes the regularizing operators of φ-calculus (proposi-
tion 4.8) and the extended φ-calcul satisfies the inclusion relation :
Ψ∞
φ (X) ⊂ Ψ∞(Γφ(X)).
4.5.1. Length functions with polynomial growth.
Definition 4.4 ([Mon03], 1.4). Let G be a Lie groupoid and µ a Haar system on
G . A length function with polynomial growth is a C∞ function ϕ : G → R+ such
that :
· ϕ is subadditive, i.e. ϕ(γ1γ2) ≤ ϕ(γ1) + ϕ(γ2),
· ∀γ ∈ G, ϕ(γ−1) = ϕ(γ),
· ϕ is proper,
· ∃c, N, ∀x ∈ G(0), ∀r ∈ R+, µx(ϕ−1([0, r])) ≤ c(rN + 1).
With such a function ϕ one can define the space
S0 = {f ∈ C0(G , Ω1/2), ∀P ∈ C[X], sup
γ∈G
P (ϕ(γ))f (γ) < ∞}
where the half-densities bundle Ω
is the vector space of applications
1
2 is the line bundle over G which fibre for γ ∈ G
ρ : ΛkTγ G r(γ) ⊗ ΛkTγ Gs(γ) → C
such that ρ(λν) = λ1/2ρ(ν), ∀λ ∈ R.
The space Sϕ(G) of functions with rapid decay over G is defined in ([Mon03],
1.5) as the subspace of S0 of functions f such that :
∀l ∈ N, ∀(v1, . . . , vl) ∈ C∞(AG)l, ∀k ≤ l, (v1 . . . vk · f · vk+1 . . . vl) ∈ S0.
Sϕ(G) is the ideal of regularizing operators of the extended pseudodifferential
c (G, G(0); Ω1/2) + Sϕ(G). It is also a subalgebra of
calculus I∞(G, G(0); Ω1/2) = I∞
C∗(G ) stable under holomorphic functionnal calculus ([LMN05], theorem 6).
4.5.2. Length function for manifolds with fibred corners. Let X be a manifold with
embedded and fibred corners defined by a fibred decoupage Eφ = (M, (Vi, φi)i∈[1,N ]).
Each puff groupoid G Viφ is diffeomorphic through the embedding is of proposition
4.1 to a closed submanifold of M × M × Rqi × R. The positive part of their fibered
product over M × M is thus diffeomorphic (through a diffeomorphism i) to a closed
Rqi × RN →
Rqi × RN be the canonical projection, we define a length function on G Eφ
Rqi × RN . Let π : X × X ×Qi∈[1,N ]
submanifold of X × X ×Qi∈[1,N ]
Qi∈[1,N ]
by ψ(γ) = kπ ◦ i(γ)k.
Lemma 4.5. ψ : G Eφ → R+ is a length function with polynomial growth.
22
LAURENT GUILLAUME
Demonstration. On every chart iD (D Uα
Vα
D (γ)k satisfies ψi(γ) = kξk + kλk where (x, ξ, λ, t) = Θ−1 ◦ φ ◦ i−1
i−1
) ⊂ G Viφ of proposition 4.1, ψi = kπΘ ◦
D (γ). Each ψi is
therefore a groupoid morphism and ψ =Pi∈[1,N ] ψi is subadditive.
π is proper as X is compact, besides i(Γφ(X)) is closed in X × X ×Qi∈[1,N ]
(R)N so ψ = kπ ◦ ik is proper.
Finally for x ∈ X the local expresion of the ψi shows that there exists two
constants c1, c2 such that for r ≥ 1, µx(ψ−1([0, r])) < c1 vol(BRM (0, r)) where
Rqi ×
M = N +Pi qi and for r < 1, µx(ψ−1([0, r])) ≤ supx∈X µx(ψ−1([0, 1])) = c2.
By denoting c = max(c1, c2) we get µx(ψ−1([0, r])) ≤ c(rM + 1). As X is compact
it can be covered with a finite number of neighbourhoods which give an inequality
on X.
Hence ψ has polyomial growth.
(cid:3)
4.5.3. Definition of the extended pseudodifferential calculus. We are now able to
define an extended pseudodifferential calculus on manifolds with fibred corners:
Definition 4.6. Let X be a manifold with fibred corners and Γφ(X) the associated
longitudinally smooth groupoid. The extended pseudodifferential calculus on X is
the algebra Ψ∞(Γφ(X)) of pseudodifferential operators on Γφ(X) defined by:
Ψ∞(Γφ(X)) = Ψ∞
c (Γφ(X)) + Sψ(Γφ(X))
where Sψ(Γφ(X)) = Sψ((G Eφ)X
respect to the length function ψ of previous lemma.
X ) is the space of functions with rapid decay with
Sψ(Γφ(X)) is the ideal of regularizing operators of the extended pseudodifferen-
tial calculus Ψ∞(Γφ(X)).
Remark 4.7. The definition of Sψ(Γφ(X)) given here is based on the existence
of a length function with polynomial growth. In more complex cases of foliated
boundary no such functions exist in general, for instance in the case of nonzero
entropy foliations. The problem is overcome in section 3.5 of [Gui12] with the
definition of a Schwartz algebra Sc(Γφ(X)) valid for any regular foliation of the
boundary. For the sake of clearness only the fibration case is developped in this
article.
4.5.4. Identification of Ψ∞(Γφ(X)) and Ψ∞
φ (X). The regularizing operators of φ-
calculus are C∞ kernels on X 2
φ which Taylor series vanish on lb ∪ rb ∪ sb. Melrose
shows in proposition 4 of [MM98] that a kernel K(x, x′) which vanishes in Taylor
series with respect to the powers of ρ(x), φ(x) is a rapid decay kernel with respect
, that is with respect to πΘ ◦
to the variables Y =
i−1
D (x, x′).
and S = ρ(x)−ρ(x′)
φ(x)− φ(x′)
ρ(x)
ρ(x)
Those kernels are thus a fortiori functions with rapid decay with respect to
ψ, so that any regularizing operator of φ-calculus defines by restriction to the
complementary of lb∪rb∪sb a function with rapid decay on Γφ(X) and Ψ−∞
(X) ⊂
Sψ(Γφ(X)).
φ
We get as a corollary the following result:
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
23
Theorem 4.8. The extended pseudodifferential calculus Ψ∞(Γφ(X)) = Ψ∞
Sψ(Γφ(X)) contains the operators of φ-calculus :
c (Γφ(X))+
Ψ∞
φ (X) ⊂ Ψ∞(Γφ(X))
4.5.5. Identification of the boundary restriction with Melrose's normal operator.
Contrary to the case of a smooth manifold without corners, regularizing operators
are not necessarly compact. The obstruction to compacity for φ-calculus is de-
scribed by introducing an indicial algebra I (A, ξ, λ) ∈ Ψsus(φ)(∂X) and a normal
operator Nφ : Ψ∞
φ (X) → Ψsus(φ)(∂X) with values in a suspended algebra of kernels
over ∂X which are invariant under translation and with rapid decay at infinity (see
[Mel93], [MM98]).
In the case of manifolds with fibred corners, the indicial family I (A, ξ, λ) is de-
fined by Melrose ([MM98], proposition 5) as the restriction of the kernel to f f (X 2
φ).
The normal operator Nφ of Melrose thus coincides with the operator ∂ of restriction
to the boundary of the groupoid Γφ(X). The translation invariance of the indicial
family seen as an element of ∂Sψ(Γφ(X)) is nothing but the invariance under T Y ×R
of a family of operators on Sψ(∂X ×Y ∂X ×Y T Y ×R) ≃ S(T Y ×R, C∞
c (∂X ×Y ∂X)).
4.6. Total ellipticity and Fredholm index. Let Ψ0(Γφ(X)) be the norm closure
of Ψ0
r (Γφ(X)). Let recall that the symbol
map σ induces the exact sequence:
c(Γφ(X)) in the multiplier algebra of C∗
0 → C∗
r (Γφ(X)) −→ Ψ0(Γφ(X)) σ−→ C0(S∗Γφ(X)) → 0
and that the analytic index inda : K 0(A∗Γφ(X)) → K0(C∗
r (Γφ(X))) takes values
in the K-theory of the C∗-algebra of the groupoid Γφ(X). The analytic index is
thus not a Fredholm index in general, its values are not elements of Z. To get
a Fredholm operator it is necessary to introduce additionnal conditions, classical
ellipticity (the property of a symbol to be invertible) not being sufficient.
The condition of total ellipticity is derived very naturally from the groupoid
∂X is a closed saturated set of Γφ(X) and the
approach. Indeed ∂Γφ = (Γφ(X))∂X
following diagram is commutative:
0
0
0
C∗
r (X ◦ × X ◦)
C∗
r (Γφ(X))
Ψ0(Γφ(X))
σ C0(S∗Γφ(X))
∂
C∗
r (∂Γφ)
Ψ0(∂Γφ)
C0(S∗∂Γφ)
0
0
0
0
The relation C∗
r (X ◦ × X ◦) = ker(σ) ∩ ker(∂) = ker(σ ⊕ ∂) shows that σ ⊕ ∂
factorizes through the map στ :
Ψ0(Γφ(X))
στ−→ C0(S∗Γφ(X)) ×∂X Ψ0(∂Γφ)
where C0(S∗Γφ(X)) ×∂X Ψ0(∂Γφ) denotes the fibered product of C0(S∗Γφ(X)) et
de Ψ0(∂Γφ) over C0(S∗∂Γφ).
24
LAURENT GUILLAUME
Definition 4.9 ([Gui12], 4.6). An operator P ∈ Ψ0(Γφ(X)) will be called totally
elliptic when the element στ (P ) is inversible.
The map στ induces from the previous diagram the exact sequence :
0 → C∗(X ◦ × X ◦) ≃ K → Ψ0(Γφ(X))
στ→ C0(S∗Γφ(X)) ×∂X Ψ0(∂Γφ)
Therefore an element in Ψ0(Γφ(X)) is invertible modulo compact operators if
an only if its image in C0(S∗Γφ(X)) ×∂X Ψ0(∂Γφ) is invertible. This last condition
is equivalent to the double condition of ellipticity for the operator and inversibility
for its bounday restriction.
In particular we get the property stated in the first part of [MR05] : an operator
of φ-calculus is totally elliptic when its symbol and the normal operator Nφ = ∂ are
jointly invertible, and the operator is Fredholm if and only if it is totally elliptic.
5. The holonomy groupoid of manifolds with fibred corners
In section 2 we constructed the puff groupoid G VF of a foliated submanifold
(V, F ) of codimension 1 in M. The module FM of vector fields tangent to F on V
defines a singular foliation on M. We prove here that G VF is the holonomy groupoid
Hol(M, FM ) of the singular foliation (M, FM ).
For that purpose we show the groupoid G VF integrates the Lie algebroid A FM
of FM and satisfies the minimality condition defining the groupoid Hol(M, FM ).
Notations from section 3.1.1 are reused.
5.1. The singular foliation of a fibred boundary. Consider the Lie algebroid
A FM = T M with anchor map p defined over each Ni by:
(fi)∗(p((fi)−1
∗ (v1, v2, t, λ))) = (v1 + tv2, t, tλ)
when (v1, v2, (t, λ)) ∈ T Fi ⊕ Ni × Tt(R), and by the identity over T (M \ N ).
A FM defines a foliation (M, FM ) with this map (remark 1.3). Let recall that a
foliation is almost regular when it is defined by a Lie algebroid with anchor map
injective over a dense open set.
Proposition 5.1. The foliation (M, FM ) is almost regular.
Demonstration. It is sufficient to show that the anchor map of A FM is injective
over the dense open set M \ V , which follows immediatly from the injectivity of the
map (t, λ) 7→ (t, tλ) for t 6= 0.
(cid:3)
A result by Claire Debord ([Deb01],Thm 4.3) valid for any almost regular folia-
tion then ensures the existence of a Lie groupoid Hol(M, FM ) which Lie algebroid
is A FM . It is a groupoid with units M which integrates the foliation defined by
A FM .
The immediate surjectivity of the map (t, λ) 7→ (t, tλ) for t 6= 0, and thus that of
the anchor map of A FM over M \ V implies the leaves of this foliation are M \ V
and the leaves of F . The groupoid G VF integrates this space of leaves and one can
ask if it is the only one. The following property answers that question.
A Lie algebroid A is said extremal for the foliation F if for any other Lie
algebroid A ′ defining F , there exists a Lie algebroid morphism from A ′ to A .
When the foliation F is almost regular, the extremality of A is equivalent ([Deb01])
to the existence of a unique "minimal" groupoid G integrating A , in the sense that
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
25
for any other Lie groupoid H defining F there exists a differentiable morphism of
groupoids from H to G. In that sense G VF can be defined as the "smallest" Lie
groupoid defining F . Indeed:
Proposition 5.2. A FM is extremal for (M, FM ).
Demonstration. It is sufficient to show that the anchor map p of A FM induces
an isomorphism between Γ(A FM ), the vector space of local sections of A FM and
Γ(T FM ), the vector space of local vector fields tangent to FM . The condition over
M \ V immediately results from the bijectivity of the anchor map pM\V . The
condition over V is obtained from the Taylor series development of a vector field
tangent on V to the leaves of the foliation F . If xi, yi, t denote local coordinates
associated to the directions respectively defined by T Fi, Ni and Tt(R), a vector field
ξ ∈ FM will be locally generated by the free family ht ∂
i, which ensures
the announced isomorphism.
∂t , ∂
∂xi
, t ∂
∂yi
5.2. The groupoid of a fibred decoupage as a holonomy groupoid.
Definition 5.3. We call holonomy groupoid of (M, FM ) the minimal Lie groupoid
Hol(M, FM ) with Lie algebroid A FM . The previous proposition shows that Hol(M, FM )
is generated from the atlas of bi-submersions near the identity of vector fields tan-
gent on V to the leaves of F (see [AS06]).
(cid:3)
Lemma 5.4. G VF is a semi-regular s-connected Lie groupoid.
Demonstration. The s-connected property of G VF is immediate by definition.
The groupoid Dϕ is endowed with the differential structure of a normal cone defor-
mation. The open sets of GV × R∗
V ) exp Z,
ΩU
V = {(x, x′, ξ, t) ∈ R2p × Rq × R × R , (x, x′, tξ) ∈ U } form a regular open sets
basis for the topology of Dϕ. Thus Dϕ is semi-regular and so is G VF according to
proposition 3.3.
+ × R∗ and the sets of the form Θ(ΩU
(cid:3)
Proposition 5.5 ([Deb01]). Let H be a semi-regular s-connected Lie groupoid.
Then H is a quasi-graphoid if and only if the set {x ∈ H (0) , H x
x = {x}} of units
with trivial isotropy group is dense in H (0).
Corollary 5.6. G VF is a quasi-graphoid.
Demonstration. It is immediate to check that the isotropy of G VF is trivial over
M ◦. G VF being s-connected and semi-regular (lemma 5.4), it is a quasi-graphoid
from proposition 5.5.
(cid:3)
Proposition 5.7. G VF is the holonomy groupoid Hol(M, FM ) of (M, FM ).
Demonstration. Let A and B be the atlas introduced in 3.1.4 to describe the
differential structure of G VF . A = {(Ωj
i =
iD (D j
i , φij )i,j∈I 2 } with φij = ϕij ◦ i−1
i ). B = {(Ωβ, ϕβ)} is an atlas of the manifold M ◦
+F M ◦
Over ΩB =S Ωβ source and target maps of G VF are defined by:
D and Ωj
(cid:26) s(x, y) = x
r(x, y) = y
+ × M ◦
− × M ◦
−.
26
LAURENT GUILLAUME
The algebroid A G VF restricted to M \N ⊂ ΩB is thus AM\N =S ker(ds{(x,x),x∈M\V }) =
T (M \ V ) endowed with the identity anchor map p : AM\N ◦ → T (M \ V ) as
drT (M\V ) = Id. So A G VF M\N = A FMM\N .
Besides the groupoid immersion ϕ : G1 → V × V induces an injective Lie al-
gebroid morphism ϕA
∗ : A1 → T V , with A1 = T F . The connection ω allows the
identification of T V with A1⊕N where N = T V /A1 and the anchor of the algebroid
Aϕ from the normal cone groupoid Dϕ is:
pD : Aϕ = A1 ⊕ N × R → T (V × R) = T V × T R
(v1, v2, t)
7→ (p2(v1 + tv2), (t, 0))
The composition law on Dϕ is given by the product of the law composition on Dϕ
and the multiplicative action of R∗
+ on R. Therefore A Dϕ = A1 ⊕ N ⊕ T R and the
anchor map pϕ of A Dϕ is obtained as:
pϕ : A = A1 ⊕ N × T R → T (V × R) = T V × T R
(v1, v2, t, λ)
7→ (p2(v1 + tv2), (t, t · exp λ))
The isomorphism di−1
then implies the equality pN = pϕ ◦ di−1
D : A G VF → A Dϕ induced by i−1
D and A G VF = A FMN .
D over ΩA = S Ωj
i = N
Thus G VF is a quasi-graphoid integrating the Lie algebroid A FM over M. A FM
being extremal, G VF is indeed the holonomy groupoid Hol(M, FM ) of the foliation
(M, FM ).
(cid:3)
Finally we get as a corollary a similar interpretation for the groupoid of a mani-
fold with fibred corners:
Corollary 5.8. Let X be a manifold with fibred corners defined by a fibred decoupage
Eφ = (M, (Vi, φi)i∈I ). Let FM be the module of vector fields tangent to the fibrations
of each face. Then G Eφ is the holonomy groupoid Hol(M, FM ) of (M, FM ).
Demonstration. Let Hol(M, FM ) be the holonomy groupoid of (M, FM ). From
previous proposition 5.7, for i, j ∈ I the minimality of G Viφ and G Vjφ implies the
existence of morphisms pi and pj such that the following diagram commutes:
Hol(M, FM )
pi
G Viφ
pj
G Vjφ
si⊕ri
sj ⊕rj
M 2
From the universal property of the fibered product G Eφ there thus exists a mor-
phism of groupoids u : Hol(M, FM ) → G Eφ. The minimality of Hol(M, FM ) then
implies u to be an isomorphism and G Eφ ≃ Hol(M, FM ) is the holonomy groupoid
of (M, FM ).
(cid:3)
The groupoid Γφ(X) of a manifold with fibred corners has therefore a natural
geometric meaning as a singular foliation holonomy groupoid, its construction is
independent of any particular description. It is an explicit example of a singular
leaf space in the sense of [AS06]. This result allows the conceptual interpretation of
φ-calculus as the pseudodifferential calculus associated with the holonomy groupoid
of the singular foliation defined by the manifold with fibred corners.
PSEUDODIFFERENTIAL CALCULUS ON MANIFOLDS WITH FIBRED CORNERS
27
References
[ALMP09] Pierre Albin, Eric Leichtnam, Rafe Mazzeo, and Paolo Piazza, The signature package
[AS63]
[AS68a]
[AS68b]
[AS06]
[AS09]
[BHS91]
[BP85]
[Con79]
[Con82]
[Con94]
[CR07]
[CS84]
[Deb01]
[DL09]
on witt spaces, i. index classes, arxiv:math.DG/09061568 (2009).
M. F. Atiyah and I. M. Singer, The index of elliptic operators on compact manifolds,
Bull. Amer. Math. Soc. 69 (1963), 422–433.
, The index of elliptic operators. I, Ann. of Math. (2) 87 (1968), 484–530.
, The index of elliptic operators. III, Ann. of Math. (2) 87 (1968), 546–604.
Iakovos Androulidakis and Georges Skandalis, The holonomy groupoid of a singular
foliation, arxiv:math.DG/0612370 (2006).
I. Androulidakis and G. Skandalis, Pseudodifferential calculus on a singular foliation,
ArXiv e-prints (2009).
J.-P. Brasselet, G. Hector, and M. Saralegi, Th´eor`eme de de rham pour les vari´et´es
stratifi´ees, Ann. Global Analy. Geom. 9 (1991), no. 3, 211–243.
B Bigonnet and Jean Pradines, Graphe d'un feuilletage singulier, C.R. cad. Sci. Paris
300 (1985), no. 13, 439–442.
Alain Connes, Sur la th´eorie non commutative de l'int´egration, Alg`ebres d'op´erateurs
(S´em., Les Plans-sur-Bex, 1978), Lecture Notes in Math., vol. 725, Springer, Berlin,
1979, pp. 19–143.
, A survey of foliations and operator algebras, Operator algebras and applica-
tions, Part I (Kingston, Ont., 1980), Proc. Sympos. Pure Math., vol. 38, Amer. Math.
Soc., Providence, R.I., 1982, pp. 521–628.
, Noncommutative geometry, Academic Press Inc., San Diego, CA, 1994.
Paulo Carrillo-Rouse, Indices analytiques `a support compact pour des groupoides de
Lie, Th`ese de Doctorat `a l'Universit´e de Paris 7 (2007).
Alain Connes and Georges Skandalis, The longitudinal index theorem for foliations,
Publ. Res. Inst. Math. Sci. 20 (1984), no. 6, 1139–1183.
Claire Debord, Holonomy groupoids of singular foliations, J. Differential Geom. 58
(2001), no. 3, 467–500.
Claire Debord and Jean-Marie Lescure, K-duality for stratified pseudomanifolds, Ge-
ometry and Topology 13 (2009), 49–86.
[DLN06] Claire Debord, Jean-Marie Lescure, and Victor Nistor, Groupoids and an index theo-
rem for conical pseudomanifolds, arxiv:math.OA/0609438 (2006).
[DLR11] C. Debord, J.-M. Lescure, and F. Rochon, Pseudodifferential operators on manifolds
[Gui12]
[Hae84]
[HS87]
with fibred corners, ArXiv e-prints (2011).
Laurent Guillaume, G´eom´etrie non-commutative et calcul pseudodiff´erentiel sur les
vari´et´es `a coins fibr´es, Ph.D. thesis, Universit´e Paul Sabatier Toulouse 3, 2012.
Andr´e Haefliger, Groupoıdes d'holonomie et classifiants, Ast´erisque (1984), no. 116,
70–97, Transversal structure of foliations (Toulouse, 1982).
Michel Hilsum and Georges Skandalis, Morphismes K-orient´es d'espaces de feuilles et
fonctorialit´e en th´eorie de Kasparov (d'apr`es une conjecture d'A. Connes), Ann. Sci.
´Ecole Norm. Sup. (4) 20 (1987), no. 3, 325–390.
[LMN00] Robert Lauter, Bertrand Monthubert, and Victor Nistor, Pseudodifferential analysis
on continuous family groupoids, Doc. Math. 5 (2000), 625–655 (electronic).
[Mat70]
[Mel93]
[LMN05] Robert Lauter, Bertrand Monthubert, and Victor Nistor, Spectral invariance for cer-
tain algebras of pseudodifferential operators, Journal of the Institute of Mathematics
of Jussieu 4 (2005), no. 03, 405–442.
J.N. Mather, Notes on topological stability, Mimeographed Notes, 1970.
Richard B. Melrose, The Atiyah-Patodi-Singer index theorem, Research Notes in
Mathematics, vol. 4, A K Peters Ltd., Wellesley, MA, 1993.
Rafe Mazzeo and Richard B. Melrose, Pseudodifferential operators on manifolds with
fibred boundaries, arxiv:math.DG/9812120 (1998).
Bertrand Monthubert, Groupoids and pseudodifferential calculus on manifolds with
corners, J. Funct. Anal. 199 (2003), no. 1, 243–286.
Bertrand Monthubert and Fran¸cois Pierrot, Indice analytique et groupoıdes de Lie, C.
R. Acad. Sci. Paris S´er. I Math. 325 (1997), no. 2, 193–198.
R. B. Melrose and F. Rochon, Index in K-theory for families of fibred cusp operators,
ArXiv Mathematics e-prints (2005).
[Mon03]
[MM98]
[MP97]
[MR05]
28
LAURENT GUILLAUME
[NWX99] Victor Nistor, Alan Weinstein, and Ping Xu, Pseudodifferential operators on differen-
[Ren80]
[Tho69]
[Ver84]
[Win83]
tial groupoids, Pacific J. Math. 189 (1999), no. 1, 117–152.
Jean Renault, A groupoid approach to C ∗-algebras, Lecture Notes in Mathematics,
vol. 793, Springer, Berlin, 1980.
Ren´e Thom, Ensembles et morphismes stratifi´es, Bull. Amer. Math. Soc. 75 (1969),
240–284.
Andrei Verona, Stratified mappings-structure and triangulability, Lecture Notes in
Mathematics 1102 (1984).
H. E. Winkelnkemper, The graph of a foliation, Ann. Global Anal. Geom. 1 (1983),
no. 3, 51–75.
|
1305.3470 | 2 | 1305 | 2013-05-23T12:49:47 | Random matrix model for free Meixner laws | [
"math.OA",
"math.PR"
] | Applying the concept of matricial freeness which generalizes freeness in free probability, we have recently studied asymptotic joint distributions of symmetric blocks of Gaussian random matrices (Gaussian Symmetric Block Ensemble). This approach gives a block refinement of the fundamental result of Voiculescu on asymptotic freeness of independent Gaussian random matrices. In this paper, we show that this framework is natural for constructing a random matrix model for free Meixner laws. We also demonstrate that the ensemble of independent matrices of this type is asymptotically conditionally free with respect to the pair of partial traces. | math.OA | math | RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
ROMUALD LENCZEWSKI
Abstract. Applying the concept of matricial freeness which generalizes freeness in
free probability we have recently studied asymptotic joint distributions of symmetric
blocks of Gaussian random matrices (Gaussian Symmetric Block Ensemble). This
approach gives a block refinement of the fundamental result of Voiculescu on asymp-
totic freeness of independent Gaussian random matrices. In this paper, we show that
this framework is natural for constructing a random matrix model for free Meixner
laws. We also demonstrate that the ensemble of independent matrices of this type is
asymptotically conditionally free with respect to the pair of partial traces.
3
1
0
2
y
a
M
3
2
]
.
A
O
h
t
a
m
[
2
v
0
7
4
3
.
5
0
3
1
:
v
i
X
r
a
1. Introduction
It is well-known that free probability is an effective tool in the study of random
matrices and their asymptotics. This approach was originated by Voiculescu in his
fundamental paper [17], where he showed that independent Gaussian random matrices
are asymptotically free (generalized to non-Gaussian entries by Dykema [9]). His result
showed that the semicircle law obtained by Wigner [19] as the limit distribution of
certain symmetric random matrices can now be viewed as an element of a much more
general probability theory involving operator algebras [16].
If the complex-valued Gaussian variables which are entries of the considered random
matrices are not identically distributed, one has to apply a more general scheme to study
their asymptotics. One approach is to use operator-valued states and the associated
notion of freeness with amalgamation, as in the paper of Shlakhtyenko on Gaussian
band matrices [15]. This approach was further developed by Benayach-Georges [5]
who described the asymptotics of blocks of random matrices and introduced a related
additive convolution. Recently, we studied asymptotic joint distributions of symmetric
blocks of random matrices by means of operatorial methods on Hilbert spaces. For
this purpose, we employed a scheme based on arrays of scalar-valued states and the
associated concept of matricial freeness introduced in [10].
In particular, we showed in [11,12] that the symmetric blocks of an ensemble of n n
complex Hermitian Gaussian random matrices Y pu, nq with block-identical variances
converge in moments under normailzed partial traces to the mixed moments of sym-
metrized Gaussian operators, namely
Tp,qpu, nq Ñ pωp,qpuq
where u P U and 1 ď p ď q ď r, with U being an index set enumerating independent
matrices. The operators pωp,qpuq are natural symmetrizations of square arrays pωp,qpuqq
of matricially free Gaussian operators playing the role of basic Gaussian operators. By
2010 Mathematics Subject Classification: 46L53, 46L54, 15B52
Key words and phrases: free probability, freeness, matricial freeness, random matrix, free Meixner law
1
2
R. LENCZEWSKI
a partial trace we understand a normalized trace over the subset of basis vectors related
to diagonal blocks.
In the random matrix context, the corresponding framework is thus a block refine-
ment of that used by Voiculescu and is closely related to his idea of decomposition of
Gaussian random matrices leading to semicircular and circular systems [18]. We studied
a deformation of this decomposition based on allowing the Gaussian variables to have
block-identical variances rather than identical and then computing their mixed moments
under (normalized) partial traces rather than under the (normalized) complete trace.
We would also like to remark that some results obtained by our methods can perhaps
be suitably reformulated in terms of freeness with amalgamation.
The key parameters of the block refinement are given by r r symmetric variance
matrices V puq " pvp,qpuqq associated with symmetric blocks of the matrices Y pu, nq,
which, in turn are defined by the partition of the set
into r disjoint intervals (they depend on n, but this is supressed in the notation), and
by the dimension matrix
rns " N1 Y N2 Y . . . Y Nr
whose entries are given by non-negative numbers
Nj
n
dj " lim
D " diagpd1, d2, . . . , drq,
nÑ8
called asymptotic dimensions. An important assumption is that we allow some of these
dimensions to vanish. Note that in our first paper, where we presented the block model
[11], we assumed that all asymptotic dimensions are positive.
It follows from the asymptotics of random symmetric blocks that the parameters of
random matrices are encoded in the products of the dimension matrix and the variance
matrices, namely
Bpuq " DV puq
and these matrices provide constants associated with blocks of colored non-crossing pair
partitions underlying the combinatorics of mixed moments of symmetrized Gaussian
operators. Let us add that we take the same dimension matrix for all random matrices.
In comparison with freeness of free probability, matricial freeness gives more flexibility
in treating such problems of random matrix theory as
(1) evaluating limit distributions of random matrices,
(2) studying asymptotic properties of random matrix ensembles,
(3) constructing random matrix models for given probability measures,
and in that respect it reminds freeness with amalgamation. Some advantage of our
approach is that we rely on operators living in Hilbert spaces. This seems quite in-
tutive especially since computations involve operators which remind free creation and
annihilation operators and therefore their moments can be easily expressed in terms
of non-crossing (pair) partitions. A sample of such computations is contained in this
paper.
In particular, this flexibility allows us to treat sums and products of rectangular ran-
dom matrices in a unified manner, including Wishart matrices [20] as well as more gen-
eral products like those leading to free Bessel laws [4] and free products of Marchenko-
In fact, we were able to
Pastur [14] distributions with arbitrary shape parameters.
RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
3
compute the moments of the latter in the explicit form (known only in very special
cases before) and introduce polynomials which can be viewed as multivariate Narayana
polynomials [13]. A number of other new applications to the random matrix theory can
be given. In particular, the matricially free Gaussian operators turned out to be effec-
tive in the construction of random matrix models for boolean independence, monotone
independence and s-freeness [12]. In this paper, we also use these operators to construct
a simple random matrix model for an important class of probability measures on the
real line called free Meixner laws and prove the asymptotic conditional independence
of the associated ensembles of random matrices.
Free Meixner systems of polynomials and the associated family of functionals were
introduced and studied by Anshelevich [2,3]. Let us remark that free Meixner laws
are free analogs of classical Meixner laws. In particular, up to affine transformations,
they belong to one of the following six classes: free Gaussian (Wigner semicircle), free
Poisson (Marchenko-Pastur), free negative binomial (free Pascal), free Gamma, free
binomial and free hyperbolic secant, following the terminology of Anshelevich. Free
Meixner laws turn out to display similar properties with respect to free independence
as do the classical Meixner laws with respect to classical independence as Bryc and
Bozejko showed in their study of the regression problem [6].
Random matrix models for certain special free Meixner laws are well-known, like
the Gaussian Unitary Ensemble for the semicircle law, the Wishart Ensemble for the
Marchenko-Pastur law or the Jacobi Ensemble for the free binomial law (see, for in-
stance, [8,17,20]). However, a natural model for the whole class of free Meixner laws
has not been given in the literature.
The paper is organized as follows. In Section 2, we recall a combinatorial formula for
the moments of free Meixner laws. An operatorial realization of their moments in terms
of matricially free Gaussian operators is proved in Section 3. A random matrix model
for free Meixner laws is constructed in Section 4. An ensemble of independent random
matrices of this type, called the Free Meixner Ensemble, is considered in Section 5,
where we prove its asymptotic conditional freeness.
2. Moments of free Meixner laws
It is well-known that every probability measure on the real line with finite moments
of all orders is characterized by two sequences of Jacobi parameters
α " pα1, α2, . . .q and β " pβ1, β2, . . .q,
where αn P R and βn ě 0 for all n P N Y t0u, with the condition that if βk " 0 for
some k, then βm " 0 for all m ą k. We will call them Jacobi sequences and we will
use the notation Jpµq " pα, ωq. The Cauchy transform of µ can then be expressed as a
continued fraction of the form
Gµpzq "
z ´ α1 ´
1
β1
z ´ α2 ´
β2
z ´ α3 ´
β3
. . .
and it is understood that if βm " 0 for some m, then the fraction terminates and, for
convenience, we set βn " αn " 0 for all n ą m.
4
R. LENCZEWSKI
This continued fraction representation of Cauchy transforms turns out useful in our
approach. Thus, let us first remark that the family of free Meixner laws is the family
of probability measures on the real line associated with the pair of Jacobi sequences of
the form
α " pα1, α2, α2, . . .q and β " pβ1, β2, β2, . . .q,
i.e. they are constant starting from the second level of the corresponding continued
fractions. If a free Meixner law corresponds to the pair of Jacobi sequences of the above
form, we will say that it corresponds to pα1, α2, β1, β2q. In particular, if α1 " 0 and
β1 " 1, we obtain the standard free Meixner laws with mean zero and variance one. In
that case, the absolutely continuous part of the associated measure µ takes the form
dµpxq " a4β2 ´ px ´ α2q2
2πpβ2 ´ 1qx2 ` α2x ` 1
on rα2 ´ 2?β2, α2 ` 2?β2s, the measure can also have one or two atoms.
There is a useful combinatorial formula which expresses moments of probability mea-
sures on the real line in terms of non-crossing partitions consisting of 1-blocks (single-
tons) and 2-blocks (pairs). Namely, let N C1,2
m be the set of non-crossing partitions of
the set rms ": t1, 2, . . . , mu consisting of singletons and pairs, namely
π " tπ1, π2, . . . , πku P N Cm
where each πj contains one or two elements, respectively, and it is not possible to have
two different 2-blocks πi " tp, qu and πj " tr, su, for which p ă r ă q ă s.
In any non-crossing partition π P N Cm, if we put all numbers from the set rms in
order and draw lines connecting all numbers which belong to the same block, the lines
corresponding to different blocks cannot intersect each other. Further, its block πi is
outer with respect to the block πj if there exist r, s P πi such that for each p P πj it
holds that r ă p ă s. If π consists of singletons and pairs, it is clear that any outer
block must be a pair. We say that the block πi of π P N C has depth d pπiq " d piq if
it has d piq ´ 1 outer blocks. Thus, blocks which do not have outer blocks are assumed
to have depth one. Note that if a block πi has at least one outer block, we can choose
among them the one which lies immediately above πi and we will call it its nearest outer
block.
Mnpµq " ÿπPN C 1,2
If µ is a probability measure on the real line with all moments finite and the pair of
Jacobi sequences Jpµq " pα, βq, its n-th moment is given by the combinatorial formula
n źi:πi"1
i.e. each block of depth d of every π P N C1,2
m contributes αd or βd if it is a singleton
or a pair, respectively. This formula was first discovered by Cabanal-Duvillard and
Ionesco for symmetric measures [7]. In that case, the first Jacobi sequence α vanishes
and only pair partitions appear in the formula. The general version is due to Accardi
and Bozejko [1].
αd piq źj:πj"2
βd pjq,
3. Operatorial realization
We will use matricially free Gaussian operators living in the matricially free Fock
space of tracial type introduced in [12] to find a realization of moments of free Meixner
RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
5
laws. This Fock space is a generalization of the matricially free Fock space M introduced
in [11].
For the purposes of this article, it suffices to consider the special case when
where both M1 and M2 are Hilbert space direct sums
M " M1 ' M2,
M1 " CΩ1 '
M2 " CΩ2 '
2 b H1q,
8àk"0pHbk
8àk"1
Hbk
2 ,
where Ω1, Ω2 are unit vectors, Hj " Cej for j P t1, 2u, where e1, e2 are unit vectors,
and Hb0 b H1 " H1. The space M is endowed with the canonical inner product.
Using the canonical basis of this Fock space,
2 b e1, ebl
B " tΩ1, Ω2, ebk
2
: k P N Y t0u, l P Nu,
we define creation operators ℘1, ℘2 P BpMq as follows. Let pβ1, β2q be a pair of non-
negative numbers. We set
℘1Ω1 "aβ1 e1,
and we assume that ℘1 sends the remaining basis vectors to zero. In turn, ℘2 sends Ω1
to zero and otherwise,
℘2Ω2 " aβ2 e2
2 q " aβ2 ebpk`1q
℘2pebk
2 b e1q " aβ2 pebpl`1q
2
2
℘2pebl
b e1q
for any k P N, l P N Y t0u. By ℘
respectively, and sums of the form
1 and ℘
2 we denote the adjoints of ℘1 and ℘2,
ω1 " ℘1 ` ℘
1 and ω2 " ℘2 ` ℘
2
will be the corresponding Gaussian operators. Note that Mj is invariant with respect
to ℘i, ℘
i , ωi for any i, j P t1, 2u.
In particular, if we set β1 " β2 " 1, then the restrictions
p℘1 ` ℘2qM1 and ℘2M2
can be identified with the standard free creation operators living in M1 and M2, respec-
tively, and both spaces are isomorphic to the free Fock space over the one-dimensional
Hilbert space.
Remark 3.1. Our Fock space M is a special case of the matricially free Fock space of
tracial type associated with an array pHp,qq of Hilbert spaces, by which we understand
the Hilbert space direct sum
M "
ràq"1
Mq,
6
R. LENCZEWSKI
where each summand is of the form
pp1,p2q‰pp2,p3q‰...‰ppm,qq
F 0
p1,p2 b F 0
p2,p3 b . . . b F 0
pm,q
with tensor products built from free and boolean Fock spaces
Mq " CΩq '
8àm"1
à
p,q "" À8
Hp,q
F 0
k"1
Hbk
q,q
if p " q
if p ‰ q
,
In this paper, we suppose the array pHp,qq consists
with vacuum spaces subtracted.
of only two one-dimensional Hilbert spaces H2,1 " H1 and H2,2 " H2. Clearly, an
assymmetry in pHp,qq leads to an assymetry in the definitions of M1 and M2.
Remark 3.2. We can identify the creation operators ℘1, ℘2 with the matricially free
creation operators
℘1 " ℘2,1 and ℘2 " ℘2,2,
where we use the matricial two-index notation of [10,11]. This notation is often helpful
(and will be used when we refer to the results of these papers) since the second index
shows onto which basis vectors the operators act non-trivially (it must match the first
index of the basis vector). Therefore, ℘p,q acts non-trivially only onto Ωq and tensor
products which begin with eq,r for any r. Thus, for instance,
℘2,1Ω2 " 0, ℘2,1e2,1 " 0, ℘2,1pe2,2 b e2,1q " 0, ℘2,2Ω1 " 0,
which stands behind the definition of ℘1, ℘2 (the fact that ℘1,2 and ℘1,1 are not used
makes the one-index notation feasible). We also have ℘
2,2, with the
corresponding scalars β1 " b2,1 and β2 " b2,2.
In turn, ω1 and ω2 can be identified
with the corresponding matricially free Gaussian operators ω2,1 and ω2,2, respectively.
Details on the arrays of such operators can be found in [10,11].
1 " ℘
2 " ℘
2,1, ℘
Using these operators, we can define operators in BpM1q whose distributions in the
state Ψ1 defined by the vector Ω1 are free Meixner laws. For that purpose, the subspace
M2 is not needed yet.
Theorem 3.1. If µ is the free Meixner law corresponding to pα1, α2, β1, β2q, where
β1 ‰ 0 and β2 ‰ 0, then its m-th moment is given by
Mmpµq " Ψ1ppω ` γqmq,
where
and
ω " ω1 ` ω2
and Ψ1 is the state defined by the vector Ω1.
γ " pα2 ´ α1qpβ´1
1 ℘1℘
1 ` β´1
2 ℘2℘
2q ` α1,
Proof. Let us first analyze the moments of ω since these were studied in [12] in the
general case of matricially free Gaussian operators. The operator ω can be identified
with
ω " ω2,1 ` ω2,2
RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
7
by Remark 3.2. Of course, if we set ω1,2 " ω1,1 " 0, we can use the combinatorial
formula for the moments of the Gaussian pseudomatrix,
ω " ÿ1ďp,qď2
ωp,q
associated with a 2 2 array pωp,qq, in which we express these moments in terms of
colored non-crossing pair partitions [11, Lemma 4.1].
By a colored non-crossing pair partition we shall understand a pair pπ, fq, where
π " tπ1, π2, . . . , πsu is a non-crossing pair partition and f is a function on the set of its
blocks with values in the set rrs. If we draw an additional 2-block which is outer with
respect to all blocks of π, called the imaginary block, and we color it by q, we obtain
the set of colored non-crossing pair partitions N C 2
m,qrrs colored by rrs under condition
that the imaginary block is colored by q. Then, we have
Ψqpωmq " ÿpπ,f qPN C 2
m,qrrs
bqpπ, fq,
where the summation is over the empty set if m is odd, and
bqpπ, fq " bqpπ1, fqbqpπ2, fq . . . bqpπs, fq
if m " 2s, where π " tπ1, π2, . . . , πsu and
bqpπk, fq " bi,j
whenever block πk is colored by i and its nearest outer block is colored by j. In this
formulation, we set b1,1 " b1,2 " 0 since there is no ω1,1 or ω1,2, but formally it holds
for all colorings.
If we set q " 1, which refers to our theorem, the imaginary block gets colored by 1.
Moreover, since b1,2 " b1,1 " 0, the non-vanishing contribution to Ψ1pωmq comes only
from those colored partitions pπ, fq P N C 2
m,qrrs in which each block of π is colored by
2. In fact, if some block πk was colored by 1 and its nearest outer block (including the
imaginary block) was colored by 1 or 2, then the corresponding bpπk, fq would have to
be b1,1 or b1,2, but these vanish. This means that to each block of depth one we assign
the number b2,1 " β1 since the imaginary block is colored by 1 and it is its nearest outer
block, whereas to each block of depth greater than one we assign the number b2,2 " β2
since each block of π is colored by 2. Namely
This gives
if d pπkq " 1
if d pπkq ą 1
b1pπk, fq "" β1
Ψ1ppω2,2 ` ω2,1qmq " ÿπPN C 2
since in this case the set N C 2
m,1r2s of colored non-crossing pair partitions of rms with
the imaginary block colored by 1 reduces to the set in which all blocks colored by 2,
which is in bijection with N C2
m. Switching back to the notations of this paper, we thus
have
b1pπ, fq
β2
.
m
Ψ1pωmq " ÿπPN C 2
m
βB1pπq
1
βB2pπq
2
,
8
R. LENCZEWSKI
where B1pπq and B2pπq are the sets of 2-blocks of π of depth 1 nad of depth greater
than 1, respectively.
In fact, the above formula for the moments of ω can be proved directly without
invoking the general statement of [11, Lemma 4.1]. It suffices to observe that pπ, fq is
uniquely determined by the sequence ǫ " pǫ1, ǫ2, . . . , ǫmq which appears in nonvanishing
mixed moments of creation and annihilation operators of type
Ψ1p℘ǫ1
q1℘ǫ2
q2 . . . ℘ǫm
qmq,
where ǫk P t1,u since the choice of ǫ uniquely determines the tuple pq1, q2, . . . , qmq due
to the 0-1 action of ℘1 and ℘2 and their adjoints. Namely, only ℘1 acts non-trivially onto
Ω1, giving e1, which corresponds to the right leg of each block of depth 1 (its adjoint
corresponds to its left leg since it sends e1 into Ω1). In turn, ℘2 acts non-trivially onto
each basis element of B except Ω and thus it corresponds to the right leg of each block
of depth greater than 1 (its adjoint corresponds to its left leg). Therefore, each block of
π of depth 1 is associated with the pair p℘
1, ℘1q producing β1, whereas the remaining
blocks are associated with the pair p℘
2, ℘2q producing β2.
It remains to check what happens when we replace ω by ω ` γ. Observe that
℘1℘
1 " β1P1 and ℘2℘
2 " β2P2,
where P1 is the canonical projection onto H1 and P2 is the canonical projection onto
the subspace
respectively. Therefore,
F2 "
8àk"1pHbk
2 b H1q,
γ " α1P ` α2pP1 ` P2q,
where P is the canonical projection onto CΩ1, which means that γ is diagonal in the
basis B, namely it multiplies Ω1 and all vectors from BzΩ1 by α1 and α2, respectively.
Therefore, if we are given a mixed moment
q1℘ǫ2
Ψ1p℘ǫ1
q2 . . . ℘ǫk
qkq,
associated with a non-crossing pair partition π of the set rks, each mixed moment of
the form
Ψ1pγn0℘ǫ1
q1γn1℘ǫ2
q2γn2 . . . ℘ǫk
qk
γnkq,
where n0, n1, . . . , nk are non-negative integers such that
k ` n0 ` n1 ` . . . ` nk " m,
which appears when we compute the m-th moment of ω`γ, is naturally associated with
a non-crossing partition rπ of the set rms obtained from π by adding m ´ k singletons
in such a way that nj singletons are placed right after the number j, with n0 singletons
placed before the number 1 belonging to the first pair. In this fashion we obtain all
non-crossing partitions of rms which have m ´ k singletons and k pairs. Further, each
rπ P N C1,2
Moreover, to each singleton of depth 1 we assign α1 and to each singleton of depth
greater than 1 we assign α2 in view of the diagonal form of γ in the basis B. Therefore,
we obtain
m is obtained exactly once in this fashion from some π P N C 2
k.
Ψ1ppω ` γqmq " ÿπPN C 1,2
m
αS1pπq
1
αS2pπq
2
βB1pπq
1
βB2pπq
2
,
RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
9
2
2
2
1
2
π
1
1
2
2
2
2
2
σ
Figure 1. Examples of colored non-crossing partitions.
where S1pπq and S2pπq are the sets of singletons of depth 1 and of depth greater than 1
in π, respectively. As we know from the combinatorial formula for the moments given
in the Introduction, this is the m-th moment of the free Meixner law. This completes
the proof.
(cid:4)
Example 3.1. Let us give some examples of non-crossing partitions and the associated
mixed moments. The diagrams are given in Figure 1. The partition π consists of 4
pairs, namely π1 " t1, 8u, π2 " t2, 5u, π3 " t3, 4u, π4 " t6, 7u, with the imaginary
block marked with a dotted line. There exists exactly one mixed moment of creation
and annihilation operators that corresponds to this partition, namely we must have
ǫ " p,,, 1, 1,, 1, 1q and the corresponding moment (the only non-trivial one which
corresponds to this ǫ) is
Ψ1p℘
1℘
2℘
2 ℘2℘2℘
2℘2℘1q " β1β 3
2
2 b e1.
2 b e1, respectively. Next, ℘
since ℘1 is the only creation operator which acts non-trivially onto Ω1, giving e1, and ℘2
is the only creation operator which acts non-trivially onto e1 and e2 b e1, giving e2 b e1
and eb2
1 is the only annihilation operator which acts non-
trivially onto e1, whereas ℘
2 is the only annihilation operator which acts non-trivially
onto e2 b e1 and eb2
The partition σ contains 3 pairs and 3 singletons, namely σ1 " t1, 8u, σ2 " t2, 7u,
σ3 " t3u, σ4 " t4, 5u, σ5 " t6u, σ6 " t9u. We assign the color 1 to all singletons of
depth one and the color 2 to all remaining singletons. The colors assigned to singletons
are to some extent arbitrary (they did not appear in [11,12], where we considered
pair partitions only), but it is convenient to color all singletons of depth 1 by 1 and
the remaining ones by 2 since this corresponds to the right Jacobi coefficients. The
associated mixed moment is
Ψ1p℘
1℘
2 γ℘
2℘2γ℘2℘1γq " α1α2
2β1β 2
2 ,
where the 2-blocks are associated with the pairs p℘
2 , ℘2q, which produce
β1 and β2, respectively (like in the case of π), whereas the singletons are associated
with γ, which produces α1 in the case of t9u (since in this case γ acts onto Ω1), and α2
in the case of t3u and t9u (since in this case γ acts onto e2 b e1).
1, ℘1q and p℘
If β1 " β2 " 0, we set ω1 " 0 and γ1 " α which leads to the Dirac measure at α1. In
turn, the case β2 " 0 is treated below.
Corollary 3.1. If µ is the free Meixner law corresponding to pα1, α2, β1, 0q, then its
m-th moment is given by
Mmpµq " Ψ1ppω1 ` γ1qmq,
10
where
R. LENCZEWSKI
and Ψ1 is the state defined by the vector Ω1.
γ1 " pα2 ´ α1qβ´1
1 ℘1℘
1 ` α1
Proof. It suffices to observe that if we disregard ℘2 and ℘
2 in all computations in the
proof of Theorem 3.1, then β2 disappears from the formula for the moments of ω ` γ
under Ψ1.
(cid:4)
Finally, we would like to compute the moments of ω ` γ in the state Ψ2. Observe
that ℘2,1 vanishes on M2 and therefore this reduces to the computation of moments of
a slightly simpler operator.
Corollary 3.2. If µ is the free Meixner law corresponding to pα1, α2, β2, β2q, where
β2 ą 0, then its m-th moment is given by
where
Mmpµq " Ψ2ppω2 ` γ2qmq,
γ2 " pα2 ´ α1qβ´1
2 ℘2℘
2 ` α1
and Ψ2 is the state defined by the vector Ω2.
Proof. Observe that the action of ℘2, ℘
2 on M2 is exactly the same as that of the free
creation and annihilation operators, respectively, on the free Fock space. This means
that the moments of ω2 under Ψ2 agree with the moments of the (centered) semicircle
law with variance β, i.e. each moment of even order m " 2s is equal to βs times the
Catalan number Cs. Represent Cs as the sum over N C2
m and observe that if we replace
ω2 by ω2 ` γ2, the effect is that N C 2
m as in the proof of Theorem
3.1, with singletons of depth 1 and 2 contributing α1 and α2, respectively. This gives
the combinatorial formula for the m-th moment of the free Meixner law corresponding
to pα1, α2, β2, β2q.
(cid:4)
m gets replaced by N C 1,2
4. Random matrix model
Using our results on asymptotic distributions of random symmetric blocks and The-
orem 3.1, we can now construct a random matrix model for free Meixner laws.
Consider the sequence of Gaussian Hermitian random matrices Y pnq, where n P N,
under the assumptions of [11, Theorem 5.1]. Namely, we assume that Y pnq is a complex
Gaussian n n random matrix of the block form
Cpnq Dpnq
Y pnq " Apnq Bpnq
where the off-diagonal blocks are adjoints of each other, whereas the diagonal blocks
are Hermitian and the sizes of blocks are defined by the partition of the set rns "
t1, 2, . . . , nu,
and
rns " N1 Y N2, where N1 X N2 " H
d1 " lim
nÑ8
N1
n " 0 and d2 " lim
nÑ8
N2
n " 1,
which corresponds to the situation in which
(1) the sequence pDpnqq is balanced,
RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
11
(2) the sequence of symmetric blocks built from pBpnqq and pCpnqq is unbalanced,
(3) the sequence pApnqq is evanescent,
according to the natural terminology introduced in [12]. Since pApnqq is evanescent, we
can equivalently assume that each block of this sequence vanishes.
Using the notation of [12], where blocks are equipped with indices, we have
Apnq " S1,1pnq, Bpnq " S1,2pnq, Cpnq " S2,1pnq, Dpnq " S2,2pnq.
It is convenient to identify all blocks Sp,qpnq as well as the symmetric blocks
Tp,qpnq "" Sq,qpnq
Sp,qpnq ` Sq,ppnq
if p " q
if p ‰ q
with their embeddings in the algebra of n n matrices, so that we can decompose
matrices in terms of their blocks, namely
Y pnq "ÿp,q
Sp,qpnq " ÿpďq
Tp,qpnq,
which allows us to write the mixed moments of blocks under any partial trace τjpnq
over basis vectors of Cn indexed by the set Nj.
Shortly speaking, we shall assume that the matrices Y pnq are Gaussian Hermitian
random matrices with block-identically distributed entries. More explicitly, we assume
that
(1) each entry Yi,jpnq of Ypnq is a complex Gaussian random variable of the form
Yi,jpnq " ReYi,jpnq ` iImYi,jpnq,
(2) the family
tReYi,jpnq, ImYi,jpnq : 1 ď i ď j ď nu
is independent for any n,
(3) the real-valued Gaussian variables have mean zero and
EpYi,jpnqYi,jpnqq "
vp,q
n
whenever pi, jq P Np Nq for p, q P t1, 2u, where the variance matrix V " pvp,qq
is symmetric.
Theorem 4.1. Under the above assumptions, let τ1pnq be the partial normalized trace
over the set of first N1 basis vectors and let β1 " v2,1 ą 0 and β2 " v2,2 ą 0. Then
lim
nÑ8
where
τ1pnqppMpnqqmq " Ψ1ppω ` γqmq
Mpnq " Y pnq ` α1I1pnq ` α2I2pnq
for any n P N, where Ipnq " I1pnq` I2pnq is the decomposition of the n n unit matrix
induced by the partition rns " N1 Y N2 and ω, γ are given by Theorem 3.1.
Proof. We decompose Y pnq in terms of symmetric random blocks as
Y pnq " T1,2pnq ` T1,1pnq ` T2,2pnq
12
R. LENCZEWSKI
and therefore, by [11, Theorem 5.1], the moments of Y pnq under any partial trace,
including τ1pnq, tend to the moments of the corresponding Gaussian pseudomatrix ω,
namely
lim
nÑ8
τ1pnqppY pnqqmq " Ψ1ppωqmq
where
ω " ω2,1 ` ω2,2
since ω1,2 " ω1,1 " 0 and that is why they do not appear in the above formula (each
ωp,q is associated with the scalar bp,q " dpvp,q and we have d1 " 0). In the random
matrix contex, this means that the sequence pT1,1pnqq is evanescent and pT1,2pnqq is
unbalanced. Moreover,
b2,1 " d2v2,1 :" β1 and b2,2 " d2v2,2 :" β2
since d2 " 1 and v2,1 " v1,2. This proves the assertion in the case when α1 " α2 " 0
(this includes Kesten laws).
Before we prove the assertion for the general case, let us observe that the block
refinement of the above asymptotics can be written in the form
lim
nÑ8
τ1pnqpTp1,q1Tp2,q2 . . . Tpm,qmq " Ψ1pωp1,q1ωp2,q2 . . . ωpm,qmq
provided we denote by T2,1 rather than by T1,2 the off-diagonal symmetric block.
Namely, by [11, Theorem 5.1], the mixed moments of symmetric blocks Tp,q under
partial traces converge to the corresponding mixed moments of symmetrized Gaussian
operators pωp,q, where pω1,1 " ω1,1 and pω2,2 " ω2,2 and, more importantly,
Since, in the case considered in this theorem, ω1,2 " d1v1,2 " 0 and thus pω1,2 " ω2,1,
we can replace each pωpi,qi by ωpi,qi, which leads to the above equation. Moreover, even
more information about these moments can be obtained. For that purpose, decompose
Cn " W1 ' W2, where Wj is the linear span of basis vectors indexed by i P Nj and
observe that
pω1,2 " ω1,2 ` ω2,1.
T2,1pW1q Ď W2, T2,1pW2q Ď W1 and Tj,jpWjq Ď Wj
for j P t1, 2u. Since τ1pnq is the partial trace over basis vectors from W1, the above
mixed moments of symmetric blocks vanishes unless it takes the form in which even
powers of T2,2 alternate with T2,1, namely
τ1pnqpT2,1T m1
2,2 T2,1 . . . T2,1T mr
2,2 T2,1q,
where m1, . . . , mr P 2N Y t0u and m1 ` m2 ` . . . ` mr ` 2r " m. Likewise, the
corresponding mixed moments of matricially free Gaussian operators vanish unless they
take the form
Ψ1pω2,1ωm1
2,2 ω2,1 . . . ω2,1ωmr
2,2 ω2,1q
since ω2,1 acts non-trivially onto Ω1 giving e1 and sends e1 back to Ω1, whereas ω2,2
kills both Ω1 and e1, leaving F2 invariant. An even more detailed inspection leads to
the formula
lim
nÑ8
τ1pnqpS1,2T m1
since ℘2,1Ω1 " e1 and ℘
not true in general that S1,2 Ñ ℘
2,2 S2,1 . . . S1,2T mr
2,1e1 " Ω1. Note that the last formula is not obvious since it is
2,1 and S2,1 Ñ ℘2,1 under the partial traces. However,
2,2 S2,1q " Ψ1p℘
2,2 ℘2,1 . . . ℘
2,2 ℘2,1q
2,1ωm1
2,1ωm1
RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
13
it is very convenient because it allows us to study the effect of inserting the diagonal
deterministic matrix
B " α1I1pnq ` α2I2pnq
between the symmetric blocks, where the dependence of B on n is supressed. We will
show that an insertion of B somewhere on the LHS of the above formula corresponds
to an insertion of the operator γ at the corresponding place on the RHS. Namely, this
local analysis gives:
(1) at the left or right end of the above moment, the matrix B reduces to α1I1
and thus it produces α1 since it acts onto W1; the corresponding γ can also be
replaced by α1 since it acts onto Ω,
(2) in products of type BS2,1 and BT2,2, the matrix B reduces to α2I2 and gives α2
since it acts onto W2; the corresponding pairs γ℘2,1 and γω2,2 can be replaced
by α2℘2,1 and α2ω2,2, respectively, since γ acts here onto vectors from F2.
Consequently, for all non-trivial mixed moments of Tp,q and B, we can write
lim
nÑ8
τ1pnqpBn0Y Bn1Y . . . Y Bnkq " Ψ1pγn0ωγn1ω . . . ωγnkq
for any nonnegative integers n0, n1, . . . , nk and any α1 and α2. This implies that
lim
nÑ8
τ1pnqppMpnqqmq " Ψ1ppω ` γqmq,
which completes the proof of our theorem.
(cid:4)
Corollary 4.1. If β1 " v2,1 ą 0 and β2 " v2,2 " 0 and under the remaining assumptions
as in Theorem 4.1, it holds that
lim
nÑ8
τ1pnqppMpnqqmq " Ψ1ppω1 ` γ1qmq
where ω1, γ1 are given by Corollary 3.1.
Proof. The proof is similar to that of Theorem 4.1. The only difference is that blocks
T2,2pu, nq disappear from the computations under the trace τ1pnq and thus non-trivial
mixed moments take the special form
τ1pnqpBn0T2,1Bn1T2,1 . . . T2,1Bnmq " τ1pnqpBn0S1,2Bn1S2,1 . . . S2,1Bnmq
where m is even and S1,2 alternates with S2,1. They tend to
Ψ1pγn0ω1γn1ω1 . . . ω1γnmq " Ψ1pγn0℘
1 γn1℘1 . . . ℘1γnmq
1 alternates with ℘1, since each BjS1,2Bk can be replaced by αj
as n Ñ 8, where ℘
2S1,2
for any j, k P N by the definition of B and, similarly, each γj℘
1γk can be replaced by
αj
1αk
1 be the definition of γ. It remains to observe that in the situation when we have
mixed moments of ω1 and γ under Ψ1, we remain within H1 ' CΩ1 and thus γ can be
repleced by γ1, which completes the proof.
2℘
1αk
(cid:4)
Corollary 4.2. Under the assumptions of Theorem 4.1, it holds that
lim
nÑ8
τ2pnqppMpnqqmq " Ψ2ppω2 ` γ2qmq
where ω2, γ2 are given by Corollary 3.2.
14
R. LENCZEWSKI
Proof. The proof is similar to that of Theorem 4.1. In this case, when we compute
the moments of Mpnq under τ2pnq, the mixed moments of T1,1pnq, T2,1pnq, T2,2pnq and
B become zero as n Ñ 8 if there is T1,1pnq or T2,1pnq among them. On the level of
matrices, this can be explained as follows: the fact that pT2,1pnqq is unbalanced and
is forced to act onto 'many' (of order nq basis vectors from W2 giving 'few' (of order
smaller than n) basis vectors from W1 makes the moment containing T2,1pnq vanish in
the limit n Ñ 8 (in other words, zero asymptotic dimensions cannot be associated
with inner blocks). Of course, the case of T1,1pnq is clear since it is evanescent. On
the operatorial level, the effect of this is that the moments involving ω1 do not con-
tribute to the limit moments since all operators act within M2, where ω1 is trivial and
thus these moments reduce to the moments of ω2 and γ under Ψ2. Moreover, it is not
hard to see that in fact γ can be replaced with γ2, which is the restriction of γ to M2. (cid:4)
5. Free Meixner Ensemble
Let us consider an ensemble of independent random matrices of type considered in
Section 4 and study their limit joint distributions under the state Ψ1 as n Ñ 8. The
situation parallels that for the case of independent Gaussian random matrices and their
asymptotic freeness [17]. As in Section 4, we will rely on the result derived in [12].
Definition 5.1. By the Free Meixner Ensemble we will understand the family of inde-
pendent n n Hermitian Gaussian random matrices tMpu, nq : n P N, u P Uu, where
matrices
Mpu, nq " Y pu, nq ` α1puqI1pnq ` α2puqI2pnq
satisfy the assumptions of Theorem 4.1 or Corollary 4.1 for any u P U, where U is
an index set, with the constants α1puq, α2puq as well as variances β1puq " v2,1puq,
β2puq " v2,2puq depending on u P U. In particular, we assume that all matrices are
decomposed into blocks in the same fashion for any fixed n and that their asymptotic
dimensions are d1 " 0 and d2 " 1 for all u.
We already know from Theorem 4.1 that the asymptotic distribution of Mpu, nq
under the partial trace τ1pnq is the free Meixner distribution associated with
pα1puq, α2puq, β1puq, β2puqq,
but we would like to find an asymptotic relation between independent random matrices
from this ensemble. This relation is expected to be of asymptotic freeness type. In fact,
we will demonstrate that the Free Meixner Ensemble is asymptotically conditionally
free. As in Section 4, we exclude the case when β1puq " 0 for some u since in this case
the corresponding matrix realization would be purely deterministic, but one can easily
extend all results to include this case.
We also know from [12] that the Hermitian Symmetric Gaussian Block Ensemble
tTp,qpu, nq : u P U, n P Nu
is asymptotically symmetrically matricially free, where symmetric matricial freeness is a
symmetrized version of matricial freeness. More precisely, its asymptotics is determined
by operators of type pωp,qpuq which are limit realizations of the corresponding symmetric
RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
15
blocks Tp,qpuq. We shall use the results of [12], where we also studied the family of their
sums
Y pu, nq " ÿpďq
Tp,qpu, nq,
in order to find the limit distributions of the Free Meixner Ensemble.
We used the mutlivariate matricially free Fock space of tracial type. The definition
of M remains the same as in Section 3, but instead of one-dimensional Hilbert spaces,
we take direct sums
Hj "àuPU
Hjpuq
ϕpa1a2 . . . amq " 0
whenever ai P Apuiq X Kerψ for any 1 ď i ď n ´ 1 and an P Apunq X Kerϕ, where
u1 ‰ u2 ‰ . . . ‰ un. This definition is equivalent to other definitions and immediately
shows that there is a relation between different levels of Hilbert spaces in their free
product and the corresponding states assigned to these levels. Consequently, there is a
relation with the depths of the blocks of noncrossing partitions which contribute to the
moments of conditionally free random variables. In more generality, we obtain freeness
with infinitely many states [7].
where Hjpuq " Cejpuq for any j P t1, 2u and u P U, where tejpuq : j P t1, 2u, u P Uu is
an orthonormal set. Let
B " tΩ1, Ω2, e2pu1, . . . , unq, e2pu1, . . . , un´1q b e1punq : u1, . . . , un P U, n P Nu
be the orthonormal basis of M, where we use a shorthand notation
Then we define the family of creation operators ℘1puq, ℘2puq by the following rules:
e2pu1, . . . , unq " e2pu1q b . . . b e2punq.
℘1puqΩ1 " aβ1puq e1puq
℘2puqΩ2 " aβ2puq e2puq
℘2puqe2pu1, . . . , unq " aβ2puq e2pu, u1, . . . , unq
℘2puqe2pu1, . . . , un´1q b e1punq " aβ2puq e2pu, u1, . . . , un´1q b e1punq
2puq we denote their adjoints, respectively, and sums of the form
and we assume that ℘1puq, ℘2puq send the remaining basis vectors to zero. By ℘
and ℘
1puq
ωjpuq " ℘jpuq ` ℘
jpuq
are the corresponding Gaussian operators. We have shown in [12] that operators of
this type give the limit realization of the mixed moments of symmetric blocks of inde-
pendent Hermitian Gaussian random matrices with block-identical variances (Gaussian
Symmetric Block Ensemble). In other words, we showed that we have convergence of
mixed moments
lim
nÑ8
τqpnqpTp1,q1pu1, nq . . . Tpm,qmpum, nqq " Ψqppωp,qpu1q . . .pωpm,qmpumqq.
where pωp,qpuq is the same symmetrization as in the case of pωp,q in Section 3.
Let us give a definition of conditional freeness which is very similar to that of freeness
and that will be helpful for us. The family of unital subalgebras tApuq : u P Uu of a
unital algebra A is conditionally free with respect to the pair of states pϕ, ψq on A if
16
R. LENCZEWSKI
Theorem 5.1. Let τjpnq be the partial trace over the set of basis vectors indexed by Nj,
where j P t1, 2u. The family of matrices
tMpu, nq : u P U, n P Nu
is asymptotically conditionally free with respect to the pair of partial traces pτ1pnq, τ2pnqq
as n Ñ 8.
Proof. In particular, if we consider the 2 2 block random matrices with asymptotic
dimensions d1 " 0 and d2 " 1, the sequence pT1,1pn, uqq is evanescent and pT1,2pn, uqq
is unbalanced and thus the corresponding arrays of symmetrized Gaussian operators
reduce to arrays containing only ω2puq " ω2,2puq and ω1puq " ω2,1puq simply because
ω1,1puq " 0 and ω1,2puq " 0. Thus, in view of the above, we have
lim
nÑ8
τ1pnqpY pu1, nq . . . Y pum, nqq " Ψ1pωpu1q . . . ωpumqq,
where
ωpuq " ω1puq ` ω2puq
for any u P U. As in the proofs of Theorems 3.1 and 4.1, this can be generalized to the
moments of matrices Mpu, nq from the Free Meixner Ensemble since all computations
presented there are based on the relations between matricial indices of the considered
blocks and of the considered operators and they depend on u only in the sense that the
blocks associated with symmetric blocks and with the corresponding operators labelled
by u give rise to parameters αjpuq, βjpuq labelled by u. Thus, we have
τ1pnqpMpu1, nq . . . Mpum, nqq " Ψ1pypu1q . . . ypumqq,
lim
nÑ8
where
and
ypuq " ωpuq ` γpuq
γpuq " pα2puq ´ α1puqqpβ´1
1 puq℘1puq℘
1puq ` β´1
2 puq℘2puq℘
2puqq ` α1puq,
for any u P U, where αqpuq P R and βqpuq ą 0 for q P t1, 2u. In a similar way one shows
that
lim
nÑ8
τ2pnqpMpu1, nq . . . Mpum, nqq " Ψ2pypu1q . . . ypumqq
where Corollaries 3.2 and 4.2 are used. Therefore, in order to prove our assertion, we
need to show that the family
typuq : u P Uu
is conditionally free with respect to the pair of states pΨ1, Ψ2q, where Ψq is the vector
state associated with Ωq. We will prove a slightly more general result, namely that the
family of unital *-algebras tApuq : u P Uu, each generated by ℘2,1puq and ℘2,2puq for
fixed u, respectively, is conditionally free with respect to pΨ1, Ψ2q. We need to show
that
for any ai P Apuiq X KerΨ2, where 1 ď i ď n ´ 1 and an P Aun X KerΨ1.
We claim that the variable an is a polynomial in noncommuting variables
Ψ1pa1a2 . . . anq " 0
℘1punq, ℘
1punq, ℘2punq, ℘
2punq
which can be written as a linear combination of P K " 1 ´ P and of monomials
℘m2
2 punq℘m1
1 punqp℘
1punqqk2p℘
2punqqk2
RANDOM MATRIX MODEL FOR FREE MEIXNER LAWS
17
where m2, k2 P N Y t0u, k1, k2 P t0, 1u are such that m1 ` m2 ` k1 ` k2 ą 0. In order to
reduce all monomials from KerΨ1 to this form, first observe that M1 is invariant under
the action of ℘1, ℘2 and their adjoints. Therefore, it suffices to consider all operators
as their restrictions to M1. Then we have the relations
1puq℘1puq " β1puqP, ℘
℘
2puq℘2puq " β2puqP K and ℘
qpuq℘qpu1q " 0
as well as
℘1puq℘2pu1q " 0, P ℘1puq " 0, P K℘1puq " ℘1puq, ℘1puqP " ℘1puq, ℘1puqP K " 0
for any q P t1, 2u and u ‰ u1, as well as their adjoints. Clearly, P K P KerΨ1. Therefore,
we can pull all starred operators to the right of the unstarred ones in an and our claim
is proved. This implies that an maps Ω1 into M1 a CΩ1.
Now, any vector from the image anpΩ1q is a linear combination of vectors which begin
with e2punq. Therefore, the action of an´1 onto these vectors is the same as its action
onto Ω2. Therefore, if we take an´1 P KerΨ2 and we apply a similar reasoning as above,
we can write it as a linear combination of monomials
2pun´1qqk2
2 pun´1qp℘
℘m2
where m2 ` k2 ą 0 since Apun´1q leaves M1 a CΩ1 invariant (recall that un´1 ‰
un) and the action of ℘
1pun´1q is trivial on this space. Moreover, the constant term
vanishes since an´1 P KerΨ2 and thus an´1anpΩ1q is a linear combination of vectors
which begin with e2pun´1q. Continuing in this fashion, we obtain a1a2 . . . anpΩ1q K Ω1,
which completes the proof.
Remark 5.1. It can be easily seen that the family of algebras tApuq : u P Uu is, in
general, not free with respect to Ψ1. For instance, in the simple case when the Jacobi
parameters are p0, 0, β1, β2q for all u P U, where 0 ‰ β1 ‰ β2 ‰ 0, then we can take two
polynomials, say w1 " ypsq P Apsq and w2 " pypuqq2 ´ β1 P Apuq, where u ‰ s, which
are in KerΨ1, but
(cid:4)
since
Ψ1pw1w2w1q " β1pβ2 ´ β1q ‰ 0
1puq℘1puqΩ1 " β1Ω1 and ℘
℘
2puq℘2puqe1 " β2e1.
Of course, if we replace w2 by w3 " pypuqq2 ´ β2 P KerΨ2, we get zero in the above
equation, which in in agreement with the conditional freeness of w1, w3 with respect to
pΨ1, Ψ2q stated in Theorem 5.1.
References
[1] L. Accardi, M. Bozejko, Interacting Fock spaces and Gaussianization of probability measures,
Infin. Dimens. Anal. Quantum Probab. Relat. Top. 4 (1998), 663-670.
[2] M. Anshelevich, Free martingale polynomials, J. Funct. Anal. 201(2003), 228-261.
[3] M. Anshelevich, Orthogonal polynomials with a resolvent-type generating function, Trans. Amer.
Math. Soc. 360 (2008), 4125-4143.
[4] T. Banica, S.T. Belinschi, M. Capitaine, B. Collins, Free Bessel laws, Canad. J. Math. 63 (2011),
3-37.
[5] F. Benaych-Georges, Rectangular random matrices, related convolution, Probab. Theory Relat.
Fields 144 (2009), 471-515.
[6] M. Bozejko, W. Bryc, On a class of free L´evy laws related to a regression problem, J. Funct. Anal.
236 (2006), 59-77.
18
R. LENCZEWSKI
[7] Th. Cabanal-Duvillard, V. Ionescu, Un th´eoreme central limite pour de variables al´eatoires non-
commutatives, C.R.A.S. 325 (1997), Serie I, 1117-1120.
[8] M. Capitaine, M. Casalis, Asymptotic freeness by generalized moments for Gaussian and Wishart
matrices. Applications to beta random matrices. Indiana Univ. Math. J. 53 (2004), 397-431.
[9] K. Dykema, On certain free product factors via an extended matrix model, J. Funct. Anal. 112
(1993), 31-60.
[10] R. Lenczewski, Matricially free random variables, J. Funct. Anal. 258 (2010), 4075-4121.
[11] R. Lenczewski, Asymptotic properties of random matrices and pseudomatrices, Adv. Math. 228
(2011), 2403-2440.
[12] R. Lenczewski, Limit distributions of random matrices, arXiv:1208.3586 [math.OA], 2012.
[13] R. Lenczewski, R. Sa lapata, Multivariate Fuss-Narayana polynomials with appplication to random
matrices, arXiv:1210.3063 [math.CO], 2012.
[14] V. Marchenko, L. Pastur, The distribution of eigenvalues in certain sets of random matrices, Math.
Sb. 72 (1967), 507-536.
[15] D. Shlyakhtenko, Random Gaussian band matrices and freeness with amalgamation, Int. Math.
Res. Notices 20 (1996), 1013-1025.
[16] D. Voiculescu , K. Dykema, A. Nica, Free random variables, CRM Monograph Series, No.1,
A.M.S., Providence, 1992.
[17] D. Voiculescu, Limit laws for random matrices and free products, Invent. Math. 104 (1991),
201-220.
[18] D. Voiculescu, Circular and semicircular systems and free product factors, Progress in Math. 92,
Birkhauser, 1990.
[19] E. Wigner, On the distribution of the roots of certain symmetric matrices, Ann. Math. 67 (1958),
325-327.
[20] J. Wishart, The generalized product moment distribution in samples from a normal multivariate
population, Biometrika 20A (1928), 32-52.
Romuald Lenczewski,
Instytut Matematyki i Informatyki, Politechnika Wroc lawska,
Wybrze ze Wyspia´nskiego 27, 50-370 Wroc law, Poland
E-mail address: [email protected]
|
1003.0908 | 2 | 1003 | 2012-03-01T22:39:46 | The matricial relaxation of a linear matrix inequality | [
"math.OA",
"math.OC"
] | Given linear matrix inequalities (LMIs) L_1 and L_2, it is natural to ask: (Q1) when does one dominate the other, that is, does L_1(X) PsD imply L_2(X) PsD? (Q2) when do they have the same solution set? Such questions can be NP-hard. This paper describes a natural relaxation of an LMI, based on substituting matrices for the variables x_j. With this relaxation, the domination questions (Q1) and (Q2) have elegant answers, indeed reduce to constructible semidefinite programs. Assume there is an X such that L_1(X) and L_2(X) are both PD, and suppose the positivity domain of L_1 is bounded. For our "matrix variable" relaxation a positive answer to (Q1) is equivalent to the existence of matrices V_j such that L_2(x)=V_1^* L_1(x) V_1 + ... + V_k^* L_1(x) V_k. As for (Q2) we show that, up to redundancy, L_1 and L_2 are unitarily equivalent.
Such algebraic certificates are typically called Positivstellensaetze and the above are examples of such for linear polynomials. The paper goes on to derive a cleaner and more powerful Putinar-type Positivstellensatz for polynomials positive on a bounded set of the form {X | L(X) PsD}.
An observation at the core of the paper is that the relaxed LMI domination problem is equivalent to a classical problem. Namely, the problem of determining if a linear map from a subspace of matrices to a matrix algebra is "completely positive". | math.OA | math |
THE MATRICIAL RELAXATION OF A
LINEAR MATRIX INEQUALITY
J. WILLIAM HELTON1, IGOR KLEP2, AND SCOTT MCCULLOUGH3
Abstract. Given linear matrix inequalities (LMIs) L1 and L2 it is natural to ask:
(Q1) when does one dominate the other, that is, does L1(X) (cid:23) 0 imply L2(X) (cid:23) 0?
(Q2) when are they mutually dominant, that is, when do they have the same solution set?
The matrix cube problem of Ben-Tal and Nemirovski [B-TN02] is an example of LMI
domination. Hence such problems can be NP-hard. This paper describes a natural relaxation
of an LMI, based on substituting matrices for the variables xj. With this relaxation, the
domination questions (Q1) and (Q2) have elegant answers, indeed reduce to constructible
semidefinite programs. As an example, to test the strength of this relaxation we specialize it
to the matrix cube problem and obtain essentially the relaxation given in [B-TN02]. Thus our
relaxation could be viewed as generalizing it.
Assume there is an X such that L1(X) and L2(X) are both positive definite, and suppose
the positivity domain of L1 is bounded. For our "matrix variable" relaxation a positive answer
to (Q1) is equivalent to the existence of matrices Vj such that
(A1)
L2(x) = V ∗1 L1(x)V1 + · · · + V ∗µ L1(x)Vµ.
As for (Q2) we show that L1 and L2 are mutually dominant if and only if, up to certain
redundancies described in the paper, L1 and L2 are unitarily equivalent. Algebraic certificates
for positivity, such as (A1) for linear polynomials, are typically called Positivstellensatze. The
paper goes on to derive a Putinar-type Positivstellensatz for polynomials with a cleaner and
more powerful conclusion under the stronger hypothesis of positivity on an underlying bounded
domain of the form {X L(X) (cid:23) 0}.
An observation at the core of the paper is that the relaxed LMI domination problem is
equivalent to a classical problem. Namely, the problem of determining if a linear map τ from
a subspace of matrices to a matrix algebra is "completely positive". Complete positivity is
one of the main techniques of modern operator theory and the theory of operator algebras.
On one hand it provides tools for studying LMIs and on the other hand, since completely
positive maps are not so far from representations and generally are more tractable than their
merely positive counterparts, the theory of completely positive maps provides perspective on
the difficulties in solving LMI domination problems.
Date: October 29, 2018.
2010 Mathematics Subject Classification. Primary 46L07, 14P10, 90C22; Secondary 11E25, 46L89, 13J30.
Key words and phrases. linear matrix inequality (LMI), completely positive, semidefinite programming, Pos-
itivstellensatz, Gleichstellensatz, archimedean quadratic module, real algebraic geometry, free positivity.
1Research supported by NSF grants DMS-0700758, DMS-0757212, and the Ford Motor Co.
2Research supported by the Slovenian Research Agency grants P1-0222, P1-0288, and J1-3608.
3Research supported by the NSF grant DMS-0758306.
1
2
HELTON, KLEP, AND MCCULLOUGH
1. Introduction and the statement of the main results
In this section we state most of our main results of the paper. We begin with essential
definitions.
1.1. Linear pencils and LMI sets. For symmetric matrices A0, A1, . . . , Ag ∈ SRd×d, the
expression
(1.1)
L(x) = A0 +
gXj=1
Ajxj ∈ SRd×dhxi
in noncommuting variables x, is a linear pencil. If A0 = I, then L is monic. If A0 = 0, then
j=1 Ajxj of a linear pencil L as in (1.1)
L is a truly linear pencil. The truly linear partPg
will be denoted by L(1).
Given a block column matrix X = col(X1, . . . , Xg) ∈ (SRn×n)g, the evaluation L(X) is
defined as
(1.2)
L(X) = A0 ⊗ In +X Aj ⊗ Xj ∈ SRdn×dn.
The tensor product in this expressions is the usual (Kronecker) tensor product of matrices.
We have reserved the tensor product notation for the tensor product of matrices and have
eschewed the strong temptation of using A⊗ xℓ in place of Axℓ when xℓ is one of the variables.
Let L be a linear pencil. Its matricial linear matrix inequality (LMI) set (also called
a matricial positivity domain) is
(1.3)
Let
(1.4)
(1.5)
(1.6)
DL := [n∈N
{X ∈ (SRn×n)g L(X) (cid:23) 0}.
DL(n) = {X ∈ (SRn×n)g L(X) (cid:23) 0} = DL ∩ (SRn×n)g,
∂DL(n) = {X ∈ (SRn×n)g L(X) (cid:23) 0, L(X) 6≻ 0},
∂DL = [n∈N
∂DL(n).
The set DL(1) ⊆ Rg is the feasibility set of the semidefinite program L(X) (cid:23) 0 and is called a
spectrahedron by algebraic geometers.
We call DL bounded if there is an N ∈ N with kXk ≤ N for all X ∈ DL. We shall see
later below (Proposition 2.4) that DL is bounded if and only if DL(1) is bounded.
1.2. Main results on LMIs. Here we state our main theorems giving precise algebraic char-
acterizations of (matricial) LMI domination. While the main theme of this article is that
matricial LMI domination problems are more tractable than their traditional scalar counter-
parts, the reader interested only in algorithms for the scalar setting can proceed to the following
subsection, §1.3, and then onto Section 4.
Suppose L ∈ SRd×dhxi,
L2(x) = V ∗(cid:0)Iµ ⊗ L1(x)(cid:1)V.
L = I +
Ajxj
gXj=1
THE MATRICIAL RELAXATION OF AN LMI
3
Theorem 1.1 (Linear Positivstellensatz). Let Lj ∈ SRdj×djhxi, j = 1, 2, be monic linear
pencils and assume DL1 is bounded. Then DL1 ⊆ DL2 if and only if there is a µ ∈ N and an
isometry V ∈ Rµd1×d2 such that
(1.7)
is a monic linear pencil. A subspace H ⊆ Rd is reducing for L if H reduces each Aj; i.e., if
AjH ⊆ H. Since each Aj is symmetric, it also follows that AjH⊥ ⊆ H⊥. Hence, with respect
to the decomposition Rd = H ⊕ H⊥, L can be written as the direct sum,
Ajxj,
L = I +
L = L ⊕ L⊥ =" L 0
L⊥# , where
0
gXj=1
and Aj is the restriction of Aj to H. (The pencil L⊥ is defined similarly.) If H has dimension
ℓ, then by identifying H with Rℓ, the pencil L is a monic linear pencil of size ℓ. We say that
L is a subpencil of L. If moreover, DL = D L, then L is a defining subpencil and if no proper
subpencil of L is defining subpencil for DL, then L is a minimal defining (sub)pencil.
Theorem 1.2 (Linear Gleichstellensatz). Let Lj ∈ SRd×dhxi, j = 1, 2, be monic linear pencils
with DL1 bounded. Then DL1 = DL2 if and only if minimal defining pencils L1 and L2 for DL1
and DL2 respectively, are unitarily equivalent. That is, there is a unitary matrix U such that
(1.8)
L2(x) = U ∗ L1(x)U.
An observation at the core of these results is that the relaxed LMI domination problem
is equivalent to the problem of determining if a linear map τ from a subspace of matrices to a
matrix algebra is completely positive.
1.3. Algorithms for LMIs. Of widespread interest is determining if
(1.9)
DL1(1) ⊆ DL2(1),
or if DL1(1) = DL2(1). For example, the paper of Ben-Tal and Nemirovski [B-TN02] exhibits
simple cases where determining this is NP-hard. We explicitly give (in Section 4.1) a certain
semidefinite program whose feasibility is equivalent to DL1 ⊆ DL2. Of course, if DL1 ⊆ DL2,
then DL1(1) ⊆ DL2(1). Thus our algorithm is a type of relaxation of the problem (1.9). The
algorithms in this section can be read immediately after reading Section 1.3.
We also have an SDP algorithm (Section 4.4) easily adapted from the first to determine
if DL is bounded, and what its "radius" is. Proposition 2.4 shows that DL is bounded if and
only if DL(1) is bounded. Thus our algorithm definitively tells if DL(1) is a bounded set; in
addition it yields an upper bound on the radius of DL(1).
4
HELTON, KLEP, AND MCCULLOUGH
In Section 4.5 we specialize our relaxation to solve a matricial relaxation of the classical
matrix cube problem, finding the biggest matrix cube contained in DL. It turns out, as shown
in Section 5, that our matricial relaxation is essentially that of [B-TN02]. Thus the our LMI
inclusion relaxation could be viewed as a generalization of theirs, indeed a highly canonical
one, in light of the precise correspondence to classical complete positivity theory shown in
§3. A potential advantage of our relaxation is that there are possibilities for strengthening it,
presented generally in Section 4.2 and illustrated on the matrix cube in Section 5.2.
Finally, given a matricial LMI set DL, Section 4.6 gives an algorithm to compute the linear
pencil L ∈ SRd×dhxi with smallest possible d satisfying DL = D L.
1.4. Positivstellensatz. Algebraic characterizations of polynomials p which are positive on
DL are called Positivstellensatze and are classical for polynomials on Rg. This theory underlies
the main approach currently used for global optimization of polynomials, cf. [Las09, Par03].
The generally noncommutative techniques in this paper lead to a cleaner and more power-
ful commutative Putinar-type Positivstellensatz [Put93] for p strictly positive on a bounded
spectrahedron DL(1). In the theorem which follows, SRd×d[y] is the set of symmetric d × d
matrices with entries from R[y], the algebra of (commutative) polynomials with coefficients
from R. Note that an element of SRd×d[y] may be identified with a polynomial (in commuting
variables) with coefficients from SRd×d.
Theorem 1.3. Suppose L ∈ SRd×d[y] is a monic linear pencil and DL(1) is bounded. Then
for every symmetric matrix polynomial p ∈ Rℓ×ℓ[y] with pDL(1) ≻ 0, there are Aj ∈ Rℓ×ℓ[y],
and Bk ∈ Rd×ℓ[y] satisfying
(1.10)
B∗
kLBk.
p =Xj
A∗
j Aj +Xk
We also consider symmetric (matrices of) polynomials p in noncommuting variables with
the property that p(X) is positive definite for all X in a bounded matricial LMI set DL; see
Section 6. For such noncommutative (NC) polynomials (and for even more general algebras of
polynomials, see Section 7) we obtain a Positivstellensatz (Theorem 6.1) analogous to (1.10).
In the case that the polynomial p is linear, this Positivstellensatz reduces to Theorem 1.1,
which can be regarded as a "Linear Positivstellensatz". For perspective we mention that the
proofs of our Positivstellensatze actually rely on the linear Positivstellensatz. For experts we
point out that the key reason LMI sets behave better is that the quadratic module associated
to a monic linear pencil L with bounded DL is archimedean.
1.5. Outline. The paper is organized as follows. Section 2 collects a few basic facts about
linear pencils and LMIs. In Section 3, inclusion and equality of matricial LMI sets are char-
acterized and our results are then applied in the algorithmic Section 4. Section 5 gives some
further details about matricial relaxations of the matrix cube problem. The last two sections
give algebraic certificates for polynomials to be positive on LMI sets.
THE MATRICIAL RELAXATION OF AN LMI
5
2. Preliminaries on LMIs
This section collects a few basic facts about linear pencils and LMIs.
Proposition 2.1. If L is a linear pencil and DL contains 0 as an interior point, i.e., 0 ∈
DL r ∂DL, then there is a monic pencil bL with DL = D bL.
Proof. As 0 ∈ DL, L(0) = A0 is positive semidefinite. Since 0 6∈ ∂DL, A0 (cid:23) εAj for some small
ε ∈ R>0 and all j. Let V = Ran A0 ⊆ Rd, and set
for
j = 0, 1, . . . , g.
eAj := AjV
Clearly, eA0 : V → V is invertible and thus positive definite. We next show that Ran A0
contains Ran Aj for j ≥ 1. If x ⊥ Ran A0, i.e., A0x = 0, then 0 = x∗A0x ≥ ± εx∗Ajx and hence
x∗Ajx = 0. Since A0 + εAj ≥ 0 and x∗(A0 + εAj )x = 0 it follows that (A0 + εAj)x = 0, and
since A0x = 0, we finally conclude that Ajx = 0, i.e., x ⊥ Ran Aj. Consequently, eAj : V → V
are all symmetric and DL = D eL for eL = eA0 +Pg
j=1 eAjxj.
To build bL, factor eA0 = B∗B with B invertible and set
bAj := B−∗eAjB−1
The resulting pencil bL = I +Pg
j=1 bAjxj is monic and DL = D bL.
Our primary focus will be on the matricial LMI sets DL. If the spectrahedron DL(1) ⊆
Rg does not contain interior points, then (as it is a convex set) it is contained in a proper
affine subspace of Rg. By reducing the number of variables we arrive at a new pencil whose
spectrahedron does have an interior point. By a translation we can ensure that 0 is an interior
point. Then Proposition 2.1 applies and yields a monic linear pencil with the same matricial
LMI set. This reduction enables us to concentrate only on monic linear pencils in the sequel.
j = 0, . . . , g.
for
another linear pencil. Set s := n(1 + g). Then:
Lemma 2.2. Let L ∈ SRd×dhxi be a linear pencil with DL bounded, and let bL ∈ SRn×nhxi be
(1) bLDL ≻ 0 if and only if bLDL(s) ≻ 0;
(2) bLDL (cid:23) 0 if and only if bLDL(s) (cid:23) 0.
Proof. In both statements the direction (⇒) is obvious. If bLDL 6≻ 0, there is an ℓ, X ∈ DL(ℓ)
j=1vj ∈ (Rℓ)n with
and v = ⊕n
hbL(X)v, vi ≤ 0.
Let
K := span(cid:0){Xivj i = 1, . . . , g, j = 1, . . . , n} ∪ {vj j = 1, . . . , n}(cid:1).
Clearly, dimK ≤ s. Let P be the orthogonal projection of Rℓ onto K. Then
Since P XP ∈ DL(s), this proves (1). The proof of (2) is the same.
hbL(P XP )v, vi = hbL(X)v, vi ≤ 0.
6
HELTON, KLEP, AND MCCULLOUGH
Lemma 2.3. Let L be a linear pencil. Then
DL is bounded ⇔ DL(cid:0)(1 + g)2(cid:1) is bounded.
Proof. Given a positive N ∈ N, consider the monic linear pencil
JN (x) =
1
N
N x1
x1 N
...
xg
··· xg
. . .
N
1
N "N x∗
x N Ig# ∈ SR(g+1)×(g+1)hxi.
=
Note that DL is bounded if and only if for some N ∈ N, JNDL (cid:23) 0. The statement of the
lemma now follows from Lemma 2.2.
To the linear pencil L we can also associate its matricial ball
{X ∈ (SRn×n)g kL(X)k ≤ 1} = {X I − L(X)2 (cid:23) 0}.
BL := [n∈N
Observe that BL = DL′ for
(2.1)
L′ ="I L
L I# .
Proposition 2.4. Let L be a linear pencil. Then:
(1) DL is bounded if and only if DL(1) is bounded;
(2) BL is bounded if and only if BL(1) is bounded.
Proof. (1) The implication (⇒) is obvious. For the converse suppose DL is unbounded. By
Lemma 2.3, this means DL(N ) is unbounded for some N ∈ N. Then there exists a sequence
(X (k)) from (SRN ×N )g such that kX (k)k = 1 and a sequence tk ∈ R>0 tending to ∞ such that
L(tkX (k)) (cid:23) 0. A subsequence of (X (k)) converges to X = (X1, . . . , Xg) ∈ (SRN ×N )g which
also has norm 1. For any t, tX (k) → tX and for k big enough, tX (k) ∈ DL by convexity. So X
satisfies L(tX) (cid:23) 0 for all t ∈ R≥0.
There is a nonzero vector v so that hXiv, vi 6= 0 for at least one i. Then with Z :=
(hX1v, vi, . . . ,hXgv, vi) ∈ Rg r{0}, and V denoting the map V : R → RN defined by V r = rv,
L(tZ) = (I ⊗ V )∗L(tX)(I ⊗ V )
is nonnegative for all t > 0, so DL(1) is unbounded.
To conclude the proof observe that (2) is immediate from (1) using (2.1).
A linear pencil L is nondegenerate, if it is one-one in that L(X) = L(Y ) implies X = Y
for all n ∈ N and X, Y ∈ (SRn×n)g. In particular, a truly linear pencil L is nondegenerate if
and only if L(X) 6= 0 for X 6= 0.
Lemma 2.5. For a linear pencil L(x) = A0 +Pg
j=1 Ajxj the following are equivalent:
(i) L is nondegenerate;
THE MATRICIAL RELAXATION OF AN LMI
7
(ii) L(Z) = L(W ) implies Z = W for all Z, W ∈ Rg;
(iii) the set {Aj j = 1, . . . , g} is linearly independent;
(iv) L(1) is nondegenerate.
Proof. Clearly, (i) ⇔ (iv). Also, (i) ⇒ (ii) and (ii) ⇒ (iii) are obvious. For the remaining
implication (iii) ⇒ (i), assume L(X) = L(Y ) for some X, Y ∈ (SRn×n)g. Equivalently, L(1)(X−
Y ) = 0. Note that L(1)(X − Y ) equalsPg
j=1(Xj − Yj) ⊗ Aj modulo the canonical shuffle. If
this expression equals 0, then the linear independence of the A1, . . . , Ag (applied entrywise)
implies X = Y .
Proposition 2.6. Let L = I +Pg
(1) BL(1) is bounded if and only if L(1) is nondegenerate;
(2) if DL is bounded then {I, Aj j = 1, . . . , g} is linearly independent; the converse fails in
j=1 Ajxj ∈ SRd×dhxi be a monic linear pencil and let L(1)
denote its truly linear part. Then:
general.
Proof. (1) Suppose L(1) is not nondegenerate, say Pg
j=1 zjAj = 0 for some zj ∈ R. Then
with Z = (z1, . . . , zg) ∈ Rg we have tZ ∈ BL(1) for every t, so BL(1) is not bounded. Let
us now prove the converse. First, if BL(1) is unbounded, then by Proposition 2.4, BL(1)(1) is
unbounded. So suppose BL(1)(1) is unbounded. Then there exists a sequence (Z (k)) from Rg
such that kZ (k)k = 1 and a sequence tk ∈ R>0 tending to ∞ such that kL(1)(tkZ (k))k ≤ 1. A
subsequence of (Z (k)) converges to Z ∈ Rg which also has norm 1; however, kL(1)(Z)k = 0 and
thus L(1) is degenerate.
For (2) assume
(2.2)
λ +Xj
xjAj = 0
with λ, xj ∈ R. We may assume xj 6= 0 for at least one index j. Let Z = (x1, . . . , xg) 6= 0. If
λ = 0, then L(tZ) = I is positive semidefinite for all t ∈ R. Thus DL is not bounded.
Now let λ ∈ R be nonzero. Then L(Z/λ) = 0. Thus, L(tZ/λ) (cid:23) 0 for all t < 0, showing
DL is unbounded.
The converse of (2) fails in general. For instance, if the Aj are positive semidefinite, then
DL contains (R≥0)g and thus cannot be bounded.
3. Matricial LMI sets: Inclusion and Equality
Given L1 and L2 monic linear pencils
(3.1)
Lj(x) = I +
Aj,ℓxℓ ∈ SRdj×djhxi,
j = 1, 2,
we shall consider the following two inclusions for matricial LMI sets:
gXℓ=1
(3.2)
DL1 ⊆ DL2;
8
HELTON, KLEP, AND MCCULLOUGH
(3.3)
Equation (3.2) is equivalent to: for all n ∈ N and X ∈ (SRn×n)g,
∂DL1 ⊆ ∂DL2.
L1(X) (cid:23) 0 ⇒ L2(X) (cid:23) 0.
Similarly, (3.3) can be rephrased as follows:
L1(X) (cid:23) 0 and L1(X) 6≻ 0 ⇒ L2(X) (cid:23) 0 and L2(X) 6≻ 0.
In this section we characterize precisely the relationship between L1 and L2 satisfying
(3.2) and (3.3). Section 3.1 handles (3.2) and gives a Positivstellensatz for linear pencils.
Section 3.3 shows that "minimal" pencils L1 and L2 satisfying (3.3) are the same up to unitary
equivalence.
Example 3.1. By Lemma 2.2 it is enough to test condition (3.2) on matrices of some fixed
(large enough) size. It is, however, not enough to test on X ∈ Rg. For instance, let
x1 x2
1
0
1
0
∈ SR3×3hxi
1 − x1# ∈ SR2×2hxi.
x2
1
x1
x2
0 0 1
0 0 0
1 0 0
x2 =
1 0# x2 ="1 + x1
1 − X 2
1 + X 2
1 + X 2
x2
2 (cid:23) 0},
2 ≤ 1},
2 ≤ 1}.
D∆ = {(X1, X2) 1 − X 2
D∆(1) = {(X1, X2) ∈ R2 X 2
DΓ(1) = {(X1, X2) ∈ R2 X 2
0 0# ,"0
" 1
3
4
3
4
0
2
0#! ∈ D∆ r DΓ,
∆(x1, x2) = I +
Γ(x1, x2) = I +"1
0 1 0
1 0 0
0 0 0
x1 +
0 −1# x1 +"0 1
0
and
Then
Thus D∆(1) = DΓ(1). On one hand,
so ∆(X1, X2) (cid:23) 0 does not imply Γ(X1, X2) (cid:23) 0. On the other hand, Γ(X1, X2) (cid:23) 0 does
imply ∆(X1, X2) (cid:23) 0. We shall prove this later below, see Example 3.4.
(cid:3)
We now introduce subspaces to be used in our considerations:
Sj = span{I, Aj,ℓ ℓ = 1, . . . , g} ⊆ SRdj×dj .
(3.4)
Lemma 3.2. Sj = span{Lj(X) X ∈ Rg}.
The key tool in studying inclusions of matricial LMI sets is the mapping τ we now define.
Definition 3.3. Let L1, L2 be monic linear pencils as in (3.1). If {I, A1,ℓ ℓ = 1, . . . , g} is
linearly independent (e.g. DL1 is bounded), we define the unital linear map
(3.5)
τ : S1 → S2, A1,ℓ 7→ A2,ℓ.
THE MATRICIAL RELAXATION OF AN LMI
9
We shall soon see that, assuming (3.2), τ has a property called complete positivity, which
we now introduce. Let Sj ⊆ Rdj×dj be unital linear subspaces invariant under the transpose,
and φ : S1 → S2 a unital linear ∗-map. For n ∈ N, φ induces the map
φn = In ⊗ φ : Rn×n ⊗ S1 = S n×n
1 → S n×n
2
, M ⊗ A 7→ M ⊗ φ(A),
called an ampliation of φ. Equivalently,
··· T1n
...
. . .
··· Tnn
T11
...
Tn1
φn
=
φ(T11)
...
φ(Tn1)
··· φ(T1n)
. . .
··· φ(Tnn)
...
for Tij ∈ S1. We say that φ is k-positive if φk is a positive map. If φ is k-positive for every
k ∈ N, then φ is completely positive. If φk is an isometry for every k, then φ is completely
isometric.
0 1# 7→
"1 0
2
, E12 7→
1 1 0
1 1 0
0 0 1
1
Example 3.4 (Example 3.1 revisited). The map τ : S2 → S1 in our example is given by
0 0 1
0 0 0
1 0 0
Consider the extension of τ to a unital linear ∗-map ψ : R2×2 → R3×3, defined by
1 0 0
0 1 0
0 0 1
0
.
, "0 1
1 0# 7→
, E22 7→
0 0
1
0 0 −1
0
1 1
1
0 1 0
1 0 0
0 0 0
, "1
0 −1# 7→
2
2
, E21 7→
C ="ψ(E11) ψ(E12)
ψ(E21) ψ(E22)# .
0
1
0
1
1 −1 0
0
0
1
(Here Eij are the 2 × 2 matrix units.) Now we show the map ψ is completely positive. To do
this, we use its Choi matrix defined as
1
2
1 −1 0
−1
0
0
1
1
0
.
E11 7→
(3.6)
[Pau02, Theorem 3.14] says ψ is completely positive if and only if C (cid:23) 0. We will use the Choi
matrix again in Section 4 for computational algorithms. To see that C is positive semidefinite,
note
C =
1
2
1
0
W ∗W for W ="1 1 0 0
0 0 1 1 −1 0# .
V2#
0 S#"V1
V2#∗"S 0
2"V1
1 −1 0# , thus W = hV1 V2i .) In
0 0 1# and V2 = "0
V ∗
2 SV2 =
V ∗
1 SV1 +
1
2
1
2
1
1
0
Now ψ has a very nice representation:
(3.7)
ψ(S) =
for all S ∈ R2×2. (Here V1 = "1 1 0
particular,
(3.8)
2∆(x, y) = V ∗
1 Γ(x, y)V1 + V ∗
2 Γ(x, y)V2.
10
HELTON, KLEP, AND MCCULLOUGH
Hence Γ(X1, X2) (cid:23) 0 implies ∆(X1, X2) (cid:23) 0, i.e., DΓ ⊆ D∆.
The formula (3.8) illustrates our linear Positivstellensatz which is the subject of the next
subsection. The construction of the formula in this example is a concrete implementation of
the theory leading up to the general result that is presented in Corollary 3.7.
(cid:3)
3.1. The map τ is completely positive: Linear Positivstellensatz. We begin by equat-
ing n-positivity of τ with inclusion DL1(n) ⊆ DL2(n). Then we use the complete positivity of
τ to give an algebraic characterization of pencils L1, L2 producing an inclusion DL1 ⊆ DL2.
Theorem 3.5. Let
Lj(x) = I +
Aj,ℓxℓ ∈ SRdj×djhxi,
j = 1, 2
gXℓ=1
be monic linear pencils and assume the matricial LMI set DL1 is bounded. Let τ : S1 → S2 be
the unital linear map A1,ℓ 7→ A2,ℓ.
(1) τ is n-positive if and only if DL1(n) ⊆ DL2(n);
(2) τ is completely positive if and only if DL1 ⊆ DL2 ;
(3) τ is completely isometric if and only if ∂DL1 ⊆ ∂DL2 ,
We remark that the binding condition (3.3) used in (3) implies (3.2) used in (2) under
the boundedness assumption; see Proposition 3.9. The proposition says that the relaxed dom-
ination problem (see the abstract) can be restated in terms of complete positivity, under a
boundedness assumption. Conversely, suppose D is a unital (self-adjoint) subspace of SRd×d
and τ : D → SRd′×d′ is completely positive. Given a basis {I, A1, . . . , Ag} for D, let Bj = τ (Aj).
Let
L1 = I +X Ajxj, L2 = I +X Bjxj.
The complete positivity of τ implies, if L1(X) (cid:23) 0, then L2(X) (cid:23) 0 and hence DL1 ⊆ DL2.
Hence the completely positive map τ (together with a choice of basis) gives rise to an LMI
domination.
To prove the theorem we need a lemma.
Lemma 3.6. Let L = I +Pg
matricial LMI set DL. Then:
(1) if Λ ∈ Rn×n and X ∈ (SRn×n)g, and if
(3.9)
is symmetric, then Λ = Λ∗;
S := I ⊗ Λ + L(1)(X)
j=1 Ajxj ∈ SRd×dhxi be a monic linear pencil with bounded
(2) if S (cid:23) 0, then Λ (cid:23) 0;
(3) if Λ ∈ Rn×n and X ∈ (SRn×n)g, and if
(3.10)
T := Λ ⊗ I +
then Λ (cid:23) 0.
gXj=1
Xj ⊗ Aj (cid:23) 0
THE MATRICIAL RELAXATION OF AN LMI
11
Proof. To prove item (1), suppose
S = I ⊗ Λ +
gXj=1
Aj ⊗ Xj
is symmetric. Then 0 = S − S∗ = I ⊗ (Λ − Λ∗). Hence Λ = Λ∗.
For (2), if Λ 6(cid:23) 0, then there is a vector v such that hΛv, vi < 0. Consider the projection
P onto Rd ⊗ Rv, and let Y = (hXjv, vi)g
j=1 ∈ Rg. Then the corresponding compression
P SP = P (I ⊗ Λ + L(1)(X))P = I ⊗ hΛv, vi + L(1)(Y ) (cid:23) 0,
which says that L(1)(Y ) ≻ 0. This implies 0 6= tY ∈ DL for all t > 0; contrary to DL being
bounded.
Finally, for (3), we note that T is, after applying a permutation (often called the canonical
shuffle), of the form (3.9). Hence Λ (cid:23) 0 by (2).
Proof of Theorem 3.5. In each of the three statements, the direction (⇒) is obvious. We focus
on the converses.
is positive definite. Then T is of the form (3.10) for some
Fix n ∈ N. Suppose T ∈ S n×n
1
Λ (cid:23) 0 and X ∈ (SRn×n)g. By applying the canonical shuffle,
S = I ⊗ Λ +X A1,j ⊗ Xj ≻ 0.
If we change Λ to Λ + εI, the resulting T = Tε is in S n×n
assume Λ ≻ 0. Hence,
1
(I ⊗ Λ− 1
2 )S(I ⊗ Λ− 1
2 ) = I ⊗ I +X A1,j ⊗ (Λ− 1
2 XjΛ− 1
2 ) ≻ 0.
Condition (3.2) thus says that
, so without loss of generality we may
Multiplying on the left and right by I ⊗ Λ
Applying the canonical shuffle again, yields
I ⊗ I +X A2,j ⊗ (Λ− 1
1
2 shows
2 XjΛ− 1
2 ) (cid:23) 0.
I ⊗ Λ +X A2,j ⊗ Xj (cid:23) 0.
τ (Tε) = Λ ⊗ I +X Xj ⊗ A2,j (cid:23) 0.
Thus we have proved, if Tε ∈ S n×n
and Tε ≻ 0, then τ (Tε) (cid:23) 0. An approximation argument
now shows if T (cid:23) 0, then τ (T ) (cid:23) 0 and hence τ is n-positive proving (1). Now (2) follows
immediately.
1
For (3), suppose T ∈ S n×n
1
has norm one. It follows that
W =" I
T ∗
T
I# (cid:23) 0.
12
HELTON, KLEP, AND MCCULLOUGH
From what has already been proved, τ (W ) (cid:23) 0 and therefore τ (W ) has norm at most one.
Moreover, since W has a kernel, so does τ (W ) and hence the norm of τ (T ) is at least one. We
conclude that τ is completely isometric.
Corollary 3.7 (Linear Positivstellensatz). Let
Lj(x) = I +
Aj,ℓxℓ ∈ SRdj×djhxi,
j = 1, 2
gXℓ=1
be monic linear pencils and assume DL1 is bounded.
implies L2(X) (cid:23) 0 for all X, then there is µ ∈ N and an isometry V ∈ Rµd1×d2 such that
(3.11)
If (3.2) holds, that is, if L1(X) (cid:23) 0
Conversely, if µ, V are as above, then (3.11) implies (3.2) holds.
L2(x) = V ∗(cid:0)Iµ ⊗ L1(x)(cid:1)V.
Remark 3.8. Before turning to the proof of the Corollary, we pause for a couple of remarks.
(1) Equation (3.11) can be equivalently written as
(3.12)
L2(x) =
V ∗
j L1(x)Vj,
where Vj ∈ Rd1×d2 and V = col(V1, . . . , Vµ). Since Pµ
j Vj = Id2, V is an isometry.
Moreover, µ can be uniformly bounded (see the proof of Corollary 3.7, or Choi's charac-
terization [Pau02, Proposition 4.7] of completely positive maps between matrix algebras).
In fact, µ ≤ d1d2.
j=1 V ∗
(2) Corollary 3.7 can be regarded as a Positivstellensatz for linear (matrix valued) polynomials,
a theme we expand upon later below. Indeed, (3.12) is easily seen to be equivalent to the
more common statement
(3.13)
L2(x) = B +
W ∗
j L1(x)Wj
µXj=1
ηXj=1
for some positive semidefinite B ∈ SRd2×d2 and Wj ∈ Rd1×d2.
If we worked over C, the proof of Corollary 3.7 would proceed as follows. First invoke
Arveson's extension theorem [Pau02, Theorem 7.5] to extend τ to a completely positive map
ψ from d1 × d1 matrices to d2 × d2 matrices, and then apply the Stinespring representation
theorem [Pau02, Theorem 4.1] to obtain
a ∈ Cd1×d1
ψ(a) = V ∗π(a)V,
(3.14)
for some unital ∗-representation π : Cd1×d1 → Cd3×d3 and isometry (since τ is unital) V :
Cd1 → Cd3. As all representations of Cd1×d1 are (equivalent to) a multiple of the identity
representation, i.e., π(a) = Iµ ⊗ a for some µ ∈ N and all a ∈ Cd1×d1, (3.14) implies (3.11).
However, in our case, the pencils Lj have real coefficients and we want the isometry V to
have real entries as well. For this reason and to aid understanding of this and our algorithm
Section S 4 we present a self-contained proof, keeping all the ingredients real.
THE MATRICIAL RELAXATION OF AN LMI
13
We prepare for the proof by reviewing some basic facts about completely positive maps.
This serves as a tutorial for LMI experts, who often are unfamiliar with complete positivity.
Linear functionals σ : Rd1×d1⊗Rd2×d2 → R are in a one-one correspondence with mappings
ψ : Rd1×d1 → Rd2×d2 given by
(3.15)
hψ(Eij)ea, ebi = hψ(eie∗
j )ea, ebi = σ(eje∗
i ⊗ eae∗
b ).
Here, with a slight conservation of notation, the ei, ej are from {e1, . . . , ed1} and ea, eb are from
{ea, . . . , ed2} which are the standard basis for Rd1 and Rd2 respectively.
Now we verify that positive functionals σ correspond precisely to completely positive ψ
and give a nice representation for such a ψ.
A positive functional σ : Rd1×d1 ⊗ Rd2×d2 = Rd1d2×d1d2 → R corresponds to a positive
semidefinite d1d2 × d1d2 matrix C via
σ(Z) = tr(ZC).
Express C = (Cpq)d1
(i, j) block entry of C is
p,q=1 as a d1 × d1 matrix with d2 × d2 entries. Thus, the (a, b) entry of the
(Cij)ab = hC(ej ⊗ ea), ei ⊗ ebi.
With Z = (ej ⊗ ea)(ei ⊗ eb)∗ observe that
hψ(Eij )ea, ebi = σ(eje∗
Hence, given S = (sij) =Pd1
b ) = tr(ZC) = hC(ej ⊗ ea), ei ⊗ ebi = hCijea, ebi.
i ⊗ eae∗
i,j=1 sijEij, by the linearity of ψ,
ψ(S) =Xi,j
sijCij.
Cij =
W ∗
jℓWiℓ.
d1Xℓ=1
(The matrix C is the Choi matrix for ψ, illustrated earlier in (3.6).) The matrix C is positive
and thus factors (over the reals) as W ∗W . Expressing W = (Wij)d1
i,j=1 as a d1× d1 matrix with
d2 × d2 entries Wij,
Define Vℓ = (Wiℓ). Then we have σ positive implies
(3.16)
ψ(S) =
sijCij =
d2Xi,j=1
d1Xℓ=1
d2Xi,j=1
W ∗
iℓsijWjℓ =
d1Xℓ=1
V ∗
ℓ SVℓ = V ∗(cid:0)(Id1 ⊗ S) ⊗ Id2(cid:1)V,
where V denotes the column with ℓ-th entry Vℓ. Hence ψ is completely positive.
Proof of Corollary 3.7. We now proceed to prove Corollary 3.7. Given τ as in Theorem 3.5,
define a linear functional σ : S1 ⊗ Rd2×d2 → R as in correspondence (3.15) by
σ(S ⊗ Y ) =Xa,b
hY eb, eaihτ (S)eb, eai.
14
HELTON, KLEP, AND MCCULLOUGH
Suppose Z =P Sk ⊗ Yk ∈ S1 ⊗ Rd2×d2 is positive semidefinite and let e =Pd2
the map τd1 = Id1 ⊗ τ , called an ampliation of τ , is positive,
0 ≤ hτd1 (Z)e, ei = σ(Z).
a=1 ea ⊗ ea. Since
Thus σ is positive and hence extends to a positive mapping σ : Rd1×d1 ⊗ Rd2×d2 → R by
the Krein extension theorem, which in turn corresponds to a completely positive mapping
ψ : Rd1×d1 → Rd2×d2 as in (3.15). It is easy to verify that ψS1 = τ . By the above,
Since ψ(I) = I, it follows that V ∗V = I.
ψ(S) = V ∗(cid:0)(Id1 ⊗ S) ⊗ Id2(cid:1)V.
3.2. Equal matricial LMI sets. In this section we begin an analysis of the binding condition
(3.3). We present an equivalent reformulation:
Proposition 3.9. Let L1, L2 be monic linear pencils. If DL1 is bounded and (3.3) holds, that
is, if ∂DL1 ⊆ ∂DL2 , then DL1 = DL2.
The proof is an easy consequence of the following elementary observation on convex sets.
Lemma 3.10. Let C1 ⊆ C2 ⊆ Rn be closed convex sets, 0 ∈ int C1 ∩ int C2. If ∂C1 ⊆ ∂C2
then C1 = C2.
Proof. By way of contradiction, assume C1 ( C2 and let a ∈ C2 r C1. The interval [0, a]
intersects C1 in [0, µa] for some 0 < µ < 1. Then µa ∈ ∂C1 ⊆ ∂C2. Since 0 ∈ int C1, C1
contains a small disk D(0, ε). Then K := co(D(0, ε) ∪ {a}) is contained in C2 and µa ∈
int K ⊆ int C2 contradicting µa ∈ ∂C2.
Proof of Proposition 3.9. Let Ci := DLi, i = 1, 2. Then
∂Ci = {X ∈ DLi Li(X) (cid:23) 0, Li(X) 6≻ 0}.
Since C1 is closed and bounded, it is the convex hull of its boundary. Thus by (3.3), C1 ⊆ C2.
Hence the assumptions of Lemma 3.10 are fulfilled and we conclude C1 = C2.
Example 3.11. It is tempting to guess that DL1 = DL2 implies L1 and L2 (or, equivalently,
L(1)
2 ) are unitarily equivalent. In fact, in the next subsection we will show this to
1
be true under a certain irreducibility-type assumption. However, in general this fails for the
trivial reason that the direct sum of a representing pencil and an "unrestrictive" pencil is also
representative.
and L(1)
Let L1 be an arbitrary monic linear pencil (with DL1 bounded) and
L2(x) = I +(cid:0)L(1)
1 (x) ⊕
I + 1
Then DL1 = DL2 but L1 and L2 are obviously not unitarily equivalent. However,
I + 1
0
1 (x)# ="L1(x)
0
2 L(1)
1 (x)# .
0
2 L(1)
L(1)
1
2
1 (x)
0
1 (x)(cid:1) ="I + L(1)
L1(x) ="I
0#∗
L2(x)"I
0#
THE MATRICIAL RELAXATION OF AN LMI
15
in accordance with Corollary 3.7.
"L1(x)
0
I + 1
Another guess would be that under DL1 = DL2, we have p = 1 in Corollary 3.7. However
this example also refutes that. Namely, there is no isometry V ∈ Rd1×2d1 satisfying
1 (x)# = L2(x) = V ∗L1(x)V.
0
2 L(1)
(Here L1 is assumed to be a d1 × d1 pencil.)
3.3. Minimal L representing DL are unique: Linear Gleichstellensatz. Let L = I +
we explain how to associate a monic linear pencil L to L with the following properties:
P Aixi be a d × d monic linear pencil and S = span{I, Aℓ ℓ = 1, . . . , g}. In this subsection
(cid:3)
(a) D L = DL;
(b) L is the minimal (with respect to the size of the defining matrices) pencil satisfying (a).
A pencil L = I +P Ajxj is a subpencil of L provided there is a nontrivial reducing
subspace H for S such that Aj = V ∗AjV , where V is the inclusion of H into Rd, where d is
the size of the matrices Aj. The pencil L is minimal if there does not exist a subpencil L
such that DL = D L.
Theorem 3.12. Suppose L and M are linear pencils of size d × d and e × e respectively. If
DL = DM is bounded and both L and M are minimal, then d = e and there is a unitary d × d
matrix U such that U ∗LU = M ; i.e., L and M are unitarily equivalent.
In particular, all minimal pencils for a given matricial LMI set have the same size (with
respect to the defining matrices) and this size is the smallest possible.
Example 3.13. Suppose L and M are only minimal with respect to the spectrahedra DL(1)
and DM (1), respectively. Then DL(1) = DM (1) does not imply that L and M are unitarily
equivalent. For instance, let L and M be the two pencils studied in Example 3.1. Then both
L and M are minimal, DL(1) = DM (1), but L and M are clearly not unitarily equivalent. (cid:3)
The remainder of this subsection is devoted to the proof of, and corollaries to, Theorem
3.12. We shall see how DL is governed by the multiplicative structure (i.e., the C ∗-algebra)
C ∗(S) generated by S as well as the embedding S ֒→ C ∗(S). For this we borrow heavily from
Arveson's noncommutative Choquet theory [Arv69, Arv08, Arv10] and to a lesser extent from
the paper of the third author with Dritschel [DM05].
We start with a basics of real C ∗-algebras needed in the proof of Theorem 3.12. First, the
well-known classification result.
Proposition 3.14. A finite dimensional real C ∗-algebra is ∗-isomorphic to a direct sum of
real ∗-algebras of the form Mn(R), Mn(C) and Mn(H). (Here the quaternions H are endowed
with the standard involution.)
Proposition 3.15. Let K ∈ {R, C, H} and let Φ : Mn(K) → Mn(K) be a real ∗-isomorphism.
16
HELTON, KLEP, AND MCCULLOUGH
Mn(K).
(1) If K ∈ {R, H}, then there exists a unitary U ∈ Mn(K) with Φ(A) = U ∗AU for all A ∈
(2) For K = C, there exists a unitary U ∈ Mn(C) with Φ(A) = U ∗AU for all A ∈ Mn(C) or
Φ(A) = U ∗ ¯AU for all A ∈ Mn(C). (Here ¯A denotes the entrywise complex conjugate of
A.)
Proof. In (1), Mn(K) is a central simple R-algebra. By the Skolem-Noether theorem [KMRT98,
Theorem 1.4], there exists an invertible matrix U ∈ Mn(K) with
(3.17)
Φ(A) = U −1AU
for all A ∈ Mn(K).
Since Φ is a ∗-isomorphism,
U −1A∗U = Φ(A∗) = Φ(A)∗ =(cid:0)U −1AU(cid:1)∗
= U ∗A∗U −∗,
leading to U U ∗ being central in Mn(K). By scaling, we may assume U U ∗ = I, i.e., U is unitary.
(2) Φ(i) is central and a skew-symmetric matrix, hence Φ(i) = αi for some α ∈ R.
Moreover, Φ(i2) = −1 yields α2 = 1. So Φ(i) = i or Φ(i) = −i. In the former case, Φ is a
∗-isomorphism over C and thus given by a unitary conjugation as in (1). If Φ(i) = −i, then Φ
composed with entrywise conjugation is a ∗-isomorphism over C. Hence there is some unitary
U with Φ(A) = U ∗ ¯AU for all A ∈ Mn(C).
Remark 3.16. For K ∈ {R, C, H}, every real ∗-isomorphism Φ : Mn(K) → Mn(K) lifts to a
unitary conjugation isomorphism Mdn(R) → Mdn(R), where d = dimR K. By Proposition 3.15,
this is clear if K ∈ {R, H}. To see why this is true in the complex case we proceed as follows.
Consider the standard real presentation of complex matrices, induced by
(3.18)
ι : C → M2(R),
a + i b 7→" a
−b a# .
b
If the real ∗-isomorphism Φ : Mn(C) → Mn(C) is itself a unitary conjugation, the claim is
obvious. Otherwise ¯Φ is conjugation by some unitary U ∈ Mn(C) and thus has a natural
extension to a ∗-isomorphism
Φ : M2n(R) → M2n(R), A 7→ ι(U )∗Aι(U ).
Then
Φ : M2n(R) → M2n(R), A 7→ In ⊗"1
0
0 −1#!−1
Φ(A) In ⊗"1
0
0 −1#!
is a unitary conjugation ∗-isomorphism of M2n(R) and restricts to Φ on Mn(C).
Let K be the biggest two sided ideal of C ∗(S) such that the natural map
(3.19)
is completely isometric on S. K is called the Silov ideal (also the boundary ideal) for S in
C ∗(S). Its existence and uniqueness is nontrivial, see the references given above. The snippet
a 7→ a := a + K
C ∗(S) → C ∗(S)/K,
THE MATRICIAL RELAXATION OF AN LMI
17
[Arv+] contains a streamlined, compared to approaches which use injectivity, presentation of
the Silov ideal based upon the existence of completely positive maps with the unique extension
property. While this snippet, as well as all of the references in the literature of which we are
aware, use complex scalars, the proofs go through with no essential changes in the real case.
A central projection P in C ∗(S) is a projection P ∈ C ∗(S) such that P A = AP for all
A ∈ C ∗(S) (alternately P A = AP for all A ∈ S). We will say that a projection Q reduces
or is a reducing projection for C ∗(S) if QA = AQ for all A ∈ C ∗(S). In particular, P is a
central projection if P reduces C ∗(S) and P ∈ C ∗(S).
Proposition 3.17. Let L be a d × d truly linear pencil and suppose DL is bounded. Then L
is minimal if and only if
(1) every minimal reducing projection Q is in fact in C ∗(S); and
(2) the Silov ideal of C ∗(S) is (0).
Proof. Assume (1) does not hold and let Q be a given minimal nonzero reducing projection for
C ∗(S), which is not an element of C ∗(S). Let P be a given minimal nonzero central projection
such that P dominates Q; i.e., Q (cid:22) P . By our assumption, Q 6= P .
Consider the real C ∗-algebra A = C ∗(S)P as a real ∗-algebra of operators on the range
H of P . First we claim that the mapping A ∋ A 7→ AQ is one-one. If not, it has a nontrivial
kernel J which is an ideal in A. The subspace K = JH reduces A and moreover, because of
finite dimensionality, the projection R onto K is in fact in A. Hence, R is a central projection.
By minimality, R = P or R = (0). In the second case the mapping is one-one. In the first
case, JH = H and thus J = C ∗(S)P ; i.e., the mapping C ∗(S)P ∋ A 7→ AQ is identically zero.
In this case, the mapping C ∗(S)P ∋ A 7→ A(I − Q) is completely isometric, contradicting the
minimality of L. Hence the map A ∋ A 7→ AQ is indeed one-one.
Therefore, the mapping C ∗(S) ∋ A 7→ A(I − P ) + AQ is faithful and in particular com-
pletely isometric. Thus the restriction of our pencil to the span of the ranges of I − P and
Q produces a pencil L′ with DL′ = DL, but of lesser dimension. Thus, we have proved, if (1)
does not hold, then L is not minimal.
It is clear that if the Silov ideal of C ∗(S) is nonzero, then L is not minimal. Suppose
J ⊆ C ∗(S) is an ideal and the quotient mapping σ : S → C ∗(S)/J is completely isometric.
As before, let K = J Rd (where the pencil L has size d). The projection P onto K is a central
projection. Because for S ∈ S we have both σ(S) = σ(S − SP ), and σ is completely isometric,
it follows that S 7→ S(I − P ) is completely isometric. By the minimality of L, it follows that
P = 0.
Conversely, suppose (1) and (2) hold. If L is not minimal, let L denote a minimal subpencil
with D L = DL, corresponding to a reducing subspace K ( Rd for S. Let Q denote the projection
onto K and T denote {SQ S ∈ S}. Note that the equality D L = DL says exactly that the
mapping S → T given by S 7→ SQ is completely isometric. In particular, if R is the projection
onto a reducing subspace which contains K, then also S 7→ SR is completely isometric.
18
HELTON, KLEP, AND MCCULLOUGH
Let P : Rd → K′ denote any minimal orthogonal projection onto a reducing subspace
of K⊥. By (1), P ∈ C ∗(S), and hence C ∗(S)P is a (minimal) two-sided ideal of C ∗(S). On
the other hand, (I − P ) is the projection onto a reducing subspace which contains K and
hence S 7→ S(I − P ) is completely isometric. Now let S = (Si,j) ∈ Mn(S) be given.
If
T = (Ti,j)(In ⊗ P ) ∈ Mn(C ∗(S))P , then
kS + Tk = kS(In ⊗ (I − P )) ⊕ (S + T )(In ⊗ P )k ≥ kS(In ⊗ (I − P ))k = kSk,
where the last equality comes from the fact that S 7→ S(I − P ) is completely isometric and
the inequality from the fact that the norm of a direct sum is the maximum of the norm of the
summands. Of course choosing T = S(In ⊗ P ) it follows that the norm of S in the quotient
C ∗(S)/C ∗(S)P is the same as kSk. Hence the induced map S → C ∗(S)/C ∗(S)P is completely
isometric and therefore C ∗(S)P is contained in the Silov ideal of S, contradicting (2).
Proof of Theorem 3.12. Write L = I +P Ajxj and M = I +P Bjxj and let C ∗(S) and C ∗(T )
denote the unital C ∗-algebras generated by {A1, . . . , Ag} and {B1, . . . , Bg} respectively. By
Proposition 3.17, both C ∗(S) and C ∗(T ) are reduced relative to S and T respectively; i.e., the
Silov ideals for S and T respectively are (0).
Moreover, for Q and P maximal families of minimal nonzero reducing projections for
C ∗(S) and C ∗(T ) respectively, we use Proposition 3.17 to obtain
C ∗(S) = ⊕Q∈QC ∗(S)Q, C ∗(T ) = ⊕P ∈PC ∗(T )P.
For later use we note that a minimal ideal in these C ∗-algebras is of the form C ∗(S)Q for
Q ∈ Q, and C ∗(T )P for P ∈ P, respectively.
The unital linear ∗-map
τ : S → T , Aj 7→ Bj
ρ : C ∗(S) → C ∗(T ).
is a completely isometric isomorphism by Theorem 3.5 and maps between reduced operator
systems. By [Arv69, Theorem 2.2.5], τ is induced by a ∗-isomorphism
Since ρ is an isomorphism of C ∗-algebras and C ∗(S)P for P ∈ P, is a minimal ideal,
(3.20)
for some Q ∈ Q. The converse is true too. That is, for each Q ∈ Q there is a unique P ∈ P
such that (3.20) holds. We conclude that d = e.
ρ(C ∗(S)P ) = C ∗(T )Q
By Proposition 3.15 and Remark 3.16 we also conclude that the C ∗-isomorphism ρ :
C ∗(S)P → C ∗(T )Q must be implemented by a unitary mapping Ran P → Ran Q.
Corollary 3.18. Let L ∈ SRd×dhxi be a monic linear pencil with bounded DL and L ∈ SRℓ×ℓhxi
its minimal pencil. Then there is a (d − ℓ) × (d − ℓ) monic linear pencil J satisfying JDL (cid:23) 0
and a unitary U ∈ Rd×d such that
L(x) = U ∗" L(x)
J(x)# U.
THE MATRICIAL RELAXATION OF AN LMI
19
Proof. Easy consequence of the construction of L.
4. Computational algorithms
In this section we present several numerical algorithms using semidefinite programming
(SDP) [WSV00], based on the theory developed in the preceding section. However, one can
read and implement these algorithms without reading anything beyond Section 1.3 of the
introduction. In each case, we first present the algorithm and then give the justification (which
a user need not read). The following section, Section 5, provides comparisons and refinements
of the matricial matrix cube algorithm of Subsection 4.5 below.
Given L1 and L2 monic linear pencils
(4.1)
Lj(x) = I +
Aj,ℓxℓ ∈ SRdj×djhxi,
j = 1, 2,
gXℓ=1
with bounded matricial LMI set DL1, we present an algorithm, the inclusion algorithm,
to test whether DL1 ⊆ DL2. Of course this numerical test yields a sufficient condition for
containment of the spectrahedra DL1(1) ⊆ DL2(1). We refer the reader to Section 4.4 for
a test of boundedness of LMI sets, which works both for commutative LMIs and matricial
LMIs, and computes the radius of a matricial LMI set based on the basic inclusion algorithm.
Subsection 4.2 contains a refinement of the basic inclusion algorithm, in the case that either L1
or L2 is a direct sum of pencils of smaller size. As an application, we then present a matricial
version of the classical matrix cube problem in Section 4.5. Analysis of the matricial matrix
cube algorithm are in Section 5 along with a comparison to the matrix cube algorithm of
Ben-Tal and Nemirovski [B-TN02]. There further algorithms, which offer improved estimates,
at the expense of additional computation, for the matrix cube problem are also discussed.
The final subsection of this section gives a (generically successful) algorithm for computation
of a minimal representing pencil and the Silov ideal, these being the only algorithms whose
statement is not self contained.
4.1. Checking inclusion of matricial LMI sets.
The inclusion algorithm
Given: A1,ℓ and A2,ℓ
for
ℓ = 1, . . . , g. Let αℓ
p,q denote the (p, q) entry of A1,ℓ.
Solve the following (feasibility) SDP:
(4.2)
(cpq)d1
p,q=1 := C (cid:23) 0,
d1Xp
cpp = Id2,
∀ℓ = 1, . . . , g :
αℓ
pqcpq = A2,ℓ,
d1Xp,q
for the unknown symmetric matrix C. Since each cpq is a Rd2×d2 matrix, the symmetric matrix
C of unknown variables (reasonably termed the Choi matrix) is of size d1d2×d1d2 and there are
2 d1d2(d1d2 + 1) (scalar) unknowns and 1
2 (1 + g)d2(d2 + 1) (scalar) linear equality constraints.
This can be, in practice, solved numerically with standard SDP solvers. In the next subsection,
1
20
HELTON, KLEP, AND MCCULLOUGH
we show that if L1 has special structure, then the number of (C) variables can be reduced,
sometimes dramatically.
Conclude: DL1 ⊆ DL2 if and only if the SDP (4.2) is feasible, i.e., has a solution.
Justification. By Theorem 3.5 and Corollary 3.7, L2 is positive semidefinite on DL1 if and
only if there is a completely positive unital map
(4.3)
satisfying
(4.4)
τ : Rd1×d1 → Rd2×d2
τ (A1,ℓ) = A2,ℓ
for
ℓ = 1, . . . , g.
To determine the existence of such a map, consider the Choi matrix C =(cid:0)τ (Eij)(cid:1)d1
i,j=1 ∈
(Rd2×d2)d1×d1 of τ . (Here, Eij are the d1×d1 elementary matrices.) For convenience of notation
we consider C to be a d1 × d1 matrix with d2 × d2 entries cij. This is the matrix C which
appears in the algorithm. It is well-known that τ is completely positive if and only if C is
positive semidefinite [Pau02, Theorem 3.14].
Note that we can write A1,ℓ =Pp,q αℓ
pqcpq.
This lays behind the last equation in (4.2). If a solution C to (4.2) has been obtained, then
a Positivstellensatz-type certificate for the inclusion of the matricial LMI sets DL1 ⊆ DL2 can
be obtained; cf. Example 3.4 or the proof of Corollary 3.7.
pqEpq. Then τ (A1,ℓ) =Pp,q αℓ
pqτ (Epq) =Pp,q αℓ
4.2. LMIs which are direct sums of LMIs. If either pencil Lj as in (4.1) is given as a
direct sum of pencils, then the Choi matrix C in the inclusion algorithm can be chosen with
many fewer unknowns, reflecting this structure. We start with L1.
Proposition 4.1. Suppose L1 = ⊕k
µ=1Mµ, where M1, . . . , Mk are monic linear pencils,
Mµ = I +
Bµ
ℓ xℓ,
gXℓ=1
where the Bµ
Bµ
ℓ . Then, DL1 ⊆ DL2 if and only if there exists a symmetric matrix C = ⊕k
j are of size δµ × δµ. Thus, A1,ℓ = ⊕k
pq denote the (p, q) entry of
µ=1C µ such that
ℓ . Let αℓ,µ
µ=1Bµ
∀µ = 1, . . . , k :
∀ℓ = 1, . . . , g :
(4.5)
C µ := (cµ
pq)δµ
p,q=1 (cid:23) 0,
kXµ=1
δµXp,q=1
kXµ=1
αℓ,µ
pq cµ
pq = A2,ℓ,
cµ
pp = Id2.
δµXp=1
Each cµ
pq is an unknown d2 × d2 matrix and (cµ
pq)∗ = (cµ
qp).
Consider the set C of 2k−1 matrices of the form
W ∗
1
...
W ∗
k
hW1
··· Wki
··· ±Wki .
C = W ∗W =
W ∗
1
±W ∗
±W ∗
...
±W ∗
k
2
3
hW1 ±W2 ±W3
THE MATRICIAL RELAXATION OF AN LMI
21
Proof. The inclusion DL1 ⊆ DL2 is equivalent to the existence of a Choi matrix C satisfying
the feasibility conditions (4.2) of the inclusion algorithm. Thus C is a d1 × d1 block matrix
with d2 × d2 entries. On the other hand, d1 =Pµ δµ and the matrix C can be viewed as a
block matrix C = (Ci,j)k
i,j where Ci,j is δi × δj block matrix whose entries are d2 × d2 matrices.
Observe that for i 6= j, the entries of Ci,j do not appear as part of the linear constraint in the
inclusion algorithm - they are unconstrained because our direct sum structure forces certain
αpq to be zero.
Since d1 =P δµ and the matrix C is a d1 × d1 block matrix with d2 × d2 blocks and is
positive semidefinite, hence there exist d1 × δµ block matrices Wµ having d2 × d2 entries such
that C factors as
Each C ∈ C solves the inclusion algorithm; i.e., validates DL1 ⊆ DL2. Hence the matrix
C obtained by averaging over C also validates the inclusion. Noting that, because each off
diagonal entry of C is the average of 2k−2 terms W ∗
i Wj, we get C is
the block diagonal matrix with diagonal entries W ∗
i Wj with 2k−2 terms −W ∗
j Wj, which completes the proof.
With the hypotheses of Proposition 4.1, the number of unknown variables in the LMI
inclusion algorithm are greatly reduced. Indeed, from 1
2 (d1d2 + 1)d1d2, to
1
2
kXµ=1
(d2δµ + 1)d2δµ.
The number of equality constraints is still 1
2 (1 + g)d2(d2 + 1).
A reduction in both the number of variables and equality constraints occurs if L2, the
range linear pencil, in the inclusion algorithm has a direct sum structure.
Proposition 4.2. In the inclusion algorithm, if the pencil L2 is a direct sum; i.e., L2 =
⊕k
µ=1Mµ, where each
Mµ = I +
Bµ
j xj
gX1
22
HELTON, KLEP, AND MCCULLOUGH
is a monic linear pencil of size δµ × δµ (so that P δµ = d2), then DL1 ⊆ DL2 if and only if
there exists a symmetric matrix C = ⊕k
µ=1C µ such that
∀µ = 1, . . . , k :
(4.6)
∀ℓ = 1, . . . , g, µ = 1, . . . , k :
C µ := (cµ
pq)d1
p,q=1 (cid:23) 0,
αℓ
pqcµ
pq = Bµ
ℓ ,
dµXp,q=1
dµXp=1
∀µ = 1, . . . , k :
cµ
pp = Iδµ.
Each cµ
pq is an unknown δµ × δµ matrix and (cµ
The count of unknowns is 1
pq)∗ = (cµ
qp).
gkδµ(δµ + 1) + kδµ(δµ + 1).
2Pk
µ=1(d1δµ + 1)d1δµ and of scalar equality constraints is
4.3. Tightening the relaxation. There is a general approach to tightening the inclusion
algorithm which relaxes DL1(1) ⊆ DL2(1), and thus applies to the algorithms in the section,
based upon the following simple lemma.
Lemma 4.3. Suppose L1, L2 and M are linear pencils and let M = L1⊕M. If DL1(1) ⊆ DM (1),
then
D M ⊆ DL1
and D M (1) = DL1(1).
In particular, if D M ⊆ DL2 , then DL1(1) ⊆ DL2(1).
Proof. The first part of the lemma is evident:
D M = DL1 ∩ DM ⊆ DL1.
Likewise, D M (1) = DL1(1) ∩ DM (1) = DL1(1), since DL1(1) ⊆ DM (1). For the last statement
note that D M ⊆ DL2 implies DL1(1) = D M (1) ⊆ DL2(1).
This lemma tells us applying our inclusion algorithm to M versus L2 is at least as accurate
as applying it to L1 versus M and it quite possibly is more accurate. The lemma is used in
the context of the matrix cube problem in Section 5.
4.4. Computing the radius of matricial LMI sets. Let L be a monic linear pencil,
(4.7)
L(x) = I +
gXℓ=1
Aℓxℓ ∈ SRd×dhxi.
We present an algorithm based on semidefinite programming to compute the radius of a matri-
cial LMI set DL (and at the same time check whether it is bounded). The idea is simply to use
the test in Section 4.1 to check if DL is contained in the ball of radius N . The smallest such
N will be the matricial radius, and also an upper bound on the radius of the spectrahedron
DL(1).
THE MATRICIAL RELAXATION OF AN LMI
23
Let
JN (x) =
1
N "N x∗
x N Ig# = I +
1
N
gXj=1
(E′
1,j+1 + E′
j+1,1)xj ∈ SR(g+1)×(g+1)hxi
be a monic linear pencil. Here E′
entry and zeros elsewhere.
ij the (g + 1)× (g + 1) elementary matrix with a 1 in the (i, j)
Then DL is bounded, and its matricial radius is ≤ N , if and only if DJN ⊇ DL.
The matricial radius algorithm
rsErs. Solve the SDP (RM):
Let αℓ
rs(crs)1,2
subject to
(RM1) (crs)d
r,s denote the (r, s) entry of Aℓ, that is, Aℓ =Pr,s αℓ
min b :=Pr,s α1
dPr=1
r,s=1 := C (cid:23) 0,
crr = Ig+1,
(RM2)
(RM3) ∀ℓ = 1, . . . , g, ∀p, q = 1, . . . , g + 1 :
for
αℓ
rs(crs)g+1,1
rs(crs)p,q = 0
(p, q) 6∈ {(1, ℓ + 1), (ℓ + 1, 1)},
Xr,s
rs(crs)1,2 =Pr,s α1
(RM4) Pr,s α1
rs(crs)1,g+1 =Pr,s αg
=Pr,s αg
for the unknown C; i.e., the d2 unknown (g + 1) × (g + 1) matrices (crs). If the optimal value
of (RM) is b ∈ R>0, then kXk ≤ 1
b for all X ∈ DL, and this bound is sharp.
rs(crs)1,3 =Pr,s α2
rs(crs)2,1 =Pr,s α2
This SDP is always feasible (for b =Pr,s α1
rs(crs)1,2 = 0). Clearly, DL is bounded if and
only if this SDP has a positive solution. In fact, any value of b > 0 obtained gives an upper
bound of 1
b for the norm of an element in DL. The size of the (symmetric) matrix of unknown
variables is d(g + 1) × d(g + 1) and there are 1
2 (g3 + 4g2 + 3g + 4) (scalar) linear constraints.
To reduce the number of unknowns, solve the linear system of 1
2 g(g2 + 3g − 2) equations given
in (RM3).
rs(crs)3,1 = ···
Checking boundedness of DL(1) is a classical, fairly basic semidefinite programming prob-
lem. Indeed, given a nondegenerate monic linear pencil L, DL(1) is bounded (equivalently, DL
is bounded) if and only if the following SDP is infeasible:
(Here, L(1) denotes the truly linear part of L.)
L(1)(X) (cid:23) 0,
tr(cid:16)L(1)(X)(cid:17) = 1.
However, computing the radius of DL(1) is harder. Thus our algorithm, yielding a conve-
nient upper bound on the radius, might be of broad interest, motivating us to spend more time
describing its implementation. The algorithm can be written entirely in a matricial form which
is both elegant and easy to code in MATLAB or Mathematica. The matricial component of
24
HELTON, KLEP, AND MCCULLOUGH
the algorithm is as follows. Let en denote the vector of length n with all ones, let En = en⊗ et
be the n × n matrix of all ones. Then (RM2) is (using •H for the Hadamard product)
n
while the left hand side of (RM3) can be presented as the (p, q) entry of
(cid:0)eg+1 ⊗ Id(cid:1)t(cid:0)(Id ⊗ Eg+1) •H C(cid:1)(cid:0)eg+1 ⊗ Id(cid:1) = Ig+1,
(cid:0)eg+1 ⊗ Id(cid:1)t(cid:0)(Aℓ ⊗ Eg+1) •H C(cid:1)(cid:0)eg+1 ⊗ Id(cid:1).
Equations (RM3) and (RM4) give constraints on these matrices.
As an example we computed the matricial radius of an ellipse, which for the example we
computed agrees with the scalar radius. The corresponding Mathematica notebook can be
downloaded from
http://srag.fmf.uni-lj.si/preprints/ncLMI-supplement.zip
Justification. As in the previous subsection, we need to determine whether there is a com-
pletely positive unital map τ : Rd×d → R(g+1)×(g+1) satisfying τ (Aj) = 1
j+1,1) for
some N . The Choi matrix here is C = (τ (Eij))i,j ∈ (R(g+1)×(g+1))d×d. Let Aℓ =Pr,s αℓ
rsErs.
Then the linear constraints we need to consider say that
1,j+1 + E′
N (E′
τ (Aℓ) =Xr,s
αℓ
rscrs
has all entries 0 except for the (1, ℓ + 1) and (ℓ + 1, 1) entries which are the same; indeed they
are all equal to 1
N . Thus we arrive at the feasibility SDP (RM) above.
4.5. The matricial matrix cube problem. This section describes our matricial matrix cube
algorithm - a test for inclusion of the matricial matrix cube (as defined below) into a given
LMI set. Variations on the algorithm and an analysis of the connection between this algorithm
and the matrix cube algorithm of [B-TN02] is the subject of Section 5.
Let L ∈ SRd×dhxi be a monic linear pencil as in (4.7). We present an algorithm that
computes the size ρ of the biggest matricial cube contained in DL. That is, ρ ∈ R is the largest
if n ∈ N and X ∈ (SRn×n)g satisfies kXik ≤ ρ for all
number with the following property:
i = 1, . . . , g, then X ∈ DL. When Xi is in R1×1 this is the classical matrix cube problem
(cf. Ben-Tal and Nemirovski [B-TN02]), which they show is NP-hard.
First we need an LMI which defines the cube. Let
Cρ(x) =
1
ρ(cid:16)(cid:0) ⊕g
j=1 ρ − xj(cid:1)M(cid:0) ⊕g
j=1 ρ + xj(cid:1)(cid:17) ∈ SR2g×2ghxi.
j=1(Ejj − Eg+j,g+j)xj, where Ei,j is a the elementary 2g × 2g matrix
ρPg
Then Cρ(x) = I + 1
with a 1 in the (i, j) entry and zeros elsewhere, and
DCρ = [n∈N(cid:8)X ∈ (SRn×n)g kXik ≤ ρ for all i = 1, . . . , g(cid:9) .
This is a matricial cube. Our algorithm uses the test in Section 4.1 to compute the largest ρ
with DCρ ⊆ DL. It also takes advantage of the fact that Cρ is a direct sum (of scalar-valued
THE MATRICIAL RELAXATION OF AN LMI
25
pencils) by using Proposition 4.1 with k = 2g, δµ = 1 and d2 = d. This immediately gives rise
to the following SDP:
max ρ
subject to
j = 1, . . . , 2g
(preMC1) C j (cid:23) 0,
(preMC2) ∀j = 1, . . . , g : C j − C g+j = ρAj.
(preMC3)
C j = Id,
2gPj=1
Each of the 2g symmetric matrices C j is in SRd×d.
Next we make this algorithm more efficient by solving the equality constraints (preMC2)
to eliminate C g+1, . . . , C 2g and (preMC3) to obtain
(4.8)
C g =
1
2(cid:16)I − 2
g−1Xj=1
C j + ρ
gXj=1
Aj(cid:17)
With this, the above SDP reduces to
The matricial matrix cube algorithm
max ρ
subject to
(MC1) C j (cid:23) 0,
(MC2) C j (cid:23) ρAj
j = 1, . . . , g − 1
(MC3) Id − 2Pg−1
where each of the g − 1 symmetric matrices C j is in SRd×d.
j C j +Pg−1
ρAj ± ρAg (cid:23) 0,
j
This SDP is always feasible (with ρ = 0). If its optimal value is ρ > 0, then DCρ ⊆ DL, and
the obtained upper bound for the size of the matricial cube is sharp. There are 1
2 (g− 1)d(d + 1)
variables and all of the linear equality constraints have been eliminated. There are 2g matrix
inequality constraints.
Example 4.4. Consider finding the largest square embedded inside the unit disk. We consider
the two pencils ∆, Γ from Example 3.1, each of which represents the unit disk, since D∆(1) =
DΓ(1) = {(X1, X2) ∈ R2 X 2
(1) is the maximal square
contained in the unit disk D∆(1). Indeed the biggest matricial cube in D∆ is DC√2/2
, but the
biggest matricial cube in DΓ is DC 1
. For details, see the Mathematica notebook available at
2 ≤ 1}. It is clear that DC√2/2
1 + X 2
2
http://srag.fmf.uni-lj.si/preprints/ncLMI-supplement.zip
We will revisit this example in Section 5.
(cid:3)
Justification. A justification for the matrix cube algorithm based on the pre-algorithm
has already been given. So it suffices to justify the pre-matricial matrix cube algorithm.
Let Bj := Ej,j − Eg+j,g+j ∈ SR2g×2g. Taking advantage of the fact that Cρ is the direct
sum of 2g scalar linear pencils, we want to determine the biggest ρ for which there exists a
26
HELTON, KLEP, AND MCCULLOUGH
j=1(cj ) ∈ ⊕2g
completely positive unital map τ : ⊕2g
1 R1×1 → Rd×d satisfying τ (Bj) = ρAj, j = 1, . . . , g.
Suppose C = ⊕2g
1 (Rd×d)1×1 (because each cj is a 1 × 1 block matrix whose entries
are (symmetric) d × d matrices, there is no need for the indexing cj
pq) is the corresponding
Choi matrix as in Proposition 4.1. Then the linear constraint τ (Bj) = ρAj translates into
1,1 − cg+j
cj
4.6. Minimal pencils and the Silov ideal. This section describes an algorithm aimed at
constructing from a given pencil L a pencil L of minimal size with DL = D L.
4.6.1. Minimal pencils. Let L be a monic linear pencil,
1,1 = ρAj which is (preMC2).
(4.9)
L(x) = I +
gXℓ=1
Aℓxℓ ∈ SRd×dhxi
with bounded DL. We present a probabilistic algorithm based on semidefinite programming
that computes a minimal pencil L with the same matricial LMI set.
The two-step procedure goes as follows.
In Step 1, one uses the decomposition of a
semisimple algebra into a direct sum of simple algebras, a classical technique in computational
algebra, cf. Friedl and Ro´nyal [FR85], Eberly and Giesbrecht [EG96], or Murota, Kanno,
Kojima, and Kojima [MKKK10] for a recent treatment. This yields a unitary matrix U ∈ Rd×d
that simultaneously transforms the Aℓ into block diagonal form, that is,
For each j, the set {I, Bj
pencils
1, . . . , Bj
j=1Bj
ℓ
U ∗AℓU = ⊕s
g} generates a simple real algebra. Define the monic linear
for all
ℓ.
Lj(x) = I +
Bj
ℓ xℓ, L′(x) = U ∗L(x)U = ⊕s
j=1Lj(x).
gXℓ=1
Given ℓ, let Lℓ = ⊕j6=ℓLj. If there is no ℓ such that
LℓD Lℓ (cid:23) 0,
(this can be tested using SDP as explained in Section 4.1) then the pencil is minimal. If there
is such an ℓ remove the (one) corresponding block from L′ to obtain a new pencil and repeat
the process. Once we have no more redundant blocks in L′, the obtained pencil L is minimal,
and satisfies D L = DL by construction.
4.6.2. Silov ideal. Thus subsection requires material from Section 3. Using our results from
Section 3.3 (cf. Proposition 3.17) and Section 4.1, one can compute the Silov ideal of a unital
matrix algebra A generated by symmetric matrices A1, . . . , Ag ∈ SRd×d. Form the monic linear
pencil
and compute the minimal pencil
L = I +X Aℓxℓ ∈ SRd×dhxi,
L = I +X Aℓxℓ
THE MATRICIAL RELAXATION OF AN LMI
27
as in the previous subsection. If
then the kernel of the canonical unital map
S = span{I, Aℓ ℓ = 1, . . . , g},
A → C ∗( S), Aℓ 7→ Aℓ
is the Silov ideal of A.
5. More on the matrix cube problem
This section provides perspective on the inclusion algorithm by focusing on the matrix
cube problem. The first subsection shows that the estimate based on the inclusion algorithm,
namely the matricial matrix cube algorithm of Subsection 4.5, is essentially identical to that
obtained by the algorithm of Ben-Tal and Nemirovski in [B-TN02].
Subsection 5.2, illustrates the tightening procedure of Lemma 4.3 on the matricial cube.
5.1. Comparison with the algorithm in [B-TN02]. Let L = I +Pg
ℓ=1 Aℓxℓ ∈ SRd×dhxi be
a monic linear pencil and recall the pencil Cρ and its corresponding positivity domain DCρ , the
matricial cube. In [B-TN02] the verifiable sufficient condition for the inclusion Dρ(1) ⊆ DL(1)
is the following: Suppose there exist symmetric matrices B1, . . . Bg such that
(S)
Bj (cid:23) ±Aj for all j = 1, 2, . . . , g;
and I − ρXj
Bj (cid:23) 0
holds. Then DCρ(1) ⊆ DL(1).
The following proposition says that the estimate of the largest cube contained in a given
spectrahedron given by the matricial relaxation based upon the matricial matrix cube algorithm
is the same as that based upon condition (S).
Proposition 5.1. Given ρ ∈ R≥0, condition (S) holds if and only if DCρ ⊆ DL. Moreover,
there is an explicit formula for converting condition (S) to a feasible point for the matricial
matrix cube algorithm and vice-versa.
Proof. Suppose we have found the optimal ρ and the corresponding C j for j = 1, . . . , 2g, in
the matricial matrix cube algorithm. From (preMC2) ρAj = C j − C g+j. Set
ρBj := C j + C g+j.
From (preMC1),
(5.1)
and
ρ(Bj − Aj) = 2C j (cid:23) 0
ρ(Bj + Aj) = 2C j+g (cid:23) 0.
Relation (preMC3) gives I = ρPj Bj. Thus we see Bj and ρ satisfy (S).
Conversely, suppose Bj, ρ are a solution to (S). Solve (5.1) for C j, C g+j. It is straightfor-
ward to check that these C j satisfy the conditions (preMCj) for j = 1, 2, 3.
28
HELTON, KLEP, AND MCCULLOUGH
Now that we know the estimate provided by our relaxation is the same as that of algorithm
(S), we look at the computational cost. (S) has 1
2 gd(d + 1) unknowns and the number of d × d
matrix inequality constraints is (2g + 1). As we saw, our matricial matrix cube algorithm had
1
2 (g − 1)d(d + 1) unknowns and 2g matrix (d × d) inequality constraints, so the costs are a
bit less than those of (S). However, (S) can be improved easily by the general trick in the
following remark which removes an unknown and a constraint, thus making the cost of (S) the
same as ours.
Remark 5.2. If, in the A1,ℓ in the inclusion algorithm of Section 4.1 all have trace 0, then the
p=1 cpp (cid:23) 0, since then
. When presented with the inequality form
p=1 cpp = 0 is equivalent to the inequality ∆ = Id2 −Pd1
condition Id2 −Pd1
the cpp can, without harm, be replaced by cpp + ∆
d1
we could convert it to equality, then eliminate one variable by solving for it.
5.2. The lattice of inclusion algorithm relaxations for the matrix cube. A virtue
of the general method of the inclusion algorithm based on matricial relaxations is that, as
alluded to in Section 4.3, it allows tightening in order to improve estimates (with added cost).
This subsection discusses and illustrates properties of this tightening procedure, mainly as
an introduction to a topic that might merit further study. We do show in an example that
tightening can produce an improved estimate.
5.2.1. General theory. The operator theory upon which this paper is based, when converted
to the language of LMIs, contains a theory of matricial relaxations of a given LMI set and thus
provides a general framework containing the tightening methods described in Section 4.3.
Suppose S ⊆ Rg is an LMI set; i.e., suppose there is a monic linear pencil Λ such that
S = DΛ(1). The collection LS of all monic linear pencils L with S = DL(1) is naturally
ordered by inclusion. Namely, if L, M ∈ LS, then L ≥ M if, DL(n) ⊆ DM (n) for every positive
integer n. If Λ′ is also a monic linear pencil and S ⊆ S′ = DΛ′, then the matricial inclusion
DM ⊆ DΛ′ implies the inclusion DL ⊆ DΛ′. Thus, the pencil L gives at least as good a test for
the inclusion S ⊆ S′, than does M . Similarly, if M ′, L′ ∈ LS′ and L′ ≥ M ′, then M ′ gives at
least as good a test for the inclusion of S into S′ as does L′.
If LS has a maximal element, denote it by Lmax and similarly for Lmin. Generally, LS
will not have either a minimal or maximal element; however, it turns out for the matrix
cube there is a minimal element. See Proposition 5.3 below. Further, in general by dropping
the requirement that the pencils L have matrix coefficients and instead allowing for operator
coefficients, it is possible to prove that Lmax and Lmin exists. (The discussion in [Pau02] on
max and min operator space structures is easily seen to carry over to the present setting.)
Thus, though typically not practical, the matricial relaxation for the inclusion of the set
S = DΛ(1) into T = DΛ′ based upon using Lmax in place of Λ produces the exact answer;
whereas the matricial relaxation based upon DLmin produces the most conservative estimate
over all possible matricial relaxations.
THE MATRICIAL RELAXATION OF AN LMI
29
5.2.2. Lmin and Lmax for the matrix cube. From the next proposition it follows that for the
matrix cube DC1 (1) the minimal pencil Lmin is C1.
Proposition 5.3. If M is a monic linear pencil and DM (1) ⊆ DCρ(1), then DM ⊆ DCρ . In
particular, if DM (1) = DCρ (1), and L is a monic linear pencil for which DCρ ⊆ DL, then
DM ⊆ DL. Hence if DM (1) = DC1 (1), then the inclusion DM ⊆ DL is at least as good a test
for DL to contain the unit cube as the inclusion DC1 ⊆ DL.
Proof. Write M = I −P Ajxj. The condition DM (1) ⊆ DCρ(1) implies, if
1
ρX Ajxj (cid:22) I,
for some xj ∈ R,
then xj ≤ 1 for each 1 ≤ j ≤ g.
Now suppose that X = (X1, . . . , Xg) ∈ (SRn×n)g and
ρX Aj ⊗ Xj (cid:22) ρI.
1
For each vector f and unit vector x (of the appropriate sizes), it follows that
With x fixed, varying f shows that
1
ρXhAjf, fihXjx, xi ≤ kfk2.
ρX AjhXj x, xi (cid:22) I.
1
It now follows that hXj x, xi ≤ 1 for each j and unit vector x. Hence, kXjk ≤ 1 for each j
and hence X ∈ DCρ(n) and the proof is complete.
We have not computed Lmax for the matrix cube (g > 2 variables), but we have for the
matrix square (g = 2 variables) and found it to be a pencil with operator (infinite dimensional)
coefficients. We do not give the calculation in this paper, rather we content ourselves with
the simple, and natural, example below which suffices to show that there are in fact choices of
M in Proposition 5.3 which do lead to improved estimates for the matrix cube problem and
that Lmin and Lmax are different for the cube. Of course, any such improved estimate comes
with additional computational cost; and, because it is in only two variables (where solving four
LMIs gives the exact answer), the example is purely illustrative.
Given η = (s, t) ∈ R2 with s2 + t2 = 1, let
A1(η) ="s
0
0 −s# , A2(η) ="0 t
t 0# ,
Lη = I +X Aj(η)xj .
Recall, a unitary matrix which is symmetric is called a signature matrix. Up to scaling the
Aj(η) are signature matrices, and further (A1(η) ± A2(η))2 = I.
It is straightforward to show that DLη (1) contains the unit square; i.e., DC1(1) ⊆ DLη (1).
Hence, with Mη = C1 ⊕ Lη,
DMη (1) = DC1(1)
30
HELTON, KLEP, AND MCCULLOUGH
and at the same time,
(5.2)
DMη ⊆ DC1.
On the other hand, for η 6= (±1, 0) or (0,±1), it is possible to check by hand that DC1(n) 6⊆
DLη (n) for each n ≥ 2. Indeed, let X = (X1, X2) denote the tuple of 2 × 2 matrices
X1 ="1
0 −1# ,
0
and X2 ="0 1
1 0# .
Then (X1, X2) ∈ DC1(2), but (X1, X2) /∈ DLη (2). Hence, the inclusion in equation (5.2) is
proper.
Another way to see the inclusion is proper is to verify that the extreme points X = (X1, X2)
of DC1(n) are exactly the pairs of n × n signature matrices X1, X2. On the other hand, for
η 6∈ {(±1, 0), (0,±1)}, the extreme points X = (X1, X2) of DC1(n) which are also in DLη (n)
are precisely the pairs of n × n signature matrices X1, X2 which commute.
Example 5.4 (Example 4.4 revisited). Recall the pencil Γ from Example 3.1. In Example 4.4
we employed the matricial matrix cube relaxation DCρ ⊆ DΓ to obtain a lower bound of 1
2 for
the biggest square inside DΓ(1). To tighten the relaxation we direct sum Lη to Cρ to obtain a
linear pencil Mη. Hand calculations for this problem tell us that with η = (√2/2,√2/2) we
obtain the exact relaxation DMη ⊆ DΓ. However, suppose we did not know this and ask: will
selecting η without much care give a reasonable improvement?
We made 100 runs with random η and considered the inclusion ρDMη ⊆ DΓ for each η
and found the average value for ρ to be approximately 0.6. This is a considerable improvement
over 0.5 obtained in the untightened problem.
(cid:3)
6. Positivstellensatze on a matricial LMI set
We give an algebraic characterization of symmetric polynomials p in noncommuting vari-
ables with the property that p(X) is positive definite for all X in a bounded matricial LMI
set DL. The conclusion of this Positivstellensatz is stronger than previous ones because of the
stronger hypothesis that DL is an LMI set. If the polynomial p is linear, then an algebraic
characterization is given by Theorem 1.1. We shall use the linear Positivstellensatz, Corollary
3.7, to prove that the quadratic module associated to a monic linear pencil L with bounded
DL is archimedean. Thereby we obtain a Putinar-type Positivstellensatz [Put93] without the
unpleasant added "bounding term". In this section, for simplicity of presentation we stick to
polynomials on the free ∗-algebra. Later in Section 7 we give this improved type of Positivstel-
lensatz on general ∗-algebras, a few examples being commuting variables, free variables and free
symmetric variables (this section). The material here is motivated by the study of positivity
of matrix polynomials in commuting variables undertaken in [KS10]; see also [HM04].
To state and prove our next string of results, we need to introduce notation pertaining to
words and polynomials in noncommuting variables.
THE MATRICIAL RELAXATION OF AN LMI
31
6.1. Words and NC polynomials. Given positive integers n, d, d′ and g, let Rd′×d denote
the d′ × d matrices with real entries and (Rn×n)g the set of g-tuples of real n × n matrices.
We write hxi for the monoid freely generated by x = (x1, . . . , xg), i.e., hxi consists of
words in the g noncommuting letters x1, . . . , xg (including the empty word ∅ which plays the
role of the identity 1). Let Rhxi denote the associative R-algebra freely generated by x, i.e.,
the elements of Rhxi are polynomials in the noncommuting variables x with coefficients in R.
Its elements are called (NC) polynomials. An element of the form aw where 0 6= a ∈ R
and w ∈ hxi is called a monomial and a its coefficient. Hence words are monomials whose
coefficient is 1. Endow Rhxi with the natural involution fixing R ∪ {x} pointwise. The
involution reverses words. For example, (3 − 2x2
1x2x3)∗ = 3 − 2x3x2x2
1.
6.1.1. NC matrix polynomials. More generally, for an abelian group R we use Rhxi to denote
the abelian group of all (finite) sums of monomials in hxi. Besides R = R, the most important
example is R = Rd′×d giving rise to NC matrix polynomials. If d′ = d, i.e., R = Rd×d, then Rhxi
is an algebra, and admits an involution fixing {x} pointwise and being the usual transposition
on Rd×d. We also use ∗ to denote the canonical mapping Rd′×dhxi → Rd×d′hxi.
A matrix NC polynomial is an NC polynomial with matrix coefficients, i.e., an element
of Rd′×dhxi for some d′, d ∈ N.
6.1.2. Polynomial evaluations. If p ∈ Rd′×dhxi is an NC polynomial and X ∈ (Rn×n)g, the
evaluation p(X) ∈ Rd′n×dn is defined by simply replacing xi by Xi. For example, if p(x) =
Ax1x2, where
Similarly, if p(x) = A and X ∈ (Rn×n)g, then p(X) = A ⊗ In.
Most of our evaluations will be on tuples of symmetric matrices X ∈ (SRn×n)g; our invo-
lution fixes the variables x element-wise, so only these evaluations give rise to ∗-representations
of NC polynomials.
6.2. Archimedean quadratic modules and a Positivstellensatz. In this subsection we
use the linear Positivstellensatz (Corollary 3.7) to prove that linear pencils with bounded
LMI sets give rise to archimedean quadratic modules. This is then used to prove a (nonlinear)
Positivstellensatz for matrix NC polynomials positive (semi)definite on bounded matricial LMI
sets.
then
p "0 1
1 0# ,"1
3
0# ,
A ="−4 2
0 −1#! =
0 −1#! = A ⊗ "0 1
1 0#"1
0
0
4
0
0 −2
0
−4
0
2
0 −3 0
0
0
0
3
0
.
32
HELTON, KLEP, AND MCCULLOUGH
Theorem 6.1. Suppose L ∈ SRd×dhxi is a monic linear pencil and DL is bounded. Then
for every symmetric polynomial f ∈ Rℓ×ℓhxi with fDL ≻ 0, there are Aj ∈ Rℓ×ℓhxi, and
Bk ∈ Rd×ℓhxi satisfying
(6.1)
f =Xj
Corollary 6.2. Keep the assumptions of Theorem 6.1. Then for a symmetric polynomial
f ∈ Rℓ×ℓhxi the following are equivalent:
(i) fDL (cid:23) 0;
(ii) for every ε > 0 there are Aj ∈ Rℓ×ℓhxi, and Bk ∈ Rd×ℓhxi satisfying
(6.2)
j Aj +Xk
kLBk.
B∗
B∗
A∗
kLBk.
f + ε =Xj
A∗
j Aj +Xk
Proof. Obviously, (ii) ⇒ (i). Conversely, if (i) holds, then f + εDL ≻ 0 and we can apply
Theorem 6.1.
We emphasize that convexity of DL implies concrete bounds on the size of the matrices
X ∈ DL that need to be plugged into f to check whether fDL ≻ 0:
Proposition 6.3 (cf. [HM04, Proposition 2.3]). Let L ∈ SRd×dhxi be a linear pencil with DL
bounded, and let f = f ∗ ∈ Rn×nhxi be of degree m. Set s := nPm
(1) fDL ≻ 0 if and only if fDL(s) ≻ 0;
(2) fDL (cid:23) 0 if and only if fDL(s) (cid:23) 0.
Proof. In both statements the direction (⇒) is obvious. If fDL 6≻ 0, there is an ℓ, X ∈ DL(ℓ)
and v = ⊕n
j=1vj ∈ (Rℓ)n with
j=0 gj . Then:
hf (X)v, vi ≤ 0.
Let
Clearly, dimK ≤ nPm
K := {w(X)vj w ∈ hxi is of degree ≤ m, j = 1, . . . , n}.
j=0 gj = s. Let P be the orthogonal projection of Rℓ onto K. Then
hf (P XP )v, vi = hf (X)v, vi ≤ 0.
Since P XP ∈ DL(s), this proves (1). The proof of (2) is the same.
The crucial step in proving Theorem 6.1 is observing that the quadratic module generated
by L in Rℓ×ℓhxi is archimedean. This is essentially a consequence of Corollary 3.7, i.e., of the
linear Positivstellensatz as we now demonstrate.
Definition 6.4. Let A be a ring with involution a 7→ a∗ and set SymA := {a ∈ A a = a∗}.
A subset M ⊆ SymA is called a quadratic module in A if
1 ∈ M, M + M ⊆ M and a∗M a ⊆ M for all a ∈ A.
THE MATRICIAL RELAXATION OF AN LMI
33
We will be mostly interested in the case A = Rℓ×ℓhxi. In this case given a subset S ⊆
S generated by S in Rℓ×ℓhxi is the smallest subset of
Sym Rd×dhxi, the quadratic module M ℓ
Sym Rℓ×ℓhxi containing all a∗sa for s ∈ S ∪ {1}, a ∈ Rd×ℓhxi, and closed under addition:
M ℓ
S =n NXi=1
a∗
i siai N ∈ N, si ∈ S ∪ {1}, ai ∈ Rd×ℓhxio.
This notion extends naturally to quadratic modules generated by S ⊆Sd∈N Sym Rd×dhxi.
Definition 6.5. A quadratic module M of a ring with involution A is archimedean if
(6.3)
To a quadratic module M ⊆ SymA we associate its ring of bounded elements
∀a ∈ A ∃N ∈ N : N − a∗a ∈ M.
HM (A) := {a ∈ A ∃N ∈ N : N − a∗a ∈ M}.
A quadratic module M ⊆ SymA is thus archimedean if and only if HM (A) = A.
The name ring of bounded elements is justified by the following proposition:
Proposition 6.6 (Vidav [Vid59]). Let A be an R-algebra with involution, and M ⊆ SymA a
quadratic module. Then HM (A) is a subalgebra of A and is closed under the involution.
Hence it suffices to check the archimedean condition (6.3) on a set of algebra generators.
Lemma 6.7. A quadratic module M ⊆ Rℓ×ℓhxi is archimedean if and only if there exists
N ∈ N with N − x∗x = N −Pi x2
Proof. The "only if" direction is obvious. For the converse, observe that Rℓ×ℓhxi is generated
as an R-algebra by x and the ℓ × ℓ matrix units Eij, i, j = 1, . . . , ℓ. By assumption,
i ∈ M .
i ) +Xj6=i
so xj ∈ HM (Rℓ×ℓhxi) for every j. On the other hand, E∗
N − x2
x2
x2
j ∈ M,
i = (N −Xi
ijEij =Xk6=j
1 − E∗
ijEij = Ejj and thus
E∗
kkEkk ∈ M.
Hence by Proposition 6.6, HM (Rℓ×ℓhxi) = Rℓ×ℓhxi so M is archimedean.
We are now in a position to give our crucial observation.
Proposition 6.8. Suppose L ∈ SRd×dhxi is a monic linear pencil and DL is bounded. Then
the quadratic module M ℓ
{L} generated by L in Rℓ×ℓhxi is archimedean.
To make the proof more streamlined we separate one easy argument into a lemma:
Lemma 6.9. For S ⊆Sd∈N Sym Rd×dhxi the following are equivalent:
S is archimedean for some ℓ ∈ N;
(i) M ℓ
34
HELTON, KLEP, AND MCCULLOUGH
(ii) M ℓ
S is archimedean for all ℓ ∈ N.
Proof. (ii) ⇒ (i) is obvious. For the converse assume (i) and let p ∈ N be arbitrary. By
assumption, there is N ∈ N with (N − x∗x)Iℓ ∈ M ℓ
denote the s × q matrix units,
then
11 ∈ M p
S.
S for all j concluding the proof
)∗(N − x∗x)IℓE(ℓ,p)
Now using permutation matrices we see (N − x∗x)E(p,p)
by the additivity of M p
S.
(N − x∗x)E(p,p)
11 = (E(ℓ,p)
S. If E(s,q)
∈ M p
11
jj
ij
Proof of Proposition 6.8. Since DL is bounded, there is N ∈ N with N ≥ kXk for all X ∈ DL.
Consider the (g + 1) × (g + 1) monic linear pencil
x N Ig# ∈ SR(g+1)×(g+1)hxi.
By taking Schur complements, we see JN (X) (cid:23) 0 if and only if N − 1
j ≥ 0, i.e., if and
only if kXk ≤ N . This means JNDL (cid:23) 0 and so by Corollary 3.7 (in the new terminology),
JN ∈ M g+1
{L} is closed under ∗-conjugation, we obtain
N "N x∗
NPj X 2
{L} . Since M g+1
JN (x) =
1
0
"N
N x 1# NJN (x)" 1
N x 1#∗
− 1
− 1
Again, using permutation matrices leads to (N 2g − x∗x)Ig+1 ∈ M g+1
M g+1
{L} is archimedean. Finally, Lemma 6.9 implies M ℓ
{L} is archimedean.
N x∗x(cid:1)Ig# =" 1
(cid:0)N − 1
0
0
0
∈ M g+1
{L} .
{L} . By Proposition 6.6,
Corollary 6.10. For a monic linear pencil L the following are equivalent:
(i) DL(1) is bounded;
(ii) the quadratic module M ℓ
(iii) the quadratic module M ℓ
{L} is archimedean for some ℓ ∈ N;
{L} is archimedean for all ℓ ∈ N.
Proof. Clearly, (iii) ⇒ (ii) ⇒ (i). On the other hand, (i) is equivalent to DL being bounded
by Proposition 2.4, so Proposition 6.8 applies and allows us to deduce (iii).
{L}. Now that the archimedeanity of the quadratic module M ℓ
Proof of Theorem 6.1; compare [HM04, Proposition 4.1]. The statement (6.1) holds if and only
if f ∈ M ℓ
{L} has been established
in Proposition 6.8, the proof is classical. We only list basic steps and refer the reader to [HM04]
for detailed proofs.
The proof is by contradiction, so assume f 6∈ M ℓ
{L} is equivalent
to the existence of an order unit (also called algebraic interior point), namely 1, of the convex
{L} ⊆ Sym Rℓ×ℓhxi. Thus the Eidelheit-Kakutani separation theorem yields a linear
cone M ℓ
map ϕ : Rℓ×ℓhxi → R satisfying
{L}. Archimedeanity of M ℓ
ϕ(f ) ≤ 0
and ϕ(M ℓ
{L}) ⊆ R≥0.
THE MATRICIAL RELAXATION OF AN LMI
35
Modding out N := {f ∈ R1×ℓhxi ϕ(p∗p) = 0} out of R1×ℓhxi leads to a vector space H0
and ϕ induces a scalar product
h , i : H0 × H0 → R,
(¯p, ¯q) 7→ ϕ(q∗p).
Completing H0 with respect to this scalar product yields a Hilbert space H. It is nonzero
sincePihei, eii = ϕ(1) = 1, where ei are the matrix units of R1×ℓ. Let e = ⊕ei ∈ Hℓ.
The induced left regular ∗-representation π : Rhxi → B(H) is bounded (since M ℓ
{L} is
:= π(Xi) and X := ( X1, . . . , Xg). The constructed scalar product
archimedean). Let Xi
extends naturally to Hℓ. For every ¯p ∈ Hℓ
0, we have
ϕ(p∗
kL(x)j,kpj) = ϕ(p∗L(x)p) ≥ 0,
hL( X)j,k ¯pj, ¯pki =Xj,k
hL( X)¯p, ¯pi =Xj,k
{L}. Hence X ∈ DL. But now
where p has been identified with a ℓ× ℓ matrix polynomial and the last inequality results from
p∗L(x)p ∈ M ℓ
a contradiction.
0 ≥ ϕ(f ) = hf ( X)e, ei > 0,
The cautious reader will have noticed that the constructed X leading to the contradiction
was (in general) not acting on a finite dimensional Hilbert space. However this is only a slight
technical difficulty; we refer the reader to Proposition 6.3 or [HM04, Proposition 2.3] for a
remedy.
6.3. More constraints. Additional constraints can be imposed on elements of a matricial
LMI set. Given S ⊆Sd∈N Sym Rd×dhxi, define
DS = [n∈N
and let
DS(n)
DS(n) = {X ∈ (SRn×n)g ∀s ∈ S : s(X) (cid:23) 0},
denote the (matrix) positivity domain. Also of interest is the operator positivity domain
S =(cid:8)X ∈ SymB(H)g ∀s ∈ S : s(X) (cid:23) 0(cid:9).
D∞
Here H is a separable Hilbert space, and SymB(H) is the set of all bounded symmetric oper-
ators on H.
Theorem 6.11. Suppose L ∈ SRd×dhxi is a monic linear pencil and DL is bounded. Let
gj ∈ Sym Rdj ×djhxi (j ∈ N) be symmetric matrix polynomials. Then for every f ∈ Sym Rℓ×ℓhxi
with fD∞
{L, gjj∈N}.
{L, gjj∈N} ≻ 0, we have f ∈ M ℓ
Proof. Since the quadratic module M ℓ
Theorem 6.1 applies.
{L, gjj∈N} ⊇ M ℓ
{L} is archimedean, the same proof as for
36
HELTON, KLEP, AND MCCULLOUGH
Remark 6.12. For a particularly appealing consequence (in commuting variables) of Theorem
6.11 see Section 7.1.
We conclude this section with a Nichtnegativstellensatz.
It is a stronger form of the
Nirgendsnegativsemidefinitheitsstellensatz [KS07] for matricial LMI sets.
Corollary 6.13. Let L ∈ SRd×dhxi be a monic linear pencil and suppose DL is bounded. Let
gj ∈ Sym Rdj ×djhxi (j ∈ N) be symmetric matrix polynomials. Then for every h ∈ Sym Rℓ×ℓhxi
the following are equivalent:
(i) hD∞
{L, gjj∈N} 6(cid:22) 0, i.e., for every (nontrivial separable) Hilbert space H and tuple of sym-
{L, gjj∈N} on H, there is a v ∈ H with hh(X)v, vi > 0;
metric bounded operators X ∈ D∞
(ii) there are Dj ∈ Rℓ×ℓhxi satisfying
X D∗
(6.4)
j hDj ∈ Iℓ + M ℓ
{L, gjj∈N}.
{L, −h, gjj∈N} = ∅.
Proof. (ii) ⇒ (i) is obvious. The converse is also easy. Just apply Theorem 6.11 with f = −1
and the positivity domain D∞
Remark 6.14. There does not seem to exist a clean linear Nichtnegativstellensatz. We found
4 × 4 monic linear pencils L1, L2 in nine variables with the following properties:
(1) DL1 and DL2 are bounded;
(2) L2DL1 6(cid:22) 0, or equivalently,
0
(X ∈ R9(cid:12)(cid:12)(cid:12) "L1(X)
j L2(x)Uj = I +Xℓ
U ∗
0
−L2(X)# (cid:23) 0) = ∅;
ℓ Wℓ +Xk
W ∗
V ∗
k L1(x)Vk.
(3) there do not exist real matrices Uj, Vk, Wℓ with
(6.5)
Xj
By Corollary (6.13), (1) and (2) imply that (6.5) holds with Uj, Vk, Wℓ ∈ R4×4hxi. A Mathe-
matica notebook with all the calculations is available at http://srag.fmf.uni-lj.si.
7. More general Positivstellensatze
In this section we present two possible modifications of our theory. First, we apply our
techniques to commuting variables and derive a "clean" classical Putinar Positivstellensatz on
a bounded spectrahedron. This is done by adding symmetrized commutation relations to our
list of constraints. In fact we can add any symmetric relation and get a clean Positivstellensatz
on a subset of a bounded LMI set (this is Theorem 7.4). In Section 7.2 we also show how to
deduce similar results for nonsymmetric noncommuting variables.
THE MATRICIAL RELAXATION OF AN LMI
37
7.1. Positivstellensatze on an LMI set in Rg. We adapt some of our previous definitions
to commuting variables. Let [y] be the monoid freely generated by y = (y1, . . . , yg), i.e., [y]
consists of words in the g commuting letters y1, . . . , yg (including the empty word ∅ which
plays the role of the identity 1). Let R[y] denote the commutative R-algebra freely generated
by y, i.e., the elements of R[y] are polynomials in the commuting variables y with coefficients
in R.
More generally, for an abelian group R we use R[y] to denote the abelian group of all R-
linear combinations of words in [y]. Besides R = R, the most important example is R = Rd′×d
giving rise to matrix polynomials. If d′ = d, i.e., R = Rd×d, then R[y] is an R-algebra, and
admits an involution fixing {y} pointwise and being the usual transposition on Rd×d. We also
use ∗ to denote the canonical mapping Rd′×d[y] → Rd×d′[y]. If p ∈ Rd′×d[y] is a polynomial
and Y ∈ Rg, the evaluation p(Y ) ∈ Rd′×d is defined by simply replacing yi by Yi.
The natural map hxi → [y] is called the commutative collapse. It extends naturally to
matrix polynomials.
For A0, A1, . . . , Ag ∈ SRd×d, a linear matrix polynomial
(7.1)
L(y) = A0 +
gXj=1
Ajyj ∈ SRd×d[y],
is a linear pencil. If A0 = I, then L is monic. If A0 = 0, then L is a truly linear pencil. Its
spectrahedron is
and for every ℓ ∈ N, L induces a quadratic module Qℓ
{L} in Rℓ×ℓ[y]:
DL(1) = {Y ∈ Rg L(Y ) (cid:23) 0},
Qℓ
{L} =n NXi=1
a∗
i ai +
NXj=1
b∗
j Lbj N ∈ N, ai ∈ Rℓ×ℓ[y], bj ∈ Rd×ℓ[y]o.
All the results on linear pencils and archimedeanity given in Section 6 carry over to the
commutative setting. For instance, given a monic linear pencil L ∈ SRd×d[y], we have:
(1) Qℓ
{L} is archimedean for all ℓ ∈ N;
(2) Qℓ
{L} is archimedean for some ℓ ∈ N if and only if Qℓ
{L} is archimedean if and only if the spectrahedron DL(1) is bounded.
Most importantly, we obtain the following clean version of Putinar's Positivstellensatz
[Put93] on a bounded spectrahedron.
Theorem 7.1. Suppose L ∈ SRd×d[y] is a monic linear pencil and DL(1) is bounded. Then
for every symmetric polynomial f ∈ Rℓ×ℓ[y] with fDL(1) ≻ 0, there are Aj ∈ Rℓ×ℓ[y], and
Bk ∈ Rd×ℓ[y] satisfying
(7.2)
B∗
kLBk.
f =Xj
A∗
j Aj +Xk
38
HELTON, KLEP, AND MCCULLOUGH
Proof. Let F ∈ Sym Rℓ×ℓhxi be an arbitrary symmetric matrix polynomial in noncommuting
variables whose commutative collapse is f . By abuse of notation, let L ∈ SRd×dhxi be the
canonical lift of L ∈ SRd×d[y]. Write
gij = −(xixj − xjxi)∗(xixj − xjxi) = (xixj − xjxi)2 ∈ Sym Rhxi
for i, j = 1, . . . , g. Note gij(X) (cid:23) 0 if and only if XiXj = XjXi. By the spectral theorem,
FD{L, giji,j=1,...,g} ≻ 0. So Theorem 6.11 implies and yields
{L, giji,j=1,...,g}.
F ∈ M ℓ
Applying the commutative collapse gives f ∈ Qℓ
Corollary 7.2. Suppose L ∈ SRd×d[y] is a monic linear pencil and DL(1) is bounded. Let
g1, . . . , gs ∈ R[y] and
{L}, as desired.
DL(g1, . . . , gs) := {Y ∈ Rg L(Y ) (cid:23) 0, g1(Y ) ≥ 0, . . . , gs(Y ) ≥ 0}.
If f ∈ R[y] satisfies fDL(g1,...,gs) > 0, then there are hij ∈ R[y], and Bk ∈ Rd×1[y] satisfying
(7.3)
where g0 := 1.
f =
sXj=0
gjXi
h2
ij +Xk
B∗
kLBk,
Corollary 7.3. Suppose L ∈ SRd×d[y] is a monic linear pencil and DL(1) is bounded. Then for
every polynomial f ∈ R[y] with fDL(1) > 0, there are hj ∈ R[y], and Bk ∈ Rd×1[y] satisfying
(7.4)
B∗
h2
kLBk.
f =Xj
j +Xk
It is clear that a Nichtnegativstellensatz along the lines of Corollary 6.13 holds in this
setting. We leave the details to the reader.
1, . . . , x∗
g), i.e., hx, x∗i consists of words in the 2g noncommuting letters x1, . . . , xg, x∗
7.2. Free (nonsymmetric) variables. In this section we explain how our theory adapts to
the free ∗-algebra. Let hx, x∗i be the monoid freely generated by x = (x1, . . . , xg) and x∗ =
1, . . . , x∗
(x∗
(including the empty word ∅ which plays the role of the identity 1). Let Chx, x∗i denote
the C-algebra freely generated by x, x∗, i.e., the elements of Chx, x∗i are polynomials in the
noncommuting variables x, x∗ with coefficients in C. As before, we introduce matrix poly-
nomials Cd′×dhx, x∗i. If p ∈ Cd′×dhx, x∗i is a polynomial and X ∈ (Cn×n)g, the evaluation
p(X, X ∗) ∈ Cd′×d is defined by simply replacing xi by Xi and x∗
i by X ∗
i .
g
For A1, . . . , Ag ∈ Cd×d, a linear matrix polynomial
(7.5)
L(x) =
gXj=1
Ajxj ∈ Cd×dhxi,
THE MATRICIAL RELAXATION OF AN LMI
39
is a truly linear pencil. (Note: none of the variables x∗ appears in such an L.) Its monic
symmetric pencil is
L(x, x∗) = I + L(x) + L(x)∗ = I +
Ajxj +
gXj=1
gXj=1
A∗
j x∗
j ∈ Sym Cd×dhx, x∗i.
The associated matricial LMI set is
DL = [n∈N
its operator-theoretic counterpart is
D∞
L = {X ∈ B(H)g L(X) (cid:23) 0},
{X ∈ (Cn×n)g L(X) (cid:23) 0},
and for every ℓ ∈ N, L induces a quadratic module M ℓ
{L} in Cℓ×ℓhx, x∗i:
M ℓ
{L} =n NXi=1
a∗
i ai +
NXj=1
b∗
jLbj N ∈ N, ai ∈ Cℓ×ℓhx, x∗i, bj ∈ Cd×ℓhx, x∗io.
Like in the previous subsection, all our main results from Section 6 carry over to this free
setting. As a sample, we give a Positivstellensatz:
Theorem 7.4. Suppose L ∈ Sym Cd×dhx, x∗i is a monic symmetric linear pencil and DL is
bounded. Let gj ∈ Sym Cdj ×djhx, x∗i (j ∈ N) be symmetric matrix polynomials. Then for every
f ∈ Sym Cℓ×ℓhx, x∗i with fD∞
{L, gjj∈N} ≻ 0, we have f ∈ M ℓ
{L, gjj∈N}.
As a special case we obtain a Positivstellensatz describing polynomials positive definite
on commuting tuples X in a matricial LMI set. (Note: we are not assuming the entries Xi
commute with the adjoints X ∗
j .)
Corollary 7.5. Suppose L ∈ Sym Cd×dhx, x∗i is a monic symmetric linear pencil and DL is
bounded. Suppose f ∈ Sym Cℓ×ℓhx, x∗i satisfies f (X, X ∗) ≻ 0 for all X ∈ D∞
L with XiXj =
XjXi for all i, j.
(1) Let cjk = xjxk − xkxj. Then
(2) Let djk = −c∗
jkcjk. Then
f ∈ M ℓ
{L, cjk+c∗jk, i(cjk−ckj)j,k=1,...,g}.
f ∈ M ℓ
{L, djkj,k=1,...,g}.
References
[Arv69]
[Arv08]
[Arv10]
[Arv+]
W.B. Arveson: Subalgebras of C∗-algebras, Acta Math. 123 (1969) 141 -- 224. 15, 18
W.B. Arveson: The noncommutative Choquet boundary, J. Amer. Math. Soc. 21 (2008) 1065 -- 1084.
15
W.B. Arveson: The noncommutative Choquet boundary III, Math. Scand. 106 (2010) 196 -- 210. 15
W.B. Arveson: Notes on the Unique Extension Property, snippet,
http://math.berkeley.edu/~arveson/texfiles.html 17
40
HELTON, KLEP, AND MCCULLOUGH
[B-TN02] A. Ben-Tal, A. Nemirovski: On tractable approximations of uncertain linear matrix inequalities
[DM05]
[EG96]
[FR85]
[HM04]
[KS07]
[KS10]
affected by interval uncertainty, SIAM J. Optim. 12 (2002) 811 -- 833. 1, 3, 4, 19, 24, 27, 41
M.A. Dritschel, S. McCullough: Boundary representations for families of representations of operator
algebras and spaces, J. Operator Theory 53 (2005) 159 -- 167. 15
W. Eberly, M. Giesbrecht: Efficient decomposition of associative algebras. In: Y.N. Lakshman,
Proceedings of the 1996 international symposium on symbolic and algebraic computation, ISSAC
'96, Zurich, Switzerland, ACM Press. (1996) 170 -- 178. 26
K. Friedl, L. Ro´nyal: Polynomial time solutions of some problems in computational algebra. In:
ACM Symposium on Theory of Computing 17 (1985) 153 -- 162. 26
J.W. Helton, S. McCullough: A Positivstellensatz for non-commutative polynomials, Trans. Amer.
Math. Soc. 356 (2004) 3721 -- 3737. 30, 32, 34, 35
I. Klep, M. Schweighofer: A Nichtnegativstellensatz for polynomials in noncommuting variables,
Israel J. Math. 161 (2007) 17 -- 27. 36
I. Klep, M. Schweighofer: Pure states, positive matrix polynomials and sums of hermitian squares,
Indiana Univ. Math. J. 59 (2010) 857 -- 874. 30
[KMRT98] M.-A. Knus, A.S. Merkurjev, M. Rost, J.-P. Tignol: The Book of Involutions, Amer. Math. Soc.,
[Las09]
1998. 16
J.B. Lasserre: Moments, Positive Polynomials and Their Applications, Imperial College Press, 2009.
4
[Par03]
[MKKK10] K. Murota, Y. Kanno, M. Kojima, S. Kojima: A numerical algorithm for block-diagonal decompo-
sition of matrix ∗-algebras with application to semidefinite programming, Japan J. Indust. Appl.
Math. 27 (2010) 125 -- 160. 26
P.A. Parrilo: Semidefinite programming relaxations for semialgebraic problems. Algebraic and geo-
metric methods in discrete optimization, Math. Program. 96B (2003) 293 -- 320. 4
V. Paulsen: Completely bounded maps and operator algebras, Cambridge University Press, 2002. 9,
12, 20, 28
M. Putinar: Positive polynomials on compact semi-algebraic sets, Indiana Univ. Math. J. 42 (1993)
969 -- 984. 4, 30, 37
I. Vidav: On some ∗-regular rings, Acad. Serbe Sci. Publ. Inst. Math. 13 (1959) 73 -- 80. 33
[Vid59]
[WSV00] H. Wolkowicz, R. Saigal, L. Vandenberghe (editors): Handbook of semidefinite programming. Theory,
[Pau02]
[Put93]
algorithms, and applications, Kluwer Academic Publishers, 2000. 19
J. William Helton, Department of Mathematics, University of California, San Diego
E-mail address: [email protected]
Igor Klep, Univerza v Ljubljani, Fakulteta za matematiko in fiziko, and Univerza v Mari-
boru, Fakulteta za naravoslovje in matematiko
E-mail address: [email protected]
Scott McCullough, Department of Mathematics, University of Florida
E-mail address: [email protected]
THE MATRICIAL RELAXATION OF AN LMI
NOT FOR PUBLICATION
Contents
Introduction and the statement of the main results
1.
1.1. Linear pencils and LMI sets
1.2. Main results on LMIs
1.3. Algorithms for LMIs
1.4. Positivstellensatz
1.5. Outline
2. Preliminaries on LMIs
3. Matricial LMI sets: Inclusion and Equality
3.1. The map τ is completely positive: Linear Positivstellensatz
3.2. Equal matricial LMI sets
3.3. Minimal L representing DL are unique: Linear Gleichstellensatz
4. Computational algorithms
4.1. Checking inclusion of matricial LMI sets
4.2. LMIs which are direct sums of LMIs
4.3. Tightening the relaxation
4.4. Computing the radius of matricial LMI sets
4.5. The matricial matrix cube problem
4.6. Minimal pencils and the Silov ideal
4.6.1. Minimal pencils
4.6.2.
5. More on the matrix cube problem
5.1. Comparison with the algorithm in [B-TN02]
5.2. The lattice of inclusion algorithm relaxations for the matrix cube
5.2.1. General theory
5.2.2. Lmin and Lmax for the matrix cube
6. Positivstellensatze on a matricial LMI set
6.1. Words and NC polynomials
6.1.1. NC matrix polynomials
6.1.2. Polynomial evaluations
6.2. Archimedean quadratic modules and a Positivstellensatz
6.3. More constraints
7. More general Positivstellensatze
7.1. Positivstellensatze on an LMI set in Rg
7.2. Free (nonsymmetric) variables
References
Silov ideal
41
2
2
2
3
4
4
5
7
10
14
15
19
19
20
22
22
24
26
26
26
27
27
28
28
29
30
31
31
31
31
35
36
37
38
39
|
1902.02379 | 2 | 1902 | 2019-08-29T18:25:06 | Free Stein Irregularity and Dimension | [
"math.OA"
] | We introduce a free probabilistic quantity called free Stein irregularity, which is defined in terms of free Stein discrepancies. It turns out that this quantity is related via a simple formula to the Murray--von Neumann dimension of the closure of the domain of the adjoint of the non-commutative Jacobian associated to Voiculescu's free difference quotients. We call this dimension the free Stein dimension, and show that it is a $*$-algebra invariant. We relate these quantities to the free Fisher information, the non-microstates free entropy, and the non-microstates free entropy dimension. In the one-variable case, we show that the free Stein dimension agrees with the free entropy dimension, and in the multivariable case compute it in a number of examples. | math.OA | math |
FREE STEIN IRREGULARITY AND DIMENSION
IAN CHARLESWORTH◦ AND BRENT NELSON•
Abstract. We introduce a free probabilistic quantity called free Stein irregularity, which is defined in terms
of free Stein discrepancies. It turns out that this quantity is related via a simple formula to the Murray -- von
Neumann dimension of the closure of the domain of the adjoint of the non-commutative Jacobian associated
to Voiculescu's free difference quotients. We call this dimension the free Stein dimension, and show that
it is a ∗-algebra invariant. We relate these quantities to the free Fisher information, the non-microstates
free entropy, and the non-microstates free entropy dimension. In the one-variable case, we show that the
free Stein dimension agrees with the free entropy dimension, and in the multivariable case compute it in a
number of examples.
Introduction.
In free probability, given an n-tuple of self-adjoint operators X := (x1, . . . , xn) in a tracial von Neumann
algebra (M, τ ), a regularity condition is some quantitative behavior of the joint distribution of X that implies
some qualitative behavior of the individual operators x1, . . . , xn or the algebras (von Neumann or otherwise)
that they generate. All of the well-studied regularity conditions fall broadly into two categories: microstates
and non-microstates. Examples of the former include Voiculescu's microstates free entropy χ(X), microstates
free entropy dimension δ(X) [Voi94], modified microstates free entropy dimension δ0(X) [Voi96], upper free
orbit dimension K2(X) [HS07], and 1-bounded entropy h(W ∗(X)) [Hay18]. Examples of the latter include
non-microstates free entropy χ∗(X), free Fisher information Φ∗(X) [Voi98], non-microstates free entropy
dimensions δ∗(X) and δ∗(X), and ∆(X) [CS05].
Roughly speaking, microstates quantities examine the joint distribution of X in terms of how well it
is approximated by finite dimensional matrix algebras, whereas non-microstates quantities consider the
behavior of certain derivations on the polynomial algebra generated by x1, . . . , xn. We recall a few of the
regularity conditions corresponding to the aforementioned free probabilistic quantities:
L3(R, m) [Voi93].
• If Φ∗(x) < ∞, then the spectral measure of x is Lebesgue absolutely continuous with density in
• If δ(x) = 1, then x is diffuse (i.e. its spectral measures has no atoms) [Voi94].
• If δ0(X) > 1, then W ∗(X) has no Cartan subalgebras and does not have property Γ [Voi96].
• If δ0(X) > 1, then W ∗(X) is prime [Ge98].
• If Φ∗(X) < ∞, then W ∗(X) does not have property Γ [Dab10].
• If δ∗(X) = n > 1, then W ∗(X) is a factor [Dab10].
• If δ∗(X) = n, then every non-constant, self-adjoint p ∈ Chx1, . . . , xni is diffuse [CS16, MSW17].
• If Φ∗(X) < ∞, then χ∗(p) > −∞ for every non-constant, self-adjoint p ∈ Chx1, . . . , xni [BM18].
In the present paper, we propose new quantities that fall into the non-microstates category: free Stein
irregularity and free Stein dimension (see Definitions 2.1 and 2.11). Motivated by work of the second
author in [FN17], these quantities are defined via the free analogues of Stein kernels and Stein discrepancy
(see [LNP15] and its references). Given an n-tuple (ξ1, . . . , ξn) ∈ L2(M )n, the free Stein discrepancy of X
relative to this n-tuple (see Subsection 1.2) is a non-negative quantity that measures how close ξ1, . . . , ξn
are to being the conjugate variables to x1, . . . , xn. In particular, the free Stein discrepancy is zero if and
only if ξ1, . . . , ξn are the conjugate variables, in which case Φ∗(X) < ∞ and so the above results tell us that
W ∗(X) does not have property Γ and χ∗(p) > −∞ for every non-constant, self-adjoint p ∈ Chx1, . . . , xni. Of
◦Department of Mathematics, University of California, Berkeley
•Department of Mathematics, Michigan State University
[email protected]
[email protected]
1
Charlesworth and Nelson
Free Stein Irregularity and Dimension
course, determining that the free Stein discrepancy was zero required preexisting knowledge of the n-tuple
(ξ1, . . . , ξn) -- or a very lucky guess.
In this paper, we explore what can be said if instead one merely supposes that the free Stein discrepancy
can be made arbitrarily small by varying the n-tuple (ξ1, . . . , ξn) ∈ L2(M )n. We are therefore naturally
driven to consider the infimum of free Stein discrepancies, which we define as the free Stein irregularity,
and the situation of interest is simply the regularity condition of having zero free Stein irregularity. One
immediately has that this is a weaker regularity condition than Φ∗(X) < ∞, but it turns out to be a stronger
condition than δ∗(X) = n (see Corollary 4.4). Interestingly, in the one variable case δ∗(X) = 1 is equivalent
to having zero free Stein irregularity. This is because for X = (x1), the square of the free Stein irregularity
can be computed explicitly and is given by the sum of the squares of masses of any atoms in the spectral
measure of x1 (see Theorem 4.5). In the general case, the free Stein irregularity is (somewhat surprisingly)
given by a formula involving the Murray -- von Neumann dimension of the domain of an unbounded operator
(see Theorem 2.10): namely, the adjoint of the non-commutative Jacobian associated to Voiculescu's free
difference quotients (see Subsection 1.1). We call this dimension the free Stein dimension of X, and are
able to further relate it to a module of closable derivations on the Chx1, . . . , xni. From this characterization
it follows that the free Stein dimension is a ∗-algebra invariant (see Theorem 3.2). Furthermore, we also
consider the above quantities when x1, . . . , xn are considered as variables over a unital ∗-subalgebra B ⊂ M .
The structure of the paper is as follows. In Section 1 we establish some notation and recall the definitions
of free Stein kernels and free Stein discrepancy. In Section 2, we define free Stein irregularity, derive some
elementary properties, and define free Stein dimension. In Section 3, we characterize free Stein dimension
through modules of closable derivations and use this to show algebraic invariance.
In Section 4, we re-
late the free Stein irregularity and dimension to free Fisher information and non-microstates free entropy
dimension(s), and compute the both explicitly in the one-variable case. In Section 5, we compute the (multi-
variable) free Stein irregularity and dimension for a tuple of generating a group algebra or finite-dimensional
algebra. We conclude the paper with a few appendices detailing interesting examples and computations.
Acknowledgments. The authors would like to thank Dimitri Shlyakhtenko for his useful comments and
suggestions; in particular, for suggesting a cleaner approach to the results in Section 3. They would also
like to thank Michael Hartglass, Benjamin Hayes, and David Jekel for helpful discussions related to this
paper. This work was initiated while the authors were attending the Park City Mathematics Institute
(PCMI) Summer Session on Random Matrices. Part of this research was performed while the authors were
visiting the Institute for Pure and Applied Mathematics (IPAM), which is supported by the National Science
Foundation. The first and second authors were supported by NSF grants DMS-1803557 and DMS-1502822,
respectively.
1. Preliminaries.
1.1 Notation. Throughout (M, τ ) denotes a tracial W ∗-probability space. We denote by L2(M ) the
GNS Hilbert space corresponding to τ and identify M with its representation on this space. We let M◦ =
{x◦ : x ∈ M} denote the opposite von Neumann algebra, represented on L2(M◦) which can be identified with
the dual Hilbert space to L2(M ). We let M ¯⊗M◦ denote the von Neumann algebra tensor product, which
is equipped with the tensor product trace τ ⊗ τ◦. We will typically repress the '◦' notation on elements of
M ¯⊗M◦.
Throughout, X will denote a tuple (x1, . . . , xn) ∈ M n (not necessarily self-adjoint), and B ⊂ M will be a
unital subalgebra. We will always assume that for each i = 1, . . . , n, x∗i = xj for some j (possibly j = i if xi
is actually self-adjoint). T will denote a family (t1, . . . , tn) of indeterminates of the same length as X, and
B hTi will be the algebra generated by t1, . . . , tn and B. Note that there is a unique unital homomorphism
B hTi → B hXi which sends each ti to xi, which we will denote evX ; evX is always surjective but may fail
to be injective. We will also use evX to denote the corresponding maps on B hTin, B hTi ⊗ B hTi◦, and
Mn(B hTi ⊗ B hTi◦).
2
Charlesworth and Nelson
Free Stein Irregularity and Dimension
For each i, the free difference quotient ∂i : B hTi → B hTi ⊗ B hTi◦ is defined to be the (unique) linear
map with ∂i(tj ) = δi=j and B ⊂ ker ∂i, satisfying the Leibniz rule; more precisely,
d
∂i(b0ti1 b1ti2 ··· bd−1tdbd) =
δik=ib0ti1 b1 ··· tk−1bk−1 ⊗ bktk+1 ··· tnbn.
Xk=1
We similarly define ∂ : B hTi → (B hTi⊗B hTi◦)n by ∂p = (∂1p, . . . , ∂np), and let J : B hTin → Mn(B hTi⊗
B hTi◦) be the non-commutative Jacobian:
J (p1, . . . , pn) =
=
∂np1
∂np2
...
∂npn
∂1p1
∂1p2
...
∂1pn
For Ξ = (ξ1, . . . , ξn), H = (η1, . . . , ηn) in either L2(M )n or L2(M ¯⊗M◦)n denote
∂p1
∂p2
...
∂pn
.
···
···
. . .
···
n
For A, B ∈ Mn(L2(M ¯⊗M◦)) denote
hΞ, Hi2 :=
hξj , ηji2 .
Xj=1
n
hA, BiHS :=
h[A]jk, [B]jki2 .
Xj,k=1
We denote by # the usual product in M ¯⊗M◦ ((a ⊗ b)#(c ⊗ d) = (ac) ⊗ (db)), the usual product in
Mn(M ¯⊗M◦), the action of M ¯⊗M◦ on L2(M ) ((a⊗ b)#c = acb), the diagonal action of M ¯⊗M◦ on L2(M )n,
and the action of Mn(M ¯⊗M◦) on L2(M )n.
In the case that X satisfy no B-algebraic relations, we can view ∂i, ∂, and J defined on polynomials in the
variables X rather than the indeterminates T , and so they become densely-defined operators on L2(B hXi)
or L2(B hXi)n with codomains L2(M ¯⊗M◦), L2(M ¯⊗M◦)n, or Mn(L2(M ¯⊗M◦)), respectively.
We denote by ∂∗i:B, ∂∗X:B, and J ∗X:B the adjoints of the implied relations on B hXi; for example, we define
∂∗i : B to be the map with domain consisting of those a ∈ L2(M ¯⊗M◦) for which there is some η ∈ L2(B hXi)
such that for all p ∈ B hTi we have
hη, evX (p)i = ha, evX ◦∂i(p)i ;
we then set ∂∗i (a) = η. Thus ∂∗i:B : L2(B hXi ⊗ B hXi◦) → L2(B hXi), ∂∗X:B : L2(B hXi ⊗ B hXi◦)n →
L2(B hXi), and J ∗X:B : Mn(L2(B hXi ⊗ B hXi◦)) → L2(B hXi)n are unbounded operators, although their
domains may fail to be dense. When X or B are clear from context, we may suppress the relevant subscript.
Lastly, let us denote
so that J (X) = 1.
1 ⊗ 1
0
1 :=
. . .
0
1 ⊗ 1
,
1.2 Free Stein kernels and free Stein discrepancy. We recall some definitions below from [FN17].
These have been modified slightly to accommodate our consideration of non-algebraically free operators X
over a unital subalgebra B, but when X is algebraically free and B = C, we recover the original definition.
By working in this broader generality, we reap a number of benefits: we are able to consider freeness with
amalgamation; we are able to compute free Stein dimensions in finite-dimensional algebras; and we are able
derive some interesting statements about the free Stein dimension of certain generators of interpolated free
group factors (see Appendix B). The reader may find it useful to gain intuition by considering (as the authors
have) the simpler case outlined in Remark 2.3, where B = C, X is algebraically free and self-adjoint, and
the free difference quotients are densely defined operators.
Given Ξ ∈ L2(M )n, we say that
A ∈ Mn(L2(B hXi ⊗ B hXi◦))
3
Charlesworth and Nelson
Free Stein Irregularity and Dimension
is a free Stein kernel of X relative to Ξ over B if A ∈ dom(J ∗X:B) and J ∗X:B(A) = Ξ: to wit, if
hΞ, evX Pi2 = hA, evX ◦J (P )iHS
∀P ∈ B hTin .
In this case we say (after [Shl04]) that Ξ is a partial conjugate variable to X corresponding to A.
The free Stein discrepancy of X relative to Ξ over B is the quantity
A kA − 1kHS,
Σ∗(X Ξ : B) := inf
(1)
where as before the infimum is over all free Stein kernels of X relative to Ξ over B. Equivalently, Σ∗(X Ξ :
B) = kΠ(A) − 1kHS where A is any free Stein kernel of X relative to Ξ and Π is the orthogonal projection
onto the closure of the range of evX ◦J .
A priori the free Stein discrepancy could be infinite, since a free Stein kernel for X need not exist. Indeed,
if Ξ is not orthogonal to Bn ⊂ L2(M )n then for some Z ∈ Bn we have
hΞ, evX Zi2 6= 0 = hA, evX ◦J (Z)iHS
∀A ∈ Mn(L2(M ¯⊗M◦)).
For general unital subalgebras B, it is not clear if the condition Ξ ⊥ Bn is sufficient to guarantee the existence
of free Stein kernels. However, in the case B = C it suffices by [CFM18, Theorem 2.1], which we state below.
Proposition 1.1 ([CFM18]). For Ξ = (ξ1, . . . , ξn) ∈ L2(M )n ⊖ Cn,
(ξi ⊗ 1 − 1 ⊗ ξi)#(xj ⊗ 1 − 1 ⊗ xj)(cid:21)n
i,j=1 ∈ Mn(L2(M ¯⊗M◦))
AΞ :=(cid:20) 1
2
is a free Stein kernel for X relative to Ξ. Consequently, Σ∗(X Ξ) < ∞ always.
Remark 1.2. For larger unital subalgebras B, AΞ given in Proposition 1.1 may fail to be a free Stein kernel.
Indeed, if x, s are freely independent semicircular variables, B = C[s], and ξ = sxs, one can compute that
hξ, sxsi = 1 while
(cid:28) 1
2
(ξ ⊗ 1 − 1 ⊗ ξ)#(x ⊗ 1 − 1 ⊗ x), s ⊗ s(cid:29) = 0.
One might hope that in nice cases AΞ is the free Stein kernel which attains the free Stein discrepancy of
X, but unfortunately this holds if and only if Ξ = 0 (see Appendix A). However, we do obtain the following
corollary:
Corollary 1.3. The map
L2(M )n ⊖ Cn ∋ Ξ 7→ Σ∗(X Ξ)
is continuous.
Proof. For Ξ, Ξ′ ∈ L2(M )n ⊖ Cn let AΞ and AΞ′ be as in Proposition 1.1. Then
Σ∗(X Ξ) − Σ∗(X Ξ′) = kΠ(AΞ) − 1kHS − kΠ(AΞ′ ) − 1kHS
≤ kΠ(AΞ) − Π(AΞ′ )kHS ≤ kAΞ − AΞ′kHS ≤ CkΞ − Ξ′k2,
where C > 0 is a constant depending only on n and X.
(cid:3)
Remark 1.4. If Σ∗(X Ξ) = 0, then 1 is a free Stein kernel for X and hence
∀P ∈ ChXin .
hΞ, evX Pi2 = h1, evX ◦J (P )iHS
That is, Ξ is the usual conjugate variable to X. In fact, this is precisely why the free Stein discrepancy is
defined to measure the distance between a free Stein kernel A and 1. We remind the reader that the free
Fisher information of X is defined as the quantity
Φ∗(X) := kΞk2
2
if Ξ is the conjugate variable to X, whereas it is defined to be +∞ if no conjugate variable exists (cf. [Voi98,
Definition 6.1]).
Furthermore, Σ∗(X X) = 0 if and only if X is the conjugate variable to X if and only if X is a free
semicircular family.
4
Charlesworth and Nelson
Free Stein Irregularity and Dimension
2. Free Stein Irregularity.
We begin with the definition of free Stein irregularity. In order to better motivate and clarify the definition,
it is followed by an examination of a special case.
Definition 2.1. Let X = (x1, . . . , xn) ∈ M n be a tuple of operators such that for each i = 1, . . . , n, x∗i = xj
for some j. Let B be a unital ∗-subalgebra of M . The free Stein irregularity of X over B is the quantity
For R > 0, the R-bounded free Stein irregularity of X over B is the quantity
Σ∗(X : B) := inf(cid:8)Σ∗(X Ξ : B) : Ξ ∈ L2(M )n(cid:9) .
Σ∗R(X : B) := inf(cid:8)Σ∗(X Ξ : B) : Ξ ∈ L2(M )n with kΞk2 ≤ R(cid:9) .
Σ∗R(X : B). In the particular case B = C, we will use the
Note that Σ∗(X : B) = inf
R>0
shorthand Σ∗(X) := Σ∗(X : C).
Σ∗R(X : B) = lim
R→∞
Remark 2.2. Notice that if B ⊆ C ⊆ M , there are fewer free Stein kernels of X over B than over C (as
there are more polynomials and so more relations must be satisfied); it follows that Σ∗(X : B) ≤ Σ∗(X : C).
More formally, if E : M ¯⊗M◦ → W ∗(B hXi⊗ B hXi◦) is the trace-preserving conditional expectation onto the
von Neumann algebra generated by B hXi ⊗ B hXi◦, then the claimed inequality follows from the inclusion
(E ⊗ In) (dom(J ∗X : C)) ⊂ dom(J ∗X : B).
Remark 2.3. Consider the following special case: let X = (x1, . . . , xn) be an n-tuple of self-adjoint operators
generating M . Assume that x1, . . . , xn are algebraically free so that evX : ChTi → ChXi is a ∗-algebra
isomorphism. This allows us to view the free difference quotients ∂j, j = 1, . . . , n, as defined directly on
ChXi, and -- moreover -- as densely defined (unbounded) operators of the form
∂j : L2(M ) → L2(M ¯⊗M◦).
Similarly, ∂ and J may be regarded as maps densely defined on the appropriate Hilbert spaces.
In this context, a free Stein kernel A of X relative to some Ξ is simply an element of dom(J ∗) with
J ∗(A) = Ξ. Consequently, the free Stein irregularity, which is given by the formula
Σ∗(X) := inf{Σ∗(X Ξ) : Ξ ∈ L2(M )n},
(see Definition 2.1), is equivalently the distance between 1 and (the closure of) dom(J ∗) in Mn(L2(M ¯⊗M◦)).
The free Stein irregularity can be thought of as quantitative measurement of how close the n-tuple X is to
having conjugate variables. Indeed, capturing such a defect was the original motivation for defining this
quantity and if we consider the following technical modification
Σ∗R(X) := inf{Σ∗(X Ξ) : Ξ ∈ L2(M )n with kΞk2 ≤ R},
R > 0
then Σ∗R(X) = 0 if and only if an n-tuple of conjugate variables to X exists and is bounded by R (see
Theorem 4.1).
It turns out that the Hilbert subspace dom(J ∗) ⊂ Mn(L2(M ¯⊗M◦)) is a left M ¯⊗M◦-module (see
Lemma 2.9) and that its Murray -- von Neumann dimension is related to the free Stein irregularity by the
following formula:
n − Σ∗(X)2 = dimM ¯⊗M ◦(cid:16)dom(J ∗)(cid:17)
(see Theorem 2.10). We are thus compelled to study the quantity on the left-hand side, which we denote
by σ(X) and call the free Stein dimension of X. Analogously to free entropy dimension, it satisfies the
inequality
where Y is another tuple of self-adjoint operators generating some (potentially larger) von Neumann algebra
along with X; equality holds if X and Y are freely independent (see Corollary 2.7). It is also a ∗-algebra
invariant (see Theorem 3.2) and compares to the non-microstates free entropy dimensions:
σ(X, Y ) ≤ σ(X) + σ(Y ),
σ(X) ≤ δ∗(X) ≤ δ∗(X)
5
Charlesworth and Nelson
Free Stein Irregularity and Dimension
(see Corollary 4.4). Moreover, it is known to agree with these other dimensions in a number of cases (see
Theorem 4.5, Proposition 5.1, and Corollary 5.2). In particular, when n = 1 and x is a self-adjoint operator
with spectral measure µx we have
It is thus natural to wonder whether these dimensions always agree. However, some basic relations still elude
us. For example, when χ∗(X) > ∞ it is known that δ∗(X) = δ∗(X) = n, but it remains open whether or
not this implies σ(X) = n as well.
σ(x) = 1 −Xt∈R
µx({t})2.
2.1 Elementary Properties. We derive some useful properties of free Stein irregularity.
Proposition 2.4. Σ∗(X : B) = Σ∗(X : W ∗(B)).
Proof. Denote N = W ∗(B). We have Σ∗(X : B) ≤ Σ∗(X : N ) by Remark 2.2, so we need only establish
the other inequality. Now, let us suppose that A is a Stein kernel for X relative to Ξ over B. Fix p =
b0ti1 b1 ··· tid bd ∈ N hTi with b0, . . . , bd ∈ N and for j = 0, . . . , d take a sequence (bj(k))k∈N ⊂ B converging
strongly to bj with norms uniformly bounded by kbjk (such exists by Kaplansky's density theorem). Then
if we let
p(k) := b0(k)ti1 b1(k)··· tid bd(k) ∈ B hTi ,
we find evX (p(k)) converges to evX (p) in L2(M ), while evX ◦∂(p(k)) converges to evX ◦∂(p) in L2(M ¯⊗M◦)n.
If P ∈ N hTin and P (k) are chosen in a similar way, it follows that
hΞ, evX Pi = lim
k→∞ hΞ, evX (P (k))i = lim
k→∞ hA, evX ◦J (P (k))i = hA, evX ◦J (P )i ;
that is, A is also a Stein kernel for X relative to Ξ over N . Hence Σ∗(X : N ) ≤ Σ∗(X : B).
Lemma 2.5. Let B, C ⊆ M be unital ∗-subalgebras, and let D ⊆ B ∩ C be a common unital ∗-subalgebra
with conditional expectation E : (B ∨ C)hXi → D, where B ∨ C is the ∗-algebra generated by B and C. If
C is free from B hXi with amalgamation over D, then
(cid:3)
Σ∗(X : B ∨ C) = Σ∗(X : B).
In particular, if X is free from C, then Σ∗(X : C) = Σ∗(X).
Proof. By Remark 2.2, it suffices to prove Σ∗(X : B ∨ C) ≤ Σ∗(X : B). We will prove this by showing
dom(∂∗X:B) ⊂ dom(∂∗X:B∨C ). Let η ∈ dom(∂∗X:B), with ξ := ∂∗X:B(η). Take c0, . . . , cd ∈ C with E(ci) = 0 for
i = 1, . . . , d − 1, and P1, . . . , Pd ∈ B hTi with E(evX Pi) = 0 for i = 1, . . . , d; set P = c0P1c1 ··· Pdcd.
We claim
hξ, evX Pi2 = hη, evX ◦∂X:B∨CPi2 .
If d = 0, the left-hand side is hξ,E(c0)i = 0 since ξ is orthogonal to D ⊆ B and free from C with amalga-
mation. The right-hand side is zero by the definition of ∂X:B∨C . If d ≥ 2, it is not hard to check that both
sides are zero due to freeness with amalgamation over D. Thus it remains to establish the claim when d = 1.
In this case, invoking freeness with amalgamation, we have:
hξ, evX (c0P1c1)i2 = hξ, c0(evX P1)c1i2
= hξ,E(c0)(evX P1)E(c1)i2
= hξ, evX (E(c0)P1E(c1))i2
= hη, evX ◦∂X : B(E(c0)P1E(c1))i2
= hη,E(c0) · evX ◦∂X : B(P1) · E(c1)i2
= hη,E(c0) · evX ◦∂X : B∨C(P1) · E(c1)i2
= hη, c0 · evX ◦∂X : B∨C (P1) · c1i2
= hη, evX ◦∂X : B∨C(c0P1c1)i2 .
This completes the proof of the claim.
Finally, since such elements P span (B ∨ C)hTi, this shows that η ∈ dom(∂∗X:B∨C), completing the
(cid:3)
proof.
6
Charlesworth and Nelson
Free Stein Irregularity and Dimension
Theorem 2.6. Let B, C ⊂ M be unital ∗-subalgebras, and let X ∈ M n, Y ∈ M m be tuples. Then
Σ∗(X : B)2 + Σ∗(Y : C)2 ≤ Σ∗(X, Y : B ∨ C)2 ≤ Σ∗(X : (B ∨ C)hY i)2 + Σ∗(Y : (B ∨ C)hXi)2.
Moreover, suppose D ⊂ B ∩ C is a common unital ∗-subalgebra with conditional expectation E : (B ∨ C)[X ∪
Y ] → D. If B hXi and C hY i are free with amalgamation over D, then the above inequalities are equalities.
Proof. Let A ∈ dom(J ∗(X,Y ) : B∨C ). Let A1 be the entry-wise projection of the top-left n × n sub-matrix of
A onto L2(B hXi ⊗ B hXi◦), and let A2 be the entry-wise projection of the bottom-right m × m sub-matrix
of A onto L2(C hY i ⊗ C hY i◦). One easily checks that A1 ∈ dom(J ∗X : B) and A2 ∈ dom(J ∗Y : C ). Hence
Σ∗(X : B)2 + Σ∗(Y : C)2 ≤ kA1 − 1k2
HS + kA2 − 1k2
HS ≤ kA − 1k2
HS.
Since A ∈ dom(J ∗(X,Y ) : B∨C ) was arbitrary, this yields the first inequality.
Next, let A1 ∈ dom(J ∗X : (B∨C)hY i
). It is easily checked that
) and A2 ∈ dom(J ∗Y : (B∨C)hXi
A =(cid:20) A1
0 A2 (cid:21) ∈ dom(J ∗(X,Y ) : B∨C ).
0
Thus
and so the second inequality follows.
Σ∗(X, Y : B ∨ C)2 ≤ kA − 1k2
HS = kA1 − 1k2
HS + kA2 − 1k2
HS,
Finally, if B hXi and C hY i are free with amalgamation over D, then by Lemma 2.5 we have
Σ∗(X : (B ∨ C)hY i) = Σ∗(X : B ∨ (C hY i)) = Σ∗(X : B).
Similarly, Σ∗(Y : (B ∨ C)hXi) = Σ∗(Y : C). This forces the claimed equality.
(cid:3)
Applying the previous theorem to the special case B = C = D(= C), yields the following corollary.
Corollary 2.7.
(1) If B hXi and B hY i are free with amalgamation over B, then
Σ∗(X, Y : B)2 = Σ∗(X : B)2 + Σ∗(Y : B)2.
(2) If X and Y are free, then
Σ∗(X, Y )2 = Σ∗(X)2 + Σ∗(Y )2.
Proposition 2.8. The function R 7→ Σ∗R(X : B) is convex.
Proof. Let 0 < R1 < R2. Let A1, A2 ∈ dom(J ∗X : B) with kJ ∗X : B(Ai)k2 ≤ Ri, i = 1, 2,. Then for t ∈ [0, 1],
(1 − t)A1 + tA2 ∈ dom(J ∗X : B) with
kJ ∗X : B((1 − t)A1 + tA2)k2 ≤ (1 − t)kJ ∗X : B(A1)k2 + tkJ ∗X : B(A2)k2 ≤ (1 − t)R1 + tR2.
Hence
Σ∗(1−t)R1+tR2
(X : B) ≤ k(1 − t)A1 + tA2 − 1kHS ≤ (1 − t)kA1 − 1kHS + tkA2 − 1kHS.
Taking the infimum over A1 and A2 completes the proof.
(cid:3)
2.2 Free Stein dimension. In this subsection we give a characterization of the free Stein irregularity in
terms of the Murray -- von Neumann dimension of the closure of dom(J ∗X : B) in Mn(L2(M ¯⊗M◦)), viewed as
a left M ¯⊗M◦-module. We first show, in the following lemma, that dom(∂∗X : B) admits a left B hXi⊗ B hXi◦
action; this is the multivariate analogue of [Voi98, Proposition 4.1] and follows by an identical proof.
Lemma 2.9. For η = (η1, . . . , ηn) ∈ dom(∂∗X : B) and p, q ∈ B hTi, evX (p ⊗ q)#η ∈ dom(∂∗X : B) with
∂∗X : B(evX (p ⊗ q)#η) = evX (p ⊗ q)#∂∗X : B(η)
n
−
Xj=1
(1 ⊗ τ◦) (evX p · [ηj # evX ◦∂j(q∗)∗]) − (τ ⊗ 1) ([ηj # evX ◦∂j(p∗)∗] · evX q) .
7
Charlesworth and Nelson
Free Stein Irregularity and Dimension
From this lemma we see that dom(∂∗X : B) is invariant under the left action of B hXi ⊗ B hXi◦. Con-
sequently, the Kaplansky density theorem implies that dom(∂∗X : B) is a closed, left M ¯⊗M◦-module. Ob-
serve that for A ∈ dom(J ∗X : B), if Ai = (Ai1, . . . , Ain) (i.e.
the i-th row of A) for i = 1, . . . , n, then
It then follows that dom(J ∗X : B) is also a closed, left Mn(M ¯⊗M◦)-module
A1, . . . , An ∈ dom(∂∗X : B).
satisfying dom(J ∗X : B) ∼= dom(∂∗X : B)
. This identification immediately gives the second equality in the
following theorem.
Theorem 2.10. For B ⊂ M a unital ∗-subalgebra and X = (x1, . . . , xn) ∈ M n such that M = W ∗(B hXi),
n
n − Σ∗(X : B)2 = dimM ¯⊗M ◦ dom(∂∗X : B) = dimMn(M ¯⊗M ◦) dom(J ∗X : B).
Proof. Let e ∈ Mn(L2(M ¯⊗M◦)) be the projection of 1 onto dom(J ∗X : B) so that Σ∗(X : B) = ke − 1kHS.
Hence
n − Σ∗(X : B)2 = n − kek2
HS + 2Re he, 1iHS − k1k2
HS = kek2
HS.
Now, identify M ¯⊗M◦ with its diagonal representation M ¯⊗M◦⊗In on L2(M ¯⊗M◦)n. Then N := (M ¯⊗M◦)′∩
B(L2(M ¯⊗M◦)) is identified with Mn(N ). Observe that
dom(J ∗X : B) ∼= dom(∂∗X : B)
n
∼=n(T v1, . . . , T vn) ∈ Mn(L2(M ¯⊗M◦)) : T ∈ Mn(N ), T · L2(M ¯⊗M◦)n ⊂ dom(∂∗X : B)o,
where vj ∈ L2(M ¯⊗M◦)n is the vector with 1⊗1 in the j-th entry and zeros elsewhere. In fact, (T v1, . . . , T vn)
in the last space is sent to its transpose in the first space. Let f be the projection of L2(M ¯⊗M◦)n onto
dom(∂∗X : B), so that f ∈ Mn(N ); then f T ∈ dom(J ∗X : B), and we further claim that f T = e. Indeed, for
A ∈ dom(J ∗X : B) let A1, . . . , An ∈ dom(∂∗X : B) be the rows of A as in the discussion preceding the theorem.
Hence f Ai = Ai and so
(cid:10)f T , A(cid:11)HS =(cid:10)1, f T A(cid:11)HS =
n
Xi,j=1(cid:10)1 ⊗ 1, [f T ]ji[A]ij(cid:11)HS
n
n
=
=
=
Thus f T = e and
n
Xi,j=1
Xi=1
Xi=1
n
h1 ⊗ 1, [f ]ij[A]ijiHS =
h1 ⊗ 1, (f Ai)iiHS =
h1 ⊗ 1, [f ]ij(Ai)jiHS
Xi,j=1
h1 ⊗ 1, (Ai)iiHS
n
Xi=1
h1 ⊗ 1, [A]iiiHS = h1, AiHS = he, AiHS .
So the result follows by our previous computation.
dimM ¯⊗M ◦ (dom(∂∗X : B)) = kfk2
HS = kf Tk2
HS = kek2
HS.
(cid:3)
In light of the above theorem, we make the following definition.
Definition 2.11. For an n-tuple X, the free Stein dimension of X over B is the quantity
We can rephrase Theorem 2.6 and Corollary 2.7 in terms of free Stein dimension as follows:
σ(X : B) := n − Σ∗(X : B)2.
Corollary 2.12. Let B, C ⊂ M be unital ∗-subalgebras, and let X ∈ M n, Y ∈ M m be tuples. Then
σ(X : (B ∨ C)hY i) + σ(Y : (B ∨ C)hXi) ≤ σ(X, Y : B ∨ C) ≤ σ(X : B) + σ(Y : C).
Moreover, suppose D ⊂ B ∩ C is a common unital ∗-subalgebra with conditional expectation E : (B ∨ C)[X ∪
Y ] → D. If B hXi and C hY i are free with amalgamation over D, then the above inequalities are equalities.
In particular, if B hXi and B hY i are free with amalgamation over B, then
σ(X, Y : B) = σ(X : B) + σ(Y : B).
Furthermore, if X and Y are free, then
σ(X, Y ) = σ(X) + σ(Y ).
8
Charlesworth and Nelson
Free Stein Irregularity and Dimension
3. Via Closable Derivations.
In this section we characterize σ(X : B) in terms of certain closable derivations on B hXi. This perspective
yields a number of invariance results; in particular, that σ(X : B) depends only on the algebras B and B hXi.
For an inclusion of two ∗-subalgebras B ⊂ C ⊂ M with M = W ∗(C), consider the set
Der1⊗1(B ⊂ C) := {d : C → L2(M ¯⊗M◦) d is a derivation with B ⊂ ker(d) and 1 ⊗ 1 ∈ dom(d∗)}.
This set of derivations admits a right C ⊗ C◦-action:
d · (a ⊗ b) := d( · )#(a ⊗ b).
Indeed, a ⊗ b ∈ dom(d∗) by the same proof as [Voi98, Proposition 4.1] and so
[d · (a ⊗ b)]∗(1 ⊗ 1) = d∗(a ⊗ b).
Lemma 3.1. For B ⊂ M a ∗-subalgebra and X = (x1, . . . , xn) ∈ M n, the conjugate linear map
φX : Der1⊗1(B ⊂ B hXi) → dom(∂∗X : B)
d 7→ (Jτ⊗τ ◦d(x1), . . . , Jτ⊗τ ◦d(xn))
is a bijection that maps the right B hXi ⊗ B hXi◦-action on Der1⊗1(B ⊂ B hXi) to the left regular B hXi ⊗
B hXi◦-action on L2(M ¯⊗M◦). Consequently, when M = W ∗(B hXi)
σ(X : B) = dimM ¯⊗M ◦ φX (Der1⊗1(B ⊂ B hXi)).
Proof. First notice that each element of Der1⊗1(B ⊂ B hXi) is determined by its values on X. Hence φX is
injective.
Now, given d ∈ Der1⊗1(B ⊂ B hXi), we have for any p ∈ B hTi
hd∗(1 ⊗ 1), evX pi2 = h1 ⊗ 1, d(evX p)i =*1 ⊗ 1,
n
evX ◦∂i(p)#d(xi)+2
Xi=1
Thus φX (d) ∈ dom(∂∗X : B).
Given a = (a1, . . . , an) ∈ dom(∂∗X : B), define
da : B hTi ∋ p 7→
n
Xi=1
evX ◦∂i(p)#Jτ⊗τ ◦ai.
= hφX (d), evX ◦∂(p)i2 .
Then for p, q, r ∈ B hTi one has
DevX (p ⊗ q), da(r)EHS
= hevX (p ⊗ q)#a, evX ◦∂(r)iHS = h∂∗X : B(evX (p ⊗ q)#a), evX (r)iHS ,
where the last equality uses Lemma 2.9. It follows that da = da ◦ evX for some da ∈ Der1⊗1(B ⊂ B hXi)
with d∗a(1 ⊗ 1) = ∂∗X : B(a). In particular, da(xi) = da(ti) = Jτ⊗τ ◦(ai). Thus a = φX (da) ∈ φX (Der1⊗1(B ⊂
B hXi)).
(cid:3)
3.1 Algebraic invariance. If Y ∈ B hXim satisfies B hY i = B hXi, then φY ◦ φ−1
X yields a left
B hXi ⊗ B hXi◦-module isomorphism of dom(∂∗X : B) ∼= dom(∂∗Y : B). This extends to a left M ¯⊗M◦-module
isomorphism dom(∂∗X : B) ∼= dom(∂∗Y : B). Using Theorem 2.10 we obtain the following theorem:
Theorem 3.2. If Y ∈ B hXim satisfies B hY i = B hXi, then
σ(Y : B) = σ(X : B).
Remark 3.3. It follows from Theorem 3.2 that for any Y ∈ B hXim, we have
σ(X, Y : B) = σ(X : B).
In particular, if Y ∈ Bm then σ(Y : B) = 0.
9
Charlesworth and Nelson
Free Stein Irregularity and Dimension
For every Y ∈ B hXim we have the following map:
ψX,Y : Der1⊗1(B ⊂ B hXi) → Der1⊗1(B ⊂ B hY i)
d 7→ d BhY i .
Of course, if B hY i = B hXi then this map is the identity map, but otherwise it is potentially neither injective
nor surjective. Nevertheless, one can therefore always consider the composition φY ◦ ψX,Y ◦ φ−1
X .
Proposition 3.4. Let Y ∈ B hXin with Y = evX F for some F ∈ B hTi. Then for a ∈ dom(∂∗X : B), we
have
with ∂∗Y : B(a# evY ◦J (F )∗) = ∂∗X : B(a). Moreover, φY ◦ ψX,Y ◦ φ−1
dom(∂∗Y : B), and when M = W ∗(B hXi) one has
φY ◦ ψX,Y ◦ φ−1
X (a) = a# evX ◦J (F )∗ ∈ dom(∂∗Y : B),
(2)
X extends to a map ρX,Y : dom(∂∗X : B) →
σ(X : B) ≤ σ(Y : B) + dimM ¯⊗M ◦ (ker(ρX,Y )).
Proof. Let Y = (y1, . . . , ym) and F = (f1, . . . , fm). For a = (a1, . . . , an) ∈ dom(∂∗X : B), we have that φ−1
X (a)
is given by the derivation da defined in the proof of Lemma 3.1. In particular, da(xi) = Jτ⊗τ ◦ai. It follows
that
φY ◦ ψX,Y ◦ φ−1
X (a) = (Jτ⊗τ ◦da(y1), . . . , Jτ⊗τ ◦da(ym))
evX ◦∂i1 (f1)#Jτ⊗τ ◦ai1 , . . . , Jτ⊗τ ◦
evX ◦∂im(fm)#Jτ⊗τ ◦aim!
n
Xim=1
n
Xi1=1
ai1 # evX ◦∂i1(f1)∗, . . . ,
= Jτ⊗τ ◦
= n
Xi1
= a# evX ◦J (F )∗,
aim # evX ◦∂im(fm)∗!
n
Xim
and so Equation (2) holds. As the right action of Mn(B hXi⊗B hXi◦) is bounded, we immediately obtain the
extension ρX,Y . Furthermore, ρX,Y commutes with the left action of M ¯⊗M◦ and so is a left M ¯⊗M◦-module
map when M = W ∗(B hXi). Hence the claimed inequality follows from Theorem 2.10 and the rank -- nullity
theorem.
(cid:3)
This structure in many cases puts restrictions on the sort of kernels that may be produced for a given
tuple. For example, in light of Theorem 2.6 (and in particular its proof) one may ask whether a kernel for
X may always be extended to a kernel for a larger system (X, Y ), as in many nice cases this can be done.
However, using the above proposition, Example B.5 shows that this is not always possible.
Remark 3.5. Theorem 3.2 and Proposition 3.4 can be generalized slightly by considering the following
non-commutative power series. After [CS16], for R > 0 we denote by B hTiR the completion of B hTi in the
norm
Note that this is in fact a Banach norm. We also denote
kpkR := infnXkb0k ···kbdkRd : p =X b0ti1 b1 ··· tid bd, b0, b1, . . . , bd ∈ Bo .
B hTi>R := [R′>R
B hTiR′ .
This space should be regarded as non-commutative power series with radius of convergence strictly greater
than R. Observe that if R ≥ maxi kxik, the evaluation evX extends continuously to a homomorphism
B hTi>R → M that sends ti to xi. We denote B hXi>R = evX (B hTi>R).
It is readily seen that the derivations ∂i, i = 1, . . . , n, extend to derivations on B hTi>R that are valued
⊗B hTi◦>R. The evaluation map on B hTi ⊗ B hTi◦ extends to
in the projective tensor product B hTi>R
⊗B hTi◦>R and is valued in M ¯⊗M◦. Consequently, when R ≥ maxi kxik, any d ∈ Der1⊗1(B ⊂
B hTi>R
B hXi) can be extended to evX p ∈ B hXi>R by
d(evX p) :=
evX ∂i(p)#d(xi).
n
Xi=1
10
Charlesworth and Nelson
Free Stein Irregularity and Dimension
That is,
Der1⊗1(B ⊂ B hXi) ⊂ Der1⊗1(B ⊂ B hXi>R).
In fact, the above inclusion is an equality. Indeed, all concerned derivations are closable by virtue of having
1 ⊗ 1 in the domain of their adjoints. Consequently, such a derivation on B hXi>R is uniquely determined
by its values on B hXi. It follows that for Y ∈ B hXim
>R, if B hY i>R = B hXi>R then σ(Y : B) = σ(X : B).
3.2 The special case of B = C. We consider now the special case B = C. Of particular interest to us
will be the case when ∂ gives a closable densely defined operator ∂ : L2(M ) → L2(M ¯⊗M◦)n, in which case
we denote its closure by ¯∂. (We will see in Corollary 4.7 that this is equivalent to the condition σ(X) = n.)
Since ∂ is a derivation which is symmetric in the sense that
ha · ∂(b), ∂(c)i2 = h∂(c∗), ∂(b∗) · a∗i2
a, b, c ∈ ChXi ,
it follows from [DL92] that ¯∂ is a symmetric derivation on dom( ¯∂) ∩ M , which is itself a ∗-algebra.
Theorem 3.6. Let M = W ∗(X). Suppose ∂ : L2(M ) → L2(M ¯⊗M◦)n gives a closable densely defined
operator. Then for any Y = (y1, . . . , ym) ∈ (cid:0)dom( ¯∂) ∩ M(cid:1)m
with ¯∂(yj) ∈ (M ¯⊗M◦)n for each j = 1, . . . m,
we have σ(X) = σ(X, Y ).
Proof. First note that since dom( ¯∂) ∩ M is a ∗-algebra, it contains ChX, Y i. Moreover, since each ¯∂yj is a
bounded operator, ¯∂p is a bounded operator for every p ∈ ChX, Y i.
Now, given d ∈ Der1⊗1(C ⊂ ChXi) define d : ChX, Y i → L2(M ¯⊗M◦) by
d(p) =
n
Xi=1
¯∂i(p)#d(xi).
We claim d ∈ Der1⊗1(C ⊂ ChX, Y i). Indeed, it is a derivation by virtue of ¯∂ being a derivation on dom( ¯∂)∩
M ⊃ ChX, Y i. To see that 1 ⊗ 1 ∈ dom( d∗), note that for any p ∈ ChX, Y i there is a sequence (pk)k∈N ⊂
ChXi converging to p in L2(M ) with (∂(pk))k∈N converging to ¯∂(p) in L2(M ¯⊗M◦)n. Consequently,
hd∗(1 ⊗ 1), pi2 = lim
= lim
k→∞ hd∗(1 ⊗ 1), pki2
k→∞ h1 ⊗ 1, d(pk)i2
n
= lim
k→∞
n
h∂i(pk)∗, d(xi)i2
Xi=1
=
Xi=1(cid:10) ¯∂i(p)∗, d(xi)(cid:11)2 =D1 ⊗ 1, d(p)E2
,
where the second-to-last equality follows from the fact that the adjoint is an isometry on L2(M ¯⊗M◦). Thus
1 ⊗ 1 ∈ dom( d∗) with d∗(1 ⊗ 1) = d∗(1 ⊗ 1). This establishes the claim.
Next consider d ∈ Der1⊗1(C ⊂ ChX, Y i). We claim d(yj) = ¯∂(yj)#d(X) for each j = 1, . . . , m. Indeed,
for each j = 1, . . . , n let (y(k)
))k∈N
)k∈N ⊂ ChXi be a sequence converging to yj in L2(M ) with (∂(y(k)
j
j
11
Charlesworth and Nelson
Free Stein Irregularity and Dimension
converging to ¯∂(yj) in L2(M ¯⊗M◦)n. Then for each j = 1, . . . , m and any a ∈ ChXi ⊗ ChXi◦ we have
= lim
hd(yj ), ai2 = hyj, d∗(a)i2
k→∞Dy(k)
k→∞Dd(y(k)
= lim
n
j
j
, d∗(a)E2
), aE2
Xi=1D∂i(y(k)
Xi=1D∂i(y(k)
n
j
j
= lim
k→∞
= lim
k→∞
n
)#d(xi), aE2
), a#Jτ⊗τ ◦d(xi)E2
This yields the claimed equality since ChXi ⊗ ChXi◦ is dense in L2(M ¯⊗M◦).
The first claim established the existence of a map
=
Xi=1(cid:10) ¯∂i(yj), a#Jτ⊗τ ◦d(xi)(cid:11)2 =(cid:10) ¯∂(yj)#d(X), a(cid:11)2 .
Der1⊗1(C ⊂ ChXi) → Der1⊗1(C ⊂ ChX, Y i)
d 7→ d.
The second claim shows that every derivation in the latter set is completely determined by its values on the
tuple X and ¯∂(yj) for j = 1, . . . , m. It follows that the above map is a bijection, and so by Lemma 3.1 we
have σ(X) = σ(X, Y ).
(cid:3)
Remark 3.7. For R ≥ maxi kxik, Theorem 3.6 applies to any y ∈ ChXi>R as in Remark 3.5.
It also
applies to f (p), where p ∈ dom( ¯∂) ∩ M is self-adjoint with ¯∂(p) ∈ (M ¯⊗M◦)n and f ∈ C1(R). In this case
¯∂(f (p)) = ∂p(f )# ¯∂(p), where ∂p(f ) is the image of the function
f (s, t) =( f (s)−f (t)
s−t
f′(s)
if s 6= t
if s=t
,
under the identification of the unital C∗-algebra generated by p ⊗ 1 and 1 ⊗ p with continuous functions on
its spectrum. Moreover, this can be further extended to Lipschitz functions f on R (see [DL92, Theorem
5.1]).
Lastly, we show that 1 is a lower bound for σ(X) as soon as W ∗(X) contains a diffuse element.
In
particular, this implies that σ(X) ≥ 1 for any generating set X of the hyperfinite II1 factor R.
Theorem 3.8. If W ∗(X) contains a diffuse element, then σ(X) ≥ 1.
Proof. We first note that for any elementary tensor a⊗ b ∈ M ⊗ M◦, d(· ) := [ · , a∗ ⊗ b∗] defines an element
of Der1⊗1(X) with
d∗(1 ⊗ 1) = aτ (b) − τ (a)b.
n
2 ≤ 2
Xi=1
φ : L2(M ¯⊗M◦) → dom(∂∗X ).
Furthermore,
k(Jτ⊗τ ◦d(x1), . . . , Jτ⊗τ ◦d(xn))k2
kxik2ka ⊗ bk2
2.
Thus we can extend the map a ⊗ b 7→ (Jτ⊗τ ◦d(x1), . . . , Jτ⊗τ ◦d(xn)) into a left M ¯⊗M◦-module map
If φ is injective, then it will follow that
σ(X) = dimM ¯⊗M ◦ dom(∂∗X ) ≥ dimM ¯⊗M ◦ L2(M ¯⊗M◦) = 1.
Suppose η ∈ L2(M ¯⊗M◦) satisfies φ(η) = 0. Consequently, [xi, η] = 0 for i = 1, . . . , n and so it follows that
[y, η] = 0 for all y ∈ W ∗(X). Let y0 ∈ W ∗(X) be a diffuse element, which exists by hypothesis. Since we
can identify L2(M ¯⊗M◦) ∼= HS(L2(M )), [y0, η] = 0 implies η = 0. Thus φ is injective.
(cid:3)
12
Charlesworth and Nelson
Free Stein Irregularity and Dimension
4. Relation to Free Entropy.
We now turn to an examination of how free Stein irregularity and dimension relate to the free Fisher
information and non-microstates free entropy dimension(s).
Theorem 4.1. For R > 0, Σ∗R(X : B) = 0 if and only if Φ∗(X : B) ≤ R2.
Proof. Suppose Σ∗R(X : B) = 0. Then there exists a sequence (Ξ(k))k∈N ⊂ L2(B hXi) such that kΞ(k)k2 ≤ R
and Σ∗(X Ξ(k) : B) < 1
k for all k ∈ N. Let Ak be a free Stein kernel of X relative to Ξ(k) over B such that
kAk − 1kHS = Σ∗(X Ξ(k) : B). Then Ak → 1. Hence for every P ∈ B hTin we have
lim
k→∞DΞ(k), evX PE2
= lim
k→∞ hAk, evX ◦J (P )iHS = h1, evX ◦J (P )iHS .
The density of B hXi in L2(B hXi) implies the sequence (Ξ(k))k∈N (since it is uniformly bounded) converges
weakly to some Ξ ∈ L2(B hXi)n. Moreover, the above limit implies Ξ is the conjugate variable of X with
respect to B and
The converse is immediate.
Φ∗(X : B) = kΞk2
2 ≤ lim inf
k→∞ kΞ(k)k2
2 ≤ R2.
(cid:3)
The following result is a minor generalization of [Shl04, Theorem 2.7] (which corresponds to the special
case B = C). We state it here using our notation and terminology, but the core idea of the proof is not
novel.
Proposition 4.2. Let S be a free semicircular family, free from B hXi. Then
tΦ∗(X + √tS : B) ≤ Σ∗(X : B)2.
lim sup
t→0
Proof. Let f : (0,∞) → (0,∞) be any decreasing function such that
(E.g. f (t) = t−1/4). Then
f (t) = +∞, but
lim
t→0
√tf (t) = 0.
lim
t→0
Σ∗f (t)(X : B) = Σ∗(X : B).
lim
t→0
For each t > 0, let Qt ∈ L2(B hXi)n be such that kQtk2 ≤ f (t) and such that there exists a free Stein kernel
At for X relative to Qt over B such that
kAt − 1kHS ≤ Σ∗f (t)(X : B) + t.
Recall that the conjugate variables to X + √tS with respect to B are Et( 1√t
S) where Et : W ∗(B hX, Si) →
W ∗(B(cid:10)X + √tS(cid:11)) is the conditional expectation (cf. [Voi98, Corollary 3.9]). By the same proof as in [Shl04,
Lemma 2.3], it follows that
Thus
√tΦ∗(X + √tS : B)
This tends to Σ∗(X : B) as t → 0.
1
1
√t
S(cid:19) .
Et(Qt) = Et(cid:18)At#
2 = √t(cid:13)(cid:13)(cid:13)(cid:13)
S(cid:19)(cid:13)(cid:13)(cid:13)(cid:13)2
Et(cid:18) 1
√t
√t(cid:13)(cid:13)(cid:13)(cid:13)
S(cid:19)(cid:13)(cid:13)(cid:13)(cid:13)2
Et(cid:18)(1 − At)#
√t(cid:13)(cid:13)(cid:13)(cid:13)
S(cid:13)(cid:13)(cid:13)(cid:13)2
1
√t
≤
(1 − At)#
= k1 − AtkHS + √tf (t)
≤ Σ∗f (t)(X : B) + t + √tf (t).
1
√t
≤
13
+ √tkEt(Qt)k2
+ √tf (t)
(3)
(cid:3)
Charlesworth and Nelson
Free Stein Irregularity and Dimension
We remind the reader that the relative non-microstates free entropy of X with respect to B is defined as
the quantity
χ∗(X : B) :=
1
2Z ∞
0
n
1 + t − Φ∗(X + √tS : B) dt +
n
2
log(2πe),
where S is a free semicircular family free from B hXi (cf. [Voi98, Definition 7.1]). The following is a minor
generalization of [Shl04, Corollary 2.8]. As with the previous result, we state it using our notation and
terminology, but the core idea of the proof is not novel.
Proposition 4.3. Let S be a free semicircular family free from X. Then
χ∗(X + √ǫS : B)
lim sup
ǫ→0
1
2 log ǫ
≤ Σ∗(X : B)2.
Proof. Using [Voi98, Corollary 6.14] and implementing the change of variable t 7→ t − ǫ in the integral
appearing in the above definition of χ∗, we obtain
χ∗(X + √ǫS : B)
1
2 log ǫ
lim sup
ǫ→0
= lim sup
ǫ→0
1
log ǫZ 1
ǫ
n
1 + t − ǫ − Φ∗(X + √tS : B) dt.
Now, for any free Stein kernel A relative to some Q over B we have
Φ∗(X + √tS : B) ≤(cid:18)kEt((1 − A)#
1
t k1 − Ak2
≤
HS +
S)k2 + kEt(A#
S)k2(cid:19)2
1
√t
2
√tk1 − AkHSkQk2 + kQk2
2.
1
√t
Thus
Z 1
ǫ
n
1 + t − ǫ − Φ∗(X + √tS : B) dt ≥ n log(2 − ǫ) + log(ǫ)k1 − Ak2
HS
− 4(1 − √ǫ)k1 − AkHSkQk2 − (1 − ǫ)kQk2
2.
Since log(ǫ) < 0 for ǫ < 1, this in turn implies
χ∗(X + √ǫS : B)
1
2 log ǫ
lim sup
ǫ→0
≤ k1 − Ak2
HS.
Since A was an arbitrary free Stein kernel over B, we obtain the desired inequality.
(cid:3)
We remind the reader that the there are two versions of the relative non-microstates free entropy dimension
of X with respect to B:
δ∗(X : B) := n − lim inf
ǫ→0
χ∗(X + √ǫS : B)
1
2 log ǫ
δ⋆(X : B) := n − lim inf
ǫ→0
tΦ∗(X + √tS : B),
where S is a free semicircular family free from B hXi; moreover, δ∗(X : B) ≤ δ⋆(X : B) (cf. [CS05, Section
4.1.1]1). Thus from Proposition 4.3 we obtain:
Corollary 4.4. For any ∗-algebra B and n-tuple X,
σ(X : B) ≤ δ∗(X : B) ≤ δ⋆(X : B).
Recall that in the self-adjoint one-variable case X = (x), one has
µ({t})2
δ∗(x) = δ⋆(x) = 1 −Xt∈R
by [Voi94, Proposition 6.3] and [Voi98, Propositions 7.5 and 7.6], where µ is the distribution of x on R. Thus,
in particular, the following theorem shows that the above inequalities are in fact equalities.
1Although this paper was interested only in the case B = C, the idea generalizes straightforwardly by using the relative versions
of χ∗ and Φ∗.
14
Charlesworth and Nelson
Free Stein Irregularity and Dimension
Theorem 4.5. Let x ∈ (M, τ ) be self-adjoint with distribution µ on R. Then
Consequently, Σ∗(x) = 0 if and only if x has no atoms.
Σ∗(x)2 =Xt∈R
µ({t})2.
Proof. Recall that in the one-variable case
χ∗(X + √ǫS : B)
1
2 log ǫ
lim inf
ǫ→0
µ({t})2.
=Xt∈R
Thus Corollary 4.4 implies
µ({t})2 ≤ Σ∗(x)2.
Xt∈R
To see the reverse inequality, consider for ǫ > 0 the function
(t − s)
gǫ(t) := 2ZR
(t − s)2 + ǫ2 dµ(s).
Observe that gǫ(t) ≤ 2
ǫ2 (t + τ (x)) ∈ L2(µ). In particular, for any polynomial p we have
Z gǫ(t)p(t) dµ(t) = 2ZZ
(t − s)p(t)
(t − s)2 + ǫ2 dµ(s)dµ(t)
=ZZ (t − s)(p(t) − p(s))
(t − s)2 + ǫ2
(t − s)2
=ZZ
t − s
(t − s)2 + ǫ2
p(t) − p(s)
dµ(s)dµ(t)
dµ(s)dµ(t).
That is, Aǫ(t, s) := (t−s)2
(t−s)2+ǫ2 is a free Stein kernel for x relative to gǫ. So we compute for δ > 0
Σ∗(x)2 ≤ kAǫ − 1k2
ǫ4
L2(µ) =ZZ Aǫ(t, s) − 12 dµ(t)dµ(s)
((t − s)2 + ǫ2)2 dµ(t)dµ(s)
=ZZ
≤ZZt−s≥δ
ǫ4
δ4 + (µ ⊗ µ)({(t, s) ∈ R2 : t − s < δ}).
≤
δ4 dµ(t)dµ(s) +ZZt−s<δ
ǫ4
1 dµ(t)dµ(s)
Letting first ǫ tend to zero and then δ, we obtain the other inequality.
(cid:3)
Remark 4.6. As a particular example of Theorem 4.5, for x ∈ MN (C)s.a. we have
Σ∗(x)2 =
m2
j
N 2 ,
k
Xj=1
where k ≤ n is the number of distinct eigenvalues of x with respective multiplicities m1, . . . , mk ∈ N.
The inequalities in Corollary 4.4 also enable us to prove the following.
Corollary 4.7. Suppose M = W ∗(X). Then σ(X) = n if and only if J gives a densely defined closable
operator
and if and only if ∂ gives a densely defined closable operator
J : L2(M )n → Mn(L2(M ¯⊗M◦)),
∂ : L2(M ) → L2(M ¯⊗M◦)n.
15
Charlesworth and Nelson
Free Stein Irregularity and Dimension
Proof. Let us suppose that σ(X) = n; since σ(X) ≤ δ∗(X) ≤ n, X has full free entropy dimension.
It
then follows from [CS16] that X satisfies no algebraic relation, and hence we may view J and ∂ as densely
defined operators with the above domains and codomains. Moreover, by Theorem 2.10, J ∗ and ∂∗ are
densely defined, whence J and ∂ are closable.
Contrariwise, when either J or ∂ gives a linear operator, its adjoint as an unbounded operator and its
adjoint arising from evaluation of polynomials in ChTi agree. The closability of J or ∂ is then equivalent
to their adjoints having dense domains, and so Theorem 2.10 yields the result.
(cid:3)
This also allows us to reword Theorem 3.6 as follows:
Corollary 4.8. Let M = W ∗(X). Suppose σ(X) = n. Then for any Y = (y1, . . . , ym) ∈ (cid:0)dom( ¯∂) ∩ M(cid:1)m
with ¯∂(yj) ∈ (M ¯⊗M◦)n for each j = 1, . . . , m, we have σ(X, Y ) = n.
4.1 Regularity hierarchy Let us relate the condition σ(X) = n to other well-studied regularity con-
ditions. We have the following picture:
Φ∗(X) < ∞
=
⇒
χ∗(X) > −∞
=
⇒
σ(X) = n
=
⇒
=
⇒
δ∗(X) = n
The top two arrows are of course well-known results: the first is [Voi98, Proposition 7.9] while the second
follows from [Voi98, Proposition 7.5] and the definition of δ∗ in [CS05, Section 4.1.1]. The bottom two arrows
follow from Theorem 4.1 and Corollary 4.4, respectively. Thus it is natural to ask what the relationship is
between having finite non-microstates free entropy and having full free Stein dimension. In the case n = 1,
we see that the former implies the latter by Theorem 4.5.
Remark 4.9. The above raises some interesting questions:
1. Does χ∗(X) > −∞ imply σ(X) = n in general?
2. Does σ(X) = δ∗(X) in general?
We begin to investigate the first question below; then, in Section 5, we exhibit some cases where the equality
in the second question holds.
In order to be begin analyzing the relationship between these two conditions, consider the the following
quantity:
α := lim sup
R→∞
ln Σ∗R(X)
ln R
∈ [−∞, 0].
(4)
That is, α compares how quickly Σ∗R(X) decays as R grows. Note that if Σ∗(X) 6= 0 we have α = 0; however,
it may be that α = 0 even when Σ∗(X) = 0. Indeed, consider the Example B.3 below.
Proposition 4.10. With α as above, if α < 0 then χ∗(X) > −∞.
Proof. Let α < β < 0. Then there exists R0 > 0 such that for all R ≥ R0 we have
Σ∗R(X) ≤ Rβ.
Let γ ∈ (0, 1). Then substituting R = 1
tγ/2 we have
Σ∗1/tγ/2(X) ≤ t−γβ/2
∀t < t0 :=
1
R2/γ
0
.
Using Equation (3) we therefore have
Φ∗(X + √tS) ≤
1
t (cid:16)t−γβ/2 + t(1−γ)/2(cid:17)2
=(cid:16)t(−γβ−1)/2 + t−γ/2(cid:17)2
Since (−γβ − 1)/2 > −1/2 and −γ/2 > −1/2 we have that the above quantity is integrable on [0, t0].
(cid:3)
16
∀t < t0.
Charlesworth and Nelson
Free Stein Irregularity and Dimension
5. Some Computations of Free Stein Dimension.
We provide some examples in which the free Stein irregularity and dimension can be explicitly computed.
In particular, we show that in these examples the free Stein dimension agrees with the non-microstates
free entropy dimensions. The first result concerns Atiyah's ℓ2-Betti numbers for discrete groups (cf. [Ati76,
CG86]). Also see [Luc02, Chapter 1] for the definition considered here, and [MS05] for the connection to free
entropy dimension.
Proposition 5.1. Let Γ be a discrete group and let x1, . . . , xn ∈ C[Γ]s.a. generate the group algebra. Then
where β(2)
0 (Γ) and β(2)
1 (Γ) are the ℓ2-Betti numbers of Γ.
σ(x1, . . . , xn) = β(2)
1 (Γ) − β(2)
0 (Γ) + 1,
Proof. It was shown in [MS05, Theorem 4.1] that
So, by Corollary 4.4, it suffices to show
δ∗(x1, . . . , xn) = δ∗(x1, . . . , xn) = β(2)
1 (Γ) − β(2)
0 (Γ) + 1.
We make use of the following space from [Shl06, Section 2]: H1 = H◦1
where
σ(x1, . . . , xn) ≥ β(2)
1 (Γ) − β(2)
0 (Γ) + 1.
HS
H◦1 := span{(z1, . . . , zn) ∈ HS(L2(M ))n : ∃Y = Y ∗ unbounded, densely defined with
1 ∈ dom(Y ), [Y, xj] = zj for each j = 1, . . . , n}.
We can identify H1 with a closed subspace in L2(M ¯⊗M◦)n using the identification
L2(M ¯⊗M◦) ∼= HS(L2(M ))
a ⊗ b◦ 7→ aP1b,
∂Z (p) =
evX ◦∂j(p)#zj.
n
Xj=1
where P1 is the rank one projection onto 1 ∈ L2(M ). By [Shl06, Theorem 1], for every Z := (z1, . . . , zn) ∈ H◦1
we have 1 ⊗ 1 ∈ dom(∂∗Z ) where ∂Z : ChTi → L2(M ¯⊗M◦) is the derivation defined by
Observe that if J = Jτ⊗τ ◦ is the Tomita conjugation operator on L2(M ¯⊗M◦), then for p ∈ ChXi we have
n
n
h1 ⊗ 1, evX ∂Z(p)i2 =
Xj=1
h1 ⊗ 1, evX ∂j(p)#zji2 =
Xj=1
hJzj, evX ∂j(p)i2 = hJZ, evX ∂(p)i2 .
Consequently, 1 ⊗ 1 ∈ dom(∂∗Z) if and only if JZ ∈ dom(∂∗). It follows that JH1 ⊂ dom(∂∗) and so
where the latter dimension is as a right M ¯⊗M◦-module. In the proof of [Shl06, Corollary 4] it was shown
that the latter dimension is β(2)
0 (Γ) + 1, and so Theorem 2.10 completes the proof.
(cid:3)
1 (Γ) − β(2)
dimM ¯⊗M ◦ (dom(∂∗)) ≥ dimM ¯⊗M ◦ (H1),
Our final example concerns finite-dimensional von Neumann algebras, for which δ, δ⋆, δ0, and ∆ are
known to agree. We show here that σ can be added to this list.
Corollary 5.2. Consider a finite-dimensional algebra for the form
d
where the λi are positive and sum to one, and trki is the normalized trace on Mki(C). Then for any tuple of
generators X = (x1, . . . , xn), we have
(M, τ ) =
(Mki(C), λitrki ) ,
Mi=1
In particular, σ(X) = δ∗(X) = δ∗(X) = δ0(X) = ∆(X) = 1 − β0(M, τ ).
17
σ(X) = 1 −
λ2
i
k2
i
.
d
Xi=1
Charlesworth and Nelson
Free Stein Irregularity and Dimension
Proof. In the proof of [Shl06, Corollary 5] it is shown that
1 −
d
Xi=1
λ2
i
k2
i
= δ∗(X) = dimM ¯⊗M ◦ (H1),
where H1 is as in the proof of Proposition 5.1. Hence equality with σ(X) follows from the proof of Proposi-
tion 5.1. The remaining equalities are then simply [Shl06, Corollary 5] (see also [CS05], namely Proposition
2.9 and Equation 3.10).
(cid:3)
Appendix A.
In this appendix we will demonstrate that for X self-adjoint and algebraically free, the Mai kernel AΞ (given
in Proposition 1.1) satisfies
kAΞ − 1kHS = Σ∗(X Ξ)
if and only if Ξ = 0. We emphasize that any free Stein kernel attaining the free Stein discrepancy of X is
necessarily contained in the closure of the range of J .
Let d : L2(M ) → L2(M ¯⊗M◦) be the derivation given by commutation against 1⊗1: ζ 7→ ζ⊗1−1⊗ζ. Given
Z = (ζ1, . . . , ζn) ∈ L2(M )n, let D : L2(M )n → L2(M ¯⊗M◦)n be given by applying d to each coordinate:
D(Z) = (d(ζ1), . . . , d(ζn)).
Lemma A.1. Suppose that A ∈ ∂ChXi ⊆ L2(M ¯⊗M◦)n. If (pk)k∈N is a sequence in ChXi so that
A = lim
k→∞
∂pk,
then
Proof. Observe that D(X) ∈ (M ¯⊗M◦)n, so that it has a bounded right action on L2(M ¯⊗M◦)n. Thus the
equation follows from a straightforward computation:
A · D(X) = lim
k→∞
(pk ⊗ 1 − 1 ⊗ pk).
A · D(X) = lim
k→∞
(∂pk) · D(X) = lim
k→∞
n
Xj=1
∂jpk#(xj ⊗ 1 − 1 ⊗ xj) = lim
k→∞
pk ⊗ 1 − 1 ⊗ pk.
(cid:3)
The lemma applies, in particular, to the rows of any free Stein kernel that attains the free Stein discrepancy
of X.
Proposition A.2. Suppose Ξ = (ξ1, . . . , ξn) ∈ L2(M )n ⊖ Cn, and let AΞ be as in Proposition 1.1:
AΞ :=(cid:20) 1
2
(ξi ⊗ 1 − 1 ⊗ ξi)#(xj ⊗ 1 − 1 ⊗ x◦j )(cid:21)n
i,j=1 ∈ Mn(L2(M ¯⊗M◦)).
If kAΞ − 1kHS = Σ∗(X Ξ), then Ξ = 0.
Proof. First note that it suffices to assume that τ (x1) = ··· = τ (xn) = 0. Indeed, let
X = ( x1, . . . , xn) = (x1 − τ (x1), . . . , xn − τ (xn)).
Then clearly Ch Xi = ChXi and consequently Ξ ∈ L2(W ∗( X))n. Moreover, AΞ is unchanged when replacing
X with X. Now for any
A ∈ Mn(L2(W ∗(X) ¯⊗W ∗(X)◦)) = Mn(L2(W ∗( X) ¯⊗W ∗( X)◦)),
Note that the i-th row of AΞ is given by 1
if A is a free Stein kernel for X relative to Ξ, then by the chain rule it is also a free Stein kernel for X relative
to Ξ, and vice versa. Hence Σ∗(X Ξ) = Σ∗( X Ξ) and so, replacing X with X if necessary, we may assume
τ (x1) = ··· = τ (xn) = 0.
2 d(ξi)#D(X) =: ri, and from the assumption that Σ∗(X Ξ) =
kAΞ − 1kHS we have that ri ∈ ∂ChXi ⊂ L2(M ¯⊗M◦)n. Now, pick (pk)k∈N in ChXi so that ∂pk → ri; since
C1 ∈ ker ∂, we may assume τ (pk) = 0, replacing pk by pk − τ (pk) if needed. Then from Lemma A.1, we have
ri · D(X) = lim
k→∞
pk ⊗ 1 − 1 ⊗ pk. Hence
(1 ⊗ τ◦)(ri · D(X)) = lim
k→∞
pk
and
18
(τ ⊗ 1)(ri · D(X)) = − lim
k→∞
pk.
Charlesworth and Nelson
Free Stein Irregularity and Dimension
We compute
n
Xj=1
ξix2
j ⊗ 1 − 2ξixj ⊗ xj + ξi ⊗ x2
j − x2
j ⊗ ξi + 2xj ⊗ xjξi − 1 ⊗ x2
j ξi
n
pk ⊗ 1 − 1 ⊗ pk
= 2ri · D(X)
= 2 lim
k→∞
= [(1 ⊗ τ◦)(2ri · D(X))] ⊗ 1 + 1 ⊗ [(τ ⊗ 1)(2ri · D(X))]
=
Xj=1
Xj=1
j )d(ξi) + 2τ (xj ξi)d(xj ) − 1 ⊗ x2
j )) + 2xjτ (xj ξi) − τ (x2
j ξi)
⊗ 1 + 1 ⊗
j ⊗ 1 + τ (x2
j + τ (x2
ξi(x2
ξix2
j ξi.
=
n
Subtracting common terms on each side, we find
n
Xj=1
τ (ξix2
j ) − 2τ (ξixj )xj − (τ (x2
j ) + x2
j )ξi
n
Xj=1
ξi ⊗ [x2
j − τ (xj )2] + 2xj ⊗ [xj ξi − τ (xj ξi)] − 2[ξixj − τ (ξixj)] ⊗ xj − [x2
j − τ (x2
j )] ⊗ ξi = 0.
(5)
As X is algebraically free, we may find polynomials p and q such that (cid:10)x2
1, p(cid:11) = 1 while p is orthogonal to
all other monomials of degree at most two, and hx1, qi = 1 while q is orthogonal to all other monomials of
degree at most three. Applying the map 1 ⊗ h · , pi2 to the above equality yields
n
ξi +
Xj=1
2xj hxj ξi, pi − [x2
j − τ (x2
j )]hξi, pi = 0,
whence ξi is a polynomial in X of degree at most two. Now, applying 1 ⊗ h · , qi2 to Equation 5 and using
the fact that xjξi is a polynomial of degree at most three, we find
n
2x1 hx1ξi, qi − 2(ξix1 − τ (ξix1)) − hξi, qi
(x2
j − τ (x2
j )) = 0.
Xj=1
From this it follows that ξix1 is a linear combination of 1, x1, x2
2, . . . , x2
combination of 1 and x1; say ξi = s + tx1. Looking at the coefficient of x2
that −2t − hξi, qi = 0; since hξi, qi = t, we have t = 0, whence ξi ∈ C. As ξi ∈ L2(M ) ⊖ C, ξi = 0.
n. But then ξi must be a linear
1 in the above equation, we find
(cid:3)
1, x2
Appendix B.
In this appendix we consider a few informative examples. The first two show that for certain tuples generating
interpolated free group factors L(Ft), the parameter t can be recovered through a formula involving the free
Stein dimension of the tuples.
Example B.1. Let s0, s1, . . . , sn be a free semicircular family. Let B = W ∗(s0) and for each j = 1, . . . , n,
let ej, fj be projections in B that are either equal or orthogonal. Define kj = 1 if ej = fj and kj = 2
otherwise. Then by [Rad94] we have
where
n
M := W ∗(s0, e1s1f1, . . . , ensnfn) ∼= L(Ft)
t = 1 +
ksτ (es)τ (fs).
Xs=1
j=1 kj and let X be the K-tuple consisting of the ejsjfj (and fjsjej if ej 6= fj). We claim
Let K =Pn
(6)
Indeed, define Pt to be the K × K diagonal matrix whose (i, i) entry is ej ⊗ fj if xi = ejsjfj. Note
that Pt is a projection, and by freeness one easily sees that Pt ∈ dom(J ∗X : B) with J ∗X : B(Pt) = X.
σ(X : B) + σ(s0) = t.
19
Charlesworth and Nelson
Free Stein Irregularity and Dimension
Consequently, Σ∗(X : B) ≤ kPt − 1kHS. On the other hand, evX T = X = Pt#X = evX (Pt#T ). Thus for
any A ∈ dom(J ∗X : B) we have
hA, 1iHS = hJ ∗X : B(A), evX Ti2 = hJ ∗X : B, evX (Pt#T )i2 = hA, PtiHS .
Consequently
Thus
kA − 1k2
HS − kPt − 1k2
HS = kAk2
= kAk2
HS − 2Re hA, 1iHS − kPtk2
HS − 2Re hA, PtiHS + kPtk2
HS + 2Re hPt, 1iHS
HS = kA − Ptk2
HS ≥ 0.
n
n
Σ∗(X : B)2 = kPt − 1k2
HS =
k2τ ⊗ τ◦(1 ⊗ 1 − ej ⊗ fj) = K −
k2τ (ej )τ (fj ) = K + 1 − t.
Xj=1
Xj=1
Equation (6) then follows since σ(s0) = 1.
Example B.2. Fix a finite, connected graph Γ = (V, E) with vertex weighting µ : V → [0, 1] satisfying
Pv∈V µ(v) = 1, and let ~Γ = (V, ~E) be the associated directed graph (cf. [HN18]). Recall that the free
graph von Neumann algebra (M(Γ, µ), τ ) is generated by operators {xǫ : ǫ ∈ ~E} and an orthogonal family
of projections {pv : v ∈ V }, which satisfy the following graph relations:
(cid:4)
• τ (pv) = µ(v) for all v ∈ V ;
• x∗ǫ = xǫop for all ǫ ∈ ~E;
• pvxǫpw = δv=s(ǫ)δw=t(ǫ)xǫ for all v, w ∈ V and ǫ ∈ ~E.
Moreover, there is a trace-preserving isomorphism between M and the interpolated free group factor L(Ft)
with parameter
where nv,w is the number of edges connecting v to w.
t := 1 −Xv∈V
µ(v)2 +Xv∈V
nv,wµ(w),
µ(v)Xw∼v
Let X := (xǫ : ǫ ∈ ~E), Y := (pv : v ∈ V ), and B := ChY i. We claim
(7)
By [Har17, Lemma 3.9] (see also [HN18, Lemma 2.1]), one has PV ∈ dom(J ∗X : B) where PV is the projection
given by the ~E × ~E diagonal matrix with (ǫ, ǫ)-entry given by ps(ǫ) ⊗ pt(ǫ). Then one has Σ∗(X : B) ≤
kPV − 1kHS. On the other hand, observe that evX T = X = PV #X = evX (PV #T ). So the other inequality
follows by precisely the same argument as in the previous example. Thus
σ(X : B) + σ(Y ) = t.
Σ∗(X : B)2 = kPV − 1k2
τ ⊗ τ◦(1 ⊗ 1 − ps(ǫ) ⊗ pt(ǫ))
HS = Xǫ∈ ~E
= Xǫ∈ ~E
= ~E − Xv∈V
1 − µ(s(ǫ))µ(t(ǫ))
µ(v)Xw∼v
nv,wµ(w),
Finally, appealing to Corollary 5.2 yields Equation (7).
(cid:4)
The next example was concocted to demonstrate explicitly that α = 0 in Equation (4) does not imply
Σ∗(x) > 0. It also demonstrates the fact that full free entropy dimension is strictly weaker than finite free
entropy, by explicitly constructing a probability measure with no atoms and infinite logarithmic energy;
while this result is already known, we are not aware of an explicit example in the literature.
Example B.3. Let In ⊂ [0, 1] be a disjoint sequence of intervals such that the Lebesgue measure λ(In) <
e−12n
. Define a function f as follows:
f : R → R≥0
∞
Xn=1
t 7→
1
2nλ(In)
χIn (t).
20
Charlesworth and Nelson
Free Stein Irregularity and Dimension
By construction f is non-negative, integrable, and has mass 1, so it is a probability density; let µ be the
measure with density given by f . We claim that the (negative) logarithmic energy of µ is infinite. Indeed,
ZZR2
log x − y dµ(x) dµ(y) ≤
log x − y dµ(x) dµ(y)
∞
n
∞
Xn=1ZZI 2
Xn=1
≤
= −∞.
log(e−12n
)4−n
Now, since supp(µ) is bounded and has a diffuse component, there exists a bounded, self-adjoint, algebraically
free operator x with spectral measure µ. It follows from Proposition 4.10 that α = 0, and by Theorem 4.5
we have Σ∗(x) = 0.
(cid:4)
if this behaviour can occur when Σ∗(X) > 0; we provide a family of examples to show that it can.
As a decreasing convex function, if R 7→ Σ∗R(X) ever plateaus it remains constant forever. This happens,
for example, when conjugate variables actually exist: Σ∗R(X) = 0 for R ≥pΦ∗(X). One may wonder, then,
√4 − t2 dt is the semicircle law. Then we will
Example B.4. Let µ = 1
show that if a > 2 and x has spectral measure µ, then there is R > 0 so that Σ∗R(x) = Σ∗(x) = 1
2 .
2 δa where dσ = χ[−2,2](t) 1
2 σ + 1
2π
As in the proof of Theorem 4.5, define the functions
As before, we have a free Stein kernel for x relative to gǫ given by
gǫ(t) := 2ZR
t − s
(t − s)2 + ǫ2 dµ(s).
(t − s)2
(t − s)2 + ǫ2 .
Notice that as ǫ → 0, Aǫ(t, s) → χt6=s =: A(s, t) which has kA − 1k2
that A is a free Stein kernel.
L2(µ×µ) = µ({a})2; so it suffices to show
Here we will use the fact that a /∈ supp(σ) to conclude that gǫ converges in L2(µ) as ǫ → 0. This can be
checked by, for example, recognizing that gǫ converges in both L2(δa) and L2(σ): in the former space,
Aǫ(t, s) :=
which converges since a is outside the support of σ; in the latter,
gǫ →Z
1
a − s
dσ(s),
gǫ(t) → Kt +
1
t − a
,
where we have used the fact that the Hilbert transform of the semicircle distribution is t while a is, once
again, outside of the support of σ. Let g = lim
ǫ→0
We claim that A, above, is a free Stein kernel for x relative to g, whereupon Σ∗R(x) = µ({a}) = 1
2 for
R ≥ kgkL2(µ). (However, note that kgkL2(µ) diverges as a → 2.) To see that, notice that ∂∗ is closed since ∂
is densely defined. Since Aǫ ∈ dom(∂∗) with ∂∗(Aǫ) = gǫ (by virtue of being a free Stein kernel) we therefore
have A ∈ dom(∂∗) with ∂∗(A) = g. That is, A is a free Stein kernel for x relative to g.
gǫ with the limit in L2(µ).
(cid:4)
One may be tempted to guess that if A is a Stein kernel for X and Y is arbitrary that there is some Stein
kernel for (X, Y ) of the form
(cid:18)A ∗
∗(cid:19) .
∗
This is true when Y ∈ ChXim or when Y is free from X. However, this does not happen in general.
Example B.5. Let µ be any measure which is diffuse and so that
s+t /∈ L2([0, 1]2, µ × µ). Note that s2 is
diffuse as s is and so 1 ∈ dom(cid:0)∂∗s2(cid:1); we will show that there is no element of the form (1, α) ∈ dom(∂∗(s2,s)).
Notice that because (0, s) and (s2, s) generate the same algebra, the map ρ(0,s),(s2,s) from Proposition 3.4
provides a bijection between the closures of the free Stein kernels. In particular, if (0, b) ∈ dom(∂∗(0,s)) then
1
21
Charlesworth and Nelson
Free Stein Irregularity and Dimension
((s + t)b, b) ∈ dom(∂∗(s2,s)), and every element is of this form. Hence if (1, α) were to be in the domain of
∂∗(s2,s), we would have α = 1
s+t , which is absurd, as we would then have α /∈ L2([0, 1]2, µ × µ).
(cid:4)
References
[Ati76] Michael F. Atiyah, Elliptic operators, discrete groups and von Neumann algebras, 43 -- 72. Ast´erisque, No. 32 -- 33.
MR 0420729
[BM18] Marwa Banna and Tobias Mai, Holder Continuity of Cumulative Distribution Functions for Noncommutative Poly-
nomials under Finite Free Fisher Information, ArXiv e-prints (2018).
[CFM18] Guillaume C´ebron, Max Fathi, and Tobias Mai, A note on existence of free Stein kernels, arXiv e-prints (2018),
[CG86]
[CS05]
[CS16]
arXiv:1811.02926.
Jeff Cheeger and Mikhael Gromov, L2-cohomology and group cohomology, Topology 25 (1986), no. 2, 189 -- 215.
MR 837621
Alain Connes and Dimitri Shlyakhtenko, L2-homology for von Neumann algebras, J. Reine Angew. Math. 586 (2005),
125 -- 168. MR 2180603 (2007b:46104)
Ian Charlesworth and Dimitri Shlyakhtenko, Free entropy dimension and regularity of non-commutative polynomials,
J. Funct. Anal. 271 (2016), no. 8, 2274 -- 2292. MR 3539353
[Dab10] Yoann Dabrowski, A note about proving non-Γ under a finite non-microstates free Fisher information assumption,
[DL92]
J. Funct. Anal. 258 (2010), no. 11, 3662 -- 3674. MR 2606868 (2011d:46135)
E. Brian Davies and J. Martin Lindsay, Noncommutative symmetric Markov semigroups, Math. Z. 210 (1992), no. 3,
379 -- 411. MR 1171180
[FN17] Max Fathi and Brent Nelson, Free Stein kernels and an improvement of the free logarithmic Sobolev inequality, Adv.
[Ge98]
Math. 317 (2017), 193 -- 223. MR 3682667
Liming Ge, Applications of free entropy to finite von Neumann algebras. II, Ann. of Math. (2) 147 (1998), no. 1,
143 -- 157. MR 1609522
[Har17] Michael Hartglass, Free product C∗-algebras associated with graphs, free differentials, and laws of loops, Canad. J.
Math. 69 (2017), no. 3, 548 -- 578. MR 3679687
[Hay18] Ben Hayes, 1-Bounded entropy and regularity problems in von Neumann algebras, Int. Math. Res. Not. IMRN (2018),
no. 1, 57 -- 137. MR 3801429
[HN18] Michael Hartglass and Brent Nelson, Free transport for interpolated free group factors, J. Funct. Anal. 274 (2018),
[HS07]
no. 1, 222 -- 251. MR 3718052
Don Hadwin and Junhao Shen, Free orbit dimension of finite von Neumann algebras, J. Funct. Anal. 249 (2007),
no. 1, 75 -- 91. MR 2338855
[LNP15] Michel Ledoux, Ivan Nourdin, and Giovanni Peccati, Stein's method, logarithmic Sobolev and transport inequalities,
Geom. Funct. Anal. 25 (2015), no. 1, 256 -- 306. MR 3320893
[Luc02] Wolfgang Luck, L2-invariants: theory and applications to geometry and K-theory, Ergebnisse der Mathematik und
ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas.
3rd Series. A Series of Modern Surveys in Mathematics], vol. 44, Springer-Verlag, Berlin, 2002. MR 1926649
Igor Mineyev and Dimitri Shlyakhtenko, Non-microstates free entropy dimension for groups, Geom. Funct. Anal. 15
(2005), no. 2, 476 -- 490. MR 2153907
[MS05]
[MSW17] Tobias Mai, Roland Speicher, and Moritz Weber, Absence of algebraic relations and of zero divisors under the
assumption of full non-microstates free entropy dimension, Adv. Math. 304 (2017), 1080 -- 1107. MR 3558227
[Rad94] Florin Radulescu, Random matrices, amalgamated free products and subfactors of the von Neumann algebra of a
[Shl04]
[Shl06]
[Voi93]
[Voi94]
[Voi96]
[Voi98]
free group, of noninteger index, Invent. Math. 115 (1994), no. 2, 347 -- 389. MR 1258909
Dimitri Shlyakhtenko, Some estimates for non-microstates free entropy dimension with applications to q-semicircular
families, Int. Math. Res. Not. (2004), no. 51, 2757 -- 2772. MR 2130608
, Remarks on free entropy dimension, Operator Algebras: The Abel Symposium 2004, Abel Symp., vol. 1,
Springer, Berlin, 2006, pp. 249 -- 257. MR 2265052
Dan-Virgil Voiculescu, The analogues of entropy and of Fisher's information measure in free probability theory. I,
Comm. Math. Phys. 155 (1993), no. 1, 71 -- 92. MR 1228526
, The analogues of entropy and of Fisher's information measure in free probability theory. II, Invent. Math.
118 (1994), no. 3, 411 -- 440. MR 1296352 (96a:46117)
, The analogues of entropy and of Fisher's information measure in free probability theory. III. The absence
of Cartan subalgebras, Geom. Funct. Anal. 6 (1996), no. 1, 172 -- 199. MR 1371236
, The analogues of entropy and of Fisher's information measure in free probability theory. V. Noncommutative
Hilbert transforms, Invent. Math. 132 (1998), no. 1, 189 -- 227. MR 1618636 (99d:46087)
22
|
1204.5820 | 2 | 1204 | 2012-10-25T22:51:34 | Functoriality of Cuntz-Pimsner correspondence maps | [
"math.OA"
] | We show that the passage from a $C^\ast$-correspondence to its Cuntz-Pimsner $C^\ast$-algebra gives a functor on a category of $C^\ast$-correspondences with appropriately defined morphisms. Applications involving topological graph $C^\ast$-algebras are discussed, and an application to crossed-product correspondences is presented in detail. | math.OA | math |
FUNCTORIALITY OF CUNTZ-PIMSNER
CORRESPONDENCE MAPS
S. KALISZEWSKI, JOHN QUIGG, AND DAVID ROBERTSON
Abstract. We show that the passage from a C ∗-correspondence
to its Cuntz-Pimsner C ∗-algebra gives a functor on a category of
C ∗-correspondences with appropriately defined morphisms. Appli-
cations involving topological graph C ∗-algebras are discussed, and
an application to crossed-product correspondences is presented in
detail.
1. Introduction
Cuntz-Pimsner algebras were first introduced by Pimsner in [Pim97]
as C ∗-algebras associated to C ∗-correspondences with injective left
actions; Katsura extended the definition in [Kat04b] to include C ∗-
correspondences with non-injective left actions. The class of Cuntz-
Pimsner algebras is very rich, containing all Cuntz-Krieger algebras,
crossed products by Z, and topological graph algebras. Accordingly,
there is a pressing need to understand how constructions at the level
of C ∗-correspondences carry over to the Cuntz-Pimsner algebras.
Now, many C ∗-correspondence constructions naturally and neces-
sarily involve multiplier correspondences (M(Y ), M(B)). For exam-
ple, if a group G acts on a correspondence (X, A), then (X, A) em-
beds in the multipliers (M(X ⋊ G), M(A ⋊ G)) of the crossed-product
correspondence, but not in general in (X ⋊ G, A ⋊ G) itself. Our
main result (Corollary 3.6) is that a C ∗-correspondence homomorphism
(X, A) → (M(Y ), M(B)) that is covariant in an appropriate sense
(Definition 3.1) induces a C ∗-algebra homomorphism OX → M(OY )
of the corresponding Cuntz-Pimsner algebras.
Given that the natural embedding of a C ∗-correspondence in its
Cuntz-Pimsner algebra is often degenerate, there is no reason to expect
that a correspondence homomorphism (X, A) → (M(Y ), M(B)) will
automatically extend to (M(X), M(A)); this makes composing two ho-
momorphisms problematic. To deal with this, our notion of covariance
Date: October 8, 2012.
2010 Mathematics Subject Classification. Primary 46L08; Secondary 46L55.
Key words and phrases. Hilbert module, C ∗-correspondence, Cuntz-Pimsner al-
gebra, action, coaction.
1
2
KALISZEWSKI, QUIGG, AND ROBERTSON
incorporates the assumption that the image of X lies in the so-called
restricted multipliers MB(Y ) that were introduced in [DKQ, Appen-
dix A]. We then can show (Theorem 3.5) that the C ∗-homomorphisms
OX → M(OY ) are obtained in a functorial way on an appropriately-
defined category of C ∗-correspondences. Some earlier work has been
done on functoriality of Cuntz-Pimsner algebras [Rob, RS], but our
approach is the first to incorporate multipliers.
In Section 4 we give three applications of our techniques: to topolog-
ical graph actions, topological graph coactions, and to crossed-product
C ∗-correspondences. The first two are really just brief indications of ap-
plications, the first to a recent result [DKQ] concerning free and proper
actions of groups on topological graphs, where the argument was essen-
tially a "bare-hands" special case of our Corollary 3.6, and the second is
a foreshadowing of how we plan to use the technique to define coactions
on Cuntz-Pimsner algebras from suitable coactions on the correspon-
dences. For the third application, we bring our methods to bear on
crossed-product C ∗-correspondences (X ⋊γ G, A ⋊α G). We show that
the canonical embedding (X, A) → (M(X ⋊γ G), M(A ⋊α G)) induces
via functoriality a covariant homomorphism (OX , G) → M(OX⋊γ G).
As a result we find a surjective homomorphism OX ⋊β G → OX⋊γ G,
which serves as an alternative approach to the result of Hao and Ng
[HN08, Theorem 2.10] that when the group G is amenable there is an
∼= OX ⋊β G. By showing there is a maximal
isomomorphism OX⋊γ G
dual coaction on OX⋊γ G, we can show that our surjection is actually an
isomorphism when the group G is amenable, though this requires back-
ground on C ∗-correspondence coactions and is saved for a forthcoming
paper [KQRb].
2. Preliminaries
Given C ∗-algebras A and B, an A − B correspondence (or a C ∗-
correspondence over A and B) is a Hilbert B-module X equipped with
a left A-module action which is implemented by a homomorphism of A
into the C ∗-algebra L(X) of adjointable operators on X. (We refer
to [EKQR06, Kat04a, RW98] for background on Hilbert modules and
further details on C ∗-correspondences.) The homomorphism is gener-
ically called ϕA, but we usually suppress this notation and just write
a · ξ for ϕA(a)(ξ), where a ∈ A and ξ ∈ X. We say X is full
if
spanhX, Xi ⊂ B is dense.
If in addition, the left action is an iso-
morphism ϕA : A → K(X) we will call X an A − B imprimitivity
bimodule. We write (A, X, B) to denote an A − B correspondence X;
when A = B we just write (X, A), and call X an A-correspondence (or
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
3
a C ∗-correspondence over A). In this paper we will make the standing
assumption that all C ∗-correspondences are nondegenerate in the sense
that A · X = X.
Given a C ∗-correspondence (A, X, B), the set K(X) = span{θξ,η
ξ, η ∈ X} is a (closed) two-sided ideal in L(X) called the compact
operators on X, where by definition θξ,η(ζ) = ξ · hη, ζi for ζ ∈ X. The
Banach space L(B, X) of adjointable operators from B to X (where
B is viewed as a Hilbert B-module in the natural way) becomes an
M(A) − M(B) correspondence when equipped with the natural left
action of M(A), right action of M(B), and M(B)-valued inner product
hm, ni = m∗n all given by composition of operators. This is called the
multiplier correspondence of X, and is denoted by M(X). There is an
embedding ξ 7→ Tξ : X → M(X) given by Tξ(b) = ξ · b for b ∈ B. The
strict topology on M(X) is generated by the seminorms
m 7→ kT · mk and m 7→ km · bk for T ∈ K(X), b ∈ B.
A correspondence homomorphism between two C ∗-correspondences
is a triple (π, ψ, ρ) consisting of C ∗-
(A, X, B) and (C, Y, D)
homomorphisms π : A → M(C) and ρ : B → M(D) and a linear map
ψ : X → M(Y ) satisfying
(i) ψ(a · ξ) = π(a) · ψ(ξ),
(ii) ψ(ξ · b) = ψ(ξ) · ρ(b), and
(iii) ρ(hξ, ηi) = hψ(ξ), ψ(η)i.
(Note that condition (ii) follows from (iii).)
A correspondence homomorphism (π, ψ, ρ) is nondegenerate when π
and ρ are nondegenerate C ∗-homomorphisms and ψ(X)·B = Y . In this
case, by [EKQR06, Theorem 1.30] there is a unique strictly continuous
extension ψ : M(X) → M(Y ) such that
(π, ψ, ρ) : (M(A), M(X), M(B)) → (M(C), M(Y ), M(D))
is a correspondence homomorphism, where π and ρ are the usual ex-
tensions of nondegenerate C ∗-homomorphisms to multiplier algebras.
If (X, A) is an A-correspondence and B is a C ∗-algebra, a corre-
spondence homomorphism of the form (π, ψ, π) : (A, X, A) → (B, B, B)
(where B is viewed as a B-correspondence in the natural way) is called
a Toeplitz representation of (X, A) in B, and is denoted (ψ, π). It is a
critical observation, usually attributed to Pimsner [Pim97, Lemma 3.2]
(see also [KPW98, Lemma 2.2]), that a Toeplitz representation of a
C ∗-correspondence determines a homomorphism of the algebra of com-
pact operators. Here we derive this fact (Corollary 2.2 below) from
the more fundamental Proposition 2.1 to emphasize that it is really a
property of the underlying Hilbert module structure on X and B.
4
KALISZEWSKI, QUIGG, AND ROBERTSON
Proposition 2.1. Let X and Y be Hilbert modules over C ∗-algebras A
and B, respectively, and suppose (ψ, ρ) : (X, A) → (M(Y ), M(B)) is a
correspondence homomorphism. Then there is a unique homomorphism
ψ(1) : K(X) → M(K(Y )) such that
ψ(1)(θξ,η) = ψ(ξ)ψ(η)∗
for
ξ, η ∈ X.
If ψ(X) ⊆ Y , then ψ(1)(K(X)) ⊆ K(Y ), with ψ(1)(θξ,η) = θψ(ξ),ψ(η).
Proof. Obviously there can be at most one such ψ(1). For existence, first
suppose ψ(X) ⊆ Y . Without loss of generality, we may assume that
X is full (otherwise, replace A by the closed span of hX, Xi); similarly,
we may assume that Y is full, and that ψ(X) = Y and ρ(A) = B.
Next, let C = K(X), so that X is a C − A imprimitivity bimod-
ule. Let I = ker ρ. The Rieffel correspondence (see [EKQR06, Def-
inition 1.7]) induces an ideal J = X- Ind I := {c ∈ C : cX ⊂ XI}
so that XI and X/XI are J − I and C/J − A/I imprimitivity bi-
modules respectively. Then the quotient maps (qC, qX , qA) comprise
a surjective imprimitivity bimodule homomorphism of (C, X, A) onto
(C/J, X/XI, A/I), and there is an imprimitivity bimodule isomor-
J) = ψ(ξ) for ξ ∈ X. Taking ψ(1) = π ◦ qC, for ξ, η ∈ X we have
ψ(1)(θξ,η) = π(qC(Chξ, ηi)) = π(Chξ, ηi + J) = π(C/J hξ + J, η + Ji)
phism (π, eψ,eρ) of (C/J, X/XI, A/I) onto (K(Y ), Y, B) such that eψ(ξ +
= K(Y )heψ(ξ + J), eψ(η + J)i = K(Y )hψ(ξ), ψ(η)i = θψ(ξ),ψ(η).
For the general case, apply the above to the Hilbert M(B)-module
M(Y ), and note that by [DKQ, Remark A.10] we have K(M(Y )) ⊂
M(K(Y )), with K(M (Y ))hm, ni = mn∗ for m, n ∈ M(Y ).
(cid:3)
Corollary 2.2 ([Pim97, KPW98]). Let (X, A) be a C ∗-correspondence,
and let (ψ, π) be a Toeplitz representation of (X, A) in a C ∗-algebra B.
Then there is a unique homomorphism ψ(1) : K(X) → B such that
ψ(1)(θξ,η) = ψ(ξ)ψ(η)∗.
For a C ∗-correspondence (X, A), we follow Katsura's convention
[Kat04a] and define an ideal JX of A by
JX = {a ∈ A ϕA(a) ∈ K(X) and ab = 0 for all b ∈ ker(ϕA)}.
A Toeplitz representation (ψ, π) of (X, A) in B is Cuntz-Pimsner co-
variant if
ψ(1)(ϕA(a)) = π(a)
for a ∈ JX.
We denote the universal Cuntz-Pimsner covariant Toeplitz representa-
tion by (kX, kA), and the Cuntz-Pimsner algebra by OX. Note that
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
5
kA : A → OX will always be nondegenerate as a C ∗-homomorphism
because of our assumption that all correspondences (X, A) are nonde-
generate. Again, we refer the reader to [Kat04b] for details.
Finally, we will need the theory of "relative multipliers" from [DKQ,
Appendix A], which is useful for extending degenerate homomorphisms.
If (X, A) is a nondegenerate correspondence and κ : C → M(A) is a
nondegenerate homomorphism, the C-multipliers of X are by definition
MC(X) = {m ∈ M(X) κ(C) · m ∪ m · κ(C) ⊂ X}.
The C-strict topology on MC(X) is generated by the seminorms
m 7→ kκ(c) · mk and m 7→ km · κ(c)k for c ∈ C.
When A is viewed as an A-correspondence over itself in the usual way,
MC(A) is a C ∗-subalgebra of M(A).
Relative multipliers possess the following elementary properties:
(i) The C-strict topology is stronger than the relative strict topol-
ogy on MC(X).
(ii) MC(X) is an MC(A)-correspondence with respect to the
restrictions of the operations of the M(A)-correspondence
M(X),
separately C-strictly
continuous.
operations
and the
are
(iii) If X = A, then MC(A) is a C ∗-subalgebra of M(A), and
the multiplication and involution on MC(A) are separately C-
strictly continuous.
(iv) K(MC(X)) ⊂ MC(K(X)).
(v) MC(X) is the C-strict completion of X.
(vi) MC(X) is an M(C)-sub-bimodule of M(X).
The main purpose of relative multipliers is the following extension
theorem [DKQ, Proposition A.11]: Suppose (X, A) and (Y, B) are
(nondegenerate) C ∗-correspondences and κ : C → M(A) and σ : D →
M(B) are nondegenerate homomorphisms. For any correspondence
homomorphism (ψ, π) : (X, A) → (MD(Y ), MD(B)), if there is a non-
degenerate homomorphism λ : C → M(σ(D)) such that
π(κ(c)a) = λ(c)π(a)
for all c ∈ C and a ∈ A,
then there is a unique C-strict to D-strictly continuous correspondence
homomorphism (ψ, π) : (MC(X), MC(A)) → (MD(Y ), MD(B)) that ex-
tends (ψ, π).
A closely related concept is the following, due to Baaj and Skandalis
[BS89].
6
KALISZEWSKI, QUIGG, AND ROBERTSON
Definition 2.3. For an ideal I of a C ∗-algebra A, let
M(A; I) = {m ∈ M(A) mA ∪ Am ⊂ I}.
Lemma 2.4. If I is an ideal of a C ∗-algebra A, then:
(i) M(A; I) is the strict closure of I in M(A);
(ii) if π : A → M(B) is a nondegenerate homomorphism such that
π(I) ⊂ B, then π(M(A; I)) ⊂ MA(B), and π : M(A; I) →
MA(B) is strict to A-strictly continuous.
Proof. This is elementary, and the techniques are similar to those of
[DKQ, Appendix A]. For (i), we first show that I is strictly dense in
M(A; I). Let m ∈ M(A; I), and let {ei} be an approximate identity
for A. Then mei ∈ I for all i, and mei → m strictly in M(A).
To see that M(A; I) is strictly closed in M(A), let {mi} be a net
in M(A; I) converging strictly to m in M(A). We must show that
m ∈ M(A; I). For a ∈ A we have
kmia − mak → 0 and kami − amk → 0,
so ma, am ∈ I because mia, ami ∈ I for all i.
For (ii), note that by (i) it suffices to show that π : I → B is con-
tinuous for the relative strict topology of I in M(A) and the relative
A-strict topology of B ⊂ MA(B). Let {ci} be a net in I converging
strictly to 0 in M(A), and let a ∈ A. Then π(ci)π(a) = π(cia) and
π(a)π(ci) = π(aci) converge to 0 in norm, so π(ci) → 0 A-strictly
in B.
(cid:3)
Remark 2.5. Let I be an ideal of a C ∗-algebra A, and let ρ : A →
M(I) be the canonical homomorphism. Then by Lemma 2.4 the canon-
ical extension ρ : M(A) → M(I) maps M(A; I) into MA(I). In fact,
the restriction ρM (A;I) is injective, although we will not need this here.
3. Functoriality
3.1. Let
Definition
ate C ∗-correspondences.
(ψ, π) : (X, A) → (M(Y ), M(B)) is Cuntz-Pimsner covariant if
A correspondence
be
nondegener-
homomorphism
(X, A)
and
(Y, B)
(i) ψ(X) ⊂ MB(Y ),
(ii) π : A → M(B) is nondegenerate,
(iii) π(JX ) ⊂ M(B; JY ), and
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
7
(iv) the diagram
π
JX
ϕA
K(X)
ψ(1)
M(B; JY )
ϕB(cid:12)(cid:12)
/ MB(K(Y ))
commutes, where ψ(1)
Proposition 2.1.
is the homomorphism provided by
The above definition simplifies when the correspondence homomor-
phism is nondegenerate:
3.2. A nondegenerate
correspondence homomorphism
Lemma
(ψ, π) : (X, A) → (M(Y ), M(B)) is Cuntz-Pimsner covariant if and
only if items (i ) and (iii ) hold in Definition 3.1.
The lemma follows immediately from the following elementary result.
Lemma 3.3 is presumably well-known, but we could not find a reference
in the literature.
Lemma 3.3. For any nondegenerate correspondence homomorphism
(π, ψ, ρ) : (A, X, B) → (M(C), M(Y ), M(D)), the diagram
A
ϕA
L(X)
π
ψ(1)
M(C)
ϕC
/ L(Y )
commutes.
Proof. Fix a ∈ A; we must show that ψ(1)(ϕA(a))η = ϕC(π(a))η for all
η ∈ Y . By nondegeneracy it suffices to consider elements of the form
η = ψ(ξ)d with ξ ∈ X and d ∈ D, in that case we have
ψ(1)(ϕA(a))η = ψ(1)(ϕA(a))ψ(ξ)d = ψ(ϕA(a)ξ)d = ψ(a · ξ)d
= π(a) · ψ(ξ)d = ϕC(π(a))ψ(ξ)d = ϕC(π(a))η.
(cid:3)
The following lemma addresses the overlap between Cuntz-Pimsner
covariant correspondence homomorphisms and Cuntz-Pimsner covari-
ant Toeplitz representations:
Lemma 3.4. Let (X, A) be a nondegenerate C ∗-correspondence, and let
(ψ, π) : (X, A) → M(B) be a Toeplitz representation of (X, A) in a C ∗-
algebra B. Then (ψ, π) is Cuntz-Pimsner covariant as a correspondence
homomorphism into (B, B) (as in Definition 3.1) if and only if π : A →
/
/
/
/
/
/
8
KALISZEWSKI, QUIGG, AND ROBERTSON
M(B) is nondegenerate and (ψ, π) is Cuntz-Pimsner covariant as a
Toeplitz representation. In particular, (kX, kA) : (X, A) → (OX , OX) is
Cuntz-Pimsner covariant in the sense of Definition 3.1.
Proof. This follows from the identifications JB = K(B) = B and
MB(B) = M(B; B) = MB(K(B)) = M(B), and the observation that
ϕB is the inclusion B ֒→ M(B).
(cid:3)
The Cuntz-Pimsner covariant homomorphisms between correspon-
dences are the morphisms in a suitable category, which we now define.
Theorem 3.5. There is a category CPCorres that has:
• nondegenerate C ∗-correspondences as objects, and
• Cuntz-Pimsner covariant homomorphisms (ψ, π) : (X, A) →
(M(Y ), M(B)) (as in Definition 3.1) as morphisms from
(X, A) to (Y, B);
• and in which the composition of (ψ, π) : (X, A) → (Y, B) and
(σ, τ ) : (Y, B) → (Z, C) is (cid:0)σ ◦ ψ, τ ◦ π).
Proof. First of all, to see that composition is well-defined, note that
since τ : B → M(C) is nondegenerate by definition, it follows from
[DKQ, Proposition A.11] that σ : Y → MC(Z) extends uniquely to a
B-strict to C-strictly continuous homomorphism
(cid:0)σ, τ(cid:1) : (MB(Y ), M(B)) → (MC(Z), M(C)).
Thus we get a correspondence homomorphism
(σ ◦ ψ, τ ◦ π) : (X, A) → (MC(Z), M(C)),
which we must check is Cuntz-Pimsner covariant in the sense of Defi-
nition 3.1. Certainly τ ◦ π : A → M(C) is nondegenerate, so it remains
to verify items (iii) -- (iv) in Definition 3.1.
For (iii), since τ is nondegenerate we have
(τ ◦ π)(JX)C = τ (π(JX))τ (B)C ⊂ τ(cid:0)M(B; JY )(cid:1)τ (B)C
= τ(cid:0)M(B; JY )B(cid:1)C ⊂ τ (JY )C ⊂ JZ,
and similarly C(τ ◦ π)(JX) ⊂ JZ. Therefore τ ◦ π(JX) ⊂ M(C; JZ).
For (iv), let a ∈ JX. We must show that
ϕC ◦ (τ ◦ π)(a) = (σ ◦ ψ)(1) ◦ ϕA(a)
in L(Z), and by nondegeneracy it suffices to show equality after mul-
tiplying on the right by ϕC(c) for an arbitrary c ∈ C. Again by non-
degeneracy we can factor c = τ (b)c′ for some b ∈ B and c′ ∈ C, and
then
ϕC ◦ (τ ◦ π)(a)ϕC(c) = ϕC(τ (π(a))))ϕC(cid:0)τ (b)c′(cid:1)
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
9
= ϕC ◦ τ(cid:0)π(a)b(cid:1)ϕC(c′)
= σ(1) ◦ ϕB(cid:0)π(a)b(cid:1)ϕC(c′)
= σ(1)(cid:0)ϕB ◦ π(a)ϕB(b)(cid:1)ϕC(c′)
= σ(1)(cid:0)ψ(1) ◦ ϕA(a)ϕB(b)(cid:1)ϕC(c′)
=(cid:0)σ ◦ ψ(cid:1)(1) ◦ ϕA(a)ϕC ◦ τ (b)ϕC(c′)
=(cid:0)σ ◦ ψ(cid:1)(1)
=(cid:0)σ ◦ ψ(cid:1)(1) ◦ ϕA(a)ϕC(c).
◦ ϕA(a)ϕC(cid:0)τ (b)c′(cid:1)
= σ(1) ◦ ψ(1) ◦ ϕA(a)σ(1) ◦ ϕB(b)ϕC(c′)
We have thus verified that composition is well-defined in CPCorres.
To see that composition is associative is a routine exercise in the
definitions and the properties of "barring" (see, e.g., [aQRW11, Ap-
pendix A]): if also (ζ, ρ) : (Z, C) → (W, D) in CPCorres then
(ζ, ρ) ◦(cid:0)(σ, τ ) ◦ (ψ, π)(cid:1) = (ζ, ρ) ◦ (σ ◦ ψ, τ ◦ π)
=(cid:0)ζ ◦ (σ ◦ ψ), ρ ◦ (τ ◦ π)(cid:1)
=(cid:0)(ζ ◦ σ) ◦ ψ, (ρ ◦ τ ) ◦ π(cid:1)
=(cid:16)ζ ◦ σ ◦ ψ, ρ ◦ τ ◦ π(cid:17)
=(cid:0)ζ ◦ σ, ρ ◦ τ(cid:1) ◦ (ψ, π)
=(cid:0)(ζ, ρ) ◦ (σ, τ )(cid:1) ◦ (ψ, π).
It is now clear that (idX , idA) is an identity morphism on each object
(cid:3)
(X, A), and therefore CPCorres is a category.
(X, A) and (Y, B) be nondegenerate C ∗-
Corollary 3.6. Let
correspondences, and let (ψ, π) : (X, A) → (M(Y ), M(B)) be a
Cuntz-Pimsner covariant correspondence homomorphism. Then there
is a unique homomorphism Oψ,π making the diagram
(X, A)
(ψ,π)
(MB(Y ), M(B))
(kX ,kA)
(kY ,kB)
OX
Oψ,π
/ MB(OY )
commute. Moreover, Oψ,π is nondegenerate, and is injective if π is.
Proof. Applying Theorem 3.5 with (Z, C) being the C ∗-algebra OY
viewed as a correspondence over itself in the canonical way, we see that
(kY ◦ ψ, kB ◦ π) is a Cuntz-Pimsner covariant Toeplitz representation
/
/
/
10
KALISZEWSKI, QUIGG, AND ROBERTSON
(σ, ν) : X → MB(OY ). Then the universal property of OX gives the
unique homomorphism
Oψ,π =(cid:0)kY ◦ ψ(cid:1) ×(cid:0)kB ◦ π(cid:1).
Nondegeneracy of Oψ,π follows from nondegeneracy of π and kB,
since
OY = kB ◦ π(A)OY = Oψ,π ◦ kA(A)OY
= Oψ,π(kA(A))OY ⊆ Oψ,π(OX )OY ⊆ OY
implies equality throughout.
If π is injective, then so is kB ◦ π, so to show Oψ,π is injective we
can apply the Gauge-Invariant Uniqueness Theorem [Kat04b, Theo-
rem 6.4]: let γ : T → Aut OY be the gauge action. It suffices to observe
that for all z ∈ T, ξ ∈ X, and a ∈ A we have
γz ◦ kY ◦ ψ(ξ) = zkY ◦ ψ(ξ)
γz ◦ kB ◦ π(a) = kB ◦ π(a).
(cid:3)
Recall from, e.g., [aQRW11], that there is a category C∗
nd that has:
• C ∗-algebras as objects, and
• nondegenerate homomorphisms π : A → M(B) as morphisms
from A to B;
• and in which the composition of π : A → B and τ : B → C is
τ ◦ π.
Theorem 3.7. The assignments X 7→ OX and (ψ, π) 7→ Oψ,π define
a functor from CPCorres to C∗
nd.
Proof. First of all, it follows from Corollary 3.6 that if (ψ, π) : (X, A) →
(Y, B) in CPCorres then Oψ,π is a morphism from OX to OY in C∗
nd.
Moreover, OidX ,idA = idOX by uniqueness.
To see that compositions are preserved, let (ψ, π) : (X, A) → (Y, B)
and (σ, τ ) : (Y, B) → (Z, C) in CPCorres. We have
O(σ,τ )◦(ψ,π)◦kX = kZ ◦ (σ ◦ ψ) = kZ ◦ σ ◦ ψ
= Oσ,τ ◦ kY ◦ ψ = Oσ,τ ◦ Oψ,π ◦ kX ,
and similarly
O(σ,τ )◦(ψ,π) ◦ kA = Oσ,τ ◦ Oψ,π ◦ kA,
so that O(σ,τ )◦(ψ,π) = Oσ,τ ◦ Oψ,π in C∗
nd.
(cid:3)
4. Applications
We give three applications of Corollary 3.6.
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
11
Topological graph actions. Our first application is historical; we
show that in [DKQ] the germ of the idea of Corollary 3.6 was introduced
in an ad-hoc way, in a very special case.
Theorem 5.6 of [DKQ] shows that if a locally compact group G acts
freely and properly on a topological graph E, then the quotient E/G
is also a topological graph, and C ∗(E) ⋊r G and C ∗(E/G) are Morita
equivalent. The strategy for proving Morita equivalence in [DKQ] is to
construct an isomorphism of C ∗(E/G) with Rieffel's generalized fixed-
point algebra C ∗(E)G, after showing that the action of G on C ∗(E) is
saturated and proper in the sense of [Rie90], and then appealing to the
imprimitivity theorem C ∗(E) ⋊r G ∼M C ∗(E)G of [Rie90].
By definition, C ∗(E)G is a C ∗-subalgebra of M(C ∗(E)), and the
isomorphism of C ∗(E/G) onto C ∗(E)G is constructed from a Cuntz-
Pimsner covariant Toeplitz representation (τ, π) of the topological-
graph correspondence (X(E/G), C0((E/G)0)) in M(C ∗(E)), which
in turn is constructed via a correspondence homomorphism (µ, ν)
from (X(E/G), C0((E/G)0)) to (M(X(E)), M(C 0(E0))). The proof
in [DKQ] that (τ, π) is Cuntz-Pimsner covariant essentially uses a
special case of the concept of Cuntz-Pimsner covariant correspondence
homomorphisms defined in Definition 3.1.
We will now explain this in more detail. First, recall that the C0(E0)-
correspondence X(E) is a completion of Cc(E1), the Katsura ideal
JX(E) can be identified with C0(E0
rg is a certain open subset
of E0, and the topological-graph algebra C ∗(E) is the Cuntz-Pimsner
algebra OX(E). It will help the exposition to introduce the following
temporary notation:
rg), where E0
• A = C0((E/G)0)
• X = X((E/G))
• Arg = JX
• B = C0(E0)
• Y = X(E)
• Brg = JY
• q : E → E/G is the quotient map (both for edges E1 and
vertices E0).
(Warning: the roles of X, A and Y, B between [DKQ] and here are
switched, to allow more convenient reference to the methods of the
current paper.) [DKQ] constructed the correspondence homomorphism
(µ, ν) : (X, A) → (MB(Y ), M(B))
starting with
• ν(f ) = f ◦ q
12
KALISZEWSKI, QUIGG, AND ROBERTSON
• (µ(ξ) · g)(e) = ξ(q(e))g(e) for ξ ∈ Cc((E/G)1), g ∈ Cc(E0),
and e ∈ E1.
Then the pair (X, A) was mapped into M(OY ) in [DKQ] by the com-
mutative diagram
(X, A)
(µ,ν)
(MB(Y ), M(B))
)❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙
(τ,π)
(kY ,kB)
MB(OY )
In [DKQ] it was then recognized that Cuntz-Pimsner covariance of
(τ, π) is expressed by commutativity of the left-hand triangle of the
following diagram, which is a version of [DKQ, page 1547, diagram (5)]
in which some of the notation has been modified to be consistent with
the current paper:
(4.1)
JX
ϕA
ν ′
MB(JY )
MB(OY )
ϕB
π
%❑❑❑❑❑❑❑❑❑❑
9sssssssss
τ (1)
kB
w♣♣♣♣♣♣♣♣♣♣♣
g◆◆◆◆◆◆◆◆◆◆◆
k(1)
Y
K(X)
µ(1)
/ MB(K(Y )),
and the strategy was to verify that the other parts of the diagram com-
mute. Here the homomorphism ν′ is constructed from the commutative
diagram
ν
M(B; JY )
JX
$❍❍❍❍❍❍❍❍❍
ν ′
xqqqqqqqqqqq
MB(JY ),
where the inclusion M(B : JY ) ֒→ MB(JY ) is given by restriction
rg.
g 7→ gY 0
In [DKQ] it was not recognized that in fact it would be better to do
away with the map ν′ altogether, so that the outer square of (4.1) is
/
/
)
/
/
%
w
9
/
g
/
/
$
K
k
x
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
13
replaced by
JX
ϕA
ν
M(B; JY )
ϕB
K(X)
µ(1)
/ MB(K(Y )),
whose commutativity is precisely our definition Definition 3.1 of Cuntz-
Pimsner covariance of the correspondence homomorphism (µ, ν). The
computations in [DKQ] were much more painstakingly "bare-hands"
than in the current paper, because, again, the techniques were entirely
ad-hoc, whereas here we take a more conceptual and systematic ap-
proach, developing appropriate machinery along the way.
Topological graph coactions. In a forthcoming paper [KQ], a con-
tinuous map ("cocycle") κ on a topological graph E with values in a lo-
cally compact group G is used to construct a coaction δ of G on C ∗(E),
with an eye toward proving that the crossed product C ∗(E) ⋊δ G is
isomorphic to the C ∗-algebra of the skew-product topological graph
E ⋊κ G, thereby generalizing [KQRa, Theorem 2.4] from the discrete
case.
To describe this application, let
• A = C0(E0)
• X = X(E),
and recall, e.g., from the discussion at the beginning of this section,
that C ∗(E) is the Cuntz-Pimsner algebra of the C ∗-correspondence
(X, A).
To be a coaction, δ must in particular be a homomorphism from
C ∗(E) to M1⊗C ∗(G)(C ∗(E) ⊗ C ∗(G)). As usual, δ is constructed from a
Cuntz-Pimsner covariant Toeplitz representation (δX , δA) of (X, A) in
the C ∗-algebra M(C ∗(E) ⊗ C ∗(G)), which in turn is constructed via
a correspondence homomorphism (σ, idA ⊗ 1) from (X, A) to (M(X ⊗
C ∗(G)), M(A ⊗ C ∗(G))). Techniques based upon Corollary 3.6 will
be used to show that the correspondence homomorphsim (σ, idA ⊗ 1)
gives rise to a coaction δ on C ∗(E).
Interestingly, however, we will
need a slight strengthening of the Cuntz-Pimsner covariance condition
of Definition 3.1. The problem is that, due to nonexactness of minimal
C ∗-tensor products, we have no reason to believe that the minimal
tensor product OX ⊗ C ∗(G) coincides with the Cuntz-Pimsner algebra
OX⊗C ∗(G) of the external-tensor-product correspondence (where C ∗(G)
is regarded as a correspondence over itself in the standard way). In fact,
/
/
/
14
KALISZEWSKI, QUIGG, AND ROBERTSON
the basic theory of coactions on Cuntz-Pimsner algebras will require a
significant amount of work, which we will do in [KQRb].
Anyway, once we have the machinery necessary to construct
coactions on Cuntz-Pimsner algebras, our application in [KQ]
will go roughly as follows: first of all, a G-valued cocycle on a
topological graph E is just a continuous map κ : E1 → G. Since
the A-correspondence X is a completion of Cc(E1), and the group
G embeds as unitary multipliers of C ∗(G), we are led to regard the
cocycle κ as an adjointable operator v on X ⊗ C ∗(G), and then we
are able to define a coaction on X by σ(ξ) = v(ξ ⊗ 1), where ξ ⊗ 1 is
regarded as a multiplier of the correspondence X ⊗ C ∗(G). This is a
continuous version of the coaction χ{e} 7→ χ{e} ⊗ κ(e) of [KQRa]. We
emphasize that the justification that this actually gives a coaction will
depend upon the preparation to come in [KQRb].
C ∗-correspondence action crossed products. Let (γ, α) be an ac-
tion of a locally compact group G on a nondegenerate correspondence
(X, A). The crossed product is the completion (X ⋊γ G, A ⋊α G) of the
pre-correspondence (Cc(G, X), Cc(G, A)) with operations
(f · ξ)(s) =ZG
(ξ · f )(s) =ZG
hξ, ηi(s) =ZG
f (t) · γt(cid:0)ξ(t−1s)(cid:1) dt
ξ(t) · αt(cid:0)f (t−1s)(cid:1) dt
αt−1(cid:0)(cid:10)ξ(t), η(ts)(cid:11)(cid:1) dt
for f, g ∈ Cc(G, A) and ξ, η ∈ Cc(G, X). (We refer to, e.g., [EKQR00],
[HN08], [EKQR06, Chapters 2 and 3], and [Kas88] for the elementary
theory of actions and crossed products for correspondences.)
Since (γs, αs) : (X, A) → (X, A) is a correspondence homomorphism
for each s ∈ G, Proposition 2.1 provides homomorphisms γ(1)
: K(X) →
s
K(X), which by uniqueness give rise to an action (γ(1), γ, α) of G on
the correspondence (K(X), X, A), and such that ϕA : A → M(K(X))
is α − γ(1) equivariant.
There is an isomorphism (see, e.g., [Kas88, 3.11])
satisfying
τ : K(X ⋊γ G)
∼=−→ K(X) ⋊
γ(1) G
τ (θξ,η)(s) =ZG
θξ(t),γs(η(s−1t))∆(s−1t)dt,
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
15
where ξ, η ∈ Cc(G, X), s ∈ G and ∆ is the modular function of G, and
moreover the diagram
A ⋊α G
commutes.
ϕA⋊α G
)❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚
ϕA⋊G
M(cid:0)K(X ⋊γ G)(cid:1)
M(cid:0)K(X) ⋊
γ(1) G(cid:1)
τ∼=
Let (iA, iG) : (A, G) → M(A ⋊α G) be the canonical covariant homo-
morphism.
Proposition 4.1. There is a Cuntz-Pimsner covariant correspondence
homomorphism
(iX, iA) : (X, A) →(cid:0)M(X ⋊γ G), M(A ⋊α G)(cid:1)
such that for x ∈ X, f ∈ Cc(G, A), s ∈ G we have
(4.2)
Proof. We first claim that for fixed x ∈ X, (4.2) uniquely determines
an operator iX (x) : A ⋊α G → X ⋊γ G with adjoint given by
(cid:0)iX(x) · f(cid:1)(s) = x · f (s).
(cid:0)iX (x)∗ξ(cid:1)(s) = hx, ξ(s)i
for ξ ∈ Cc(G, X) and s ∈ G. Indeed, (4.2) certainly defines a right
Cc(G, A)-module map Cc(G, A) → Cc(G, X), and we can check the
adjoint property on generators: for any ξ ∈ Cc(G, X), f ∈ Cc(G, A),
and s ∈ G we simply calculate
(cid:10)iX(x)f, ξ(cid:11)(s) =ZG
=ZG
=ZG
=ZG
αt−1(cid:0)(cid:10)(iX (x)f )(t), ξ(ts)(cid:11)(cid:1) dt
αt−1(cid:0)f (t)∗hx, ξ(ts)i(cid:1) dt
αt(cid:0)f (t−1)∆(t−1)hx, ξ(t−1s)i(cid:1) dt
f ∗(t)αt(cid:0)hx, ξ(t−1s)i(cid:1) dt,
proving the claim.
It is now straightforward to verify that the pair (iX , iA) is a corre-
spondence homomorphism. For example, for x, y ∈ X, f ∈ Cc(G, A)
and s ∈ G we calculate
(cid:0)hiX (x), iX(y)if(cid:1)(s) =(cid:0)iX (x)∗iX(y) · f(cid:1)(s) =(cid:10)x, (iX(y) · f (s)(cid:11)
= hx, y · f (s)i = hx, yif (s) =(cid:0)iA(hx, yi)f(cid:1)(s)
/
/
)
16
KALISZEWSKI, QUIGG, AND ROBERTSON
as required.
To show that (iX, iA) is Cuntz-Pimsner covariant, by Lemma 3.2 it
suffices to show that (iX , iA) is nondegenerate and satisfies items (i)
and (iii) in Definition 3.1.
We already know that the coefficient map iA is nondegenerate. Then
for x ∈ X, a ∈ A, g ∈ Cc(G) we have
iX(x)·(cid:0)iA(a)iG(g)(cid:1) = iX(x · a)iG(g),
so nondegeneracy of (iX , iA) follows since X · A = X and
X ⋊γ G = iX(X) · iG(Cc(G)).
Item (i), that iX(X) ⊂ MA⋊αG(X ⋊γ G), is clear from the definition.
To verify item (iii), that iA(JX) ⊂ M(A ⋊α G; JX⋊γG), fix a ∈ JX.
Firstly, [HN08, Lemma 2.6(a)] says that JX is α-invariant, and we
know from [HN08, Proposition 2.7] that JX ⋊α G ⊂ JX⋊γG. For any
f ∈ Cc(G, A) and s ∈ G we have
so iA(a)f ∈ Cc(G, JX). Hence we get
(cid:0)iA(a)f(cid:1)(s) = af (s) ∈ JX
iA(JX )(A ⋊α G) ⊂ JX ⋊α G ⊂ JX⋊γ G
as required.
(cid:3)
Proposition 4.2. With iX as in Proposition 4.1, let
OiX ,iA : OX → M(OX⋊γ G)
be the nondegenerate homomorphism vouchsafed by Corollary 3.6. Also
define a strictly continuous unitary homomorphism u : G → M(OX⋊γ G)
by
u = kA⋊αG ◦ iG.
Then the pair (OiX ,iA, u) defines a covariant homomorphism of the C ∗-
dynamical system (OX , G, β) in the Cuntz-Pimsner algebra OX⋊γ G.
Proof. We only need to verify the covariance condition, namely that
for each t ∈ G we have
Ad u(t) ◦ OiX ,iA = OiX ,iA ◦ βt,
and it is enough to check this on the generators from X and A. For
x ∈ X we have
u(t)OiX ,iA ◦ kX(x) = kA⋊αG(cid:0)iG(t)(cid:1)kX⋊γ G(cid:0)iX(x)(cid:1)
= kX⋊γG(cid:0)iG(t) · iX(x)(cid:1)
= kX⋊γG(cid:16)iX(cid:0)γt(x)(cid:1) · iG(t)(cid:17)
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
17
= kX⋊γG(cid:16)iX(cid:0)γt(x)(cid:1)(cid:17)kA⋊αG(cid:0)iG(t)(cid:1)
= OiX ,iA ◦ kX ◦ γt(x)u(t)
= OiX ,iA ◦ βt ◦ kX(x)u(t),
and the calculation for generators from A is similar.
(cid:3)
Proposition 4.3. The integrated form
OiX ,iA × u : OX ⋊β G → OX⋊γ G
of the covariant pair (OiX ,iA, u) is surjective.
Proof. Let f ∈ Cc(G, A). Then kA ◦ f ∈ Cc(G, OX), and since (iA, iG)
is a covariant pair we have
(OiX ,iA × u)(kA ◦ f ) =ZG
OiX ,iA(kA(f (t)))u(t) dt
iA(f (t))iG(t) dt(cid:19)
= kA⋊αG(cid:18)ZG
= kA⋊αG(cid:0)(iA × iG)(f )(cid:1)
= kA⋊αG(f ).
Therefore the image of OiX ,iA × u contains kA⋊αG(A ⋊α G).
Now let ξ ∈ Cc(G, X). A similar calculation shows that
Now, for any f ∈ Cc(G, A), s ∈ G we can calculate
(OiX ,iA × u)(kX ◦ ξ) = kX⋊γG(cid:18)ZG
iX(ξ(t)) · iG(t) dt f(cid:19) (s) =(cid:18)ZG
(cid:18)ZG
iX (ξ(t)) · iG(t) dt(cid:19) .
iX (ξ(t)) · iG(t)f dt(cid:19) (s)
=ZG(cid:0)iX (ξ(t)) · iG(t)f(cid:1)(s) dt
=ZG
ξ(t) ·(cid:0)iG(t)f(cid:1)(s) dt
=ZG
ξ(t) · αt(f (t−1s)) dt
and so we have
= (ξ · f )(s),
ZG
iA(ξ(t)) · iG(t) dt = ξ.
Thus the image of OiX ,iA × u also contains kX⋊γ G(X ⋊γ G), and it now
follows that OiX ,iA is surjective.
(cid:3)
18
KALISZEWSKI, QUIGG, AND ROBERTSON
Remark 4.4. Ideally we would like the map in Proposition 4.3 to be
injective, which would give OX ⋊β G ∼= OX⋊γ G for an arbitrary locally
compact group G. In a forthcoming paper [KQRb] we will use our tech-
niques to prove this when the group G is amenable, thereby recovering
the result previously obtained by Hao and Ng [HN08, Theorem 2.10]
using the theory of correspondence coactions.
References
[BS89]
[DKQ]
S. Baaj and G. Skandalis, C ∗-alg`ebres de Hopf et th´eorie de Kasparov
´equivariante, K-Theory 2 (1989), 683 -- 721.
V. Deaconu, A. Kumjian, and J. Quigg, Group actions on topological
graphs, Ergodic Theory Dynam. Systems 32 (2012), 1527 -- 1566.
[EKQR06]
[EKQR00] S. Echterhoff, S. Kaliszewski, J. Quigg, and I. Raeburn, Naturality and
induced representations, Bull. Austral. Math. Soc. 61 (2000), 415 -- 438.
, A Categorical Approach to Imprimitivity Theorems for C*-
Dynamical Systems, vol. 180, Mem. Amer. Math. Soc., no. 850, Ameri-
can Mathematical Society, Providence, RI, 2006.
G. Hao and C.-K. Ng, Crossed products of C ∗-correspondences by
amenable group actions, J. Math. Anal. Appl. 345 (2008), no. 2, 702 --
707.
[HN08]
[aQRW11] A. an Huef, J. Quigg, I. Raeburn, and D. P. Williams, Full and reduced
coactions of locally compact groups on C ∗-algebras, Expositiones Math.
29 (2011), 3 -- 23.
[KQ]
[KQRa]
[KPW98] T. Kajiwara, C. Pinzari, and Y. Watatani, Ideal structure and simplicity
of the C ∗-algebras generated by Hilbert bimodules, J. Funct. Anal. 159
(1998), no. 2, 295 -- 322.
S. Kaliszewski and J. Quigg, Skew products and coactions for topological
graphs, in preparation.
S. Kaliszewski, J. Quigg and I. Raeburn, Skew products and crossed
products by coactions, J. Operator Theory 46, (2001), 411 -- 433.
S. Kaliszewski, J. Quigg, and D. Robertson, Coactions on Cuntz-
Pimsner algebras, preprint.
G. G. Kasparov, Equivariant KK-theory and the Novikov conjecture,
Invent. Math. 91 (1988), no. 1, 147 -- 201.
[KQRb]
[Kas88]
[Kat04a] T. Katsura, A class of C ∗-algebras generalizing both graph algebras
and homeomorphism C ∗-algebras. I. Fundamental results, Trans. Amer.
Math. Soc. 356 (2004), no. 11, 4287 -- 4322.
[Kat04b] T. Katsura, On C ∗-algebras associated with C ∗-correspondences, J.
[Lan95]
Funct. Anal. 217 (2004), no. 2, 366 -- 401.
E. C. Lance, Hilbert C ∗-modules, London Math. Soc. Lecture Note Ser.,
vol. 210, Cambridge University Press, 1995.
[Pim97] M. V. Pimsner, A class of C ∗-algebras generalizing both Cuntz-Krieger
algebras and crossed products by Z, Free probability theory (Waterloo,
ON, 1995), Fields Inst. Commun., vol. 12, Amer. Math. Soc., Provi-
dence, RI, 1997, pp. 189 -- 212.
FUNCTORIALITY OF CUNTZ-PIMSNER CORRESPONDENCE MAPS
19
[RW98]
[Rie90]
[Rob]
[RS]
I. Raeburn and D. P. Williams, Morita equivalence and continuous-
trace C ∗-algebras, Math. Surveys and Monographs, vol. 60, American
Mathematical Society, Providence, RI, 1998.
M. A. Rieffel, Proper actions of groups on C ∗-algebras, Mappings of
operator algebras (Philadelphia, PA, 1988) (Boston, MA), Birkhauser
Boston, 1990.
D. Robertson, Extensions of Hilbert bimodules and associated Cuntz-
Pimsner algebras, preprint.
D. Robertson and W. Szyma´nski, C ∗-algebras associated to C ∗-
correspondences and applications to mirror quantum spheres, preprint.
(S. Kaliszewski) School of Mathematical and Statistical Sciences,
Arizona State University, Tempe, Arizona 85287
E-mail address: [email protected]
(John Quigg) School of Mathematical and Statistical Sciences, Ari-
zona State University, Tempe, Arizona 85287
E-mail address: [email protected]
(David Robertson) School of Mathematics and Applied Statistics, Uni-
versity of Wollongong, NSW 2522, AUSTRALIA
E-mail address: [email protected]
|
1708.04105 | 1 | 1708 | 2017-08-14T13:05:48 | Opposite algebras of groupoid C*-algebras | [
"math.OA"
] | We show that every groupoid C*-algebra is isomorphic to its opposite, and deduce that there exist C*-algebras that are not stably isomorphic to groupoid C*-algebras, though many of them are stably isomorphic to twisted groupoid C*-algebras. We also prove that the opposite algebra of a section algebra of a Fell-bundle over a groupoid is isomorphic to the section algebra of a natural opposite bundle. | math.OA | math |
OPPOSITE ALGEBRAS OF GROUPOID C ∗-ALGEBRAS
ALCIDES BUSS AND AIDAN SIMS
Abstract. We show that every groupoid C ∗-algebra is isomorphic to its oppo-
site, and deduce that there exist C ∗-algebras that are not stably isomorphic to
groupoid C ∗-algebras, though many of them are stably isomorphic to twisted
groupoid C ∗-algebras. We also prove that the opposite algebra of a section
algebra of a Fell bundle over a groupoid is isomorphic to the section algebra
of a natural opposite bundle.
1. Introduction
Groupoids are among the most widely used models for operator algebras. It is
therefore a basic question whether a given C ∗-algebra A can be realised as C ∗(G)
r (G) for some locally compact topological groupoid G. Many classes of C ∗-al-
or C ∗
for example, graph C ∗-algebras and higher-rank
gebras have groupoid models:
graph C ∗-algebras, C ∗-algebras of actions of inverse semigroups, and C ∗-algebras
associated to foliations. Moreover, it follows from the main results in [5] that ev-
ery Kirchberg C ∗-algebra (that is, every separable, simple, nuclear, purely infinite
C ∗-algebra) has an étale groupoid model.
We show in this paper that not every C ∗-algebra has a groupoid model. We
achieve this by showing that all groupoid C ∗-algebras are self-opposite in the sense
that they are isomorphic to their opposite C ∗-algebras.
Several examples of non-self-opposite C ∗-algebras are already known. The first,
produced by Connes [2], is a non-self-opposite von Neumann factor. Later, examples
of non-self-opposite separable C ∗-algebras were found by Phillips in [10]. All of
Phillips' examples are continuous-trace C ∗-algebras, hence nuclear. Simple and
separable non-self-opposite C ∗-algebras are constructed in [11, 12]; these examples
are non-nuclear, though the one in [12] is exact. It remains open whether there
exists a simple, separable and nuclear non-self-opposite C ∗-algebra [3]. This is
related to Elliott's conjecture (see [15]) because the Elliott invariant (essentially K-
groups) used in the conjecture cannot distinguish a C ∗-algebra A from its opposite
Aop.
Although our result implies the existence of C ∗-algebras with no groupoid model,
it is still possible that such C ∗-algebras can be realised as twisted groupoid C ∗-alge-
bras. That is, they could be isomorphic to C ∗(G, Σ) or C ∗
r (G, Σ), for some twist Σ
over a groupoid G. A twist over G is essentially the same thing as a Fell line bundle
L over G, and C ∗(G, Σ) and C ∗
r (G, Σ) are then the corresponding full and reduced
cross-sectional C ∗-algebras C ∗(G, L) and C ∗
r (G, L). Renault proves in [14] that
every C ∗-algebra A admitting a Cartan subalgebra C0(X) ⊆ A is isomorphic to
C ∗
r (G, Σ) for some (second countable, locally compact Hausdorff) étale essentially
principal groupoid G with G0 = X and some twist Σ on G; furthermore, the pair
(G, Σ) is uniquely determined by the Cartan pair (A, C0(X)).
2010 Mathematics Subject Classification. 46L05, 22A22.
The first author is supported by CNPq (Brazil). The second author was supported by the
Australian Research Council grant DP150101598. We would like to thank Chris Phillips and Ilijas
Farah for helpful discussions and references to examples of non-self-opposite C*-algebras.
1
2
ALCIDES BUSS AND AIDAN SIMS
Kumjian, an Huef and Sims proved in [7] that every Fell C ∗-algebra (in particular,
every continuous-trace C ∗-algebra) is Morita equivalent to a C ∗-algebra with a
diagonal subalgebra in the sense of Kumjian [8]. These diagonal subalgebras are
exactly the Cartan subalgebras (in the sense of Renault) with the unique extension
property: every pure state of the Cartan subalgebra C0(X) extends uniquely to
A. The corresponding twist (G, Σ) that describes (A, C0(X)) is over a principal,
not just essentially principal, groupoid G. After stabilisation, these results imply
that all continuous-trace C ∗-algebras have a twisted groupoid model-including
the examples of Phillips in [10] that do not admit untwisted groupoid models. The
point is that the opposite algebra of C ∗(G, Σ) arises as the C ∗-algebra C ∗(G, Σ) of
the conjugate twist, and this corresponds to taking the negative of the associated
Dixmier–Douady invariant.
We elucidate the above phenomenon by describing the opposite C ∗-algebras
C ∗(G, A)op and C ∗
r (G, A)op of the cross-sectional algebras of arbitrary Fell bundles
A over locally compact groupoids. Specifically, given a Fell bundle A over G, we
construct an appropriate opposite bundle Ao over G, and prove that C ∗(G, A)op ∼=
C ∗(G, Ao). This can also be described in terms of the conjugate Fell bundle ¯A. In
the special case of a Fell line bundle L (that is, a twist over G), this corresponds to
the conjugate line bundle. When L is the trivial line bundle, ¯L = L, and C ∗
r (G; L)
r (G) and C ∗(G), so we recover our earlier result as a
and C ∗(G; L) coincide with C ∗
special case.
For a Fell bundle associated to an action α of a locally compact group G on a
C ∗-algebra A, our result is equivalent to the statement that the opposite C ∗-alge-
bras of the full and reduced crossed products A ⋊α G and A ⋊α,r G are isomorphic
to Aop ⋊αop G and Aop ⋊αop,r G (where αop is the action of G on Aop determined
by α upon identifying A and Aop as linear spaces); this was proved for full crossed
products in [3].
2. Groupoid C*-algebras and their opposites
In this section we show that the full and reduced C*-algebras of a locally compact,
locally Hausdorff groupoid with Haar system are self-opposite. We first briefly recall
how these C ∗-algebras are defined.
Let G be a locally compact and locally Hausdorff groupoid with Hausdorff unit
space G0 and a (continuous) left invariant Haar system λ = {λx}x∈G0. Let Cc(G, λ)
be the ∗-algebra of compactly supported, quasi-continuous sections, that is, the
linear span of continuous functions with compact support f : U → C on open
Hausdorff subsets U ⊆ G. These functions are extended by zero off U and hence
viewed as functions G → C. The continuity of λ means that every such function is
mapped to a continuous function λ(f ) : G0 → C via λ(f )(x) := RG f (g) dλx(g). By
definition, λx is a Radon measure on G with support Gx := r−1(x) for all x ∈ G0.
Recall that the product and involution on Cc(G, λ) are defined by
(f1 ∗ f2)(g) := ZG
f1(h)f2(h−1g) dλr(g)(h)
and
f ∗(g) := f (g−1).
Under these operations and the inductive-limit topology, Cc(G, λ) is a topological
∗-algebra. The I-norm on Cc(G, λ) is defined by
kf kI := max{kλ(f )k∞, kλ(f ∗)k∞}.
The L1-Banach ∗-algebra of G is the completion of Cc(G, λ) with respect to k · kI;
we denote it by L1
I (G, λ). The full C ∗-algebra of G is the universal enveloping
C ∗-algebra of L1
I (G, λ); in other words, it is the C ∗-completion of Cc(G, λ) with
OPPOSITE ALGEBRAS OF GROUPOID C ∗-ALGEBRAS
3
respect to the maximum k · kI-bounded C ∗-norm:
kf ku := sup{kπ(f )k : π is an I-norm decreasing ∗-representation of Cc(G, λ)}.
The regular representations of (G, λ) are the representations
πx : Cc(G, λ) → B(L2(Gx, λx)),
x ∈ G(0)
given by πx(f )ξ(g) := (f ∗ ξ)(g) = RG f (gh)ξ(h−1) dλx(h). Here λx is the image
of λx under the inversion map G → G, g 7→ g−1; so it is a measure with support
Gx = s−1(x). The system of measures (λx)x∈G0 is a right invariant Haar system
on G.
The regular representations of G give rise to a k · kI-bounded C ∗-norm called the
reduced C ∗-norm:
kf kr := sup
x∈G0
kπx(f )k.
The reduced C ∗-algebra of G is the completion of Cc(G, λ) with respect to k · kr. It
is denoted by C ∗
r (G, λ).
Given a groupoid G, we write Gop for the opposite groupoid, equal to G as a
topological space, but with (Gop)(2) = {(h, g) : (g, h) ∈ G(2)} and composition
given by h ·op g = gh. We write λop for the Haar system on Gop defined as the
image of λ under the inversion map regarded as a homeomorphism of G onto Gop.
Theorem 2.1. Let (G, λ) be a locally compact, locally Hausdorff groupoid with Haar
system. The inversion map g 7→ g−1 defines an isomorphism (G, λ) ∼= (Gop, λop) of
topological groupoids with Haar systems. Given f ∈ Cc(G, λ), define f op : G → C by
f op(g) := f (g−1). Then f 7→ f op is an isomorphism Cc(G, λ)
−→ Cc(G, λ)op of topo-
logical ∗-algebras. This isomorphism extends to a Banach ∗-algebra isomorphism
L1
−→ C ∗(G, λ)op
and C ∗
I(G, λ)op and to C ∗-algebra isomorphisms C ∗(G, λ)
∼
−→ C ∗
I (G, λ)
−→ L1
∼
∼
∼
r (G, λ)
r (G, λ)op.
∼
Proof. The map g 7→ g−1 is a homeomorphism G → Gop and satisfies g−1 ·op h−1 =
h−1g−1 = (gh)−1, so it is an isomorphism G
−→ Gop of topological groupoids. The
range (resp. source) map of Gop is the source (resp. range) map of G, and the
inversion map sends the left invariant Haar system λ = (λx)x∈G0 on G to the right
invariant Haar system (λx)x∈G0, which is precisely λop. This yields the isomorphism
(G, λ) ∼= (Gop, λop). The map f 7→ f op is a linear involution (in particular, a
bijection) which is clearly a homeomorphism with respect to the inductive-limit
topology. It is also clearly isometric for the k·kI-norms on Cc(G, λ) and Cc(Gop, λop).
So to prove that it is topological ∗-algebra isomorphism Cc(G, λ) ∼= Cc(G, λ)op and
I (G, λ)op and C ∗(G, λ) ∼= C ∗(G, λ)op, it
extends to isomorphisms L1
suffices to show that f 7→ f op is a ∗-homomorphism. For f ∈ Cc(G, λ),
I (G, λ) ∼= L1
(f op)∗(g) = f op(g−1) = f (g) = f ∗(g−1) = (f ∗)op(g).
So f 7→ f op preserves involution. If f1, f2 ∈ Cc(G, λ), then
(2.2)
while
(2.3)
(f1 ∗ f2)op(g) = (f1 ∗ f2)(g−1) = ZG
f1(h)f2(h−1g−1) dλs(g)(h),
(f op
2 ∗ f op
1 )(g) = ZG
= ZG
f op
2 (h)f1(h−1g) dλr(g)(h)
f1(g−1h)f2(h−1) dλr(g)(h).
Making the change of variables h 7→ gh and applying left invariance of λ shows
that (2.2) and (2.3) are equal.
4
ALCIDES BUSS AND AIDAN SIMS
r (G, λ) ∼= C ∗
To prove that C ∗
r (G, λ)op, observe that the map f 7→ f op gives
an isomorphism L2(Gx, λx) ∼= L2(Gx, λx) = L2(Gop
x ) which induces a unitary
equivalence between the regular representations πx : Cc(G, λ) → B(L2(Gx, λx) and
x ). This yields the equality kf kr = kf opkr which
πop
x : Cc(Gop, λop) → B(L2(Gop
shows that f 7→ f op extends to an isomorphism C ∗
(cid:3)
x , λop
x , λop
r (G, λ) ∼= C ∗
r (Gop, λop).
Remark 2.4. Similar arguments to those above show that the identity map on G, re-
garded as an anti-multiplicative homeomorphism from G to Gop, induces (by compo-
sition) an anti-multiplicative linear isomorphism Cc(G, λ) ∼= Cc(Gop, λop), and there-
fore a topological ∗-algebra isomorphism Cc(G, λ)op ∼= Cc(Gop, λop). This latter
I (Gop, λop), C ∗(G, λ)op ∼= C ∗(Gop, λop),
extends to isomorphisms L1
and C ∗
I(G, λ)op ∼= L1
r (G, λ)op ∼= C ∗
r (Gop, λop).
Another way to prove Theorem 2.1 is to work with conjugate algebras. If A is
a ∗-algebra, its conjugate ∗-algebra ¯A is the conjugate vector space of A endowed
with the same algebraic operations as A. Involution, a 7→ a∗ is then a linear anti-
multiplicative isomorphism A → ¯A and therefore an isomorphism Aop ∼= ¯A. We
have Cc(G, λ) ∼= Cc(G, λ) via ξ 7→ ¯ξ and this extends to isomorphisms L1
I (G, λ) ∼=
I (G, λ), C ∗(G, λ) ∼= C ∗(G, λ) and C ∗
L1
Corollary 2.5. There are (nuclear, separable) C ∗-algebras that are not isomorphic
to either C ∗(G, λ) or C ∗
r (G, λ) for any locally compact, locally Hausdorff groupoid
with Haar system.
r (G, λ) ∼= C ∗
r (G, λ).
Proof. It is known that there are examples of nuclear and separable C ∗-algebras
that are not self-opposite [3, 10].
(cid:3)
Let us say that a ∗-algebra A is self-opposite if A ∼= Aop. Our main result says
that given a topological groupoid with Haar system (G, λ), the ∗-algebras Cc(G, λ),
L1
r (G, λ) are all self-opposite. Both the minimal and the
maximal tensor product of self-opposite C ∗-algebras are again self-opposite because
(A ⊗ B)op ∼= Aop ⊗ Bop.
I (G, λ), C ∗(G, λ) and C ∗
Let K denote the C ∗-algebra of compact operators on a separable, infinite di-
mensional Hilbert space; writing R for the equivalence relation N × N regarded as
a discrete principal groupoid, we have K ∼= C ∗(R) = C ∗
r (R). Hence the preced-
ing paragraph shows that every self-opposite C ∗-algebra is also stably self-opposite.
The converse fails in general: Phillips constructs in [10] examples of (separable,
continuous-trace) non-self-opposite C ∗-algebras which are stably self-opposite. But
Phillips also constructs examples of (separable, continuous-trace) C ∗-algebras that
are not stably self-opposite. This yields the following:
Corollary 2.6. There are separable continuous-trace C ∗-algebras that are not sta-
bly isomorphic to any groupoid C ∗-algebra.
Remark 2.7. By the Brown–Green–Rieffel theorem [1], Corollary 2.6 implies that
there exist separable C ∗-algebras that are not Morita equivalent to a separable (or
even σ-unital) groupoid C ∗-algebra. However, it is unclear whether these examples
could be Morita equivalent to a non-σ-unital groupoid C ∗-algebra.
In [6], in the framework of ZFC enriched with Jensen's diamond principle (a
strengthening of the continuum hypothesis), Farah and Hirshberg construct exam-
ples of non-separable approximately matricial algebras (uncountable direct limits
of the CAR algebra) that are non-self-opposite, so we can also state:
Corollary 2.8. It is consistent with ZFC that there are non-separable approx-
imately matricial (so simple, nuclear) C ∗-algebras that are not isomorphic to a
groupoid C ∗-algebra.
OPPOSITE ALGEBRAS OF GROUPOID C ∗-ALGEBRAS
5
Recall that the ordinary separable AF-algebras admit groupoid models: it is even
known that they are always crossed products for a partial action of the integers,
see [4].
By [7, Theorem 6.6(1)], every separable continuous-trace C ∗-algebra (indeed,
every Fell algebra) is Morita equivalent to a separable C ∗-algebra with a diagonal
subalgebra in the sense of Kumjian [8]. Kumjian shows in [8] that C ∗-algebras
containing diagonals are, up to isomorphism, the C ∗-algebras obtained from twists
on étale principal groupoids. More precisely, this means a pair (G, Σ) consisting
of a (second countable) locally compact Hausdorff étale groupoid G and another
(locally compact Hausdorff, second countable) topological groupoid Σ that fits into
a central groupoid extension of the form
T × G0 ֒→ Σ ։ G,
where T denotes the circle group, and T × G0 is viewed as a (trivial) group bundle,
and hence as a topological groupoid. To a twisted groupoid (G, Σ) one can assign
a full C ∗-algebra C ∗(G, Σ) and a reduced C ∗-algebra C ∗
r (G, Σ), and then every
separable C ∗-algebra containing a diagonal subalgebra has the form C ∗
r (G, Σ) for
some twist Σ over a principal groupoid G. Moreover, the pair (G, Σ) is unique,
up to isomorphism of twisted groupoids. This follows from the more general re-
sult, proved by Renault in [14], that isomorphism classes of Cartan subalgebras
correspond bijectively to isomorphism classes of twisted essentially principal étale
groupoids (meaning twisted groupoids where G is not necessarily principal, but
only essentially principal; see [14] for details). Using these results, we arrive at the
following consequence:
Corollary 2.9. There are separable stable continuous-trace C ∗-algebras that are
not isomorphic to any groupoid C ∗-algebra but which are isomorphic to the reduced
C ∗-algebra of a twisted principal étale groupoid.
Proof. Let A be a separable continuous-trace C ∗-algebra which is not stably iso-
morphic to any groupoid C ∗-algebra as in Corollary 2.6. Let B := A ⊗ K be the
stabilisation of A. Then B is a separable stable continuous-trace C ∗-algebra which
is not isomorphic to any groupoid C ∗-algebra. By [7, Theorem 6.(1)] A is Morita
equivalent to C ∗
r (G, Σ), for some twisted principal étale groupoid (G, Σ). It follows
from the Brown–Green–Rieffel theorem that A ∼= C ∗
r (G, Σ) ⊗ K. To finish the proof
we observe that, again writing R for the discrete equivalence relation N × N, we
have C ∗
(cid:3)
r (G, Σ) ⊗ K ∼= C ∗
r (G × R, Σ × R).
3. Section C ∗-algebras of Fell bundles and their opposites
Let G be a locally compact and locally Hausdorff groupoid endowed with a
continuous Haar system λ, which we fix throughout the rest of the section.
In
this section we generalise our previous result and describe the opposite C ∗-algebras
of the section C ∗-algebras of a Fell bundles over G. Our result generalises the
observation in [3] that (A ⋊α G)op ∼= Aop ⋊αop G for any action α of a locally
compact group G on a C ∗-algebra A.
Fell bundles over topological groupoids are defined in [9]. Only Hausdorff groupoids
are considered there, but the same definition makes sense for locally Hausdorff
groupoids. A Fell bundle over G consists of an upper semicontinuous Banach bun-
dle A over G endowed with multiplications Ag × Ah → Agh, (a, b) 7→ a · b, for
every composable pair (g, h) ∈ G2 and involutions Ag → Ag−1 , a 7→ a∗, for every
g ∈ G. These operations are required to be continuous (with respect to the given
topology on A) and satisfy algebraic conditions similar to those in the definition of
a C ∗-algebra.
6
ALCIDES BUSS AND AIDAN SIMS
We next recall, briefly, how to define the full and reduced C ∗-algebras of a Fell
bundle. Consider the space Cc(G, A) of compactly supported continuous sections
ξ : U → A defined on open Hausdorff subspaces U ⊆ G and extended by zero
outside U and hence viewed as sections ξ : G → A. The continuity of the algebraic
operations on A implies that for ξ, η ∈ Cc(G, A), the formulas
(ξ ∗ η)(g) := ZG
ξ(h) · η(h−1g) dλr(g)(h),
and
ξ∗(g) := ξ(g−1)∗.
define elements ξ ∗ η, ξ∗ ∈ Cc(G, A) and so determine a convolution product ∗ and
an involution ∗ on Cc(G, A). Under these operations, Cc(G, A) is a ∗-algebra; and
indeed, a topological ∗-algebra in the inductive-limit topology.
Since the norm function on A is upper semicontinuous, the function g 7→ kξ(g)k
from G to [0, ∞) is upper semicontinuous and hence measurable. So we can define
the I-norm on Cc(G, A) by
kξkI := sup
x∈G(0)
maxnZGx
ξ(g) dλx(g),ZGx
ξ∗(g) dλx(g)o.
The L1-Banach algebra of A, denoted L1
I (G, A), is defined as the completion of
Cc(G, A) with respect to k · kI . The full C ∗-algebra C ∗(G, A) of A is defined as
the universal enveloping C ∗-algebra of L1
I (G, A): the completion of Cc(G, A) with
respect to the C ∗-norm
kξku := sup{kπ(ξ)k : π is an I-norm decreasing ∗-representation of Cc(G, A)}.
That this is indeed a norm on Cc(G, A), and not just a seminorm, follows from the
existence of the following regular representations.
For each x ∈ G(0), let L2(Gx, A) be the Hilbert Ax-module completion of the
space Cc(Gx, A) of quasi-continuous sections Gx → A with respect to the norm
induced by the Ax-valued inner product
hξηiAx := ZG
ξ(h)∗η(h) dλx(h) = ZG
ξ(h−1)∗η(h−1) dλx(h).
Then for each x ∈ G(0), the regular representation πx : Cc(G, A) → B(L2(Gx, A))
is defined by
(cid:0)πx(ξ)η(cid:1)(g) := ZG
ξ(gh)η(h−1) dλx(h) = ZG
ξ(gh−1)η(h) dλx(h),
for all ξ ∈ Cc(G, A), η ∈ Cc(Gx, A) and g ∈ Gx. The reduced C ∗-norm on Cc(G, A)
is defined by
This is, indeed, a norm: if πx(ξ) = 0 then (ξ ∗ η)(g) = 0 for all η ∈ Cc(Gx, A) and
kξkr := sup
x∈G0
kπx(ξ)k.
g ∈ Gx; so ξ ∗ ξ∗(x) = RG ξ(h)ξ(h)∗ dλx(h) = 0 for all x ∈ G(0), forcing ξGx = 0
for all x. A standard computation shows that kξkr ≤ kξkI. Therefore k · ku is also
a C ∗-norm and k · kr ≤ k · ku. The completion of Cc(G, A) with respect to k · kr is
the reduced section C ∗-algebra of A, and is denoted by C ∗
r (G, A).
Our goal here is to describe the opposite C ∗-algebras C ∗(G, A)op and C ∗
r (G, A)op.
We show that C ∗(G, A)op) ∼= C ∗(G, Ao), for an appropriate opposite Fell bundle Ao
over G associated to A. It is more natural to first define an opposite Fell bundle Aop
over the opposite groupoid Gop and then later use the canonical anti-isomorphism
G ∼= Gop induced by the inversion map to obtain the desired Fell bundle Ao over
G.
The opposite Fell bundle Aop over Gop is defined as follows. As a Banach bundle,
Aop does not differ from A. In particular, the fibres are equal, Aop
g = Ag for all
g ∈ G, and also the topology on Aop is equal to that on A. Moreover, Aop is also
OPPOSITE ALGEBRAS OF GROUPOID C ∗-ALGEBRAS
7
endowed with the same involution as A, which makes sense because G and Gop carry
the same inversion map. The only thing that changes in Aop is the multiplication:
given g, h ∈ Gop the condition sop(g) = rop(h) means r(g) = s(h), so we can use
the multiplication map µh,g : Ah × Ag → Ahg and define the multiplication maps
µop : Aop
g × Aop
h = Ag × Ah → Aop
g·oph = Ahg by µop(a, b) := µ(b, a).
In other words, a ·op b := b · a if we use · and ·op to denote the multiplications on A
and Aop, respectively. It is straightforward to see that Aop is indeed a Fell bundle
over Gop. Now we use the anti-isomorphism Gop ∼= G induced by the inversion
map g 7→ g−1 to form the pullback Fell bundle of Aop. In other words, Ao is a
Fell bundle over G with fibres Ao
g = Ag−1 and the topology induced by the sections
ξo(g) := ξ(g−1) for ξ : U → A a continuous section defined on a Hausdorff open
subset U ⊆ G. The involution map Ao
g−1 is the involution map Ag−1 → Ag
from A and the multiplication map Ao
h → Ao
gh is given by (a, b) 7→ b · a for
all b ∈ Ao
h = Ah−1 and g, h ∈ G with s(g) = r(h).
g = Ag−1 , b ∈ Ao
g → Ao
g × Ao
Theorem 3.1. Let A be a Fell bundle over a locally compact, locally Hausdorff
groupoid with Haar system (G, λ), and consider the Fell bundle Ao over (G, λ)
described above. The map ξ 7→ ξo defined by ξo(g) := ξ(g−1) gives an isomorphism
of topological ∗-algebras Cc(G, A)op ∼
−→ Cc(G, Ao). Moreover, this isomorphism
I (G, Ao) and
extends to an isomorphism of Banach ∗-algebras L1
C ∗-algebras C ∗(G, A)op ∼
Proof. We prove the equivalent assertion that Cc(G, A)op ∼= Cc(Gop, Aop) via the
canonical linear isomorphism Cc(G, A) ∋ ξ 7→ ξop := ξ ∈ Cc(Gop, Aop), and that
this isomorphism extends to isomorphisms
−→ L1
r (G, Ao).
−→ C ∗(G, Ao) and C ∗
I(G, A)op ∼
r (G, A)op ∼
−→ C ∗
I(G, A)op ∼
L1
−→ L1
I (Gop, Aop), C ∗(G, A)op ∼
−→ C ∗(Gop, Aop),
and
r (G, A)op ∼
C ∗
−→ C ∗
r (Gop, Aop).
Since the topologies on A and Aop are the same, the map ξ 7→ ξop is clearly a
linear bijection Cc(G, A) → Cc(Gop, Aop) which is a homeomorphism with respect
to the inductive-limit topologies. Also, this map preserves the involution, that
is, (ξop)∗ = ξ∗ on Cc(G, A) (which is the same as the involution on Cc(G, A)op),
and on Cc(Gop, Aop) because the involutions on A and on Aop are the same. It
remains to check that the map is a homomorphism Cc(G, A)op → Cc(Gop, Aop).
But, remembering that the left Haar system λop on Gop is the right Haar system
(λx)x∈G0 on G, we get
ξop ∗ ηop(g) = ZGop
ξ(h) ·op η(h−1g) d(λop)rop(g)(h)
= ZG
η(gh−1)ξ(h) dλs(g)(h) = ZG
η(gh)ξ(h−1) dλr(g)(h) = (η ∗ ξ)(g)
for all ξ, η ∈ Cc(G, A) and g ∈ G. This shows that the identity map is an anti-
homomorphism Cc(G, A) → Cc(Gop, Aop), that is, a homomorphism Cc(G, A)op →
Cc(Gop, Aop), as desired. A similar computation shows that kξopkI = kξkI and that
I (Gop, Aop)
therefore the identity map extends to an isomorphism L1
and hence also to the corresponding universal enveloping C ∗-algebras C ∗(G, A)op ∼
−→
C ∗(Gop, Aop).
I (G, A)op ∼
−→ L1
r (G, A)op ∼= C ∗
Finally, to check that C ∗
r (Gop, Aop), fix x ∈ G0. The regular
x defines a representation of C ∗(Gop, Aop) by adjointable operators
representation πop
on the right-Hilbert Aop-module L2(Gop
x , Aop). Any right Hilbert module E over
the opposite Bop of a C ∗-algebra B determines a left Hilbert B-module E with left
B-action b·ξ := ξ·b and left B-valued inner product Bhξηi := hηξiBop . This process
8
ALCIDES BUSS AND AIDAN SIMS
x -module L2(Gop
preserves the C ∗-algebras of adjointable operators, meaning that the identity map
on E yields an isomorphism B(EBop ) ∼= B(BE). Applying this to the right Hilbert
x , Aop) we get a left Hilbert Ax-module with left Ax-action given
Aop
x , Aop); the right hand side denotes the right
by a · ξ = ξ ·op a for all ξ ∈ L2(Gop
x , Aop), so it is given by (a ·op ξ)(g) = ξ(g) ·op a(sop(g)) =
Aop
a(r(g)) · ξ(g). The left Ax-valued inner product on L2(Gop, Aop) is given by
x -action on L2(Gop
Ax hξηi = hηξiAop
x = ZGop
η(h)∗ ·op ξ(h) dλop
x (h) = ZG
ξ(h)η(h)∗ dλx(h)
x -module L2(Gop
x , Aop) = Cc(Gx, A). Therefore the left Hilbert Ax-module
for all ξ, η ∈ Cc(Gop
obtained from the right Hilbert Aop
x , Aop) in this way equals the left
Hilbert Ax-module L2(Gx, A) defined as the completion of Cc(Gx, A) with respect
to the norm associated to the left Ax-valued inner product given by the above
formula and the left Ax-action also defined above. Therefore we may view πop
x as a
representation of C ∗(Gop, Aop) on B(Ax L2(Gx, A)) ∼= B(L2(Gop
x ). Under
the isomorphism C ∗(Gop, Aop) ∼= C ∗(G, A)op, this corresponds to the canonical
representation πx of C ∗(G, A)op on Ax L2(Gx, A) via the formula
x , Aop)Aop
πx(ξ)η(g) := (η ∗ ξ)(g) = ZG
η(h)ξ(h−1g) dλx(h)
for ξ ∈ Cc(G, A), η ∈ Cc(Gx, A) and g ∈ Gx. Straightforward computations show
that the above formula defines a representation πx : C ∗(G, A)op → B(Ax L2(Gx, A))
of the opposite C ∗-algebra C ∗(G, A)op.
Given a left Hilbert B-module E, let E denote the dual right Hilbert B-module
of E: as a vector space E = { ξ : ξ ∈ E} is the conjugate of E and the right B-action
and right B-valued inner product are defined by ξ · b := ]b∗ · ξ and hξηiB :=B hξηi.
Then each representation π : Aop → B(BE) of an opposite C ∗-algebra Aop on the
C ∗-algebra of adjointable operators B(BE) of a left Hilbert B-module E induces a
representation πop : A → B(BE)op ∼= B( EB). The isomorphism B(BE)op ∼= B( EB)
we used above is induced by the involution, that is, it sends an operator T ∈ B(BE)
to T ∈ B( EB) defined by T ( ξ) := (T ∗(ξ))∼.
For ξ ∈ Cc(Gx, A), the formula ξ∗(g) := ξ(g−1)∗ determines an element ξ∗ ∈
∼= L2(Gx, A)Ax
Cc(Gx, A). The map ξ 7→ ξ∗ induces an isomorphism (L2(Gx, A))∼
Ax
from the dual Hilbert Ax-module of Ax L2(Gx, A) to the right Hilbert Ax-module
L2(Gx, A) that carries the regular representation πx : C ∗(G, A) → B(L2(Gx, A)Ax).
This isomorphism intertwines the representations πx : C ∗(G, A) → B(L2(Gx, A)Ax )
and πop
). We conclude that
x : C ∗(G, A) → B(Ax L2(Gx, A))op ∼= B((L2(Gx, A)))∼
Ax
kπop
x (ξ)k = kπx(ξo)k = kπop
x (ξo)k = kπx(ξo)k.
Since x ∈ G0 was arbitrary, we get the equality kξokr = kξkr and therefore the
desired isomorphism C ∗
(cid:3)
r (Gop, Aop) ∼= C ∗
r (G, A)op.
Remark 3.2. As in the case of groupoid C ∗-algebras, we can rephrase the preceding
result in terms of conjugate bundles as well. Let A be a Fell bundle over a groupoid
G. For g ∈ G, let Ag be the conjugate vector space of Ag; that is, Ag is a copy
{a : a ∈ Ag} as an abelian group under addition, but with scalar multiplication
given by λa = ¯λa. Via the map a 7→ ¯a, the operations on the Fell bundle A induce
operations on A := Fg∈G Ag: ab = ab, and a∗ = a∗. Under these operations, A is
a Fell bundle over G, called the conjugate bundle of A.
Let Ao be the opposite bundle of A defined above; so Ao
g = Ag−1 . Then the
maps Ao
g ∋ a 7→ a∗ ∈ Ag are linear isometries because the maps a 7→ a∗ and
a 7→ ¯a are both conjugate linear. We have (a ·op b)∗ = (ba)∗ = a∗b∗ = a∗b∗
and (a∗))∗ = a = (a∗)∗, so a 7→ a∗ determines an isomorphism Ao ∼= A of Fell
OPPOSITE ALGEBRAS OF GROUPOID C ∗-ALGEBRAS
9
bundles over G. Thus Theorem 3.1 shows that there is a topological-∗-algebra
isomorphism ξ 7→ ξ from Cc(G, A) to Cc(G, A) given by ξ(g) := ξ(g−1)∗ that
extends to isomorphisms
I(G, A) ∼= L1
L1
I (G, A), C ∗(G, A) ∼= C ∗(G, A),
and C ∗
r (G, A) ∼= C ∗
r (G, A).
Remark 3.3. Remark 3.2 is closely related to the idea behind Phillips' construction
in [10] of non-self-opposite continuous-trace C ∗-algebras A; the observation under-
lying his construction is that the Dixmier–Douady class of the opposite algebra
Aop is the inverse of the Dixmier–Douady class of A. To see how this relates to our
results, fix a compact Hausdorff space X, and let S denote the sheaf of germs of
continuous T-valued functions on X. The Raeburn–Taylor construction [13] shows
that (after identifying H 3(X, Z) with H 2(X, S)) any class δ ∈ H 2(X, S) can be re-
alised as the Dixmier–Douady invariant of a twisted groupoid C ∗-algebra C ∗(G, σ)
associated to a continuous 2-cocycle σ on a principal étale groupoid G with unit
space G(0) = Fi,j Uij for some precompact open cover {Ui} of X. The cocycle σ
determines, and is determined up to cohomology by, the Fell line-bundle Lσ over G
given by Lσ = G × T with twisted multiplication (g, w)(h, z) = (gh, c(g, h)wz) and
the obvious involution. Remark 3.2 shows that C ∗(G, σ)op is given by the conjugate
bundle Lσ, so the corresponding class in H 2(X, S) is determined by the pointwise
conjugate of the class δ; that is, δ(C ∗(G, σ)) = δ(C ∗(G, σ)op)−1.
References
[1] Lawrence G. Brown, Philip Green, and Marc A. Rieffel, Stable isomorphism and strong
Morita equivalence of C ∗-algebras, Pacific J. Math. 71 (1977), no. 2, 349–363, available at
http://projecteuclid.org/euclid.pjm/1102811432.
[2] Alain Connes, A factor not anti-isomorphic to itself, Ann. Math. (2) 101 (1975), 536–554.
[3] Marius Dadarlat, Ilan Hirshberg, and N. Christopher Phillips, Simple nuclear C*-algebras
not equivariantly isomorphic to their opposites (2016), preprint. arXiv: 1602.04612.
[4] Ruy Exel, Approximately finite C ∗-algebras and partial automorphisms, Math. Scand. 77
(1995), no. 2, 281–288, doi: 10.7146/math.scand.a-12566. MR 1379271
[5] Ruy Exel and Enrique Pardo, Self-similar graphs, a unified treatment of Katsura and Nekra-
shevych C∗-algebras (2014), eprint. arXiv: 1409.1107.
[6] Ilijas Farah and Ilan Hirshberg, Simple nuclear C*-algebras not
isomorphic to their
opposites, Proceedings Nat. Acad. Sciences USA 114 (2017), no. 24, 6244–6249,
doi: 10.1073/pnas.1619936114.
[7] Astrid an Huef, Alex Kumjian, and Aidan Sims, A Dixmier–Douady theorem for Fell algebras,
J. Funct. Anal. 260 (2011), no. 5, 1543–1581, doi: 10.1016/j.jfa.2010.11.011. MR 2749438
[8] Alexander Kumjian, On C ∗-diagonals, Canad. J. Math. 38 (1986), no. 4, 969–1008,
doi: 10.4153/CJM-1986-048-0. MR 854149
[9] Alex Kumjian, Fell bundles over groupoids, Proc. Amer. Math. Soc. 126 (1998), no. 4, 1115–
1125, doi: 10.1090/S0002-9939-98-04240-3. MR 1443836
[10] N. Christopher Phillips, Continuous-trace C*-algebras not isomorphic to their opposite alge-
bras, International J. Math. 12 (2001), 263–275.
[11] N. Christopher Phillips, A simple separable C ∗-algebra not
isomorphic to its op-
posite algebra, Proc. Amer. Math. Soc. 132 (2004), no. 10, 2997–3005 (electronic),
doi: 10.1090/S0002-9939-04-07330-7.
[12] N. Christopher Phillips and M. G. Viola, A simple separable exact C*-algebra not anti-
isomorphic to itself, Math. Ann. 355 (2013), 783–799.
[13] Iain Raeburn and Joseph L. Taylor, Continuous trace C ∗-algebras with given Dixmier-Douady
class, J. Austral. Math. Soc. Ser. A 38 (1985), no. 3, 394–407. MR779202
[14] Jean Renault, Cartan subalgebras in C ∗-algebras, Irish Math. Soc. Bull. 61 (2008), 29–63,
available at http://www.maths.tcd.ie/pub/ims/bull61/S6101.pdf.
[15] Mikael Rørdam, Classification of nuclear, simple C ∗-algebras, Classification of nuclear
C ∗-algebras. Entropy in operator algebras, Encyclopaedia of Mathematical Sciences, vol. 126,
Springer, Berlin, 2002, pp. 1–145, doi: 10.1007/978-3-662-04825-2_1.
E-mail address: [email protected]
10
ALCIDES BUSS AND AIDAN SIMS
Departamento de Matemática, Universidade Federal de Santa Catarina, 88.040-900
Florianópolis-SC, Brazil
E-mail address: [email protected]
School of Mathematics and Applied Statistics, The University of Wollongong, NSW
2522, Australia
|
1104.1398 | 1 | 1104 | 2011-04-07T18:32:56 | Semicrossed products of the disc algebra | [
"math.OA"
] | If $\alpha$ is the endomorphism of the disk algebra, $\AD$, induced by composition with a finite Blaschke product $b$, then the semicrossed product $\AD\times_{\alpha} \bZ^+$ imbeds canonically, completely isometrically into $\rC(\bT)\times_{\alpha} \bZ^+$. Hence in the case of a non-constant Blaschke product $b$, the C*-envelope has the form $ \rC(\S_{b})\times_{s} \bZ$, where $(\S_{b}, s)$ is the solenoid system for $(\bT, b)$. In the case where $b$ is a constant, then the C*-envelope of $\AD\times_{\alpha} \bZ^+$ is strongly Morita equivalent to a crossed product of the form $ \rC(\S_{e})\times_{s} \bZ$, where $e \colon \bT \times \bN \longrightarrow \bT \times \bN$ is a suitable map and $(\S_{e}, s)$ is the solenoid system for $(\bT \times \bN, \, e)$ . | math.OA | math |
SEMICROSSED PRODUCTS OF THE DISK ALGEBRA
KENNETH R. DAVIDSON AND ELIAS G. KATSOULIS
Abstract. If α is the endomorphism of the disk algebra, A(D),
induced by composition with a finite Blaschke product b, then the
semicrossed product A(D) ×α Z+ imbeds canonically, completely
isometrically into C(T)×α Z+. Hence in the case of a non-constant
Blaschke product b, the C*-envelope has the form C(Sb) ×s Z,
where (Sb, s) is the solenoid system for (T, b). In the case where
b is a constant, then the C*-envelope of A(D) ×α Z+ is strongly
Morita equivalent to a crossed product of the form C(Se) ×s Z,
where e : T × N −→ T × N is a suitable map and (Se, s) is the
solenoid system for (T × N, e) .
1. Introduction
If A is a unital operator algebra and α is a completely contractive en-
domorphism, the semicrossed product is an operator algebra A ×α Z+
which encodes the covariant representations of (A, α): namely com-
pletely contractive unital representations ρ : A → B(H) and contrac-
tions T satisfying
ρ(a)T = T ρ(α(a))
for all a ∈ A.
Such algebras were defined by Peters [9] when A is a C*-algebra.
One can readily extend Peter's definition [9] of the semicrossed prod-
uct of a C*-algebra by a ∗-endomorphism to unital operator algebras
and unital completely contractive endomorphisms. One forms the poly-
tiai,
where ai ∈ A, with multiplication determined by the covariance rela-
tion at = tα(a) and the norm
nomial algebra P(A, t) of formal polynomials of the form p = Pn
i=0
kpk =
sup
(ρ,T ) covariant(cid:13)(cid:13)
n
Xi=0
T iρ(ai)(cid:13)(cid:13).
2000 Mathematics Subject Classification. 47L55.
Key words and phrases. semicrossed product, crossed product, disk algebra, C*-
envelope.
First author partially supported by an NSERC grant.
Second author was partially supported by a grant from ECU.
1
2
K.R.DAVIDSON AND E.G. KATSOULIS
This supremum is clearly dominated by Pn
i=0 kaik; so this norm is well
defined. The completion is the semicrossed product A ×α Z+. Since
this is the supremum of operator algebra norms, it is also an opera-
tor algebra norm. By construction, for each covariant representation
(ρ, T ), there is a unique completely contractive representation ρ × T of
A ×α Z+ into B(H) given by
ρ × T (p) =
n
Xi=0
T iρ(ai).
This is the defining property of the semicrossed product.
In this note, we examine semicrossed products of the disk algebra
by an endomorphism which extends to a ∗-endomorphism of C(T).
In the case where the endomorphism is injective, these have the form
α(f ) = f ◦ b where b is a non-constant Blaschke product. We show
that every covariant representation of (A(D), α) dilates to a covari-
ant representation of (C(T), α). This is readily dilated to a covariant
representation (σ, V ), where σ is a ∗-representation of C(T) (so σ(z)
is unitary) and V is an isometry. To go further, we use the recent
work of Kakariadis and Katsoulis [6] to show that C(T) ×α Z+ imbeds
completely isometrically into a C*-crossed product C(Sb) ×α Z.
In
e(C(T) ×α Z+) = C(Sb) ×α Z and as a consequence, we obtain
fact, C∗
that (ρ, T ) dilates to a covariant representation (τ, W ), where τ is a
∗-representation of C(T) (so σ(z) is unitary) and W is a unitary.
In contrast, if α is induced by a constant Blashcke product, we can no
e(C(T) ×α Z+) up to isomorphism. In that case, α is
longer identify C∗
evaluation at a boundary point. Even though every covariant represen-
tation of (A(D), α) dilates to a covariant representation of (C(T), α),
the theory of [6] is not directly applicable since α is not injective. In-
stead, we use the process of "adding tails to C*-correspondences" [8],
e(C(T) ×α Z+) up to strong
as modified in [3, 7] and we identify C∗
Morita equivalence as a crossed product. In Theorem 2.6 we show that
e(C(T) ×α Z+) is strongly Morita equivalent to a C*-algebra of the
C∗
form C(Se) ×s Z, where e : T × N −→ T × N is a suitable map and
(Se, s) is the solenoid system for (T × N, e).
Semi-crossed products of the the disc algebra were introduced and
first studied by Buske and Peters in [1], following relevant work of
Hoover, Peters and Wogen [5]. The algebras A(D) ×α Z+, where α is
an arbitrary endomorphism, where classified up to algebraic endomor-
phism in [2]. Results associated with their C*-envelope can be found in
[1, Proposition III.13] and [10, Thoorem 2]. The results of the present
paper subsume and extend these earlier results.
SEMICROSSED PRODUCTS OF THE DISK ALGEBRA
3
2. The Disk Algebra
The C*-envelope of the disk algebra A(D) is C(T), the space of
continuous functions on the unit circle. Suppose that α is an endomor-
phism of C(T) which leaves A(D) invariant. We refer to the restriction
of α to A(D) as α as well. Then b = α(z) ∈ A(D); and has spectrum
σA(D)(b) ⊂ σA(D)(z) = D and σC(T)(b) ⊂ σC(T)(z) = T.
It follows that b is a finite Blaschke
Thus kbk = 1 and b(T) ⊂ T.
product. Therefore α(f ) = f ◦ b for all f ∈ C(T). When b is not
constant, α is completely isometric.
A (completely) contractive representation ρ of A(D) is determined
by ρ(z) = A, which must be a contraction. The converse follows from
the matrix von Neumann inequality; and shows that ρ(f ) = f (A) is a
complete contraction. A covariant representation of (A(D), α) is thus
determined by a pair of contractions (A, T ) such that AT = T b(A).
The representation of A(D) ×α Z+ is given by
ρ × T(cid:0)
n
Xi=0
tifi(cid:1) =
n
Xi=0
T ifi(A),
which extends to a completely contractive representation of the semi-
crossed product by the universal property.
A contractive representation σ of C(T) is a ∗-representation, and
is likewise determined by U = σ(z), which must be unitary; and all
unitary operators yield such a representation by the functional calculus.
A covariant representation of (C(T), α) is given by a pair (U, T ) where
U is unitary and T is a contraction satisfying UT = T b(U). To see
this, multiply on the left by U ∗ and on the right by b(U)∗ to obtain the
identity
U ∗T = T b(U)∗ = T¯b(U) = T α(¯z)(U).
The set of functions {f ∈ C(T) : f (U)T = T α(f )(U)} is easily seen to
be a norm closed algebra. Since it contains z and ¯z, it is all of C(T).
So the covariance relation holds.
Theorem 2.1. Let b be a finite Blaschke product, and let α(f ) = f ◦ b.
Then A(D) ×α Z+ is (canonically completely isometrically isomorphic
to) a subalgebra of C(T) ×α Z+.
Proof. To establish that A(D) ×α Z+ is completely isometric to a
subalgebra of C(T) ×α Z+, it suffices to show that each (A, T ) with
AT = T b(A) has a dilation to a pair (U, S) with U unitary and S a
contraction such that US = Sb(U) and PHSnU mH = T nAm for all
4
K.R.DAVIDSON AND E.G. KATSOULIS
m, n ≥ 0. This latter condition is equivalent to H being semi-invariant
for the algebra generated by U and S.
The covariance relation can be restated as
0
(cid:20)A
0 b(A)(cid:21)(cid:20)0 T
0 0(cid:21) = (cid:20)0 T
0 0(cid:21)(cid:20)A
0 b(A)(cid:21)
0
we may dilate [ 0 T
0 0 ] to a contraction of the form [ ∗ S
Dilate A to a unitary U which leaves H semi-invariant. Then (cid:2) A 0
0 b(A)(cid:3)
dilates to(cid:2) U 0
0 b(U )(cid:3). By the Sz.Nagy-Foia¸s Commutant Lifting Theorem,
with (cid:2) U 0
0 α(U )(cid:3) and has H ⊕ H as a common semi-invariant subspace.
Clearly, we may take the ∗ entries to all equal 0 without changing
things. So (U, S) satisfies the same covariance relations US = Sb(U).
Therefore we have obtained a dilation to the covariance relations for
(C(T), α).
∗ ∗ ] which commutes
Once we have a covariance relation for (C(T), α), we can try to
dilate further. Extending S to an isometry V follows a well-known
path. Observe that
b(U)S ∗S = S ∗US = S ∗Sb(U).
Thus D = (I − S ∗S)1/2 commutes with b(U). Write b(n) for the com-
position of b with itself n times, Hence we can now use the standard
Schaeffer dilation of S to an isometry V and simultaneously dilate U
to U1 as follows:
V =
S 0 0
D 0 0
I 0
0
0 I
0
...
...
...
0
0
0
0
. . .
. . .
. . .
. . .
. . .
. . .
and U1 =
0
U
0 b(U1)
0
0
...
0
0
...
0
0
b(2)(U1)
0
...
0
0
0
. . .
. . .
. . .
b(3)(U1) . . .
. . .
...
.
A simple calculation shows that U1V = V b(U1). So as above, (U, V )
satisfies the covariance relations for (C(T), α).
We would like to make V a unitary as well. This is possible in the case
where b is non-constant, but the explicit construction is not obvious.
Instead, we use the theory of C*-envelopes and maximal dilations. First
we need the following.
Lemma 2.2. Let b be a finite Blaschke product, and let α(f ) = f ◦ b.
Then
C∗
e(A(D) ×α Z+) ≃ C∗
e(C(T) ×α Z+).
SEMICROSSED PRODUCTS OF THE DISK ALGEBRA
5
Proof. The previous Theorem identifies A(D) ×α Z+ completely iso-
metrically as a subalgebra of C(T) ×α Z+. The C*-envelope C of
C(T)×αZ+ is a Cuntz-Pimsner algebra containing a copy of C(T) which
is invariant under gauge actions. Now C is a C*-cover of C(T) ×α Z+,
so it is easy to see that it is also a C*-cover of A(D) ×α Z+. Since
A(D) ×α Z+ is invariant under the same gauge actions, its Shilov ideal
J ⊆ C will be invariant by these actions as well. If J 6= 0 then by
gauge invariance J T C(T) 6= 0. Since the quotient map
A(D) −→ C(T)/(J ∩ C(T))
is completely isometric, we obtain a contradiction. Hence J = 0 and
the conclusion follows.
We now recall some of the theory of semicrossed products of C*-
algebras. When A is a C*-algebra, the completely isometric endomor-
phisms are the faithful ∗-endomorphisms. In this case, Peters shows
[9, Prop.I.8] that there is a unique C*-algebra B, a ∗-automorphism
β of B and an injection j of A into B so that β ◦ j = jα and B is
the closure of Sn≥0 β −n(j(A)). It follows [9, Prop.II.4] that A ×α Z+
is completely isometrically isomorphic to the subalgebra of the crossed
product algebra B ×β Z generated as a non-self-adjoint algebra by
an isomorphic copy j(A) of A and the unitary u implementing β in
the crossed product. Actually, Kakariadis and the second author [6,
Thm.2.5] show that B ×β Z is the C*-envelope of A ×α Z+.
In the case where A = C(X) is commutative and α is induced by an
injective self-map of X, the pair (B, β) has an alternative description.
Definition 2.3. Let X be a Hausdorff space and ϕ a surjective self-
map of X. We define the solenoid system of (X, ϕ) to be the pair
(Sϕ, s), where
Sϕ = {(xn)n≥1 : xn = ϕ(xn+1), xn ∈ X, n ≥ 1}
equipped with the relative topology inherited from the product topol-
i=1 Xi, Xi = X, i = 1, 2, . . . , and s is the backward shift on
ogy on Q∞
Sϕ.
It is easy to see that in the case where A = C(X) and α is induced
by an injective self-map ϕ of X, the pair (B, β) for (A, α) described
above, is conjugate to the solenoid system (Sϕ, s). Therefore, we obtain
Corollary 2.4. Let b be a non-constant finite Blaschke product, and
let α(f ) = f ◦ b on C(T). Then
C∗
e(A(D) ×α Z+) = C∗
e(C(Sb) ×s Z).
where (Sb, s) is the solenoid system of (T, b).
6
K.R.DAVIDSON AND E.G. KATSOULIS
It is worth restating this theorem as a dilation result.
Corollary 2.5. Let α be an endomorphism of A(D) induced by a non-
constant finite Blaschke product and let A, T ∈ B(H) be contractions
satisfying AT = T α(A). Then there exist unitary operators U and
W on a Hilbert space K ⊃ H which simultaneously dilate A and T ,
in the sense that PHW mU nH = T mAn for all m, n ≥ 0, so that
UW = W α(U).
Proof. Every covariant representation (A, T ) of (A(D), α) dilates to
a covariant representation (U1, V ) of (C(T), α). This in turn dilates
to a maximal dilation τ of C(T) ×α Z+, in the sense of Dritschel and
McCullough [4]. The maximal dilations extend to ∗-representations of
the C*-envelope. Then A is dilated to τ (j(z)) = U is unitary and T
dilates to the unitary W which implements the automorphism β on B,
and restricts to the action of α on C(T).
The situation changes when we move to non-injective endomorphisms
α of A(D). Indeed, let λ ∈ T and consider the endomorphism αλ of
A(D) induced by evaluation on λ, i.e., αλ(f )(z) = f (λ), ∀z ∈ D. (Thus
αλ is the endomorphism of A(D) corresponding to a constant Blaschke
product.) If two contractions A, T satisfy AT = T αλ(A) = λT , then
the existence of unitary operators U, W , dilating A and T respectively,
implies that A = λI.
It is easy to construct a pair A, T satisfying
AT = λT and yet A 6= λI. This shows that the analogue Corollary 2.5
Z+)
fails for α = αλ and therefore one does not expect C∗
to be isomorphic to the crossed product of a commutative C*-algebra,
at least under canonical identifications. However as we have seen, a
weakening of Corollary 2.5 is valid for α = αλ if one allows W to be
an isometry instead of a unitary operator. In addition, we can iden-
e(A(D) ×α Z+) as being strongly Morita equivalent to a crossed
tify C∗
product C*-algebra. Indeed, if
e(A(D) ×αλ
e : T × N −→ T × N
is defined as
then
e(z, n) = (cid:26) (1, 1)
(z, n − 1) otherwise,
if n = 1
Theorem 2.6. Let α = αλ be an endomorphism of A(D) induced by
e(A(D) ×α Z+) is strongly Morita
evaluation at a point λ ∈ T. Then C∗
equivalent to C(Se) ×s Z, where e : T × N −→ T × N is defined above
and (Se, s) is the solenoid system of (T × N, e).
SEMICROSSED PRODUCTS OF THE DISK ALGEBRA
7
Proof. In light of Lemma 2.2, it suffices to identify the C*-envelope
of C(T) ×α Z+. As α is no longer an injective endomorphism of C(T),
we invoke the process of adding tails to C*-correspondences [8], as
modified in [3, 7].
Indeed, [7, Example 4.3] implies that the C*-envelope of the tensor
algebra associated with the dynamical system (C(T), α) is strongly
Morita equivalent to the Cuntz-Pimsner algebra associated with the
injective dynamical system (T × N, e) defined above. Therefore by
invoking the solenoid system of (T × N, e), the conclusion follows from
the discussion following Lemma 2.2.
References
[1] D. Buske and J. Peters, Semicrossed products of the disk algebra: contractive
representations and maximal ideals, Pacific J. Math. 185 (1998), 97 -- 113.
[2] K. Davidson, E. Katsoulis, Isomorphisms between topological conjugacy alge-
bras, J. reine angew. Math. 621 (2008), 29 -- 51.
[3] K. Davidson and J. Roydor, C*-envelopes of of tensor algebras for multivariable
dynamics, Proc. Edinb. Math. J. 53 (2010), 333-351.
[4] M. Dritschel and S. McCullough, Boundary representations for families of
representations of operator algebras and spaces, J. Operator Theory 53 (2005),
159 -- 167.
[5] T. Hoover, J. Peters and W. Wogen, Spectral properties of semicrossed prod-
ucts, Houston J. Math. 19 (1993), 649 -- 660.
[6] E. Kakariadis and E. Katsoulis, Semicrossed products of operator algebras and
their C*-envelopes, manuscript.
[7] E. Kakariadis and E. Katsoulis, Contributions to the theory of C*-
correspondences with applications to multivariable dynamics, manuscript.
[8] P. S. Muhly, M. Tomforde, Adding tails to C*-correspondences, Doc. Math. 9
(2004), 79 -- 106.
[9] J. Peters, Semicrossed products of C*-algebras, J. Funct. Anal. 59 (1984), 498 --
534.
[10] S. Power, Completely contractive representations for some doubly generated
antisymmetric operator algebras, Proc. Amer. Math. Soc. 126 (1998), 2355 --
2359.
Pure Mathematics Department, University of Waterloo, Waterloo,
ON N2L -- 3G1, CANADA
E-mail address: [email protected]
Department of Mathematics, University of Athens, 15784 Athens,
GREECE
Alternate address: Department of Mathematics, East Carolina Uni-
versity, Greenville, NC 27858, USA
E-mail address: [email protected]
|
1009.4778 | 3 | 1009 | 2015-05-26T20:15:27 | Classifying $C^*$-algebras with both finite and infinite subquotients | [
"math.OA"
] | We give a classification result for a certain class of $C^{*}$-algebras $\mathfrak{A}$ over a finite topological space $X$ in which there exists an open set $U$ of $X$ such that $U$ separates the finite and infinite subquotients of $\mathfrak{A}$. We will apply our results to $C^{*}$-algebras arising from graphs. | math.OA | math |
CLASSIFYING C ∗-ALGEBRAS WITH BOTH FINITE AND
INFINITE SUBQUOTIENTS
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
Abstract. We give a classification result for a certain class of C ∗-algebras A
over a finite topological space X in which there exists an open set U of X such
that U separates the finite and infinite subquotients of A. We apply our results
to C ∗-algebras arising from graphs.
1. Introduction
Just like a finite group, any C ∗-algebra A with finitely many ideals has a decom-
position series
0 = I0 ⊳ I1 ⊳ · · · ⊳ In = A
in such a way that all subquotients are simple. As in the group case, the simple
subquotients are unique up to permutation of isomorphism classes, but far from
determine the isomorphism class of A.
If we further assert that all simple subquotients are classifiable by algebraic in-
variants such as K-theory we are naturally lead to the pertinent question of which
algebraic invariants, if any, classify all of A. This question has previously been stud-
ied, leading to a complete solution when all subquotients are AF (cf. [19]), and a
partial solution when all are purely infinite (cf. [26], [6], [30]), but in the case when
some are of one type and some of another, only sporadic results have been found. It
is the purpose of this paper to provide a general framework in which classification
of C ∗-algebras can be proved for a large class of C ∗-algebras of mixed type.
We are able to do so by combining several recent important developments in
classification theory, notably
• Kirchberg's isomorphism result [20]
• The corona factorization property [21]
• The universal coefficient theorem of Meyer and Nest [26]
with refinements of our previous work [16] inspired by an idea of Rørdam [32].
As useful as these problems may be to test the borders of our understanding
of classification we are driven to this project by one class of examples. A graph
C ∗-algebra has the property that all simple subquotients are either AF or purely
infinite, and examples of mixed type occur even for very small graphs. For instance,
consider the graphs {En}n∈N given below
Date: May 28, 2018.
2000 Mathematics Subject Classification. Primary: 46L35, 37B10 Secondary: 46M15, 46M18.
Key words and phrases. Classification, Extensions, Graph algebras.
1
2
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
0 0 0
n 3 0
1 1 3
•
?⑦
•
✸S
⑦
n
⑦
⑦
✸S
•
For any n > 0, C ∗(En) decomposes into a linear lattice with simple subquotients
K, O3 ⊗ K, O3.
The results presented in this paper show that C ∗(En) ⊗ K ≃ C ∗(En+4) ⊗ K and that
there are three stable isomorphism classes
[C ∗(E1)] = [C ∗(E3)], [C ∗(E2)], [C ∗(E4)]
As in our previous paper [16] the technical focal point in this work is the general
question of when one can deduce from the fact that A and B in the extension
0
/ B
/ E
/ A
/ 0
are classifiable by K-theory, that the same is true for E. We shall fix a finite (not
necessarily Hausdorff) T0 topological space X with a non-trivial open subset U ⊆ X
and require that E is a C ∗-algebra over X -- associating ideals in E with open subsets
of X -- in such a way that B is the C ∗-algebra corresponding to U . Assuming then
that A and B are KK-strongly classifiable by their filtered and ordered K-theories
over X\U and U , respectively, we supply conditions on the extension securing that
also E is classifiable by filtered and ordered K-theory. Our key technical result to
this effect, Theorem 3.3 below, provides stable isomorphism in this context under,
among other things, the provision of fullness of the extension and KK-liftability of
morphisms of filtered K-theory. KK-liftability follows in many cases by the UCT
of Meyer and Nest [26] as generalized by Bentmann and Kohler [6], and we develop
in Section 5 several useful tools to establish fullness.
Although we are confident that Theorem 3.3 will apply in other settings as well,
we restrict ourselves to demonstrating how the results lead to classification of certain
graph C ∗-algebras up to stable isomorphism, generalizing results by the first named
author and Tomforde in [18]. As a consequence of the results in [18], all graph C ∗-
algebras with exactly one non-trivial ideal are classifiable up to stable isomorphism
by the six-term exact sequence in K-theory together with the positive cone of the
ideal, algebra, and quotient, irrespective of the types of the simple subquotients.
Indicating in a (Hasse) diagram of the ideal lattice a simple AF subquotient with
a straight line, and a purely infinite subquotient with a curly line (the zero ideal
indicated by "◦"), the results in [18] solved all the cases
•
•
◦
•
•
◦
•
•
◦
•
•
◦
?
a
O
O
o
o
a
/
/
/
/
O
O
O
O
O
O
O
O
CLASSIFYING C∗-ALGEBRAS
3
Further, for graph C ∗-algebras with a maximal proper ideal which is AF, indicated
diagrammatically as
•
•
❄❄❄❄❄❄
⑧⑧⑧⑧⑧⑧
are classifiable up to stable isomorphism by the six-term exact sequence in K-theory
together with the positive cone of the ideal, algebra, and quotient. In this paper we
generalize to the settings
⑧⑧⑧⑧⑧
❄❄❄❄❄
•
...
◦
•
...
•
⑧⑧⑧⑧⑧⑧
❄❄❄❄❄❄
where the ideal lattice at one end (top or bottom) has a finite linear lattice of purely
infinite C ∗-algebras, and at the other has an AF algebra. In particular, we cover
all graph C ∗-algebras of finite linear lattices which has no more than one transition
from finite to infinite subquotient, such as C ∗(En) defined above. We speculate that
this condition is not necessary (although it certainly is for our approach) and in [17]
give partial evidence for this, under extra conditions on the algebraic nature of the
involved K-theory.
2. Notation and Conventions
We first start with some definitions and conventions that will be used throughout
the paper.
2.1. C ∗-algebras over topological spaces. Let X be a topological space and let
O(X) be the set of open subsets of X, partially ordered by set inclusion ⊆. A subset
Y of X is called locally closed if Y = U \ V where U, V ∈ O(X) and V ⊆ U . The set
of all locally closed subsets of X will be denoted by LC(X). The set of all connected,
non-empty locally closed subsets of X will be denoted by LC(X)∗.
The partially ordered set (O(X), ⊆) is a complete lattice, that is, any subset S
of O(X) has both an infimum V S and a supremum W S. More precisely, for any
subset S of O(X),
For a C ∗-algebra A, let I(A) be the set of closed ideals of A, partially ordered by
⊆. The partially ordered set (I(A), ⊆) is a complete lattice. More precisely, for any
U = \U ∈S
^U ∈S
U!◦
and _U ∈S
U = [U ∈S
U
O
O
O
O
O
O
O
O
O
O
4
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
subset S of I(A),
^I∈S
I = \I∈S
I and _I∈S
I =XI∈S
I
Definition 2.1. Let A be a C ∗-algebra. Let Prim(A) denote the primitive ideal
space of A, equipped with the usual hull-kernel topology, also called the Jacobson
topology.
Let X be a topological space. A C ∗-algebra over X is a pair (A, ψ) consisting
of a C ∗-algebra A and a continuous map ψ : Prim(A) → X. A C ∗-algebra over X,
(A, ψ), is separable if A is a separable C ∗-algebra. We say that (A, ψ) is tight if ψ
is a homeomorphism.
We always identify O(Prim(A)) and I(A) using the lattice isomorphism
U 7→ \p∈Prim(A)\U
p
Let (A, ψ) be a C ∗-algebra over X. Then we get a map ψ∗ : O(X) → O(Prim(A)) ∼=
I(A) defined by
U 7→ {p ∈ Prim(A) : ψ(p) ∈ U } = A(U )
For Y = U \ V ∈ LC(X), set A(Y ) = A(U )/A(V ). By Lemma 2.15 of [25], A(Y )
does not depend on U and V .
In this paper, we will be mainly interested in the following examples:
Example 2.2. For any C ∗-algebra A, the pair (A, idPrim(A)) is a tight C ∗-algebra
over Prim(A). For each U ∈ O(Prim(A)), the ideal A(U ) equals Tp∈Prim(A)\U p.
Example 2.3. Let Xn = {1, 2, . . . , n} partially ordered with ≤. Equip Xn with the
Alexandrov topology, so the non-empty open subsets are
[a, n] = {x ∈ X : a ≤ x ≤ n}
for all a ∈ Xn; the non-empty closed subsets are [1, b] with b ∈ Xn, and the non-
empty locally closed subsets are those of the form [a, b] with a, b ∈ Xn and a ≤ b.
Let (A, φ) be a C ∗-algebra over Xn. We will use the following notation throughout
the paper:
A[k] = A({k}), A[a, b] = A([a, b]), and A(i, j] = A[i + 1, j].
Using the above notation we have ideals A[a, n] such that
{0} ✂ A[n] ✂ A[n − 1, n] ✂ · · · ✂ A[2, n] ✂ A[1, n] = A
Definition 2.4. Let A and B be C ∗-algebras over X. A homomorphism φ : A → B
is X-equivariant if φ(A(U )) ⊆ B(U ) for all U ∈ O(X). Hence, for every Y = U \ V ,
φ induces a homomorphism φY : A(Y ) → B(Y ). Let C∗-alg(X) be the category
whose objects are C ∗-algebras over X and whose morphisms are X-equivariant ho-
momorphisms.
Remark 2.5. Suppose A and B are tight C ∗-algebras over Xn. Then it is clear
that a ∗-homomorphism φ : A → B is an isomorphism if and only if φ is an Xn-
equivariant isomorphism. We will use this fact in Theorem 6.8.
CLASSIFYING C∗-ALGEBRAS
5
Remark 2.6. Let ei : 0 → Bi → Ei → Ai → 0 be an extension for i = 1, 2.
Note that Ei can be considered as a C ∗-algebra over X2 = {1, 2} by sending ∅ to
the zero ideal, {2} to the image of Bi in Ei, and {1, 2} to Ei. Hence, there exists
a one-to-one correspondence between X2-equivariant homomorphisms φ : E1 → E2
and homomorphisms from e1 and e2.
2.2. Filtered ordered K-theory.
Definition 2.7. Let X be a T0 topological space and let A be a C ∗-algebra over
X. For open subsets U1, U2, U3 of X with U1 ⊆ U2 ⊆ U3, set Y1 = U2 \ U1, Y2 =
U3 \ U1, Y3 = U3 \ U2 ∈ LC(X). Then we have a six term exact sequence
K0(A(Y1))
ι∗
/ K0(A(Y2))
π∗
/ K0(A(Y3))
∂∗
∂∗
K1(A(Y3))
π∗
K1(A(Y2))
ι∗
K1(A(Y1))
The filtered K-theory FKX(A) of A is the collection of all K-groups thus occurring
and the natural transformations {ι∗, π∗, ∂∗}. The filtered, ordered K-theory FK+
X(A)
of A is FKX(A) of A together with K0(A(Y ))+ for all Y ∈ LC(X).
Let A and B be C ∗-algebras over X. An isomorphism α : FKX(A) → FKX(B)
is a collection of group isomorphisms
αY,∗ : K∗(A(Y )) → K∗(B(Y ))
for each Y ∈ LC(X) preserving all natural transformations. An isomorphism α :
FK+
X(B) is an isomorphism α : FKX(A) → FKX(B) which satisfies
that αY,0 is an order isomorphism for all Y ∈ LC(X).
X (A) → FK+
If Y ∈ LC(X) such that Y = Y1 ⊔ Y2 with two disjoint relatively open subsets
Y1, Y2 ∈ O(Y ) ⊆ LC(X), then A(Y ) ∼= A(Y1) × A(Y2) for any C ∗-algebra over
X. Moreover, there is a natural isomorphism K∗(A(Y )) to K∗(A(Y1)) ⊕ K∗(A(Y2))
which is a positive isomorphism from K0(A(Y )) to K0(A(Y1)) ⊕ K0(A(Y2)). If X
is finite, any locally closed subset is a disjoint union of its connected components.
Therefore, we lose no information when we replace LC(X) by the subset LC(X)∗.
This observation reduces computation and will be used in Section 7.
3. The KK (X2; −, −) functor
Let a be an element of a C ∗-algebra A. We say that a is norm-full in A if a is
not contained in any norm-closed proper ideal of A. The word "full" is also widely
used, but since we will often work in multiplier algebras, we emphasize that it is
the norm topology we are using, rather than the strict topology. We say that a
sub-C ∗-algebra B of a C ∗-algebra A is norm-full if the norm-closed ideal generated
by B is A.
Definition 3.1. An extension e is said to be full if the associated Busby invariant
τe has the property that τe(a) is a norm-full element of Q(B) = M(B)/B for every
a ∈ A\{0}.
/
/
O
O
o
o
o
o
6
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
We now define some functors that will be used throughout the rest of the paper.
Let X and Y be T0 topological spaces. For every continuous function f : X → Y
we have a functor
(1) Define g1
f : C∗-alg(X) → C∗-alg(Y ),
X : X → X1 by g1
X (x) = 1. Then g1
(A, ψ) 7→ (A, f ◦ ψ)
X is continuous. Note that the
induced functor g1
X : C∗-alg(X) → C∗-alg(X1) is the forgetful functor.
(2) Let U be an open subset of X. Define g2
U,X : X → X2 by g2
U,X(x) = 1 if
U,X is continuous. Thus the induced
x /∈ U and g2
functor
U,X(x) = 2 if x ∈ U . Then g2
g2
U,X : C∗-alg(X) → C∗-alg(X2)
is just specifying the extension 0 → A(U ) → A → A/A(U ) → 0.
(3) We can generalize (2) to finitely many ideals. Let U1 ⊆ U2 ⊆ · · · ⊆ Un = X
be open subsets of X. Define gn
U1,U2,...,Un,X(x) =
n − k + 1 if x ∈ Uk \ Uk−1. Then gn
U1,U2,...,Un,X is continuous. Therefore,
any C ∗-algebra with ideals 0 ✂ I1 ✂ I2 ✂ · · · In = A can be made into a
C ∗-algebra over Xn.
U1,U2,...,Un,X : X → Xn by gn
(4) For all Y ∈ LC(X), rY
X : C∗-alg(X) → C∗-alg(Y ) is the restriction functor
defined in Definition 2.19 of [25]
Let KK(X) be the category whose objects are separable C ∗-algebras over X and the
set of morphisms is KK (X; A, B). By Proposition 3.4 of [25], these functors induce
functors from KK(X) to KK(Z), where Z = Y, X1, Xn.
Lemma 3.2. Let U be an open set of X and Y = X \ U . Then
r{2}
X2 ◦ g2
U,X = g1
U ◦ rU
X
and
r{1}
X2 ◦ g2
U,X = g1
Y ◦ rY
X
from C∗-alg(X) to C∗-alg(X1). Consequently, the induced functors from KK(X) to
KK(X1) will be equal.
Proof. Let A be a C ∗-algebra over X. Then
r{2}
X2 ◦ g2
U,X(A) = A(U )
and
g1
U ◦ rU
X(A) = g1
U (A(U )) = A(U ).
Suppose B is a C ∗-algebra over X and φ : A → B is an X-equivariant homomor-
phism. Then
and
r{2}
X2 ◦ g2
U,X(φ) = φU
Therefore, r{2}
X2 ◦ g2
U,X = g1
X(φ) = g1
U (φU ) = φU .
g1
U ◦ rU
U ◦ rU
X.
Similar computation shows that r{1}
X2 ◦ g2
U,X = g1
Y ◦ rY
X .
(cid:3)
The following theorem is a generalization of Theorem 2.3 of [16]. This is the key
technical theorem of the paper.
CLASSIFYING C∗-ALGEBRAS
7
Theorem 3.3. Let A1 and A2 be separable, nuclear C ∗-algebras over X2. Suppose
If
ei
x ∈ KK (X2; A1, A2), then
0 → Ai[2] → Ai → Ai[1] → 0 is an essential extension for i = 1, 2.
:
X2 (x) × [τe2] = [τe1] × r{2}
r{1}
X2 (x)
in KK 1(A1[1], A2[2]).
Proof. Let C∗-algnuc(X2) be the category whose objects are nuclear, separable C ∗-
algebras over X2 and morphisms are X2-equivariant homomorphisms. Let KKnuc(X2)
be the category whose objects are nuclear, separable C ∗-algebras over X2 and the
morphisms from A to B are the elements of KK (X2; A, B). Let KKnuc be the cat-
egory whose objects are nuclear, separable C ∗-algebras and the morphisms from A
to B are the elements of KK (A, B).
Let A be in C∗-algnuc(X2) and let πA be the natural projection from A to A[1].
Let iA : SA → CπA and jA : A[2] → CπA be the natural embeddings, where
CπA = {(a, f ) ∈ A ⊕ C0((0, 1], A[1]) : πA(a) = f (1)}
is the mapping cone of πA. Recall that KK (jA) is invertible in KK (A[2], CπA ). Then
the isomorphism from KK 1(A[1], B[2]) to KK (SA[1], B[2]) sends the class induced
by
0 → A[2] → A → A[1] → 0
to KK (iA)× KK (jA)−1 (see [13] and Theorem 19.5.7 of [7]). Using this isomorphism
from KK 1(A[1], B[2]) to KK (SA[1], B[2]), the equation
r{1}
X2 (x) × [τe2] = [τe1] × r{2}
X2 (x)
in KK 1(A1[1], A2[2]) becomes
Sr{1}
X2 (x) × KK (iA2) × KK (jA2)−1 = KK (iA1) × KK (jA1)−1 × r{2}
X2 (x)
in KK (SA[1], B[2]), where S is the natural isomorphism from KK (A, B) to KK (SA, SB).
Let m = 1 or 2. Define Fm : C∗-algnuc → KKnuc by
Fm(A) =(A[2], m = 2
SA[1], m = 1
and Fm(φ) =(φ{2}, m = 2
Sφ{1}, m = 1
We claim that η : F1 → F2 given by ηA = KK (iA) × KK (jA)−1 is a natural trans-
formation between the functors F1 and F2. Let A and B be in C∗-algnuc(X2) and
let φ be an X2-equivariant homomorphism. By the definition of the mapping cone
sequence, there exists a homomorphism ψ : CπA → CπB such that the diagrams
0
0
iA
/ SA[1]
Sφ{1}
CπA
ψ
/ SB[1]
/ CπB
iB
A
φ
/ B
0
/ 0
/
/
/
/
/
/
/
/
/
/
/
8
and
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
0
0
jA
/ A[2]
φ{2}
CπA
ψ
CA
0
Cφ
/ B[1]
/ CπB
jB
/ CB
/ 0
are commutative. Thus,
F1(φ) × KK (iB) × KK (jB)−1 = KK (Sφ{1}) × KK (iB) × KK (jB)−1
= KK (iA) × KK (ψ) × KK (jB)−1
= KK (iA) × KK (jA)−1 × KK (φ{2}).
= KK (iA) × KK (jA)−1 × F2(φ).
Hence, η : F1 → F2 is a natural transformation.
Since F1, F2 are stable, split exact, and homotopy invariant functors, by the
universal property of KK , we have that Fm induces a functor F m : KKnuc(X2) →
KKnuc and η induces a natural transformation η : F 1 → F 2. In particular, for each
x ∈ KK (X2; A1, A2), we have that
Sr{1}
X2 (x) × KK (iA2) × KK (jA2 )−1 = KK (iA1) × KK (jA1)−1 × r{2}
X2 (x).
By the comments made in the second paragraph of the proof, we have that
X2 (x) × [τe2] = [τe1] × r{2}
r{1}
X2 (x)
in KK 1(A1[1], A2[2]).
(cid:3)
4. Classification
In this section we prove general classification results for several classes of extension
algebras. The result of this section is a generalization of the classification results
obtained by the authors in [16].
Definition 4.1. For a T0 topological space X, we will consider classes CX of sepa-
rable, nuclear C ∗-algebras in the bootstrap category of Rosenberg and Schochet N
such that
(1) any element in CX is a C ∗-algebra over X;
(2) if A and B are in CX and there exists an invertible element α in KK (X; A, B)
X(B), then there exists
which induces an isomorphism from FK+
an isomorphism φ : A → B such that KK (φ) = g1
X(A) to FK+
X (α).
We concentrate on the following two classes satisfying (1) -- (3) above:
Example 4.2. Let A and B be separable, nuclear, stable, O∞-absorbing tight C ∗-
algebras over X. Let α be an invertible element in KK (X; A, B). By Kirchberg
[20], there exists an isomorphism φ : A → B such that KK (X; φ) = α. Hence,
KK (φ) = g1
X (KK (X; φ)) = KK (α). Thus, if CX is the class of all stable, separable,
nuclear C ∗-algebras over X which are O∞-absorbing in N , then CX satisfies the
properties of Definition 4.1.
/
/
/
/
/
/
/
/
/
/
/
CLASSIFYING C∗-ALGEBRAS
9
Example 4.3. Let A and B be stable AF algebras. Let α be an invertible element
in KK (X1; A, B) = KK (A, B) which induces an isomorphism from FK+
X1(A) =
(K0(A), K0(A)+) to FK+
X1(B) = (K0(B), K0(B)+). Then by the classification of
AF algebras [19], there exists an isomorphism φ : A → B such that K0(φ) = K0(α).
Since KK (A, B) ∼= Hom(K0(A), K0(B)), we have that KK (φ) = α. Thus, if CX1 is
the class of all stable, AF algebras, then CX1 satisfies the properties of Definition
4.1.
Remark 4.4. The condition
(2') if A and B are in CX and there exists an isomorphism β from FK+
X(A) to
FK+
X(B), then there exists an isomorphism φ : A → B such that φ∗ = β.
is more closely suited to our purposes, and (1),(2') is true in general in Example
4.3, but not always in Example 4.2, as pointed out in [26]. In fact, there exists a
space X with four points such that (3') fails in Example 4.2.
Lemma 4.5. For i = 1, 2, let ei : 0 → Ii → Ei → Ai → 0 be non-unital full
extensions. Suppose Ii is a stable C ∗-algebra satisfying the corona factorization
If there exist an isomorphism φ0 : I1 → I2 and an isomorphism φ2 :
property.
A1 → A2 such that KK (φ2) × [τe2] = [τe1] × KK (φ0), then E1 is isomorphic to E2.1
Proof. Note that e1 ∼= e1·φ0 and e2 ∼= φ2·e2, where e1·φ0 is the push-out of e1 along φ0
and φ2 · e2 is the pull-back of e2 along φ2 (cf. [32]). Since [τe1·φ0] = [τe1] × KK (φ0) =
KK (φ2) × [τe2] = [τφ2·e2] in KK 1(A1, I2), we have that [τe1·φ0] = [τφ2·e2]. Since τe1·φ0
and τφ2·e2 are non-unital full extensions and I2 satisfies the corona factorization
property, by Theorem 3.2(2) of [27], there exists a unitary u in M(I2) such that
Ad(π(u)) ◦ τe1·φ0 = τφ2·e2. Hence, (Ad(u), Ad(u), idA1) is an isomorphism between
e1 · φ0 and φ2 · e2. Thus, E1 is isomorphic to E2.
(cid:3)
We will apply the theorem below to a certain class of C ∗-algebras arising from
graphs. See Proposition 6.3, Corollary 6.4, Proposition 6.5, and Theorem 6.8.
Theorem 4.6. Let X be a finite topological space and let U ∈ O(X). Set Y =
X \ U ∈ LC(X). For i = 1, 2, let Ai be a C ∗-algebra over X such that Ai is a stable,
separable, nuclear C ∗-algebra and every simple sub-quotient of Ai is in the bootstrap
category N .
Let CI,U and CQ,Y be classes of C ∗-algebras that satisfy the properties of Defi-
nition 4.1. Suppose Ai(U ) is a stable C ∗-algebra in CI,U and satisfying the corona
factorization property, and Ai(Y ) is a stable C ∗-algebra in CQ,Y . Suppose for i = 1, 2
ei : 0 → Ai(U ) → Ai → Ai(Y ) → 0
are full extensions. If there exists an isomorphism α : FKX (A1) → FKX (A2) such
that αU : FK+
Y (A2(Y )) are
isomorphisms, and α lifts to an invertible element in KK (X; A1, A2), then A1 ∼= A2.
U (A2(U )) and αY : FK+
Y (A1(Y )) → FK+
U (A1(U )) → FK+
1This lemma is not correct as stated. See arXiv:1505.05951 for the corrected version of Lemma 4.5.
As noted in arXiv:1505.05951, the main results of this paper are true verbatim.
10
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
U (A1(U )) → FK+
U (A2(U )) and αY : FK+
Proof. Suppose there exists an isomorphism α : FKX(A1) → FKX(A2) such that
αU : FK+
Y (A2(Y )) are
isomorphisms, and α lifts to an invertible element in KK (X; A1, A2). Let x ∈
KK (X; A1, A2) be this lifting. Then rU
and rY
X (x) is an invertible element in KK (Y ; A1(Y ), A2(Y )). Since A1(U ) and
A2(U ) are in CI,U and A1(Y ) and A2(Y ) are in CQ,Y , there exists an isomor-
phism φ0 : A1(U ) → A2(U ) which induces rU
X(x) and there exists an isomorphism
φ2 : A1(Y ) → A2(Y ) which induces rY
X(x). By Theorem 3.3 and by Lemma 3.2
Y (A1(Y )) → FK+
X(x) is an invertible element in KK (U ; A1(U ), A2(U ))
KK (φ2) × [τe2] = (g1
Y ◦ rY
X(x)) × [τe2] = (r{1}
X2 ◦ g2
U,X(x)) × [τe2]
= [τe1] × (r{2}
X2 ◦ g2
U,X )(x) = [τe1] × (g1
U ◦ rU
X(x)) = [τe1] × KK (φ0)
in KK 1(A1(Y ), A2(U )). Since e1 and e2 are full non-unital extensions and Ai(U )
has the corona factorization property, by Lemma 4.5 we have that A1 ∼= A2.
(cid:3)
5. Full Extensions
In this section, we prove that certain extensions arising from graph C ∗-algebras
are necessarily full, allowing one to use the results in Section 4.
Let I be an ideal of a C ∗-algebra B. Set
M(B; I) = {x ∈ M(B) : xB ⊆ I}
It is easy to check that M(B; I) is a (norm-closed, two-sided) ideal of M(B).
Definition 5.1. Let {fn} be an approximate identity consisting of projections for
K, where f0 = 0 and fn − fn−1 is a projection of dimension one. Let A be a unital
C ∗-algebra and set en = 1A ⊗ fn. Note that {en} is an approximate identity of A ⊗ K
consisting of projections.
As in [31], we call an element X ∈ M(A⊗ K) diagonal with respect to {en} if there
exists a strictly increasing sequence {α(n)} of integers with α(0) = 0 such that
X(eα(n) − eα(n−1)) − (eα(n) − eα(n−1))X = 0
for all n ∈ N. We write X = diag(x1, x2, . . . ), where
xn = X(eα(n) − eα(n−1))
Conversely, if {xn} is a bounded sequence with xn ∈ Mkn(A), then upon identi-
fying Mkn(A) with
(eα(n) − eα(n−1))(A ⊗ K)(eα(n) − eα(n−1))
for an appropriate α(n), we have that X = diag(x1, x2, . . . ) for some X ∈ M(A⊗ K).
Let ǫ > 0. Define fǫ : R+ → R+ by
0,
ǫ−1(t − ǫ),
1,
if t ≤ ǫ
if ǫ ≤ t ≤ 2ǫ
if t ≥ 2ǫ
fǫ(t) =
CLASSIFYING C∗-ALGEBRAS
11
Theorem 5.2. Let B0 be a unital, O∞-absorbing C ∗-algebra. Let I be the largest
proper non-trivial ideal of B = B0 ⊗ K. If x ∈ M(B) such that x is not an element
of M(B; I) + B, then I(x) + B = M(B), where I(x) is the norm-closed ideal of
M(B) generated by x.
Consequently, every nonzero element x ∈ Q(B) that is not an element of M(B; I)/I
is norm-full in Q(B).
Proof. First note that by the proof of Theorem 3.2 of [31], M(B; I) is a proper ideal
of M(B). Let x ∈ M(B) \ (M(B; I) + B). By Proposition 2.8 (i) of [31], we may
assume that x = diag(x1, x2, . . . ) with respect to {en}, i.e. there exists a strictly
k=1 xk, where
xk ∈ (eα(k) −eα(k−1))B(eα(k) −eα(k−1)) and the sum converges in the strict topology.
Let m ∈ N. Since x is not an element of M(B; I) and x is not an element
k=m xk is not an element of (eα(m′) −
k=m fδ(xk)
k=m fδ(xk)
is norm-full in (eα(m′) − eα(m−1))B(eα(m′ ) − eα(m−1)). By Proposition 2.2 of [24],
there exists z ∈ (eα(m′) − eα(m−1))B(eα(m′ ) − eα(m−1)) such that
increasing sequence of integers {α(n)} with α(0) = 0 such that x =P∞
of B, there exists m′ ≥ m such that Pm′
eα(m−1))I(eα(m′ ) − eα(m−1)). Hence, there exists δ > 0 such that Pm′
is not an element of (eα(m′) − eα(m−1))I(eα(m′ ) − eα(m−1)). Therefore, Pm′
Therefore,
fδ(xk)! z∗
eα(m′) − eα(m−1) = z m′
Xk=m
1B0 ≤ eα(m′) − eα(m−1) = z m′
Xk=m
fδ(xk)! z∗
(cid:3)
By Corollary 2.7 of [31], I(x) = M(B).
Corollary 5.3. Let B0 be a unital, O∞-absorbing C ∗-algebra. Let I be the largest
non-trivial proper ideal of B = B0 ⊗ K. Suppose B is an ideal of A such that
e′ : 0 → B/I → A/I → A/B → 0 is an essential extension. Then the extension
e : 0 → B → A → A/B → 0 is a full extension.
Proof. Note that the canonical projection from A to A/I is an X2-equivariant ho-
momorphism. Therefore, by Theorem 2.2 of [15], the diagram
A/B
τe
Q(B)
$■■■■■■■■■
τe′
Q(B/I)
is commutative.
We will first show that e is an essential extension. Suppose D is a nonzero ideal
of A. Note that
((I + D)/I) ∩ B/I = 0 ⇐⇒ (I + D)/I = 0
and the second equality occurs exactly when D ⊆ I. Suppose that D ⊆ I. Then
it is clear that D ∩ B = D 6= 0. Suppose D is not a subset of I. By the above
/
/
$
12
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
equivalence, ((I + D)/I) ∩ B/I 6= 0. Hence, there exists x ∈ D such that x ∈ B \ I.
Therefore, x ∈ D ∩ B and x 6= 0. Hence, e is an essential extension.
We now prove that e is a full extension. Note that since B/I is a stable purely
infinite simple C ∗-algebra and e′ is an essential extension, we have that e′ is a full
extension. Let a ∈ A/B be a nonzero element. Then the ideal generated by τe′(a)
in Q(B/I) is Q(B/I).
Since
0 → M(B; I)/I → Q(B) → Q(B/I) → 0
is an exact sequence, by the above commutative diagram, τe(a) is not an element of
M(B; I)/I. Hence, by Theorem 5.2, τe(a) is norm-full in Q(B).
(cid:3)
Proposition 5.4. Let A be a C ∗-algebra and let I and D be ideals of A with I ⊆ D.
Suppose D/I is an essential ideal of A/I. Then e1 : 0 → I → A → A/I → 0 is a
full extension if and only if e2 : 0 → I → D → D/I → 0 is a full extension.
Proof. Note that the natural embedding ιD : D → A is an X2-equivariant homo-
morphism with ιI = idI. Hence, the following diagram is commutative
Therefore, the diagram
0
0
/ I
/ D
D/I
ιD/I
/ I
/ A
/ A/I
/ 0
/ 0
τe2
D/I
Q(I)
ιD/I
A/I τe1
/ Q(I)
is commutative.
Suppose e1 is a full extension. Then it is clear from the above diagram that
e2 is a full extension. Suppose that e2 is a full extension. Let a ∈ A/I be a
non-zero element. Let I be the ideal generated by a in A/I. Since D/I is an
essential ideal of A/I, there exists a non-zero b ∈ D/I such that ιD/I(b) ∈ I. Hence,
τe2(b) = (τe1 ◦ ιD/I)(b) is norm-full in Q(I). Since (τe1 ◦ ιD/I)(b) is in the ideal
generated by τe1(a) in Q(I), we have that τe1(a) is norm-full in Q(I). Thus, e1 is a
full extension.
(cid:3)
Proposition 5.5. Let A be a graph C ∗-algebra satisfying Condition (K). Suppose
I1 ✂ I2 ✂ A such that I1 is an AF algebra, I1 is the largest proper ideal of I2, I2/I1
: 0 → I1 ⊗ K → I2 ⊗ K → I2/I1 ⊗ K → 0 is a full
is purely infinite. Then e
extension.
Proof. By [14], A ⊗ K ∼= C ∗(E) ⊗ K, where E is a graph satisfying Condition (K) and
has no breaking vertices. Hence, by Theorem 3.6 of [4] and Proposition 3.4 of [4],
I2 ⊗ K is isomorphic to a C ∗(E1) ⊗ K where E1 has no breaking vertices and satisfies
Condition (K). Note that C ∗(E1) has a largest proper ideal D1, that C ∗(E1)/D1 is
purely infinite, D1 is an AF algebra, and D1 ⊗ K ∼= I1 ⊗ K. Thus, by Proposition 3.10
/
/
/
/
/
/
/
/
/
/
/
/
CLASSIFYING C∗-ALGEBRAS
13
of [18], there exists a projection p ∈ C ∗(E1) such that pC ∗(E1)p ⊗ K ∼= C ∗(E1) ⊗ K
and pD1p is stable. Since pC ∗(E1)p/pD1p is unital simple C ∗-algebra and
0 → pD1p → pC ∗(E1)p → pC ∗(E1)p/pD1p → 0
is a unital essential extension, the extension is full, cf. Lemma 1.5 of [16]. Since
pD1p is stable, Proposition 1.6 of [16] implies that the extension
0 → pD1p ⊗ K → pC ∗(E1)p ⊗ K → (pC ∗(E1)p/pD1p) ⊗ K → 0
is full. The proposition now follows since the isomorphism between pC ∗(E1)p ⊗ K
and C ∗(E1)p⊗K maps the ideal pD1p⊗K onto the ideal D1 ⊗K by Brown's Theorem,
cf. [9].
(cid:3)
Corollary 5.6. Let A be a graph C ∗-algebra satisfying Condition (K). Suppose that
I is an AF algebra such that I is an ideal of A, for all ideals J of A we have that J ⊆ I
or I ⊆ J, and A/I is O∞-absorbing. Then e : 0 → I ⊗ K → A ⊗ K → A/I ⊗ K → 0
is a full extension.
Proof. Let {Cn : n ∈ N} be the set of all minimal ideals of A/I and let An be an
ideal of A such that I ⊆ An and An/I = Cn.
Let J be an ideal of An. Then J is an ideal of A. Hence, J ⊆ I or I ⊆ J. Suppose
I ⊆ J but I 6= J. Then, J/I = An/I = Cn since Cn is simple. Hence, I is the largest
proper ideal of An, I is an AF algebra, and An/I is purely infinite. Therefore, by
Proposition 5.5, 0 → I ⊗ K → An ⊗ K → Cn ⊗ K → 0 is a full extension.
Let D =P∞
n=1 An. Then D is an ideal of A such that D/I is an essential ideal of
A/I. Since Ci ∩Cj = {0} for i 6= j, we have that 0 → I⊗K → D⊗K → D/I ⊗K → 0.
is a full extension. The corollary now follows from Proposition 5.4.
(cid:3)
6. Applications to graph C ∗-algebras
Recall our definition of Xn from Example 2.3 above. We now apply the results
of Section 4 and Section 5 to classify a certain class of graph C ∗-algebras that are
tight C ∗-algebras over Xn.
Proposition 6.1. Suppose A is a C ∗-algebra with finitely many ideals. If the sta-
bilization of every simple sub-quotient of A ⊗ K satisfies the corona factorization
property, then A ⊗ K satisfies the corona factorization property. Consequently, any
graph C ∗-algebra with finitely many ideals has the corona factorization property.
Proof. We will prove the result of the proposition by induction.
If A is simple,
then by our assumption, A ⊗ K has the corona factorization property. Suppose that
the proposition is true for any C ∗-algebra B with at most n ideals such that the
stabilization of any simple sub-quotient of B ⊗ K satisfies the corona factorization
property.
Let A be a C ∗-algebra with n + 1 ideals such that the stabilization of every
simple sub-quotient of A ⊗ K satisfies the corona factorization property. Let I be a
proper non-trivial ideal of A ⊗ K. Then I and A/I are C ∗-algebras with at most n
ideals such that the stabilization of every simple sub-quotient of I and A/I satisfies
the corona factorization property. Hence, I ⊗ K and A/I ⊗ K satisfy the corona
14
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
factorization property. Therefore, by Theorem 3.1(1) of [22], A ⊗ K satisfies the
corona factorization property.
(cid:3)
The authors have been recently informed by Eduard Ortega that his joint work
with Francesc Perera and Mikael Rørdam (see [28]) implies that the stabilization of
any graph C ∗-algebra has the corona factorization property.
Theorem 6.2. (Meyer-Nest [26]) For the topological space Xn, if A and B are
separable, nuclear, C ∗-algebras over Xn such that A[k] and B[k] are in the bootstrap
category N , then any isomorphism α : FKXn(A) → FKXn(B) lifts to an invertible
element in KK (Xn; A, B).
Proposition 6.3. Let A1 and A2 separable, nuclear, C ∗-algebras over Xn. Suppose
Ai[1] is an AF algebra and Ai[2, n] is a tight stable O∞-absorbing C ∗-algebra over
[2, n], and Ai[2] is an essential ideal of Ai[1, 2]. Then A1 ⊗ K ∼= A2 ⊗ K if and only if
there exists an isomorphism α : FKXn(A1) → FKXn(A2) such that α{1} is positive.
Proof. Since Ai[2, n] is a tight C ∗-algebra over [2, n], by Theorem 3.14 of [11], there
exists a norm-full projection p in Ai[2, n]. By Brown's Theorem [9], pAi[2, n]p ⊗ K ∼=
Ai[2, n] ⊗ K. Since Ai[2, n] is an O∞-absorbing C ∗-algebra, by Corollary 3.1 of [34],
pAi[2, n]p is an O∞-absorbing C ∗-algebra. By Corollary 5.3,
0 → Ai[2, n] ⊗ K → Ai ⊗ K → Ai[1] ⊗ K → 0
is a full extension for n = 3. Suppose n = 2. Then Ai[2, 2] is a purely infinite simple
C ∗-algebra, hence Q(Ai[2, 2] ⊗ K) is a simple C ∗-algebra. Thus,
0 → Ai[2, 2] ⊗ K → Ai ⊗ K → Ai[1] ⊗ K → 0
is a full extension. By Proposition 6.1, Ai[2, n] has the corona factorization property.
The theorem now follows from Theorem 4.6.
(cid:3)
The following specialization of the result above is in a certain sense dual to The-
orem 4.7 of [18].
Corollary 6.4. Let A1 and A2 be graph C ∗-algebras satisfying Condition (K) and
A1 and A2 are C ∗-algebras over X2. Suppose Ai[2] is O∞-absorbing, Ai[2] is the
smallest ideal of Ai, and Ai[1] is an AF algebra. Then A1 ⊗ K ∼= A2 ⊗ K if and only if
there exists an isomorphism α : FKXn(A1) → FKXn(A2) such that α{1} is positive.
Proposition 6.5. Let A1 and A2 be graph C ∗-algebras satisfying Condition (K).
Suppose Ai is a C ∗-algebra over Xn such that Ai[n] is an AF algebra, for every
ideal I of Ai we have that I ⊆ Ai[n] or Ai[n] ⊆ I, and Ai[1, n − 1] is a tight, O∞-
absorbing C ∗-algebra over [1, n − 1]. Then A1 ⊗ K ∼= A2 ⊗ K if and only if there exists
an isomorphism α : FKXn(A1) → FKXn(A2) such that α{n} is positive.
Proof. By Corollary 5.6, 0 → Ai[n] ⊗ K → Ai ⊗ K → Ai[1, n − 1] ⊗ K → 0 is a
full extension. By Lemma 3.10 of [16], Ai[n] ⊗ K satisfies the corona factorization
property. The result now follows from Theorem 4.6.
(cid:3)
Remark 6.6. Note that the above propositions do not assume that the ideal lattice
of Ai is linear. For example, Proposition 6.5 can be applied to C ∗-algebras A that
CLASSIFYING C∗-ALGEBRAS
15
are tight C ∗-algebras over X = {1, 2, 3, 4} with O(X) = {∅} ∪ {U ⊆ X : 4 ∈ U }
such that
(1) A({4}) is purely infinite
(2) A({1, 2, 3}) is an AF algebra.
In a forthcoming paper, we study graph C ∗-algebras with small ideal structures
similar to the C ∗-algebra described above. Proposition 6.3, Proposition 6.5, and
related results will be used to classify these graph C ∗-algebras.
Definition 6.7. For each n ∈ N, we define a class Cn of graph C ∗-algebras as
follows: A is in Cn if
(1) A is a graph C ∗-algebra;
(2) A is a tight C ∗-algebra over Xn; and
(3) there exists U ∈ O(Xn) such that either A(U ) is an AF algebra and A(Xn\U )
is O∞-absorbing or A(U ) is O∞-absorbing and A(X \ U ) is an AF algebra.
Note that if C ∗(E) is an element in Cn, then by the proof of Lemma 3.1 of [18],
E satisfies Condition (K).
Theorem 6.8. Let E1 and E2 be graphs such that C ∗(E1) and C ∗(E2) are in Cn
for some n ∈ N. Then the following are equivalent:
(1) C ∗(E1) ⊗ K ∼= C ∗(E2) ⊗ K
(2) There exists an isomorphism α : FK+
Xn
(C ∗(E1)) → FK+
Xn
(C ∗(E2))
Proof. Suppose there exists an isomorphism α : FK+
(C ∗(E2)).
Xn
Note that by Cuntz [12], if A is an O∞-absorbing with a norm-full projection,
then K0(A) = K0(A)+. Since K0(B) 6= K0(B)+ for any AF algebra, there is no
positive isomorphism from the K0-group of an AF algebra to the K0-group of an
O∞-absorbing C ∗-algebra with a norm-full projection.
(C ∗(E1)) → FK+
Xn
With the above observation, one of the following four cases must happen:
(i) C ∗(E1) and C ∗(E2) are AF algebras;
(ii) C ∗(E2) and C ∗(E2) are O∞-absorbing;
(iii) there exists 2 ≤ k ≤ n such that C ∗(Ei)[k, n] is an AF algebra and C ∗(Ei)[1, k−
1] is O∞-absorbing for i = 1, 2;
(iv) there exists 2 ≤ k ≤ n such that C ∗(Ei)[k, n] is O∞-absorbing and C ∗(Ei)[1, k−
1] is an AF algebra for i = 1, 2.
Case (i) follows from the classification of AF algebras. Case (ii) follows from
Theorem 4.14 of [26]. Case (iii) follows from Proposition 6.5 and Case (iv) follows
from Proposition 6.3.
(cid:3)
7.1. Case I. Fix a prime p and consider the class of graph C ∗-algebras given by
adjacency matrices
7. Examples
0
0
z p + 1
y
x
0
0
p + 1
16
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
for y, z > 0. Theorem 6.8 applies directly as the resulting graph C ∗-algebra has a
finite linear ideal lattice 0 ⊳ I1 ⊳ I2 ⊳ A with subquotients I1 = K, I2/I1 = Op+1 ⊗ K,
and A/I2 = Op+1. All K1-groups in the filtered K-theory vanish, and the K0-groups
and the natural transformations
K0(I1)
/ K0(I2)
K0(I2/I1)
K0(I1)
/ K0(A)
K0(A/I1)
K0(A/I2)
K0(A/I2)
may be computed as
Z
Z
/ cok [ z
p ]
x p
/ cokh z y
p 0i
cok(cid:2)p(cid:3)
cok(cid:2)p(cid:3)
cok(cid:2) x p
p 0(cid:3)
cok(cid:2)p(cid:3)
with all maps induced by the canonical maps from Zr into Zs for suitably chosen r
and s.
Checking when two such filtered K-theories are isomorphic is not easy. Of course
it would be necessary that
p x ⇔ p x′
p z ⇔ p z′
but depending upon the invertibility of x and z in Z/p we get varying conditions on
y. The work in [2] (see [1, p. 33, 39]) explains how to reduce this task to checking
isomorphism only in the part of the invariant enclosed in dashed lines. With this,
it is easy to conclude:
Example 7.1. With graphs E and E′ given by matrices
0
0
z p + 1
y
x
0
0
p + 1
0
0
z′ p + 1
y′
x′
,
0
0
p + 1
respectively, we have C ∗(E) ⊗ K ≃ C ∗(E′) ⊗ K precisely when
(1) p x ⇔ p x′, and
(2) p z ⇔ p z′, and
(3)
(a) p y ⇔ p y′ when p x and p z
(b) p [y − xz/p] ⇔ p [y′ − x′z′/p] when p ∤ x and p z
/
/
/
/
/
/
/
/
/
/
/
/
CLASSIFYING C∗-ALGEBRAS
17
7.2. Case II. We now consider graphs given by
0
0
x p + 1
y
z
0
0
0
0
p + 1
0
0
0
0
p + 1
with x, y, z > 0. The resulting ideal lattice is not linear; in fact we have an extension
0
/ K
/ A
/ Op+1 ⊕ Op+1 ⊕ Op+1
/ 0
(7.1)
showing that the ideal lattice is precisely of the type demonstrated by Meyer and
Nest in [26] to not generally allow a UCT for filtered K-theory. But since our C ∗-
algebras have real rank zero, we may appeal to [3] to see that isomorphisms of the
filtered K-theory, which in this case has the form
cok [ x
p ]
Z
cok [ y
p ]
A✄✄✄✄✄✄✄✄✄✄✄
❀❀❀❀❀❀❀❀❀❀❀❀
cok [ z
p ]
cok(cid:2)p(cid:3)
;①①①①①①①①①①
#❋❋❋❋❋❋❋❋❋❋
cok(cid:2)p(cid:3)
cok(cid:2)p,(cid:3)
"❉❉❉❉❉❉❉❉❉❉❉❉
<③③③③③③③③③③③
"❉❉❉❉❉❉❉❉❉❉❉
<③③③③③③③③③③③③
p 0
p 0
cokh x y
0 pi
#❍❍❍❍❍❍❍❍❍
cokh x z
0 pi
;✈✈✈✈✈✈✈✈✈
cokh y z
0 pi
p 0
/ cok" x y z
0 0 p#
p 0 0
0 p 0
lift to invertible elements of KK , so since the ideal I1 ∼= K is a least ideal with
A/I1 absorbing O∞ we may apply Theorem 4.6. Further, as explained in [1, p.
33, 39] it suffices by [2] to check the existence of isomorphisms on the part of the
invariant enclosed in dashed lines, and then it is straightforward to determine when
the filtered K-theory for two such matrices are the same; indeed this amounts to
p x ⇔ p x′
p y ⇔ p y′
p z ⇔ p z′.
Taking into account the homeomorphims of Prim(A) we arrive at
Example 7.2. With graphs E and E′ given by matrices
0
0
x p + 1
y
z
0
0
0
0
p + 1
0
0
0
0
p + 1
0
0
x′ p + 1
y′
z′
0
0
0
0
p + 1
0
0
0
0
p + 1
,
respectively, we have C ∗(E) ⊗ K ≃ C ∗(E′) ⊗ K if and only if the number of entries
in (x, y, z) which are multiples of p agrees with the number of entries in (x′, y′, z′)
which are multiples of p.
Remark 7.3. Note that KK 1(Op+1 ⊕ Op+1 ⊕ Op+1, K) does not tell the full story
about stable isomorphism among the possible extensions fitting in (7.1), as indeed,
we have only 4 stable isomorphism classes among the p3 different extensions.
/
/
/
/
/
/
"
#
A
/
/
<
"
/
/
/
;
#
/
/
<
;
18
SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ
8. Acknowledgments
This work was supported by the NordForsk Research Network "Operator algebras
and dynamics" grant # 11580 and by the Faroese Research Council. The second
and third named author wishes to thank the Department of Mathematical Sciences
at the University of Copenhagen for their support and hospitality.
References
[1] S. Arklint, Classification of non-simple C ∗-algebras of real rank zero, PhD Thesis, University
of Copenhagen, 2012.
[2] S. Arklint, R. Bentmann, and T. Katsura, Reduction of filtered K-theory and a charac-
terization of Cuntz-Krieger algebras, preprint.
[3] S. Arklint, G. Restorff, and E. Ruiz, Filtrated K-theory for real rank zero C ∗-algebras,
preprint.
[4] T. Bates, J. H. Hong, I. Raeburn, and W. Szyma´nski, The ideal structure of the C ∗-
algebras of infinite graphs, Illinois J. Math., 46 (2002), pp. 1159 -- 1176.
[5] T. Bates, D. Pask, I. Raeburn, and W. Szyma´nski, The C ∗-algebras of row-finite graphs,
New York J. Math., 6 (2000), pp. 307 -- 324 (electronic).
[6] R. Bentmann and M. Kohler, Universal coefficient theorems for C ∗-algebras over finite
topological spaces, 2001. arXiv:1101.5702.
[7] B. Blackadar, K-theory for operator algebras, vol. 5 of Mathematical Sciences Research
Institute Publications, Cambridge University Press, Cambridge, second ed., 1998.
[8] A. Bonkat, Bivariante K-Theorie fur Kategorien projektiver Systeme von C ∗-Algebren, 2002.
Ph.D. thesis, Westfalische Wilhelms Universitat Munster, 2002, Preprintreihe SFB 478, heft
319.
[9] L. G. Brown, Stable isomorphism of hereditary subalgebras of C ∗-algebras, Pacific J. Math.,
71 (1977), pp. 335 -- 348.
, Semicontinuity and multipliers of C ∗-algebras, Canad. J. Math., 40 (1988), pp. 865 -- 988.
[10]
[11] L. G. Brown and G. K. Pedersen, C ∗-algebras of real rank zero, J. Funct. Anal., 99(1)
(1991), pp. 131 -- 149.
[12] J. Cuntz, K-theory for certain C ∗-algebras, Ann. of Math. (2), 113 (1981), pp. 181 -- 197.
[13] J. Cuntz and G. Skandalis, Mapping cones and exact sequences in KK-theory, J. Operator
Theory, 15 (1986), pp. 163 -- 180.
[14] D. Drinen and M. Tomforde, The C ∗-algebras of arbitrary graphs, Rocky Mountain J.
Math., 35 (2005), pp. 105 -- 135.
[15] S. Eilers, T. A. Loring, and G. K. Pedersen, Morphisms of extensions of C ∗-algebras:
pushing forward the Busby invariant, Adv. Math., 147 (1999), pp. 74 -- 109.
[16] S. Eilers, G. Restorff, and E. Ruiz, Classification of extensions of classifiable C ∗-algebras,
Adv. Math., 222 (2009), pp. 2153 -- 2172.
[17]
, On graph C ∗-algebras with a linear ideal lattice, Bull. Malaysian Math. Sci. Soc. (2), 33
(2010), 233 -- 241.
[18] S. Eilers and M. Tomforde, On the classification of nonsimple graph C ∗-algebras, Math.
Ann., 346 (2010), pp. 393 -- 418.
[19] G. A. Elliott, On the classification of inductive limits of sequences of semisimple finite-
dimensional algebras, J. Algebra, 38 (1976), pp. 29 -- 44.
[20] E. Kirchberg, Das nicht-kommutative Michael-Auswahlprinzip und die Klassifikation nicht-
einfacher Algebren, in C ∗-algebras (Munster, 1999), Springer, Berlin, 2000, pp. 92 -- 141.
[21] D. Kucerovsky and P. W. Ng, The corona factorization property and approximate unitary
equivalence, Houston J. Math., 32 (2006), pp. 531 -- 550.
[22]
, S-regularity and the corona factorization property, Math. Scand., 99 (2006), pp. 204 --
216.
[23] H. Lin, A separable Brown-Douglas-Fillmore theorem and weak stability, Trans. Amer. Math.
Soc., 356 (2004), pp. 2889 -- 2925 (electronic).
CLASSIFYING C∗-ALGEBRAS
19
[24]
, Full extensions and approximate unitary equivalence, Pacific J. Math., 229 (2007),
pp. 389 -- 428.
[25] R. Meyer and R. Nest, C ∗-algebras over topological spaces: the bootstrap class, Munster J.
Math., 2 (2009), pp. 215 -- 252.
[26]
[27] P. W. Ng, The corona factorization property,
, C ∗-algebras over topological spaces: Filtrated K-theory, 2009. arXiv:0810:0096v2.
in Operator theory, operator algebras, and
applications, vol. 414 of Contemp. Math., Amer. Math. Soc., Providence, RI, 2006, pp. 97 --
110.
[28] E. Ortega, F. Perera, and M. Rørdam, The corona factorization property and refinement
monoids, 2009. arXiv:0904.0541.
[29] C. Pasnicu and M. Rørdam, Purely infinite C ∗-algebras of real rank zero, J. Reine Angew.
Math., 613 (2007), pp. 51 -- 73.
[30] G. Restorff, Classification of Cuntz-Krieger algebras up to stable isomorphism, J. Reine
Angew. Math., 598 (2006), pp. 185 -- 210.
[31] M. Rørdam, Ideals in the multiplier algebra of a stable C ∗-algebra, J. Operator Theory, 25
(1991), pp. 283 -- 298.
[32]
, Classification of extensions of certain C ∗-algebras by their six term exact sequences in
K-theory, Math. Ann., 308 (1997), pp. 93 -- 117.
[33] M. Tomforde, Structure of graph C ∗-algebras and their generalizations, 2006. Graph Alge-
bras: Bridging the gap between analysis and algebra (Gonzalo Aranda Pino, Francesc Perera
Dom`enech, and Mercedes Siles Molina, eds.), Servicio de Publicaciones de la Universidad de
M´alaga, M´alaga, Spain.
[34] A. Toms and W. Winter, Strongly self-absorbing C ∗-algebras, Trans. Amer. Math. Soc., 359
(2007), pp. 3999 -- 4029 (electronic).
Department of Mathematical Sciences, University of Copenhagen, Universitetsparken 5,
DK-2100 Copenhagen, Denmark
E-mail address: [email protected]
Faculty of Science and Technology, University of Faroe Islands, N´oat´un 3, FO-100
T´orshavn, Faroe Islands
E-mail address: [email protected]
Department of Mathematics, University of Hawaii, Hilo, 200 W. Kawili St., Hilo,
Hawaii, 96720-4091 USA
E-mail address: [email protected]
|
1609.04001 | 3 | 1609 | 2017-03-31T07:55:08 | A noncommutative version of Farber's topological complexity | [
"math.OA"
] | Topological complexity for spaces was introduced by M. Farber as a minimal number of continuity domains for motion planning algorithms. It turns out that this notion can be extended to the case of not necessarily commutative C*-algebras. Topological complexity for spaces is closely related to the Lusternik--Schnirelmann category, for which we do not know any noncommutative extension, so there is no hope to generalize the known estimation methods, but we are able to evaluate the topological complexity for some very simple examples of noncommutative C*-algebras. | math.OA | math |
A NONCOMMUTATIVE VERSION OF FARBER'S TOPOLOGICAL
COMPLEXITY
V. MANUILOV
Abstract. Topological complexity for spaces was introduced by M. Farber as a mini-
mal number of continuity domains for motion planning algorithms. It turns out that this
notion can be extended to the case of not necessarily commutative C ∗-algebras. Topolog-
ical complexity for spaces is closely related to the Lusternik -- Schnirelmann category, for
which we do not know any noncommutative extension, so there is no hope to generalize
the known estimation methods, but we are able to evaluate the topological complexity
for some very simple examples of noncommutative C ∗-algebras.
Introduction
Gelfand duality between compact Hausdorff spaces and unital commutative C ∗ -- algebras
allows to translate some topological constructions and invariants into the noncommutative
setting. The most successful example is K-theory, which became a very useful tool in
C ∗-algebra theory. Homotopies between ∗-homomorphisms of C ∗-algebras also play an
important role, but there is no nice general homotopy theory for C ∗-algebras due to the
fact that the loop functor has no left adjoint [11], Appendix A. Nevertheless, there are
some homotopy invariants that allow noncommutative versions.
The aim of our work is to show that M. Farber's topological complexity [4] is one of
those. In Section 1 we recall the original commutative definition of topological complexity,
and in Section 2 we use Gelfand duality to reverse arrows in this definition, and show
that the resulting noncommutative definition generalizes the commutative one.
In the
remaining two sections we calculate topological complexity for some simple examples of
C ∗-algebras. In particular, we show that introducing noncommutative coefficients may
decrease topological complexity. Although in most our examples topological complexity
is either 1 or ∞, we provide a noncommutative example with topological complexity 2.
The author is grateful to A. Korchagin for helpful comments.
1. Farber's topological complexity
The topological approach to the robot motion planning problem was initiated by M.
Farber in [4]. Let us recall his basic construction. Let X be the configuration space
of a mechanical system. A continuous path γ : [0, 1] → X represents a motion of the
system, with γ(0) and γ(1) being the initial and the final state of the system. If X is
path-connected then the system can be moved to an arbitrary state from a given state.
Let P X denote the space of paths in X with the compact-open topology, and let
π : P X → X × X
(1)
The author acknowledges partial support by the RFBR grant No. 16-01-00357.
1
2
V. MANUILOV
be the map given by π(γ) = (γ(0), γ(1)). A continuous motion planning algorithm is a
continuous section
s : X × X → P X
of π. Typically, there may be no continuous motion planning algorithm, so one may take
a covering of X × X by sets V1, . . . , Vn (domains of continuity) and require existence of
continuous sections
si : Vi → P XVi
of maps πi : P XVi → Vi, i = 1, . . . , n. Here P XVi denotes the restiction of π onto Vi, i.e.
the subset of paths γ : [0, 1] → X such that (γ(0), γ(1)) ∈ Vi. In this case, the collection of
the sections si, i = 1, . . . , n, is called a (discontinuous) motion planning algorithm. There
are several versions of the definition, which use various kinds of coverings, e.g. coverings
by open or closed sets, or by Euclidean neighborhood retracts, etc., but most of them
agree on simplicial polyhedra (cf. [5], Theorem 13.1). The topological complexity T C(X)
of X is the minimal number n of domains of continuity, i.e. the minimal number n, for
which there exists a covering V1, . . . , Vn and continuous sections si as above. This number
measures the complexity of the problem of navigation in X.
2. Noncommutative version of topological complexity
For a compact Hausdorff space X we can rewrite the above construction in terms of
unital commutative C ∗-algebras and their unital ∗-homomorphisms using Gelfand duality.
Let C(X) denote the commutative C ∗-algebra of complex-valued continuous functions on
X. A closed covering V1, . . . , Vn of X × X corresponds to n surjective ∗-homomorphisms
ji : C(X) ⊗ C(X) → C(Vi),
i = 1, . . . , n, with ∩n
i=1 Ker ji = {0}. As the path space P X is not locally compact, it is
not Gelfand dual to any C ∗-algebra, but we can bypass this, replacing the sections si by
∗-homomorphisms
defined by
σi : C(X) → C(Vi) ⊗ C[0, 1]
σi(f )(x, t) = f (si(x)(t)),
where x ∈ Vi, t ∈ [0, 1], f ∈ C(X). Let us denote by evt the ∗-homomorphism of
evaluation at t ∈ [0, 1], and let us consider the compositions
Let
ev0 ◦σi, ev1 ◦σi : C(X) → C(Vi).
π0, π1 : X × X → X
denote the projections onto the first and the second copy of X respectively, and let
p0, p1 : C(X) → C(X) ⊗ C(X)
be the corresponding ∗-homomorphisms. The condition π ◦ si = idVi can be written as
πk ◦ π ◦ si = πk : Vi → X, k = 0, 1, which allows rewriting, in terms of C ∗-algebras and
∗-homomorphisms, as ji ◦ p0 = ev0 ◦σi, ji ◦ p1 = ev1 ◦σi. Thus we have
A NONCOMMUTATIVE VERSION OF FARBER'S TOPOLOGICAL COMPLEXITY
3
Lemma 2.1. Continuous sections si : Vi → P XVi exist iff there exist ∗-homomorphisms
σi making the diagrams
C(X)
σi
pk
C(X) ⊗ C(X)
ji
C(Vi) ⊗ C[0, 1]
/ C(Vi),
evk
(2)
k = 0, 1, commute.
Thus, we may define the topological complexity T C(A) for a unital C ∗-algebra A as the
minimal number n of quotient C ∗-algebras B1, . . . , Bn of A ⊗ A with the quotient maps
qi : A ⊗ A → Bi, such that
i=1 Ker qi = {0};
(1) ∩n
(2) there exist ∗-homomorphisms
σi : A → Bi ⊗ C[0, 1],
i = 1, . . . , n,
making the diagrams
pk
A
σi
A ⊗ A
qi
Bi ⊗ C[0, 1]
/ Bi,
evk
(3)
k = 0, 1, commute for each i = 1, . . . , n, where p0(a) = a ⊗ 1, p1(a) = 1 ⊗ a, a ∈ A.
Here and further we always use ⊗ to denote the minimal tensor product of C ∗-algebras.
If there is no such n then we set T C(A) = ∞.
Corollary 2.2. For a compact Hausdorff space X, one has T C(C(X)) = T C(X) if
T C(X) is defined using closed coverings.
Proof. Commutativity of A = C(X), hence of A ⊗ A, implies commutativity of Bi, hence
Bi = C(Vi) for some Vi. Surjectivity of qi implies that Vi is a closed subset of X × X.
The condition ∩n
i=1 Ker qi = {0} means that {V1, . . . , Vn} is a covering for X × X.
(cid:3)
As we shall see later, topological complexity is not well suited for general C ∗-algebras,
e.g. it is infinite for topologically non-trivial simple C ∗-algebras, but there are two good
classes of C ∗-algebras, for which this characterization may be interesting -- the class of
noncommutative CW complexes introduced in [3] and the class of C(X)-algebras. Most
of our examples are from the first class.
Note that in the commutative case, topological complexity makes sense only for path-
connected spaces -- otherwise any two points may be not connected by a path, i.e. the
map (1) is not surjective. There is no good C ∗-algebraic analog for that, but the following
holds:
Lemma 2.3. Let A = A1 ⊕ A2. Then T C(A) = ∞.
/
/
/
/
/
/
4
V. MANUILOV
Proof. One has A ⊗ A = ⊕2
k,l=1Ak ⊗ Al. Let qi : A ⊗ A → Bi, i = 1, . . . , n, and σ : A →
Bi ⊗ C[0, 1] be as in the definition of topological complexity, and let e1 = qi(1A1 ⊗ 1A1),
e2 = qi(1A1 ⊗ 1A2), e3 = qi(1A2 ⊗ 1A1), e4 = qi(1A2 ⊗ 1A2). Then e1, . . . , e4 are projections
k=1 ekbek. In particular, if
e ∈ Bi is a projection then each ekeek is a projection, and if e(t), t ∈ [0, 1], is a homotopy
of projections, then we have four homotopies eke(t)ek.
in Bi, and, as qi is surjective, any element of Bi has the form P4
Let a = 1A1 ⊕ 0A2 ∈ A. Then qi(p0(a)) = e1 + e2 and qi(p1(a)) = e1 + e3 should be
connected by a homotopy. This is possible only if e2 = e3 = 0. As this argument does not
depend on i, we conclude that 1A1 ⊗ 1A2, 1A2 ⊗ 1A1 ∈ ∩n
i=1 Ker qi = {0} -- a contradiction.
(cid:3)
The topological complexity of a space X can be estimated from above by using covering
dimension of X, and from below using multiplicative structure in cohomology. Regretfully,
these estimates cannot work in the noncommutative case, thus making the problem of
evaluating topological complexity even more difficult.
3. Case T C(A) = 1
The condition T C(A) = 1 means that the two inclusions of A into A ⊗ A, p0 : a 7→ a ⊗ 1
and p1 : a 7→ 1 ⊗ a, are homotopic. This property is similar to, but different from that of
approximately inner half flip [10], which means that p0 and p1 are approximately unitarily
equivalent, i.e. there exist unitaries un ∈ A⊗A such that limn→∞ kp1(a)−Adun p0(a)k = 0
for any a ∈ A.
The condition T C(A) = 1 imposes restrictions on the K-theory groups of A. Let
K∗(A) denote the graded K-theory group of A, and let 1 ∈ K0(A) be the class of the
unit element. Recall that if A is in the bootstrap class [7] then it satisfies the Kunneth
formula, hence K∗(A) ⊗ K∗(A) ⊂ K∗(A ⊗ A). The bootstrap class is the smallest class
which contains all separable type I C ∗-algebras and is closed under extensions, strong
Morita equivalence, inductive limits, and crossed products by R and by Z.
Lemma 3.1. Let A satisfy K∗(A) ⊗ K∗(A) ⊂ K∗(A ⊗ A). If K∗(A) ⊗ 1 6= 1 ⊗ K∗(A)
then T C(A) > 1.
Proof. This follows from homotopy invariance of K-theory groups. If T C(A) = 1 then
the flip on K∗(A ⊗ A) must induce the identity map.
(cid:3)
For spaces, it is known that T C(X) = 1 iff X is contractible. For C ∗-algebras, it
is reasonable to call a unital C ∗-algebra A contractible to a point if there exists a ∗-
homomorphism h : A → A ⊗ C[0, 1] and a ∗-homomorphism i : A → C such that
ev1 ◦h = idA and ev0 ◦h = j ◦ i, where j : C → A is defined by j(1) = 1A. If B is a
non-unital contractible C ∗-algebra then its unitalization B+ is contractible to a point.
Lemma 3.2. Let A be contractible to a point. Then T C(A) = 1.
Proof. Let α : A ⊗ A ⊗ C[0, 1] be the flip, α(a1 ⊗ a2 ⊗ f ) = a2 ⊗ a1 ⊗ f , where a1, a2 ∈ A,
f ∈ C[0, 1]. Let h : A → A ⊗ C[0, 1] be the homotopy as above. We write ht for evt ◦h.
Define a ∗-homomorphism
σ : A → A ⊗ A ⊗ C[−1, 1]
A NONCOMMUTATIVE VERSION OF FARBER'S TOPOLOGICAL COMPLEXITY
5
by setting, for a ∈ A,
σ(a)(t) = (cid:26) α(1 ⊗ ht(a)),
1 ⊗ h−t(a),
if t ∈ [0, 1];
if t ∈ [−1, 0].
Then ev1 ◦σ(a) = a ⊗ 1, ev−1 ◦σ(a) = 1 ⊗ a. Continuity of σ at t = 0 follows from the
equality i(a) ⊗ 1 = 1 ⊗ i(a).
(cid:3)
Corollary 3.3. If An = {f ∈ C([0, 1]; Mn) : f (1) is scalar} then T C(An) = 1.
Lemma 3.4. Let T C(A) = 1. If there exists a unital ∗-homomorphism i : A → C then
A is contractible to a point.
Proof. Let σ : A → A ⊗ A ⊗ C[0, 1] satisfy ev0 ◦σ(a) = a ⊗ 1 and ev1 ◦σ(a) = 1 ⊗ a, a ∈ A.
Let ¯ι : A ⊗ A ⊗ C[0, 1] → A ⊗ C[0, 1] be the map defined by ¯ι(a1 ⊗ a2 ⊗ f ) = i(a1) · a2 ⊗ f ,
where a1, a2 ∈ A, f ∈ C[0, 1]. Set h = ¯ι ◦ σ : A → A ⊗ C[0, 1]. Then ev0 ◦h(a) = i(a) · 1,
ev1 ◦h(a) = a, hence h is the required homotopy.
(cid:3)
Below we list three examples of C ∗-algebras with topological complexity 1. The proofs
are known to specialists, but we could not find exact references.
Proposition 3.5. One has T C(Mn) = 1.
Proof. Let U be a unitary in Mn2 ∼= Mn ⊗ Mn such that AdU is an automorphism of Mn2
that interchanges Mn ⊗ 1 with 1 ⊗ Mn. If Mn acts on an n-dimensional space Hn with the
orthonormal basis {ei}n
i=1 then U interchanges vectors ei ⊗ ej and ej ⊗ ei when i 6= j. Let
Ut, t ∈ [0, 1], be the path connecting U with 1 constructed using the standard rotation
formula. Define σ : Mn → Mn ⊗ Mn ⊗ C[0, 1] by σ(a)(t) = AdUt(a ⊗ 1), a ∈ Mn.
(cid:3)
The above example can be extended to UHF algebras:
Proposition 3.6. If A is a UHF algebra then T C(A) = 1.
Proof. Let n, k be integers, ϕ : Mn → Mkn a unital ∗-homomorphism. Let σ′ : Mn →
Mn ⊗ Mn ⊗ C[0, 1] and σ′′ : Mkn → Mkn ⊗ Mkn ⊗ C[0, 1] be the maps constructed in
t (a′′ ⊗ 1), a′ ∈ Mn,
the proof of Lemma 3.5, σ′(a′)(t) = AdU ′
a′′ ∈ Mkn.
t (a′ ⊗ 1), σ′′(a′′)(t) = AdU ′′
Then the diagram
Mn
ϕ
σ′
Mn ⊗ Mn ⊗ C[0, 1]
ϕ⊗ϕ⊗id
Mkn
σ′′
/ Mkn ⊗ Mkn ⊗ C[0, 1]
(4)
commutes. Let A be the direct limit of matrix algebras An = Mmn, where mn divides
mn+1, n ∈ N. Commutativity of the diagram (4) shows that the maps σ(n) : An → An ⊗
An ⊗ C[0, 1] agree, so, for any t ∈ [0, 1] one can define the limit map σt : A → A ⊗ A such
that (σt)An = evt ◦σ(n). Since kσt(a)k ≤ kak for any a ∈ A and any t ∈ [0, 1], continuity
of σt(a) with respect to t for a ∈ ∪∞
n=1An implies continuity of σt(a) for any a ∈ A. This
/
/
/
6
V. MANUILOV
means that the family {σt}t∈[0,1] defines a ∗-homomorphism σ : A → A ⊗ A ⊗ C[0, 1],
which provides the required homotopy.
(cid:3)
Let O2 be the Cuntz algebra generated by two isometries s1, s2 satisfying s1s∗
1+s2s∗
2 = 1.
Proposition 3.7. One has T C(O2) = 1.
Proof. Let u = s∗
generators that p0 = Adu p1 (cf.
5.2.1), and the unitary group of O2 is contractible, hence, p0 and p1 are homotopic.
2 ⊗ s2 ∈ O2 ⊗ O2. It is unitary, and it suffices to check on
∼= O2 ([6], Theorem
[6], Theorem 5.1.2). But O2 ⊗ O2
1 ⊗ s1 + s∗
(cid:3)
4. General case
Let K+ be the unitalized algebra of compact operators. In contrast with Lemma 3.5,
its topological complexity is infinite. This often happens for C ∗-algebras with few ideals.
Lemma 4.1. One has T C(K+) = ∞.
Proof. Let B1, . . . , Bn be the quotients of K+ ⊗ K+.
If they satisfy the definition of
topological complexity then one of them must coincide with K+ ⊗ K+ itself, in which case
other quotients are redundant. Therefore, if T C(K+) 6= ∞ then T C(K+) = 1. To show
that this is not the case, recall that K0(K+) ∼= Z2 and use Lemma 3.1.
(cid:3)
Lemma 4.2. Let T C(A) > 1. If A is simple then T C(A) = ∞.
Proof. It follows from [8] that A ⊗ A is simple, hence any possible quotient B must equal
A ⊗ A.
(cid:3)
It follows that topological complexity distinguishes commutative C ∗-algebras from their
non-commutative deformations. For example, consider an irrational rotation algebra Aθ,
θ ∈ [0, 1] \ Q, often called a non-commutative torus. It is simple and has the same K-
theory as the usual torus T2 [2], hence T C(Aθ) = ∞, while for a usual torus T2 one has
T C(C(T2)) = 3 (cf. [5], Example 16.4).
Nevertheless, tensoring by matrices does not increase topological complexity.
Proposition 4.3. For any compact Hausdorff space X, one has T C(C(X) ⊗ Mn) ≤
T C(C(X)).
Proof. Let T C(C(X)) = k, and let qi : C(X) ⊗ C(X) → Bi and σi : C(X) → Bi ⊗ C[0, 1],
i = 1, . . . , k, be as in the definition of topological complexity. Set Bi = Bi ⊗ Mn ⊗ Mn,
qi = qi ⊗ id : C(X) ⊗ C(X) ⊗ Mn ⊗ Mn → Bi. Define σi : C(X) ⊗ Mn → Bi ⊗ C[0, 1] by
σi(f ⊗m)(t) = σi(f )⊗AdUt(m⊗1) ∈ Bi⊗C[0, 1]⊗Mn⊗Mn, f ∈ C(X), m ∈ Mn, t ∈ [0, 1],
and Ut as in the proof of Lemma 3.5. Then the maps qi, σi make the corresponding
diagrams commute, hence T C(C(X) ⊗ Mn) ≤ T C(C(X)).
(cid:3)
More generally, one has
Proposition 4.4. Let T C(A) = n, T C(C) = m. Then T C(A ⊗ C) ≤ nm.
A NONCOMMUTATIVE VERSION OF FARBER'S TOPOLOGICAL COMPLEXITY
7
: A ⊗ A → Bi, σA
i
Proof. Let qA
: C ⊗ C → Dj,
i
σC
: C → Dj ⊗ C[0, 1], j = 1, . . . , m, be as in the definition of topological complexity. Let
j
∆ : C([0, 1]2) → C[0, 1] be the map induced by the diagonal embedding [0, 1] → [0, 1]2
and define the composition
: A → Bi ⊗ C[0, 1], i = 1, . . . , n, and qC
j
σij : A ⊗ C
σA
i
⊗σC
j
/ Bi ⊗ Dj ⊗ C([0, 1]2)
id ⊗∆
/ Bi ⊗ Dj ⊗ C[0, 1].
Then the diagram
pA
k
⊗pC
k
A ⊗ C
σij
A ⊗ C ⊗ A ⊗ C
qA
i
⊗qC
j
Bi ⊗ Dj ⊗ C[0, 1]
evk
/ Bi ⊗ Dj,
k = 0, 1, commutes for all i, j.
(cid:3)
Remark that in the commutative case the tensor product of C ∗-algebras is Gelfand dual
to the product of spaces, and there is a much better estimate T C(A ⊗ C) ≤ n + m − 1
([4], Theorem 11).
We have no examples with T C(C(X) ⊗ Mn) < T C(C(X)), but tensoring by a more
general C ∗-algebra may decrease topological complexity. Let U(A) denote the group of
unitaries of a C ∗-algebra A. Recall that U(O2) is contractible [9].
Let S denote the circle. It is known that T C(C(S)) = 2.
Theorem 4.5. Let A satisfy T C(A) = 1, π0(U(A)) = π1(U(A)) = 0 (e.g. A = O2).
Then T C(C(S) ⊗ A) = 1.
Proof. We have to connect by a homotopy the two ∗-homomorphisms σi : C(S) ⊗ A →
C(S)⊗A⊗C(S)⊗A, i = 0, 1, given by σ0(f ⊗a) = f ⊗a⊗1⊗1 and σ1(f ⊗a) = 1⊗1⊗f ⊗a,
f ∈ C(S), a ∈ A. Note that these maps are determined by their values on u ⊗ a, where
u(x) = e2πix, u ∈ C(S). By assumption, any unitary in C(S) ⊗ A has a homotopy that
connects it with 1⊗1. Let ut, t ∈ [2/3, 1], be a homotopy, in the unitary group of C(S)⊗A,
that connects u ⊗ 1 with 1 ⊗ 1. Then the homotopy σt, given by σt(u ⊗ a) = 1 ⊗ ut ⊗ a
connects σ1 with σ2/3 given by σ2/3(u ⊗ a) = 1 ⊗ 1 ⊗ 1 ⊗ a. Similarly, one can connect σ0
with σ1/3 given by σ1/3(u ⊗ a) = 1 ⊗ a ⊗ 1 ⊗ 1. Finally, as T C(A) = 1, σ1/3 and σ2/3 are
homotopic.
(cid:3)
Our next examples show how sensitive topological complexity may be. Let
A2 = {f ∈ C([0, 1]; M2) : f (1) is diagonal}.
This algebra is considered as a noncommutative version of the non-Hausdorff T1 space X2
obtained from two intervals {(x, y) ∈ [0, 1]2 : y = 0 or 1} by identifying the points (x, 0)
and (x, 1) for each x ∈ [0, 1) [1]. Although X2 is not Hausdorff, it is contractible, hence
T C(X2) = 1.
Lemma 4.6. One has T C(A2) = ∞.
/
/
/
/
/
8
V. MANUILOV
Proof. Suppose that T C(A2) = n < ∞. Let qi : A2 ⊗ A2 → Bi, i = 1, . . . , n, be as in
the definition of topological complexity. There are two ∗-homomorphisms from A2 to C,
given by r0(f ) = f11(1) and r1(f ) = f22(1), where f ∈ A2. It is easy to see that each
quotient map from A2 factorizes through the restriction map on a closed subset of [0, 1]2.
As ∩n
i=1 Ker qi = {0}, there is at least one i such that r0 ⊗r1 factorizes through qi. Further,
we may argue as in Lemma 2.3: the maps (r0 ⊗ r1) ◦ p0 and (r0 ⊗ r1) ◦ p1 from A2 to C
shoud be homotopic. Let a = ( 1 0
0 0 ) ∈ A2. Then (r0 ⊗ r1) ◦ p0(a) = 1, (r0 ⊗ r1) ◦ p1(a) = 0,
which makes homotopy between (r0 ⊗ r1) ◦ p0 and (r0 ⊗ r1) ◦ p1 impossible.
(cid:3)
Let
Dn = {f ∈ C([0, 1]; Mn) : f (0), f (1) are scalars}
be a (unital) dimension-drop algebra.
Lemma 4.7. If n > 1 then T C(Dn) = ∞.
Proof. We identify Dn ⊗Dn with the subalgebra of functions f = f (x, y) in C([0, 1]2; Mn ⊗
Mn) satisfying the obvious boundary conditions. As above, if there exist k quotients
B1 . . . , Bk of Dn⊗Dn then at least one of them surjects onto a copy of C that identifies with
restrictions of functions f onto the point (1, 0) ∈ [0, 1]2. Denote this map by µ : Bi0 → C.
If there is a homotopy σi0 : Dn → B ⊗ C[0, 1] then it restricts to a homotopy Dn → C ⊗
C[0, 1]. If the diagram (3) commutes then µ◦ev0 ◦σi0(f ) = f (1) and µ◦ev1 ◦σi0(f ) = f (0),
f ∈ Dn. But these two maps are not homotopic.
(cid:3)
In both examples, T C infinite means that there is no "path" connecting 0 and 1 in the
noncommutative versions of an interval. In contrast with these examples is our next one.
Let
Sn = {f ∈ C([0, 1]; Mn) : f (0) = f (1) is scalar}.
This is an algebra of matrix-valued functions on a circle, with the dimension drop at one
point. If n = 1 then S1 is exactly the algebra of continuous functions on a circle.
Theorem 4.8. For any n ∈ N, T C(Sn) = 2.
Proof. We identify Sn ⊗ Sn with the algebra of Mn ⊗ Mn-valued functions on [0, 1]2 with
obvious boundary conditions. Let
Y1 = {(x, y) ∈ [0, 1]2 : x − y ≤ 2/3},
Y2 = {(x, y) ∈ [0, 1]2 : x ≥ 2/3, y ≤ 1/3} ∪ {(x, y) ∈ [0, 1]2 : x ≤ 1/3, y ≥ 2/3}.
Then Y1 ∪ Y2 = [0, 1]2. Let Bi, i = 1, 2, be the algebras of continuous Mn ⊗ Mn-valued
functions with the same boundary conditions as in Sn ⊗ Sn, and let qi : Sn ⊗ Sn → Bi be
the quotient ∗-homomorphisms induced by restrictions onto Yi.
We have to construct homotopies σi : Sn → Bi ⊗ C[0, 1] such that
ev0 ◦σi(f )(x, y) = f (x) ⊗ 1,
ev1 ◦σi(f ) = 1 ⊗ f (y).
(5)
For i = 1, ev0 ◦σ1 is homotopic to σ′ defined by
σ′(f )(x, y) = (cid:26)
f (0) ⊗ 1,
2/3 − 1) ⊗ 1,
f ( x+y
for x + y ≥ 4/3 or x + y ≤ 2/3;
for 2/3 ≤ x + y ≤ 4/3.
A NONCOMMUTATIVE VERSION OF FARBER'S TOPOLOGICAL COMPLEXITY
9
Similarly, ev1 ◦σ1 is homotopic to σ′′ = AdU (σ′), where U intertwines Mn ⊗ 1 and 1 ⊗ Mn.
Finally, σ′ is homotopic to σ′′, as AdUt maps scalars into scalars for any t, where Ut is a
path connecting U with 1, so the boundary conditions on Y1 hold.
For i = 2, as
{(x, y) ∈ [0, 1]2 : x ≥ 2/3, y ≤ 1/3} ∩ {(x, y) ∈ [0, 1]2 : x ≤ 1/3, y ≥ 2/3} = ∅,
so after identifying 0 and 1, there is a single common point (0, 1) = (1, 0). That's why we
can construct the required homotopy separately for each of the C ∗-algebras corresponding
to these sets, but with the additional requirement that the two homotopies should agree at
this common point. And as these sets are symmetric, it suffices to construct a homotopy
for only one of them. Let B0 denote the C ∗-algebra of Mn2-valued functions on
{(x, y) ∈ [0, 1]2 : x ≥ 2/3, y ≤ 1/3}
with the obvious boundary conditions, and let q0 : Sn ⊗ Sn → B0 be the restriction
quotient map.
Note that the maps ev0 ◦σ0 and ev1 ◦σ0 (5) factorize through A0 and A1 respectively,
where
A0 = {f ∈ C([2/3, 1]; Mn) : f (1) is scalar},
A1 = {f ∈ C([0, 1/3]; Mn) : f (0) is scalar}
(i.e. with no restrictions at one of the end-points), hence the map ev0 ◦σ0 is homotopic
to σ′
0 given by
and the map ev1 ◦σ0 is homotopic to σ′′
0 given by
σ′
0(f )(x, y) = f (1) ⊗ 1,
σ′
0(f )(x, y) = 1 ⊗ f (0).
But, as f (0) = f (1), they are homotopic. Along all these homotopies, their values at the
point (1, 0) are the same. Thus, T C(Sn) ≤ 2.
To show that T C(Sn) 6= 1, let us calculate its K-theory groups. As Sn is a split
extension of C by the suspension SMn over Mn, one has K0(Sn) ∼= K1(Sn) ∼= Z. Then
K1(Sn ⊗ Sn) ∼= K0(Sn) ⊗ K1(Sn) ⊕ K1(Sn) ⊗ K0(Sn).
Let (pk)∗ : K1(Sn) → K1(Sn ⊗ Sn) be the maps induced by the ∗-homomorphisms pk :
Sn → Sn ⊗ Sn, k = 0, 1, and let e and u be generators for K0(Sn) and for K1(Sn)
respectively. Then
(p0)∗(u) = u ⊗ e ∈ K1(Sn) ⊗ K0(Sn) ⊂ K1(Sn ⊗ Sn),
(p1)∗(u) = e ⊗ u ∈ K0(Sn) ⊗ K1(Sn) ⊂ K1(Sn ⊗ Sn).
As these elements are different, there is no homotopy that connects p0 with p1.
(cid:3)
10
V. MANUILOV
References
[1] Connes, A. Noncommutative Geometry, Academic Press, San Diego (1994).
[2] Davidson, K. C ∗-Algebrs by Example, Fields Institute Monographs, AMS, Providence (1996).
[3] Eilers, S., Loring, T.A., Pedersen, G.K. Stability of anticommutation relations: an application of
noncommutative CW complexes, J. Reine Angew. Math. 99 (1998), 101-143.
[4] Farber, M. Topological complexity of motion planning, Discrete Comput. Geom. 29, (2003), 211 -- 221.
[5] Farber, M. Topology of robot motion planning, in: Morse Theoretic Methods in Nonlinear Analysis
and in Symplectic Topology. NATO Science Series II: Mathematics, Physics and Chemistry, Springer,
217 (2006), 185 -- 230.
[6] Rørdam M. Classification of nuclear, simple C ∗-algebras, in: Classification of nuclear C ∗-algebras.
Entropy in operator algebras, Springer, Encycl. Math. Sci. 126 (VII) (2002), 1 -- 145.
[7] Schochet, C. Topological methods for C ∗-algebras. II. Geometry resolutions and the Kunneth formula,
Pacific J. Math. 98 (1982), 443 -- 458.
[8] Takesaki, M. On the cross-norm of the direct product of C ∗-algebras, Tohoku Math. J. 16 (1964),
111-122.
[9] Thomsen, K. Nonstable K-theory for operator algebras, K-Theory 4 (1991), 245 -- 267.
[10] Toms, A. S., Winter, W. Strongly self-absorbing C ∗-algebras, Trans. Amer. Math. Soc. 359 (2007),
3999-4029.
[11] Uuye, O. Homotopical algebra for C ∗-algebras, J. Noncommut. Geom. 7 (2013), 981-1006.
Moscow State University, Leninskie Gory, Moscow, 119991, Russia
E-mail address: [email protected]
|
1812.01042 | 3 | 1812 | 2019-12-11T14:59:21 | Tensor-product coaction functors | [
"math.OA"
] | For a discrete group $G$, we develop a `$G$-balanced tensor product' of two coactions $(A,\delta)$ and $(B,\epsilon)$, which takes place on a certain subalgebra of the maximal tensor product $A\otimes_{\max} B$. Our motivation for this is that we are able to prove that given two actions of $G$, the dual coaction on the crossed product of the maximal-tensor-product action is isomorphic to the $G$-balanced tensor product of the dual coactions. In turn, our motivation for this is to give an analogue, for coaction functors, of a crossed-product functor originated by Baum, Guentner, and Willett, and further developed by Buss, Echterhoff, and Willett, that involves tensoring an action with a fixed action $(C,\gamma)$, then forming the image inside the crossed product of the maximal-tensor-product action. We prove that composing our tensor-product coaction functor with the full crossed product of an action reproduces the tensor-crossed-product functor of Baum, Guentner, and Willett. We prove that every such tensor-product coaction functor is exact, thereby recovering the analogous result for the tensor-crossed-product functors of Baum, Guentner, and Willett. When $(C,\gamma)$ is the action by translation on $\ell^\infty(G)$, we prove that the associated tensor-product coaction functor is minimal, generalizing the analogous result of Buss, Echterhoff, and Willett for tensor-crossed-product functors. | math.OA | math |
TENSOR-PRODUCT COACTION FUNCTORS
S. KALISZEWSKI, MAGNUS B. LANDSTAD, AND JOHN QUIGG
Dedicated to the memory of J. M. G. Fell, 1923 -- 2016.
Abstract. Recent work by Baum, Guentner, and Willett, and
further developed by Buss, Echterhoff, and Willett introduced a
crossed-product functor that involves tensoring an action with a
fixed action (C, γ), then forming the image inside the crossed prod-
uct of the maximal-tensor-product action. For discrete groups, we
give an analogue, for coaction functors. We prove that composing
our tensor-product coaction functor with the full crossed product
of an action reproduces their tensor-crossed-product functor. We
prove that every such tensor-product coaction functor is exact and
if (C, γ) is the action by translation on ℓ∞(G), we prove that the
associated tensor-product coaction functor is minimal; thereby re-
covering the analogous result by the above authors. Finally, we
discuss the connection with the E-ization functor we defined ear-
lier, where E is a large ideal of B(G).
1. Introduction
For a fixed locally compact group G, the full and the reduced crossed-
product functors each take an action of G on a C ∗-algebra and produce
a C ∗-algebra. Baum, Guentner, and Willett [BGW16] studied exotic
crossed-product functors that are intermediate between the full and re-
duced crossed products, as part of an investigation of the Baum-Connes
conjecture. In Section 5 of that paper, the authors introduced a natu-
ral class of crossed products arising from tensoring with a fixed action.
Their general construction starts with an arbitrary crossed-product
functor, but we only need the version for the full crossed product.
They prove that their tensor-crossed-product functor is exact. Buss,
Echterhoff, and Willett [BEW18] further the study of these tensor-
crossed-product functors, and in Section 9 of that paper they prove
Date: December 10, 2019.
2000 Mathematics Subject Classification. Primary 46L55; Secondary 46M15.
Key words and phrases. Crossed product, action, coaction, tensor product, Fell
bundle.
1
2
KALISZEWSKI, LANDSTAD, AND QUIGG
that the case with ℓ∞(G) produces the smallest of all tensor-crossed-
product functors. This leads them to ask whether tensoring with ℓ∞(G)
in fact produces the minimal exact correspondence crossed product.
Thus the ℓ∞(G)-tensor-crossed-product functor takes on substan-
tial importance. We have initiated in [KLQ16, KLQ18] a new ap-
proach to exotic crossed products, applying a coaction functor to the
full crossed product. We have shown that this procedure reproduces
many (perhaps all of the important?) crossed-product functors, and
we believe that fully utilizing the coactions makes for a more robust
theory. In [KLQ16, KLQ18] we have shown that the theory of coaction
functors is in numerous aspects parallel to that of the crossed-product
functors of [BGW16, BEW18]. In this paper, using the techniques of
[BGW16, BEW18] as a guide, we initiate an investigation into an ana-
logue for coaction functors of the tensor-crossed-product functors for
actions (see Section 3 for details). Our development must of course have
many differences from that of crossed products by actions, since coac-
tions are different from actions, and also, according to our paradigm for
crossed-product functors, our coaction functors form the second part
of such a crossed product.
To give a more precise overview of our tensor-coaction functors,
we first outline (with slightly modified notation), the construction of
tensor-crossed-products from [BGW16, BEW18]. For technical rea-
sons, our techniques currently only apply to discrete groups, so from
now on we suppose that the group G is discrete.1 Fix an action (C, γ)
of G. Both papers [BGW16] and [BEW18] require C to be unital. For
every action (B, α) of G, first form the diagonal action α⊗γ of G on the
maximal tensor product B ⊗max C. The embedding b 7→ b⊗1 from B to
B ⊗max C is G-equivariant, and its crossed product is a homomorphism
from B ⋊α G to (B ⊗max C) ⊗α⊗γ G. The C-crossed product B ⋊α,C G is
the image of B ⋊α G in (B ⊗max C) ⋊α⊗γ G under this crossed-product
homomorphism. We want an analogue of this construction for coaction
functors. Our previous work indicates that there should be a coaction
on B ⋊α,C G that is the result of applying a coaction functor to the dual
coaction (B ⋊α G, bα), and presumably this should involve the fixed dual
coaction (C ⋊γ G,bγ). Abstractly, we are led to search for a coaction
functor formed by somehow combining a coaction (A, δ) with a fixed
coaction (D, ζ), with D unital, to form a coaction (AD, δD) in such a
manner that if the two coactions are (B ⋊α G, bα) and (C ⋊γ G,bγ) then
1It is certainly a draw-back of our techniques in this paper that we only handle
discrete groups. It is imperative to find some way to extend all this to arbitrary
locally compact G, and we will investigate this in future research.
TENSOR-PRODUCT COACTION FUNCTORS
3
(B ⋊ G)C⋊G is the natural image of B ⋊α G in (B ⊗max C) ⋊α⊗γ G, and
(bα)C⋊G is the restriction of the dual coaction \α ⊗ γ. Since we require
C to be unital, the crossed product C ⋊γ G is unital too, so we incur
no penalty by supposing that D is unital as well.
We accomplish our goal via a "G-balanced Fell bundle" A ⊗G D,
whose cross-sectional C ∗-algebra embeds faithfully in the maximal ten-
sor product A ⊗max D.
In Section 2 we record our notation and terminology for coactions,
Fell bundles, and coaction functors.
Sections 3 -- 5 contain our main results, and begin, as we mentioned
above, by proving in Theorem 3.2 the existence of the tensor D coaction
functor, for a fixed dual coaction (D, ζ). For a maximal coaction (A, δ),
we define an equivariant homomorphism from A to the G-balanced ten-
sor product A⊗GD, and then for an arbitrary coaction we first compose
with maximalization. In Theorem 3.6 we prove that when we compose
with the full crossed product we recover the tensor-crossed-product
functors of [BGW16, BEW18]. We prove in Theorem 5.2 that the case
D = ℓ∞(G) ⋊ G, with ζ the dual of the translation action, gives the
smallest of these coaction functors. We point out that our methods are
in many cases drawn from those of [BGW16, BEW18], but we mod-
ified the proof of minimality -- [BEW18] chooses an arbitrary state
and temporarily uses completely positive maps as opposed to homo-
morphisms, and we managed to avoid the need for these techniques.
Before that, we prove a general lemma involving embeddings into ex-
act functors, from which in Theorem 4.2 we deduce that all tensor D
coaction functors are exact.
We close in Section 6 with a few concluding remarks. First of all,
we acknowledge that our standing assumption that the group G is
discrete was heavily used, and we hope to generalize in future work to
arbitrary locally compact groups. We also mention that it is certainly
necessary to use a mixture of Fell bundles and coactions -- Fell bundles
by themselves are insufficient for our purposes. We then describe a
tantalizing connection with the coaction functors determined by large
ideals E of the Fourier-Stieltjes algebra.
We added a very short appendix containing a Fell-bundle version of
Lemma 4.1, which could be proved using the lemma and which would
lead to a quick proof of Theorem 4.2. However, we felt that it would
interrupt the flow too much to actually use Proposition A.1 in the
main development of Section 4, and it would in fact have lengthened
the exposition.
4
KALISZEWSKI, LANDSTAD, AND QUIGG
2. Preliminaries
Throughout, G will be a discrete group, with identity element de-
noted by e.
We refer to [EKQR06, Appendix A] and [EKQ04] for background
material on coactions, and to [KLQ16, KLQ18] for coaction functors.
For an action (A, α) of G we use the following notation:
A, iα
• (iα
G) is the universal representation of (A, α) in M(A ⋊α G);
this is abbreviated to (iA, iG) when the action α is clear from
context.
• bα is the dual coaction of G on A ⋊α G.
Fell bundles. We work as much as possible in the context of Fell bun-
dles over G, and the primary references are [FD88, Exe17, Qui96]. The
canonical Fell bundle over G is the line bundle C×G, whose C ∗-algebra
is naturally isomorphic to C ∗(G). If A = {As}s∈G and B = {Bs}s∈G
are Fell bundles over G, we say a map φ : A → B is a homomorphism
if it preserves all the structure (in particular, multiplication and invo-
lution).
We call an operation-preserving map π from a Fell bundle A into a
C ∗-algebra B a representation of A in B. We call a representation φ
nondegenerate if φ(Ae)B = B. We write C ∗(A) for the cross-sectional
C ∗-algebra of a Fell bundle A, and iA : A → C ∗(A) for the universal
representation, so that for every nondegenerate representation π : A →
B there is a unique (nondegenerate) homomorphism eπ : C ∗(A) → B,
which we call the integrated form of π, making the diagram
B
<①
A π
①
①
eπ
①
iA
①
C ∗(A)
commute. If φ : A → B is a Fell-bundle homomorphism, then iB ◦ φ is
a nondegenerate representation, so by the universal property there is a
unique C ∗-homomorphism C ∗(φ) making the diagram
A
iA
φ
B
iB
C ∗(A)
/ C ∗(B)
C ∗(φ)
commute. Thus, the assignment A 7→ C ∗(A) is functorial from Fell
bundles to the nondegenerate category of C ∗-algebras, in which a mor-
phism from A to B is a nondegenerate homomorphism π : A → M(B).
/
/
<
/
/
/
TENSOR-PRODUCT COACTION FUNCTORS
5
We frequently suppress the universal representation iA, and regard the
fibres As of the Fell bundle A as sitting inside C ∗(A), so that the pas-
sage from a representation to its integrated form can be regarded as
extending from A to C ∗(A), and in fact we will frequently use the same
notation for both the representation and its integrated form.
Recall from [Ng96, Qui96] that for every Fell bundle A over G there
is a coaction δA : C ∗(A) → C ∗(A)⊗C ∗(G) (where unadorned ⊗ denotes
the minimal C ∗-tensor product) given by
δA(as) = as ⊗ s
for s ∈ G, as ∈ As,
and conversely for every coaction (A, δ) of G the spectral subspaces
As = {a ∈ A : δ(a) = a ⊗ s}
give a Fell bundle A = {As}s∈G. Moreover, if (B, ε) is another coaction,
with associated Fell bundle B, then a homomorphism φ : A → B is
δ − ε equivariant if and only if it restricts to a homomorphism A → B.
By [EKQ04, Proposition 4.2], the coaction δ is maximal if and only
if the integrated form of the inclusion representation As ֒→ A is an
isomorphism C ∗(A) ≃ A. Thus, for maximal coactions (A, δ) we can
define a homomorphism from A to another C ∗-algebra B simply by
giving a representation of the Fell bundle A in B.
Remark 2.1. We will need the following result [AEK13, Corollary 6.3]:
if A = {As}s∈G is a Fell bundle over G and H is a subgroup of G,
then the canonical map C ∗(AH) → C ∗(A) is injective (where AH =
{Ah}h∈H is the restriction to a Fell bundle over H).
3. Tensor D functors
We will be particularly interested in the case of a homomorphism
from the canonical bundle C × G to another Fell bundle D = {Ds}s∈G,
and we will just say that we have a homomorphism V : G → D. Note
that this will require the unit fibre C ∗-algebra De to be unital, and the
elements Vs for s ∈ G will have to be unitary.
Given Fell bundles A, D over G, with cross-sectional algebras A =
C ∗(A) and D = C ∗(D), we form a new Fell bundle A ⊗G D over G as
follows: the fibre over s ∈ G is the closure in A ⊗max D of the algebraic
tensor product As ⊙ Ds, and we write this fibre as As ⊗max Ds. We
write
A ⊗G D = C ∗(A ⊗G D).
We then define a Fell-bundle homomorphism
φA : A → A ⊗G D
6
by
KALISZEWSKI, LANDSTAD, AND QUIGG
Then the image φA(A) is a Fell subbundle
φA(as) = as ⊗ Vs
for as ∈ As.
AD = {As ⊗ Vs}s∈G
of A ⊗G D. Applying the C ∗-functor gives a homomorphism
QA = C ∗(φA) : A → A ⊗G D,
and we write
AD = QA(A).
Occasionally, if A is understood we will just write Q for QA. On the
other hand, if D, and so D, is ambiguous, we write QD
A . There is a
subtlety: although the fibres As ⊗max Ds give a linearly independent
family of Banach subspaces of A⊗max D with dense linear span, making
A ⊗max D a graded C ∗-algebra over G in the sense of Exel [Exe17,
Definition 6.2], it is not a priori obvious that the the inclusion map
A ⊗G D ֒→ A ⊗max D gives a faithful embedding of the Fell-bundle
C ∗-algebra A ⊗G D in A ⊗max D. The following theorem establishes
this fact.
Theorem 3.1. If π : A ⊗G D → A ⊗max D is the representation given
by inclusions of the subspaces As ⊗max Ds, the integrated form
is injective.
eπ : A ⊗G D → A ⊗max D
Proof. First, consider the Fell bundle
A ⊗max D = {As ⊗max Dt}(s,t)∈G×G,
where, similarly to the definition of A ⊗G D, we define As ⊗max Dt as
the closure of the algebraic tensor product As ⊙ Dt in A ⊗max D. By
[AV, Proposition 4.6] the integrated form of the representation of the
Fell bundle A ⊗max D in A ⊗max D given by the inclusions
As ⊗max Dt ֒→ A ⊗max D
is an injective homomorphism
C ∗(A ⊗max D) → A ⊗max D.
In view of this, we identify C ∗(A ⊗max D) = A ⊗max D.
Now, the diagonal subgroup
∆ = {(s, s) : s ∈ G}
of G×G is isomorphic to G in the obvious way, and thus the restriction
(A ⊗max D)∆ = {As ⊗max Ds}s∈G
TENSOR-PRODUCT COACTION FUNCTORS
7
of the Fell bundle A ⊗max D to ∆ is canonically isomorphic to our
Fell bundle A ⊗G D over G. By [AEK13, Corollary 6.3] (which we
mentioned in Remark 2.1) the canonical map
C ∗(cid:0)(A ⊗max D)∆(cid:1) → C ∗(A ⊗max D)
is injective, and it follows by the above isomorphism that eπ is injective
also.
(cid:3)
In view of Theorem 3.1 we can identify A⊗GD with the C ∗-subalgebra
of A⊗max D given by the closed span of the subspaces {As ⊗max Ds}s∈G.
Note that the homomorphism QA : A → A ⊗max D is nondegenerate.
For any Fell bundle A, let δ = δA be the canonical coaction of G on
A = C ∗(A). By functoriality, QA is δA − δA⊗G D equivariant, and hence
is equivariant for δ and a unique coaction δD on the image AD.
Theorem 3.2. There is a functor σD from the category of maximal
coactions to the category of all coactions, defined as follows:
(i) On objects: (A, δ) 7→ (AD, δD).
(ii) On morphisms: given maximal coactions (A, δ) and (B, ε) and
a δ − ε equivariant homomorphism φ : A → B, let A and B
be Fell bundles such that A = C ∗(A) and B = C ∗(B), let
ψ : A → B be the unique Fell-bundle homomorphism such that
φ = C ∗(ψ), and define φD by the commutative diagram
A
QA
AD
φ
φD
/❴❴❴❴❴❴❴❴
B
QB
BD
A ⊗G D
C ∗(ψ⊗Gid)
/ B ⊗G D,
where it is clear that C ∗(ψ ⊗G id) maps AD into BD.
Moreover, for any maximal coaction (A, δ) we have
ker QA ⊆ ker Λ,
where Λ : A → An is the normalization.
Proof. We obviously have a functor A 7→ A ⊗max D on the category of
C ∗-algebras. As discussed above, if δ is a maximal coaction on A, then
we have a homomorphism QA : A → A ⊗max D taking as to as ⊗ Vs. If
φ : A → B is equivariant for maximal coactions δ and ε, then obviously
/
/
/
_
_
/
8
KALISZEWSKI, LANDSTAD, AND QUIGG
we get a commutative diagram
φ
A
QA
B
QB
A ⊗max D
/ B ⊗max D,
φ⊗maxid
(φ ⊗max id)(AD) ⊆ BD,
so
and hence the restriction gives a homomorphism φD making the dia-
gram
φ
A
B
QA
QB
AD
/ BD
φD
commute. Moreover, by considering elements of spectral subspaces
it is obvious that φD is δD − εD equivariant. Since A 7→ A ⊗max D is
functorial, it follows that we now have a functor A 7→ AD from maximal
coactions to coactions.
We turn to the inclusion ker QA ⊆ ker Λ. The composition δD ◦ QA
maps A into (A ⊗max D) ⊗ C ∗(G). Composing with the homomorphism
Υ ⊗ λ : (A ⊗max D) ⊗ C ∗(G) → A ⊗ D ⊗ C ∗
r (G),
where Υ is the canonical surjection
A ⊗max D → A ⊗ D,
we get a homomorphism
(3.1)
(Υ ⊗ λ) ◦ δD ◦ QA : A → A ⊗ D ⊗ C ∗
r (G)
which takes any element as ∈ As to
(3.2)
as ⊗ Vs ⊗ λs.
Representing faithfully on Hilbert space, we can apply Fell's absorption
trick to the representation V ⊗ λ to construct an endomorphism τ of
A ⊗ D ⊗ C ∗
r (G) that takes any element of the form (3.2) to
as ⊗ 1D ⊗ λs.
Then
τ ◦ (Υ ⊗ λ) ◦ δD ◦ QA : A → A ⊗ 1D ⊗ C ∗
r (G)
takes any element as ∈ As to
as ⊗ 1 ⊗ λs.
/
/
/
/
/
/
TENSOR-PRODUCT COACTION FUNCTORS
9
Then composing with the obvious isomorphism
θ : A ⊗ 1D ⊗ C ∗
r (G)
we get a homomorphism
≃
−→ A ⊗ C ∗
r (G),
θ ◦ τ ◦ (Υ ⊗ λ) ◦ δD ◦ QA : A → A ⊗ C ∗
r (G)
taking any element as ∈ As to as ⊗ λs.
On the other hand, Λ = (id ⊗ λ) ◦ δ, and for any element as ∈ As we
have
Thus θ ◦ τ ◦ (Υ ⊗ λ) ◦ δD ◦ QA = Λ, so
(id ⊗ λ) ◦ δ(as) = as ⊗ λs.
ker QA ⊆ ker Λ,
completing the proof.
(cid:3)
The second part of Theorem 3.2 justifies the following:
Definition 3.3. We define a coaction functor τ D on the category of
coactions by
τ D = σD ◦ (maximalization).
Proposition 3.4. Let (B, α) and (C, γ) be two actions of G. Then the
map
φ : (B × G) ⊗G (C × G) → (B ⊗max C) × G
defined by
φ(cid:0)(b, s) ⊗ (c, s)(cid:1) = (b ⊗ c, s)
is a Fell-bundle homomorphism, and consequently
C ∗(φ) : (B ⋊α G) ⊗G (C ⋊γ G) → (B ⊗max C) ⋊α⊗γ G
is a C ∗-isomorphism.
Proof. The first statement is easily verified, and for the second we pro-
duce an inverse: define
π : B ⊗max C → (B ⋊α G) ⊗G (C ⋊γ G)
as the unique homomorphism associated to the commuting homomor-
phisms iB ⊗max 1 and 1 ⊗max iC, and define a unitary homomorphism
U : G → M(cid:0)(B ⋊α G) ⊗G (C ⋊γ G)(cid:1)
by
Us = iα
G(s) ⊗max iγ
G(s).
Routine computations show that (π, U) is a covariant representation of
the action (B ⊗max C, α ⊗max γ), so its integrated form gives a homo-
morphism
Π = π × U : (B ⊗max C) ⋊α⊗γ G → (B ⋊α G) ⊗G (C ⋊γ G).
10
KALISZEWSKI, LANDSTAD, AND QUIGG
One checks without pain, using the identity
Π(b ⊗ c, s) = (cid:0)(b, s) ⊗ (c, s)(cid:1)
for all b ∈ B, c ∈ C, s ∈ G,
that Π ◦ C ∗(φ) is the identity on the Fell bundle (B × G) ⊗G (C × G),
and that C ∗(φ) ◦ Π is the identity on generators (b ⊗ c, s) of the Fell
bundle (B ⊗max C) × G, and hence on the entire Fell bundle. Thus the
associated C ∗-homomorphisms give inverse isomorphisms, finishing the
proof.
(cid:3)
Remark 3.5. Although we will not need it here, we point out that
the technique used in the above proof can also be used to show that
for actions (B, G, α) and (C, K, γ),
(B × G) ⊗max (C × K) ≃ (B ⊗max C) × (G × K),
which in turn implies
(B ⋊α G) ⊗max (C ⋊γ K) ≃ (B ⊗max C) ⋊α⊗γ (G × K).
Theorem 3.6. Let (C, γ) be an action of G, with C unital, and let
D be the associated semidirect-product Fell bundle, with D = C ∗(D) =
C ⋊γ G. Then
is naturally isomorphic to the C-crossed product functor
τ D ◦ (crossed product)
(B, α) 7→ B ⋊α,C G.
Proof. Let (B, α) be an action, and define ψ : B → B ⊗max C by b 7→
b⊗1. In the notation of Proposition 3.4, we will show that the diagram
B ⋊α G
Q
(B ⋊α G)D
+❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲
ψ⋊G
/❴❴❴❴❴❴❴❴❴❴❴❴
≃
θB
B ⋊α,C G
C ∗((B × G) ⊗G (C × G))
≃
C ∗(φ)
/ C ∗((B ⊗max C) × G)
commutes at the perimeter and that the bottom isomorphism takes
(B ⋊α G)D onto B ⋊α,C G, giving the desired isomorphism θB. For the
first, it suffices to compute on the Fell bundle B × G:
C ∗(φ) ◦ Q(b, s) = C ∗(φ)(cid:0)(b, s) ⊗ (1, s)(cid:1)
= (b ⊗ 1, s)
= (ψ ⋊ G)(b, s).
+
_
/
_
/
TENSOR-PRODUCT COACTION FUNCTORS
11
This computation also makes it clear that C ∗(φ) maps (B ⋊α G)D onto
B ⋊α,C G.
We still need to verify naturality: let π : (B, α) → (E, β) be a mor-
phism of actions. We must show that the diagram
(B ⋊α G)D
(π⋊G)D
(E ⋊β G)D
θB
θE
B ⋊α,C G
π⋊C G
/ E ⋊β,C G
commutes. Again, it suffices to compute on the Fell bundle (B × G)D:
(π ⋊C G) ◦ θB(cid:0)(b, s) ⊗ (1, s)(cid:1) = (π ⋊C G)(b ⊗ 1, s)
= (π(b) ⊗ 1, s)
= θE(cid:0)(π(b), s) ⊗ (1, s)(cid:1)
= θE ◦ (π ⋊ G)D(cid:0)(b, s) ⊗ (1, s)(cid:1). (cid:3)
4. Exactness
We now want to show that the tensor D functor is exact. We separate
out the following abstract lemma because we feel that it might be useful
in other similar situations. Actually, we suspect that it is folklore, and
we include the proof only for completeness.
Lemma 4.1. Let
0
0
/ I
η
/ J
ψ
φ
A
ζ
B
ω
/ C ρ
π
/ D
0
/ 0
be a commutative diagram of C ∗-algebras and homomorphisms. Sup-
pose that the bottom row is exact, φ(I) is an ideal of A, ψ is surjective,
and the vertical maps are nondegenerate injections. Then the top row
is exact.
Proof. The top row is exact at B. Also, φ is injective because π ◦ η is,
so the top row is exact at I; we must show that it is exact at A. Since
ω ◦ ψ ◦ φ = ρ ◦ π ◦ η = 0 and ω is injective we have ψ ◦ φ = 0. It remains
to show that ker ψ ⊆ φ(I). Let a ∈ ker ψ. Then by commutativity
ζ(a) ∈ ker ρ, so by exactness there is c ∈ J such that ζ(a) = π(c).
Choose an approximate identity (ei) for I. Then by nondegeneracy
(η(ei)) is an approximate identity for J. We have
ζ(a) = π(c)
/
/
/
/
/
/
/
/
/
/
/
/
/
/
12
KALISZEWSKI, LANDSTAD, AND QUIGG
= lim
i
= lim
i
π(cid:0)η(ei)c(cid:1)
π ◦ η(ei)π(c)
ζ(cid:0)φ(ei)a(cid:1)
i
= lim
∈ ζ(cid:0)φ(I)(cid:1),
because φ(I) is an ideal of A. Since ζ is an injective homomorphism
between C ∗-algebras, we get a ∈ φ(I), as desired.
(cid:3)
Theorem 4.2. Every tensor D functor τ D is exact.
Proof. Since maximalization is an exact functor, it suffices to show that
if
0
/ I
φ
/ A
ψ
/ B
/ 0
is a short exact sequence of C ∗-algebras carrying compatible maximal
coactions of G, then the image under τ D is also exact (we don't need
notation for the coactions, since they will take care of themselves in
this proof). We apply Lemma 4.1 to the diagram
0
0
/ I D
φD
AD
ψD
BD
/ I ⊗max D
/ A ⊗max D
φ⊗maxid
ψ⊗maxid
/ B ⊗max D
0
/ 0.
Properties of the maximal tensor product guarantee that the bottom
row is exact, and we have noted that the vertical inclusion maps are
nondegenerate.
We have a commutative diagram
I
QI
φ
A
QA
I D
/ AD.
φD
Since QI and QA are surjective, φD(I D) = QA(φ(I)) is an ideal of AD.
On the other hand, if ψ : A → B is a surjection that is equivariant
for maximal coactions, then the commutative diagram
ψ
A
B
QA
QB
AD
/ BD
ψD
/
/
/
/
/
/
/
_
/
/
_
/
/
_
/
/
/
/
/
/
/
/
/
/
TENSOR-PRODUCT COACTION FUNCTORS
13
shows that ψD is surjective since QB ◦ ψ is.
Thus the hypotheses (i) -- (iv) of Lemma 4.1 are satisfied, so the con-
(cid:3)
clusion follows.
Remark 4.3. In [KLQ16, Theorem 4.12] we gave necessary and suf-
ficient conditions for a coaction functor to be exact, expressing it as a
quotient of the functor maximalization, which is exact. However, for
the functor τ D it turned out to be easier to use Lemma 4.1, which was
inspired by [BGW16, proof of Lemma 5.4].
5. Minimal tensor D functor
Recall from [KLQ16, Definition 4.7, Lemma 4.8] that if σ and τ are
coaction functors then τ ≤ σ means that for every coaction (A, δ) there
is a homomorphism Γ making the diagram
A
qσ
③③③③③③③③
A
qτ
!❉❉❉❉❉❉❉❉
Am
Aσ
/❴❴❴❴❴❴❴❴
Γ
Aτ
commute. If S is a family of coaction functors, and τ is an element of
S, we say that τ is the smallest element of S if τ ≤ σ for all σ ∈ S.
The above partial ordering of coaction functors is compatible with
the partial ordering of crossed-product functors (see [BGW16, p. 8])
in the sense that if ρ and µ are crossed-product functors associated to
coaction functors τ and σ, respectively, then τ ≤ σ implies ρ ≤ µ.
[BEW18, Lemma 9.1] shows that the smallest of the C-crossed-
product functors is for (C, γ) = (ℓ∞(G), lt) (when G is discrete). For
our purposes it will be more convenient to use right translation, so
we replace lt by rt, which obviously causes no harm. The tensor D
functor with (D, ζ, V ) = (ℓ∞(G) ⋊rt G, brt, iG) reproduces the ℓ∞(G)-
crossed product upon composing with full crossed product, so clearly
we should expect that the tensor ℓ∞(G) ⋊rt G coaction functor is the
smallest among all tensor D functors. We verify this in Theorem 5.2
below.
Lemma 5.1. Let (C, γ) be an action of G. Define a homomorphism
ψ : C → C ⊗max ℓ∞(G) = ℓ∞(G, C) by
ψ(c)(s) = γs(c).
Then ψ is γ − (id ⊗ rt) equivariant, and the crossed-product homomor-
phism
Ψ : C ⋊γ G → (C ⊗max ℓ∞(G)) ⋊γ⊗rt G = C ⊗max (ℓ∞(G) ⋊rt G)
!
/
14
KALISZEWSKI, LANDSTAD, AND QUIGG
satisfies
Proof. Folklore.
Ψ(iγ
G(s)) = 1 ⊗ irt
G(s)
for s ∈ G.
(cid:3)
Theorem 5.2. Let R be the semidirect-product Fell bundle of the ac-
tion (ℓ∞(G), rt), and let R = C ∗(R) = ℓ∞(G) ⋊rt G, with unitary
homomorphism W = irt
G : G → R. Then τ R is the smallest among all
tensor D functors.
Proof. Let V : G → D be a homomorphism to a Fell bundle, and let
D = C ∗(D). Since by definition the coaction functors τ D and τ R are
formed by first maximalizing and then applying the surjections QD and
QR, by [KLQ16, Lemma 4.8] it suffices to show that for every maximal
coaction (A, δ) there is a homomorphism Γ making the diagram
(5.1)
A
QD
~⑤⑤⑤⑤⑤⑤⑤
QR
❆❆❆❆❆❆❆
AD
Γ
/❴❴❴❴❴❴❴
AR
commute. By Landstad duality, we can assume that D is a semidirect-
product Fell bundle associated to an action (C, γ), and V = iG, so that
D = C ⋊γ G. By Lemma 5.1 we have a homomorphism
such that
Ψ : D → C ⊗max R
Ψ(Vs) = 1 ⊗ Ws.
This gives a homomorphism
id ⊗max Ψ : A ⊗max D → A ⊗max C ⊗max R
taking as ⊗ Vs to as ⊗ 1 ⊗ Ws. Thus the composition
Φ = (id ⊗max Ψ) ◦ QD : A → A ⊗max C ⊗max R
has image in A ⊗max 1 ⊗max R, and so id ⊗max Ψ maps AD into A ⊗max
1 ⊗max R. Using the obvious isomorphism
θ : A ⊗max 1 ⊗max R
≃
−→ A ⊗max R,
we see for s ∈ G and as ∈ As,
θ ◦ (id ⊗max Ψ) ◦ QD(as) = θ ◦ (id ⊗max Ψ)(as ⊗ Vs)
= θ(as ⊗ 1 ⊗ Ws)
= as ⊗ Ws
= QR(as),
~
/
TENSOR-PRODUCT COACTION FUNCTORS
so we can take
Γ = θ ◦ (id ⊗max Ψ)AD.
15
(cid:3)
The last part of the above proof is very similar to an argument in
the proof of Theorem 3.2.
6. Concluding remarks
Throughout, we have taken advantage of our standing assumption
that the group G is discrete. In particular, this allowed us to do almost
everything with Fell bundles. In future research we will investigate the
case of arbitrary locally compact G.
Remark 6.1. It would not be useful to try to do everything in the
context of functors from Fell bundles to Fell bundles, because in our
most important construction
A 7→ A ⊗max D
the image Fell bundle AD is isomorphic to A. We must ultimately take
the target of the functor to be a C ∗-algebra to get anything of interest.
For an equivariant maximal coaction (D, ζ, V ), we will now show
how the tensor D coaction functor τ D is tantalizingly close to a functor
coming from a large ideal E of the Fourier-Stieltjes algebra B(G) (see
[KLQ16]).
Recall that a large ideal E is the annihilator of an ideal I of C ∗(G)
that is δG-invariant and contained in the kernel of the regular repre-
sentation, where invariance means that the quotient map
qE : C ∗(G) → C ∗
E(G) = C ∗(G)/I
takes δG to a coaction on C ∗
E(G). For any maximal coaction (A, δ) we
let AE be the quotient of A by the kernel of the composition (id⊗qE)◦δ.
Then the quotient map QE = QE
A : A → AE is equivariant for δ and
a coaction δE, and moreover the assignments (A, δ) 7→ (AE, δE) give a
functor from maximal coactions to all coactions. Composing with the
maximalization functor gives a coaction functor that we call E-ization.
Apply this to the ideal I = ker V , where we also write V : C ∗(G) →
D for the integrated form of the unitary homomorphism V : G → D.
The annihilator E = I ⊥ is a large ideal of B(G) since V is δD − ζ
invariant and nonzero. A cursory glance at the situation might lead one
to ask, "Are the tensor D functor and E-ization naturally isomorphic?"
One obvious obstruction is that (for maximal coactions) the tensor
D functor goes into a maximal tensor product A ⊗max D, while E-
ization goes into the minimal tensor product A ⊗ C ∗
E(G). We can make
16
KALISZEWSKI, LANDSTAD, AND QUIGG
a closer connection by modifying the coaction δ so that it becomes a
homomorphism δM that makes the diagram
δM
A
(◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗
δ
A ⊗max C ∗(G)
ψ
A ⊗ C ∗(G)
commute, where ψ is the canonical surjection of the maximal tensor
product onto the minimal one, and satisfies the other axioms for a
coaction.
Here is a commutative diagram illustrating how the various maps
are related:
A
QD
δM
A ⊗max C ∗(G)
id⊗V
A ⊗max D
id⊗maxι
A ⊗max V (C ∗(G)),
where ι : V (C ∗(G)) ֒→ D is the inclusion map. Since V (C ∗(G)) is nat-
urally isomorphic to C ∗
E(G), we see that the tensor D functor seems to
be closer to a version of "E-ization" but using the modified δM . How-
ever, there is yet another stumbling block: we do not know whether the
homomorphism id ⊗max ι is injective, due to the mysteries of maximal
tensor products. A bit more succinctly, we could view the composition
(id ⊗max V ) ◦ δM : A → A ⊗max V (C ∗(G))
(preceded by maximalization) as a sort of "maximalized version" of
E-ization, and then we could ask whether it is naturally isomorphic to
τ D.
Here is a particularly important special case:
Question 6.2. Is the minimal tensor D functor (the case D = R =
ℓ∞(G) ⋊G) isomorphic to a maximalized version of E-ization as above?
Appendix A. Exactness of Fell Bundle Functors
Although we do not need it, we mention here how the abstract
Lemma 4.1 could be used to deduce a corresponding exactness result
for Fell-bundle functors, quite similarly to how we proved Theorem 4.2.
If σ is a functor from Fell bundles over G to C ∗-algebras, in this
appendix we will write Aσ for the image under σ of a Fell bundle A,
and φσ for the image under σ of a homomorphism φ : A → B. Recall
/
/
(
/
/
o
o
TENSOR-PRODUCT COACTION FUNCTORS
17
from [Exe17, Definition 21.10] that a Fell bundle I that is a subbundle
of a Fell bundle A is called an ideal of A if
IsAt ⊆ Ist
and AtIs ⊆ Its
for all s, t ∈ G.
The following could be proved similarly to Theorem 4.2.
Proposition A.1. Let σ and ρ be two functors from Fell bundles to
C ∗-algebras, Assume that:
• for every short exact sequence
0
/ I
φ
/ A
ψ
/ B
/ 0
of Fell bundles, φσ(I σ) is an ideal of Aσ and ψσ is surjective;
• there is a natural transformation η from σ to ρ such that for
e injectively
every Fell bundle A the homomorphism ηA maps Aσ
onto a nondegenerate subalgebra of Aρ
e;
• ρ is exact.
Then σ is exact.
In fact, Theorem 4.2 could be deduced almost immediately from
Proposition A.1, but we decided to avoid this approach.
References
[AV]
F. Abadie-Vicens, Tensor products of Fell bundles over discrete groups,
arXiv:funct-an/9712006.
[AEK13] P. Ara, R. Exel, and R. Katsura, Dynamical systems of type (m, n) and
their C∗-algebras, Ergodic Theory Dynam. Systems 33 (2013), no. 5,
1291 -- 1325.
[BGW16] P. Baum, E. Guentner, and R. Willett, Expanders, exact crossed prod-
ucts, and the Baum-Connes conjecture, Annals of K-Theory 1 (2016),
no. 2, 155 -- 208.
[BEW18] A. Buss, S. Echterhoff, and R. Willett, Exotic crossed products and the
Baum-Connes conjecture, J. Reine Angew. Math. 740 (2018), 111 -- 159.
[EKQ04] S. Echterhoff, S. Kaliszewski, and J. Quigg, Maximal coactions, Internat.
J. Math. 15 (2004), 47 -- 61.
[Exe17]
[EKQR06] S. Echterhoff, S. Kaliszewski, J. Quigg, and I. Raeburn, A Categorical
Approach to Imprimitivity Theorems for C*-Dynamical Systems, vol.
180, Mem. Amer. Math. Soc., no. 850, American Mathematical Society,
Providence, RI, 2006.
R. Exel, Partial dynamical systems, Fell bundles and applications, Math-
ematical Surveys and Monographs, vol. 224, American Mathematical
Society, Providence, RI, 2017.
J. M. G. Fell and R. S. Doran, Representations of ∗-algebras, locally com-
pact groups, and Banach ∗-algebraic bundles. Vol. 2, Pure and Applied
Mathematics, vol. 126, Academic Press Inc., Boston, MA, 1988.
[FD88]
[KLQ16] S. Kaliszewski, M. B. Landstad, and J. Quigg, Coaction functors, Pacific
J. Math. 284 (2016), no. 1, 147 -- 190.
/
/
/
/
18
KALISZEWSKI, LANDSTAD, AND QUIGG
[KLQ18]
, Coaction functors, II, Pacific J. Math. 293 (2018), no. 2, 301 --
[Ng96]
[Qui96]
339.
C.-K. Ng, Discrete coactions on C ∗-algebras, J. Austral. Math. Soc. Ser.
A 60 (1996), no. 1, 118 -- 127.
J. C. Quigg, Discrete C ∗-coactions and C ∗-algebraic bundles, J. Austral.
Math. Soc. Ser. A 60 (1996), 204 -- 221.
School of Mathematical and Statistical Sciences, Arizona State
University, Tempe, Arizona 85287
E-mail address: [email protected]
Department of Mathematical Sciences, Norwegian University of
Science and Technology, NO-7491 Trondheim, Norway
E-mail address: [email protected]
School of Mathematical and Statistical Sciences, Arizona State
University, Tempe, Arizona 85287
E-mail address: [email protected]
|
1308.2126 | 3 | 1308 | 2013-12-19T08:54:52 | Classification of certain continuous fields of Kirchberg algebras | [
"math.OA",
"math.KT"
] | We show that the K-theory cosheaf is a complete invariant for separable continuous fields with vanishing boundary maps over a finite-dimensional compact metrizable topological space whose fibers are stable Kirchberg algebras with rational K-theory groups satisfying the universal coefficient theorem. We provide a range result for fields in this class with finite-dimensional K-theory. There are versions of both results for unital continuous fields. | math.OA | math |
CLASSIFICATION OF CERTAIN CONTINUOUS FIELDS OF
KIRCHBERG ALGEBRAS
RASMUS BENTMANN
Abstract. We show that the K-theory cosheaf is a complete invariant for sep-
arable continuous fields with vanishing boundary maps over a finite-dimensional
compact metrizable topological space whose fibers are stable Kirchberg al-
gebras with rational K-theory groups satisfying the universal coefficient the-
orem. We provide a range result for fields in this class with finite-dimensional
K-theory. There are versions of both results for unital continuous fields.
1. Introduction
The present article is part of a programme aimed at deciding when two C ∗-alge-
bras over a (second countable) topological space X are equivalent in ideal-related
KK-theory. In consequence of a fundamental result due to Eberhard Kirchberg [20,
Folgerung 4.3], this is a central question in the classification theory of non-simple
purely infinite C ∗-algebras.
We briefly review the existing results in the literature. These are concerned with
two classes of spaces, namely finite (non-Hausdorff) spaces on the one hand and
finite-dimensional compact metrizable spaces on the other hand. Universal coeffi-
cient theorems (UCT), which compute the KK(X)-groups in terms of K-theoretic
invariants and which imply a solution to the given classification problem, have been
established for certain classes of finite T0-spaces in [28, 8, 26, 22, 3, 5]. A solution for
finite unique path spaces using a more complicated kind of invariant is provided
in [6]. For arbitrary finite spaces, the problem remains unsolved and seems rather
unfeasible because certain wildness phenomena occur; see [2].
In the context of
finite-dimensional compact metrizable spaces the strongest results are available in
the totally disconnected case [15, 14] and in the case of the unit interval [12, 13, 4].
As these examples illustrate, the feasibility of the classification problem under
consideration depends critically on the space X. However, it is possible to obtain
solutions for more general base spaces under additional K-theoretical assumptions.
For instance, Kirchberg's isomorphism theorem [20, Folgerung 2.18] states that
two separable nuclear stable O2-absorbing C ∗-algebras are isomorphic once their
primitive ideal spaces are homeomorphic (O2-absorption entails in particular the
vanishing of all K-theoretic data). In [1], a UCT for C ∗-algebras with vanishing
boundary maps (as we shall define in §3) over an arbitrary finite T0-space is proven.
The main result of the present article is the following; it is based on the UCT
in [1] and Dadarlat -- Meyer's approximation of ideal-related E-theory over an infinite
space X by ideal-related E-theory over finite quotient spaces of X from [14], together
with Kirchberg's theorem [20].
2010 Mathematics Subject Classification. 46L35, 46L80, 46M20, 19K35.
Key words and phrases. Classification of C ∗-algebras, Continuous fields, Kirchberg algebras,
K-theory.
The author was supported by the Danish National Research Foundation through the Centre
for Symmetry and Deformation (DNRF92) and by the Marie Curie Research Training Network
EU-NCG.
1
2
RASMUS BENTMANN
Theorem 1.1. Let A and B be separable continuous fields over a finite-dimensional
compact metrizable topological space X whose fibers are stable Kirchberg algebras
that satisfy the UCT and have rational K-theory groups. Assume that A and B have
vanishing boundary maps. Then any isomorphism of K-theory cosheaves OK(A) ∼=
OK(B) lifts to a C(X)-linear ∗-isomorphism A ∼= B.
The K-theory cosheaf OK is a rather simple (but large) K-theoretic invariant
which we shall define in §3; it comprises the K-theory groups of all (distinguished)
ideals of the algebra, together with the maps induced by all inclusions of such ideals.
The proof of this theorem is concluded in §4. We provide a version of the theorem
for unital algebras in Theorem 4.3. An Abelian group G is rational if it is isomorphic
to its tensor product with the field of rational numbers Q; this is equivalent to G
being torsion-free and divisible and to G being a vector space over Q.
Our method of proof is largely parallel to the one in [4], where the UCT from [5]
for C ∗-algebras over finite accordion spaces was used, based on the observation
that the unit inverval has sufficiently many finite quotients of accordion type. The
main result of the present article is instead valid for an arbitrary finite-dimensional
compact metrizable base space, but this comes at the expense of the assumption of
vanishing boundary maps.
Using a result from [18], Kirchberg's isomorphism theorem for O2-absorbing
C ∗-algebras mentioned above implies in particular that a separable continuous
C ∗-bundle over a finite-dimensional compact metrizable space X whose fibers are
stable UCT Kirchberg algebras with trivial K-theory groups is isomorphic to the
trivial C ∗-bundle C(X, O2 ⊗ K); see also [11]. Our classification result may be con-
sidered as an extension of this automatic triviality theorem for continuous O2 ⊗ K-
bundles: instead of asking the C ∗-bundles to have entirely trivial K-theory, we only
require the collection of the K-theory groups of all ideals in the algebra to form a
flabby cosheaf of Z/2-graded Q-vector spaces (this terminology is explained in §3).
In Section 5, we determine the range of the invariant in the classification res-
ult above under the additional assumption of finite-dimensional K-theory. More
precisely, we show:
Theorem 1.2. Let M be a flabby cosheaf of Z/2-graded Q-vector spaces on X such
that M (X) is finite-dimensional. Then M is a direct sum of a finite number of
skyscraper cosheaves and M ∼= OK(A) for a continuous field A as in Theorem 1.1.
This theorem also has an analogue for unital continuous fields. The range ques-
tion in the general case where M (X) may be countably infinite-dimensional remains
open. We provide an example to illustrate the greater complexity in this case.
Acknowledgement. The author is grateful to Marius Dadarlat for helpful conver-
sations on the topic of the present article. He thanks the referee for a number of
valuable suggestions.
2. Preparations
Throughout this article, we let X denote a finite-dimensional compact metrizable
topological space (arbitrary topological spaces will be denoted by Y ). The topology
of X (its lattice of open subsets) is denoted by O(X). We choose an ordered basis
(Un)n∈N for O(X) and consider the (finite) T0-quotient Xn of X equipped with the
topology O(Xn) generated by the family {U1, . . . , Un} (see [14, §3]).
Our reference for continuous fields of C ∗-algebras (or, synonymically, C ∗-bun-
dles) is [16]. For basic definitions, facts and notation concerning C ∗-algebras over
(possibly non-Hausdorff) topological spaces, we refer to [23]. Versions of KK-theory
and E-theory for separable C ∗-algebras over second countable topological spaces
CLASSIFICATION OF CERTAIN CONTINUOUS FIELDS
3
have been constructed in [23] and [14], respectively. By [14, Theorem 3.2], there is
a natural short exact sequence of Z/2-graded Abelian groups
(2.1)
lim
←−
1 E∗+1(Xn; A, B) E∗(X; A, B) ։ lim
←−
E∗(Xn; A, B)
for every pair A, B of separable C ∗-algebras over X.
Recall from [23, §3.2] that there is an exterior product for KK(X)-theory. In
particular, we can form the (minimal) tensor product of a C ∗-algebra A over X
with a C ∗-algebra D and obtain a C ∗-algebra A ⊗ D over X. We let MQ denote
the universal UHF-algebra. Hence K0(MQ) ∼= Q and K1(MQ) = 0. A C ∗-algebra B
(over X) is called MQ-absorbing if B ∼= B ⊗ MQ. If A and B are C ∗-algebras over X
and B is MQ-absorbing, then the exterior product
(2.2) KK∗(X; A, B) ⊗ KK∗(C, MQ) → KK∗(X; A ⊗ C, B ⊗ MQ) ∼= KK∗(X; A, B)
turns KK∗(X; A, B) into a rational vector space.
3. Vanishing boundary maps and flabby cosheaves
A C ∗-algebra A over an arbitrary topological space Y is said to have vanishing
boundary maps if the natural map iV
open subsets U ⊆ V ⊆ Y (it suffices to consider the case V = Y ); this is equivalent
to the condition in [1, Definition 3.2] because of the six-term exact sequence.
U : K∗(cid:0)A(U )(cid:1) → K∗(cid:0)A(V )(cid:1) is injective for all
If A has vanishing boundary maps, one can deduce from the Mayer -- Vietoris
sequence and continuity of K-theory that, for every covering (Vi)i∈I of an open
subset V ⊆ Y by open subsets Vi ⊆ V , one has an exact sequence
(3.1) Mj,k∈I
K∗(cid:0)A(Vj ∩ Vk)(cid:1)
see [9, Proposition 1.3].
(i
Vj
Vj ∩Vk
−i
Vk
Vj ∩Vk
)
−−−−−−−−−−−→Mi∈I
K∗(cid:0)A(Vi)(cid:1)
(iU
Vi
)
−−−→ K∗(cid:0)A(V )(cid:1) −→ 0;
Definition 3.2. The K-theory cosheaf OKY (A) of a C ∗-algebra A over Y with
vanishing boundary maps consists of the collection of Z/2-graded Abelian groups
(cid:0)K∗(cid:0)A(U )(cid:1) U ∈ O(Y )(cid:1) together with the collection (cid:0)iV
of graded group homomorphisms.
U V ∈ O(Y ), U ∈ O(V )(cid:1)
In Theorem 1.1 we briefly wrote OK(A) for OKX (A). For an arbitrary C ∗-alge-
bra A over Y , OKY (A) would only define a precosheaf, that is, a covariant functor
on O(X). However, if A has vanishing boundary maps, then by (3.1), OKY (A)
is indeed a flabby cosheaf of Z/2-graded Abelian groups in the technical sense
of [9, §1]. We reproduce the definition below for the reader's convenience. The
partially ordered set O(Y ) is considered as a category with morphisms given by
inclusions.
Definition 3.3. A precosheaf on Y is a covariant functor M from O(Y ) to the
category of modules over some ring. For open subsets U ⊆ V ⊆ Y , the induced
map M (U ) → M (V ) is denoted by iV
U . A precosheaf M is a cosheaf if the sequence
(3.4) Mj,k∈I
M (Vj ∩ Vk)
(i
Vj
Vj ∩Vk
−i
Vk
Vj ∩Vk
)
−−−−−−−−−−−→Mi∈I
M (Vi)
)
(iU
Vi
−−−→ M (V ) −→ 0
is exact for every open covering (Vi)i∈I of an open subset V ⊆ Y . It is flabby if the
map iV
U : M (U ) → M (V ) is injective for all open subset U ⊆ V ⊆ Y . A morphism
of cosheaves is a natural transformation of the corresponding functors.
4
RASMUS BENTMANN
Remark 3.5. The invariant OK is necessarily large (in a non-technical sense), as its
purpose is to classify certain C ∗-bundles that need not be locally trivial. However,
as we shall see in the proof of Theorem 1.1, we may minimize its size without losing
essential information by restricting to a fixed countable basis of O(X).
4. Proof of Theorem 1.1
We are now prepared to prove our main result. As in Theorem 1.1, we ab-
breviate OKX by OK. Assume that A and B as in Theorem 1.1 are given. By
[4, Propositions 2.8 and 2.10], A and B are stable, nuclear and O∞-absorbing, and
belong to the E(X)-theoretic bootstrap class BE(X) defined in [14, Definition 4.1].
Hence, by the comparison theorem [14, Theorem 5.4], Kirchberg's classification
theorem [20, Folgerung 4.3] and the invertibility criterion [14, Theorem 4.6], it suf-
fices to show that a given homomorphism OK(A) → OK(B) lifts to an element in
E0(X; A, B).
Consider now one of the approximating spaces Xn (this may have the homeo-
morphism type of any finite T0-space). We apply the machinery of homological al-
gebra in triangulated categories [24] to the category E(Xn)⊗Q whose objects are sep-
arable C ∗-algebras over Xn and whose morphism groups are given by E0(Xn, ␣, ␣)⊗
Q. This is indeed a triangulated category because E(Xn) is triangulated (see [14])
and Q is flat (compare [19]). In [1, §4 -- 5], we defined a homological functor XnK
on KK(Xn) and indicated that it is universal with respect to the homological ideal
given by the kernel of XnK on morphisms. Similarly, we have a homological functor
XnK ⊗ Q on E(Xn) ⊗ Q that is universal with respect to its kernel. Since we are
working over the field Q, an argument as in [1, Lemma 4.6] shows that XnK(A) is
projective since A has vanishing boundary maps. From [24, Theorem 3.41], we
obtain an isomorphism E0(Xn; A, B) ⊗ Q ∼= Hom(cid:0)XnK(A) ⊗ Q, XnK(B) ⊗ Q(cid:1).
By [1, Lemma 4.3] and the natural isomorphisms Colim ◦ XnK ∼= OKXn and
Res ◦ OKXn ∼= XnK, we may replace XnK with OKXn and obtain the identific-
ation
(4.1)
E0(Xn; A, B) ⊗ Q ∼= Hom(cid:0)OKXn (A) ⊗ Q, OKXn (B) ⊗ Q(cid:1).
Since the fibers of A and B have rational K-theory groups, they are MQ-absorbing
by the Kirchberg -- Phillips classification theorem [27, §8.4]. By [18], the algebras A
and B themselves are MQ-absorbing. Hence, by (2.2), (4.1), the comparison the-
orem [14, Theorem 5.4] and the Künneth formula for tensor products,
(4.2)
because Q ⊗ Q ∼= Q. The continuity of K-theory with respect to inductive limits
E0(Xn; A, B) ∼= Hom(cid:0)OKXn (A), OKXn (B)(cid:1)
shows that the canonical map Hom(cid:0)OK(A), OK(B)(cid:1) → lim
is an isomorphism. Hence lim
←−
now follows from the exact sequence (2.1).
E0(Xn; A, B) ∼= Hom(cid:0)OK(A), OK(B)(cid:1). The claim
←−
Hom(cid:0)OKXn (A), OKXn (B)(cid:1)
4.1. Classification of unital C ∗-bundles. For a unital C ∗-bundle over X, we
may equip OK(A) with the unit class [1A] ∈ K0(A). This pair is denoted by
OK+(A); it is a pointed cosheaf, that is, a cosheaf M with a distinguished element
in the degree-zero part of M (X). Morphisms of such pointed cosheaves are of
course required to preserve the distinguished element. By [17, Theorem 3.3], we
immediately obtain the following version of our main result for unital algebras.
Theorem 4.3. Let A and B be separable unital continuous fields over X whose
fibers are UCT Kirchberg algebras with rational K-theory groups. Assume that
A and B have vanishing boundary maps. Then any isomorphism OK+(A) ∼=
OK+(B) lifts to a C(X)-linear ∗-isomorphism A ∼= B.
CLASSIFICATION OF CERTAIN CONTINUOUS FIELDS
5
5. Range results
We investigate the question of the range of the invariant in Theorem 4.3 (the
same considerations apply mutatis mutandis and without keeping track of the unit
class in the stable case and yield a proof for Theorem 1.2). For the results in this
section, it would suffice to assume that X is a compact metrizable space.
If a unital C ∗-bundle A of the form classified by our result is locally trivial on an
open subset U of X, then A(U ) must be isomorphic to C0(U, O2). Hence interesting
examples cannot be locally trivial (around every point in X).
Example 5.1. We will now describe some basic non-trivial examples of C ∗-bundles
satisfying the conditions in our classification theorem. Let D1, . . ., Dn be unital
UCT Kirchberg algebras. By the Exact Embedding Theorem [21, Theorem 2.8], we
may find unital ∗-monomorphisms γi : Di → O2. For points x1, . . ., xn in X, we
define
(5.2)
A = {f ∈ C(X, O2) f (xi) ∈ γi(Di) for i = 1, . . . , n}.
This is clearly a continuous field of Kirchberg algebras, with fiber Di at xi and
fiber O2 at all other points. A simple computation using excision shows that
K∗(cid:0)A(U )(cid:1) ∼= Mi : xi∈U
K∗(Di).
Hence OK(A) is the direct sum of so-called skyscraper cosheaves ixi(cid:0)K∗(Di)(cid:1) based
at xi with coefficient group K∗(Di). Here ix(G) is defined by
ix(G)(U ) =(G if x ∈ U ,
else.
0
These cosheaves are indeed flabby.
It follows that the continuous field A has
vanishing boundary maps. So, if the algebras Di have rational K-theory groups,
then A satisfies the conditions of Theorem 4.3. Under the identification K0(A) ∼=
i=1[1Di ]. Using the range result for K-theory on
unital UCT Kirchberg algebras [27, §4.3], it follows that an arbitrarily pointed
finite direct sum of skyscraper cosheaves whose coefficient groups are countable
Z/2-graded Abelian groups can be realized as the pointed K-theory cosheaf of a
unital continuous field as in (5.2).
Ln
i=1 K0(Di), we have [1A] = Pn
The following proposition shows that, if A is a unital continuous field as in
Theorem 4.3 and the Q-vector space K∗(A) is finite-dimensional, then A must be
of the form (5.2).
Proposition 5.3. Let F be a field and let Y be an arbitrary topological space. Let M
be a flabby cosheaf of F-vector spaces over Y . If M (Y ) is finite-dimensional, then
M is a direct sum of a finite number of skyscraper cosheaves.
Proof. We proceed by induction on the dimension of M (Y ). If the dimension is
zero, then there is nothing to prove. Otherwise, by [10, V. Proposition 1.5], there
exists y ∈ Y such that M (Y \ {y}) is a proper subspace of M (Y ). By assumption,
the subcosheaf N of M defined by
N (U ) = M (U \ {y})
for U ∈ O(Y ) is a direct sum of skyscraper cosheaves. Since the quotient Q = M/N
vanishes on Y \ {y}, it follows from the exact sequence (3.4) that Q is a skyscraper
cosheaf of the form Q = iy(V ) for some F-vector space V . It remains to show that
6
RASMUS BENTMANN
the extension N M ։ Q splits. We have Hom(iy(V ), N ) ∼= Hom(cid:0)V, lim
←−
U ∋x
and thus
N (U )(cid:1)
Ext1(Q, N ) ∼= Hom(cid:0)V, lim
←−
U ∋x
1
N (U )(cid:1) = 0
by the Mittag -- Leffler condition using that N (U ) is a finite-dimensional vector space
for every U .
(cid:3)
The considerations above are summarized in the following version of Theorem 1.2
for unital continuous fields:
Theorem 5.4. Let (M, m) be a pointed flabby cosheaf of Z/2-graded Q-vector
spaces on X such that M (X) is finite-dimensional. Then M is a direct sum of
a finite number of skyscraper cosheaves and (M, m) ∼= OK+(A) for a continuous
field A as in Theorem 4.3.
Combining the range result above with our classification results, we obtain an
explicit description of the isomorphism classes of the classified continuous fields A
whose K-theory K∗(A) is finite-dimensional over Q.
In the case that K∗(A) is
an arbitrary (countable) Q-vector space the situation is unclear: it remains open
whether a countable direct sum of skyscraper cosheaves whose coefficient groups
are countable Z/2-graded Q-vector spaces can be realized as the K-theory cosheaf
OK(A) of a continuous field A as in Theorem 1.1.
We generalize the previous example by replacing the finite set of singularities
with a totally disconnected subset. This demonstrates that the assumption of
finite-dimensionality in Proposition 5.3 cannot be dropped, that is, a flabby cosheaf
of vector spaces of countable dimension need not be a direct sum of skyscraper
cosheaves.
Example 5.5. Let Y ⊆ X be a closed, totally disconnected subset. Then C(Y )
is a tight C ∗-algebra over Y with vanishing boundary maps. This follows because
the C ∗-algebra C(Y ) as well as all of its ideals and quotients are AF-algebras and
thus have vanishing K1-groups. In fact, one readily sees that OK(cid:0)C(Y )(cid:1) ∼= Cc(␣, Z)
(compactly supported, locally constant functions).
We let A be the tight C ∗-algebra over Y given by C(Y ) ⊗ O∞ ⊗ MQ. This is
a unital continuous field of Kirchberg algebras over Y . By Blanchard's embedding
theorem [7], there is a unital ∗-monomorphism α : A ֒→ C(Y ) ⊗ O2 over Y . We
define
B = {f ∈ C(X) ⊗ O2 f Y ∈ α(A)}.
Excision by the obvious extension C0(X \ Y ) ⊗ O2 B ։ A together with the
Künneth formula shows that B has vanishing boundary maps. By construction,
B is a unital separable continuous field over X whose fibers are UCT Kirchberg
algebras with rational K-theory groups; more precisely, we have B(x) ∼= O∞ ⊗ MQ
for x ∈ Y and B(x) ∼= O2 for x 6∈ Y . Hence the restriction of B to any closed
subset that intersects both Y and X \ Y is nontrivial. The K-theory cosheaf of B
is given explicitly by
U 7→ Cc(U ∩ Y, Z) ⊗ Q.
This is a flabby cosheaf of rational vector spaces of countable dimension. The stalks
of OK(B) are given by Q at points in Y and by 0 at points in X \ Y . (The stalk
Mx of a cosheaf M on X at a point x ∈ X is the quotient M (X)/M (X \ {x}).)
I know no separable continuous field C over X with vanishing boundary maps such
that the set {x ∈ X OK(C)x 6= 0} is not zero-dimensional.
CLASSIFICATION OF CERTAIN CONTINUOUS FIELDS
7
6. Further remarks
6.1. Real rank zero. We briefly comment on the relationship of the assumptions
in our classification theorem to real rank zero, a property that has often proved
useful for classification purposes. It was shown in [25, Theorem 4.2] that a separable
purely infinite C ∗-algebra A has real rank zero if and only if the primitive ideal
space of A has a basis consisting of compact open subsets and A is K0-liftable
(meaning, in our terminology, that A has "vanishing exponential maps"). While a
C ∗-bundle (with non-vanishing fibers) over a compact metrizable space of positive
dimension cannot satisfy the first condition, the second condition of K0-liftability is
built into our assumptions (we also assume that A has "vanishing index maps"). As
Theorem 1.2 shows, at least in the case of finite-dimensional K-theory, the K-theory
cosheaf of a separable continuous field with vanishing boundary maps has a very
zero-dimensional flavour.
6.2. Cosheaves versus sheaves. The following explanations clarify the relation-
ship (in the setting of fields with vanishing boundary maps) between our K-theory
cosheaf and the K-theory sheaf defined in [12] for C ∗-bundles over the unit interval.
In [10, Propositions V.1.6 and V.1.8], Glen Bredon provides a structure result for
flabby cosheaves: the compact sections functor yields a one-to-one correspondence
between soft sheaves and flabby cosheaves on O(X). A sheaf is soft if sections over
closed subsets can be extended to global sections. If \OK(A) denotes the soft sheaf
corresponding to the flabby cosheaf OK(A), then we have \OK(A)(Z) ∼= K∗(cid:0)A(Z)(cid:1)
for every closed subset Z ⊆ X. Regarding the range question considered in §5, we
remark that [13, Theorem 5.8] provides a range result for unital C ∗-bundles over
the unit interval, but it is not clear when the constructed algebras have vanishing
boundary maps.
6.3. Another classification result. We conclude the note by stating one more
result which follows in essentially the same way as our main result. We comment
below on the required modifications in the proof. Again, a version for unital algebras
can be obtained from [17, Theorem 3.3].
Theorem 6.1. Fix i ∈ {0, 1}. Let A and B be separable continuous fields of stable
UCT Kirchberg algebras over a finite-dimensional compact metrizable topological
space X. Assume that Ki(cid:0)A(Z)(cid:1) = 0 for all locally closed subsets Z ⊆ X. Then
any isomorphism OK(A) ∼= OK(B) lifts to a C(X)-linear ∗-isomorphism A ∼= B.
Notice that we do not assume that the fibers of A and B have rational K-theory
groups. Example 5.5 provides an example of a non-trivial C ∗-bundle falling under
the classification in Theorem 6.1.
The K-theoretical assumption in the theorem implies that A and B have van-
ishing boundary maps. Hence the universal coefficient theorem [1, Theorem 5.2]
applies (we may write OKXn instead of XnK by [1, Lemma 4.3]) and simplifies to
an isomorphism because the relevant Ext1-term vanishes for parity reasons. The
remainder of the proof is analogous.
References
[1] Rasmus Bentmann, Kirchberg X-algebras with real rank zero and intermediate cancellation
(2013), available at arXiv:math/1301.6652.
[2]
[3]
, Algebraic models in rational equivariant KK-theory. in preparation.
, Filtrated K-theory and classification of C ∗-algebras (University of Göttingen, 2010).
Diplom thesis, available online at: www.uni-math.gwdg.de/rbentma/diplom_thesis.pdf.
[4] Rasmus Bentmann and Marius Dadarlat, One-parameter continuous fields of Kirchberg al-
gebras with rational K-theory (2013), available at arXiv:math/1306.1691.
8
RASMUS BENTMANN
[5] Rasmus Bentmann and Manuel Köhler, Universal Coefficient Theorems for C ∗-algebras over
finite topological spaces (2011), available at arXiv:math/1101.5702.
[6] Rasmus Bentmann and Ralf Meyer, Circle actions on C ∗-algebras up to KK-equivalence. in
preparation.
[7] Etienne Blanchard, Subtriviality of continuous fields of nuclear C ∗-algebras, J. Reine Angew.
Math. 489 (1997), 133 -- 149, DOI 10.1515/crll.1997.489.133. MR 1461207
[8] Alexander Bonkat, Bivariante K-Theorie für Kategorien projektiver Systeme von C ∗-Alge-
bren, Ph.D. Thesis, Westf. Wilhelms-Universität Münster, 2002 (German). Available at the
Deutsche Nationalbibliothek at http://deposit.ddb.de/cgi-bin/dokserv?idn=967387191.
[9] Glen E. Bredon, Cosheaves and homology, Pacific J. Math. 25 (1968), 1 -- 32.
MR 0226631
[10]
, Sheaf theory, 2nd ed., Graduate Texts in Mathematics, vol. 170, Springer-Verlag,
New York, 1997. MR 1481706
[11] Marius Dadarlat, Continuous fields of C ∗-algebras over finite dimensional spaces, Adv. Math.
222 (2009), no. 5, 1850 -- 1881, DOI 10.1016/j.aim.2009.06.019. MR 2555914
[12] Marius Dadarlat and George A. Elliott, One-parameter continuous fields of Kirchberg algeb-
ras, Comm. Math. Phys. 274 (2007), no. 3, 795 -- 819, DOI 10.1007/s00220-007-0298-z.
MR 2328913
[13] Marius Dadarlat, George A. Elliott, and Zhuang Niu, One-parameter continuous fields of
Kirchberg algebras. II, Canad. J. Math. 63 (2011), no. 3, 500 -- 532, DOI 10.4153/CJM-2011-
001-6. MR 2828531
[14] Marius Dadarlat and Ralf Meyer, E-theory for C∗-algebras over topological spaces, J. Funct.
Anal. 263 (2012), no. 1, 216 -- 247, DOI 10.1016/j.jfa.2012.03.022. MR 2920847
[15] Marius Dadarlat and Cornel Pasnicu, Continuous fields of Kirchberg C ∗-algebras, J. Funct.
Anal. 226 (2005), no. 2, 429 -- 451. MR 2160103
[16] Jacques Dixmier, C ∗-algebras, North-Holland Publishing Co., Amsterdam, 1977. Translated
from the French by Francis Jellett; North-Holland Mathematical Library, Vol. 15.
MR 0458185
[17] Søren Eilers, Gunnar Restorff, and Efren Ruiz, Strong classification of extensions of classifi-
able C ∗-algebras (2013), available at arXiv:math/arXiv:1301.7695.
[18] Ilan Hirshberg, Mikael Rørdam, and Wilhelm Winter, C0(X)-algebras, stability and strongly
self-absorbing C ∗-algebras, Math. Ann. 339 (2007), no. 3, 695 -- 732, DOI 10.1007/s00208-007-
0129-8. MR 2336064
[19] Hvedri Inassaridze, Tamaz Kandelaki, and Ralf Meyer, Localisation and colocalisation of
KK-theory, Abh. Math. Semin. Univ. Hambg. 81 (2011), no. 1, 19 -- 34, DOI 10.1007/s12188-
011-0050-7. MR 2812030
[20] Eberhard Kirchberg, Das nicht-kommutative Michael-Auswahlprinzip und die Klassifikation
nicht-einfacher Algebren, C ∗-algebras (Münster, 1999), Springer, Berlin, 2000, pp. 92 -- 141
(German, with English summary). MR 1796912
[21] Eberhard Kirchberg and N. Christopher Phillips, Embedding of exact C ∗-algebras in the
Cuntz algebra O2, J. Reine Angew. Math. 525 (2000), 17 -- 53, DOI 10.1515/crll.2000.065.
MR 1780426
[22] Ralf Meyer and Ryszard Nest, C∗-algebras over topological spaces: filtrated K-theory, Canad.
J. Math. 64 (2012), no. 2, 368 -- 408, DOI 10.4153/CJM-2011-061-x. MR 2953205
[23]
[24]
, C ∗-algebras over topological spaces: the bootstrap class, Münster J. Math. 2 (2009),
215 -- 252. MR 2545613
, Homological algebra in bivariant K-theory and other triangulated categories. I, Tri-
angulated categories (Thorsten Holm, Peter Jørgensen, and Raphaël Rouqier, eds.), London
Math. Soc. Lecture Note Ser., vol. 375, Cambridge Univ. Press, Cambridge, 2010, pp. 236 -- 289,
DOI 10.1017/CBO9781139107075.006, (to appear in print). MR 2681710
[25] Cornel Pasnicu and Mikael Rørdam, Purely infinite C ∗-algebras of real rank zero, J. Reine
Angew. Math. 613 (2007), 51 -- 73, DOI 10.1515/CRELLE.2007.091. MR 2377129
[26] Gunnar Restorff, Classification of Non-Simple C∗-Algebras, Ph.D. Thesis, Københavns Uni-
versitet, 2008, http://www.math.ku.dk/~restorff/papers/afhandling_med_ISBN.pdf.
[27] Mikael Rørdam, Classification of nuclear, simple C ∗-algebras, Classification of nuclear
C ∗-algebras. Entropy in operator algebras, Encyclopaedia of Mathematical Sciences, vol. 126,
Springer, Berlin, 2002, pp. 1 -- 145.
[28] Jonathan Rosenberg and Claude Schochet, The Künneth theorem and the universal coefficient
theorem for Kasparov's generalized K-functor, Duke Math. J. 55 (1987), no. 2, 431 -- 474, DOI
10.1215/S0012-7094-87-05524-4. MR 894590
CLASSIFICATION OF CERTAIN CONTINUOUS FIELDS
9
Department of Mathematical Sciences, University of Copenhagen, Universitetsparken 5,
2100 Copenhagen Ø, Denmark
E-mail address: [email protected]
|
1510.02061 | 1 | 1510 | 2015-10-07T18:57:22 | KK-equivalence for amalgamated free product C*-algebras | [
"math.OA"
] | We prove that any reduced amalgamated free product C*-algebra is KK-equivalent to the corresponding full amalgamated free product C*-algebra. The main ingredient of its proof is Julg--Valette's geometric construction of Fredholm modules with Connes's view for representation theory of operator algebras. | math.OA | math | KK-EQUIVALENCE FOR AMALGAMATED FREE PRODUCT
C∗-ALGEBRAS
KEI HASEGAWA
Abstract. We prove that any reduced amalgamated free product C∗-algebra is KK-equivalent
to the corresponding full amalgamated free product C∗-algebra. The main ingredient of its
proof is Julg -- Valette's geometric construction of Fredholm modules with Connes's view for
representation theory of operator algebras.
1. Introduction
5
1
0
2
t
c
O
7
]
.
A
O
h
t
a
m
[
1
v
1
6
0
2
0
.
0
1
5
1
:
v
i
X
r
a
In [2][3] Cuntz gave a strategy of computing the K-theory of the reduced C∗-algebra C∗
red(Γ)
of a given discrete group Γ. The strategy consists of two parts:
(1) proving that the canonical surjection λ : C∗(Γ) → C∗
red(Γ) (where C∗(Γ) denotes the full
C∗-algebra of Γ) gives a KK-equivalence, that is, has an inverse in KK-theory, and
(2) computing the K-theory of C∗(Γ).
In fact, usual computations in K-theory are made by establishing suitable exact sequences, and
the full group C∗-algebra C∗(Γ) is easier to handle than the reduced one C∗
red(Γ). By the strategy,
Cuntz indeed gave a much simpler proof of celebrated Pimsner -- Voiculescu's result of the K-theory
of C∗
red(Fn) ([15]). Then Julg and Valette [11] achieved part (1) of the strategy when Γ acts on a
tree with amenable stabilizers. In the direction, Pimsner gave in [13] an optimal result, but his
strategy looks different from Cuntz's one.
It is very natural (at least for us) to try to adapt Cuntz's strategy to amalgamated free
product C∗-algebras. Part (2) of the strategy was achieved by Thomsen [17] under a very weak
assumption. Thus, our main problem is part (1) of the strategy. It was Germain [5][6] who first
tried to examine the strategy for plain free product C∗-algebras, and he obtained the desired KK-
equivalence result for plain free product C∗-algebras of nuclear C∗-algebras. Following Germain's
idea in [5][7] we recently proved in [9] (also see [8]) that the canonical surjection onto a given
reduced amalgamated free product C∗-algebra from the corresponding full one gives a KK-
equivalence under the assumption that every free component is "strongly relative nuclear" against
the amalgamated subalgebra. This was is a byproduct of our attempt to seek for a suitable
formulation of "relative nuclearity" for inclusions of C∗-algebras.
In this paper we adapt, unlike [5][7][8][9], Julg -- Valette's geometric idea to the problem, and
establish the optimal KK-equivalence result for amalgamated free product C∗-algebras. We
emphasize that the core part of the proof is very simple and just 3 pages long (though this paper
is rather self-contained). To state our main result precisely, let us give a few terminologies. Let
{(B ⊂ Ai, EAi
B )}i∈I be a countable family of quasi-unital inclusions of separable C∗-algebras with
nondegenerate conditional expectations from Ai onto B. Here B ⊂ Ai is quasi-unital if BAiB is
norm-dense in Ai, and also EAi
B : Ai → B is nondegenerate if the associated GNS representation
is faithful. Let (A, E) = ⋆B,i∈I (Ai, EAi
i ) be the reduced amalgamated free product and we
call A the reduced amalgamated free product C∗-algebra. Also, let A = ⋆B,i∈I Ai be the full
amalgamated free product C∗-algebra. With the notation we will prove the following:
Key words and phrases. amalgamated free product, KK-theory.
1
2
KEI HASEGAWA
Theorem A. The canonical surjection λ : A → A gives a KK-equivalence without any extra
assumption.
The proof is done by translating the "geometric" construction of Fredholm modules due to
Julg -- Valette (and its quantum group analog due to Vergnioux [20]) into a C∗-algebraic language
following Connes's view of correspondences. This is similar to our previous work [9] on relative
nuclearity. More precisely, we will easily prove that the canonical surjection λ gives a KK-
subequivalence like Julg -- Valette [11] and Vergnioux [20]. Then we will directly prove that λ
indeed gives a KK-equivalence. The latter is unnecessary in the amenable (quantum) group case
[11][20] thanks to the existence of counits, and is the most original part of the present paper. As
a bonus of the present approach we obtain the following corollary:
Corollary B. Both A and A are K-nuclear if all the Ai are nuclear.
Throughout this paper, we employ the following standard notation: For a Hilbert space H,
we denote by B(H) the bounded linear operators on H and by K(H) the compact ones on H.
For C∗-algebras A and B, A ⊗ B stands for the minimal tensor product. We use the symbol ⊙
for algebraic tensor products. For a subset S of a normed space X, we denote by [S] the closed
linear subspace of X generated by S.
2. preliminaries
2.1. C∗-correspondneces. For the theory of Hilbert C∗-modules, we refer to Lance's book [12].
Let A and B be C∗-algebras. An A-B C∗-correspondence is a pair (X, πX ) such that X is a
Hilbert B-module and πX is a ∗-homomorphism from A into the C∗-algebra LB(X) of right B-
linear adjointable operators on X. We denote by KB(X) the C∗-ideal of LB(X) generated by
"rank one operators" θξ,η, ξ, η ∈ X defined by θξ,η(ζ) := ξhη, ζi. The identity C∗-correspondence
over A is the pair (A, λA), where A is equipped with the A-valued inner product hx, yi = x∗y for
x, y ∈ A and λA : A → LA(A) is defined by the left multiplication. It is known that LA(A) is
naturally isomorphic to the multiplier algebra M(A) of A.
We use the following two notions of tensor products for Hilbert C∗-modules. Let X be a Hilbert
B-module and (Y, φ) be a B-C C∗-correspondence. We denote by X ⊗φ Y the interior tensor
product of X and (Y, φ), which is given by separation and completion of X ⊙ Y with respect to
the C-valued inner product hξ ⊗ η, ξ′ ⊗ η′i = hη, φ(hξ, ξ′i)ηi. There is a canonical map LB(X) →
LC (X ⊗φ Y ) sending T to the operator T ⊗φ 1Y : ξ ⊗ η 7→ (T ξ) ⊗ η. For a given ∗-homomorphism
πX : A → LB(X) we define πX ⊗φ 1Y : A → LC (X ⊗φ Y ) by (πX ⊗φ 1Y )(a) = πX (a) ⊗φ 1Y .
When no confusion may arise, we use the notations X ⊗B Y and πX ⊗B 1Y for short.
For a Hilbert D-module Z, we denote by X ⊗ Z the exterior tensor product of X and Y , which
is the completion of X ⊙ Y with respect to the B ⊗ D-valued inner product hξ ⊗ ζ, ξ′ ⊗ ζ ′i =
hξ, ξ′i ⊗ hζ, ζ ′i. When (Z, πZ ) is a C-D C∗-correspondence, there is a natural ∗-homomorphism
πX ⊗ πZ : A⊗ C → LB⊗D(X ⊗ Z) so that (X ⊗ Z, πX ⊗ πZ) is an A⊗ C-B ⊗ D C∗-correspondence.
Let B ⊂ A be a quasi-unital inclusion of C∗-algebras (i.e., BAB is norm-dense in A) with
conditional expectation E : A → B. We denote by L2(A, E) the Hilbert B-module given by
separation and completion of A with respect to the B-valued inner product hx, yi = E(x∗y)
for x, y ∈ A, and by πE : A → LB(L2(A, E)) the ∗-homomorphism induced from the left mul-
tiplication. The conditional expectation E is said to be nondegenerate if πE is faithful. We
denote by 1A the unit of the multiplier algebra of A. Since the inclusion B ⊂ A is quasi-unital,
B contains an approximate unit for A.
In particular, A is unital if and only if so is B, and
they should have a common unit. Thus, we can uniquely extend E to a conditional expectation
eE : A + C1A → B + C1A. Let ξE be the vector in L2(A + C1A, eE) corresponding to 1A. We
always identify L2(A, E) with [π eE(A)ξE] ⊂ L2(A + C1A,eE) and call the triple (L2(A, E), πE, ξE)
the GNS-representation associated with the conditional expectation E. Notice that ξE need not
to be in L2(A, E) when A is non-unital.
KK-EQUIVALENCE FOR AMALGAMATED FREE PRODUCT
3
2.2. KK-theory. Throughout this subsection, all C∗-algebras are assumed to be separable for
simplicity. We refer the reader to [1] for KK-theory.
Definition 2.1. For (trivially graded) C∗-algebras A and B, a Kasparov A-B bimodule is a triple
(X, φ, F ) such that X is a countably generated graded Hilbert B-module, φ : A → LB(X) is a
∗-homomorphism of degree 0, and F ∈ LB(X) is of degree 1 and satisfies the following condition:
• [F, φ(a)] ∈ KB(X) for a ∈ A,
• (F − F ∗)φ(a) ∈ KB(X) for a ∈ A,
• (1 − F 2)φ(a) ∈ KB(X) for a ∈ A.
When [F, φ(a)] = (F − F ∗)φ(a) = (1 − F 2)φ(a) = 0 holds for every a ∈ A, we say that (X, φ, F )
is degenerate. We denote by E(A, B) and D(A, B) the corrections of Kasparov A-B bimodules
and degenerate ones, respectively.
We say that two Kasparov A-B bimodules (X, φ, F ) and (Y, ψ, G) are unitarily equivalent,
denoted by (X, φ, F ) ∼= (Y, ψ, G), if there exists a unitary U ∈ L(X, Y ) of degree 0 such that
ψ = Ad U ◦ φ and G = U F U ∗.
For any Hilbert B-module X, we set IX := C([0, 1])⊗X. In particular, we set IB = C([0, 1])⊗
B. For each t ∈ [0, 1] we still denote by t the surjectiove ∗-homomorphism IB ∼= C([0, 1], B) ∋
f 7→ f (t) ∈ B. Note that we have a natural isomorphism IX ⊗t B ∼= X for every t ∈ [0, 1].
Definition 2.2. Two Kasparov A-B bimodules (X0, φ0, F0) and (X1, φ1, F1) are said to be ho-
motopic if there exists a Kasparov A-IB bimodule (Y, ψ, G) such that (Y ⊗t B, ψ⊗t 1B, G⊗t 1B) ∼=
(Xt, φt, Ft) for t = 0, 1. The KK-group KK(A, B) is the set of homotopy equivalence classes of
all Kasparov A-B bimodules.
The next technical lemma will be used later.
Lemma 2.3. Let P, Q and R be separable C∗-algebras and let (X, ψi, F ) ∈ E(Q, R) be given
for i = 0, 1. Suppose that there exist a surjective ∗-homomorphism π : P → Q and a family of
Kasparov P -R bimodules (X, φt, F ) for t ∈ [0, 1] satisfying
(i) the mapping [0, 1] ∋ t 7→ φt(a) is strictly continuous for each a ∈ P ;
(ii) φt factors through π : P → Q for every t ∈ [0, 1];
(iii) φi = ψi ◦ π holds for i = 0, 1.
Then, (X, ψ0, F ) and (X, ψ1, F ) are homotopic.
Proof. By assumption, there exists a ∗-homomorphism φ : P → LIR(IX) such that (IX, φ, F ⊗
1C[0,1]) ∈ E(P, IR) and φ ⊗t 1R = φt for t ∈ [0, 1]. Since one has kφ(a)k = sup0≤t≤1 kφt(a)k ≤
kπ(a)k for a ∈ P , there exists ψ : Q → LIR(IX) such that φ = ψ ◦ π. We then have (IX, ψ, F ⊗
1C([0,1])) ∈ E(Q, IR) and the evaluations of this Kasparov bimodule at endpoints are exactly
(X, ψi, F ), i = 0, 1.
(cid:3)
The KK-group becomes an additive group in the following way: For α, β ∈ KK(A, B) im-
plemented by (X, φ, F ), (Y, ψ, G), respectively, α + β is the element implemented by (X ⊕ Y, φ ⊕
ψ, F ⊕ G). All degenerate Kasparov bimodules are homotopic to the trivial bimodule 0 = (0, 0, 0)
and define the zero element in KK(A, B). Let X0 and X1 be the even and odd parts of X so
that X = X0 ⊕ X1 and let −X be the graded Hilbert B-module with the even part X1 and the
odd part X0. The inverse of α is implemented by (−X, Ad U ◦ φ, U F U ∗), where U : X → −X is
the natural unitary.
For any ∗-homomorphism φ : A → B, we have (B ⊕ 0, φ ⊕ 0, 0) ∈ E(A, B) and still denote by
φ the corresponding element in KK(A, B).
For α ∈ KK(A, B) and γ ∈ KK(B, C), the Kasparov product of α and γ is denoted by γ ◦ α
(or α ⊗B γ). When one of α and β comes from a ∗-homomorphism, the construction of the
Kasparov product is very simple (and we will use Kasparov products only in these special cases).
Indeed, if γ comes from a ∗-homomorphism γ : B → C with [γ(B)C] = C and α is implemented
4
KEI HASEGAWA
by (X, φ, F ), then the Kasparov product γ ◦ α is implemented by (X ⊗γ C, φ ⊗γ 1C , F ⊗γ 1C ).
Similarly, when α is a ∗-homomorphism from A into B and γ is implemented by (Y, ψ, G) with
[ψ(B)Y ] = Y , the Kasparov product γ ◦ α is implemented by (Y, ψ ◦ α, G).
Definition 2.4. An element α ∈ KK(A, B) is said to be a KK-equivalence if there exists
β ∈ KK(B, A) such that idA = β ◦ α and idB = α ◦ β. In this case, A and B are said to be
KK-equivalent.
Note that KK-equivalence between A and B implies KK(A, C) ∼= KK(B, C) and KK(C, A) ∼=
KK(C, B) for any separable C∗-algebra C.
Finally, we recall the notion of K-nuclearity in the sense of Skandalis [16].
Theorem 2.5 ([16, Theoreme 1.5]). Let A and B be separable C∗-algebras and let π : A → B(H)
be a faithful and essential representation on a separable Hilbert space H. For a given A-B C∗-
correspondence (X, σ) with X countably generated, the following are equivalent:
(i) For any unit vector ξ ∈ X the c.c.p. (completely contractive positive) map A ∋ a 7→
hξ, σ(a)ξi ∈ B is nuclear.
(ii) For any x ∈ KB(X) of norm 1, the c.c.p. map A ∋ a 7→ x∗σ(a)x ∈ KB(X) is nuclear.
(iii) There exists a sequence of isometries Vn ∈ LB(X, H ⊗ B) such that σ(a) − V ∗
n (π(a) ⊗
1A)Vn ∈ KB(X) and limn→∞ kσ(a) − V ∗
n (π(a) ⊗ 1A)Vnk = 0 for all a ∈ A.
When any of these three conditions holds, we say that (X, σ) is nuclear.
Note that any C∗-correspondence of the form (X ⊗B Y, πX ⊗B 1Y ) is nuclear whenever B is
nuclear (see e.g. [9, Remark 2.11]).
Definition 2.6. A separable C∗-algebra A is said to be K-nuclear if idA in KK(A, A) is imple-
mented by a Kasparov bimodule (X, φ, F ) such that (X, φ) is nuclear.
2.3. Amalgamated free products. Let {B ⊂ Ai}i∈I be a family of inclusions of C∗-algebras.
Recall that the full amalgamated free product of {Ai}i∈I over B is a C∗-algebra A generated by the
images of injective ∗-homomorphisms fi : Ai ֒→ A such that fiB = fjB for i, j ∈ I and satisfying
for any C∗-algebra C and ∗-homomorphisms πi : Ai → C
the following universal property:
satisfying πiB = πj B for i, j ∈ I, there exists a unique ∗-homomorphism ⋆i∈I πi : A → C
such that (⋆i∈I πi) ◦ fi = πi for i ∈ I. Since the full amalgamated free product is unique up to
isomorphism, we denote it by ⋆B,i∈I Ai. We identify Ai with fi(Ai) so that Ai ⊂ ⋆B,i∈I Ai for
every i ∈ I.
Further assume that, the inclusion B ⊂ Ai is quasi-unital and there exists a nondegenerate
conditional expectation EAi
B : Ai → B for each i ∈ I. In [21], Voiculescu introduced reduced
amalgamated free products of unital inclusions of C∗-algebras with conditional expectations. To
reduce Theorem A to the case when I = {1, 2}, we need to extend Voiculescu's definition to
quasi-unital inclusions. For any m ∈ N, set Im := {ι : {1, . . . , m} → I ι(k) 6= ι(k + 1) for 1 ≤
k ≤ m − 1}. Recall that the reduced amalgamated free product of {(Ai, EAi
B )}i∈I is a pair (A, E)
such that
• A is a C∗-algebra generated by the images of injective ∗-homomorphisms gi : Ai ֒→ A
such that giB = gjB for i, j ∈ I;
• E is a nondegenerate conditional expectation from A onto gi(B) (independent of i);
• one has E(gι(1)(x1)gι(2)(x2) · · · gι(m)(xm)) = 0 for any m ≥ 1, ι ∈ Im, and xk ∈ ker E
Aι(k)
B
for 1 ≤ k ≤ m.
We will also identify Ai with gi(Ai) for every i ∈ I. Since the pair (A, E) is determined by
the above three conditions, we will write ⋆B,I (Ai, EAi
B ) := (A, E). Clearly, we have a canonical
surjection λ : A → A satisfying that λ ◦ fi = gi for every i ∈ I.
We recall the construction of (A, E) ([21]). Let (Xi, πXi , ξi) be the GNS-representation as-
i )ξi] for
i = Xi ⊖ ξiB = [πXi (A◦
sociated with EAi
B (see §§2.1) and set A◦
i
:= ker EAi
B and X ◦
KK-EQUIVALENCE FOR AMALGAMATED FREE PRODUCT
5
i ∈ I. Recall that the free product of {(Xi, ξi)}i∈I is the Hilbert B-module (X, ξ0) defined to
ι(m), where we define hξ0b, ξ0ci = b∗c for b, c ∈ B. We
i so that Xi ⊂ X. For each i ∈ I, we consider the following submodules:
be ξ0B ⊕Lm≥1Lι∈Im X ◦
identify Xi with ξ0B ⊕ X ◦
ι(1) ⊗B · · · ⊗B X ◦
X ◦
ι(1) ⊗B · · · ⊗B X ◦
ι(m),
X(ℓ, i) := ξ0B ⊕Mm≥1 Mι∈Im
X(r, i) := ξ0B ⊕Mm≥1 Mι∈Im
ι(1)6=i
ι(m)6=i
X ◦
ι(1) ⊗B · · · ⊗B X ◦
ι(m).
Then, there is a natural unitary Vi
:= Ad Vi ◦
(πXi ⊗B 1X(ℓ,i)) and A = C∗(gi(Ai) i ∈ I). Then, the gi's agree with each other on B and
the compression map to ξ0B gives the desired conditional expectation E : A → B. Note that
the representation A ⊂ LB(X) with ξ0 is nothing but the GNS-representation associated with E,
and hence E must be nondegenerate.
: X ∼= Xi ⊗B X(ℓ, i) (see [21]). We set gi
The following lemma is probably well-known, but we give its proof for the reader's convenience.
Lemma 2.7. Let A = ⋆B,i∈I Ai and (A, E) = ⋆B,i∈I (Ai, EAi
B ) be as above. Let C be a C∗-
algebra and (Z, πZ ) be an A-C C∗-correspondence. Suppose that for each i ∈ I there exists a
subset Si ⊂ A◦
i as a normed space, and there exists a cyclic subspace Γ ⊂ Z (i.e.,
[πZ (A)ΓC] = Z holds) which satisfies the freeness condition: for any m ∈ N, ι ∈ Im, xk ∈ Sι(k)
for 1 ≤ k ≤ m and ξ, η ∈ Γ, one has hξ, πZ (x1x2 · · · xm)ηi = 0. Then, πZ factors through
λ : A → A.
i generating A◦
Proof. We will show that ker λ ⊂ ker πZ . Choose and fix z ∈ ker λ arbitrarily. By assumption, it
suffices to show that hξ, πZ (xzy)ηi = 0 for all x, y ∈ A and ξ, η ∈ Γ. We may assume that x and
y are in ∗-alg(Si∈I Ai). Take a sequence zn in ∗-alg(Si∈I Ai) such that limn→∞ kz − znk = 0.
For each n ≥ 1 there exists bn ∈ B such that xzny − bn is a sum of some elements of the form
x1 · · · xm for some m ≥ 1, ι ∈ Im and xk ∈ A◦
ι(k) for 1 ≤ k ≤ m so that bn = E(xzny). The
assumption on Γ implies that hξ, πZ (xzny − bn)ηi = 0, and hence we have khξ, πZ (xzy)ηik =
limn→∞ khξ, πZ(bn)ηik ≤ lim supn→∞ kξkkηkkE(xzny)k = 0.
(cid:3)
3. Proof
B ) ⋆B (A2, EA2
3.1. Case of two free components. We first deal with the case when I = {1, 2}. Let
(A, E) = (A1, EA1
B ), A = A1 ⋆B A2 and λ : A → A be as in Theorem A. As in
the previous section, let (X, πX , ξ0) be the GNS-representation associated with E and identify
Xi := L2(Ai, EAi
for i = 1, 2. Let EAi : A → Ai be the canonical conditional
expectation given by the compression map to Xi and let (Yi, πYi , ηi) be the GNS-representation
associated with EAi for i = 1, 2. Note that any vector of the form ξ0 ⊗ a with a ∈ Ai sits in
X(r, i) ⊗B Ai for each i ∈ I thanks to the assumption that B ⊂ A is quasi-unital. The following
lemma can be shown in the same manner as [20, Lemma 3.1], but we give its proof for the reader's
convenience.
B ) with ξ0B ⊕ X ◦
i
Lemma 3.1. The exists a unitary Si : X(r, i) ⊗B Ai → Yi satisfying that Si(ξ0 ⊗ y) = ηiy and
Si(x1 · · · xmξ0 ⊗ y) = x1 · · · xmηiy for all y ∈ Ai and m ∈ N, ι ∈ Im with ι(m) 6= i, and xk ∈ A◦
for 1 ≤ k ≤ m.
ι(k)
Proof. Note that if Si is bounded, then it must be surjective. Thus, it suffices to show that Si is
an isometry. By the polarization trick, we only have to verify that EAi (x∗x) = E(x∗x) for all x =
x1 · · · xm with m ∈ N, ι ∈ Im, ι(m) 6= i and xk ∈ A◦
ι(k) for 1 ≤ k ≤ m. When m = 1, this follows
from the fact that EAi is given by the compression to Xi. Assume that we have shown for k =
1, . . . , m. For ι ∈ Im+1 with ι(m + 1) 6= i, take xk ∈ A◦
ι(k) arbitrarily. If we put y = x2 · · · xm+1
6
KEI HASEGAWA
and b = E(x∗
have EAi (x∗x) = EAi(y∗E(x∗
E(y∗E(x∗
1x1), then the induction hypothesis implies that EAi(y∗by) = E(y∗by). Thus, we
1x1))y) = EAi (y∗by) = E(y∗by) =
(cid:3)
1x1))y) = E(x∗x). Hence, we are done.
1x1)y) + EAi(y∗(x∗
1x1)y) + E(y∗(x∗
1x1 − E(x∗
1x1 − E(x∗
Consider two A-A C∗-correspondences (Z +, π+) := (X ⊗B A, πX ⊗B 1) and (Z −, π−) :=
i=1(Yi ⊗Ai A, πYi ⊗Ai 1). Notice that the vector ζi := ηi ⊗ 1A is not necessarily in Z −, but one
has ζiA ⊂ Z −. Define the isometry S : Z + → Z − by
L2
(S1 ⊗A1 1 : X(r, 1)◦ ⊗B A → Y ◦
S2 ⊗A2 1 : X(r, 2) ⊗B A → Y2 ⊗A2 A.
1 ⊗A1 A;
Lemma 3.2 (c.f. [20, Theorem 3.3 (2)]). The operator S satisfies that ker S∗ = ζ1A and π−(a)S −
Sπ+(a) is compact for all a ∈ A. Consequently, the triple (Z + ⊕ Z −, π+ ⊕ π−,(cid:2) 0 S ∗
Kasparov bimodule.
S 0 (cid:3)) is an A-A
Proof. The first assertion is obvious. Thus, it suffices to show π−(x)S − Sπ+(x) is compact for
all x ∈ A1 ∪ A2. In fact, since each x ∈ A1 enjoys xX(r, 1)◦ ⊂ X(r, 1)◦ and xX(r, 2) ⊂ X(r, 2),
one has π−(x)S = Sπ+(x) for x ∈ A1. If we define S′ : Z + → Z − by S′ξ0 ⊗ a = ζ1a for a ∈ A
and by S on X ◦ ⊗B A, then S′ intertwines the actions of A2 by the above argument. Since S is
a compact perturbation of S′, we are done.
(cid:3)
Remark 3.3. Recall that the Bass -- Serre tree associated with an amalgamated free product
group G = G1 ⋆H G2 is the graph whose vertex and edge sets are given by ∆0 = G/G1 ⊔ G/G2
and ∆1 = G/H, respectively. Consider the unitary representations of G on ℓ2(∆0) and ℓ2(∆1)
induced from the action of G on (∆0, ∆1). In [9], we saw that C∗-correspondences play a role
of unitary representations for groups. In our theory, the canonical representation G on ℓ2(G/H)
corresponds to (L2(A, E)⊗B A, πE ◦ λ⊗B 1) (c.f. [10]). Thus, the C∗-correspondences (Z +, π+ ◦ λ)
and (Z −, π− ◦ λ) should play a role of the Bass -- Serre tree in C∗-algebra theory. Also, the adjoint
of S corresponds to the co-isometry of Julg -- Valette in [11].
Here is the main technical result of this paper.
Theorem 3.4. With the notation above, let α be the element in KK(A, A) implemented by
S 0 (cid:3)). Then, we have α ◦ λ + idA = 0 and λ ◦ α + idA = 0.
(Z + ⊕ Z −, π+ ⊕ π−,(cid:2) 0 S ∗
((Z + ⊕ A) ⊕ Z −, (ρ+ ⊕ λA) ⊕ ρ−,(cid:2) 0 U ∗
Proof. We first prove that α ◦ λ + idA = 0 following the proof of [20, Theorem 3.3 (3)]. Set
ρ+ := π+ ◦ λ and ρ− := π− ◦ λ. Define the unitary U : Z + ⊕ A → Z − by S on Z + and by
U (0 ⊕ a) := ζ1a for a ∈ A. Since S is a compact perturbation of U , α ◦ λ + idA is implemented by
U ∗ 0 (cid:3)) (see §§2.2). Take a norm continuous path (vt)0≤t≤1
of unitaries in M2(C) such that v0 = 1 and v1 = [ 0 1
1 0 ]. With the natural identification M2(C) ⊂
M2(C) ⊗ M(A) = LA(ζ1A ⊕ ζ2A) we define the unitary ut ∈ LA(Z −) by vt on ζ1A ⊕ ζ2A and by
the identity operator on Z − ⊖(ζ1A⊕ζ2A). Since the restriction of B to ζ1A⊕ζ2A is just C1⊗B ⊂
M2(C) ⊗ M(A) with the above identification, the family (ut)0≤t≤1 forms a norm continuous path
of unitaries in π−(B)′ ∩(C1+KB(Z −)) satisfying that u0 = 1 and u1 switches ζ1a and ζ2a for each
a ∈ A. Let ιi : Ai ֒→ A be the inclusion map for i = 1, 2. Since π− ◦ ι1 agrees with Ad ut ◦ π− ◦ ι2
on B, we have the natural ∗-homomorphism φt := (π− ◦ ι1) ⋆ (Ad ut ◦ π− ◦ ι2) : A → LA(Z −)
thanks to the universality of A. Then, the Kasparov bimodules
(cid:0)(Z + ⊕ A) ⊕ Z −, (ρ+ ⊕ λA) ⊕ φt,(cid:2) 0 U ∗
U 0 (cid:3)(cid:1) ,
t ∈ [0, 1]
degenerate, that is,
satisfy conditions (i) and (ii) in Lemma 2.3 (with P = Q = A), and its evaluation at t = 0
implements α ◦ λ + idA. Thus, we need to show that ((Z + ⊕ A) ⊕ Z −, (ρ+ ⊕ λA) ⊕ φ1,(cid:2) 0 U ∗
(1)
Since U is unitary, we may assume that x is in A1 ∪ A◦
2. When x is in A1, the above equation is
trivial because S intertwines π+(x) and π−(x). Let S′ be as in the proof of the previous lemma.
U (ρ+(x) ⊕ λA(x)) = φt(x)U for x ∈ A.
U 0 (cid:3)) is
KK-EQUIVALENCE FOR AMALGAMATED FREE PRODUCT
7
Then, we have u∗
of A2, we have U (π+(x) ⊕ λA(x)) = u1π−(x)u∗
and hence Lemma 2.3 shows α ◦ λ + idA = 0.
1U = S′ on Z + and u∗
1U (0 ⊕ a) = ζ2a for a ∈ A. Since S′ intertwines the actions
1U for every x ∈ A2. Thus we obtain equation (1),
We next prove that λ ◦ α + idA = 0 in KK(A, A). Note that λ ◦ α + idA is implemented by
the Kasparov A-A bimodule
(see §§2.2). We observe that the family of Kasparov A-A bimodules
(cid:16)((Z + ⊗λ A) ⊕ A) ⊕ (Z − ⊗λ A), ((π+ ⊗λ 1A) ⊕ λA) ⊕ (π− ⊗λ 1A),h
(cid:16)((Z + ⊗λ A) ⊕ A) ⊕ (Z − ⊗λ A), ((ρ+ ⊗λ 1A) ⊕ λ) ⊕ (φt ⊗λ 1A),h
U ∗ ⊗λ1
0
0
U ∗ ⊗λ1A
U ⊗λ1A
0
i(cid:17)
U ⊗λ1
0
i(cid:17) ,
t ∈ [0, 1]
satisfy conditions (i) and (ii) in Lemma 2.3 (with P = A and Q = A) and its evaluations at
endpoints implement (λ ◦ α + idA) ◦ λ and 0. Thus, by Lemma 2.3 and the fact that πX : A →
LB(X) is faithful, it suffices to show that φt ⊗πX ◦λ 1X : A → LB(Z − ⊗πX ◦λ X) factors through
i=1 πYi ⊗Ai 1X : A → LB((Y1 ⊗A1 X)⊕ (Y2 ⊗A2 X))
and wt := ut ⊗πX ◦λ 1X ∈ LB((Y1 ⊗A1 X) ⊕ (Y2 ⊗A2 X)), then φt ⊗πX ◦λ 1X coincides with
ψt := (σ ◦ ι1) ⋆ (Ad wt ◦ σ ◦ ι2). Note that ψ0 = σ ◦ λ and ψ1 ∼= (ρ+ ⊗B 1X ) ⊕ πX ◦ λ apparently
factor through λ. Thus, we only have to deal with 0 < t < 1 and we write w = wt for short.
λ : A → A for every t ∈ [0, 1]. If we set σ :=L2
For the convenience, we identify X(r, i) ⊗B X with Yi ⊗Ai X via Si ⊗Ai 1 as right B-modules.
To distinguish between vectors in X(r, 1) ⊗B X and X(r, 2) ⊗B X, we use the symbols ⊗ and ⊗ as
markers in such a way that, for ζ ∈ X we denote by ξ0 ⊗ζ ∈ X(r, 1) ⊗B X and ξ0 ⊗ζ ∈ X(r, 2) ⊗B X
the vectors corresponding to η1 ⊗ ζ and η2 ⊗ ζ, respectively. Thanks to Lemma 2.7, the proof
will be completed by proving the following claim:
Claim 3.5. The subspace Γ := w(ξ0B ⊗B X(ℓ, 1)) + ξ0B ⊗B X(ℓ, 2) satisfies the assumption of
Lemma 2.7.
We first show that Γ is cyclic for ψt(A). Let Λ := [ψt(A)Γ]. We set X0 := ξ0B, Xm :=
ι(m), Xm(ℓ, i) = Xm ∩ X(ℓ, i) and Xm(r, i) = Xm ∩ X(r, i) for m ∈ N
ι(i) ⊗B · · · ⊗B X ◦
and i = 1, 2. It suffices to show that, for any m ∈ {0} ∪ N, Λ contains
Nι∈Im X ◦
Ym := mMk=0
Xk(r, 1) ⊗B Xm−k! ⊕ mMk=0
Xk(r, 2) ⊗B Xm−k! .
We will show this by induction. When m = 0, this is trivial because Y0 = (ξ0B ⊗B ξ0B) ⊕
(ξ0B ⊗B ξ0B) = w(ξ0B ⊗B ξ0B) + ξ0B ⊗B ξ0B. Suppose that Λ contains Yk for 0 ≤ k ≤ m.
Since w is equal to 1 on the complement of (ξ0B ⊗B X) ⊕ (ξ0B ⊗B X), it is easily seen that
m+1Mk=2
Xk(r, 1) ⊗B Xm+1−k! ⊕ m+1Mk=2
Xk(r, 2) ⊗B Xm+1−k! ⊂ Λ.
Thus, we only have to check that Λ contains the following six subspaces:
X ◦
X ◦
2 ⊗B Xm,
1 ⊗B Xm,
ξ0B ⊗B Xm+1(λ, 1),
ξ0B ⊗B Xm+1(λ, 1),
ξ0B ⊗B Xm+1(ℓ, 2),
ξ0B ⊗B Xm+1(ℓ, 2).
By assumption of induction, one has w(ξ0B ⊗B Xm) ⊂ Ym ⊂ Λ, and hence X ◦
[wσ(A◦
that w(ξ0B ⊗B Xm+1(ℓ, 1)) = [wσ(A◦
by the definition of Γ. Thus, one has
2 ⊗B Xm =
1 ⊗B Xm = [σ(A1)(ξ0B ⊗B Xm)] ⊂ Λ. We observe
2)w∗w(ξ0B ⊗B Xm(ℓ, 2))] ⊂ Λ and w(ξ0B ⊗B Xm+1(ℓ, 1)) ⊂ Γ
2)ww∗(ξ0B ⊗B Xm)] ⊂ Λ. We also have X ◦
ξ0B ⊗B Xm+1(ℓ, 1) + ξ0B ⊗B Xm+1(ℓ, 1) = w(cid:0)ξ0B ⊗B Xm+1(ℓ, 1) + ξ0B ⊗B Xm+1(ℓ, 1)(cid:1) ⊂ Λ.
Finally we obtain that ξ0B ⊗B Xm+1(ℓ, 2) = [σ(A1)(ξ0B ⊗BXm(ℓ, 1))] ⊂ Λ and ξ0B ⊗BXm+1(ℓ, 2) ⊂
Γ by the definition of Γ again. Therefore, by induction, it follows that Γ is cyclic for ψt(A).
8
KEI HASEGAWA
We next show that Γ = w(ξ0B ⊗B X(ℓ, 1)) + ξ0B ⊗B X(ℓ, 2) satisfies the freeness condition.
Let Γ1 := ξ0B ⊗B X(ℓ, 2)◦ and Γ2 := w(ξ0B ⊗B X(ℓ, 1)◦). We then claim that the following
inclusions hold:
Indeed, for any x ∈ A◦
σ(A◦
1)Γ ⊂ Γ1 + X ◦
1 one has
1 ⊗B X and wσ(A◦
2)w∗Γ ⊂ Γ2 + X ◦
2 ⊗B X.
(2)
σ(x)w(ξ0B ⊗B X(ℓ, 1)) ⊂ σ(x)(ξ0B ⊗B X(ℓ, 1) + ξ0B ⊗B X(ℓ, 1))
⊂ ξ0B ⊗B X(ℓ, 2)◦ + X ◦
⊂ Γ1 + X ◦
1 ⊗B X
1 ⊗B X(ℓ, 1)
and σ(x)(ξ0B ⊗B X(ℓ, 2)) ⊂ X ◦
wσ(y)w∗w(ξ0B ⊗B X(ℓ, 1)) ⊂ w(X ◦
1 ⊗B X(ℓ, 2). Similarly, for any y ∈ A◦
2 ⊗B X(ℓ, 1)) = X ◦
2 one has
2 ⊗B X(ℓ, 1)
and
wσ(y)w∗(ξ0B ⊗B X(ℓ, 2)) ⊂ wσ(y)(ξ0B ⊗B X(ℓ, 2) + ξ0B ⊗B X(ℓ, 2))
2 ⊗B X(ℓ, 2)) + w(ξ0B ⊗B X(ℓ, 1)◦)
⊂ w(X ◦
= X ◦
2 ⊗B X(ℓ, 2) + Γ2.
The subspaces on the right hand side in both equations (2) are apparently orthogonal to Γ, and
2 ⊗B X. Since
one can easily verify that σ(A◦
w = 1 on the complement of (ξ0B ⊗B X) ⊕ (ξ0B ⊗B X), the above observations show that
for any xk ∈ Aι(k) for k = 1, . . . , m with ι ∈ Im, the subspace ψt(x1 · · · xm)Γ is contained in
Γ1 + X(ℓ, 2)◦ ⊗B X + X(ℓ, 2)◦ ⊗B X when ι(1) = 1, and in Γ2 + X(ℓ, 1)◦ ⊗B X + X(ℓ, 1)◦ ⊗B X
when ι(1) = 2. This implies that Γ satisfies the freeness condition.
(cid:3)
1 ⊗B X and wσ(A◦
2)w∗Γ1 ⊂ Γ2 + X ◦
1)Γ2 ⊂ Γ1 + X ◦
3.2. Case of countably many free components. Let I be a general countable set and let
A = ⋆B,i∈I Ai and (A, E) = ⋆B,i∈I(Ai, EAi
B ) be as in Theorem A. We set c0 := c0(I) and
K := K(ℓ2(I)).
π−−−−→ K ⊗ A
Proposition 3.6. With the notation above, there exist nondegenerate conditional expectations
B :Pi Ai → c0⊗B and Ec0 ⊗idB : K⊗B → c0⊗B. If we seteA := (Pi Ai)⋆c0⊗B (K⊗B) and
B )⋆c0⊗B (K ⊗ B, Ec0 ⊗ idB), then there exist isomorphisms π :eA → K ⊗ A
Pi EAi
(eA,eE) := (Pi Ai,Pi EAi
and πred : eA → K ⊗ A such that the following diagram
eA
eλy
eA
commutes, where eλ is the canonical surjection.
units for the canonical basis {δi}i≥1 of ℓ2(N), and set fn := enn. We realizePn≥1 An and c0 ⊗ B
Proof. Since the proof in the case when I is finite is essentially same as (and easier than) the
case when I = N, we may and do assume that I = N. Let {eij}i,j≥1 be the system of matrix
yidK ⊗λ
πred−−−−→ K ⊗ A
inside K ⊗ A as
An = C∗{fn ⊗ a n ≥ 1, a ∈ An}
and c0 ⊗ B = C∗{fn ⊗ b n ≥ 1, b ∈ B}.
B )(fi ⊗ a) = fi ⊗ EAi
Consider two conditional expectationsPn EAn
defined by (Pn EAn
a ∈ Ai and b ∈ B. Set eA := (Pn≥1 An) ⋆c0⊗B (K ⊗ B) and (eA, eE) := (Pn≥1 An,Pn EAn
(K ⊗ B, Ec0 ⊗ idB) and leteλ :eA → eA be the canonical surjection.
B :Pn An → c0 ⊗B and Ec0 ⊗idB : K⊗B → c0 ⊗B
B (a) and (Ec0 ⊗ idB)(eij ⊗ b) = δi,jfi ⊗ b for i, j ∈ N,
B ) ⋆c0⊗B
Xn≥1
KK-EQUIVALENCE FOR AMALGAMATED FREE PRODUCT
9
The inclusion maps Pn An ֒→ K ⊗ A and K ⊗ B ֒→ K ⊗ A induce a ∗-homomorphism π :
eA → K ⊗ A. For any n, i, j ∈ N, a ∈ An and b, c ∈ B, one has eij ⊗ bac = π(ein ⊗ b)π(fn ⊗
a)π(enj ⊗ c) ∈ π(eA). Since [BAnB] = An holds, π is surjective. Define σn : An → eA by
σn(bac) = (e1n⊗b)(fn⊗a)(en1⊗c) for a ∈ An and b, c ∈ B. We then obtain σ = ⋆n≥1σn : A →eA.
Define eσ : K ⊗ A →eA byeσ(eij ⊗ bac) = (ei1 ⊗ b)σ(a)(e1j ⊗ c) for a ∈ A and i, j ≥ 1. Then, it is
easy to see that eσ ◦ π = id eA, and hence π is bijective.
We next see that (idK ⊗λ) ◦ π andeλ ◦eσ factor througheλ :eA → eA and idK ⊗λ : K ⊗ A → K ⊗ A,
respectively. Note that (Pn An)◦ and (K ⊗ B)◦ are generated by S1 = {fn ⊗ a a ∈ A◦
n, n ≥ 1}
and S2 = {eij ⊗b i, j ≥ 1, i 6= j, b ∈ B} as normed spaces, respectively. We represent K⊗A on the
Hilbert B-module ℓ2(N) ⊗ L2(A, E) faithfully and show that Cδ1 ⊗ ξEB satisfies the assumption
of Lemma 2.7 for S1 and S2. Let m ≥ 1, x1, . . . , xm ∈ S1 and y1, . . . , ym ∈ S2 be arbitrarily
given. If x1y2 · · · xmym is a nonzero element, then it should be of the form
(fι(1) ⊗ a1)(eι(1)ι(2) ⊗ b1) · · · (fι(m) ⊗ am)(eι(m)ι(m+1) ⊗ bm) = eι(1)ι(m+1) ⊗ (a1b1 · · · ambm)
for some ι ∈ Im+1, ak ∈ A◦
ι(k) and bk ∈ B for 1 ≤ k ≤ m. Clearly, this implies that
(x1y1 · · · xmym)(δ1⊗ξEB) ⊥ δ1⊗ξEB. A similar assertion holds for x1y1 · · · ymxm+1, y0x1 · · · xmym
and y0x1 · · · ymxm+1 for any xm+1 ∈ S1 and y0 ∈ S2. Therefore, Lemma 2.7 guarantees that
condition for S ′
n := BA◦
nB, n ≥ 1. Indeed, for m ≥ 1, ι ∈ Im, yk ∈ A◦
(idK ⊗λ) ◦ π factors througheλ.
Similarly, representing eA on L2(eA, eE) faithfully we observe that ξ eE(c0 ⊗B) satisfies the freeness
which belongs to kereE. Thus,eλ ◦ σ : A → eA factors through λ : A → A, which implies thateλ ◦eσ
σ((b1y1c1) · · · (bmymcm)) = (e1ι(1) ⊗ b1)(fι(1) ⊗ y1)(eι(1)ι(2) ⊗ c1b2) · · · (fι(m) ⊗ ym)(eι(m)1 ⊗ cm),
factors through idK ⊗λ : K ⊗ A → K ⊗ A.
ι(k), and bk, ck ∈ B, we have
(cid:3)
The following general fact is well-known (see, e.g. [1, Proposition 17.8.7]).
Proposition 3.7. Let K be as above and let ι : K ֒→ B(ℓ2(I)) be the inclusion map. Fix a
minimal projection e ∈ K. For any separable C∗-algebras A and B, the mapping E(A, B) ∋
(X, φ, F ) 7→ (K ⊗ X, λK ⊗ φ, 1K ⊗ F ) ∈ E(K ⊗ A, K ⊗ B) induces an isomorphism τ : KK(A, B) →
KK(K ⊗ A, K ⊗ B). The inverse of τ is given by the mapping E(K ⊗ A, K ⊗ B) ∋ (Y, ψ, G) 7→
(Y ⊗ι⊗λB (ℓ2(I) ⊗ B), (ψ ⊗ι⊗λB 1) ◦ σ, G ⊗ι⊗λB 1) ∈ E(A, B), where σ(a) = e ⊗ a for a ∈ A.
We are now ready to prove Theorem A and Corollary B.
Proof of Theorem A and Corollary B. We use the notation in the proof of Proposition 3.6. By
Theorem 3.4 and Proposition 3.6, there exists β ∈ KK(K ⊗ A, K ⊗ A) such that β ◦ (idK ⊗λ) =
idK⊗A and (idK ⊗λ) ◦ β = idK⊗A. Let τ be as in Proposition 3.7. We then have idA =
τ −1(idK⊗A) = τ −1(β ◦ (idK ⊗λ)) = τ −1(β) ◦ λ and idA = τ −1(idK⊗A) = τ −1((idK ⊗λ) ◦ β) =
λ ◦ τ −1(β). Thus, λ gives a KK-equivalence.
Moreover, by Theorem 3.4 and Proposition 3.7 again, τ −1(β) is implemented by a Kasparov
A-A bimodule whose "C∗-correspondence part" is the direct sum of three C∗-correspondences of
the form (Y ⊗D Z, πY ⊗D 1Z), where D is either c0 ⊗ B,Pi Ai or K ⊗ B. Thus, if Ai is nuclear
for every i ∈ I, then idA = τ −1(β) ◦ λ is also implemented a Kasparov bimodule consisting of a
nuclear C∗-correspondence (see the remark just after Theorem 2.5), and hence A is K-nuclear. (cid:3)
Remark 3.8. Theorem A generalizes the previous K-amenability results for amalgamated free
products of amenable discrete (quantum) groups [11] and [20]. However, we should remark that
our result does not imply Pimsner's result that K-amenability is closed under amalgamated free
products. Similarly, Corollary B does not imply that K-nuclearity is closed under amalgamated
free products (even for plain free products). The latter seems a next interesting question in the
direction.
10
KEI HASEGAWA
4. Six-term exact sequences
Let (A, E) = (A1, EA1
B ) is as in Theorem 3.4. We denote by ik : B → Ak and
jk : Ak → A, k = 1, 2 the inclusion maps. As we mentioned in the introduction, our KK-
equivalence and K-nuclearity results with Thomsen's result [17] imply the following:
B ) ⋆ (A2, EA2
Corollary 4.1. With the notation above, there is a cyclic six-term exact sequence
K0(B)
x
K1(A)
(i1∗ ,i2∗)
−−−−−→ K0(A1) ⊕ K0(A2)
j1∗−j2∗←−−−−− K1(A1) ⊕ K1(A2)
j1∗ −j2∗
−−−−−→ K0(A)
(i1∗ ,i2∗)
←−−−−− K1(B)
y
If A1 and A2 are further assumed to be nuclear, then for any separable C∗-algebras D there is a
cyclic exact sequence
KK(B, D)
KK(A, SD)
i∗
1 −i∗
2←−−−− KK(A1, D) ⊕ KK(A2, D)
j∗
1 +j∗
2−−−−→ KK(A1, SD) ⊕ KK(A2, SD)
j∗
1 +j∗
2←−−−− KK(A, D)
i∗
1 −i∗
2−−−−→ KK(B, SD)
Note that the second exact sequence of KK-groups is new even in the full case. We also obtain
the next corollary from Theorem A and [4].
Corollary 4.2. With the notation above, suppose that B is a direct sum of finite dimensional
C∗-algebras. Then, for any separable C∗-algebra D there are two cyclic exact sequences:
KK(D, B)
KK(SD, A)
KK(B, D)
(i1∗,i2∗)
−−−−−→ KK(D, A1) ⊕ KK(D, A2)
j1∗−j2∗←−−−−− KK(SD, A1) ⊕ KK(SD, A2)
i∗
1 −i∗
2←−−−− KK(A1, D) ⊕ KK(A2, D)
j1∗−j2∗
−−−−−→ KK(D, A)
(i1∗,i2∗)
←−−−−− KK(SD, B)
j∗
1 +j∗
2←−−−− KK(A, D)
y
x
y
x
x
y
x
y
KK(A, SD)
j∗
1 +j∗
2−−−−→ KK(A1, SD) ⊕ KK(A2, SD)
i∗
1 −i∗
2−−−−→ KK(B, SD)
Finally, we would like to point out that a similar result holds for HNN extensions. We refer
the reader to [18, 19] for HNN extensions of C∗-algebras. The next corollary follows from "the
C∗-version of Proposition 3.1", Proposition 3.3 and Proposition 4.2 in [19] and Theorem A.
Corollary 4.3. Let B ⊂ A be unital inclusion of separable C∗-algebras with nondegenerate
conditional expectation E : A → B, and θ : B → A be an injective ∗-homomorphism whose image
is the range of a conditional expectation Eθ : A → θ(B). Then, the full HNN-extension A ⋆univ
B θ
and the reduced one (A, E) ⋆B (θ, Eθ) are KK-equivalent via the canonical surjection, and there
is a six-term exact sequence:
K0(B)
(θ∗−ιB∗)
−−−−−−→ K0(A)
ιA∗−−−−→ K0((A, E) ⋆B (θ, Eθ))
K1((A, E) ⋆B (θ, Eθ))
ιA∗←−−−− K1(A)
(θ∗−ιB∗)
←−−−−−−
K1(B)
Here ιB : B → A and ιA : A → (A, E) ⋆B (θ, Eθ) are inclusion maps. Further assume that A is
nuclear. Then, these HNN-extensions are K-nuclear.
KK-EQUIVALENCE FOR AMALGAMATED FREE PRODUCT
11
Remark 4.4. Using Proposition 3.7 we can generalize the results in this section to amalgamated
free products and HNN extensions of countably many C∗-algebras and countably many injec-
tive ∗-homomorphisms, respectively. Such generalizations for HNN extensions include Pimsner --
Voiculescu's six-term exact sequence for crossed products by free groups ([14, 15]) as special cases
(see also [19]).
References
[1] B. Blackadar, K-theory for Operator Algebras. Second edition. Mathematical Sciences Research Institute
Publications, 5.
[2] J. Cuntz, The K-groups for free products of C ∗-algebras. Operator algebras and applications, Part I, Kingston,
Ont., Proc. Sympos. Pure Math. 38 (1982) 81 -- 84.
[3] J. Cuntz, K-theoretic amenability for discrete groups. J. Reine Angew. Math. 344 (1983), 180 -- 195.
[4] E. Eliasen, KK-theory of amalgamated free products of C ∗-algebras. Math. Nachr. 280 (2007), no. 3, 271 -- 280.
[5] E. Germain, KK-theory of reduced free-product C ∗-algebras. Duke Math. J. 82 (1996), no. 3, 707 -- 723.
[6] E. Germain, KK-theory of full free-product C ∗-algebras. J. Reine Angew. Math. 485 (1997), no. 10, 1 -- 10.
[7] E. Germain, Amalgamated free product C ∗-algebras and KK-theory. in Free Probability Theory, 89 -- 103,
Fields Inst. Commun., 12, Amer. Math. Soc., Providence, RI, 1997.
[8] E. Germain and A. Sarr, KK-theory for some graph C ∗-algebras. C. R. Math. Acad. Sci. Paris, Ser. I 353
(2015) 443 -- 444.
[9] K. Hasegawa, Relative nuclearity for C∗-algebras and KK-equivalences of amalgamated free products.
J. Func. Anal. to appear.
[10] K. Hasegawa, Relative nuclearity for C∗-algebras Part II., in preparation.
[11] P. Julg and A. Valette, K-theoretic amenability for SL2(Qp), and the action on the associated tree. J. Funct.
Anal. 58 (1984), no. 2, 194 -- 215.
[12] E.C. Lance, Hilbert C∗-modules. A toolkit for operator algebraists. London Mathematical Society Lecture
Note Series, 210. Cambridge University Press, Cambridge, 1995.
[13] M. Pimsner, KK-groups of crossed products by groups acting on trees. Invent. Math. 86 (1986), no. 3, 603 --
634.
[14] M. Pimsner and D. Voiculescu, Exact sequences for K-groups and Ext-groups of certain cross-product C ∗-
algebras. J. Operator Theory 4 (1980), no. 1, 93 -- 118.
[15] M. Pimsner and D. Voiculescu, K-groups of reduced crossed products by free groups. J. Operator Theory 8
(1982), no. 1, 131 -- 156.
[16] G. Skandalis, Une notion de nucl´earit´e ´en K-th´eorie (d'apr`es J. Cuntz). K-Theory 1 (1988), no. 6, 549 -- 573.
[17] K. Thomsen, On the KK-theory and the E-theory of amalgamated free products of C ∗-algebras. J. Funct.
Anal. 201 (2003), no. 1, 30 -- 56.
[18] Y. Ueda, HNN extensions of von Neumann algebras. J. Funct. Anal. 225 (2005), no. 2, 383 -- 426.
[19] Y. Ueda, Remarks on HNN extensions in operator algebras. Illinois J. Math. 52 (2008), no. 3, 705 -- 725.
[20] R. Vergnioux, K-amenability for amalgamated free products of amenable discrete quantum groups. J. Funct.
Anal. 212 (2004), no. 1, 206 -- 221.
[21] D. Voiculescu, Symmetries of some reduced free product C ∗-algebras. in Operator Algebras and Their Con-
nections with Topology and Ergodic Theory, 556 -- 588, Lecture Notes in Math., 1132, Springer, Berlin, 1985.
Graduate School of Mathematics, Kyushu University, Fukuoka 819-0395, Japan
E-mail address: [email protected]
|
math/0607488 | 4 | 0607 | 2016-04-16T10:38:17 | TRO equivalent algebras | [
"math.OA"
] | In this work we study a new equivalence relation between w* closed algebras of operators on Hilbert spaces. The algebras A and B are called TRO equivalent if there exists a ternary ring of operators M (i.e. MM*M\subset M) such that A is the w*-closed span of M*BM and B is the w*-closed span of MAM*. We prove that two reflexive algebras are TRO equivalent if and only if there exists a * isomorphism between the commutants of their diagonals mapping the invariant projection lattice of the first algebra onto the lattice of the second one. We explore some consequences of TRO equivalence for CSL algebras. We also prove that TRO equivalence is stronger than "spatial Morita equivalence". Two CSL algebras are "spatially Morita equivalent" if and only if their lattices are isomorphic. In this case if one of them is synthetic then so is the other. | math.OA | math |
TRO EQUIVALENT ALGEBRAS
G.K. ELEFTHERAKIS
Abstract. In this work we study a new equivalence relation between
w∗ closed algebras of operators on Hilbert spaces. The algebras A and
B are called TRO equivalent if there exists a ternary ring of opera-
tors M (i.e. MM∗M ⊂ M) such that A = span(M∗BM)−w∗
and
B = span(MAM∗)−w∗
. We prove that two reflexive algebras are TRO
equivalent if and only if there exists a ∗ isomorphism between the com-
mutants of their diagonals mapping the invariant projection lattice of
the first algebra onto the lattice of the second one.
A linear space M of operators between two Hilbert spaces satisfying
1. Introduction
MM∗M ⊂ M
is called a ternary ring of operators (TRO).
TRO's were introduced in [12] and constitute a generalisation of selfad-
joint operator algebras [11], [20]. They have many properties similar to
C ∗-algebras and von Neumann algebras. Recently, these objects have been
studied from the point of view of operator space theory, in which they play
an important role [5], [13], [19].
In [15], TRO's were studied from a different angle, namely as normalisers
of operator algebras:
If A ⊂ B(H) and B ⊂ B(K) are w∗ closed operator algebras, not neces-
sarily selfadjoint, an operator T ∈ B(H, K) is said to normalise the algebra
A into B if T ∗BT ⊂ A and T AT ∗ ⊂ B. It is shown in [15] that such a
normaliser T defines a TRO MT consisting of normalisers from A into B :
M∗
T BMT ⊂ A and MT AM∗
T ⊂ B.
In the present paper we are interested in a stronger situation, namely in
the existence of a TRO M so that
A = span(M∗BM)−w∗
and B = span(MAM∗)−w∗
.
2000 Mathematics Subject Classification. 47L05 (primary), 47L35, 46L10, 16D90
(secondary).
Key words and phrases. Operator algebras, reflexive algebras, CSL, TRO, Morita
equivalence.
This research was partly supported by Special Account Research Grant No. 70/3/7463
of the University of Athens.
1
2
G.K. ELEFTHERAKIS
In this case we call the algebras A and B TRO equivalent. Note that TRO
equivalence is a generalisation of unitary equivalence. We show (section 2)
that it is indeed an equivalence relation.
In the selfadjoint case, TRO equivalence coincides with the existence of
an "equivalence bimodule" for the algebras (see section 2) thus TRO equiv-
alence implies "Morita equivalence" in the sence of Rieffel [17]. This crucial
observation partly motivated our work.
In a companion paper we prove
that TRO equivalence is the appropriate context in which Rieffel's theory
for Morita equivalence of W ∗-algebras can be generalized to the class of
possibly nonselfadjoint (abstract) dual operator algebras [6]. Using results
of the present paper this theory is applied in [7] to the class of reflexive
algebras.
Also TRO equivalence is related to the very important notion of stable
isomorphism between dual operator algebras: In another paper [8] jointly
written with V.I. Paulsen we prove that two unital dual operator algebras
are stably isomorphic if and only if they have completely isometric normal
representations with TRO equivalent images. These results are generalised,
with V.I.Paulsen and I.G. Todorov, to dual operator spaces [9].
In the present paper we are concerned with the notion of TRO equivalence
within the class of reflexive (not necessarily selfadjoint) algebras. We show
(section 3) that two such algebras are TRO equivalent if and only if there
exists a ∗ isomorphism between the commutants of their diagonals mapping
the invariant projection lattice of the first algebra onto that of the second.
This may be thought as a generalisation to the non-selfadjoint case of the
remark of Connes [2] that two W ∗ algebras are Morita equivalent in the
sense of Rieffel if and only if they have faithful normal representations with
isomorphic commutants.
In section 5 we specialise to the case of TRO equivalence of separably
acting CSL algebras. Given the above criterion for TRO equivalence of
reflexive algebras the problem is the following:
If A, B are separably acting CSL algebras and φ : Lat(A) → Lat(B) is
a lattice isomorphism, under what conditions does φ extend to a ∗ isomor-
phism between the generated von Neumann algebras Lat(A)′′ and Lat(B)′′?
The interesting fact is that while φ always extends to a ∗ isomorphism be-
tween the generated C ∗ algebras (Lemma 5.1), it does not always extend to
the w∗ closures of these algebras (Remark 4.6).
In this paper we also consider an equivalence relation strictly weaker than
TRO equivalence, which we call spatial Morita equivalence (section 4). Two
w∗ closed operator algebras A, B are called spatially Morita equivalent if
there exist an A, B bimodule U and a B, A bimodule V such that A =
span(UV)−w∗
. We show that two CSL algebras are
spatially Morita equivalent if and only if they have isomorphic lattices. In
this case if one of the algebras is "synthetic" then so is the other.
and B = span(VU )−w∗
We present some definitions and concepts used in this work.
TRO EQUIVALENT ALGEBRAS
3
By an algebra A we shall mean an algebra of operators on some Hilbert
space; the diagonal of A is ∆(A) = A ∩ A∗.
A set of projections of a Hilbert space is called a lattice if it contains
the zero and identity projections and is closed under arbitrary suprema and
infima. If A is a subalgebra of B(H) for some Hilbert space H, the set
Lat(A) = {L ∈ pr(B(H)) : L⊥AL = 0}
is a lattice. Dually if L is a lattice the space
Alg(L) = {A ∈ B(H) : L⊥AL = 0 ∀ L ∈ L}
is an algebra.
A lattice L such that P ∈ L ⇔ P ⊥ ∈ L is called an ortholattice. A
commutative subspace lattice (CSL) is a projection lattice L whose elements
commute; the algebra Alg(L) is called a CSL algebra. A totally ordered
CSL is called a nest.
An order-preserving bijection between two lattices is called a lattice iso-
morphism. If the lattices L1, L2 are ortholattices and φ : L1 → L2 is a
lattice isomorphism satisfying φ(P ⊥) = φ(P )⊥ for all P ∈ L1 we call φ an
ortholattice isomorphism.
Let H1, H2 be Hilbert spaces and U a subset of B(H1, H2). The reflexive
hull of U is defined [16] to be the space
Ref(U ) = {T ∈ B(H1, H2) : T x ∈ span(U x) for each x ∈ H1}.
Simple arguments show that
Ref(U ) ={T ∈ B(H1, H2) for all projections E, F : EU F = 0 ⇒ ET F = 0}
={T ∈ B(H1, H2) for all operators A, B : AU B = 0 ⇒ AT B = 0}.
A subspace U is called reflexive if U = Ref(U ). A unital algebra is reflexive if
and only if A = Alg(Lat(A)). CSL algebras are reflexive. Every CSL algebra
contains a maximal abelian selfadjoint algebra (masa in the sequel). Hence
we can view a CSL algebra as a masa bimodule.
Moreover, an algebra is a CSL algebra if and only if it is reflexive and
contains a masa. If U is a reflexive masa bimodule, then there exists [3],
[18] a smallest w∗ closed masa bimodule which is contained in U and whose
reflexive hull is the space U . We denote this space by Umin. Whenever Umin =
U we call the space U synthetic. When A is a separably acting CSL algebra,
the space Amin is an algebra which contains the diagonal of A and whose
lattice is Lat(A) [1]. The first example of a nonsynthetic CSL algebra was
given in [1].
Now we present some concepts introduced in [10].
Let Pi = pr(B(Hi)), i = 1, 2. Define φ = Map(U ) to be the map φ :
P1 → P2 which associates to every P ∈ P1 the projection onto the subspace
span(T P y : T ∈ U , y ∈ H1)−. The map φ is ∨−continuous (that is, it
preserves arbitrary suprema) and 0 preserving.
4
G.K. ELEFTHERAKIS
Let φ∗ = Map(U ∗), S1,φ = {φ∗(P )⊥ : P ∈ P2}, S2,φ = {φ(P ) : P ∈ P1}
and observe that S1,φ = S ⊥
2,φ∗. Erdos proved that S1,φ is ∧-complete and
contains the identity projection, S2,φ is ∨-complete and contains the zero
projection, while φS1,φ : S1,φ → S2,φ is a bijection. We call the families
S1,φ, S2,φ the semilattices of U .
In fact
Ref(U ) = {T ∈ B(H1, H2) : φ(P )⊥T P = 0 for each P ∈ S1,φ}.
When φ(I) = I and φ∗(I) = I we call the space U essential.
In [15] it is proved that a TRO M is w∗ closed if and only if it is wot
closed if and only if it is reflexive. In this case, if χ = Map(M)
M = {T ∈ B(H1, H2) : T P = χ(P )T for all P ∈ S1,χ}.
In the following theorem we isolate some consequences of [15, Theorem 2.10].
Theorem 1.1. (i) A TRO M is essential if and only if the algebras span(M∗M)−w∗
span(MM∗)−w∗
contain the identity operators.
,
(ii) If M is an essential TRO and χ = Map(M) then S1,χ = pr((M∗M)′),
S2,χ = pr((MM∗)′) and the map χS1,χ : S1,χ → S2,χ is an ortholattice iso-
morphism with inverse χ∗S2,χ .
If L ⊂ B(H) we denote by L′ the set of operators which commute with
the elements of L and the set of projections in L by pr(L).
2. TRO equivalent algebras
Definition 2.1. Let A, B be w∗ closed algebras acting on Hilbert spaces H1
and H2 respectively. If there exists a TRO
M ⊂ B(H1, H2) such that A = span(M∗BM)−w∗
and B = span(MAM∗)−w∗
we write A M∼ B. We say that the algebras A, B are TRO equivalent if there
exists a TRO M such that A M∼ B.
A simple example of TRO equivalent, not necessarily selfadjoint algebras,
is the following. Take a unital w∗ closed algebra A ⊂ B(H) and let
B =(cid:20) A A
A A (cid:21) ⊂ B(H ⊕ H), M =(cid:20) ∆(A)
∆(A) (cid:21) ⊂ B(H, H ⊕ H).
It is easy to see that A M∼ B.
Proposition 2.1. Let A ⊂ B(H1), B ⊂ B(H2) be w∗ closed algebras. The
following are equivalent:
(i) The algebras A, B are TRO equivalent.
(ii) There exists an essential TRO M ⊂ B(H1, H2) such that M∗BM ⊂
A and MAM∗ ⊂ B.
If (ii) holds then A M∼ B.
TRO EQUIVALENT ALGEBRAS
5
Proof Let N ⊂ B(H1, H2) be a TRO such that A = span(N ∗BN )−w∗
and B = span(N AN ∗)−w∗
.
If P is the projection onto ker N and Q is the projection onto ker N ∗,
it is clear that M ≡ N + QB(H1, H2)P is an essential TRO such that
M∗BM ⊂ A and MAM∗ ⊂ B.
Conversely, if M is an essential TRO such that M∗BM ⊂ A and MAM∗ ⊂
B, then span(M∗BM)−w∗ ⊂ A and, since MAM∗ ⊂ B, we have
M∗MAM∗M ⊂ M∗BM ⇒
span(M∗M)−w∗
Aspan(M∗M)−w∗
⊂ span(M∗BM)−w∗
.
Since I ∈ span(M∗M)−w∗
it follows that A ⊂ span(M∗BM)−w∗
.
Symmetrically we have B = span(MAM∗)−w∗
.
(cid:3)
We wish to prove that TRO equivalence is an equivalence relation. We
need the following lemma.
Lemma 2.2. Let Si be a set of projections on the Hilbert space Hi, i = 1, 2,
χ : S1 → S2 a map onto S2, and
M = {T ∈ B(H1, H2) : T L = χ(L)T for all L ∈ S1}.
Observe that the space M is a reflexive TRO [15]. Moreover, if we have the
information that it is essential, then
span(M∗M)−w∗
= (S1)′ and span(MM∗)−w∗
= (S2)′.
Proof Let φ = Map(M). We can observe that M(S1)′ ⊂ M, so if P is
a projection then (S1)′M∗P (H2) ⊂ M∗P (H2) therefore (S1)′φ∗(P )(H1) ⊂
φ∗(P )(H1). Since S1 is selfadjoint, it follows that φ∗(P ) ∈ (S1)′′.
We proved that S2,φ∗ ⊂ (S1)′′; thus S1,φ ⊂ (S1)′′ and so (S1,φ)′ ⊃ (S1)′.
= (S1,φ)′ (Theorem 1.1) since M is an essential TRO.
But span(M∗M)−w∗
We proved that span(M∗M)−w∗
⊃ (S1)′.
Clearly, M∗M ⊂ (S1)′ and so span(M∗M)−w∗
Since χ maps onto S2 we see that M∗(S2)′ ⊂ M∗ and similar arguments
= (S1)′.
show that span(MM∗)−w∗
= (S2)′.
(cid:3)
Theorem 2.3. TRO equivalence is an equivalence relation.
Proof We only have to prove transitivity. Let A, B, C be w∗ closed al-
gebras, acting on the Hilbert spaces H1, H2, H3 respecively, and M, N be
essential TRO's such that B M∼ A and B N∼ C. Thus
and
Define
and note that
span(MBM∗)−w∗
span(M∗AM)−w∗
= A, span(N BN ∗)−w∗
= B = span(N ∗CN )−w∗
.
= C
S = pr((M∗M)′ ∩ (N ∗N )′)
S ′ = ((M∗M) ∪ (N ∗N ))′′.
6
G.K. ELEFTHERAKIS
Since M∗MBM∗M ⊂ B and similarly for N it follows that S ′BS ′ ⊂ B. Let
χ = Map(M) and φ = Map(N ). Define the TRO's
Z = {T ∈ B(H2, H1) : T L = χ(L)T, L ∈ S}
Y = {T ∈ B(H2, H3) : T L = φ(L)T, L ∈ S}.
The map χ is an ortholattice isomorphism from pr((M∗M)′) onto pr((MM∗)′)
(Theorem 1.1). Since S ⊂ pr((M∗M)′) it follows that M ⊂ Z. Similarly
N ⊂ Y and thus both Z and Y are essential TRO's. From the previous
lemma we obtain
span(Y ∗Y)−w∗
= S ′ = span(Z ∗Z)−w∗
.
We claim that
span(Z ∗AZ)−w∗
= B and span(ZBZ∗)−w∗
= A.
Indeed since Z ∗Z ⊂ S ′ and M ⊂ Z we have
Z ∗ZBZ ∗Z ⊂ B ⇒MZ ∗(ZBZ ∗)ZM∗ ⊂ MBM∗ ⊂ A
⇒MM∗(ZBZ ∗)MM∗ ⊂ A.
Since M is essential and span(MM∗) (resp. span(M∗M)) is a ∗−algebra,
one can find a bounded net in span(MM∗) (resp. span(M∗M)) converg-
ing strongly to the identity operator. Since A is w∗−closed it follows that
ZBZ∗ ⊂ A and hence span(ZBZ∗)−w∗ ⊂ A.
On the other hand A = span(MBM∗)−w∗ ⊂ span(ZBZ ∗)−w∗
span(ZBZ∗)−w∗
Z ∗ZBZ ∗Z ⊂ B. It follows by Proposition 2.1 that B = span(Z ∗AZ)−w∗
. Therefore Z ∗AZ = Z ∗span(ZBZ∗)−w∗
.
hence A =
Z ⊂ B because
In the same way we have
span(Y ∗CY)−w∗
Now put L = span(YZ∗)−w∗
we have YZ ∗Z ⊂ Y. It follows that
= B and span(YBY ∗)−w∗
. Since span(Y ∗Y)−w∗
= C.
= S ′ = span(Z ∗Z)−w∗
(YZ ∗)(YZ ∗)∗(YZ ∗) = YZ ∗ZY ∗YZ ∗ ⊂ YY ∗YZ ∗ ⊂ YZ ∗
since Y is a TRO, hence LL∗L ⊂ L. Thus the space L is a TRO and it is
essential because the spaces Y and Z ∗ are essential TRO's.
To complete the proof it remains to show that
L∗CL ⊂ A and LAL∗ ⊂ C.
Indeed, since Y ∗CY ⊂ B we have ZY ∗CYZ ∗ ⊂ ZBZ ∗ ⊂ A and thus L∗CL ⊂
A. On the other hand, YZ ∗AZY ∗ ⊂ YBY ∗ ⊂ C and therefore LAL∗ ⊂ C.
(cid:3)
Remark 2.4. From the previous proof it follows that, if A, B are TRO
equivalent algebras and B, C are also TRO equivalent algebras, then there
exist essential TRO's Z, Y generating the same von Neumann algebra such
that B Z∼ A, B Y∼ C and the space L = span(YZ ∗)−w∗
is an essential TRO
satisfying A L∼ C.
TRO EQUIVALENT ALGEBRAS
7
Proposition 2.5. Let A, B be w∗ closed algebras and M an essential TRO
such that A M∼ B. Then ∆(A)
M∼ ∆(B).
Proof It is obvious that M∗∆(B)M ⊂ ∆(A) and M∆(A)M∗ ⊂ ∆(B).
By Proposition 2.1 follows that ∆(A)
M∼ ∆(B).
(cid:3)
Lemma 2.6. Let A, B be unital w∗ closed algebras, M be an essential TRO
such that A M∼ B and χ = Map(M). Then Ref(A)
M∼ Ref(B). Also the
map χ : pr(∆(A)′) → pr(∆(B)′) is an orthoisomorphism and χ(Lat(A)) =
Lat(B).
Proof By the above proposition ∆(A)
M∼ ∆(B). From [15, Corollary 5.9]
it follows that χ(pr(∆(A)′)) = pr(∆(B)′). Since M∗M ⊂ ∆(A) we have
pr((M∗M)′) ⊃ pr((∆(A))′). So by Theorem 1.1 the map χ : pr(∆(A)′) →
pr(∆(B)′) is an orthoisomorphism.
If E, F are projections such that EAF = 0 then EM∗BMF = 0 so
EM∗Ref(B)MF = 0. It follows that
M∗Ref(B)M ⊂ Ref(A).
M∼
Similarly we can prove that MRef(A)M∗ ⊂ Ref(B), hence Ref(A)
Ref(B). Since Lat(Ref(A)) = Lat(A) and similarly for B, using again [15,
Corollary 5.9] we have χ(Lat(A)) = Lat(B).
(cid:3)
Remark 2.7. (i) Let A, B be TRO equivalent w∗-closed algebras and sup-
pose that the algebra A is reflexive. Then the algebra B is reflexive. Indeed
if M is an essential TRO such that A M∼ B as in the proof of Lemma 2.6 it
follows that A M∼ Ref(B), hence Ref(B) = B.
(ii) An orthoisomorphism χ : pr(C) → pr(D), where C and D are von Neu-
mann algebras, does not necessarily extend to a ∗−homomorphism between
the algebras. For example choose [14] nonabelian ∗ anti-isomorphic von
Neumann algebras C, D, θ : C → D a ∗ anti-isomorphism and let χ = θpr(C).
Compare now Lemma 2.6 and Theorem 3.3.
Proposition 2.8. Let A, B be unital w∗ closed algebras acting on the Hilbert
spaces H1, H2 respectively and M be an essential TRO such that A M∼ B.
Then there exists a TRO N which contains M such that A N∼ B and ∆(A) =
span(N ∗N )−w∗
, ∆(B) = span(N N ∗)−w∗
.
Proof Let χ = Map(M). From Lemma 2.6 follows that χ(pr(∆(A)′) =
pr(∆(B)′). Let
N = {T ∈ B(H1, H2) : T L = χ(L)T for all L ∈ pr(∆(A)′)}.
Since S1,χ = pr((M∗M)′) ⊃ pr(∆(A)′) we have that M ⊂ N so the TRO
N is essential.
Using Lemma 2.2 we have ∆(A) = span(N ∗N )−w∗
, ∆(B) = span(N N ∗)−w∗
.
We shall show that A = span(N ∗BN )−w∗
. Since M∗BM ⊂ N ∗BN , we get
8
G.K. ELEFTHERAKIS
A ⊂ span(N ∗BN )−w∗
. Now we have
N N ∗BN N ∗ ⊂ B ⇒M∗N N ∗BN N ∗M ⊂ M∗BM ⊂ A
⇒M∗MN ∗BN M∗M ⊂ A
hence span(M∗M)−w∗N ∗BN span(M∗M)−w∗ ⊂ A.
It follows that N ∗BN ⊂ A therefore A = span(N ∗BN )−w∗
N N ∗ ⊂ B we have N AN ∗ = N span(N ∗BN )−w∗
A N∼ B.
(cid:3)
. Now since
N ∗ ⊂ B. We proved that
We isolate the following consequence of the above proposition:
Corollary 2.9. If the unital w∗ closed algebras A, B are TRO equivalent
then the diagonal algebras ∆(A), ∆(B) are Morita equivalent in the sense of
Rieffel [17].
The following proposition says that if two non-unital w∗ closed algebras
are TRO equivalent, there exist TRO equivalent unital algebras which con-
tain the previous algebras as ideals.
Proposition 2.10. Let A, B be non unital w∗ closed algebras and M be
an essential TRO such that A M∼ B. If AM = span(A, M∗M)−w∗
, BM =
span(B, MM∗)−w∗
, then
(i)The spaces AM, BM are unital algebras.
(ii)The algebra A (respectively B) is an ideal of AM (respectively BM).
(iii)AM
(iv)There exists an essential TRO N which contains M such that A N∼ B,
. (Observe that AM =
, ∆(BM) = span(N N ∗)−w∗
M∼ BM.
∆(AM) = span(N ∗N )−w∗
AN and BM = BN ).
Proof Claims (i),(ii) are consequences of the relations AM∗M ⊂ A,
M∗MA ⊂ A, BMM∗ ⊂ B, MM∗B ⊂ B.
Also, since MAM∗ ⊂ B and M(M∗M)M∗ ⊂ MM∗ it easily follows that
MAMM∗ ⊂ BM. Similarly we get MBMM∗ ⊂ AM.
(iv) Since AM
M∼ BM, by the previous proposition there exists an essential
TRO N containing M such that AM
∆(AM) = span(N ∗N )−w∗
N∼ BM and
, ∆(BM) = span(N N ∗)−w∗
.
It remains to show that A N∼ B.
Since N N ∗ ⊂ BM and B is an ideal of BM we have
N N ∗BN N ∗ ⊂ B ⇒ M∗N N ∗BN N ∗M ⊂ M∗BM ⊂ A
⇒M∗MN ∗BN M∗M ⊂ A ⇒ span(M∗M)−w∗
N ∗BN span(M∗M)−w∗
⊂ A
hence N ∗BN ⊂ A. Similarly we can prove that N ∗AN ⊂ B.
(cid:3)
TRO EQUIVALENT ALGEBRAS
9
Proposition 2.11. Let A, B be w∗ closed algebras and M be an essential
TRO such that A M∼ B. If J is a w∗ closed A−bimodule then the space
is a B−bimodule and J M∼ F (J ). The map F
F (J ) = span(MJ M∗)−w∗
is a bijection between w∗ closed bimodules of A and those of B. Moreover
the restriction of F to the set of two sided w∗ closed ideals of A maps onto
those of B.
Proof Since AM∗M ⊂ A we have MAM∗MJ M∗ ⊂ MJ M∗. Hence
BF (J ) ⊂ F (J ). Similarly we have F (J )B ⊂ F (J ).
If I is a w∗ closed bimodule of B, the space J = span(M∗IM)−w∗
is a
bimodule of A and F (J ) = I. So the map F is onto. Clearly, F is an
injection. Also observe that if J is a two sided w∗ closed ideal of A then
the space F (J ) is a two sided w∗ closed ideal of B.
(cid:3)
The following proposition is proved easily.
Proposition 2.12. Let A, B be w∗ closed algebras and M be an essential
TRO such that A M∼ B. We denote by K(A) (respectively K(B)) the set of
compact operators in A (resp. B), by F(A) (resp. F(B)) the set of finite
rank operators in A (resp. B), by R1(A) (resp. R1(B)) the set of rank 1
operators in A(resp. B). Then it follows
K(A)−w∗ M∼ K(B)−w∗
span(R1(A))−w∗ M∼ span(R1(B))−w∗
, F(A)−w∗ M∼ F(B)−w∗
.
,
3. TRO equivalent reflexive algebras
The goal of this section is to determine sufficient and necessary conditions
for TRO equivalence of reflexive algebras. The following lemma is known.
See for example [2, 8.5.32]. We include a proof for completeness.
Lemma 3.1. Let C, E be von Neumann algebras acting on the Hilbert spaces
H1, H2 respectively , θ : C → E be a ∗ isomorphism and
M = {T ∈ B(H1, H2) : T A = θ(A)T for all A ∈ C}.
Then the space M is an essential TRO.
Proof Let D = {A ⊕ θ(A) : A ∈ C}. Since θ is w∗ continuous, as a
∗ isomorphism between von Neumann algebras [4, I.4.3, Corollaire 2], the
space D is a von Neumann algebra. The commutant of D is the algebra
(cid:20) C′ M∗
M E ′
(cid:21) .
Let φ = Map(M). Since E ′M ⊂ M we have that φ(I)⊥E ′φ(I) = 0, hence
φ(I) ∈ E. Let P = 0 ⊕ φ(I)⊥. Since φ(I)M = M and φ(I) ∈ E we can verify
that P ⊥D′P = 0, hence P ∈ D. It follows that the projection P is of the
form A ⊕ θ(A) for an operator A ∈ C. Thus φ(I) = I. Similarly we can prove
that φ∗(I) = I, so the space M is an essential TRO.
(cid:3)
10
G.K. ELEFTHERAKIS
We give a new proof of Connes' remark (see the introduction) and also
show that the isomorphism between the commutants extends the map of the
Morita equivalence bimodule. This fact will be useful below.
Theorem 3.2. Let A, B be von Neumann algebras acting on the Hilbert
spaces H1, H2 respectively , M be an essential TRO such that A = span(M∗M)−w∗
B = span(MM∗)−w∗
θ : A′ → B′ which extends the map χpr(A′).
Conversely if the algebras A′, B′ are ∗ isomorphic, the algebras A, B are
TRO equivalent.
and χ = Map(M). Then there exists a ∗ isomorphism
,
Proof By Theorem 1.1,
M = {T ∈ B(H1, H2) : T L = χ(L)T for all L ∈ pr(A′)}
Let L = {L ⊕ χ(L) : L ∈ pr(A′)}. We can verify that
C = L′ =(cid:20) A M∗
M B
(cid:21) .
So the algebra C is a von Neumann algebra acting on the direct sum of the
corresponding Hilbert spaces.
An easy calculation shows that the commutant of C is the set
(cid:26)(cid:20) T 0
0 S (cid:21) : T ∈ A′, S ∈ B′ such that SM = M T ∀M ∈ M(cid:27) .
Let
π1 : C′ → A′ :(cid:20) T 0
π2 : C′ → B′ :(cid:20) T 0
0 S (cid:21) → T
0 S (cid:21) → S.
We shall show that the maps π1, π2 are surjective. Clearly π1(C′) is a von
Neumann algebra, so it suffices to show that π1(C′)′ ⊂ A.
0 S (cid:21) ∈ C′. Thus
0 (cid:21) for all (cid:20) T 0
0 S (cid:21) ∈ C′,
If A ∈ π1(C′)′ then AT = T A for all (cid:20) T 0
0 S (cid:21)(cid:20) A 0
0 (cid:21)(cid:20) T 0
0 S (cid:21) =(cid:20) T 0
0 (cid:21) ∈ C, and so A ∈ A.
0
0
(cid:20) A 0
hence (cid:20) A 0
If (cid:20) T 0
0
0 S1 (cid:21) ,(cid:20) T 0
0 S2 (cid:21) ∈ C′ then S1M = M T = S2M for all M ∈ M.
Since the TRO M is essential we have S1 = S2.
The conclusion is that we can define a map θ : A′ → B′ such that
θ(T ) = S ⇔(cid:20) T 0
0 S (cid:21) ∈ C′. The map θ is a ∗ isomorphism. We shall show
that θ is an extension of χpr(A′).
If L ∈ pr(A′) we have M L = θ(L)M for all M ∈ M. It follows θ(L)⊥ML =
TRO EQUIVALENT ALGEBRAS
11
0 and from this χ(L) ≤ θ(L). Also θ(L)ML⊥ = 0 hence χ(L⊥) ≤ θ(L)⊥.
By Theorem 1.1 χ(L⊥) = χ(L)⊥ so χ(L) = θ(L).
Conversely, let θ : A′ → B′ be a ∗ isomorphism and
M = {T : T A = θ(A)T for all A ∈ A′}.
The space M is an essential TRO by the previous lemma. It is obvious that
M∗BM ⊂ A and MAM∗ ⊂ B.
(cid:3)
Theorem 3.3. Two unital reflexive algebras A, B are TRO equivalent if
and only if there exists a ∗ isomorphism θ : ∆(A)′ → ∆(B)′ such that
θ(Lat(A)) = Lat(B).
Proof Let A, B be TRO equivalent algebras, acting on the Hilbert spaces
H1, H2 respectively . By Proposition 2.8 there exists an essential TRO M
such that A M∼ B and ∆(A) = span(M∗M)−w∗
.
From the previous theorem there exists a ∗ isomorphism θ : ∆(A)′ →
∆(B)′ which extends the map χpr(∆(A)′). From Lemma 2.6 θ(Lat(A)) =
χ(Lat(A)) = Lat(B).
Conversely, let θ : ∆(A)′ → ∆(B)′ be a ∗ isomorphism such that θ(Lat(A)) =
Lat(B) and define
, ∆(B) = span(MM∗)−w∗
M = {T ∈ B(H1, H2) : T A = θ(A)T for all A ∈ ∆(A)′}.
By Lemma 3.1 the space M is an essential TRO. It remains to show that
A M∼ B. Let A ∈ A, L ∈ Lat(A), M1, M2 ∈ M then M1AM ∗
2 θ(L) =
M1ALM ∗
2 ∈ B. We
proved that MAM∗
⊂ B. Similarly we can prove that M∗BM ⊂ A.
2 θ(L). Hence M1AM ∗
(cid:3)
2 = M1LALM ∗
2 = θ(L)M1AM ∗
4. TRO equivalence and spatial Morita equivalence
The following definition is due to I. Todorov (personal communication).
Definition 4.1. Let H1, H2 be Hilbert spaces, A ⊂ B(H1), B ⊂ B(H2) be w∗
closed algebras. If there exist linear spaces U ⊂ B(H1, H2), V ⊂ B(H2, H1)
such that BUA ⊂ U , AVB ⊂ V, span(VU )−w∗
= B
we say that the algebras A, B are spatially Morita equivalent and the
system (A, B, U , V) is a spatial Morita context.
= A and span(UV)−w∗
As we prove in Theorem 4.5 and in remark 4.6 spatial Morita equivalence
is strictly weaker relation than TRO equivalence.
Theorem 4.1. Let (A, B, U , V) be a spatial Morita context. Moreover we
assume that the algebras A, B are unital. If φ = Map(U ) and ψ = Map(V)
then
(i) S1,φ = Lat(A), S2,φ = Lat(B), so the map φ : Lat(A) → Lat(B) is a
lattice isomorphism.
(ii) ψLat(B) = (φLat(A))−1.
12
G.K. ELEFTHERAKIS
Proof Let ζ1 = Map(A) and ζ2 = Map(B). Since span(UV)−w∗
= B we
get ζ2 = φ ◦ ψ, hence ζ2(pr(B(H2))) ⊂ S2,φ or equivalently Lat(B) ⊂ S2,φ.
Since ζ1 = ψ ◦ φ, if P ∈ pr(B(H1)), then
UVφ(P )(H2) ⊂ U ψ(φ(P ))(H1) = U ζ1(P )(H1).
Also
UAP ⊂ U P ⇒ U ζ1(P )(H1) ⊂ φ(P )(H2) ⇒ UVφ(P )(H2) ⊂ φ(P )(H2).
We proved that φ(P )⊥UVφ(P ) = 0 and therefore φ(P )⊥Bφ(P ) = 0 for all
P ∈ pr(B(H1)). It follows that S2,φ ⊂ Lat(B), hence we have equality.
Since (A∗, B∗, U ∗, V ∗) is a spatial Morita context, using the previous ar-
guments we have S1,φ = Lat(A).
Observe that φ : Lat(A) → Lat(B) is a bijection which preserves or-
der. Since ψ ◦ φ = ζ1, which is the identity on Lat(A), it follows that
ψ ◦ (φLat(A)) = IdLat(A). Similarly φ ◦ (ψLat(B)) = IdLat(B). The conclusion
is that ψLat(B) = (φLat(A))−1.
(cid:3)
Remark 4.2. If A, B are spatially Morita equivalent unital algebras and the
algebra B is reflexive, the algebra A is reflexive too. Indeed, let (A, B, U , V)
If E, F are projections such that EBF =
be a spatial Morita context.
0, we have EU (VU )VF = E(UV)(UV)F = 0 hence EUAVF = 0 and so
EU Ref(A)VF = 0. We proved that U Ref(A)V ⊂ B. But
U Ref(A)V ⊂ B ⇒ VU Ref(A)VU ⊂ VBU ⊂ A ⇒ ARef(A)A ⊂ A.
Since the algebra A is unital we have Ref(A) = A.
In the following theorem we prove that the converse of the above theorem
is true for the case of CSL algebras.
Theorem 4.3. Two CSL algebras A and B are spatially Morita equivalent
if and only if they have isomorphic lattices.
Proof By Theorem 4.1 it suffices to show that a lattice isomorphism
between CSL's induces spatial Morita equivalence of the corresponding al-
gebras. Suppose that A ⊂ B(H1) and B ⊂ B(H2).
Let S1 = Lat(A), S2 = Lat(B), let φ : S1 → S2 be a lattice isomorphism and
U = {T ∈ B(H1, H2) : φ(L)⊥T L = 0 for all L ∈ S1},
V = {S ∈ B(H2, H1) : L⊥Sφ(L) = 0 for all L ∈ S1}.
It is easily verified that span(VU) is an ideal of A. Indeed if V ∈ V, U ∈ U
and A ∈ A then for all L ∈ S1 we have
AV φ(L) = ALV φ(L) = LALV φ(L)
so L⊥AV φ(L) = 0, hence AV ∈ V. Similarly U A ∈ U , showing that
Aspan(VU)A ⊂ span(VU). Also
V U L = V φ(L)U L = LV φ(L)U L
TRO EQUIVALENT ALGEBRAS
13
and so L⊥V U L = 0; hence V U ∈ A. It follows that Ref(VU ) ⊂ A. We shall
prove that equality holds.
By Theorems 3.3, 4.4 in [10] we have that
S1,Map(U ) = S1, S2,Map(U ) = S2, Map(U )S1 = φ,
S1,Map(V) = S2, S2,Map(V) = S1, Map(V)S2 = φ−1.
Let W = Ref(VU ) and ζ = Map(W). It follows that ζ = Map(V) ◦ Map(U ).
Also since W ∗ = Ref(U ∗V ∗) we have ζ ∗ = Map(U ∗)◦Map(V ∗), hence S2,ζ ∗ ⊂
S2,Map(U ∗) = (S1,Map(U ))⊥ = (S1)⊥.
We conclude that S1,ζ ⊂ S1. So if L ∈ S1,ζ we have Map(U )(L) ∈ S2 and
ζ(L) = Map(V) ◦ Map(U )(L) = φ−1 ◦ φ(L) = L,
hence
W ={T : ζ(L)⊥T L = 0 for all L ∈ S1,ζ}
={T : L⊥T L = 0 for all L ∈ S1,ζ}
⊃{T : L⊥T L = 0 for all L ∈ S1} = A
Since VU ⊂ A we obtain the equality A = Ref(VU ).
Observe that the space span(VU )−w∗
is a masa bimodule, so it con-
tains the space Amin. But the space Amin is a unital space and the space
span(VU )−w∗
. Similarly
we can prove that B = span(UV)−w∗
is an ideal of A. It follows that A = span(VU )−w∗
(cid:3)
.
Remark 4.4. We do not know whether the 'product' span(VU )−w∗
reflexive spaces,or even reflexive masa bimodules, is necessarily reflexive.
of two
Theorem 4.5. If A, B are w∗ closed TRO equivalent algebras then they are
spatially Morita equivalent.
Proof Let M be an essential TRO such that A M∼ B and put AM =
. We recall from Proposition
M∼ BM and A, (respectively B) is an ideal of AM, (respectively
, BM = span(B, MM∗)−w∗
span(A, M∗M)−w∗
2.10 that AM
BM).
Let U = span(BM)−w∗
and V = span(M∗BM)−w∗
. We shall show that
the system (A, B, U , V) is a spatial Morita context.
(i) Since B is an ideal of BM we have B(BM)(M∗BM) ⊂ BM ⊂ U .
, this implies that
and A = span(M∗BM)−w∗
Since U = span(BM)−w∗
BUA ⊂ U .
(ii) Similarly, the relation (M∗BM)(M∗BM)B ⊂ M∗B implies AVB ⊂ V
since V = span(M∗BM)−w∗
.
(iii) Observe that M∗ ⊂ V, hence M∗BM ⊂ VU . It follows that
span(VU )−w∗ ⊃ A. Since B is an ideal of BM we have M∗BMBM ⊂
M∗BM ⊂ A. But V = span(M∗BM)−w∗
. It follows
that VU ⊂ A. Therefore A = span(VU )−w∗
and U = span(BM)−w∗
.
14
G.K. ELEFTHERAKIS
(iv) Since B is an ideal of BM we have BMM∗BM ⊂ B hence UV ⊂ B.
Now, observe that
MM∗BM ⊂ MV ⇒ span(MM∗)−w∗
BM ⊂ span(MV)−w∗
.
BM ⊂ span(MV)−w∗
⇒
Also since UV ⊃ BMV, it follows that
span(UV)−w∗
⊃ Bspan(MV)w∗
⊃ BBM = B.
We proved that B = span(UV)−w∗
. The proof is complete.
(cid:3)
Remark 4.6. Spatial Morita equivalence does not imply TRO equivalence.
There exist multiplicity free nests S1, S2 which are isomorphic but the alge-
bras S ′′
2 are not isomorphic. For an example, see [3, Example 7.19].
1 , S ′′
Thus isomorphism of the lattices does not guarantee TRO equivalence,
even for multiplicity free nest algebras.
Theorem 4.7. Let A ⊂ B(H1), B ⊂ B(H2) be separably acting CSL
algebras with isomorphic lattices. Then A is synthetic if and only if B is
synthetic. In fact, if φ : Lat(A) → Lat(B) is a lattice isomorphism and
U = {T ∈ B(H1, H2) : φ(L)⊥T L = 0 for all L ∈ Lat(A)}.
Then A (and B) is synthetic if and only if U is synthetic.
Proof Let S1 = Lat(A),
L = {φ(L) ⊕ L : L ∈ S1}
and
V = {S ∈ B(H2, H1) : L⊥Sφ(L) = 0 for all L ∈ S1}.
By Theorem 4.3 we have that
A = span(VU )−w∗
and B = span(UV)−w∗
.
It is shown in [15, Proposition 4.2] that, if C = Alg(L),
C =(cid:20) B U
V A (cid:21) and Cmin ⊂(cid:20) Bmin Umin
Vmin Amin (cid:21) .
We shall show that
(1)
Cmin =(cid:20) Bmin Umin
Vmin Amin (cid:21) .
Indeed if W is any w∗ closed masa bimodule such that Ref(W) = C and
if Q = 0 ⊕ I then
It follows that
Ref(Q⊥WQ) = Q⊥CQ =(cid:20) 0 U
0 0 (cid:21) .
(cid:20) 0 Umin
0 0
(cid:21) =(cid:20) 0 U
0 0 (cid:21)min
⊂ Q⊥WQ ⊂ W.
TRO EQUIVALENT ALGEBRAS
15
Now taking W = Cmin we obtain
(cid:20) 0 Umin
0 0
(cid:21) ⊂ Cmin.
Similarly we can prove that
(cid:20) 0 0
0 Amin (cid:21) ,(cid:20)
and (1) follows.
0
0
Vmin 0 (cid:21) ,(cid:20) Bmin 0
0 (cid:21) ⊂ Cmin
0
If E, F are projections such that EUminVF = 0 then EUVF = 0 hence
EBF = 0. It follows that B ⊂ Ref(UminV) and therefore B = Ref(UminV).
Similarly A = Ref(VminU ).
Now suppose that the algebra A is synthetic. Since B = Ref(UminV) we
have that Bmin ⊂ span(UminV)−w∗
(2)
U ⊂ BminU ⊂ span(UminVU )−w∗
so
⊂ span(UminA)−w∗
= span(UminAmin)−w∗
.
Using (1) we have
(cid:20) 0 UminAmin
0 0
(cid:21) =(cid:20) 0 Umin
0 0
(cid:21)(cid:20) 0 0
0 Amin (cid:21) ⊂ Cmin.
It follows that UminAmin ⊂ Umin, hence U ⊂ Umin (from equation (2))
and so the bimodule U is synthetic.
For the opposite direction we suppose that the bimodule U is synthetic.
Since
(cid:20) 0 0
0 VminUmin (cid:21) =(cid:20)
0
0
Vmin 0 (cid:21)(cid:20) 0 Umin
0 0
(cid:21) ,
again using (1) we conclude that VminUmin ⊂ Amin and therefore VminU ⊂
Amin. Since UA ⊂ U it follows that VminUA ⊂ Amin. But also Amin ⊂
span(VminU )−w∗
since A = Ref(VminU ), and hence AminA ⊂ Amin; there-
fore A ⊂ Amin since Amin is unital.
We have proved that the algebra A is synthetic if and only if the bimodule
U is synthetic. Similarly one shows that U is synthetic if and only if the
algebra B is synthetic.
(cid:3)
5. TRO equivalence and CSL algebras
In this section we assume that all Hilbert spaces are separable. Thus
the w∗ topology on bounded sets of operators is metrisable. We are going
present some results on TRO equivalence of CSL algberas.
Definition 5.1. If S is a CSL and L ∈ S we denote by L♭ the projection
∨{M ∈ S : M < L}. Whenever L♭ < L we call the projection L − L♭
an atom of S. If the CSL S has no atoms we say that it is a continuous
CSL. If the identity operator is the sum of the atoms we say that S is totally
atomic.
16
G.K. ELEFTHERAKIS
Lemma 5.1. Let Hi, i = 1, 2 be Hilbert spaces, Si, i = 1, 2 be commutative
lattices (not necessarily complete) containing zero and the identity and let
θ : S1 → S2 be a lattice isomorphism. Then the map θ extends to a ∗
isomorphism ρ : span(S1)−k·k → span(S2)−k·k.
{1, ..., n − 1}.
i=1 ciθ(Pi).
Proof Using induction we shall prove that if P1, ...Pn are projections of S1
i=1 ciθ(Pi) = 0.
The claim clearly holds for n = 1, 2. Assume that it holds for k ∈
i=1 ciPi = 0 where ci 6= 0 for 1 ≤ i ≤ n then Pn
i=1 ciPi = 0 where ci 6= 0 for 1 ≤ i ≤ n and put A =Pn
It suffices to show that θ(Pk)A = 0 for all k ∈ {1, ...n}.
such that Pn
Let Pn
Multiply the equation Pn
(c1 + cn)(P1 ∧ Pn) + c2(P2 ∧ P1 ∧ Pn) + . . . + cn−1(Pn−1 ∧ P1 ∧ Pn) = 0.
Let B = θ(Pn)A. We shall show that B = 0.
i=1 ciPi = 0 with P1 ∧ Pn. This gives
By the inductive hypothesis we have
(c1 + cn)θ(P1 ∧ Pn) + c2θ(P2 ∧ P1 ∧ Pn) + . . . + cn−1θ(Pn−1 ∧ P1 ∧ Pn) = 0,
hence θ(P1)B = 0.
Similarly we can prove that θ(Pi)B = 0 for 1 ≤ i ≤ n − 1.
Since Pn = (−cn)−1Pi6=n ciPi it follows that Pn ≤ ∨i6=nPi hence θ(Pn) ≤
∨i6=nθ(Pi), so θ(Pn)B = 0 and therefore B = 0. We proved that θ(Pn)A = 0.
Using the same method we have θ(Pk)A = 0, k = 1, ...n. So the claim holds.
The conclusion is that the map
n
ciθ(Pi)
ciPi! =
is well defined, and it is clearly a ∗ isomorphism.
ρ : span(S1) → span(S2) : ρ n
Xi=1
Xi=1
We shall show that ρ is norm continuous. Let T =Pn
i=1 ciPi ∈ span(S1)
and let c be in the spectrum σ(ρ(T )) of ρ(T ). Let S0 be the smallest lattice
containing the set {0, P1, ...Pn, I}. Then the space span(S0) is a C ∗-algebra
which is contained in span(S1). If c is not in the spectrum σ(T ) of T , the
operator S = (cI − T )−1 is contained in span(S0) and hence in span(S1).
Since S(cI−T ) = (cI−T )S = I we have ρ(S)(cI−ρ(T )) = (cI−ρ(T ))ρ(S) =
I, contradicting c ∈ σ(ρ(T )).
We proved that σ(ρ(T )) ⊂ σ(T ). Therefore kρ(T )k ≤ kT k.
We conclude that the map ρ extends to a ∗ isomorphism from the C ∗-
algebra span(S1)−k·k onto span(S2)−k·k.
(cid:3)
Lemma 5.2. Let S1, S2 be CSL's, φ : S1 → S2 be a lattice isomorphism, P
be the sum of the atoms of S1 and Q be the sum of the atoms of S2. Then
there exists a ∗ isomorphism
such that ρ(LP ) = φ(L)Q for all L ∈ S1.
ρ : S ′′
1 P → S ′′
2 Q
TRO EQUIVALENT ALGEBRAS
17
Proof For any atom A in a CSL S, we have
A = ∧{L ∈ S : AL = A} − ∨{L ∈ S : AL = 0}.
Letting Ai be the set of atoms of Si, we then find that the lattice isomor-
phism φ : S1 → S2 induces a bijection ρα : A1 → A2, given by
ρα(A) = ∧{φ(L) : L ∈ S : AL = A} − ∨{φ(L) : L ∈ S : AL = 0}.
1 P and S ′′
2 Q are ∗ isomorphic to l∞(A1) and l∞(A2) respectively;
Now S ′′
2 Q with l∞(A2). The map
identify S ′′
ρα extends to an isomorphism ρ : l∞(A1) → l∞(A2), which satisfies our
requirements, because any projection in l∞(Ai) is the sum of the atoms it
contains.
1 P with l∞(A1) and likewise identify S ′′
(cid:3)
The following theorem is consequence of the above lemma.
Theorem 5.3. Let S1, S2 be totally atomic CSL's. The algebras Alg(S1), Alg(S2)
are TRO equivalent if and only if the CSL's S1, S2 are isomorphic.
Proof If Alg(S1) and Alg(S2) are TRO equivalent, then S1 and S2 are
isomorphic by Theorem 3.3. Conversely, by the above lemma any lattice iso-
morphism φ : S1 → S2 extends to a ∗ isomorphism from S ′′
1 = (∆(Alg(S1)))′
2 = (∆(Alg(S2)))′. Using again Theorem 3.3 we conclude that the
onto S ′′
algebras Alg(S1) and Alg(S2) are TRO equivalent.
(cid:3)
For the general case of CSL algebras we present the following theorem.
Theorem 5.4. Let S1, S2 be CSL's acting on Hilbert spaces H1, H2 respec-
tively, P the sum of the atoms of S1, Q the sum of the atoms of S2 and
A = Alg(S1), B = Alg(S2), A0 = span(S1)−k·k, B0 = span(S2)−k·k. The
following are equivalent:
(i) The algebras A, B are TRO equivalent.
(ii)There exists a lattice isomorphism φ : S1 → S2 whose extension
(Lemma 5.1) φ : A0 → B0 is w∗-bicontinuous on the unit balls of A0, B0.
(iii) There exists a lattice isomorphism φ : S1 → S2 such that if L =
{L ⊕ φ(L) : L ∈ S1} then
L′′ ∩ (0 ⊕ B′′
0 ) = 0, L′′ ∩ (A′′
0 ⊕ 0) = 0.
(iv) There exists a lattice isomorphism φ : S1 → S2 such that if L is as
in (iii) then
L′′ ∩ (0 ⊕ B′′
0 Q⊥) = 0, L′′ ∩ (A′′
0P ⊥ ⊕ 0) = 0.
Moreover if these conditions hold and
∆(φ) = {T ∈ B(H1, H2) : T L = φ(L)T for all L ∈ S1}
then A
∆(φ)
∼ B.
Proof (i)⇒(ii)
This is obvious by Theorem 3.3
18
G.K. ELEFTHERAKIS
(ii)⇒(i)
Suppose that φ : S1 → S2 is a lattice isomorphism whose extension by
Lemma 5.1, φ : A0 → B0 is w∗-bicontinuous on the unit balls. By [14,
) extends to a w∗-continuous
Lemma 10.1.10] the map φ (respectively φ
homomorphism from (S1)′′ to (S2)′′ (respectively from (S2)′′ to (S1)′′). One
can check that the extensions are mutual inverses. (The assumption that the
map φ is w∗-continuous doesn't guarantee that its inverse is w∗-continuous.
See exercise 10.5.30 in [14]).
−1
Now Theorem 3.3 shows that the algebras A and B are TRO equivalent.
(i)⇒(iii)
If the algebras A, B are TRO equivalent, by Theorem 3.3 there exists a lattice
1 → S ′′
isomorphism φ : S1 → S2 which extends to a ∗ isomorphism ρ : S ′′
2 .
We can verify that L′′ = {A ⊕ ρ(A) : A ∈ A′′
0} hence
L′′ ∩ (0 ⊕ B′′
0 ) = 0, L′′ ∩ (A′′
0 ⊕ 0) = 0.
(iii)⇒(iv)
This is obvious.
−1
(iv)⇒(i)
It suffices to show that φ extends to a ∗ isomorphism from S ′′
2 . If
it does not by (ii) one of the maps φ : A0 → B0, φ
: B0 → A0 will not
be w∗ continuous on the unit ball. Suppose that φ is not w∗ continuous on
the unit ball. Then there exists a net (Ai) ⊂ Ball(A0) which converges in
the w∗ topology to 0 while the net (φ(Ai)) converges to a nonzero operator
B ∈ B′′
0 . Since the restriction of φ on the lattice S1P extends (Lemma 5.2)
to a ∗ isomorphism from A′′
0 Q and the net (AiP ) converges to 0
the net (φ(Ai)Q) converges to 0 too. Therefore BQ = 0.
0P onto B′′
1 onto S ′′
Observe that (L − L♭) ⊕ (φ(L) − φ(L)♭) ∈ L′′ for all L ∈ S1, hence
P ⊕ Q ∈ L′′. It follows that AiP ⊥ ⊕ φ(Ai)Q⊥ ∈ L′′ for every index i and so
0 ⊕ BQ⊥ ∈ L′′. This is a contradiction because BQ⊥ 6= 0. The proof for the
case where φ
is not w∗ continuous on the unit ball is similar.
−1
Now suppose that conditions (ii) to (v) hold and let ρ : S ′′
1 → S ′′
2 be the
extension of φ. By Lemma 3.1 the space
M = {T ∈ B(H1, H2) : T A = ρ(A)T for all A ∈ S ′′
2 }
is an essential TRO. Since the space ∆(φ) contains M it is essential too.
We can easily verify that ∆(φ)∗B∆(φ) ⊂ A and ∆(φ)A∆(φ)∗ ⊂ B. By
Proposition 2.1 we have A
∆(φ)
∼ B.
(cid:3)
Remark 5.5. By Theorem 3.3, if the algebras A and B are TRO equivalent
and A is a CSL algebra then so is B.
In the special case of nest algebras we have the following result:.
Theorem 5.6. All nest algebras with continuous nests are TRO equivalent.
TRO EQUIVALENT ALGEBRAS
19
Proof If X is a subset of some B(H) we denote by X ∞ the set of all
operators of the form X∞ = X ⊕ X ⊕ ...., where X ∈ X , acting on B(H ∞).
If S1, S2 are continuous nests, the nests S∞
2 are also continuous and
1 , S∞
of multiplicity ∞. It follows from [3, Theorem 7.24] that the nests S∞
2
1 )∞ onto
are unitarily equivalent. So there exists a ∗ isomorphism from (S ′′
(S ′′
i →
1 onto S ′′
(S ′′
2
mapping S1 onto S2. The conclusion comes from Theorem 3.3.
2 )∞ mapping S∞
i )∞, X → X∞, i = 1, 2 we obtain a ∗ isomorphism from S ′′
1 , S∞
1 onto S∞
2 . Now taking compositions with the maps S ′′
(cid:3)
Acknowledgement: I would like to express appreciation to Prof. A. Katavo-
los for his helpful comments and suggestions during the preparation of
this work, which is part of my doctoral thesis.
I am also indebted to
Dr. I. Todorov for helpful discussions and important suggestions. Finally, I
am grateful to the referee for carefully reading the paper and making signif-
icant improvements. In particular I wish to thank him for pointing out an
error in a previous version and for providing a quicker proof of Lemma 5.2.
References
[1] W. B. Arveson, Operator algebras and invariant subspaces, Ann. of Math. 100 (2):
433-532, 1974.
[2] D.P. Blecher, C. Le Merdy, Operator algebras and their modules, London Mathemat-
ical Society Monographs, 2004.
[3] K.R. Davidson, Nest algebras, volume 191 of Pitman Research Notes in Mathematics
Series. Longman Scientific & Technical, Harlow, 1988. Triangular forms for operator
algebras on Hilbert space.
[4] J. Dixmier, Les Alg`ebres d' Op´erateurs dans l' Espace Hilbertien, Gauthier-Villars,
Paris, 1969.
[5] E. Effros, N. Ozawa and Z-J Ruan, On injectivity and nuclearity for operator spaces,
Duke Math. J. 110(2001), 489-521.
[6] G.K. Eleftherakis, A Morita type equivalence for dual operator algebras, J. Pure
Appl. Algebra 212 (2008), no 5, 1060-1071.
[7] G.K. Eleftherakis, Morita type equivalences and reflexive algebras, to appear J. Op-
erator Theory, Arxiv: 0709.0600.
[8] G.K. Eleftherakis, V.I. Paulsen, Stably isomorphic dual operator algebras, Math.
Annalen 341 (2008), no 1, 99-112.
[9] G.K. Eleftherakis, V.I. Paulsen, I.G. Todorov, Stable isomorphism of dual operator
spaces, to appear J. Funct. Anal., Arxiv: 0812.2639.
[10] J.A. Erdos, Reflexivity for subspace maps and linear spaces of operators, Proc. London
Math. Soc. 52 (3): 582-600, 1986.
[11] L. Harris, A generalisation of C ∗-algebras, Proc. London Math. Soc. 42(1981), 331-
361.
[12] M. Hestenes, A ternary algebra with applications to matrices and linear transforma-
tions, Arch. Rational Mech. Anal. 11 (1961), 1315-1357.
[13] M. Kaur and Z-J. Ruan, Local properties of ternary rings of operators and their
linking C ∗-algebras, J. Funct. Anal. (195)2002, 262-305.
[14] R.V. Kadison, J.R. Ringrose, Fundamentals of the theory of operator algebras, vol. I
and II, Academic Press, Inc., 1986.
[15] A. Katavolos and I.G. Todorov, Normalisers of operator algebras and reflexivity, Proc.
London Math. Soc. (3) 86: 463-484, 2003.
20
G.K. ELEFTHERAKIS
[16] A.I. Loginov and V.S. Shulman, Hereditary and intermediate reflexivity of W*-
algebras, (Russian) Izv. Akad. Nauk. SSSR Ser. Mat., 39 1260 -- 1273, 1975 (English
translation : Math. USSR-Izv., 9, 1189 -- 1201, 1975.
[17] M.A. Rieffel, Morita equivalence for C ∗−algebras and W ∗−algebras, Journal of Pure
and Applied Algebra 5: 51-96, 1974.
[18] V. Shulman and L. Turowska, Operaror synthesis. I. Synthetic sets, bilattices and
tensor algebras, Journal of Functional Analysis 209(2004) 293-331.
[19] Zhong-Jin Ruan, Type Decomposition and the Rectangular AFD Property for W ∗-
TRO's, Canad. J. Math. 56(2004), no 4, 843-870.
[20] H. Zettl, A characterization of ternary rings of operators, Adv. in Math., 48(1983),
117-143.
GEORGE K. ELEFTHERAKIS, Department of Mathematics,, University of
Athens,, Panepistimioupolis 157 84, Athens Greece
E-mail address: [email protected]
|
1803.02050 | 1 | 1803 | 2018-03-06T07:48:29 | Characterizations of Jordan derivations on algebras of locally measurable operators | [
"math.OA"
] | We prove that if $\mathcal M$ is a properly infinite von Neumann algebra and $LS(\mathcal M)$ is the local measurable operator algebra affiliated with $\mathcal M$, then every Jordan derivation from $LS(\mathcal M)$ into itself is continuous with respect to the local measure topology $t(\mathcal M)$. We construct an extension of a Jordan derivation from $\mathcal M$ into $LS(\mathcal M)$ up to a Jordan derivation from $LS(\mathcal M)$ into itself. Moreover, we prove that if $\mathcal M$ is a properly von Neumann algebra and $\mathcal A$ is a subalgebra of $LS(\mathcal M)$ such that $\mathcal M\subset\mathcal A$, then every Jordan derivation from $\mathcal A$ into $LS(\mathcal M)$ is continuous with respect to the local measure topology $t(\mathcal M)$. | math.OA | math |
CHARACTERIZATIONS OF JORDAN DERIVATIONS ON
ALGEBRAS OF LOCALLY MEASURABLE OPERATORS
GUANGYU AN AND JUN HE
Abstract. We prove that if M is a properly infinite von Neumann algebra
and LS(M) is the local measurable operator algebra affiliated with M, then
every Jordan derivation from LS(M) into itself is continuous with respect
to the local measure topology t(M). We construct an extension of a Jordan
derivation from M into LS(M) up to a Jordan derivation from LS(M) into
itself. Moreover, we prove that if M is a properly von Neumann algebra and
A is a subalgebra of LS(M) such that M ⊂ A, then every Jordan derivation
from A into LS(M) is continuous with respect to the local measure topology
t(M).
1. Introduction
Let R be an associative ring. For an integer n ≥ 2, R is said to be n-torsion-
free if na = 0 implies that a = 0 for every a in R. Recall that a ring R is prime
if aRb = (0) implies that either a = 0 or b = 0 for each a and b in R; and is
semiprime if aRa = (0) implies that a = 0 for every a in R. For each a and b in
R, we denote by [a, b] = ab − ba.
Suppose that M is a R-bimodule. An additive mapping δ from R into M is
called a derivation if δ(ab) = δ(a)b + aδ(b) for each a and b in R; δ is called a
Jordan derivation if δ(a2) = δ(a)a + aδ(a) for every a in R; and δ is called an
inner derivation if there exists an element m in M, such that δ(a) = [m, a] for
every a in R. Obviously, every derivation is a Jordan derivation and every inner
derivation is a derivation. The converse is, in general, not true.
A classical result of Herstein [15] proves that every Jordan derivation on a 2-
torsion-free prime ring is a derivation; in [9], Bresar and Vukman give a brief proof
of [15, Theorem 3.1]. In [12], Cusack generalizes [15, Theorem 3.1] to 2-torsion-
free semiprime rings; in [8], Bresar gives an alternative proof of [12, Corollary 5].
Let A be a C ∗-algebra and M be a Banach A-bimodule, in [20], Ringrose shows
that every derivation from A into M is continuous; in [14], Hejazian and Niknam
prove that if A is commutative, then every Jordan derivation from A into M is
continuous; in [16], Johnson shows that every continuous Jordan derivation from
A into M is a derivation; by Cunts [11, Corollary 1.2] and Johnson [16, Theorem
6.3], we know that every Jordan derivation from A into M is a derivation. In
Date: Received: xxxxxx; Revised: yyyyyy; Accepted: zzzzzz.
∗ Corresponding author.
2010 Mathematics Subject Classification. 46L57, 46L51, 46L52.
Key words and phrases. Jordan derivation, von Neumann algebra, locally measurable oper-
ators algebra, continuous.
1
2
G. AN AND J. HE
[21], Sakai proves that every derivation on a von Neumann algebra is an inner
derivation.
Compare with the characterizations of derivations on Banach algebras, inves-
tigation of derivations on unbounded operator algebras begin much later.
In [22], Segal studies the theory of noncommutative integration, and introduces
various classes of non-trivial ∗-algebras of unbounded operators. In this paper,
we mainly consider the ∗-algebra S(M) of all measurable operators and the ∗-
algebra LS(M) of all locally measurable operators affiliated with a von Neumann
algebra M. In [22], Segal shows that the algebraic and topological properties of
the measurable operators algebra S(M) are similar to the von Neumann algebra
M. If M is a commutative von Neumann algebra, then M is ∗-isomorphic to the
algebra L∞(Ω, Σ, µ) of all essentially bounded measurable complex functions on a
measure space (Ω, Σ, µ); and S(M) is ∗-isomorphic to the algebra L0(Ω, Σ, µ) of
all measurable almost everywhere finite complex-valued functions on (Ω, Σ, µ). In
[4], Ber, Chilin and Sukochev show that there exists a derivation on L0(0, 1) is not
an inner derivation, and the derivation is discontinuous in the measure topology.
This result means that the properties of derivations on S(M) are different from
the derivations on M.
In [1, 2], Albeverio, Ayupov and Kudaybergenov study the properties of deriva-
tions on various classes of measurable algebras. If M is a type I von Neumann
algebra, in [1], the authors prove that every derivation on LS(M) is an inner
derivation if and only if it is Z(M) linear; in [2], the authors give the decompo-
sition form of derivations on S(M) and LS(M); they also prove that if M is a
type I∞ von Neumann algebra, then every derivation on S(M) or LS(M) is an
inner derivation. If M is a properly infinite von Neumann algebra, in [5], Ber,
Chilin and Sukochev prove that every derivation on LS(M) is continuous with
respect to the local measure topology t(M); and in [6], the authors show that
every derivation on LS(M) is an inner derivation.
Similar to the derivations and Jordan derivations, in [10], M. Bresar and J.
Vukman introduce the concepts of left derivations and Jordan left derivations.
Let R be a ring and M be a left R-module. A linear mapping δ from R into
M is called a left derivation if δ(ab) = aδ(b) + bδ(a) for each a, b in R; and δ is
called a Jordan left derivation if δ(a2) = 2aδ(a) for every a in R.
In [24], Vukman shows that every Jordan left derivation from a complex semisim-
In [7], we prove that every Jordan left
ple Banach algebra into itself is zero.
derivation from a C ∗-algebra into its Banach left module is zero.
If M is a type In von Neumann algebra, in [2], Albeverio, Ayupov and Kuday-
bergenov show that S(M) (resp. LS(M)) is ∗-isomorphic to a matrix algebra, by
[3, Theorem 3.1], we know that every Jordan derivation on S(M) (resp. LS(M))
is a derivation. If M is a type III von Neumann algebra, in [17], Murator and
Chilin show that S(M) = M, then every Jordan derivation on S(M) is an inner
derivation.
This paper is organized as follows. In Section 2, we introduce some notations
and recall the definitions of measurable operator algebras and local measurable
operator algebras.
JORDAN DERIVATIONS ON SOME ALGEBRAS
3
In Section 3, we prove that if M is a properly infinite von Neumann algebra,
then every Jordan derivation from LS(M) into itself is continuous with respect
to the local measure topology t(M).
In Section 4, we construct an extension of a Jordan derivation from M into
LS(M) up to a Jordan derivation from LS(M) into itself. Moreover, we prove
that if M is a properly von Neumann algebra and A is a subalgebra of LS(M)
such that M ⊂ A, then every Jordan derivation from A into LS(M) is continuous
with respect to the local measure topology t(M).
2. Preliminaries
Let H be a complex Hilbert space and B(H) be the algebra of all bounded
linear operators on H. Suppose that M is a von Neumann algebra on H and
Z(M) = M ∩ M′ is the center of M, where
M′ = {a ∈ B(H) : ab = ba for every b in M}.
Denote by P(M) = {p ∈ M : p = p∗ = p2} the lattice of all projections in M and
by Pf in(M) the set of all finite projections in M. For each p and q in P(M), if
we define the inclusion relation p ⊂ q by p ≤ q, then P(M) is a complete lattice.
Suppose that {pl}l∈λ is a family of projections in M, we denote by
sup
l∈λ
pl = [l∈λ
plH and inf
l∈λ
pl = \l∈λ
plH,
if {pl}l∈λ is an orthogonal family of projections in M, then we have that
sup
l∈λ
pl = Xl∈λ
pl.
Let x be a closed densely defined linear operator on H with the domain D(x),
where D(x) is a linear subspace of H. x is said to be affiliated with M, denote
by xηM, if u∗xu = x for every unitary element u in M′.
A closed densely defined linear operator xηM is said to be measurable with
respect to M, if there exists a sequence {pn}∞
n=1 ⊂ P(M) such that pn ↑ 1,
pn(H) ⊂ D(x) and p⊥
n = 1 − pn ∈ Pf in(M) for every n ∈ N, where N is the
set of all natural numbers. Denote by S(M) the set of all measurable operators
affiliated with the von Neumann algebra M.
A closed densely defined linear operator xηM is said to be locally measurable
with respect to M, if there exists a sequence {zn}∞
n=1 ⊂ P(Z(M)) such that
zn ↑ 1 and znx ∈ S(M) for every n ∈ N. Denote by LS(M) the set of all locally
measurable operators affiliated with the von Neumann algebra M.
In [17], Murator and Chilin prove that S(M) and LS(M) are both unital ∗-
algebras and M ⊂ S(M) ⊂ LS(M); the authors also show that if M is a finite
von Neumann algebra or dim(Z(M)) < ∞, then S(M) = LS(M); if M is a
type III von Neumann algebra and dim(Z(M)) = ∞, then S(M) = M and
LS(M) 6= M.
Let E be a subset of LS(M). Denote by Eh and E+ the sets of all self-adjoint
elements and all positive elements in E, respectively. The partial order in LS(M)
is defined by its cone LS+(M) and is denoted by ≤.
4
G. AN AND J. HE
Suppose that x is a closed operator with a dense domain D(x) in H. Let
x = ux be the polar decomposition of x, where x = (x∗x)
2 and u is a partial
isometry in B(H). Denote by l(x) = uu∗ the left support of x and by r(x) = u∗u
the right support of x, clearly, l(x) ∼ u(x). In [17], Muratov and Chilin show that
x ∈ S(M) (resp. x ∈ LS(M)) if and only if x ∈ S(M) (resp.
x ∈ LS(M))
and u ∈ M. If x is a self-adjoint operator affiliated with M, then the spectral
family of projections {Eλ(x)}λ∈R for x belongs to M. In [17], the authors also
prove that a locally measurable operator x is measurable if and only if there exists
λ > 0 such that E⊥
λ (x) is a finite projection in M.
1
In the following, we recall the definition of the local measure topology. Let
M be a commutative von Neumann algebra, in [23], Takesaki proves that there
exists a ∗-isomorphism from M onto the ∗-algebra L∞(Ω, Σ, µ), where µ is a
measure satisfying the direct sum property. The direct sum property means
that the Boolean algebra of all projections in L∞(Ω, Σ, µ) is total order, and for
every non-zero projection p in M, there exists a non-zero projection q ≤ p with
µ(q) < ∞. Consider LS(M) = S(M) = L0(Ω, Σ, µ) of all measurable almost
everywhere finite complex-valued functions on (Ω, Σ, µ). Define the local measure
topology t(L∞(Ω)) on L0(Ω, Σ, µ), that is, the Hausdorff vector topology, whose
base of neighborhoods of zero is given by
W (B, ε, δ) ={f ∈ L0(Ω, Σ, µ) : there exists a set E ∈ Σ such that
E ⊂ B, µ(B\E) ≤ δ, fχE ∈ L∞(Ω, Σ, µ), kfχE kL∞(Ω,Σ,µ) ≤ ε},
where ε, δ > 0, B ∈ Σ, µ(B) < ∞ and χE(ω) = 1 when ω ∈ E, χE(ω) = 0 when
ω /∈ E. Suppose that {fα} ⊂ L0(Ω, Σ, µ) and f ∈ L0(Ω, Σ, µ), if fαχB → f χB in
t(L∞(Ω))
the measure µ for every B ∈ Σ with µ(B) < ∞, then we denote by fα
−−−−−→ f .
In [25], Yeadon show the topology t(L∞(Ω)) dose not change if the measure µ is
replaced with an equivalent measure.
If M is an arbitrary von Neumann algebra and Z(M) is the center of M,
then there exists a ∗-isomorphism ϕ from Z(M) onto the ∗-algebra L∞(Ω, Σ, µ),
where µ is a measure satisfying the direct sum property. Denote by L+(Ω, Σ, µ)
the set of all measurable real-valued positive functions on (Ω, Σ, µ). In [22], Segal
shows that there exists a mapping ∆ from P(M) into L+(Ω, Σ, µ) satisfying the
following conditions:
(D1) ∆(p) ∈ L0
(D2) ∆(p ∨ q) = ∆(p) + ∆(q) if pq = 0;
(D3) ∆(u∗u) = ∆(uu∗) for every partial isometry u ∈ M;
(D4) ∆(zp) = ϕ(z)∆(p) for every z ∈ P(Z(M)) and every p ∈ P(M);
(D5) if pα, p ∈ P(M), α ∈ Γ and pα ↑ p, then ∆(p) = supα∈Γ∆(pα).
In addition, ∆ is called a dimension function on P(M) and ∆ also satisfies the
following two conditions:
(D6) if {pn}∞
+(Ω, Σ, µ) if and only if p ∈ Pf in(M);
n=1 ∆(pn); moreover, if pnpm =
n=1 ⊂ P(M), then ∆(supn≥1pn) ≤ P∞
0 when n 6= m, then ∆(supn≥1pn) = P∞
(D7) if {pn}∞
n=1 ∆(pn);
n=1 ⊂ P(M) and pn ↓ 0, then ∆(pn) → 0 almost everywhere.
JORDAN DERIVATIONS ON SOME ALGEBRAS
5
For arbitrary scalars ε, γ > 0 and a set B ∈ Σ, µ(B) < ∞, we let
V (B, ε, γ) ={x ∈ LS(M) : there exist p ∈ P(M) and z ∈ P(Z(M) such that
xp ∈ M, kxpkM ≤ ε, ϕ(z⊥) ∈ W (B, ε, γ) and ∆(zp⊥) ≤ εϕ(z)},
where k · kM is the C ∗-norm on M. In [25], Yeadon shows that the system of sets
{x + V (B, ε, γ) : x ∈ LS(M), ε, γ > 0, B ∈ Σ and µ(B) < ∞}
defines a Hausdorff vector topology t(M) on LS(M) and the sets
{x + V (B, ε, γ), ε, γ > 0, B ∈ Σ and µ(B) < ∞}
form a neighborhood base of a local measurable operator x in LS(M). In [25],
Yeadon also proves that (LS(M), t(M)) is a complete topological ∗-algebra, and
the topology t(M) does not depend on the choices of dimension function ∆ and
∗-isomorphism ϕ. The topology t(M) on LS(M) is called the local measure
topology. Moreover, if M = B(H), then LS(M) = M and the local measure
topology topology t(M) coincides with the uniform topology k · kB(H).
3. Continuity of Jordan derivations on algebras of locally
measurable operators
In this section, we give a characterization of Jordan derivations on LS(M) of
all locally measurable operators affiliated with a von Neumann algebra M. We
show that if M is a properly infinite von Neumann algebra, then every Jordan
derivation from LS(M) into itself is continuous with respect to the local measure
topology t(M).
To prove the main theorem, we need the following lemmas.
Lemma 3.1. Suppose that M is a von Neumann algebra and A is a subalgebra
of LS(M) such that P(Z(M)) ⊂ A. If δ is a Jordan derivation from A into
LS(M), then for every a in A and every central projection z in M, we have the
following two statements:
(1) δ(ab + ba) = δ(a)b + aδ(b) + δ(b)a + bδ(a);
(2) δ(z) = 0;
(3) δ(za) = zδ(a).
Suppose that M is a von Neumann algebra and η is a linear mapping from
LS(M) into itself such that η(zx) = zη(x) for every z in P(Z(M)). Thus we
can define a linear mapping ηz from LS(zM) into LS(zM) by
ηz(y) = η(y)
for every y in LS(zM). By [5, Proposition 2.6], we can obtain the following
result.
Theorem 3.2. [5] Suppose that M is a von Neumann algebra and {zi}i∈I is a
orthogonal family of central projections such that supi∈Izi = 1. If η is a linear
mapping from LS(M) into itself such that η(zix) = ziη(x) for every x ∈ LS(M)
and every i ∈ I, then the following two statements are equivalent:
(1) The mapping η : (LS(M), t(M)) → (LS(M), t(M)) is continuous;
6
G. AN AND J. HE
(2) The mapping ηzi : (LS(ziM), t(ziM)) → (LS(ziM), t(ziM)) is continuous
for every i ∈ Γ.
Suppose that M is a von Neumann algebra. The projection lattice P(M)
is said to have a countable type, if every family of non-zero pairwise orthogonal
projections in P(M) is countable.
If M is a commutative von Neumann algebra and P(M) has a countable type,
then M is ∗-isomorphic to a ∗-algebra L∞(Ω, Σ, µ) with µ(Ω) < ∞. Hence the
topology t(L∞(Ω)) is metrizable and has a base of neighborhoods of 0 consisting
of the sets W (Ω, 1/n, 1/n), n ∈ N.
If M is a commutative von Neumann algebra and P(M) does not have a
countable type. Denote by ϕ a ∗-isomorphism from M onto L∞(Ω, Σ, µ), where
µ is a measure such that the direct sum property. Hence there exists a family of
non-zero pairwise orthogonal projections {pi}i∈Γ ⊂ P(M) such that supi∈Γpi = 1
and µ(ϕ(pi)) < ∞, in particular, P(pi(Z(M))) has a countable type.
Let A be a ∗-subalgebra of LS(M) and δ be a Jordan derivation from A into
LS(M). We can define a linear mapping δ∗ from A into LS(M) by
δ∗(a) = (δ(a∗))∗
for every a in A. It is easy to show that δ∗ is also a Jordan derivation.
A Jordan derivation δ from A into LS(M) is said to be self-adjoint if δ = δ∗.
Moreover, every Jordan derivation δ can be represented in the form
δ = Re(δ) + iIm(δ),
where Re(δ) = (δ + δ∗)/2 is the real part of δ and Im(δ) = (δ − δ∗)/2i is the
imaginary part of δ. Obviously, Re(δ) and Im(δ) are also two Jordan derivations.
Since (LS(M), t(M)) is a topological ∗-algebra, we have the following result.
Lemma 3.3. Suppose that M is a von Neumann algebra and A is a subalgebra of
LS(M). If δ is a Jordan derivation δ from A into LS(M), then δ is continuous
with respect to the topology t(M) if and only if the self-adjoint Jordan derivations
Re(δ) and Im(δ) are continuous with respect to the topology t(M).
The following theorem is the main result in this section.
Theorem 3.4. Suppose that M is a properly infinite von Neumann algebra and
δ is a Jordan derivation from LS(M) into itself. Then δ is automatically con-
tinuous with respect the local measure topology t(M).
Proof. By Lemma 3.3, we can assume that δ is a self-adjoint Jordan derivation,
that is δ = δ∗.
Since Z(M) is a commutative von Neumann algebra, there exists an orthogonal
family of central projections {zi}i∈Γ ⊂ P(Z(M)) such that supi∈Γzi = 1 and the
Boolean algebra P(ziZ(M)) has a countable type for every i ∈ Γ. By Lemma
3.2 (2), we have that δ(zix) = ziδ(x) for every x in LS(M) and every i in Γ, by
Theorem 3.2, it is sufficient to prove that δzi on LS(ziM) is continuous with the
local measure topology t(ziM) for every i ∈ I. Thus without loss of generality,
we can assume that the Boolean algebra P(Z(M)) has a countable type.
It
follows that the topology t(M) is metrizable, and the sets V (Ω, 1/n, 1/n), n ∈ N
JORDAN DERIVATIONS ON SOME ALGEBRAS
7
form a countable base of neighborhoods of 0, in particular, (LS(M), t(M)) is an
F -space.
By contradiction, we suppose that δ is discontinuous with respect the local
measure topology t(M). It means that there exist a sequence {an} ⊂ LS(M)
t(M)
−−−→
n→∞
t(M)
−−−→
n→∞
and a non-zero element a ∈ LS(M) such that an
Since (LS(M), t(M)) is a topological ∗-algebra and δ is a self-adjoint Jordan
derivation, we can assume that an = a∗
n and a = a∗ for every n in N. Hence
a = a+ − a−, where a+ and a− are positive and negative parts of a in LS+(M),
respectively, we can assume that a+ 6= 0, otherwise, we replace an with −an.
0 and δ(an)
a.
Now we select two scalars 0 < λ1 < λ2 such that the projection pa = Eλ2(a) −
Eλ1(a) 6= 0. It implies that 0 < λ1pa ≤ paapa ≤ λ2pa and kpaapakM ≤ λ2, it
means that paapa is a positive element in M.
Next we denote by xn = paanpa and x = paapa, by Lemma 3.1 (3) we have that
δ(xn) = δ(paanpa) = δ(pa)anpa + paδ(an)pa + paanδ(pa),
it follows that xn
0 and δ(xn)
t(M)
−−−→
n→∞
t(M)
−−−→
n→∞
x.
It is obvious that x is a positive element in M, we can select two scalars
0 < µ1 < µ2 such that the projection px = Eµ2(x) − Eµ1(x) 6= 0. It implies that
0 < µ1px ≤ pxxpx ≤ µ2px and kpxxpxkM ≤ µ2.
Without loss of generality, we can assume that µ1 = 1, then pxxpx ≥ px, other-
wise, we replace xn with xn/µ1, Since M is a properly infinite von Neumann alge-
bra, there exists a sequence of non-zero pairwise orthogonal projections { pm}∞
in M such that P∞
m=1
m=1 pm = 1, pm ∼ 1 and px (cid:22) pm for every m ∈ N, it means
that there exists a sequence of non-zero pairwise orthogonal projections {pm}∞
m=1
in M such that px ∼ pm ≤ pm for every m ∈ N. It follows that there is a partial
isometry vm in M such that v∗
mvm = px and vmv∗
m = pm. Denote by
p0 =
∞
Xm=1
pm,
it is easy to see that p0 is an infinite projection in M.
Since x is a positive element in M, we know that pmvmxv∗
mpm and pmv∗
mxvmpm
are also two positive elements in M, and by pxxpx ≥ px, we have that
pm(vmxv∗
mpm ≥ pmvmpxv∗
mxvm)pm ≥ pmvmxv∗
mpm = pmvmpxxpxv∗
m + v∗
mpm = pm
and
kpm(vmxv∗
m + v∗
mxvm)pmkM ≤ kpmvmxv∗
mpmkM + kpmv∗
mxvmpmkM
≤ 2kxkM ≤ 2λ2.
It means that the series P∞
the topology τso to some operator y in M satisfying
m=1 pm(vmxv∗
m + v∗
mxvm)pm converges with respect to
kykM = supm≥1kpm(vmxv∗
m + v∗
mxvm)pmkM ≤ 2λ2 and y ≥ p0.
(3.1)
In the following, we assume that the central carrier Cp0 of the projection p0 is
equal to 1, otherwise, we replace the algebra M with the algebra Cp0MCp0.
8
G. AN AND J. HE
Let ϕ be a ∗-isomorphism from Z(M) onto L∞(Ω, Σ, µ). By the assumption,
the Boolean algebra P(Z(M)) has a countable type, and we assume that µ(Ω) =
RΩ 1L∞(Ω)dµ = 1, where 1L∞(Ω) is the identity of the ∗-algebra L∞(Ω, Σ, µ). In this
case, the countable base of neighborhoods of 0 in the topology t(M) is formed
by the sets V (Ω, 1/n, 1/n), n ∈ N.
By xn
t(M)
−−−→
n→∞
0 and δ(xn)
t(M)
−−−→
n→∞
x, we can obtain the following four inequalities:
pm(vmxnv∗
m + v∗
mxnvm)pm
t(M)
−−−→
n→∞
0;
δ(pm)(vmxnv∗
m + v∗
mxnvm)pm
t(M)
−−−→
n→∞
0;
pm(δ(vm)xnv∗
m + δ(v∗
m)xnvm)pm
t(M)
−−−→
n→∞
0;
and
pm(vmδ(xn)v∗
m + v∗
mδ(xn)vm)pm
t(M)
−−−→
n→∞
pm(vmxv∗
m + v∗
mxvm)pm.
(3.2)
(3.3)
(3.4)
(3.5)
In the following fix k in N. By (3.2), we can select an index n1(m, k), a
m,n ∈ P(Z(M)), such that the
m,n ∈ P(M) and a central projection z(1)
projection q(1)
following three inequalities:
kpm(vmxnv∗
mxnvm)pmq(1)
m,nkM ≤ 2−m(k + 1)−1;
ϕ(1 − z(1)
m,n)dµ ≤ 4−12−m−k−1;
m + v∗
ZΩ
m + v∗
ZΩ
and
∆(z(1)
m,n(1 − q(1)
m,n)) ≤ 4−12−m−k−1ϕ(z(1)
m,n)
for every n ≥ n1(m, k).
By (3.3), we can select an index n2(m, k), a projection q(2)
m,n ∈ P(M) and a
central projection z(2)
m,n ∈ P(Z(M)), such that the following three inequalities:
kδ(pm)(vmxnv∗
mxnvm)pmq(2)
m,nkM ≤ 5−12−m(k + 1)−1;
ϕ(1 − z(2)
m,n)dµ ≤ 4−12−m−k−1;
and
∆(z(2)
m,n(1 − q(2)
m,n)) ≤ 4−12−m−k−1ϕ(z(2)
m,n)
for every n ≥ n2(m, k).
By (3.4) we can select an index n3(m, k), a projection q(3)
m,n ∈ P(M) and a
m,n ∈ P(Z(M)), such that the following three inequalities:
central projection z(3)
kpm(δ(vm)xnv∗
m + δ(v∗
m)xnvm)pmq(3)
m,nkM ≤ 5−12−m(k + 1)−1;
ZΩ
ϕ(1 − z(3)
m,n)dµ ≤ 4−12−m−k−1;
(3.6)
(3.7)
(3.8)
(3.9)
(3.10)
(3.11)
(3.12)
(3.13)
JORDAN DERIVATIONS ON SOME ALGEBRAS
9
and
∆(z(3)
m,n(1 − q(3)
m,n)) ≤ 4−12−m−k−1ϕ(z(3)
m,n)
(3.14)
for every n ≥ n3(m, k).
By (3.5), we can select an index n4(m, k), a projection q(4)
m,n ∈ P(M) and a
central projection z(4)
m,n ∈ P(Z(M)), such that the following three inequalities:
kpm((vmδ(xn)v∗
m + v∗
mδ(xn)vm)−(vmxv∗
m + v∗
mxvm))pmq(4)
m,nkM
≤5−12−m(k + 1)−1;
ZΩ
ϕ(1 − z(4)
m,n)dµ ≤ 4−12−m−k−1;
and
∆(z(4)
m,n(1 − q(4)
m,n)) ≤ 4−12−m−k−1ϕ(z(4)
m,n)
(3.15)
(3.16)
(3.17)
for every n ≥ n4(m, k).
Let n(m, k) = maxi=1,2,3,4ni(m, k), qm = inf i=1,2,3,4q(i)
m,n and zm = inf i=1,2,3,4z(i)
By (3.6), (3.9), (3.12) and (3.15), we have that the following four inequalities:
m,n.
kpm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmqmkM ≤ 2−m(k + 1)−1;
kδ(pm)(vmxn(m,k)v∗
=kqmpm(vmxn(m,k)v∗
≤5−12−m(k + 1)−1;
m + v∗
m + v∗
mxn(m,k)vm)pmqmkM
mxn(m,k)vm)δ(pm)kM
kpm(δ(vm)xn(m,k)v∗
=kqmpm(vmxn(m,k)δ(v∗
≤5−12−m(k + 1)−1;
m + δ(v∗
m) + v∗
m)xn(m,k)vm)pmqmkM
mxn(m,k)δ(vm))pmkM
(3.18)
(3.19)
(3.20)
and
kpm((vmδ(xn(m,k))v∗
m + v∗
mδ(xn(m,k))vm)−(vmxv∗
m + v∗
mxvm))pmqmkM
By (3.7), (3.10), (3.13) and (3.16), we have that
≤5−12−m(k + 1)−1.
(3.21)
1 − ZΩ
ϕ(zm)dµ = ZΩ
ϕ(1 − zm)dµ ≤
ϕ(1 − z(i)
m,n(m,k))dµ
4
ZΩ
Xi=1
≤ 2−m−k−1.
(3.22)
By (3.8), (3.11), (3.14), (3.17) and Condition D6 we have that
∆(zm(1 − qm)) ≤
4
Xi=1
∆(zm(1 − q(i)
m,n(m,k))) ≤ 2−m−k−1ϕ(zm).
(3.23)
Let m1 and m2 be in N with m1 < m2, denote by
qm1,m2 =
inf
m1<m≤m2
qm and zm1,m2 = inf
m1<m≤m2
zm.
10
Thus
G. AN AND J. HE
1 − qm1,m2 = sup
1 − qm and 1 − zm1,m2 = sup
1 − zm.
m1<m≤m2
m1<m≤m2
It follows that
and
ϕ(1 − qm1,m2) = sup
ϕ(1 − qm)
m1<m≤m2
ϕ(1 − zm1,m2) = sup
ϕ(1 − zm).
m1<m≤m2
Hence by (3.22), we can obtain that
1 −ZΩ
ϕ(zm1,m2)dµ = ZΩ
ϕ(1 − zm1,m2)dµ ≤
m2
ZΩ
Xm=m1+1
≤ 2−m1−k−1.
ϕ(1 − zm)dµ
By (3.23) and Condition D6, we can obtain that
m2
∆(zm1,m2(1 − qm1,m2)) ≤
∆(zm1,m2(1 − qm))
Xm=m1+1
≤ 2−m1−k−1ϕ(zm1,m2).
Moreover, by (3.18), we have that
m2
pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmqm1,m2kM
kpm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmqmkM
k
m2
Xm=m1+1
Xm=m1+1
≤2−m1(k + 1)−1.
≤
(3.24)
(3.25)
(3.26)
By (3.24), (3.25) and (3.26), it follows that
Sl,k =
l
Xm=1
pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pm, l ≥ 1
is a Cauchy sequence in the F -space (LS(M), t(M)) for every k ∈ N. Hence,
there exists an element yk ∈ LS(M) such that Sl,k
t(M)
−−−→
l→∞
yk. Thus the series
yk =
∞
Xm=1
pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pm
(3.27)
converges in LS(M) with respect to the topology t(M). Since the involution is
continuous in topology t(M) and Sl,k = S ∗
l,k, it implies that yk = y∗
k.
Denote by
rm = pm ∧ qm, m ∈ N.
(3.28)
JORDAN DERIVATIONS ON SOME ALGEBRAS
11
Since zm(pm − pm ∧ qm) ∼ zm(pm ∨ qm − qm), by Condition D3 and (3.23) we have
that
∆(zm(pm − rm)) =∆(zm(pm − pm ∧ qm)) = ∆(zm(pm ∨ qm − qm))
≤∆(zm(1 − qm)) ≤ 2−m−k−1ϕ(zm).
(3.29)
Denote by
q(k)
0 = sup
m≥1
rm and z(k)
0 = inf
m≥1
zm,
by (3.1), (3.28) and (3.30), we have that
y ≥ p0 ≥ q(k)
0 .
By (3.24), we have that
(3.30)
(3.31)
1 −ZΩ
ϕ(z(k)
0 )dµ = ZΩ
ϕ(1 − z(k)
0 )dµ ≤ 2−k−1.
(3.32)
Since pmpj = 0 when m 6= j and rm ≤ pm, by (3.30) we can obtain that
p0 − q(k)
0 = supm≥1(pm − rm). By Condition D6 and (3.29) we have that
∞
0 (pm − rm)) ≤ 2−k−1ϕ(zk
0 ).
(3.33)
∆(zk
0 (p0 − q(k)
Xm=1
0 = rmq(k)
By (3.28), we have that pmq(k)
0 )) =
∆(zk
0 = rm = rmqm for every m ∈ N. It
follows that
pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmq(k)
0 = pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmrm
and by (3.18), we can obtain that
∞
kykq(k)
0 kM =k
pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmq(k)
0 kM
pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmq(k)
0 kM
kpm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmrmkM
kpm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmqmkM
∞
∞
=k
Xm=1
Xm=1
Xm=1
Xm=1
≤(k + 1)−1
≤
≤
∞
(3.34)
0 + q(k)
0 pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)δ(pm)q(k)
0
0 + q(k)
m)xn(m,k)vm + vmxn(m,k)δ(v∗
mδ(xn(m,k))vm) − (vmxv∗
m + v∗
0 pm(vmxn(m,k)v∗
m) + v∗
mxvm))pmq(k)
0
mxn(m,k)vm)δ(pm)q(k)
m + v∗
mxn(m,k)δ(vm))pmq(k)
0
0
12
G. AN AND J. HE
0
0
=q(k)
m + v∗
m + v∗
m + v∗
m + v∗
mxn(m,k)vm)pm)q(k)
mxn(m,k)vm)pmq(k)
mxn(m,k)vm)pmq(k)
mxn(m,k)vm)pmq(k)
By Lemma 3.1 (2) and (3), we have that
q(k)
0 δ(pm(vmxn(m,k)v∗
=q(k)
0 δ(pm)(vmxn(m,k)v∗
+ q(k)
0 pmδ(vmxn(m,k)v∗
0 δ(pm)(vmxn(m,k)v∗
+ q(k)
+ q(k)
mxvm)pmq(k)
+ q(k)
m + v∗
0 δ(pm)(vmxn(m,k)v∗
m + v∗
+ rmqmpm(δ(vm)xn(m,k)v∗
+ rmqmpm((vmδ(xn(m,k))v∗
+ q(k)
Consider the following formal series:
0 pm(δ(vm)xn(m,k)v∗
0 pm((vmδ(xn(m,k))v∗
0 pm(vmxv∗
m + δ(v∗
m + v∗
0 pm(vmxv∗
m + v∗
=q(k)
0
mxn(m,k)vm)pmqmrm + rmqmpm(vmxn(m,k)v∗
m)xn(m,k)vm + vmxn(m,k)δ(v∗
m + δ(v∗
m + v∗
mδ(xn(m,k))vm) − (vmxv∗
m + v∗
mxvm)pmq(k)
0 .
m) + v∗
mxvm))pmqmrm
m + v∗
mxn(m,k)vm)δ(pm)q(k)
0
mxn(m,k)δ(vm))pmqmrm
∞
Xm=1
∞
Xm=1
q(k)
0 δ(pm)(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pmqmrm;
rmqmpm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)δ(pm)q(k)
0 ;
∞
Xm=1
∞
Xm=1
rmqmpm(δ(vm)xn(m,k)v∗
m + δ(v∗
m)xn(m,k)vm)pmqmrm;
rmqmpm(vmxn(m,k)δ(v∗
m) + v∗
mxn(m,k)δ(vm))pmqmrm;
∞
Xm=1
rmqmpm(vmδ(xn(m,k))v∗
m + v∗
mδ(xn(m,k))vm
− (vmxv∗
m + v∗
mxvm))pmqmrm;
and
∞
Xm=1
q(k)
0 pm(vmxv∗
m + v∗
mxvm)pmq(k)
0 .
(3.35)
(3.36)
(3.37)
(3.38)
(3.39)
(3.40)
By (3.19), we know that the first series (3.35) and the second series (3.36)
converge with respect to the norm k · kM to some elements a and b in M, re-
spectively. Moreover, kakM ≤ 5−1(k + 1)−1 and kbkM ≤ 5−1(k + 1)−1; by (3.20),
we know that the third series (3.37) and the fourth series (3.38) converge with
respect to the norm k·kM to some elements c and d in M; respectively. Moreover,
kckM ≤ 5−1(k + 1)−1 and kdkM ≤ 5−1(k + 1)−1; by (3.21), we know that the fifth
JORDAN DERIVATIONS ON SOME ALGEBRAS
13
series (3.39) converge with respect to the norm k · kM to some elements e and
kekM ≤ 5−1(k + 1)−1.
Finally, since y = P∞
mxvm)pm (the convergence of the latter
series is taken in the topology), we see that the last series (3.40) converges with
respect to the topology τso to some element q(k)
m=1 pm(vmxv∗
0 . Hence the series
0 yq(k)
m + v∗
∞
Xm=1
q(k)
0 δ(pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pm)q(k)
0
converges with respect to the topology τso to some element ak ∈ M, and in
addition, we have that
kak − q(k)
0 yq(k)
0 kM ≤ (k + 1)−1.
(3.41)
Next we show that
ak = q(k)
0 δ(yk)q(k)
0
where yk = P∞
m=1 pm(vmxn(m,k)v∗
Denote by wm = pm(vmxn(m,k)v∗
m + v∗
m + v∗
mxn(m,k)vm)pm.
mxn(m,k)vm)pm, for each m1 and m2 in N,
by (3.30), we have that
=q(k)
=q(k)
0 rm1
0 δ(yk)q(k)
0 rm2 + rm2q(k)
rm1q(k)
0 δ(yk)q(k)
0 (rm1δ(yk)rm2 + rm2δ(yk)rm1)q(k)
0 (δ(rm1ykrm2 + rm2ykrm1) − δ(rm1)ykrm2 − δ(rm2)ykrm1
− rm1ykδ(rm2) − rm2ykδ(rm1))q(k)
0
0
=q(k)
0 (−δ(rm1)wm2rm2 − δ(rm2)wm1rm1
− rm1wm1δ(rm2) − rm2wm2δ(rm1))q(k)
0
Since the series P∞
with respect to the topology τso, it follows that the series
m=1 q(k)
0 δ(pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pm)q(k)
0
(3.42)
converges
∞
Xm=1
rm1q(k)
0 δ(pm(vmxn(m,k)v∗
m + v∗
mxn(m,k)vm)pm)q(k)
0 rm2
14
G. AN AND J. HE
also converges with respect to this topology, in addition, we have the following
equalities:
∞
0 δ(wm)q(k)
0 )rm1
rm1akrm2 + rm2akrm1 =
=
=
0 (rm1δ(wm)rm2 + rm2δ(wm)rm1)q(k)
q(k)
0
∞
rm1(q(k)
0 δ(wm)q(k)
0 )rm2 + rm2(q(k)
Xm=1
Xm=1
Xm=1
− rm1wmδ(rm2) − rm2wmδ(rm1))q(k)
0
∞
q(k)
0 (δ(rm1wmrm2) − δ(rm1)wmrm2 − δ(rm2)wmrm1
=q(k)
0 (−δ(rm1)wm2rm2 − δ(rm2)wm1rm1
− rm1wm1δ(rm2) − rm2wm2δ(rm1))q(k)
0 .
(3.43)
By (3.42) and (3.43), we have that
0 rm2 + rm2q(k)
0 δ(yk)q(k)
rm1q(k)
0 δ(yk)q(k)
0 rm1 = rm1akrm2 + rm2akrm1,
that is
It follows that
rm1(δ(yk) − ak)rm2 = rm2(ak − δ(yk))rm1.
rm1(δ(yk) − ak)rm2pm2 = rm2(ak − δ(yk))rm1pm2 = 0.
Hence, we can obtain that
rm1(δ(yk) − ak)rm = 0
for every m ∈ N. Suppose that r(rm1(δ(yk) − ak)) is the right support of
rm1(δ(yk) − ak). Hence
for every m ∈ N. Therefore by (3.30), we have that
r(rm1(δ(yk) − ak)) ≤ 1 − rm
r(rm1(δ(yk) − ak)) ≤ inf(1 − rm) = 1 − q(k)
0 .
It implies that rm1(δ(yk) − ak)q(k)
q(k)
0 (δ(yk) − ak)q(k)
0 = 0. Since q(k)
0 = 0 for every m1 ∈ N. Similarly, we have that
0 akq(k)
ak = q(k)
0 = ak, we can obtain that
0 δ(yk)q(k)
0 .
By (3.41) we have that
kq(k)
0 (δ(yk) − y)q(k)
0 kM ≤ (k + 1)−1
By (3.34) and (3.44), we have that
k(k + 1)ykq(k)
0 kM = k(k + 1)q(k)
0 ykkM ≤ 1
and
kq(k)
0 δ((k + 1)yk)q(k)
0 − (k + 1)q(k)
0 yq(k)
0 kM ≤ 1.
(3.44)
(3.45)
(3.46)
JORDAN DERIVATIONS ON SOME ALGEBRAS
15
By (3.45) and (3.46), we know that
(k + 1)q(k)
0 − q(k)
0 δ((k + 1)yk)q(k)
0 ≤ (k + 1)q(k)
0 yq(k)
0 − q(k)
0 δ((k + 1)yk)q(k)
0 ≤ q(k)
0 ,
that is
kq(k)
0 ≤ q(k)
0 δ((k + 1)yk)q(k)
0 .
(3.47)
Next we let q0 = inf k≥1q(k)
0
and z0 = inf k≥1zk
0 .
It follows that p0 − q0 =
supk≥1(p0 − q(k)
0 ). By (3.33) and Condition D6, we can obtain that
∆(z0(p0 − q0)) = ∆(supk≥1z0(p0 − q(k)
0 )) ≤
∞
Xk=1
∆(z0(p0 − q(k)
0 )) ≤ ϕ(z0),
by Condition D1, we know that z0(p0 − q0) is a finite projection. Moreover, by
(3.45) and (3.47), we obtain that
k(k + 1)q0ykkM = k(k + 1)ykq0kM ≤ 1
kq0 ≤ q0δ((k + 1)yk)q0
(3.48)
(3.49)
and
for every k in N.
Since ϕ is a ∗-isomorphism from Z(M) onto L∞(Ω, Σ, µ), by (3.32) we have
that
ZΩ
ϕ(1 − z0)dµ = ZΩ
supk≥1ϕ(1 − z(k)
0 )dµ ≤
∞
Xk=1
ZΩ
ϕ(1 − z(k)
0 )dµ ≤ 2−1,
it means that z0 6= 0. Since Cp0 = 1 and Cz0p0 = z0Cp0 = z0 6= 0, we have that
z0p0 6= 0, and there exists n ∈ N such that z0pn 6= 0. Since z0pn ∼ z0pm, we
have that z0pm 6= 0 for every m ∈ N. Hence, z0p0 is an infinite projection. Since
z0(p0 − q0) is finite, we know that z0q0 is a infinite projection. By [18, Proposition
6.3.7], we know that there exists a non-zero central projection e0 ≤ z0 such that
e0q0 is properly infinite, in particular, there exist pairwise orthogonal projections
for every n in N. In addition,
en ≤ e0q0 and en ∼ e0q0
ZΩ
ϕ(Cq0e0)dµ 6= 0.
For every n in N, the operator
bn = δ(en)en
(3.50)
(3.51)
is locally measurable, and there exists a sequence {z(n)
z(n)
m ↑ 1 when m → ∞ and z(n)
m bn ∈ S(M) for every m in N. Since
m } ⊂ P(Z(M)) such that
ϕ(z(n)
m ) ↑ ϕ(1) = 1L∞(Ω),
16
G. AN AND J. HE
it follows that RΩ ϕ(z(n)
(3.51), there exists a central projection z(n) such that z(n)bn ∈ S(M) and
m )dµ ↑ µ(1L∞(Ω)) = 1 when m → ∞. For every n in N, by
1 − 2−n−1ZΩ
ϕ(Cq0e0)dµ < ZΩ
ϕ(z(n))dµ.
(3.52)
Consider the central projection
g0 = inf
n≥1
z(n).
Since z(n)bn ∈ S(M) and g0 = g0z(n), we have that g0bn ∈ S(M) for every n in
N. By (3.52), we can obtain that
1 −ZΩ
∞
ϕ(g0)dµ =ZΩ
ϕ(1 − g0)dµ = ZΩ
ZΩ
Xn=1
≤2−1ZΩ
ϕ(Cq0e0)dµ.
≤
ϕ(1 − z(n))dµ =
supϕ(1 − z(n))dµ
∞
Xn=1
(1 −ZΩ
ϕ(z(n))dµ)
It implies that
Thus
1 − 2−1ZΩ
ϕ(Cq0e0)dµ ≤ ZΩ
ϕ(g0)dµ.
1 + 2−1ZΩ
ϕ(Cq0e0)dµ ≤ ZΩ
ϕ(g0)dµ + ZΩ
ϕ(Cq0e0)dµ.
(3.53)
By (3.51) and (3.53), it follows that RΩ ϕ(g0Cq0e0)dµ > 0, that is, g0Cq0e0 6= 0 and
so g0e0q0 6= 0. Since e0q0 is a properly infinite projection, we have that g0e0q0 is
properly, since g0en ∼ g0e0q0, we have that g0en is a properly infinite projection
for every n in N. Since
Cg0en = g0Cen ≤ g0Cq0e0 = g0Cq0e0,
it follows that zen is also a properly infinite projection for every 0 6= z ∈
In fact, if z′ ∈ P(Z(M)) and z′zen 6= 0, then
P(Z(M)) with z ≤ g0Cq0e0.
0 6= z′zen = (z′zCq0e0)g0en. Since the projection g0en is properly infinite, we have
z′zCq0e0)g0en /∈ Pf in(M). Hence, the projection zen is also a properly infinite
projection.
We may assume that g0Cq0e0 = 1, otherwise, we replace the algebra M with the
algebra g0Cq0e0M. In this case, we also assume that bn ∈ S(M), en ∼ q0, Cen = 1
and zen is a properly infinite projection for every non-zero central projection
z ∈ P(Z(M)).
m ↓ 0 when m → ∞ and bn(1 − p(n)
Since bn ∈ S(M), fixed n ∈ N, there exists a sequence {p(n)
m=1 ⊂ Pf in(M)
m ) ∈ M for every m ∈ N. By
m ) ↓ 0, it follows that
n=1 converges in measure µ to zero. Hence we can choose a central pro-
mn as to guarantee ∆(fnsn) < 2−nϕ(fn),
such that p(n)
Condition D7, we have that ∆(p(n)
{∆(p(n)
jection fn and a finite projection sn = p(n)
1 − 2−n−1 < RΩ ϕ(fn)dµ and fnbn(1 − sn) ∈ M for every n ∈ N.
m ) ∈ L0(Ω, Σ, µ) and ∆(p(n)
m )}∞
m }∞
JORDAN DERIVATIONS ON SOME ALGEBRAS
17
Denote by
f = inf
n≥1
fn and s = sup
n≥1
sn.
By Condition D6, we have that
1/2 < ZΩ
ϕ(f )dµ and ∆(f s) ≤
∞
Xn=1
∆(f sn) ≤ ϕ(f ),
it means that f 6= 0 and by Condition D1, we know that f s is a finite projection.
Moreover, since f ≤ fn and (1 − s) ≤ (1 − sn), it follows that f bn(1 − s) ∈ M
for every n ∈ N.
Let t = f (1 − s) and gn = f (en ∧ (1 − s)) for every n ∈ N. By (3.50), we have
that
gn ≤ f en ≤ q, bngn ∈ M and gn ≤ t
(3.54)
for every n ∈ N. In addition, we can obtain that
f en − gn = f (en − en ∧ (1 − s)) ∼ f (en ∨ (1 − s) − (1 − s)) ≤ f s,
that is f en − gn is a finite projection. Hence, for every non-zero central projection
z ≤ f , we have that the projection zen − zgn is a finite projection. Since the
projection zen is infinite, the projection zgn is also infinite, that is
zgn /∈ Pf in(M)
(3.55)
for every 0 6= z ∈ P(Z(M)) and every n ∈ N.
Since bnt = f bn(1 − s) ∈ M, there exists an increasing sequence {ln} ⊂ N such
that ln > n + 2kbnkM for every n ∈ N. By (3.48) and (3.54), we have that
kgn(ln + 1)ylnδ(en)engnkM ≤ kgn(ln + 1)ylnkMkδ(en)engnkM
≤ kq0(ln + 1)ylnkMkδ(en)entkM
< (ln − n)/2.
Hence,
kgnenδ(en)(ln + 1)ylngn + gn(ln + 1)ylnδ(en)engnkM ≤ ln − n.
(3.56)
For every x = x∗ ∈ M, we have that −kxkM1 ≤ x ≤ kxkM1, moreover,
−gnkxkM ≤ gnxgn ≤ gnkxkM. By (3.56), it follows that
gnenδ(en)(ln + 1)ylngn + gn(ln + 1)ylnδ(en)engn ≥ (n − ln)gn.
(3.57)
Since enem = 0 when n 6= m. By (3.48) and (3.54), we have that the series
n=1 en(ln + 1)ylnen converges with the topology τso to a self-adjoint operator
P∞
h0 ∈ M such that
kh0kM ≤ sup
n≥1
ken(ln + 1)ylnenkM ≤ 1.
18
G. AN AND J. HE
By (3.49), (3.54) and (3.57), we have that
ngn =lngn + (n − ln)gn
≤gn(ln + 1)δ(yln)gn + gnenδ(en)(ln + 1)ylngn + gn(ln + 1)ylnδ(en)engn
=(ln + 1)(gnδ(yln)gn + gnenδ(en)ylngn + gnylnδ(en)engn)
=(ln + 1)(gnδ(yln)gn + gnδ(en)ylnengn + gnenylnδ(en)gn)
=(ln + 1)gnδ(enylnen)gn
=(ln + 1)(δ(gnenylnengn) − δ(gn)enylnengn − gnenylnenδ(gn))
=δ(gnh0gn) − δ(gn)h0gn − gnh0δ(gn)
=gnδ(h0)gn.
Thus,
for every n in N.
ngn ≤ gnδ(h0)gn
(3.58)
Let g(0)
n = gn ∧ En−1(δ(h0)), n ∈ N, where {Eµ(δ(h0))} is the spectral family of
projections for self-adjoint operator δ(h0). For every n in N, by (3.58) we have
that
ng(0)
n =ng(0)
=g(0)
≤g(0)
n ≤ g(0)
n = g(0)
n gng(0)
n δ(h0)g(0)
n (n − 1)En−1(δ(h0))g(0)
n (gnδ(h0)gn)g(0)
n En−1(δ(h0))δ(h0)g(0)
n
n = (n − 1)g(0)
n .
n
Hence, gn ∧ En−1(δ(h0)) = g(0)
gn = gn − gn ∧ En−1(δ(h0)) ∼ gn ∨ En−1(δ(h0)) − En−1(δ(h0)) ≤ 1 − En−1(δ(h0)).
n = 0, it implies that
Thus
gn (cid:22) 1 − En−1(δ(h0)).
By (3.54), we have that gn ≤ f gn (cid:22) f (1 − En−1(δ(h0))), and by Condition D3,
we can obatain that
∆(gn) ≤ ∆(f (1 − En−1(δ(h0)))).
(3.59)
for every n in N.
Since f δ(h0) ∈ LS(M), there exists a non-zero central projection f0 ≤ f ,
such that f0δ(h0) is a self-adjoint element in S(M). By [17], we can select
λ0 > 0 such that
(f0 − Eλ(f0δ(h0))) ∈ Pf in(M)
for every λ > λ0. Thus ∆(f0(1 − Eλ(f0δ(h0)))) ∈ L0
+(Ω, Σ, µ) for every λ > λ0.
Since f0(1 − Eλ(f0δ(h0))) = f0(1 − Eλ(δ(h0))), by (3.59) we have that
∆(f0gn) ∈ L0
+(Ω, Σ, µ)
for every n ≥ λ0+1, by Condition D1, we know that f0gn is a finite projection. By
(3.55), we know that f0gn is an infinite projection, this leads to a contradiction.
Hence, every Jordan derivation δ from LS(M) into itself is continuous with
(cid:3)
respect the local measure topology t(M).
JORDAN DERIVATIONS ON SOME ALGEBRAS
19
4. Extension of a Jordan derivation up to a Jordan derivation on
LS(M)
In this section, we construct an extension of a Jordan derivation from a von
Neumann algebra M into LS(M) up to a Jordan derivation from LS(M) into
itself.
Let M be a von Neumann algebra and {zn}∞
tions in M such that zn ↑ 1. A sequence {xn}∞
with the sequence {zn}∞
n=1 be a sequence of central projec-
n=1 in LS(M) is called consistent
n=1 if xmzn = xnzn for each m, n ∈ N with m > n.
n=1, {yn}∞
Lemma 4.1. [5, Proposition 4.1] Suppose that M is a von Neumann algebra,
n=1, {z′
{xn}∞
n=1 are two sequences in LS(M), and {zn}∞
n=1 are two
sequences of central projections in M such that zn ↑ 1 and z′
n ↑ 1, respectively. If
{xn}∞
n=1, respectively, then we
have the following two statements:
(1) there exists a unique element x ∈ LS(M) such that xzn = znx for every
n=1 consistent with {zn}∞
n=1 and {yn}∞
n=1 and {z′
n}∞
n}∞
n ∈ N, moreover, xn
(2) if xnznz′
m = ymznz′
x;
t(M)
−−−→
n→∞
m for each m, n ∈ N, then (xnzn − ynz′
n)
t(M)
−−−→
n→∞
0.
In the following we extend a Jordan derivation from S(M) into LS(M) up
to a Jordan derivation from LS(M) into itself. For every x in LS(M), we
can select a sequence {zn}∞
n=1 of non-zero central projections in M such that
zn ↑ 1 and xzn ∈ S(M) for every n in N, By Lemma 3.2 (2) we know that
δ(xzn)zm = δ(xznzm) = δ(xznzn) = δ(xzn)zn for each m, n ∈ N with m > n.
It means that the sequence {δ(xzn)}∞
n=1.
By Proposition 4.1 (1), there exists a unique element yx ∈ LS(M) such that
n=1 is consistent with the sequence {zn}∞
δ(xzn)
t(M)
−−−→
n→∞
yx. Hence we can define a mapping from LS(M) into itself by
δ(x) = yx.
By Proposition 4.1 (2), we know that δ(x) does not depend on a choice of a
n=1 ⊂ P(Z(M)). Thus the mapping δ is well-defined. Moreover, if
sequence {zn}∞
x ∈ S(M), we let zn = 1 for every n ∈ N, then δ(x) = δ(x).
Theorem 4.2. Suppose that M is a von Neumann algebra. If δ is a Jordan
derivation from S(M) into LS(M), then δ is a unique Jordan derivation from
LS(M) into LS(M) such that δ(x) = δ(x) for every x ∈ S(M).
Proof. First, we prove that δ is a linear mapping from LS(M) into itself. Suppose
that x and y are two elements in LS(M), we can select two sequences {zn}∞
n=1 and
{z′
n ∈
S(M) for every n in N, respectively. It is clear that {znz′
n=1 is also a sequence
of non-zero central projections in M such that
n=1 of non-zero central projections in M such that zn ↑ 1, z′
n ↑ 1, xzn, yz′
n}∞
n}∞
znz′
n ↑ 1, xznz′
n, yznz′
n and (x + y)znz′
n ∈ S(M)
20
G. AN AND J. HE
for every n in N. It follows that
δ(x + y) =t(M) − lim
δ((x + y)znz′
n)
n→∞
=(t(M) − lim
n→∞
δ(xznz′
n)) + (t(M) − lim
n→∞
δ(yznz′
n))
=δ(x) + δ(y).
Similarly, we can show that δ(λx) = λδ(x) for every λ ∈ C. Thus δ is a linear
mapping from LS(M) into itself.
Next, we prove that δ is a Jordan derivation from LS(M) into itself. Since
xzn
t(M)
−−−→
n→∞
x, δ(xzn)
t(M)
−−−→
n→∞
δ(x), and x2zn ∈ S(M), we have that
δ(x2) =t(M) − lim
δ(x2zn) = t(M) − lim
n→∞
δ(xznxzn)
(δ(xzn)(xzn) + (xzn)δ(xzn))
n→∞
=t(M) − lim
n→∞
=δ(x)x + xδ(x).
It means that δ is a Jordan derivation from LS(M) into itself. Clearly, δ(x) =
δ(x) for every x in S(M).
Let x be in LS(M), {zn}∞
Finally, we show the uniqueness of the δ. Suppose that δ1 is also a Jordan
derivation from LS(M) into itself such that δ1(x) = δ(x) for every x in S(M).
n=1 be a sequence of non-zero central projections in
M such that zn ↑ 1, and xzn ∈ S(M) for every n in N. By Lemma 3.2 (2) and
Proposition 4.1 (1), we have that
δ(x) = t(M) − lim
δ1(xzn)
δ(xzn) = t(M) − lim
n→∞
n→∞
= t(M) − lim
n→∞
δ1(x)zn = δ1(x)
(cid:3)
Next, we extend a Jordan derivation from M into LS(M) up to a Jordan
derivation from S(M) into LS(M). Let x be in LS(M) and x = ux be the
polar decomposition of x, l(x) = uu∗ is the left support of x and r(x) = u∗u is
the right support of x, clearly, l(x) ∼ u(x). Denote by s(x) = l(x) ∨ r(x).
The following lemmas will be used repeated.
Lemma 4.3. Suppose that ∆ is a dimension function of a von Neumann algebra
M. If δ is a Jordan derivation from M into LS(M), then we have the following
inequality
∆(s(δ(x))) ≤ 6∆(s(x))
for every x ∈ M.
Proof. Let x be in M, we have the following three statements:
l(δ(s(x))x) ∼ r(δ(s(x))x) = r(δ(s(x))xs(x)) ≤ s(x);
r(xδ(s(x))) ∼ l(xδ(s(x))) = l(s(x)xδ(s(x))) ≤ s(x);
(4.1)
(4.2)
JORDAN DERIVATIONS ON SOME ALGEBRAS
21
and
r(s(x)δ(x)s(x)) ∼ l(s(x)δ(x)s(x)) ≤ s(x).
(4.3)
By (4.1), we have that
l(δ(s(x))x) (cid:22) s(x) and r(δ(s(x))x) (cid:22) s(x);
by (4.2), we have that
l(xδ(s(x))) (cid:22) s(x) and r(xδ(s(x))) (cid:22) s(x);
by (4.3), we have that
It implies that the following three inequalities:
r(s(x)δ(x)s(x)) (cid:22) s(x).
∆(l(δ(s(x))x)) = ∆(r(δ(s(x))x)) ≤ ∆(s(x));
∆(l(xδ(s(x)))) = ∆(r(xδ(s(x)))) ≤ ∆(s(x));
(4.4)
(4.5)
and
∆(r(s(x)δ(x)s(x))) = ∆(l(s(x)δ(x)s(x))) ≤ ∆(s(x)).
(4.6)
Since δ is a Jordan derivation and by Lemma 3.1 (2) we have that
δ(x) = δ(s(x)xs(x)) =δ(s(x))xs(x) + s(x)δ(x)s(x) + s(x)xδ(s(x))
=δ(s(x))x + s(x)δ(x)s(x) + xδ(s(x)).
By (4.4), (4.5) and (4.6) we can obtain that
s(δ(x)) =s(δ(s(x))x + s(x)δ(x)s(x) + xδ(s(x)))
=r(δ(s(x)x) ∨ l(δ(s(x)x) ∨ r(xδ(s(x))) ∨ l(xδ(s(x)))
∨ r(s(x)δ(x)s(x)) ∨ l(s(x)δ(x)s(x)).
It implies that
for every x in M.
∆(s(δ(x))) ≤ 6∆(s(x)).
(cid:3)
The following result is proved by Ber, Chilin and Sukochev in [5].
Lemma 4.4. [5, Lemma 4.4] Suppose that {xn}∞
s(xn) ∈ Pf in(M) and ∆(s(xn))
t(L∞(Ω))
−−−−−→
0, then xn
n=1 is a sequence in LS(M). If
0. In particular, if
t(M)
−−−→
n→∞
n→∞
n=1 ⊂ Pf in(M) and pn ↓ 0, then pn
{pn}∞
t(M)
−−−→
n→∞
0.
By Lemmas 4.3 and 4.4, we have the following result.
n=1, {qn}∞
Lemma 4.5. Suppose that M is a von Neumann algebra. Let x be in S(M),
{pn}∞
n=1 be two sequences of non-zero projections in M such that pn ↑
1, qn ↑ 1, xpn, xqn ∈ M and p⊥
n ∈ Pf in(M) for every n in N. If δ is a Jordan
derivation from M into LS(M), then there exists an element δ(x) in LS(M)
such that
n , q⊥
t(M) − lim
n→∞
δ(xpn) = δ(x) = t(M) − lim
n→∞
δ(xqn).
22
G. AN AND J. HE
Proof. Let m and n be in N with m > n, we have that
l(x(pm − pn)) ∼ r(x(pm − pn)) ≤ pm − pn.
By Lemma 4.3, Conditions D2 and D3, we can obtain that
∆(s(δ(xpm − xpn))) = ∆(s(δ(x(pm − xpn)))) ≤ 6∆(s(x(pm − pn)))
≤ 6∆(l(x(pm − pn)) ∨ (pm − pn)) ≤ 12∆(pm − pn)
≤ 12∆(p⊥
n ).
(4.7)
By Condition D1, we know that ∆(p⊥
know that ∆(p⊥
that
n ) ∈ L0
n ) ↓ 0, it follows that ∆(p⊥
n )
+(Ω, Σ, µ), and by Condition D7, we
t(L∞(Ω))
0. By (4.7) we can obtain
−−−−−→
n→∞
∆(s(δ(xpm) − δ(xpn)))
t(L∞(Ω))
−−−−−→
m,n→∞
0.
By Lemma 4.4, we have that
(δ(xpm) − δ(xpn))
t(L∞(Ω))
−−−−−→
m,n→∞
0.
It means that {δ(xpn)}∞
exists an element δ(x) in LS(M) such that
n=1 is a Cauchy sequence in (LS(M), t(M)). Hence, there
Next we show that
t(M) − lim
n→∞
δ(xpn) = δ(x).
t(M) − lim
n→∞
δ(xqn) = δ(x).
For every n ∈ N, we have that
(pn − qn)((pn − pn ∧ qn) ∨ (qn − pn ∧ qn))
=((pn − pn ∧ qn) − (qn − pn ∧ qn))((pn − pn ∧ qn) ∨ (qn − pn ∧ qn))
=(pn − pn ∧ qn) − (qn − pn ∧ qn) = pn − qn.
It follows that
r(pn−qn) = r((pn−qn)((pn−pn∧qn)∨(qn−pn∧qn))) ≤ ((pn−pn∧qn)∨(qn−pn∧qn)).
Since
and
r(x(pn − qn)) ≤ r(pn − qn)
l(x(pn − qn)) ∼ r(x(pn − qn)).
JORDAN DERIVATIONS ON SOME ALGEBRAS
23
By Condition D6 we can obtain that
∆(s(x(pn − qn))) =∆(l(x(pn − qn)) ∨ r(x(pn − qn)))
≤∆(l(x(pn − qn))) + ∆(r(x(pn − qn)))
=2∆(r(x(pn − qn)))
≤2∆((pn − pn ∧ qn) ∨ (qn − pn ∧ qn))
≤2∆(pn − pn ∧ qn) + 2∆(qn − pn ∧ qn)
≤4∆(1 − pn ∧ qn)
=4∆(p⊥
n ) ≤ 4(∆(p⊥
n ∨ q⊥
n ) + ∆(q⊥
n )).
(4.8)
By Lemma 4.3, we have that
∆(s(δ(xpn) − δ(xqn))) = ∆(s(δ(x(pn − qn)))) ≤ 6∆(s(x(pn − qn))),
(4.9)
By (4.8) and (4.9), we can obtain that
∆(s(δ(xpn) − δ(xqn))) ≤ 24(∆(p⊥
n ) + D(q⊥
n )) ↓ 0
By Lemma 4.4, we obtain that
t(M) − lim
n→∞
δ(xqn) = δ(x) = t(M) − lim
n→∞
δ(xpn)
for every x in S(M).
(cid:3)
By Lemma 4.5, we can extend every Jordan derivation δ from M into LS(M)
up to a Jordan derivation δ from S(M) into LS(M).
For every x in S(M), there exists a sequence {pn}∞
in M such that pn ↑ 1, xpn ∈ M and p⊥
4.5, there exists an element δ(x) in LS(M) such that
n=1 of non-zero projections
n ∈ Pf in(M) for every n in N. By Lemma
δ(x) = t(M) − lim
n→∞
δ(xpn).
Moreover, we know that the definition of δ(x) dose not depend on a choice of
n=1, which satisfies the above properties, in particular, δ(x) = δ(x)
sequence {pn}∞
for every x in M.
Theorem 4.6. Suppose that M is a von Neumann algebra and δ is a Jordan
derivation from M into LS(M). Then δ is a unique Jordan derivation from
S(M) into LS(M) such that δ(x) = δ(x) for every x ∈ M.
Proof. First, we show that δ is a linear mapping from S(M) into LS(M). For
each x and y in S(M), we can select two sequences {pn}∞
n=1 ⊂ P(M),
such that
n=1, {qn}∞
pn ↑ 1, qn ↑ 1, xpn, yqn ∈ M and p⊥
n , q⊥
n ∈ Pf in(M)
for every n ∈ N. Denote by
en = pn ∧ qn.
It is easy to show that {en}∞
n=1 is an increase sequence, and we have that
xen = xpnen ∈ M, yen = yqnen ∈ M,
24
and
G. AN AND J. HE
e⊥
n = p⊥
n ∨ q⊥
n ∈ Pf in(M), ∆(e⊥
n ) ≤ (∆(p⊥
n ) + ∆(q⊥
n )) ↓ 0.
By Condition D7, we can obtain that
e⊥
n ↓ 0 and en ↑ 1.
By the definition of δ, we have that
δ(x + y) =t(M) − lim
δ((x + y)en)
n→∞
=(t(M) − lim
n→∞
δ(xen)) + (t(M) − lim
n→∞
δ(yen))
=δ(x) + δ(y).
Similarly, we can show that δ(λx) = λδ(x) for every λ ∈ C.
Next we show that δ is a Jordan derivation from S(M) into LS(M). By the
polar decomposition x = ux, u∗u = r(x), we have that xn = xEn(x) ∈ M for
every n ∈ N. Denote by
gn = 1 − r(E⊥
n (x)xn) and sn = gn ∧ En(x).
It implies that
xngn = En(x)xngn + E⊥
n (x)xngn = En(x)xngn.
(4.10)
By (4.10), we have that
x2sn = x2En(x)sn = xxnsn = xxngnsn
= xEn(x)xngnsn = xEn(x)xEn(x)sn.
(4.11)
n = r(E⊥
Since g⊥
n (x). Since x ∈ S(M), there exists n0 ∈ N such that E⊥
n (cid:22)
n (x) ∈ Pf in(M) for
n ∈ Pf in(M) for every n ≥ n0. Hence we have that
E⊥
every n ≥ n0, it follows that g⊥
n (x), we obtain that g⊥
n (x)xn) ∼ l(E⊥
n (x)xn) ≤ E⊥
s⊥
n = g⊥
n ∨ E⊥
n (y) ∈ Pf in(M)
for every n > n0, and
n ) ≤ D(g⊥
D(s⊥
n ) + D(E⊥
n (y)) ≤ (D(E⊥
n (x)) + D(E⊥
n (y))) ↓ 0.
It means that s⊥
n ↓ 0 and sn ↑ 1. By (4.10), we have that
E⊥
n+1(x)xn+1sn =E⊥
=E⊥
=E⊥
=E⊥
=0.
n+1(x)E⊥
n+1(x)(E⊥
n+1(x)(E⊥
n+1(x)(E⊥
n (xxn+1En(x))sn
n (x)xnEn(x))sn
n (x)xnsn)
n (x)xngn)sn
It follows that
sn ≤ 1 − r(E⊥
n+1(x)xn+1) = gn+1
for every n ∈ N. By the inequalities sn ≤ En(y) ≤ En+1(y), we have that
sn ≤ sn+1.
JORDAN DERIVATIONS ON SOME ALGEBRAS
25
By (4.11), Lemmas 4.4 and 4.5, we can obtain that
δ(x2) = t(M) − lim
n→∞
δ(x2sn) = t(M) − lim
n→∞
snδ(x2sn)sn
=t(M) − lim
n→∞
=t(M) − lim
n→∞
=t(M) − lim
n→∞
[δ(snx2sn) − δ(sn)x2sn − snx2snδ(sn)]
[δ(snxEn(x)xEn(x)sn) − δ(sn)x2sn − snx2snδ(sn)]
[snδ(xEn(x)xEn(x))sn + δ(sn)xEn(x)xEn(x)sn
+ snxEn(x)xEn(x)δ(sn) − δ(sn)x2sn − snx2snδ(sn)]
=t(M) − lim
n→∞
[sn(δ(xEn(x))xEn(x) + xEn(x)δ(xEn(x))sn]
=t(M) − lim
n→∞
[sn(δ(x)x + xδ(x))sn]
=δ(x)x + xδ(x).
It means that δ is a Jordan derivation from S(M) into LS(M). Moreover, we
have that δ(x) = δ(x) for every x ∈ M.
Finally, we show the uniqueness of δ. Suppose that δ1 is a Jordan derivation
from S(M) into LS(M) such that δ1(y) = δ(y) for every y ∈ M.
Let y be in M, then En(y) ↑ 1, yEn(y) ∈ M, for every n ∈ N. There exists
n (y) ∈ Pf in(M) for every n ≥ n1. By Lemma 4.4, we have
1. Since (LS(M), t(M)) is a topological algebra, it follows
n1 ∈ N such that E⊥
that En(y)
that
t(M)
−−−→
n→∞
δ1(y) =t(M) − lim
n→∞
=t(M) − lim
n→∞
En(y)δ1(y)En(y)
[δ1(En(y)yEn(y)) − δ1(En(y))yEn(y)
− En(y)yδ1(En(y))]
=t(M) − lim
n→∞
[δ1(En(y)yEn(y)En(y)) − δ1(En(y))yEn(y)
− En(y)yδ1(En(y))]
=t(M) − lim
n→∞
[En(y)δ1(yEn(y))En(y) + δ1(En(y))yEn(y)
+ En(y)yEn(y)δ1(En(y)) − δ1(En(y))yEn(y) − En(y)yδ1(En(y))]
=δ(y) + t(M) − lim
n→∞
[En(y)yEn(y)δ1(En(y))
− En(y)yδ1(En(y))].
(4.12)
Since δ(1) = 0, s(y) = s(−y) for every y ∈ LS(M), by Lemma 4.3, it follows
that
∆(s((δ(En(y)))) = ∆(s((δ(−En(y)))) = ∆(s((δ(1 − En(y)))) ≤ 6∆(E⊥
n (y)) ↓ 0.
By Lemma 4.4, we obtain δ(En(y))
δ(y).
t(M)
−−−→
n→∞
0, by (4.12), we have that δ1(y) =
(cid:3)
Theorem 4.7. Suppose that M is a von Neumann algebra and A is a subalgebra
of LS(M) such that M ⊂ A. If δ is a Jordan derivation from A into LS(M),
26
G. AN AND J. HE
then there exists a unique Jordan derivation δA from LS(M) into LS(M) such
that δA(x) = δ(x) for every x ∈ A.
Proof. Since M ⊂ A, the restriction δ0 of the Jordan derivation δ on M is a
well-defined Jordan derivation from M into LS(M). By Theorems 4.2 and 4.7,
δ is a unique Jordan derivation from LS(M) into itself such
we know that δA =
that δA(x) = δ0(x) for every x in M.
Next we show that
δA(a) = δ(a)
for every a in A. Let a be in A, there exists a sequence {zn}∞
n=1 ⊂ P(Z(M)),
such that zn ↑ 1 and azn ∈ S(M) for every n ∈ N. By Lemma 4.1 (1), we have
that zn
t(M)
−−−→
n→∞
1. By Lemma 3.2 (2), we can obtain that
δA(a) = t(M) − lim
n→∞
δA(a)zn = t(M) − lim
n→∞
δA(azn).
Similarly, we have that δ(a) = t(M) − limn→∞ δ(azn).
Since δA(x) = δ0(x) = δ(x) for every x in M and by the proof of uniqueness
of the derivation δ from Proposition 4.7, we can obtain that δA(azn) = δ(azn) for
every n in N, it implies that δA(a) = δ(a) for every a in A.
(cid:3)
By Theorems 3.4 and 4.7, we have the following corollary.
Corollary 4.8. Suppose that M is a properly infinite von Neumann algebra, A
is a subalgebra of LS(M) such that M ⊂ A. If δ is a Jordan derivation from
A into LS(M), then δ is continuous with respect to the local measure topology
t(M).
References
1. S. Albeverio, S. Ayupov, K. Kudaybergenov. Derivations on the algebra of measurable
operators affiliated with a type I von Neumann algebra. Siberian Adv. Math., 2008, 18:
86-94.
2. S. Albeverio, S. Ayupov, K. Kudaybergenov. Structure of derivations on various algebras
of measurable operators for type I von Neumann algebras. J. Func. Anal., 2009, 256: 2917-
2943.
3. R. Alizadeh. Jordan derivations of full matrix algebras. Linear Algebra Appl., 2009, 430:
574-578.
4. A. Ber, V. Chilin, F. Sukochev. Non-trivial derivation on commutative regular algebras.
Extracta Math., 2006, 21: 107-147.
5. A. Ber, V. Chilin, F. Sukochev. Continuity of derivations of algebras of locally measurable
operators. Integr. Equ. Oper. Theory, 2013, 75: 527-557.
6. A. Ber, V. Chilin, F. Sukochev. Continuous derivations on algebras of locally measurable
operators are inner. Proc. London Math. Soc., 2014, 109: 65-89.
7. G. An, Y. Ding, J. Li. Characterizations of Jordan left derivations on some algebras. Banach
J. Math. Anal., 2016, 10: 466-481.
8. M. Bresar. Jordan derivations on semiprime rings. Bull. Aust. Math. Soc., 1988, 104: 1003-
1006.
9. M. Bresar, J. Vukman. Jordan derivations on prime rings. Bull. Aust. Math. Soc., 1988,
37: 321-322.
10. M. Bresar, J. Vukman. On left derivations and related mappings. Proc. Amer. Math. Soc.,
1990, 11:, 7-16.
JORDAN DERIVATIONS ON SOME ALGEBRAS
27
11. J. Cuntz. On the continuity of Semi-Norms on operator algebras. Math. Ann., 1976, 220:
171-183.
12. J. Cusack. Jordan derivations on rings. Proc. Amer. Math. Soc., 1975, 53: 321-324.
13. Q. Deng. On Jordan left derivations. Math. J. Okayama Univ., 1992, 34: 145-147.
14. S. Hejazian, A. Niknam. Modules, annihilators and module derivations of J B ∗-algebras.
Indian J. Pure Appl. Math., 1996, 27: 129-140.
15. I. Herstein, Jordan derivations of prime rings. Proc. Amer. Math. Soc., 1957, 8: 1104-1110.
16. B. Johnson. Symmetric amenability and the nonexistence of Lie and Jordan derivations.
Math. Proc. Cambd. Philos. Soc., 1996, 120: 455-473.
17. M. Muratov, V. Chilin. Algebras of measurable and locally measurable operators. Kyiv,
Pratse In-ty matematiki NAN ukraini., 2007, 69, 390 pp, (Russian).
18. R. Kadison, J. Ringrose, Fundamentals of the Theory of Operator Algebras, I, Pure Appl.
Math. 100, Academic Press, New York, 1983.
19. J. Li, J. Zhou. Jordan left derivations and some left derivable maps. Oper. Matrices, 2010,
4: 127-138.
20. J. Ringrose. Automatically continuous of derivations of operator algebras. J. London Math.
Soc., 1972, 5: 432-438.
21. S. Sakai. Derivations of W ∗-algebras. Ann. Math., 1966, 83: 273 -- 279.
22. I. Segal. A non-commutative extension of abstract integration. Ann, Math., 1953, 57: 401-
457.
23. M. Takesaki. Theory of operator algebras I, New York, Springer-Verlag, 1979.
24. J. Vukman. On left Jordan derivations of rings and Banach algebras. Aequations Math.,
2008, 75: 260-266.
25. F. Yeadon. Convergence of measurable operators. Proc. Camb. Phil. Soc,, 1973, 74: 257-268.
Department of Mathematics, Shaanxi University of Science and Technology,
Xi'an 710021, China.
E-mail address: [email protected]
Department of Mathematics and Physics, Anhui Polytechnic University, Wuhu
241000, China.
E-mail address: [email protected]
|
1503.08517 | 2 | 1503 | 2016-03-03T02:08:42 | Real rank and topological dimension of higher rank graph algebras | [
"math.OA"
] | We study dimension theory for the $C^*$-algebras of row-finite $k$-graphs with no sources. We establish that strong aperiodicity - the higher-rank analogue of condition (K) - for a $k$-graph is necessary and sufficient for the associated $C^*$-algebra to have topological dimension zero. We prove that a purely infinite $2$-graph algebra has real-rank zero if and only if it has topological dimension zero and satisfies a homological condition that can be characterised in terms of the adjacency matrices of the $2$-graph. We also show that a $k$-graph $C^*$-algebra with topological dimension zero is purely infinite if and only if all the vertex projections are properly infinite. We show by example that there are strongly purely infinite $2$-graphs algebras, both with and without topological dimension zero, that fail to have real-rank zero. | math.OA | math |
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK
GRAPH ALGEBRAS
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
Abstract. We study dimension theory for the C ∗-algebras of row-finite k-graphs with
no sources. We establish that strong aperiodicity -- the higher-rank analogue of condi-
tion (K) -- for a k-graph is necessary and sufficient for the associated C ∗-algebra to have
topological dimension zero. We prove that a purely infinite 2-graph algebra has real-rank
zero if and only if it has topological dimension zero and satisfies a homological condition
that can be characterised in terms of the adjacency matrices of the 2-graph. We also
show that a k-graph C ∗-algebra with topological dimension zero is purely infinite if and
only if all the vertex projections are properly infinite. We show by example that there are
strongly purely infinite 2-graphs algebras, both with and without topological dimension
zero, that fail to have real-rank zero.
1. Introduction
Dimension theory of C ∗-algebras has played an important role in the subject over the
last few decades.
In particular, notions of dimension such as real rank, stable rank,
decomposition rank and nuclear dimension have become more and more prominent in
classification theory. But these notions of dimension can be difficult to compute for
concrete classes of examples. In this paper we focus on dimension theory of higher rank
graph C ∗-algebras. Specifically we consider topological dimension zero and real-rank zero
of C ∗-algebras associated to higher rank graphs.
A C ∗-algebra A has topological dimension zero if the primitive ideal space of A endowed
with the Jacobsen topology has a basis of compact open sets. This definition can be found
in Brown and Pedersen's work [5], but the study of the concept goes back much further.
For example Bratteli and Elliott showed [1] that a separable C ∗-algebra A of type I is
AF if and only if it has topological dimension zero. For more recent work on topological
dimension see [3, 5, 16, 28, 30, 31, 35, 38].
Recently, sufficient conditions on a k-graph Λ for its C ∗-algebra to have topological
dimension zero were established by Kang and the first named author. They proved in [16,
Theorem 4.2] that for a row-finite k-graph Λ with no sources, the C ∗-algebra associated
to Λ has topological dimension zero if Λ is strongly aperiodic. Strong aperiodicity is the
higher-rank analogue of the well-known condition (K) for directed graphs (every vertex
that is the basepoint of a first-return path is the basepoint of at least two such paths). It
can be characterised combinatorially in terms of the k-graph, and can be rephrased in a
number of other ways. For example, it is equivalent to the condition on the infinite-path
groupoid GΛ of Λ, that for every closed invariant subset of the unit space, the points with
Date: September 29, 2018.
2010 Mathematics Subject Classification. 46L05.
Key words and phrases. Graph C ∗-algebra; real rank; topological dimension; purely infinite; higher-
rank graph.
This research was supported by the Australian Research Council.
1
2
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
trivial isotropy are dense.
It is also equivalent to the property that every ideal of the
k-graph C ∗-algebra C ∗(Λ) is gauge-invariant. We prove in Section 3 that this sufficient
condition is also necessary (Theorem 3.2). Our approach uses the fact that topological
dimension zero passes to ideals and quotients, together with recent results on P -graphs,
induced algebras, and primitive-ideal structure of k-graph algebras [6, 39]. We finish the
section with Corollary 3.9 which describes a number of properties of Λ and C ∗(Λ) that
are all equivalent to strong aperiodicity of Λ, and hence topological dimension zero for
C ∗(Λ).
We then consider real-rank zero for purely infinite k-graph C ∗-algebras. A C ∗-algebra
A has real-rank zero if the set of invertible self-adjoint elements of the minimal unitisation
of A is dense in the set of all self-adjoint elements in the minimal unitisation of A [4].
For k = 1, it was established in the locally finite case in [14, 15], and in general by Hong
and Szymański in [13, Theorem 2.5], that a graph C ∗-algebra C ∗(E) has real-rank zero
if and only if the graph satisfies Condition (K). However, the methods of [14, 15, 13] do
not generalise to k-graphs. So we restrict our attention to the special case of k-graphs Λ
for which C ∗(Λ) is purely infinite.
Simple purely infinite C ∗-algebras are automatically of real-rank zero by [4, 40], but non-
simple purely infinite C ∗-algebras need not have real-rank zero (see Examples 6.1 and 6.2).
Indeed, Pasnicu and Rørdam have proved that a purely infinite C ∗-algebra has real-
rank zero if and only if it has topological dimension zero and satisfies a K-theoretic
criterion called K0-liftability [31]. Using this result we derive a homological condition
characterising when a purely infinite 2-graph C ∗-algebra has real-rank zero. We do this
by examining Evans' spectral-sequence calculation of K-theory for k-graph C ∗-algebras.
This sequence is based on a complex DΛ
∗ in which the terms are the exterior powers of
Zk tensored with ZΛ0. When k = 2, Evans proves that K1(C ∗(Λ)) is isomorphic to the
first homology group of this complex. We show that the inclusion jH : H ֒→ Λ0 of a
saturated hereditary set H induces a morphism from the spectral sequence for Γ := HΛ
to the spectral sequence for Λ. Using naturality of Kasparov's spectral sequence, we show
that for k = 2 the map in K1 induced by the inclusion C ∗(Γ) ֒→ C ∗(Λ) coincides with the
map H1(jH) := H1(DΓ
∗ ) induced by the inclusion H ֒→ Λ0. Combining this
with Pasnicu and Rørdam's result yields our characterisation of real-rank zero: a purely
infinite 2-graph C ∗-algebra has real rank zero if and only if Λ is strongly aperiodic and
H1(jH) is injective for each saturated hereditary subset H of Λ0 (Theorem 4.3). Both
of these conditions are necessary for C ∗(Λ) to have real rank zero even when C ∗(Λ) is
not purely infinite. Moreover, the injectivity of each H1(jH) can be reformulated as an
elementary algebraic condition involving the adjacency matrices of the ambient 2-graph.
Our resulting condition on Λ is much more subtle than the corresponding result for
1-graphs, reflecting the more nuanced K-theory of k-graph C ∗-algebras in which, for
example, the K0-classes of vertex projections do not necessarily generate the whole K0-
group (see [10]).
∗ ) → H1(DΓ
Since pure infiniteness is a hypothesis in Theorem 4.3, we develop in Section 5 a char-
acterisation of pure infiniteness for the C ∗-algebras of strongly aperiodic k-graphs. This
topic has been intensively studied, and for 1-graphs this culminated in [13], where suffi-
cient and necessary conditions for pure infiniteness of a 1-graph C ∗-algebra (in terms of
the 1-graph) where established. Building on recent results for cofinal k-graphs in [2], we
establish a number of properties equivalent to pure infiniteness of C ∗(Λ). In particular,
generalising [2, Corollary 5.1], we prove that C ∗(Λ) is purely infinite if and only if the
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
3
vertex projections {sv : v ∈ Λ0} are all properly infinite (without assuming Λ is cofinal).
We also show that this is equivalent to asking that for any saturated hereditary H ⊆ Λ0,
the vertex projections of C ∗(Λ \ ΛH) are all infinite (not, a priori, properly infinite).
Using this, we prove that if a k-graph has an aperiodic quartet [16] at every vertex, then
its C ∗-algebra is purely infinite (Proposition 5.1). Our result adds to the list (see [11, 21,
37]) of known sufficient conditions on a k-graph Λ for pure infiniteness of its C ∗-algebra.
The utility of our result becomes apparent in Section 6, where we use Proposition 5.1
to construct two examples of purely infinite k-graph C ∗-algebras without real-rank zero.
The first example is a 2-graph Λ such that C ∗(Λ) is purely infinite and has topological
dimension zero, but fails to have real-rank zero because the morphism H1(jH), for a
suitable H ⊆ Λ0, is not injective. Our second example is also purely infinite, but fails to
have real-rank zero because the graph is not strongly aperiodic, and so its C ∗-algebra does
not have topological dimension zero. Finally we present an example of a 2-graph which
is both strongly aperiodic and satisfies the injectivity condition, but whose C ∗-algebra is
not purely infinite, and fails to have real rank zero.
2. Preliminaries and notation
We let Prim(A) denote the set of all primitive ideals in a C ∗-algebra A. The Jacobsen
topology on Prim(A) is defined as follows. For each ideal I of A, we define the hull of I
by hull(I) := {P ∈ Prim(A) : I ⊆ P }; and for each subset S ⊆ Prim(A), we define the
kernel of S to be ker(S) :=TP ∈S P . The Jacobsen topology is then determined by the
closure operation: S := hull(ker(S)) for every S ⊆ Prim(A).
Following [21, 25, 32] we briefly recall the notion of a k-graph and the associated
notation. For k ≥ 0, a k-graph is a nonempty countable small category equipped with a
functor d : Λ → Nk that satisfies the factorisation property: for all λ ∈ Λ and m, n ∈ Nk
such that d(λ) = m + n there exist unique µ, ν ∈ Λ such that d(µ) = m, d(ν) = n, and
λ = µν. When d(λ) = n we say λ has degree n, and we write Λn = d−1(n). The standard
generators of Nk are denoted e1, . . . , ek, and we write ni for the ith coordinate of n ∈ Nk.
For m, n ∈ Nk, we write m ∨ n for their coordinate-wise maximum, and define a partial
order on Nk by m ≤ n if mi ≤ ni for all i.
If Λ is a k-graph, its vertices are the elements of Λ0. The factorisation property implies
that these are precisely the identity morphisms, and so can be identified with the objects.
For α ∈ Λ the source s(α) is the domain of α, and the range r(α) is the codomain of
α (strictly speaking, s(α) and r(α) are the identity morphisms associated to the domain
and codomain of α). Given λ, µ ∈ Λ and E ⊆ Λ, we define
λE = {λν : ν ∈ E, r(ν) = s(λ)}, Eµ = {νµ : ν ∈ E, s(ν) = r(µ)},
and
λEµ = λE ∩ Eµ.
For X, E, Y ⊆ Λ, we write XEY forSλ∈X,µ∈Y λEµ. We say the k-graph Λ is row-finite
if the set vΛm is finite for each m ∈ Nk and v ∈ Λ0. Also, Λ has no sources if vΛei 6= ∅
for all v ∈ Λ0 and i ∈ {1, . . . , k}.
For λ ∈ Λ and 0 ≤ m ≤ n ≤ d(λ), we write λ(m, n) for the unique path in Λ such that
λ = λ′λ(m, n)λ′′, where d(λ′) = m, d(λ(m, n)) = n − m and d(λ′′) = d(λ) − n. We write
λ(n) for λ(n, n) = s(λ(0, n)). We denote by Ωk the k-graph with vertices Ω0
k := Nk, paths
Ωm
k := {(n, n + m), n ∈ Nk} for m ∈ Nk, r((n, n + m)) = n and s((n, n + m)) = n + m.
4
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
Let Λ be a row-finite k-graph with no sources. A Cuntz -- Krieger Λ-family in a C ∗-
algebra B is a function s : λ 7→ sλ from Λ to B such that
(CK1) {sv : v ∈ Λ0} is a collection of mutually orthogonal projections;
(CK2) sµsν = sµν whenever s(µ) = r(ν);
(CK3) s∗
λsλ = ss(λ) for all λ ∈ Λ; and
(CK4) sv =Pλ∈vΛn sλs∗
λ for all v ∈ Λ0 and n ∈ Nk.
The k-graph C ∗-algebra C ∗(Λ) is the universal C ∗-algebra generated by a Cuntz -- Krieger
Λ-family. For more details see [21].
3. Topological dimension zero
In this section we study when k-graph C ∗-algebras have topological dimension zero.
This requires some machinery, which we now introduce. Let Λ be a row-finite k-graph
with no sources. A subset H of Λ0 is hereditary if s(HΛ) ⊆ H. We say H ⊆ Λ0 is
saturated if for all v ∈ Λ0
{s(λ) : λ ∈ vΛei} ⊆ H for some i ∈ {1, . . . , k}
=⇒
v ∈ H.
(Since Λ has no sources, we do not need to worry about whether r−1(v) is empty -- cf. [32].)
For a hereditary H ⊆ Λ0 we write IH for the ideal in C ∗(Λ) generated by {sv : v ∈ H}.
If H is hereditary, then HΛ is a row-finite k-graph with no sources, and if H is also
saturated, then Λ \ ΛH is also a row-finite k-graph with no sources (see [21]).
Let Λ be a row-finite k-graph with no sources. The set
Λ∞ := {x : Ωk → Λ x is a degree-preserving functor}
is called the infinite-path space of Λ. For v ∈ Λ0, we write vΛ∞ := {x ∈ Λ∞ : x(0) = v},
and for x ∈ Λ∞, we write r(x) := x(0) ∈ Λ0. For p ∈ Nk, the shift map σp : Λ∞ → Λ∞
defined by (σpx)(m, n) = x(m + p, n + p) for (m, n) ∈ Ωk is a local homeomorphism. For
λ ∈ Λ and x ∈ Λ∞ with s(λ) = r(x) we write λx for the unique element y ∈ Λ∞ such
that λ = y(0, d(λ)) and x = σd(λ)y, see [21].
Definition 3.1. Let Λ be a row-finite k-graph with no sources. We say that Λ is aperiodic
(or satisfies the aperiodicity condition) if for every vertex v ∈ Λ0 there exist an infinite
path x ∈ vΛ∞ such that σm(x) 6= σn(x) for all m 6= n ∈ Nk, see [21, 32, 36]. If Λ \ ΛH is
aperiodic for every hereditary saturated H ( Λ0, we say Λ is strongly aperiodic, see [16].
We can now state our main result about topological dimension of k-graph C ∗-algebras.
Let us emphasise that half of the hard work has already been done for us: The implication
(ii)⇒(i) was established in [16] by Kang and the first named author. As we shall see the
reverse implication (i)⇒(ii) also requires some non-trivial work.
Theorem 3.2. Let Λ be a row-finite k-graph with no sources. Then the following are
equivalent:
(i) The C ∗-algebra C ∗(Λ) has topological dimension zero.
(ii) The k-graph Λ is strongly aperiodic.
Before we prove Theorem 3.2 we present six lemmata. The key ideas, namely the use of
P -graphs and induced algebras, are introduced in Lemmas 3.3 and 3.4. We briefly recall
the notation involved.
Following [16] we recall the definition of a maximal tail. Let Λ be a row-finite k-graph
with no sources. A nonempty subset T of Λ0 is called a maximal tail if
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
5
(a) for all v1, v2 ∈ T there exists w ∈ T such that v1Λw 6= ∅ and v2Λw 6= ∅,
(b) for every v ∈ T and 1 ≤ i ≤ k there exist e ∈ vΛei such that s(e) ∈ T , and
(c) for all w ∈ T and v ∈ Λ0 with vΛw 6= ∅ we have v ∈ T .
Following [6] we recall the definition of Per(Λ) and HPer. Let Λ be a row-finite k-graph
with no sources such that Λ0 is a maximal tail. Define a relation ∼ on Λ by µ ∼ ν if
and only if s(µ) = s(ν) and µx = νx for all x ∈ s(µ)Λ∞. Then [6, Lemma 4.2(1)] says
that the set Per(Λ) := {d(µ) − d(ν) : µ, ν ∈ Λ and µ ∼ ν} is a subgroup of Zk, called the
periodicity group of Λ. Moreover, it follows from [6, Lemma 4.6], that
Per(Λ) = {l ∈ Zk : there exists w ∈ Λ0 and n, m ∈ Nk such that l = n − m and
σn(y) = σm(y) for all y ∈ wΛ∞}.
Let
HPer := {v ∈ Λ0 :
for all λ ∈ vΛ and m ∈ Nk such that d(λ) − m ∈ Per(Λ),
there exists µ ∈ vΛm such that λ ∼ µ}.
Lemma 4.2 of [6] shows that HPer is a non-empty hereditary set (not always saturated).
We will also need to use the P -graphs which were introduced in [6]. The definitions of
P -graphs, where P is a finitely generated cancellative abelian monoid, and the associated
C ∗-algebras are obtained by replacing Nk with P in the definitions for k-graphs given
above, see [6, Section 2] for details. If Γ is a P -graph we will continue to call a family
satisfying (CK1) -- (CK4) for Γ with P replacing Nk a Cuntz -- Krieger Γ-family. If
P := {m + Per(Λ) : m ∈ Nk} ⊆ Zk/ Per(Λ),
then (HPerΛ)/∼ is a P -graph with operations inherited from Λ and degree map given by
d([λ]) = d(λ) + Per(Λ).
Lemma 3.3. Let Λ be a row-finite k-graph with no sources such that Λ0 is a maximal
tail. Define H := HPer and let Γ := (HΛ)/∼ be the P -graph described above. There exist
an action α of the annihilator Per(Λ)⊥ := {h ∈ Hom(Zk, T) : hPer(Λ) = 1} on C ∗(Γ) such
that
αh(s[λ]) = h(d(λ))−1s[λ],
for all
h ∈ Per(Λ)⊥ and λ ∈ HΛ.
Proof. Fix h ∈ Per(Λ)⊥ and λ, µ ∈ HΛ such that [λ] = [µ]. We prove that h(d(λ)) =
h(d(µ)). Since λ ∼ µ we have s(λ) = s(µ) and λx = µx for all x ∈ s(λ)(HΛ)∞. Set
w = s(λ) ∈ H (since H is hereditary). For any x ∈ wΛ∞ we have x ∈ s(λ)(HΛ)∞ so
σd(λ)(x) = σd(λ)+d(µ)(µx) = σd(λ)+d(µ)(λx) = σd(µ)(x). In particular d(λ) − d(µ) ∈ Per(Λ).
Since h ∈ Per(Λ)⊥ we see that h(d(λ) − d(µ)) = 1, so h(d(λ)) = h(d(µ)).
Fix h ∈ Per(Λ)⊥. By the preceding paragraph, for [λ] ∈ Γ, we can define
t[λ] := h(d(λ))−1s[λ].
Let P := {m + Per(Λ) : m ∈ Nk} ⊆ Zk/ Per(Λ). Since Γ is a P -graph and C ∗(Γ) is the
universal algebra generated by a Cuntz -- Krieger Γ-family {s[λ] : [λ] ∈ Γ = (HΛ)/∼}, the
set {t[λ] : [λ] ∈ Γ} is also a Cuntz -- Krieger Γ-family. So the universal property yields a
homomorphism αh : C ∗(Γ) → C ∗(Γ) such that αh(s[λ]) = t[λ] = h(d(λ))−1s[λ].
For each g, h ∈ Per(Λ)⊥, αhg = αh ◦ αg and αhh−1 = idC ∗(Γ) on the generators of C ∗(Γ),
and hence on all of C ∗(Γ). It follows that h 7→ αh is a group homomorphism from Per(Λ)⊥
into the automorphisms of C ∗(Γ).
6
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
Finally we show h 7→ αh(a) is continuous for each a ∈ C ∗(Γ). Let a = s[λ] for some
λ ∈ HΛ and suppose hi → h in Per(Λ)⊥. By definition of the topology on Hom(Zk, T)
we have hi(d(λ)) → h(d(λ)) in T, and so αhi(a) → αh(a) in C ∗(Γ). Since addition is
continuous, we see that αhi(a) → αh(a) for a ∈ span{tηt∗
ν : η, ν ∈ Γ}; an ε/3-argument
then gives continuity for each a ∈ C ∗(Γ).
(cid:3)
Following [39, p. 100] let X be a right G-space and (A, G, α) a dynamical system. If
f : X → A is a continuous function such that
(3.1)
f (x · s) = α−1
s (f (x)),
for all x ∈ X and s ∈ G
then x 7→ kf (x)k is continuous on the orbit space X/G := {x · G : x ∈ X}. The induced
algebra is
IndX
G (A, α) = {f ∈ Cb(X, A) : f satisfies (3.1) and x · G 7→ kf (x)k is in C0(X/G)}.
For n ∈ Nk and z ∈ Tk, we write zn forQk
In what follows we identify Per(Λ)⊥ with {z ∈ Tk : zn = 1 for all n ∈ Per(Λ)} ⊆ Tk.
Since Per(Λ)⊥ acts on Tk by right multiplication z · w := zw [39, Example 3.34], Tk is a
right Per(Λ)⊥-space. So, using the action α : Per(Λ)⊥ → Aut(C ∗(Γ)) of Lemma 3.3, we
can form the induced algebra Ind(α) := IndTk
Per(Λ)⊥(C ∗(Γ), α).
i=1 zni
i .
Recall that if Λ is a row-finite k-graph with no sources, then the universal property of
C ∗(Λ) ensures that it carries a canonical action γ, called the gauge action, of Tk satisfying
γz(sλ) = zd(λ)sλ (see [21]).
Lemma 3.4. Let Λ, H, α and Γ be as in Lemma 3.3. Then there exists an isomorphism
π : C ∗(HΛ) → Ind(α) such that π(sλ)(z) = zd(λ)s[λ] for each λ ∈ HΛ and z ∈ Tk. We
have
Prim(C ∗(HΛ)) ∼=(cid:16)Tk × Prim(C ∗(Γ))(cid:17)/ Per(Λ)⊥,
where the action of Per(Λ)⊥ on Tk × Prim(C ∗(Γ)) is given by (z, I) · w = (zw, αw−1(I)).
Proof. For each λ ∈ HΛ define tλ : Tk → C ∗(Γ) by tλ(z) = zd(λ)s[λ]. We show that each
tλ belongs to the C ∗-algebra Ind(α) described above.
For fixed λ ∈ HΛ, z 7→ zd(λ) is continuous from Tk to T, so tλ belongs to C(Tk, C ∗(Γ)).
Fix z ∈ Tk and w ∈ Per(Λ)⊥. We have
α−1
w (tλ(z)) = α−1
w (zd(λ)s[λ]) = zd(λ)αw−1(s[λ])
= zd(λ)wd(λ)s[λ] = (zw)d(λ)s[λ] = tλ(zw) = tλ(z · w).
So tλ satisfies (3.1). The function on Tk/ Per(Λ)⊥ given by z · Per(Λ)⊥ 7→ ktλ(z)k is
constant, hence continuous. Since Tk/ Per(Λ)⊥ is compact and Hausdorff (see [39, p. 101]),
the map z · Per(Λ)⊥ 7→ ktλ(z)k belongs to C0(Tk/ Per(Λ)⊥), so tλ ∈ Ind(α).
Since HΛ is a k-graph and C ∗(HΛ) is the universal algebra generated by the Cuntz --
Krieger family {sλ : λ ∈ HΛ} it easily follows that {tλ : λ ∈ HΛ} satisfies (CK1) -- (CK3).
To verify (CK4) fix v ∈ H and n ∈ Nk, let p := n + Per(Λ), and define ϕ : vΛn → vΓp, by
ϕ(λ) = [λ]. It suffices to show that ϕ is a bijection. To check that ϕ is surjective (this is
non-trivial because [λ] ∈ vΓp 6⇒ d(λ) = n), choose [λ] ∈ vΓp. Since d([λ]) = d(λ) + Per(Λ)
there exists µ ∼ λ such that d(µ) = n and r(µ) = r(λ) (see [6, Theorem 4.2]). We obtain
µ ∈ vΛn and ϕ(µ) = [λ]. To check that ϕ is injective, suppose that ϕ(λ) = ϕ(µ) for some
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
7
λ, µ ∈ vΛn. Since λ ∼ µ and d(λ) = d(µ) it follows from [6, Theorem 4.2] that λ = µ. So
ϕ is a bijection as claimed.
By the universal property of C ∗(HΛ) there exists a homomorphism π : C ∗(HΛ) →
Ind(α) such that
(3.2)
π(sλ) = tλ =(cid:0)z 7→ zd(λ)s[λ](cid:1) ,
for each λ ∈ HΛ.
We now show π is injective. Similarly to [39, Lemma 3.48], we have a dynamical system
(Ind(α), Tk, lt) where ltz(f )(x) = f (xz) for all z ∈ Tk and f ∈ Ind(α). Writing γ for the
gauge action on C ∗(HΛ), for x, z ∈ Tk and λ ∈ HΛ, we have
ltz ◦ π(sλ)(x) = ltz(tλ)(x) = tλ(xz) = (xz)d(λ)s[λ] = zd(λ)xd(λ)s[λ]
= zd(λ)tλ(x) = zd(λ)π(sλ)(x) = π(zd(λ)sλ)(x) = π ◦ γz(sλ)(x).
So the homomorphisms ltz ◦ π and π ◦ γz agree on generators, and hence are equal. Since
each tλ is non-zero it follows from the gauge-invariant uniqueness theorem [21, Theo-
rem 3.4] that π is injective.
We now show that π is surjective. By [39, Proposition 3.49], Ind(α) can be regarded
as a C0(Tk/ Per(Λ)⊥)-algebra in the sense of [39], where
(3.3)
ϕ · f (z) = ϕ(z · Per(Λ)⊥)f (z)
for all ϕ ∈ C0(Tk/ Per(Λ)⊥), f ∈ Ind(α) and z ∈ Tk.
In particular it follows from
[39, Proposition C.24 and Theorem C.26] that π is surjective provided the following two
properties hold:
(i) If f ∈ im(π) and ϕ ∈ C0(Tk/ Per(Λ)⊥), then ϕ · f ∈ im(π).
(ii) For each z ∈ Tk, {f (z) : f ∈ im(π)} is dense in C ∗(Γ).
To prove (i), observe that for n ∈ Per(Λ), the map z 7→ zn vanishes on Per(Λ)⊥, and so
descends to an element εn of C0(Tk/ Per(Λ)⊥) satisfying
(3.4)
εn(z · Per(Λ)⊥) = zn.
The set {εn : n ∈ Per(Λ)} generates C0(Tk/ Per(Λ)⊥) as a C ∗-algebra. So (i) follows once
we prove that im(π) is invariant for the action on every εn. For f ∈ Ind(α) and z ∈ Tk,
equations (3.3) and (3.4) give
(εn · f )(z) = εn(z · Per(Λ)⊥)f (z) = znf (z).
Fix n ∈ Per(Λ) and f ∈ im(π). We show that the function z 7→ znf (z) belongs to im(π).
Write n = p − q with p, q ∈ Nk and fix λ, µ ∈ HΛ. By [6, Theorem 4.2(3)] there exist a
bijection θ : s(λ)Λp → s(λ)Λq such that ν ∼ θ(ν) for all ν ∈ s(λ)Λp. Since {s[λ] : [λ] ∈ Γ}
8
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
satisfies (CK4) and the map ν 7→ [ν] is a bijection from s(λ)Λp to s(λ)Γp+Per(Λ) we have,
ts(λ)(z) = zd(s(λ))s[s(λ)] = s[s(λ)] = X[ν]∈s(λ)Γp+Per(Λ)
s[ν]s∗
[ν]
[ν] = Xν∈s(λ)Λp
s[ν]s∗
[θ(ν)]
s[ν]s∗
= Xν∈s(λ)Λp
= Xν∈s(λ)Λp
= z−(p−q) Xν∈s(λ)Λp
z−d(ν)tν(z)(z−d(θ(ν))tθ(ν)(z))∗
tνt∗
θ(ν)(z).
Hence (cid:0)z 7→ zntλt∗
f ∈ span{tλt∗
µ(z)(cid:1) = (cid:16)z 7→(cid:0)Pν∈s(λ)Λp tλtνt∗
µ : λ, µ ∈ HΛ} we obtain (z 7→ znf (z)) ∈ im(π).
θ(ν)t∗
µ(cid:1)(z)(cid:17) ∈ im(π). Finally using that
To prove (ii), fix z ∈ Tk. For each λ ∈ HΛ define fλ := z−d(λ)tλ ∈ im(π). Using (3.2)
we get s[λ] = fλ(z), so s[λ] ∈ {f (z) : f ∈ im(π)}. Since im(π) is a C ∗-algebra and since
span{fλf ∗
µ(z) : λ, µ ∈ HΛ} = C ∗(Γ), the set {f (z) : f ∈ im(π)} is dense in C ∗(Γ). We
conclude that C ∗(HΛ) ∼= Ind(α).
By [39, p. 100], the primitive ideal space of the induced algebra Ind(α) is homeomorphic
with the orbit space (Tk × Prim(C ∗(Γ)))/ Per(Λ)⊥, where the right action of Per(Λ)⊥ is
given by (z, I)·w := (zw, αw−1(I)) [39, Lemma 2.8]. In particular, since C ∗(HΛ) ∼= Ind(α),
we obtain Prim(C ∗(HΛ)) ∼= Prim(Ind(α)) ∼= (Tk × Prim(C ∗(Γ)))/ Per(Λ)⊥.
(cid:3)
Lemma 3.5. Let Λ and Γ be as in Lemma 3.3. The formula
ϕ((z, I) · Per(Λ)⊥) = z · Per(Λ)⊥,
determines a continuous open surjection ϕ :(cid:0)Tk×Prim(C ∗(Γ))(cid:1)/ Per(Λ)⊥ → Tk/ Per(Λ)⊥.
Proof. For completeness let us mention that Prim(C ∗(Γ)) is non-empty (for example, the
zero ideal in C ∗(Γ) is primitive) so ϕ makes sense. If (x1, I1) · Per(Λ)⊥ = (x2, I2) · Per(Λ)⊥
for (xi, Ii) ∈ Tk × Prim(C ∗(Γ)), then x1w = x2 for some w ∈ Per(Λ)⊥. So the map ϕ is
well defined.
Let ψ : Tk × Prim(C ∗(Γ)) → Tk be the projection onto the first coordinate and write
p for both the quotient map Tk → Tk/ Per(Λ)⊥ and the orbit map Tk × Prim(C ∗(Γ)) →
(Tk × Prim(C ∗(Γ)))/ Per(Λ)⊥ (cf. [39, Definition 3.21]). Then the diagram
(Tk × Prim(C ∗(Γ)))/ Per(Λ)⊥
p
Tk × Prim(C ∗(Γ))
ϕ
ψ
/ Tk/ Per(Λ)⊥
,
p
Tk
commutes. The topologies on the orbit spaces
(Tk × Prim(C ∗(Γ)))/ Per(Λ)⊥
and Tk/ Per(Λ)⊥
are the weakest topologies making the maps p continuous. So both maps p are continuous
and open (see [39, Lemma 3.25]). Now ϕ is continuous and open because ψ is. The map
ϕ is surjective because Prim(C ∗(Γ)) is non-empty.
(cid:3)
/
/
/
O
O
O
O
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
9
We will need an elementary lemma from topology. This is standard, but we give a
complete statement and proof for convenience.
Lemma 3.6. Let X, Y be topological spaces and let ϕ : X → Y be a continuous open
surjective map. If X admits a basis for its topology consisting of compact and open sets,
then so does Y .
Proof. Let B be a basis for the topology on X consisting of compact and open sets. Set
F := {ϕ(B) : B ∈ B}. Since ϕ is continuous and open each element of F is compact
and open. Since ϕ is surjective Y = ϕ(SB∈B B) =SF ∈F F , so F is an open cover of Y .
Suppose y ∈ F1 ∩ F2 for F1, F2 ∈ F. Take x ∈ X such that ϕ(x) = y. Since x ∈ ϕ−1(y) ⊆
ϕ−1(Fi) and Fi is open, there exists B ∈ B such that x ∈ B ⊆ ϕ−1(F1) ∩ ϕ−1(F2). With
F := ϕ(B) ∈ F we get y = ϕ(x) ∈ ϕ(B) = F and F = ϕ(B) ⊆ ϕ(ϕ−1(Fi)) = Fi. So F is
a basis for a topology on Y .
To see that this topology coincides with the given topology on Y , observe that each set
in F is open, and for any open set V ⊆ Y , the preimage ϕ−1(V ) is open, hence a union of
elements Bi ∈ B. Now each ϕ(Bi) ∈ F and V = ϕ(ϕ−1(V )) = ϕ(Si Bi)) =Si ϕ(Bi). (cid:3)
We write M(A) for the multiplier algebra of a C ∗-algebra A.
Lemma 3.7. Let Λ be a row-finite k-graph with no sources. Suppose that C ∗(Λ) has
topological dimension zero. Suppose that H ⊆ Λ0 is hereditary. Let I be the ideal of C ∗(Λ)
generated by {pv : v ∈ H}, and let PH := Pv∈H pv ∈ M(I). There is an isomorphism
C ∗(HΛ) ∼= PHIPH carrying the generator sλ of C ∗(HΛ) to the corresponding generator
sλ of C ∗(Λ) for every λ ∈ HΛ, and PHIPH is a full corner of I. Furthermore, C ∗(HΛ)
has topological dimension zero. If H is also saturated then C ∗(Λ \ ΛH) is isomorphic to
C ∗(Λ)/I and has topological dimension zero.
Proof. Recall from [8, Proposition 3.2.1] and [5, Proposition 2.8] that topological dimen-
sion zero passes to ideals and quotients.
It also passes to full corners (since Morita
equivalent C ∗-algebras have homeomorphic primitive ideal spaces, see [7, p. 156]).
Let H be any hereditary subset of Λ0. By [32, Theorem 5.2], C ∗(HΛ) is isomorphic to
PHIPH. This is a full corner because I is generated by PH. Since topological dimension
zero passes to ideals and full corners, we deduce that C ∗(HΛ) has topological dimension
zero. If H is saturated as well, then C ∗(Λ \ ΛH) is isomorphic to C ∗(Λ)/I, and hence it,
too, has topological dimension zero.
(cid:3)
Lemma 3.8. Let Λ be a row-finite k-graph with no sources. Suppose that Λ is not strongly
aperiodic. Then there exists a (possibly empty) hereditary and saturated set H ⊆ Λ0 such
that the vertex set of Λ \ ΛH is a maximal tail and Per(Λ \ ΛH) 6= {0}.
Proof. Choose a (possibly empty) hereditary saturated set H ′ ⊆ Λ0 such that Λ \ ΛH ′ is
not aperiodic. By [36, Lemma 3.2(iii)] there exist a vertex v ∈ Λ \ ΛH ′ and n 6= m ∈ Nk
such that σm(y) = σn(y) for every infinite path y ∈ v(Λ \ ΛH ′)∞. Fix x ∈ v(Λ \ ΛH ′)∞,
define the shift-tail equivalence class
(3.5)
[x] := {λσp(x) : p ∈ Nk, λ ∈ (Λ \ ΛH ′)x(p)},
and let T := r([x]) ⊆ Λ0 (possibly all of Λ0) and H := Λ0 \ T .
We prove that T ⊆ Λ0 is a maximal tail. First we consider property (a). If v, w ∈ r([x]),
say v = r(ασm(x)), w = r(βσn(x)), then u = x(m ∨ n) ∈ T satisfies αx(m, m ∨ n) ∈ vΛu,
and βx(n, m ∨ n) ∈ wΛu. For property (b), fix v ∈ r([x]), say v = r(ασm(x)), and fix
10
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
i ≤ k. By replacing α with αx(m, m + ei) and m with m + ei, we may assume that
d(α) ≥ ei, say α = f α′ with f ∈ Λei. Then f ∈ vΛei, and s(f ) = r(α′σm(x)) ∈ r([x]).
Finally for property (c), fix w ∈ r([x]) and v ∈ Λ0 with vΛw 6= ∅, say α ∈ vΛw. Write
w = r(βσm(x)). Then v = r((αβ)σm(x)) ∈ r([x]).
Now H ⊆ Λ0 is hereditary and saturated because it is the complement of a maximal tail
(see the proof of [16, Theorem 3.12]). Since T (Λ \ ΛH)T = Λ \ ΛH, the set T = (Λ \ ΛH)0
is a maximal tail in the k-graph Λ \ ΛH.
Since Per(Λ \ ΛH) = {a − b : there exists w ∈ (Λ \ ΛH)0 such that σa(y) = σb(y) for
all y ∈ w(Λ \ ΛH)∞} and since (Λ \ ΛH)0 = r([x]) ⊆ {r(λ) : λ ∈ Λ \ ΛH ′} = (Λ \ ΛH ′)0
we conclude that n − m ∈ Per(Λ \ ΛH) 6= {0}.
(cid:3)
Proof of Theorem 3.2. As mentioned before, (ii)⇒(i) follows from [16, Theorem 4.2]. For
(i)⇒(ii), let Λ′ be a row-finite k-graph with no sources such that C ∗(Λ′) has topological
dimension zero. We suppose that Λ′ is not strongly aperiodic, and derive a contradiction.
By Lemma 3.8 there is a hereditary and saturated set H ′ ⊆ (Λ′)0 such that the vertex set
Λ0 of Λ := Λ′ \Λ′H ′ is a maximal tail and such that Per(Λ) is nontrivial. Define H := HPer
Lemma 3.5, there is a continuous open surjective map
and Γ := (HΛ)/∼. By Lemma 3.4, Prim(C ∗(HΛ)) ∼=(cid:0)Tk × Prim(C ∗(Γ))(cid:1)/ Per(Λ)⊥. By
Lemma 3.7 implies that C ∗(HΛ) has topological dimension zero, and it follows that(cid:0)Tk ×
Prim(C ∗(Γ))(cid:1)/ Per(Λ)⊥ has a basis of compact open sets. Using Lemma 3.6 we conclude
ϕ :(cid:0)Tk × Prim(C ∗(Γ))(cid:1)/ Per(Λ)⊥ → Tk/ Per(Λ)⊥.
that Tk/ Per(Λ)⊥ has a basis of compact open sets. But Tk/ Per(Λ)⊥ is the Pontryagin
dual of a free abelian group, and hence homeomorphic to Tl where l is the rank of Per(Λ).
So Tk/ Per(Λ)⊥ is connected and not a singleton, a contradiction.
(cid:3)
Recall that a topological groupoid G is topologically principal if the set
{u ∈ G(0) : Gu
u = {u}}
of units with trivial isotropy is dense in G(0) [34, Definition 1.2], and essentially principal
if for every closed invariant X ⊆ G(0), the set {u ∈ X : Gu
u = {u}} is dense in X [33,
Definition 4.3]. For each k-graph Λ with no sources we let GΛ denote the infinite-path
groupoid of Λ introduced in [21, Definition 2.7].
Corollary 3.9. Let Λ be a row-finite k-graph with no sources. Then the following are
equivalent:
(i) The k-graph Λ is strongly aperiodic.
(ii) For every hereditary saturated subset H ( Λ0, the groupoid GΛ\ΛH is topologically
principal.
(iii) The groupoid GΛ is essentially principal.
(iv) The C ∗-algebra C ∗(Λ) has topological dimension zero.
(v) Every ideal of C ∗(Λ) is gauge invariant.
(vi) The map I 7→ {v ∈ Λ0 : sv ∈ I} is a bijection between ideals of C ∗(Λ) and
hereditary saturated subsets H ⊆ Λ0.
Proof. See [21, Proposition 4.5] for (i)⇔(ii). For (ii)⇔(iii), one checks that the nonempty
closed invariant subsets of G(0)
Λ are precisely the sets (Λ \ ΛH)∞ ⊆ Λ∞ associated to
hereditary saturated sets H ( Λ0. This follows from the argument of [22, Proposition 6.5]
mutatis mutandis.
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
11
Theorem 3.2 gives (i)⇔(iv). For (i)⇔(v) see [36, Proposition 3.6] and see [37, Theo-
rem 5.5] for (v)⇒(vi). For (vi)⇒(v) observe that the sv are fixed under the gauge action,
so any ideal generated by vertex projections is gauge invariant.
(cid:3)
4. Real rank zero
In this section we study when k-graph C ∗-algebras have real-rank zero. We can give a
complete answer for 2-graphs Λ such that C ∗(Λ) is strongly purely infinite, and we are
able to say some things in greater generality. We start with a necessary condition that
follows easily from work of Brown -- Pedersen (it could also be deduced easily from work
of Pasnicu [29]) and requires no additional hypotheses. It also connects the question of
real-rank zero to our work in the preceding section.
Lemma 4.1. Let Λ be a row-finite k-graph with no sources. If the C ∗-algebra C ∗(Λ) has
real-rank zero then Λ is strongly aperiodic.
Proof. Suppose that C ∗(Λ) has real-rank zero. Then it has generalised real-rank zero in
the sense of [5, 2.1(iv)], and so Proposition 2.7 of [5] implies that it also has topological
dimension zero. Hence (i) =⇒ (ii) of Theorem 3.2 implies that Λ is strongly aperiodic. (cid:3)
Our objective in the rest of the section is to strengthen this necessary condition and
obtain a necessary and sufficient condition for real-rank zero when C ∗(Λ) is purely infinite
and k = 2.
an elementary wedge product. If {ǫ1, ǫ2, . . . , ǫN } constitutes a basis for M (with a ≤ N)
i.e. the R-module M ⊗a/Ja where Ja is the submodule of M ⊗a spanned by elements of the
form m1 ⊗ m2 ⊗ · · · ⊗ ma such that mi = mj for some i 6= j. For any m1, m2, . . . , ma ∈ M,
For an R-module M and an integer a ≥ 2 letVaM denote the ath exterior power of M,
the coset of m1 ⊗ m2 ⊗ · · · ⊗ ma inVaM is denoted m1 ∧ m2 ∧ · · · ∧ ma and we call it
then {ǫi1 ∧ ǫi2 ∧ · · · ∧ ǫia : 1 ≤ i1 < · · · < ia ≤ N} is a basis forVaM. We regard R as an
R-module and setV0 M = R andV1 M = M.
Let Λ be a row finite k-graph with no sources. Following [23] we introduce the chain
complex DΛ
∗ . Let ZΛ0 denote the set of finitely supported functions from Λ0 to Z, regarded
as an abelian group under pointwise addition. For v ∈ Λ0 define δv ∈ ZΛ0 by δv(u) = δu,v
(the Kronecker delta). For each i = 1, . . . , k let Mi (or Mi,Λ for emphasis) be the vertex
connectivity matrix given by Mi(u, v) = uΛeiv. Regard Mi as the group endomorphism
∗ be the chain complex such that
of ZΛ0 given by (Mif )(u) =Pv∈Λ0 Mi(u, v)f (v). Let DΛ
if 0 ≤ a ≤ k,
if a > k,
with differentials ∂a : DΛ
a → DΛ
a for emphasis) defined for 1 ≤ a ≤ k by
DΛ
0
a =(Va Zk ⊗ ZΛ0
aXj=1
a−1 (or ∂Λ
∂a(ǫi1 ∧ · · · ∧ ǫia ⊗ δv) =
(−1)j+1ǫi1 ∧ · · · ∧bǫij ∧ · · · ∧ ǫia ⊗ (1 − M t
j )δv,
basis for Zk and M t
where the symbol "b· " denotes deletion of an element, v ∈ Λ0, {ǫ1, . . . , ǫk} is the canonical
Example 4.2. Let Λ be a row finite k-graph with no sources, H ( Λ0 a hereditary saturated
subset of Λ0 and Γ = HΛ. Recall that a morphism of chain complexes f : C∗ → D∗ is
j is the transpose of Mj.
12
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
given by a family of morphisms fn : Cn → Dn that commute with the differentials. In
this example we construct a morphism of chain complexes jH : DΓ
∗ → DΛ
∗ .
For each f ∈ ZΓ0 let iH(f ) be the extension of f to Λ0 by zero. Since H is hereditary,
for u ∈ H and v 6∈ H we have Mj,Λ(u, v) = 0. It follows that for each v ∈ Λ0
(M t
j,Λ(iH (f )))(v) = Xu∈Λ0
Mj,Λ(u, v)(iH(f )(u)) = Xu∈Γ0
Mj,Λ(u, v)(f (u))
Mj,Λ(u, v)(f (u)) = (iH(M t
j,Γf ))(v).
= 1Γ0(v)Xu∈Γ0
We conclude that iH ((1−M t
jH
a : DΓ
produces the commuting diagram
a → DΛ
a be the morphism given by jH
j,Γ)f ) = (1−M t
j,Λ)(iH (f )) for each f ∈ ZΓ0. For 0 ≤ a ≤ k let
a (d ⊗ f ) = d ⊗ iH(f ). The above calculation
0
0
∂Γ
k
∂Λ
k
/ DΓ
k
jH
k
/ DΛ
k
/ DΓ
k−1
/ · · ·
/ DΓ
1
jH
k−1
jH
1
/ DΛ
k−1
/ · · ·
/ DΛ
1
∂Γ
1
∂Λ
1
/ DΓ
0
jH
0
/ DΛ
0
/ 0
/ 0
and hence yields a morphism of chain complexes jH : DΓ
∗ → DΛ
∗ .
Any morphism of chain complexes induces a morphism of the homology groups of the
∗ ) →
In particular Example 4.2 yields the morphism H∗(jH) : H∗(DΓ
chain complexes.
H∗(DΛ
∗ ). We can now state our main result.
Theorem 4.3. Let Λ be a row-finite 2-graph with no sources.
(1) Suppose C ∗(Λ) is purely infinite. Then the following are equivalent:
(i) C ∗(Λ) has real rank zero
(ii) Λ is strongly aperiodic and H1(jH) is injective for every saturated hereditary
subset H of Λ0.
(2) The implication (i) ⇒ (ii) holds even if C ∗(Λ) is not purely infinite.
We will prove Theorem 4.3 at the end of the section, after establishing some preliminary
results. First though, we present a reformulation of the condition that H1(jH) is injective
in terms of the connectivity matrices of the 2-graph Λ. This result will prove useful
when applying Theorem 4.3 in practice. To state it, we need a little bit more notation.
Suppose that Λ is a 2-graph and H ⊆ Λ0 is a hereditary set. It follows that, for each i,
Mi,Λ(u, v) = 0 whenever u ∈ H and v 6∈ H. This means that each M t
i,Λ has a block-upper-
triangular decomposition with respect to the decomposition Λ0 = H ⊔ (Λ0 \ H). We will
use the following notation for this decomposition:
i,H M t
i,H,Λ0\H
M t
0
(4.1)
M t
i,Λ =(cid:18) M t
i,Λ0\H (cid:19) .
Proposition 4.4. Let Λ be a row-finite 2-graph with no sources, and let H ⊆ Λ0 be a
saturated hereditary set. With H1(jH), ∂1, ∂2 and Γ as in Example 4.2, the following are
equivalent:
(i) The morphism H1(jH ) is injective.
(ii) We have im ∂Λ
1 (ker ∂Γ
1 ) ⊆ jH
2 ∩ jH
1 (im ∂Γ
2 ).
/
/
/
/
/
/
/
/
/
/
/
/
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
13
(iii) We have
1,H,Λ0\H
2,H,Λ0\H(cid:19)(cid:0) ker(M t
(cid:18)M t
M t
1,Λ0\H − 1) ∩ ker(M t
2,Λ0\H − 1)(cid:1) ⊆(cid:18)M t
M t
1,H − 1
2,H − 1(cid:19)ZH.
Proof. Before proving (i)⇔(ii), we do some initial work to reformulate condition (i). Since
jH
1 ◦ ∂Γ
2 . Hence there is a well-defined homomor-
phism H1(jH) : ker ∂Γ
2 , we have jH
2 ) ⊆ im ∂Λ
1 / im ∂Λ
1 (im ∂Λ
2 → ker ∂Λ
2 such that
1 / im ∂Γ
2 = ∂Λ
2 ◦ jH
H1(jH )(cid:0)a + im ∂Γ
2(cid:1) = jH
1 (a) + im ∂Λ
2
for all a ∈ ker ∂Γ
1 .
2 , we have a ∈ im ∂Γ
2 .
By linearity, it follows that H1(jH) is injective if and only if whenever a ∈ ker ∂Γ
jH
1 (a) ∈ im ∂Λ
Now, to prove (i)⇒(ii), suppose that H1(jH) is injective, and fix a ∈ ker ∂Γ
jH
1 (a) ∈ im ∂Λ
jH
1 (im ∂Γ
To verify (ii)⇒(i), we suppose that (ii) holds, and fix a ∈ ker ∂Γ
2 . Then the preceding paragraph shows that a ∈ im ∂Γ
1 (a) ∈ im ∂Λ
2 .
1 (a) ∈
1 (c) for some c ∈ im ∂Γ
2 .
1 (d ⊗ f ) = d ⊗ iH (f ), where iH(f ) is the extension of f ∈ ZΓ0 to Λ0 by zero, jH
1
By the first paragraph of the proof, it suffices to show that a ∈ im ∂Γ
im ∂Λ
Since jH
is injective, so a = c ∈ im ∂Γ
2 .
1 such that
1 (a) ∈
1 ). By (ii), jH
2 . We have jH
2 , giving jH
1 (a) = jH
1 (a) ∈ jH
1 with jH
2 ), so jH
1 satisfies
1 (ker ∂Γ
1 (im ∂Γ
2 ∩ jH
2 ).
We now prove that (ii)⇔(iii). By [10, 23], the complex DΛ
Γ = HΛ. To ease notation in the calculations that follow, we write T := Λ0 \ H, and we
write A := M t
1,Λ and B := M t
2,Λ, so that
K1 := im ∂Λ
2 ,
K2 := jH
1 (ker ∂Γ
1 ),
and
L1 := jH
1 (im ∂Γ
2 ),
so that (ii) is satisfied if and only if K1 ∩ K2 ⊆ L1. We have
0
0
L1 = jH
K2 = jH
BT − 1(cid:17)(cid:16)a
1 − AT(cid:17)(cid:16)a
b(cid:17) ∈ ZΛ0(cid:27),
1 ({(u, v) : u, v ∈ ZH and AH u + BH v = u + v})
b(cid:17),(cid:16)1 − AH −AH,T
K1 =(cid:26)(cid:18)(cid:16)BH − 1 BH,T
=n(cid:16)(cid:0) u
0(cid:1),(cid:0) v
1 (cid:16)(cid:8)(cid:0)(BH − 1)a, (1 − AH)a(cid:1) : a ∈ ZH(cid:9)(cid:17)
=n(cid:16)(cid:0) (BH −1)a
0(cid:1)(cid:17) has the
0(cid:1),(cid:0) v
2(cid:0) a
b(cid:1) where b ∈ ker(1−BT )∩ker(1−AT ). Conversely, if b ∈ ker(1−BT )∩ker(1−AT )
b(cid:17)(cid:19) :(cid:16)a
0(cid:1)(cid:17) : u, v ∈ ZH and AH u + BH v = u + vo, and
(cid:1),(cid:0) (1−AH )a
0(cid:1)(cid:17) ∈ K1∩K2. The description of K1 shows that(cid:16)(cid:0) u
(cid:1)(cid:17) : a ∈ ZHo.
0(cid:1),(cid:0) v
0
0
Suppose that(cid:16)(cid:0) u
form ∂Λ
written as follows:
0
/ ZΛ0
∂Λ
2
/ ZΛ0 ⊕ ZΛ0
where ∂Λ
1 = (1 − M t
1,Λ, 1 − M t
2,Λ) and ∂Λ
2 =(cid:18)M t
2,Λ − 1
1 − M t
1 − M t
1,Λ =(cid:18) 1 − AH −AH,T
1 − AT (cid:19)
0
and
Define
∂Λ
1
/ 0
/ ZΛ0
a = Va Zk ⊗ ZΛ0 may be
1,Λ(cid:19). The same applies to the 2-graph
2,Λ =(cid:18) 1 − BH −BH,T
1 − BT (cid:19) .
1 − M t
0
/
/
/
/
14
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
and a ∈ ZH, then we have
∂Λ
b(cid:1) =(cid:16)(cid:0) (BH −1)a+BH,T b
2(cid:0) a
0
(cid:1),(cid:0) (1−AH )a−AH,T b
0
(cid:1)(cid:17).
For a ∈ ZΛ0, we write aH for the component (av)v∈H of a belonging to ZH. Using that
M t
2,Λ commute, we have
1,Λ and M t
1,Λ(M t
2,Λ(1 − M t
AH(cid:0)(BH − 1)a + BH,T b(cid:1) + BH(cid:0)(1 − AH )a − AH,T b(cid:1)
b(cid:1)(cid:17)H
+(cid:16)M t
2,Λ − 1)(cid:0) a
b(cid:1)(cid:17)H
+(cid:16)(1 − M t
2,Λ − 1)(cid:0) a
2(cid:0) a
b(cid:1) belongs to K1 ∩ K2. That is, we have showed that
L2 := {(cid:0)(cid:0) BH,T b
b(cid:1)(cid:17)H
=(cid:16)M t
1,Λ)(cid:0) a
b(cid:1)(cid:17)H
=(cid:16)(M t
1,Λ)(cid:0) a
=(cid:0)(BH − 1)a + BH,T b(cid:1) +(cid:0)(1 − AH )a − AH,T b(cid:1).
2(cid:0) a
b(cid:1) : a ∈ ZH and b ∈ ker(1 − BT ) ∩ ker(1 − AT )}.
(cid:1)(cid:1) : b ∈ ZT and b ∈ ker(1 − BT ) ∩ ker(1 − AT )}.
(cid:1),(cid:0) −AH,T b
K1 ∩ K2 = {∂Λ
0
0
Thus ∂Λ
Let
We have shown that K1 ∩ K2 = L1 + L2, and hence (ii) is satisfied if and only if L2 ⊆ L1.
As the pairs of vectors in L1 and L2 have zeros at the second row, it follows that (ii) holds
if and only if
{(BH,T b, −AH,T b) : b ∈ ZT and b ∈ ker(1 − BT ) ∩ ker(1 − AT )}
⊆(cid:8)(cid:0)(BH − 1)a, (1 − AH)a(cid:1) : a ∈ ZH(cid:9).
Multiplying by −1 in the second coordinate and recalling the definitions of the matrices
A and B gives (ii)⇔(iii).
(cid:3)
We now embark on the preliminary results needed for the proof of Theorem 4.3. Let Λ
be a row finite k-graph with no sources. As usual, given an action α of a locally compact
group G on a C ∗-algebra A, we write iG : C ∗(G) → M(A ×α G) and iA : A → M(A ×α G)
for the canonical inclusions into the multiplier algebra of the crossed product.
We will abbreviate the crossed product C ∗(Λ)×γ Tk of C ∗(Λ) by its gauge action by BΛ.
For each n ∈ Zk and f ∈ Cc(Tk, C ∗(Λ)) define αΛ
n extends to
an automorphism of BΛ and produces a dynamical system (BΛ, Zk, αΛ). The action αΛ
is called the dual action, see [39, p. 190].
n (f )(z) := z−nf (z). Then αΛ
Takai duality implies that C ∗(Λ) is isomorphic to a full corner of BΛ ×αΛ Zk. Specif-
ically, there is an inclusion map θΛ : C ∗(Λ) → BΛ ×αΛ Zk, which is given by θΛ(sλ) :=
iBΛ(iT(1)iC ∗(Λ)(sλ))iZk (d(λ))∗, and which induces the identity map in K-theory1.
Lemma 4.5. Let Λ be a row finite k-graph with no sources, H ( Λ0 a hereditary saturated
subset of Λ0 and Γ = HΛ. There is a homomorphism ϕ : C ∗(Γ) → C ∗(Λ) satisfying
(4.2)
ϕ(sλ) = sλ
for each
λ ∈ Γ.
1This is discussed in detail in terms of skew-product graphs in [24, Lemma 5.2]; to translate back to
the language of crossed-products, recall that there is an isomorphism C ∗(Λ) ×γ Tk ∼= C ∗(Λ ×d Zk) that
carries each iC ∗(Λ)(sλ)iTk (zn) to s(λ,n) [21, Corollary 5.3].
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
15
This ϕ extends to a homomorphism ϕ : BΓ ×αΓ Zk → BΛ ×αΛ Zk such that the following
diagram commutes:
(4.3)
C ∗(Γ)
θΓ
/ BΓ ×αΓ Zk
ϕ
ϕ
C ∗(Λ)
θΛ /
/ BΛ ×αΛ Zk.
Proof. Lemma 3.7 gives a homomorphism ϕ : C ∗(Γ) → C ∗(Λ) satisfying (4.2). This ϕ
is equivariant for the gauge actions on C ∗(Γ) and C ∗(Λ), so it induces a homomorphism
ϕ × id : BΓ → BΛ mapping Cc(Tk, C ∗(Γ)) to Cc(Tk, C ∗(Λ)) via ϕ × id(f )(z) = ϕ(f (z)),
see [39, Corollary 2.48].
By construction, we have (ϕ × id) ◦ iC ∗(Γ)(sλ) = iC ∗(Λ) ◦ ϕ(sλ) for every λ ∈ Γ. The
commutativity of the diagram
Cc(Tk, C ∗(Γ))
ϕ×id
Cc(Tk, C ∗(Λ))
αΓ
n
αΛ
n
Cc(Tk, C ∗(Γ))
ϕ×id
/ Cc(Tk, C ∗(Λ)),
shows that ϕ × id is equivariant for the dual actions of Zk on BΓ and BΛ. Hence ϕ × id
induces a homomorphism ϕ : BΓ ×αΓ Zk → BΛ ×αΛ Zk. Using the description of θΛ and
θΓ above, we see that the diagram (4.3) commutes.
(cid:3)
Following [23] a homology spectral sequence (Er, dr, αr) or simply (Er, dr) consists of
the following data:
(i) A family of Z-modules Er
(ii) Maps dr
(iii) Maps αr
a,b : Er
a,b : Er+1
a,b → Er
a,b → H(Er
a,b)/ im(dr
module ker(dr
a+r,b−r+1).
a,b defined for all integers a, b and r > 0.
a−r,b+r−1 that are differentials, i.e, dr
a−r,b+r−1 ◦ dr
a,b = 0
a,b) that are isomorphisms, where H(Er
a,b) denotes the
A morphism of spectral sequences f : (Er, dr) → ( Er, dr) is a family of morphisms
f r : Er → Er such that (1) the morphisms f r commute with the differentials, and (2)
the induced homomorphisms H(f r
a,b) make the following diagram
commute:
a,b) → H( Er
a,b) : H(Er
αr
a,b
Er+1
a,b
/ H(Er
a,b)
f r+1
a,b
H(f r
a,b)
αr
a,b
Er+1
a,b
/ H( Er
a,b).
Recall that the spectral sequence Er
a,b is bounded if for each r and n there are just finitely
many nonzero terms Er
a,b such that a + b = n (note that this holds for all r if it holds for
∼= Er
r = 1). We then eventually have Er+1
a,b for all a, b, and we write E∞
a,b for this limiting
a,b
term. We say that the spectral sequence Er
a,b converges to a sequence K∗ = {Kn : n ∈ Z} of
modules if each Kn admits a finite filtration 0 = Fs(Kn) ⊆ Fs+1(Kn) ⊆ · · · ⊆ Ft(Kn) = Kn
such that E∞
a,b
∼= Fa(Ka+b)/Fa−1(Ka+b) for s < a ≤ t.
/
/
/
/
/
/
16
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
The following result essentially follows from the argument of [10, Theorem 3.15] (though
we are following the notation of [23, Proposition 5.4 and Theorem 5.5]), but we need a
more detailed statement.
Lemma 4.6. Let Λ, Γ and ϕ be as in Lemma 4.5. There exist spectral sequences (Er, dr)
and ( Er, dr) and a morphism of spectral sequences f : (Er, dr) → ( Er, dr) such that
(a) (Er, dr) converges to K∗(BΓ ×αΓ Zk) and ( Er, dr) converges to K∗(BΛ ×αΛ Zk),
(b) E1
(c) d1
a,b = ∂Λ
(d) the isomorphisms E∞
a,b
a for 1 ≤ a ≤ k and b even,
∼= K∗(BΓ ×αΓ Zk) and E∞
a for 0 ≤ a ≤ k and b even,
a and E1
a and d1
a,b = DΓ
a,b = ∂Γ
a,b = DΛ
∼= K∗(BΛ ×αΛ Zk) of part (a)
a,b and the homomorphisms K∗( ϕ) :
a,b
intertwine the morphisms f ∞
K∗(BΓ ×αΓ Zk) → K∗(BΛ ×αΛ Zk), and
a,b : DΓ
a coincides with the map jH
a,b : E∞
a → DΛ
a,b → E∞
(e) f 1
a of Example 4.2.
Proof. Firstly we briefly recall parts of the proof of [23, Theorem 5.5] that we will need.
Since the vertex projections sv ∈ C ∗(Λ) are fixed by the gauge action, the map that sends
a ∈ C ∗(Λ) to the constant function z 7→ a ∈ C(Tk, C ∗(Λ)) determines a homomorphism
εΛ : C0(Λ0) ∼= C ∗({sv : v ∈ Λ0}) → BΛ; so εΛ(δv) is the constant function z 7→ sv on Tk.
Let εΛ
∗ : ZΛ0 → K0(BΛ) be the induced map in K0. The map
is a map of complexes that induces an isomorphism on homology [10, Theorem 3.14].
Thus, as in [10, 23],
It follows that the spectral sequence ( Er, dr) converges to K∗(BΛ ×αΛ Zk) as required.
We now diverge slightly from the proof of [23, Theorem 5.5] and apply Lemma 4.5: Fix
a hereditary saturated set H ( Λ0, and let Γ = HΛ. Repeating the previous argument
on Γ (using E rather than E for the new sequence) we get spectral sequences (Er, dr),
( Er, dr) satisfying properties (a) -- (c).
Define εΓ
∗ in the same way as εΛ
∗ and let
ι : C0(Γ0) → C0(Λ0)
be the extension map extending by zero. Using Lemma 4.5 and the canonical identi-
fications K0(C0(Γ0)) ∼= ZΓ0 and K0(C0(Λ0)) ∼= ZΛ0, we have the commuting diagram
below:
(4.5)
ZΓ0
K0(ι)
ZΛ0
εΓ
∗
εΛ
∗
K0(BΓ)
K0(ϕ×id)
/ K0(BΛ).
(4.4)
Therefore setting E1
even (and zero otherwise), yields
a,b = DΛ
id ⊗εΛ
∗ :V∗ Zk ⊗ ZΛ0 →V∗ Zk ⊗ K0(BΛ)
H∗(Zk, K0(BΛ)) ∼= H∗(V∗Zk ⊗ ZΛ0).
∼=(cid:26) Ha(Zk, K0(BΛ))
a for 0 ≤ a ≤ k and b even, and d1
0
if 0 ≤ a ≤ k and b even,
otherwise.
E2
a,b
a,b = ∂Λ
a for 1 ≤ a ≤ k and b
/
/
/
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
17
Taking the tensor product of the chain complexV∗ Zk with the groups appearing in (4.5)
and applying the homology functor, we get
H∗(DΓ
H∗(DΛ
∗ ) = H∗(V∗ Zk ⊗ ZΓ0)
∗ ) = H∗(V∗ Zk ⊗ ZΛ0)
H∗(id ⊗K0(ι))
H∗(id ⊗εΓ
∗ )
H∗(id ⊗εΛ
∗ )
H∗(V∗ Zk ⊗ K0(BΓ))
/ H∗(V∗ Zk ⊗ K0(BΛ)).
H∗(id ⊗K0(ϕ×id))
As in the displayed equation at the top of [23, p. 187] we have isomorphisms H∗(V∗ Zk ⊗
K0(BΓ)) → H∗(Zk, K0(BΓ)) and H∗(V∗ Zk ⊗ K0(BΛ)) → H∗(Zk, K0(BΛ)) such that the
following diagram commutes
(4.6)
H∗(DΓ
∗ )
H∗(id ⊗εΓ
∗ )
H∗(id ⊗K0(ι))
H∗(id ⊗εΛ
∗ )
H∗(DΛ
∗ )
H∗(V∗ Zk ⊗ K0(BΓ))
/ H∗(V∗ Zk ⊗ K0(BΛ))
H∗(id ⊗K0(ϕ×id))
/ H∗(Zk, K0(BΓ))
/ H∗(Zk, K0(BΛ)).
The composition of the horizontal maps in the diagram (4.6) is the isomorphism described
by Evans in [10, Theorem 3.14]. Hence H∗(id ⊗εΓ
∗ ) are isomorphisms.
Thus the left-hand vertical map on homology (induced by id ⊗K0(ι)) coincides with the
right-hand map (induced by id ⊗K0(ϕ × id)).
∗ ) and H∗(id ⊗εΛ
Applying the naturality of [23, Proposition 5.4] yields a morphism f of spectral se-
a →
a from Example 4.2 giving (d) -- (e)
(cid:3)
quences compatible with K∗( ϕ) : K∗(BΓ ×αΓ Zk) → K∗(BΛ ×αΛ Zk) such that f 1
DΛ
as required.
a is given by id ⊗K0(ι). The latter map agrees with jH
a,b : DΓ
Recall that a C ∗-algebra A is K0-liftable if for every pair of ideals I ⊆ J in A, the
extension
0
/ I
ϕ
/ J
π
/ J/I
/ 0
has the property that the induced map K0(π) : K0(J) → K0(J/I) is surjective, or equiv-
alently that K1(ϕ) : K1(I) → K1(J) is injective (see [31, Definition 3.1]).
In fact, it
suffices to consider J = A, a fact that the authors of [31] attribute to Larry Brown. So A
is K0-liftable if K1(ϕ) : K1(I) → K1(A) is injective for every ideal I in A.
The next lemma allows us to relate K0-liftability of C ∗(Λ) to the homology of the
complex DΛ described earlier.
Lemma 4.7. Let Λ be a row-finite 2-graph with no sources. Suppose that H ⊆ Λ0 is
saturated and hereditary. Let jH : DΓ
∗ be the map of complexes induced by the
inclusion Γ0 = H ֒→ Λ0, and let ϕ : C ∗(Γ) → C ∗(Λ) be the inclusion that carries
a generator sλ of C ∗(Γ) to the corresponding generator sλ of C ∗(Λ). Then there are
isomorphisms K1(C ∗(Γ)) ∼= H1(DΓ) and K1(C ∗(Λ)) ∼= H1(DΛ) making the following
diagram commute:
∗ → DΛ
K1(C ∗(Γ))
K1(ϕ)
K1(C ∗(Λ))
∼=
∼=
H1(DΓ
∗ )
H1(jH )
/ H1(DΛ
∗ ).
In particular, K1(ϕ) is injective if and only if H1(jH) is injective.
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
18
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
Proof. Define Γ = HΛ. By Lemma 4.6 we have the spectral sequence ( Er, dr) converging
to K0(BΛ ×αΛ Z2). Since Λ is a 2-graph and by the convergence from Lemma 4.6(a) we
have the following filtration of K1 = K1(BΛ ×αΛ Z2):
0 = F0( K1) ⊆ F1( K1) ⊆ F2( K1) = K1,
1,0, F2( K1) ∼= K1 and F2( K1)/F1( K1) = {0} (see [10]). The same applies
with F1( K1) ∼= E2
to the convergent spectral sequence (Er, dr). We let K∗ = K∗(BΓ ×αΓ Z2) denote its limit.
Since f is compatible with K∗( ϕ) : K∗ → K∗ by Lemma 4.6(d), we have the commuting
diagram
(4.7)
K1
E2
1,0
K0( ϕ)
f 2
1,0
K1
/ E2
1,0,
and the horizontal maps are isomorphisms. Using that E2
1,0
E2
1,0
∼= H( E1
∗ ) we obtain the following commuting diagram:
1,0) = H1(DΓ
∼= H(E1
1,0) = H1(DΛ
∗ ) and
α1
1,0
α1
1,0
E2
1,0
f 2
1,0
E2
1,0
/ H1(DΓ
∗ )
H1(jH )
/ H1(DΛ
∗ ).
Applying the K1-functor to the commuting diagram
C ∗(Γ)
/ BΓ ×αΓ Z2
ϕ
ϕ
C ∗(Λ)
/ BΛ ×αΛ Z2
of Lemma 4.5 we get a commuting diagram in which the horizontal maps are invertible.
Combining this and that K1 = K1(BΛ ×αΛ Z2) and K1 = K1(BΓ ×αΓ Z2) with the previous
diagrams we obtain the commuting diagram
K1(C ∗(Γ))
K1(BΓ ×αΓ Z2) = K1
E2
1,0
H1(DΓ
∗ )
K1(ϕ)
K0( ϕ)
f 2
1,0
H1(jH )
K1(C ∗(Λ))
/ K1(BΛ ×αΛ Z2) = K1
/ E2
1,0
/ H1(DΛ
∗ ),
where the horizontal maps are all isomorphisms. The final statement follows immediately.
(cid:3)
Proof of Theorem 4.3. We start by proving (2), so assume that C ∗(Λ) has real-rank zero.
Lemma 4.1 implies that Λ is strongly aperiodic. As alluded to immediately after [31,
Definition 3.1], all C ∗-algebras of real rank zero are K0-liftable: Theorem 2.10 of [4]
shows that if A has real-rank zero, then so does every Mn(A), and then applying [4,
Theorem 3.14] to each Mn(A) shows that if I is an ideal of A then every projection in
Mn(A/I) lifts to a projection in Mn(A). Since A, and hence also A/I, has real-rank zero,
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
19
K0(A/I) is generated by classes of projections over A/I, and it follows that A is K0-
liftable. Hence every inclusion I ⊆ C ∗(Λ) of a proper ideal induces an injective map on
K1-groups. In particular, for every hereditary saturated subset H ⊆ Λ0, the map H1(jH )
is injective by Lemma 4.7.
We now prove (1). Suppose that C ∗(Λ) is purely infinite. The preceding paragraph
establishes (i)⇒(ii). To prove the reverse implication (ii)⇒(i) suppose that Λ is strongly
aperiodic and that H1(jH) is injective for each H. Theorem 3.2 implies that C ∗(Λ) has
topological dimension zero. So [31, Theorem 4.2] implies that C ∗(Λ) has real-rank zero if
and only if it is K0-liftable, and it suffices to establish the latter.
Fix a non-trivial ideal I of C ∗(Λ). We need to show that the inclusion ι : I ֒→ C ∗(Λ)
induces an injective map at the level of K1-groups. Since Λ is strongly aperiodic it follows
from Corollary 3.9 that I = IH for some nonempty hereditary saturated H ( Λ0. Let
Γ := HΛ. As in Lemma 3.7, the C ∗-algebra C ∗(Γ) is isomorphic to a full corner of I, and
so the inclusion map C ∗(Γ) ֒→ I induces an isomorphism in K-theory. We obtain the
following commuting diagram:
K1(C ∗(Γ))
K1(I)
K1(ϕ)
K1(ι)
K1(C ∗(Λ))
id
/ K1(C ∗(Λ)).
So K1(ι) is injective if and only if K1(ϕ) is injective. By condition (ii), the map H1(jH )
is injective, and so Lemma 4.7 shows that K1(ϕ) is injective as required.
(cid:3)
To finish off the section, we show how to reduce checking real-rank zero for 2-graph
algebras with finite ideal lattice to checking real-rank zero for simple 2-graph algebras,
without any assumption of pure infiniteness. We thank an anonymous referee for pointing
out this application of our arguments.
Proposition 4.8. Let Λ be a row-finite 2-graph with no sources, and suppose that C ∗(Λ)
has finite ideal lattice. Then Λ is strongly aperiodic, and the following are equivalent:
(1) C ∗(Λ) has real-rank zero;
(2) Both of the following hold:
(i) for every saturated hereditary H ⊆ Λ0, the map H1(jH) of Example 4.2 is
injective; and
(ii) for every pair K ⊆ H where K is a saturated hereditary subset of Λ0 and H \K
is a saturated hereditary subset of Λ \ ΛK such that HΛ \ ΛK is cofinal, the
C ∗-algebra C ∗(HΛ \ ΛK) has real-rank zero.
Proof. Every C ∗-algebra with finite ideal lattice has topological dimension zero, and so
Theorem 3.2 shows that Λ is strongly aperiodic.
First suppose that C ∗(Λ) has real-rank zero. Then part (2) of Theorem 4.3 gives (2i).
Fix a nested pair K ⊆ H of saturated hereditary sets such that HΛ \ ΛK is cofinal. Then
C ∗(Λ \ ΛH) is a quotient of C ∗(Λ) by [32, Theorem 5.2(b)] and therefore has real-rank
zero. Moreover, C ∗(HΛ \ ΛK) is isomorphic to a full corner of an ideal in C ∗(Λ \ ΛK) by
[32, Theorem 5.2(c)], and therefore itself also has real-rank zero, giving (2ii).
Now suppose that (2) holds. We prove that C ∗(Λ) has real-rank zero by induction on
Prim C ∗(Λ). If Prim C ∗(Λ) is a singleton, then Λ itself is cofinal [21, Proposition 4.8],
so we can take K = ∅ and H = Λ0 in (2) to see that C ∗(Λ) = C ∗(HΛ \ ΛK) has real-rank
/
/
/
20
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
zero by hypothesis. Now fix Λ and suppose that our assertion holds for every strongly
aperiodic k-graph with fewer primitive ideals than C ∗(Λ). Since Prim C ∗(Λ) is finite, we
can fix an ideal I of C ∗(Λ) such that C ∗(Λ)/I is simple. Since Λ is strongly aperiodic,
I is generated by {pv : v ∈ K} for some saturated hereditary K ⊆ Λ0. We then have
C ∗(Λ)/I ∼= C ∗(Λ\ΛK) by [32, Theorem 5.2(b)] again, and since this C ∗-algebra is simple,
Λ \ ΛK is cofinal. So (2ii) with H = Λ0 says that C ∗(Λ)/I has real-rank zero. Lemma 4.7
and (2i) show that C ∗(Λ) is K0-liftable, and it follows from [4, Proposition 3.15] that
if I has real-rank zero, then every projection in C ∗(Λ)/I lifts to a projection in C ∗(Λ).
Therefore, by [4, Theorem 3.14], to prove that C ∗(Λ) has real-rank zero, it now suffices
to prove that I has real-rank zero.
Consider the k-graph Γ := KΛ. By Lemma 3.7, C ∗(Γ) is isomorphic to a full corner of
I, so [4, Theorem 3.8] implies that it suffices to prove that C ∗(Γ) has real-rank zero. Since
C ∗(Λ) is K0-liftable, so is I, and hence also C ∗(Γ) since the inclusion map C ∗(Γ) ֒→ I
induces an isomorphism in K-theory. The k-graph Γ is strongly aperiodic because Λ
is strongly aperiodic. Suppose that K ′ and H ′ are saturated hereditary subsets of Γ0
as in (2ii) for Γ; so H ′ \ K ′ is saturated and hereditary in Γ \ ΓK ′, and H ′Γ \ ΓK ′ is
cofinal. Then H ′ and K ′ also satisfy these hypotheses with respect to Λ (because Γ0 = K
is saturated in Λ). Hence hypothesis (2ii) for C ∗(Λ) implies that C ∗(H ′Γ \ ΓK ′) =
C ∗(H ′Λ \ ΛK ′) has real-rank zero. Thus Γ satisfies (2i) and (2ii). By construction,
C ∗(Γ) = C ∗(Λ)/I has fewer primitive ideals than C ∗(Λ), and so the inductive hypothesis
implies that C ∗(Γ) has real-rank zero as required.
(cid:3)
5. Purely infinite k-graph C ∗-algebras
In this section we turn our attention to pure infiniteness of higher-rank graph C ∗-
algebras, which is one of the assumptions in Theorem 4.3. In [19], Kirchberg and Rørdam
introduced three separate notions of purely infinite C ∗-algebras; weakly purely infinite,
purely infinite and strongly purely infinite. As the names suggest, strong pure infiniteness
implies pure infiniteness which implies weak pure infiniteness. Of these notions, strong
pure infiniteness is perhaps the most useful in the classification of non-simple C ∗-algebras.
Indeed, Kirchberg showed in [17] that two separable, nuclear, stable, strongly purely
infinite C ∗-algebras with the same primitive ideal space X are isomorphic if and only if
they are KKX-equivalent.
Following [18], we denote the set of positive elements in a C ∗-algebra A by A+. The
ideal in A generated by an element b is denoted AbA. Recall that for positive elements
a ∈ Mn(A) and b ∈ Mm(A), we say that a is Cuntz below b, denoted a - b, if there exists
a sequence of elements xk in Mm,n(A) such that x∗
kbxk → a in norm. We say A is purely
infinite if there are no characters on A and for all a, b ∈ A+, we have a - b if and only if
a ∈ AbA (see [18, Definition 4.1]).
Let Λ be a 2-graph and u ∈ Λ0. As introduced in [16], given a, b ∈ Z+, an (a, b)-aperiodic
quartet, or just an aperiodic quartet, at u consists of distinct paths α1, α2 ∈ uΛae1u and
distinct paths β1, β2 ∈ uΛbe2u such that β2α1 = α1β2, β2α2 = α2β2, β1α1 = α2β1, and
β1α2 = α1β1.
Proposition 5.1. Let Λ be a row-finite 2-graph with no sources. Suppose that there is an
aperiodic quartet at u for each u ∈ Λ0. Then C ∗(Λ) is strongly purely infinite.
Before giving a proof of Proposition 5.1 we present a characterisation of when the
C ∗-algebra C ∗(Λ) of a row-finite strongly aperiodic k-graph Λ with no sources is purely
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
21
infinite. Recall that a projection p in a C ∗-algebra A is infinite if it is Murray -- von
Neumann equivalent to a proper subprojection of itself, and properly infinite if there
are two mutually orthogonal subprojections of p in A, each one Murray -- von Neumann
equivalent to p.
Lemma 5.2. Let Λ be a row-finite k-graph with no sources. Suppose that Λ is strongly
aperiodic. Then the following are equivalent
(i) The C ∗-algebra C ∗(Λ) is purely infinite.
(ii) For every v ∈ Λ0 the projection sv is properly infinite in C ∗(Λ).
(iii) For every hereditary saturated H ⊆ Λ0 and every v ∈ Λ0 \ H, the projection sv is
infinite in C ∗(Λ \ ΛH).
(iii') For every hereditary saturated H ⊆ Λ0 and every v ∈ Λ0 \ H, the projection sv is
properly infinite in C ∗(Λ \ ΛH).
(iv) For every hereditary saturated H ⊆ Λ0 and every v ∈ Λ0 \ H, the C ∗-algebra
svC ∗(Λ \ ΛH)sv contains an infinite projection.
(iv') For every hereditary saturated H ⊆ Λ0 and every v ∈ Λ0 \ H, the C ∗-algebra
svC ∗(Λ \ ΛH)sv contains a properly infinite projection.
(v) Every non-zero hereditary sub-C ∗-algebra in any quotient of C ∗(Λ) contains an
infinite projection.
(v') Every non-zero hereditary sub-C ∗-algebra in any quotient of C ∗(Λ) contains a prop-
erly infinite projection.
Proof of Lemma 5.2. We first prove (i)⇔(ii)⇔(iii)⇔(iv)⇔(v).
The implication (i)⇒(ii) is known, see [18, Theorem 4.16]. For (ii)⇒(iii), observe that
[32, Theorem 5.2] shows that C ∗(Λ \ ΛH) is isomorphic to a quotient of C ∗(Λ), and that
the quotient map carries sv to zero if and only if v ∈ H. Corollary 3.15 of [18] says that the
image of a properly infinite projection under a C ∗-homomorphism is either zero or infinite,
giving (ii)⇒(iii). The implication (iii)⇒(iv) is trivial because sv ∈ svC ∗(Λ \ ΛH)sv for
each v ∈ Λ0 \ H.
For (iv)⇒(v), fix a proper ideal I of C ∗(Λ). Let B be a non-zero hereditary subalgebra
of C ∗(Λ)/I. By Corollary 3.9 and [32, Theorem 5.2], there is a saturated hereditary
subset H ⊆ Λ0 such that Γ := Λ \ ΛH satisfies C ∗(Λ)/I ∼= C ∗(Γ). So B is isomorphic to
a hereditary subalgebra of C ∗(Γ). We identify the two henceforth.
λ : λ ∈ Γ}) ∼= C0(G(0)
Select any non-zero positive element a ∈ B. Since Λ is strongly aperiodic, Theorem 3.9
implies that GΓ is topologically principal. Hence [2, Lemma 3.2] provides a non-zero
positive element h ∈ C ∗({sλs∗
Γ ) such that h - a. Recall that the
cylinder sets {Z(λ) : λ ∈ Γ} form a basis for the topology on G(0)
Λ (see [21, Definition
2.4]). Choose λ ∈ Γ such that h(x) ≥ khk/2 > 0 for all x ∈ Z(λ), and let v := s(λ) ∈ Γ0.
Let g ∈ C0(G(0)
Γ ) be the function g(x) := h(x)−1/2 for x ∈ Z(λ) and zero otherwise. Then
ghg = 1Z(λ) = sλs∗
λsλ = sv. In particular, sv - h (take xk := gsλ
for each k). By assumption (iv) there is an infinite projection p ∈ svC ∗(Γ)sv. By [18,
Proposition 2.7] we have p - sv. Hence p - sv - h - a, so p - a. As explained after
[18, Proposition 2.6], there exists x ∈ C ∗(Γ) such that p = x∗ax. Let y := a1/2x. Then
y∗y = x∗ax = p, and q := yy∗ = a1/2xx∗a1/2 ∈ B. Since p is infinite and Murray -- von
Neumann equivalent to q, we conclude that q ∈ B is infinite.
The implication (v)⇒(i) follows from [18, Proposition 4.7].
λ and (gsλ)∗hgsλ = s∗
λsλs∗
22
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
It now suffices to prove (ii)⇒(iii')⇒(iv')⇒(v')⇒(v). Recall that if p is properly infinite
and φ is a C ∗-homomorphism such that φ(p) 6= 0, then φ(p) is properly infinite. This and
the second statement of Lemma 3.7 gives (ii)⇒(iii'). The implication (iii')⇒(iv') is trivial,
and (iv')⇒(v') follows from the argument for (iv)⇒(v) above because any projection p that
is Murray -- von Neumann equivalent to a properly infinite projection is itself is properly
infinite. Since proper infiniteness is stronger than infiniteness, condition (v') implies (v),
completing the proof.
(cid:3)
Corollary 5.3. Let Λ be a row-finite k-graph with no sources and suppose that C ∗(Λ) has
real-rank zero. Then C ∗(Λ) is strongly purely infinite if and only if sv is properly infinite
for every v ∈ Λ0.
Proof. This follows from [31, Corollary 6.9] combined with and Lemmas 4.1 and 5.2. (cid:3)
Following [31], recall that a C ∗-algebra A has the ideal property, abbreviated (IP), if
projections in A separate ideals in A; that is, whenever I, J are distinct ideals in A, there
is a projection in I \ J.
Lemma 5.4. Let Λ be a row-finite k-graph with no sources. Suppose that Λ is strongly
aperiodic. Then C ∗(Λ) has the ideal property.
Proof. Let I, J be ideals of C ∗(Λ) such that I 6⊆ J. Recall that for H ⊆ Λ0, IH denotes
the ideal in C ∗(Λ) generated by {sw : w ∈ H}. By Corollary 3.9, I = IH and J = IK for
some saturated hereditary H, K ⊆ Λ0. Since I 6⊆ J, we have H 6⊆ K, say v ∈ H \ K. [32,
Theorem 5.2(b)] shows that H = {w : sw ∈ IH} and similarly for K, so we deduce that
sv is a projection in IH \ IK = I \ J.
(cid:3)
Remark 5.5. It not clear whether strong aperiodicity of Λ is equivalent to property (IP)
for C ∗(Λ). However, we obtain some easy partial results. First, suppose C ∗(Λ) is AF -- so
automatically has property (IP). By [11, Proposition 3.12], every ideal of C ∗(Λ) is gauge-
invariant, so Corollary 3.9, shows that Λ is strongly aperiodic. Second, suppose that
C ∗(Λ) has (IP) and is purely infinite. By [31, Proposition 2.11], C ∗(Λ) has topological
dimension zero, so Corollary 3.9 shows that Λ is strongly aperiodic.
Proof of Proposition 5.1. Since every vertex of Λ has an aperiodic quartet, [16, Propo-
sition 3.9] shows that Λ is strongly aperiodic. Hence Lemma 5.4 shows that C ∗(Λ) has
property (IP). By [31, Proposition 2.14], a C ∗-algebra with property (IP) is purely infinite
if and only if it is strongly purely infinite. So it suffices to prove that C ∗(Λ) is purely
infinite.
To show that C ∗(Λ) is purely infinite it suffices to verify property (ii) of Lemma 5.2;
that is, it suffices to show that every sv is properly infinite.
Fix any v ∈ Λ0. By assumption there exist a, b ∈ Z+, distinct α1, α2 ∈ vΛae1v and
distinct β1, β2 ∈ vΛbe2v such that β2α1 = α1β2, β2α2 = α2β2, β1α1 = α2β1, and β1α2 =
α1β1. By (CK3) -- (CK4), for i = 1, 2 we have
sv = s∗
αisαi,
sαis∗
Consequently, there exist distinct mutually orthogonal subprojections of sv in C ∗(Λ), each
Murray -- von Neumann equivalent to sv. We conclude that sv is properly infinite.
(cid:3)
αi ≤ Xλ∈vΓae1
sλs∗
λ = sv.
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
23
6. Examples
In this section we present three examples highlighting the necessity of the hypotheses
in Theorem 4.3(1). All three examples are constructed using 2-graphs. Our result says
that the combination of strong aperiodicity of Λ, injectivity of each H1(jH ), and pure
infiniteness of C ∗(Λ) is sufficient to guarantee that C ∗(Λ) has real rank zero. Our three
examples show that no combination of two of these conditions is strong enough.
We describe our examples using the 2-coloured graphs of [12]. A 2-coloured graph is a
directed graph endowed with a map c : E1 → {c1, c2}. We think of c as determining a
colour map from E∗ to the free abelian semigroup F2 generated by {c1, c2} and for w ∈ F2,
we say that λ ∈ E∗ is w-coloured if c(λ) = w. A collection of factorisation rules for E
is a range- and source-preserving bijection θ from the c1c2-coloured paths in E∗ to the
c2c1-coloured paths. For k = 2 the associativity condition of [12] is trivial, and so [12,
Theorems 4.4 and 4.5] say that for every 2-coloured graph (E, c) with a collection θ of
factorisation rules, there is a unique 2-graph Λ with Λei = c−1(ci), Λ0 = E0, and ef = f ′e′
in Λ whenever θ(ef ) = f ′e′ in E∗.
Example 6.1. Strong aperiodicity of Λ (and hence topological dimension zero for C ∗(Λ))
combined with strong pure infiniteness of C ∗(Λ) do not suffice for C ∗(Λ) to have real-rank
zero, even in the special case k = 2. To see this, fix n ≥ 3 and let (E, c) be the following
2-coloured graph:
u •
• v
...
• w
...
n + 1
n + 1
We define factorisation rules as follows. First, for each pair of vertices x, y of E, list the
blue edges in xE1y as {αx,y
xE1y/2}. When
x = y, we write αx
i and βx
. For each vertex x, we define four
i
factorisation rules by
xE1y/2} and the red edges as {βx,y
instead of αx,x
1 , . . . , αx,y
, . . . , βx,y
and βx,x
1
i
i
2 = αx
2βx
2 ,
βx
1 αx
1 = αx
2βx
1 ,
2 = αx
1βx
1 .
1 αx
βx
i βy,z
(6.1)
βx
2 αx
1 = αx
i βy,z
1βx
2 ,
j = βx,y
2 αx
βx
i αy,z
j
for every blue-red path αx,y
We then specify αx,y
j not appearing in the
left-hand side of any of the factorisation rules in (6.1). With these factorisation rules
we obtain a 2-graph Λ, which is row finite with no sources. The factorisation rules (6.1)
yield a (cid:0)(1, 0), (0, 1)(cid:1)-aperiodic quartet at each vertex x of Λ. Proposition 5.1 and [16,
Proposition 3.9] ensures that C ∗(Λ) is strongly purely infinite and Λ is strongly aperiodic
for each n ≥ 1. It is easy to check that H := {w} is a hereditary saturated subset of Λ0.
With the notation of (4.1) we have
M t
i,H = (n + 1),
M t
i,H,Λ0\H = (0 1),
and M t
i,Λ0\H =(cid:0) 2 1
1 2(cid:1)
24
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
for each i. With L2 := {M t
1,Λ0\H − 1)b = 0} and
L1 := {(M t
1,H − 1)a : a ∈ ZH}, we have L1 = nZ and L2 = Z. So Proposition 4.4 implies
that H1(jH) is injective if and only if n = 1. Hence Theorem 4.3(1) implies that C ∗(Λ)
has real-rank zero for n = 1, but not for n > 1.
1,H,Λ0\Hb : b ∈ Z(Λ0 \ H) and (M t
Example 6.2. The injectivity of each H1(jH) combined with strong pure infiniteness of
C ∗(Λ) do not suffice for C ∗(Λ) to have real-rank zero, even in the special case k = 2. To
see this, consider the following 2-coloured graph:
e2
e1
v
f1
f2
Define factorisation rules on (E, c) by eifj = fiej, and let Λ be the resulting 2-graph.
By [21, Corollary 3.5(iii)], we have C ∗(Λ) ∼= O2 ⊗ C(T). As O2 is purely infinite simple,
C ∗(Λ) is strongly purely infinite, see [20, Theorem 1.3]. There are no non-trivial saturated
hereditary subsets of Λ0, so injectivity of each H1(jH) is trivial. The C ∗-algebra C ∗(Λ)
is non-simple, but none of the proper ideals of C ∗(Λ) contain any vertex projections.
Consequently, it follows from Corollary 3.9 that Λ is not strongly aperiodic (alternatively,
for each x ∈ vΛ∞, σ(1,−1)(x) = x). Using Theorem 4.3 we conclude that C ∗(Λ) fails to
have real-rank zero (cf. [31]).
Example 6.3. Strong aperiodicity combined with the injectivity of each H1(jH) does not
suffice for C ∗(Λ) to have real-rank zero, even in the special case k = 2. To see this,
consider the 2-coloured graph:
e1
v1
e2
v2
e3
v3
f3,0
...
f3,31
f2,0
...
f2,7
f1,0
f1,1
e4
v4
. . .
that has 22n−1 edges labelled fn,0, . . . , fn,22n−1−1 from vn+1 to vn for each n ≥ 1. There is
a unique 2-graph for this 2-coloured graph with factorisation rules given by
so factorisation through a blue loop cyclicly permutes the first 2n−1 edges in vnΛe2vn+1,
and fixes the remaining ones.
For this example, in the notation of [26, Theorem 7.2], the An are all nonzero 1 × 1
n is the permutation of vnΛe2vn+1 induced by the factorisation rules.
matrices, and each F 1,1
In particular, the constants β1, as in [26, Definition 7.1], are given by
β1 =
1
22n−1 · {i : 2n−1 < i < 22n−1},
and so each 1 − β1(F 1,1
converges to 1 < ∞. Thus [26, Theorem 7.2(2)] implies that C ∗(Λ) has real-rank 1. The
graph Λ is clearly cofinal. The numbers κ(F 1,1
n ) denoting the maximal orders of elements
n ) is 2n−1/22n−1 = 2−n. So α1 = P∞
n )) = P 2−n
n=1(1 − β1(F 1,1
enfn,i =
fn,i+1en+1
fn,0en+1
fn,ien+1
if 0 ≤ i < 2n−1 − 1
if i = 2n−1 − 1
if 2n−1 < i < 22n−1;
REAL RANK AND TOPOLOGICAL DIMENSION OF HIGHER RANK GRAPH ALGEBRAS
25
of vnΛe2vn+1 under F 1,1
n ) = 2n−1 → ∞. Hence the first statement of [26,
n
Theorem 7.2] says that C ∗(Λ) is simple. It follows that Λ is strongly aperiodic, and each
H1(jH) is injective because H = Λ0 is the only non-empty saturated hereditary subset of
Λ0, in which case H1(jH) is the identity map.
satisfy κ(F 1,1
References
[1] O. Bratteli and G.A. Elliott Structure spaces of approximately finite-dimensional C ∗-algebras. II, J.
Funct. Anal. 30 (1978), 74 -- 82.
[2] J. Brown, L.O. Clark, and A. Sierakowski, Purely infinite C ∗-algebras associated to etale groupoids,
Ergodic Theory Dynam. Systems 35 (2015), 2397 -- 2411.
[3] L.G. Brown, On Higher Real and Stable Ranks for CCR C ∗-algebras, preprint 2007 (preprint available
at arXiv:0708.3072).
[4] L.G. Brown and G.K. Pedersen, C ∗-algebras of real rank zero, J. Funct. Anal. 99 (1991), 131 -- 149.
[5] L.G. Brown and G.K. Pedersen, Limits and C ∗-algebras of low rank or dimension, J. Operator
Theory 61 (2009), 381 -- 417.
[6] T.M. Carlsen, S. Kang, J. Shotwell, and A. Sims, The primitive ideals of the Cuntz -- Krieger algebra
of a row-finite higher-rank graph with no sources, J. Funct. Anal. 266 (2014), 2570 -- 2589.
[7] A. Connes, Noncommutative geometry, Academic Press, Inc., San Diego, CA, 1994. ISBN: 0-12-
185860-X.
[8] J. Dixmier, C ∗-algebras, Translated from the French by Francis Jellett. North-Holland Mathematical
Library, Vol. 15. North-Holland Publishing Co., Amsterdam-New York-Oxford, 1977. xiii+492 pp.
ISBN: 0-7204-0762-1
[9] G.A. Elliott and D.E. Evans, The structure of the irrational rotation C ∗-algebra, Ann. of Math. (2)
138 (1993), 477 -- 501.
[10] D.G. Evans, On the K-theory of higher rank graph C ∗-algebras, New York J. Math. 14 (2008), 1 -- 31.
[11] D.G. Evans and A. Sims, When is the Cuntz -- Krieger algebra of a higher-rank graph approximately
finite-dimensional?, J. Funct. Anal. 263 (2012), 183 -- 215.
[12] R. Hazlewood, I. Raeburn, A. Sims, and S.B.G. Webster, Remarks on some fundamental results
about higher-rank graphs and their C ∗-algebras, Proc. Edinb. Math. Soc. (2) 14 (2013), 575 -- 597.
[13] J.H. Hong and W. Szymanski, Purely infinite Cuntz -- Krieger algebras of directed graphs, Bull. Lond.
Math. Soc. (5) 35 (2003), 689 -- 696.
[14] J.A. Jeong, G.H. Park, and D.Y. Shin Stable rank and real rank of graph C ∗-algebras, Pacific J.
Math. 200 (2001), 331 -- 343.
[15] J.A. Jeong and G.H. Park. Graph C ∗-algebras with real rank zero, J. Funct. Anal. 188 (2002),
216 -- 226.
[16] S. Kang and D. Pask, Aperiodic and primitive ideals of row-finite k-graphs, Int. J. Math. 25 (2014),
1450022 [25 pages].
[17] E. Kirchberg, Das nicht-kommutative Michael-Auswahlprinzip und die Klassifikation nicht-einfacher
Algebren, C ∗-algebras (Münster, 1999), Springer, Berlin, 2000, pp. 92 -- 141.
[18] E. Kirchberg and M. Rørdam, Non-simple purely infinite C ∗-algebras, Amer. J. Math. 122 (2000),
637 -- 666.
[19] E. Kirchberg and M. Rørdam, Infinite non-simple C ∗-algebras: absorbing the Cuntz algebras O∞,
Adv. Math. 167 (2002), 195 -- 264.
[20] E. Kirchberg and A. Sierakowski, Filling families and strong pure infiniteness, preprint, 2014.
[21] A. Kumjian and D. Pask, Higher rank graph C ∗-algebras, New York J. Math. 6 (2000), 1 -- 20.
[22] A. Kumjian, D. Pask, I. Raeburn, and J. Renault, Graphs, groupoids, and Cuntz -- Krieger algebras,
J. Funct. Anal. 144 (1997), 505 -- 541.
[23] A. Kumjian, D. Pask, and A. Sims, C ∗-algebras associated to coverings of k-graphs, Doc. Math. 13
(2008), 161 -- 205.
[24] A. Kumjian, D. Pask, and A. Sims, On the K-theory of twisted higher-rank-graph C ∗-algebras, J.
Math. Anal. Appl., 401 (2013), 104 -- 113.
[25] D. Pask, J. Quigg, and I. Raeburn, Coverings of k-graphs, J. Algebra 289 (2005), 161 -- 191.
26
DAVID PASK, ADAM SIERAKOWSKI, AND AIDAN SIMS
[26] D. Pask, I. Raeburn, M. Rørdam and A. Sims Rank-two graphs whose C ∗-algebras are direct limits
of circle algebras, J. Funct. Anal. 239 (2006), 137 -- 178.
[27] C. Pasnicu, Extensions of AH algebras with the ideal property, Proc. Edinburgh Math. Soc. (2) 42
(1999), 65 -- 76.
[28] C. Pasnicu, The ideal property, the projection property, continuous fields and crossed products, J.
Math. Anal. Appl. 323 (2006), 1213 -- 1224.
[29] C. Pasnicu, Real rank zero and continuous fields of C ∗-algebras, Bull. Math. Soc. Sci. Math.
Roumanie (N.S.) 48 (2005), 319 -- 325.
[30] C. Pasnicu, and N.C. Phillips Permanence properties for crossed products and fixed point algebras of
finite groups., Trans. Amer. Math. Soc. 366 (2014), 4625 -- 4648.
[31] C. Pasnicu and M. Rørdam, Purely infinite C ∗-algebras of real rank zero, J. reine angew. Math. 613
(2007), 51 -- 73.
[32] I. Raeburn, A. Sims, and T. Yeend, Higher-rank graphs and their C ∗-algebras, Proc. Edinb. Math.
Soc. (2) 46 (2003), 99 -- 115.
[33] J. Renault, A groupoid approach to C ∗-algebras, Lecture Notes in Mathematics, 793. Springer,
Berlin, 1980. ISBN: 3-540-09977-8.
[34] J. Renault, Examples of masas in C ∗-algebras, Operator structures and dynamical systems, 259 -- 265,
Contemp. Math., 503, Amer. Math. Soc., Providence, RI, 2009.
[35] L. Robert and A. Tikuisis, Nuclear dimension and Z-stability of non-simple C ∗-algebras, Trans.
Amer. Math. Soc., to appear (preprint available at arXiv:1308.2941).
[36] D. Robertson and A. Sims, Simplicity of C ∗-algebras associated to higher-rank graphs., Bull. Lond.
Math. Soc. (2) 39 (2007), 337 -- 344.
[37] A. Sims, Gauge-invariant ideals in the C ∗-algebras of finitely aligned higher-rank graphs, Canad. J.
Math. 58 (2006), 1268 -- 1290.
[38] H. Thiel The topological dimension of type I C ∗-algebras, Operator algebra and dynamics, 305 -- 328,
Springer Proc. Math. Stat., 58, Springer, Heidelberg, 2013.
[39] D.P. Williams, Crossed Products of C ∗-algebra, Mathematical Surveys and Monographs, 134. Amer.
Math. Soc., Providence, RI, 2007. ISBN: 978-0-8218-4242-3.
[40] S. Zhang, Certain C ∗-algebras with real rank zero and their corona and multiplier algebras, Part I,
Pacific J. Math. 155 (1992), 169 -- 197.
E-mail address: dpask/asierako/[email protected]
(D. Pask, A. Sierakowski and A. Sims) School of Mathematics and Applied Statistics, Uni-
versity of Wollongong, Wollongong NSW 2522, AUSTRALIA
|
1309.4142 | 2 | 1309 | 2019-03-12T22:29:20 | The K-Theory of a Simple Separable Exact C*-Algebra Not Isomorphic to Its Opposite Algebra | [
"math.OA"
] | We construct uncountably many mutually nonisomorphic simple separable stably finite unital exact C$^\ast$-algebras which are not isomorphic to their opposite algebras. In particular, we prove that there are uncountably many possibilities for the $K_0$-group, the $K_1$-group, and the tracial state space of such an algebra. We show that these C*-algebras satisfy the Universal Coefficient Theorem. This is new even for the already known example of an exact C*-algebra nonisomorphic to its opposite algebra produced in earlier work. | math.OA | math |
THE K-THEORY OF A SIMPLE SEPARABLE EXACT
C*-ALGEBRA NOT ISOMORPHIC TO ITS OPPOSITE
ALGEBRA
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
Abstract. We construct uncountably many mutually nonisomorphic
simple separable stably finite unital exact C*-algebras which are not
isomorphic to their opposite algebras. In particular, we prove that there
are uncountably many possibilities for the K0-group, the K1-group, and
the tracial state space of such an algebra. We show that these C*-
algebras satisfy the Universal Coefficient Theorem. This is new even for
the already known example of an exact C*-algebra nonisomorphic to its
opposite algebra produced in earlier work.
1. Introduction
In [29] we constructed an example of a simple separable exact C*-algebra A
not isomorphic to its opposite algebra. The algebra A has a number of nice
properties: it is stably finite, approximately divisible, and it has real rank
zero, stable rank one, and a unique tracial state. The order on projections
over A is determined by traces, and A tensorially absorbs the Jiang-Su al-
3(cid:3) and K1(A) = 0. Its Cuntz
gebra Z. Its K-theory is given by K0(A) ∼= Z(cid:2) 1
semigroup is W (A) ∼= Z(cid:2) 1
3(cid:3)+ ∐ (0, ∞).
The purpose of this article is to exhibit many examples of simple separable
exact C*-algebras not isomorphic to their opposite algebras, and to prove
that they satisfy the Universal Coefficient Theorem. (This is new even for
the example in [29].) In particular, we prove that there are uncountably
many possibilities for the K-theory of such an algebra, while still preserving
most of the good properties of the algebra in [29]. For p = 2 and for
any odd prime p such that −1 is not a square mod p, and for any UHF
algebra B stable under tensoring with the p∞ UHF algebra (the algebra
∞On=1
Mp), we give a simple separable exact C*-algebra D, not isomorphic
to its opposite algebra, with real rank zero and a unique tracial state such
that K∗(D) ∼= K∗(B). For any p and B as above, and for any Choquet
simplex ∆, we give a simple separable exact C*-algebra D, not isomorphic
to its opposite algebra, with real rank one, such that K∗(D) ∼= K∗(B),
and whose tracial state space is isomorphic to ∆. For any p and B as
Date: 2 December 2015.
2000 Mathematics Subject Classification. 46L35, 46L37, 46L40, 46L54.
N. Christopher Phillips was partially supported by NSF Grants DMS-0701076 and
DMS-1101742. Maria Grazia Viola was partially supported by a Research Development
Fund from Lakehead University and an NSERC Discovery Grant.
1
2
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
above, and for any countable abelian group G, we give a simple separable
exact C*-algebra D, not isomorphic to its opposite algebra, with real rank
zero and a unique tracial state such that K0(D) ∼= K0(B) and such that
K1(D) ∼= G ⊗Z Z(cid:2) 1
p(cid:3). We show that all the C*-algebras we construct satisfy
the Universal Coefficient Theorem. We give further information on the
algebras described above, including showing that the order on projections is
determined by traces, computing the Cuntz semigroups, and showing that
the algebras have stable rank one and tensorially absorb the p∞ UHF algebra
and the Jiang-Su algebra. The examples described above are not the most
general that can be obtained with our method, but are chosen to illustrate
the possibilities. There are infinitely many primes p such that −1 is not
a square mod p, so there are infinitely many choices for p covered by our
examples.
Question 8.1 in [29] asked whether for any UHF algebra B there exists a
simple separable exact C*-algebra D not isomorphic to its opposite algebra
that has the same K-theory as B, and the same properties as the algebra
of [29]. Our results provide a partial positive answer to this question.
The paper is organized as follows. Section 2 contains various prelimi-
naries. In particular, we recall some relevant definitions and constructions
involving von Neumann algebras.
In Section 3 we recall the definition of
the continuous Rokhlin property for an action of a finite group G on a sep-
arable unital C*-algebra. The model action of G on the UHF algebra of
type card(G)∞ is an example of an action with this property. Our main
result is that if A is a separable, unital C*-algebra satisfying the Univer-
sal Coefficient Theorem, and α is an action of a finite abelian group G on
A with the continuous Rokhlin property, then the fixed point algebra Aα
and the crossed product C ∗(G, A, α) satisfy the Universal Coefficient Theo-
rem. In Section 4 we construct our basic example, one algebra D for each
prime p such that p = 2 or −1 is not a square mod p, satisfying the same
properties as the algebra in [29]. Moreover, we show that D satisfies the
Universal Coefficient Theorem. Section 5 contains the main step towards
the proof that these algebras are not isomorphic to their opposites. Each
of them has a unique tracial state. We prove that the weak closure of D in
the Gelfand-Naimark-Segal representation associated with this tracial state
is not isomorphic to its opposite algebra. In Section 6, we tensor these basic
examples with other simple separable nuclear unital C*-algebras. The main
result of Section 5 also applies to such tensor products, and we thus obtain
the examples described above. Section 7 contains some open problems.
The authors would like to thank E. Gardella for useful discussions about
the Universal Coefficient Theorem and for suggesting the argument used
to show that the algebra D in Section 4 satisfies the Universal Coefficient
Theorem.
2. Preliminaries
In this section, we provide some background material about opposite al-
gebras, automorphisms of II1 factors, the Connes invariant, and the Cuntz
semigroup.
K-THEORY
3
First we recall the definition of the opposite algebra and the conjugate
algebra of a C*-algebra A.
Definition 2.1. Let A be a C*-algebra. The opposite algebra Aop is the C*-
algebra which has the same vector space structure, norm, and adjoint as A,
while the product of x and y in Aop, which we denote by x⋆y when necessary,
is given by x ⋆ y = yx. If ω : A → C is a linear functional, then we let ωop
denote the same map but regarded as a linear functional ωop : Aop → C.
The conjugate algebra Ac is the C*-algebra whose underlying vector space
structure is the conjugate of A, that is, the product of λ ∈ C and x ∈ Ac
is equal to λx (as evaluated in A), and whose ring structure, adjoint, and
norm are the same as for A.
Remark 2.2. The map x 7→ x∗ is an isomorphism from Ac to Aop.
Notation 2.3. Let A be a C*-algebra, and let ω be a state on A. We
denote the triple consisting of the Gelfand-Naimark-Segal representation,
its Hilbert space, and its standard cyclic vector by (πω, Hω, ξω).
Also, for any C*-algebra or von Neumann algebra A and any tracial state
τ on A, we denote the usual L2-norm by kxk2,τ = (τ (x∗x))1/2 for x ∈ A.
When no confusion can arise about the tracial state used, we write kxk2.
It seems useful to make explicit the following fact, which has been used
implicitly in previous papers.
Lemma 2.4. Let A be a C*-algebra, and let τ be a tracial state on A.
Then τ op is a tracial state on Aop and, as von Neumann algebras, we have
πτ op(Aop)′′ ∼= [πτ (A)′′]op.
Proof. The functional τ op is a state because A and Aop have the same norm
and positive elements. It is immediate that τ op is tracial.
Next, we claim that kxk2,τ op = kxk2,τ for all x ∈ A. Indeed, using the
trace property at the third step,
(kxk2,τ op )2 = τ op(x∗ ⋆ x) = τ (xx∗) = τ (x∗x) = (kxk2,τ )2.
We can identify πτ (A)′′ with the set of elements in the Hausdorff comple-
tion of A in k·k2,τ which are limits in k·k2,τ of norm bounded sequences in A,
and similarly with πτ op(Aop)′′. It follows from the claim that the identity
map of A extends to a linear isomorphism πτ (A)′′ → πτ op(Aop)′′, which is
easily seen to preserve adjoints and reverse multiplication.
(cid:3)
To prove that our C*-algebras are not isomorphic to their opposite alge-
bras, we will need some terminology and results for the automorphisms of a
II1 factor.
Definition 2.5. For any von Neumann algebra M , we denote by Inn(M )
the group of inner automorphisms of M , that is, the automorphisms of the
form Ad(u) for some unitary u ∈ M . Let M be a II1 factor with separable
predual. Denote by τ the unique tracial state on M . An automorphism ϕ
of M is approximately inner if there exists a sequence of unitaries (un)n∈Z>0
in M such that Ad(un) → ϕ pointwise in k · k2. Denote by Inn(M) the group
of approximately inner automorphisms of M .
4
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
Another important class of automorphisms consists of the centrally trivial
automorphisms of M .
Definition 2.6. Let M be a II1 factor with separable predual. Let τ be the
unique tracial state on M . Recall that a bounded sequence (xn)n∈Z>0 in M
kxna − axnk2 = 0 for all a ∈ M . An automorphism ϕ of
is central if
kϕ(xn) − xnk2 = 0 for every central
M is said to be centrally trivial if
sequence (xn)n∈Z>0 in M . Let Ct(M ) denote the set of all centrally trivial
automorphisms of M .
lim
n→∞
lim
n→∞
By the comments following Definition 3.1 in [7], the set Ct(M ) is a normal
subgroup of Aut(M ). It is obviously closed.
We recall from [6] the definition of Connes invariant χ(M ) of a II1 fac-
tor M .
In [6] Connes used centralizing sequences to define the centrally
trivial automorphisms. For ϕ ∈ M∗ and x ∈ M , we define [ϕ, x] ∈ M∗ by
[ϕ, x](y) = ϕ(xy − yx) for y ∈ M . A sequence (xn)n∈Z>0 is then said to
be centralizing if
k[ϕ, xn]k = 0 for all ϕ ∈ M∗. In general, centralizing
sequences are the right ones to use to define the Connes invariant.
lim
n→∞
A bounded sequence (xn)n∈Z>0 in a II1 factor M is central if and only
if for some (equivalently, for any) strong operator dense subset S ⊆ M , we
have [xn, y] → 0 in the strong operator topology for all y ∈ S. In a II1 fac-
tor, we claim that the central sequences are the same as the centralizing
sequences. The implication from (β) to (γ) in Proposition 2.8 of [5] shows
that centralizing sequences in M are central. For the reverse, in Proposition
2.8(α) of [5], we take ϕ to be the tracial state τ . Since [τ, y] = 0 for all
y ∈ M , the implication from (α) to (γ) there shows that central sequences
in M are centralizing.
Definition 2.7. Let M be a II1 factor with separable predual. The Connes
invariant χ(M ) is the subgroup of the outer automorphism group Out(M ) =
Aut(M )/Inn(M ) obtained as the center of the image under the quotient map
of the group of approximately inner automorphisms.
Remark 2.8. If M is isomorphic to its tensor product with the hyperfinite
II1 factor, then χ(M ) is the image in Aut(M )/Inn(M ) of Ct(M ) ∩ Inn(M).
See [6].
In general it is not easy to compute the Connes invariant of a II1 factor.
For the hyperfinite II1 factor R, every centrally trivial automorphism is inner
by Theorem 3.2(1) in [8], so χ(R) = {0}. Moreover, any approximately inner
automorphism of the free group factor on n generators L(Fn) is inner, so
χ(L(Fn)) = {0}. (See [20], or Lemma 3.2 in [40].)
A useful tool to compute the Connes invariant of some II1 factors is the
short exact sequence introduced in [6]. Assume that N is a II1 factor without
nontrivial hypercentral sequences, that is, central sequences that asymp-
totically commute in the L2-norm with every central sequence of N . Let
G be a finite subgroup of Aut(N ) such that G ∩ Inn(N ) = {1}. Define
K = G ∩ Ct(N ) and let K ⊥ be its annihilator, that is
K ⊥ =(cid:8)f : G → S1 : f is a homomorphism and f K = 1(cid:9) ⊂ bG.
K-THEORY
5
Let ξ : Aut(N ) → Out(N ) denote the standard quotient map and let H ⊆
Aut(N ) be the subgroup
H =(cid:8)Ad(u) : u ∈ N is unitary and ρ(u) = u for all ρ ∈ G(cid:9).
Let G ∨ Ct(N ) be the subgroup of Aut(N ) generated by G ∪ Ct(N ). (It is
closed since G is finite and Ct(N ) is closed and normal.) Taking the closure
in the topology of pointwise L2-norm convergence, let
L = ξ(cid:0)(G ∨ Ct(N )) ∩ H(cid:1) ⊆ Out(N ).
Then the Connes short exact sequence (Theorem 4 of [6]) is given by
(2.1)
{1} −→ K ⊥ ∂
−→ χ(N ⋊ G)
Π
−→ L −→ {1}.
See Section 5 in [21] for the definition of the maps ∂ and Π in (2.1).
The obstruction to lifting, defined by Connes in Section 1 of [8], will play
a key role in showing that our algebras are not isomorphic to their opposites.
Definition 2.9. Let M be a II1 factor and let α be an automorphism of M .
Let n be the smallest nonnegative integer such that there is a unitary u ∈ M
with αn = Ad(u). If no power of α is inner, we set n = 0. Since M is a factor,
it is easy to check that there is λ ∈ C such that λn = 1 and α(u) = λu.
(Simply apply αn+1 = α ◦ αn = αn ◦ α to any element x in M .) We call λ
the obstruction to lifting of α.
Next, we recall what it means for the order on projections to be deter-
mined by traces. Let A be a C*-algebra and denote by Mn(A) the n × n
matrices with entries in A. Let M∞(A) denote the algebraic direct limit of
the sequence (Mn(A), ϕn)n∈Z>0, in which ϕn : Mn(A) → Mn+1(A) is defined
by a 7→ ( a 0
0 0 ). Denote by T (A) the set of tracial states of A.
Definition 2.10. We say that the order on projections over A is determined
by traces if whenever p, q ∈ M∞(A) are projections such that τ (p) < τ (q)
for every τ in T (A), then p - q.
We conclude this section by recalling the definitions of Cuntz subequiva-
lence and the Cuntz semigroup. See Section 2 of [4] and the references there
for the definitions below and the proofs of the assertions made here.
Definition 2.11. Let A be a C*-algebra and let a, b ∈ M∞(A)+. We say
that a is Cuntz subequivalent to b, denoted a - b, if there exists a sequence
(vn)n∈Z>0 in M∞(A) such that lim
n − ak = 0. If a - b and b - a we
n→∞
say that a is Cuntz equivalent to b and write a ∼ b.
kvnbv∗
Cuntz equivalence is an equivalence relation, and we write hai for the
equivalence class of a.
Definition 2.12. Let A be a C*-algebra. The Cuntz semigroup of A is
W (A) = M∞(A)+/∼. We define a semigroup operation on W (A) by
hai + hbi =(cid:28)(cid:18)a 0
0 b(cid:19)(cid:29)
and a partial order by hai ≤ hbi if and only if a - b. With this structure
W (A) becomes a positively ordered abelian semigroup with identity.
6
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
Usually, it is hard to compute the Cuntz semigroup of a C*-algebra, but
the following remark computes W (A) for the C*-algebras A of interest here.
Remark 2.13. Denote the Jiang-Su algebra by Z. Assume that A is a
simple unital exact stably finite C*-algebra which is Z-stable. Let V (A)
be the Murray-von Neumann semigroup of A. For any compact convex
set ∆, let LAff b(∆)++ denote the set of bounded strictly positive lower
semicontinuous affine functions on ∆. By Corollary 5.7 of [4], we have
W (A) ∼= V (A) ∐ LAff b(T (A))++.
The addition and order on the disjoint union are defined as follows. On each
part of the disjoint union, the addition and order are the usual ones. For
τ ∈ T (A). Now let x ∈ V (A) and y ∈ LAff b(T (A))++. Then x + y is the
the other cases, for x ∈ V (A) define bx : T (A) → [0, ∞) by bx(τ ) = τ (x) for
functionbx + y ∈ LAff b(T (A))++. Also, x ≤ y if and only if bx(τ ) < y(τ ) for
all τ ∈ T (A), and y ≤ x if and only if y(τ ) ≤bx(τ ) for all τ ∈ T (A).
3. The continuous Rokhlin property and the Universal
Coefficient Theorem
We recall the definition of asymptotic homomorphism, and what means
for an action of a finite group on a C*-algebra to have the continuous Rokhlin
property. The main result of this section is that if A is a separable unital
C*-algebra satisfying the Universal Coefficient Theorem, and α is an action
of a finite group G on A satisfying the continuous Rokhlin property, then
the fixed point algebra Aα and the crossed product C ∗(G, A, α) satisfy the
Universal Coefficient Theorem.
Definition 3.1. Let A be a separable, unital C*-algebra, and let α : G →
Aut(A) denote an action of a finite group G on A. We say that α has the
continuous Rokhlin property if there exist continuous functions t → e(t)
from
g
[0, ∞) to A, for g ∈ G, such that:
g (cid:1)g∈G is a family of mutually orthogonal
e(t)
g = 1.
(1) For each t ∈ [0, ∞), (cid:0)e(t)
projections such that Xg∈G
t→∞(cid:13)(cid:13)αg(cid:0)e(t)
h (cid:1) − e(t)
(2) lim
gh(cid:13)(cid:13) = 0 for every g, h ∈ G.
(3) For any given g ∈ G and a ∈ A, we have lim
g a − ae(t)
t→∞(cid:13)(cid:13)e(t)
g (cid:13)(cid:13) = 0.
Lemma 3.2. Let A and B be separable unital C*-algebras, let G be a finite
group, and let α : G → Aut(A) and β : G → Aut(B) be actions of G on A
and B. Assume that α has the continuous Rokhlin property. Let A ⊗ B be
any C∗ tensor product on which the tensor product action g 7→ αg ⊗ βg is
defined. Then α ⊗ β has the continuous Rokhlin property.
Proof. Let (cid:0)e(t)
tion 3.1 for the action α. Then(cid:0)e(t)
g (cid:1)g∈G, t∈[0,∞) be a family of projections in A as in Defini-
g ⊗ 1(cid:1)g∈G, t∈[0,∞) is a family of projections
in A ⊗ B as in Definition 3.1 for the action α ⊗ β.
(cid:3)
K-THEORY
7
The following example of an action of a finite group with the continuous
Rokhlin property will be needed in Proposition 3.9.
Example 3.3. Let G be a topological group, and let d1, d2, . . . ∈ Z>0. Let
ρ =(cid:0)ρ(1), ρ(2), . . .(cid:1) be a sequence of unitary representations ρ(k) : G → L(Cdk )
of G. Define actions ν(k) : G → Aut(L(Cdk )) by ν(k)
for k ∈ Z>0, g ∈ G, and a ∈ L(Cdk ). Let Bρ be the UHF algebra
g (a) = ρ(k)(g)aρ(k)(g)∗
and let µρ : G → Aut(Bρ) be the product type action given by
Bρ =
L(Cdk ),
∞Ok=1
∞Ok=1
µρ
g =
ν(k)
g
.
∞Ok=1
∞Ok=1
∞Ok=1
∞Ok=1
For a fixed unitary representation ρ : G → L(Cd), we abbreviate (ρ, ρ, . . .)
to ρ, so that Bρ is the d∞ UHF algebra, and the action is given by
g 7→ µρ
g =
Ad(ρ(g)) ∈ Aut(Bρ).
∞Ok=1
When G is finite with card(G) = d, and ρ is the regular representation
λ : G → L(l2(G)), we write
BG =
L(l2(G))
and g ∈ G → µG
g =
Ad(λ(g)) ∈ Aut(BG)
and when ρ is the direct sum λm of m copies of λ, we write
BG,m =
L(l2(G)m) and g ∈ G → µG,m
g =
Ad(λm(g)) ∈ Aut(BG,m).
These are product type actions of G on the d∞ and (md)∞ UHF algebras.
Lastly, we fix some notation. For any index set S and s ∈ S, we denote
by δs ∈ l2(S) the standard basis vector, determined by
δs(t) =(cid:26) 1
0
t = s
t 6= s.
Lemma 3.4. Let G be a finite group. Then the action µG,m : G → Aut(BG,m)
of Example 3.3 has the continuous Rokhlin property.
Proof. We use the notation above. Recall that λm is the direct sum of m
copies of the regular representation of G, and define ν : G → Aut(L(l2(G)m))
by νg = Ad(λm(g)).
We begin with a construction involving just two tensor factors. Let
v : l2(G)m ⊗ l2(G)m → l2(G)m ⊗ l2(G)m
be the unitary determined by v(ξ ⊗ η) = η ⊗ ξ for ξ, η ∈ l2(G)m. Equip
L(cid:0)l2(G)m ⊗ l2(G)m(cid:1) = L(l2(G)m) ⊗ L(l2(G)m)
8
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
is finite dimensional, there is a continuous path t 7→ zt of G-invariant uni-
with the action g 7→ νg⊗νg. Then v is G-invariant. Since L(cid:0)l2(G)m ⊗ l2(G)m(cid:1)G
taries in L(cid:0)l2(G)m ⊗ l2(G)m(cid:1) such that z0 = 1 and z1 = v.
For every g ∈ G, let δg be the corresponding standard basis vector in
l2(G), and let pg ∈ L(l2(G)m) be the projection on the m dimensional
subspace spanned by the standard basis vectors δg,j = (0, . . . , 0, δg, 0, . . . , 0)
for 1 ≤ j ≤ m, where δg is in the j-th position. Then νg(ph) = pgh for every
pg = 1. For n ∈ Z≥0, t ∈ [n, n + 1], and g ∈ G, we define
g, h ∈ G, andXg∈G
e(t)
g = 1 ⊗ 1 ⊗ · · · ⊗ 1 ⊗ zt−n(pg ⊗ 1)z∗
t−n ⊗ 1 ⊗ 1 ⊗ 1 ⊗ · · · ∈ BG,m,
with the expression zt−n(pg ⊗ 1)z∗
n + 2 in the tensor product. It is clear that e(t)
g
t−n occupying the two positions n + 1 and
is a projection and that
e(t)
g = 1 for all t ∈ [0, ∞). Since (νg ⊗ νg)(zt−n) = zt−n and νg(ph) = pgh,
one easily checks that µG,m
Finally, if
g
gh for all g, h ∈ G and t ∈ [0, ∞).
Xg∈G
h (cid:1) = e(t)
(cid:0)e(t)
∞Ok=1
a ∈
NOk=1
L(l2(G)m) ⊂
L(l2(G)m)
and t ≥ N,
then e(t)
g
exactly commutes with a.
Now take b ∈ BG,m. For every ε > 0 there exist N ≥ 1 and a ∈
L(l2(G)m)
g a = ae(t)
NOk=1
g (cid:13)(cid:13) = 2kb − ak < ε.
2 . Now suppose t ≥ N . Then e(t)
g by the pre-
such that kb − ak < ε
vious paragraph, so
g b − be(t)
(cid:13)(cid:13)e(t)
g (cid:13)(cid:13) ≤ 2kb − ak +(cid:13)(cid:13)e(t)
g a − ae(t)
This completes the proof.
(cid:3)
The following result will not be used, but it is easy to derive from known
results and provides motivation for the idea that the action we construct in
Section 4 should have the continuous Rokhlin property. The result we actu-
ally need is in Proposition 3.9 below. It is known that there are actions on
simple C*-algebras which have the Rokhlin property but not the continuous
Rokhlin property. Giving an example here would take us too far afield.
Proposition 3.5. Let G be a finite group. Let A be a simple separable uni-
tal nuclear C*-algebra satisfying the Universal Coefficient Theorem which,
in addition, is either purely infinite or tracially AF in the sense of [27]. Let
α : G → Aut(A) be an action with the Rokhlin property. Assume that for
all g ∈ G, the maps (αg)∗ ∈ Aut(K∗(A)) are the identity maps. Then α has
the continuous Rokhlin property.
Proof. Apply Theorem 3.4 of [18] in the purely infinite case, and Theorem 3.5
of [18] in the tracially AF (tracial rank zero) case, to show that α is conjugate
to its tensor product with the action µG of Example 3.3. Since µG has the
K-THEORY
9
continuous Rokhlin property (by Lemma 3.4), it follows from Lemma 3.2
that α has the continuous Rokhlin property.
(cid:3)
We now recall the definition of an asymptotic homomorphism.
Definition 3.6. Let A and B be C*-algebras. An asymptotic homomor-
phism from A to B is a family of maps ψt : A → B, indexed by t ∈ [0, ∞),
satisfying the following conditions:
(1) For all a ∈ A the map t 7→ ψt(a), from [0, ∞) to B, is continuous.
(2) For all a, b ∈ A and λ ∈ C one has
lim
t→∞
kψt(a + b) − ψt(a) − ψt(b)k = 0,
lim
t→∞
kψt(λa) − λψt(a)k = 0,
lim
t→∞
kψt(ab) − ψt(a)ψt(b)k = 0,
lim
t→∞
kψt(a∗) − ψt(a)∗k = 0.
and
Next we show that given a separable unital C*-algebra and an action of
a finite group G on A with the continuous Rokhlin property, there exists
a unital completely positive asymptotic homomorphism t 7→ ψt from A to
Aα which is a left inverse for the inclusion. The following argument was
suggested by E. Gardella. It replaces an earlier argument in which ψt was
not completely positive.
Proposition 3.7. Let A be a separable unital C*-algebra. Let G be a fi-
nite group and let α : G → Aut(A) be an action with the continuous Rokhlin
property. Denote by Aα the fixed point algebra, and let ι : Aα → A be the
canonical inclusion. Then there exists a unital completely positive asymp-
totic homomorphism t 7→ ψt : A → Aα for t ∈ [0, ∞) such that
for all a ∈ Aα.
lim
t→∞
k(ψt ◦ ι)(a) − ak = 0
Proof. Given C*-algebras A and B and a map ψ : A → B, we denote by
ψ(n) the map from Mn(A) to Mn(B) defined by ψ(n)(a) = (ψ(aj,k))n
j,k=1 for
a = (aj,k)n
j,k=1 ∈ Mn(A).
g (cid:1)g∈G, t∈[0,∞) be a family of projections as in Definition 3.1. For
t ∈ [0, ∞) define a map ρt : A → A by
Let (cid:0)e(t)
ρt(a) =Xg∈G
e(t)
g αg(a)e(t)
g
for a ∈ A. We claim that t 7→ ρt is a unital completely positive asymptotic
homomorphism from A to A such that
for all a ∈ Aα.
lim
t→∞
k(ρt ◦ ι)(a) − ak = 0
10
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
Obviously ρt is unital. Moreover, ρt is completely positive since if a =
(aj,k)n
j,k=1 ∈ Mn(A)+ and
g , . . . , e(t)
denotes the diagonal matrix with the element e(t)
nal, then
g , e(t)
p(t)
g = diag(cid:0)e(t)
(a) =Xg∈G
p(t)
g (cid:0)αg(aj,k)(cid:1)n
ρ(n)
t
g (cid:1)
j,k=1p(t)
g ≥ 0
g everywhere on the diago-
for every t ∈ [0, ∞). To show that ρ = (ρt)t∈[0,∞) is an asymptotic homo-
morphism, observe that t 7→ ρt(a) is clearly continuous for every a ∈ A, and
that for every a, b ∈ A and λ ∈ C we have ρt(λa + b) = λρt(a) + ρt(b) and
ρt(a∗) = ρt(a)∗. Moreover,
(3.1)
lim
t→∞
kρt(ab) − ρt(a)ρt(b)k
g (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
e(t)
g αg(a)e(t)
g αg(b)e(t)
g − e(t)
= lim
≤ lim
e(t)
g αg(ab)e(t)
t→∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Xg∈G
g −Xg∈G
t→∞Xg∈G
kαg(a)k(cid:13)(cid:13)αg(b)e(t)
t→∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Xg∈G
t→∞Xg∈G(cid:13)(cid:13)ae(t)
≤ lim
e(t)
g ae(t)
g αg(b)(cid:13)(cid:13) = 0.
g a(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
g −Xg∈G
g a(cid:13)(cid:13) = 0.
e(t)
g − e(t)
Lastly, for every a ∈ Aα we have
(3.2)
lim
t→∞
kρt(a) − ak = lim
The claim is proved.
It follows from the definition of ρt and the relation (2) in Definition 3.1
that
(3.3)
lim
t→∞
kαg(ρt(a)) − ρt(a)k = 0
for all a ∈ A and g ∈ G.
Now let E : A → Aα be the conditional expectation given by
E(a) =
αg(a)
1
card(G)Xg∈G
for a ∈ A. For t ∈ [0, ∞) define a map ψt : A → Aα by ψt(a) = E(ρt(a)) for
a ∈ A.
Obviously ψt(a) ∈ Aα for all a ∈ A, and ψt is unital, linear, and com-
pletely positive. Also, t 7→ ψt(a) is clearly continuous for every a ∈ A. Since
kρt(a) − ψt(a)k = 0 for all a ∈ A.
G is finite, it follows from (3.3) that lim
t→∞
kψt(ab) − ψt(a)ψt(b)k = 0 for
kψt(a) − ak = 0 for all
(cid:3)
all a, b ∈ A. Combining it with (3.3) gives lim
t→∞
a ∈ Aα. This completes the proof.
Combining this relation with (3.1) gives lim
t→∞
K-THEORY
11
Proposition 3.8. Let A be a separable unital C*-algebra, and let G be a
finite group. Let α : G → Aut(A) be an action with the continuous Rokhlin
property. Assume that A satisfies the Universal Coefficient Theorem. Then
Aα and A ⋊α G satisfy the Universal Coefficient Theorem.
If A is simple and nuclear, then one does not need the continuous Rokhlin
property; the Rokhlin property suffices. See Corollary 3.9 of [28] for the
crossed product and, for actions of second countable compact groups, see
Theorem 3.13 of [12].
Proof of Proposition 3.8. Denote the suspension of a C*-algebra A by SA.
Let K be the algebra of compact operators. By Theorem 4.2 in [16], for ev-
ery pair of separable C*-algebras A and B, the group KK(A, B) is canon-
ically isomorphic to the group of homotopy classes of completely positive
asymptotic homomorphisms from K ⊗ SA to K ⊗ SB. Let ι : Aα → A be
the inclusion. Proposition 3.7 now implies that the group homomorphism
ψ∗ : KK(Aα, B) → KK(A, B) induced by the unital completely positive
asymptotic homomorphism (ψt)t∈[0,∞) obtained there satisfies ι∗ ◦ ψ∗ =
In particular, ψ∗ is naturally split injective with left inverse
idKK(Aα,B).
ι∗ : KK(A, B) → KK(Aα, B).
By hypothesis A satisfies the Universal Coefficient Theorem (Theorem 1.17
of [34]). That is, let B be any separable C*-algebra. Let
γA,B : KK(A, B) → Hom(K∗(A), K∗(B))
and
κA,B : Ker(γA,B) → Ext(K∗(A), K∗+1(B))
be as described before Theorem 1.17 of [34] (and called γ(A, B) and κ(A, B)
in [34] when A and B must be specified). Then δA,B = κ−1
A,B exists, and
there is a (natural) short exact sequence
0 −→ Ext(K∗(A), K∗+1(B))
δA,B
KK(A, B)
γA,B
Hom(K∗(A), K∗(B)) −→ 0.
(Naturality is Theorem 4.4 of [34].)
Consider the commutative diagram
KK(A, B)
γA,B
Hom(K∗(A), K∗(B))
ι∗
ι∗
KK(Aα, B)
γAα ,B
Hom(K∗(Aα), K∗(B)),
in which the vertical maps are induced by ι : Aα → A and the horizontal
ones are from the Universal Coefficient Theorem.
Because ι∗ is surjective, for any f ∈ Hom(K∗(Aα), K∗(B)) there exists
c ∈ Hom(K∗(A), K∗(B)) such that ι∗(c) = f . Since we are assuming that
A satisfies the Universal Coefficient Theorem, the top horizontal map is
surjective. Thus, there exists b ∈ KK(A, B) such that γA,B(b) = c. Then
x = ι∗(b) satisfies γAα,B(x) = [ι∗ ◦ γA,B](b) = f , so γAα,B is surjective.
Now consider the following commutative diagram, in which the vertical
maps ψ∗, induced by the asymptotic homomorphism (ψt)t∈[0,∞) of Proposi-
tion 3.7, have left inverses given by ι∗, while the horizontal maps are from
12
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
the Universal Coefficient Theorem:
Ker(γA,B)
κA,B
Ext(K∗(A), K∗+1(B))
ψ∗
ψ∗
Ker(γAα,B)
κAα,B
Ext(K∗(Aα), K∗+1(B)).
Take x ∈ Ker(γAα,B) such that κAα,B(x) = 0. Then
[κA,B ◦ ψ∗](x) = [ψ∗ ◦ κAα,B](x) = 0.
Since we are assuming that A satisfies the Universal Coefficient Theorem,
κA,B is an isomorphism. Therefore ψ∗(x) = 0, and by the injectivity of ψ∗
we conclude that x = 0. So κAα,B is injective. Lastly, the argument used to
prove that γAα,B is surjective, applied to the commutative diagram
Ker(γA,B)
κA,B
Ext(K∗(A), K∗+1(B))
ι∗
ι∗
Ker(γAα,B)
κAα,B
Ext(K∗(Aα), K∗+1(B)),
shows that κAα,B is surjective. Therefore Aα satisfies the Universal Coeffi-
cient Theorem.
Now we consider the crossed product A ⋊α G. By the Proposition in [33],
Aα is isomorphic to a corner of A ⋊α G. When α has the Rokhlin property,
Corollary 2.15 of [12] implies that this corner is strongly Morita equivalent to
A ⋊α G. (This is saturation of the action, a weaker condition that hereditary
saturation as proved in [12].) Since strong Morita equivalence preserves the
class of algebras satisfying the Universal Coefficient Theorem, we conclude
that A ⋊α G satisfies the Universal Coefficient Theorem.
(cid:3)
The following argument was suggested by E. Gardella.
Proposition 3.9. Let G be a finite group. Let A be a unital separable
C*-algebra which absorbs the UHF algebra of type card(G)∞. Suppose the
action α : G → Aut(A) has the Rokhlin property. Then:
(1) Taking µG to be the product type action of Example 3.3, we have
(G, A, α) ∼=(cid:0)G, Mcard(G)∞ ⊗ A, µG ⊗ α(cid:1).
(2) The action α has the continuous Rokhlin property.
Proof. We prove (1). We will need to cite theorems which use central se-
quence algebras, so we state notation for them. For a separable unital
C*-algebra A, we define A∞ = Cb(Z>0, A)/C0(Z>0, A), and we regard A
as a subalgebra of A∞ via its embedding in Cb(Z>0, A) as the algebra of
constant sequences. Then A′ ∩ A∞ is the relative commutant of this image
of A. (It is written A∞ in [17]. See Section 2.1 there.) For ω ∈ βZ>0 \ Z>0,
if in place of C0(Z>0, A) we use
we call the quotient Aω. The image of the constant sequences here can also
be identified with A, and we again get a relative commutant A′ ∩ Aω. There
na = (an)n∈Z>0 ∈ Cb(Z>0, A) :
lim
n→ω
an = 0o ,
K-THEORY
13
is an obvious surjective map A∞ → Aω, which gives a unital homomorphism
A′ ∩ A∞ → A′ ∩ Aω.
We now fix any ω ∈ βZ>0 \ Z>0.
Since A absorbs Mcard(G)∞ , Proposition 2.8 and the comment at the be-
ginning of the proof of Proposition 2.9 in [17] provide an injective unital
homomorphism Mcard(G)∞ → A∞ ∩ A′. Since Mcard(G)∞ is simple, it follows
from the previous paragraph that there is an injective unital homomorphism
Mcard(G)∞ → Aω ∩ A′. Lemma 3.12, the proof of Proposition 3.13 in [25],
and the fact that in Lemma 0.5 in [25] the isomorphism is approximately
unitarily equivalent to the given homomorphism (see the proof of Proposi-
tion A in [32]), now provide a unital isomorphism ϕ : Mcard(G)∞ ⊗ A → A
and unitaries wn in A for n ∈ Z>0 such that
kwnϕ(1 ⊗ a)w∗
n − ak = 0
(3.4)
lim
n→∞
for every a ∈ A.
Define an action β : G → Aut(A) by βg = ϕ ◦ (µG
g ⊗ αg) ◦ ϕ−1 for g ∈ G.
We claim that βg is approximately unitarily equivalent to αg for every g ∈ G.
Set vn = wnβg(w∗
n) for n ∈ Z>0. Let g ∈ G and a ∈ A. For n ∈ Z>0 we have
n − αg(a)k = kwnβg(w∗
kvnβg(a)v∗
≤ kwnβg(w∗
n − αg(a)k
n − wnβg(ϕ(1 ⊗ a))w∗
nk
nawn)w∗
nawn)w∗
+ kwnβg(ϕ(1 ⊗ a))w∗
n − αg(a)k
= kβgkkw∗
nawn − ϕ(1 ⊗ a)k
+ kwnϕ(1 ⊗ αg(a))w∗
n − αg(a)k.
Applying (3.4) to a and αg(a), we find that lim
n→∞
This proves the claim.
kvnβg(a)v∗
n − αg(a)k = 0.
By Theorem 3.5 in [17], there exists an approximately inner automor-
phism θ such that θ ◦ αg ◦ θ−1 = βg for every g ∈ G. Part (1) follows.
Part (2) is now immediate from Lemma 3.4 and Lemma 3.2.
(cid:3)
4. The construction
In this section we describe a method to construct simple separable C*-
algebras not isomorphic to their opposite algebras. We also show that these
C*-algebras satisfy the Universal Coefficient Theorem. Throughout this
section q is a fixed integer with q ≥ 2. The construction is a generalization
of the construction of [29] for q = 3. In Section 5, we will restrict q to being
an odd prime such that −1 is not a square mod q, or q = 4.
Definition 4.1. Let q ∈ {2, 3, . . .}. Define the C*-algebra Aq to be the
reduced free product of q copies of C([0, 1]) and the C*-algebra Cq, amalga-
mated over C, taken with respect to the states given by Lebesgue measure
µ on each copy of C([0, 1]) and the state given by
ω(c1, c2, . . . , cq) =
1
q(cid:0)c1 + c2 + · · · + cq(cid:1)
on Cq. That is,
Aq = C([0, 1]) ⋆r C([0, 1]) ⋆r · · · ⋆r C([0, 1]) ⋆r Cq.
14
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
For k = 1, 2, . . . , q we denote by εk : C([0, 1]) → Aq the inclusion of the
k-th copy of C([0, 1]) in Aq. Set
(4.1)
u0 =(cid:0)e2πi/q, 1, e2(q−1)πi/q, e2(q−2)πi/q, . . . , e4πi/q(cid:1) ∈ Cq,
and regard u0 as a unitary in Aq via the obvious inclusion.
Lemma 4.2. There exists a unique automorphism α ∈ Aut(Aq) such that
for all f ∈ C([0, 1]) we have
(4.2) α(ε1(f )) = ε2(f ), α(ε2(f )) = ε3(f ),
. . . , α(εq−1(f )) = εq(f ),
(4.3)
and
α(εq(f )) = Ad(u0)(ε1(f )),
(4.4)
Moreover, αq = Ad(u0).
α(u0) = e−2πi/qu0.
Proof. The proof is the same as that of Lemma 4.6 of [29].
(cid:3)
Remark 4.3. The C*-algebra Aq is unital, separable, simple, exact, and
has a unique tracial state. Exactness follows from Theorem 3.2 of [10].
Simplicity and uniqueness of the tracial state follow by applying the corollary
on page 431 of [1] several times. Lastly, simplicity and the existence of a
faithful tracial state imply that Aq is stably finite.
Definition 4.4. Define ϕn : Mqn → Mqn+1 by ϕn(x) = diag(x, x, . . . , x) for
x ∈ Mqn. Denote by Bq the UHF algebra obtained as the direct limit of the
system (Mqn, ϕn)n∈Z>0. We identify Mqn with
For k ≥ 1 let πk : Mq → Bq be the map
nOk=1
Mq, and Bq with
Mq.
∞Ok=1
πk(x) = 1 ⊗ 1 ⊗ · · · ⊗ 1 ⊗ x ⊗ 1 ⊗ 1 ⊗ · · · ,
with x in position k. Let λ : Bq → Bq denote the shift endomorphism
of Bq, determined by λ(πk(x)) = πk+1(x) for all k and all x ∈ Mq. Denote
by (ej,k)q
j,k=1 the standard system of matrix units in Mq. Define unitaries
v, u ∈ Bq by
defines an outer automorphism of Bq such that βq = Ad(v) and β(v) =
e2πi/qv.
Proof. The proof is the same as that of Lemma 4.3 in [29].
(cid:3)
Lemma 4.6. There exists a unitary w ∈ C ∗(u0 ⊗ v) ⊆ Aq ⊗ Bq such that
wq = u0 ⊗ v and (α ⊗ β)(w) = w. Moreover, if we set γ = Ad(w) ◦ (α ⊗ β) ∈
Aut(Aq ⊗ Bq), then γ generates an action of Zq which has the Rokhlin
property.
v = π1 qXk=1
e2πik/qek,k! and u = π1(eq,1)λ(v∗) +
Ad(cid:0)uλ(u) · · · λn(u)(cid:1)(x)
β(x) = lim
n→∞
Lemma 4.5. The formula
π1(ek, k+1).
q−1Xk=1
K-THEORY
15
Proof. The proof of the first statement is the same as that of Lemma 4.8
of [29]. It is straightforward to show that (cid:2)Ad(w) ◦ (α ⊗ β)(cid:3)q = idAq ⊗Bq .
The proof of the Rokhlin property is the same as that of Proposition 6.3
of [29].
(cid:3)
Definition 4.7. Set Cq = Aq⊗Bq, and let γ be the automorphism of Lemma
4.6. We also write γ for the action of Zq generated by this automorphism,
and define the C*-algebra Dq by Dq = Cq ⋊γ Zq.
Proposition 4.8. Let q ∈ {2, 3, . . .}. The C*-algebra Dq = Cq ⋊γ Zq of
Definition 4.7 is simple, separable, unital, and exact. It tensorially absorbs
the q∞ UHF algebra Bq and the Jiang-Su algebra Z. Moreover, Dq is
approximately divisible, stably finite, has real rank zero and stable rank
one, and has a unique tracial state which determines the order on projections
over Dq. Also,
q(cid:3)
K0(Dq) ∼= Z(cid:2) 1
and K1(Dq) = 0,
where the first isomorphism sends [1] to 1, and is an isomorphism of ordered
groups. Finally, letting Z(cid:2) 1
R, the Cuntz semigroup of Dq is given by
q(cid:3)+ be the set of nonnegative elements in Z(cid:2) 1
q(cid:3) ⊆
W (Dq) ∼= Z(cid:2) 1
q(cid:3)+ ∐ (0, ∞).
Proof. We first consider the algebra Cq = Aq ⊗ Bq in place of Dq, and we
prove that it has most of the properties listed for Dq. The exceptions are
that we do not prove stable finiteness or that the order on projections over
Cq is determined by traces, the K-theory is different (and we postpone its
calculation), and we do not compute the Cuntz semigroup.
It is obvious that Cq is separable and unital. To prove simplicity of Cq,
use simplicity of Aq (Remark 4.3), simplicity and nuclearity of the UHF
algebra Bq, and the corollary on page 117 of [36]. (We warn that [36] sys-
tematically refers to tensor products as "direct products".) Exactness of Cq
follows from exactness of Aq (Remark 4.3), exactness of Bq, and Proposi-
tion 7.1(iii) of [24]. Since Aq and Bq have unique tracial states (the first
by Remark 4.3), Corollary 6.13 of [9] (or Lemma 6.1 below) implies that Cq
has a unique tracial state. Since Aq is stably finite (Remark 4.3), and Bq
is a UHF algebra, Corollary 6.6 of [31] implies that tsr(Aq ⊗ Bq) = 1. The
algebra Bq is approximately divisible by Proposition 4.1 of [2], so Aq ⊗ Bq is
approximately divisible. Since Cq is simple, approximately divisible, exact,
and has a unique tracial state, it has real rank zero by Theorem 1.4(f) of [2].
The algebra Bq tensorially absorbs Bq, and tensorially absorbs the Jiang-Su
algebra Z by Corollary 6.3 of [19]. Therefore Cq tensorially absorbs both
algebras.
The algebra Dq is separable and unital because Cq is. Exactness of Dq
follows from Proposition 7.1(v) of [24]. Since β has period q in Out(Bq),
by Theorem 1 in [41], for k = 1, 2, . . . , q − 1 the automorphism γk is outer.
Theorem 3.1 of [26] now implies that Dq is simple. Since γ has the Rokhlin
property by Lemma 4.6, Dq has a unique tracial state by Proposition 4.14
of [28], tsr(Dq) = 1 by Proposition 4.1(1) of [28], Dq is approximately di-
visible by Proposition 4.5 of [28], and Dq has real rank zero by Proposition
4.1(2) of [28]. Combining Corollary 3.4(1) of [15] with the Rokhlin property,
16
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
we see that Dq absorbs both Bq and Z. Simplicity of Dq and existence of a
tracial state imply stable finiteness.
It now follows from Proposition 2.6 of [29] that the order on projections
over Dq is determined by traces.
The computation of K0(Dq) is done in the same way as in the proof of
Proposition 7.2 of [29], and we refer the reader to that article for the many
details we omit in the following computation. Here we have
q(cid:3),
K0(Aq) ∼= Zq, K1(Aq) = 0, K0(Bq) ∼= Z(cid:2) 1
so that the Kunneth formula (see [35]) gives
and K1(Bq) = 0,
Moreover, by the argument used in the proof of Proposition 7.2 of [29],
K0(Cq) ∼= Z(cid:2) 1
q(cid:3)q
q−1\m=0
K∗(Dq) ∼=
and K1(Cq) = 0.
ker(id − K∗(γm)).
For j = 1, 2, . . . , q, define rj = (0, . . . , 0, 1, 0, . . . , 0) ∈ Cq where 1 is in the
j-th position. Then the image of the unitary
u0 = e2πi/qr1 + r2 + e2(q−1)πi/qr3 + · · · + e4πi/qrq
under the map α is given by
α(u0) = r1 + e2(q−1)πi/qr2 + e2(q−2)πi/qr3 + · · · + e2πi/qrq.
This implies that α(rj) = rj−1 for j = 2, 3, . . . , q and that α(r1) = rq. Since
Ad(w) and β are trivial on K-theory, it follows that K0(γ) : Z(cid:2) 1
is given by
q(cid:3)q → Z(cid:2) 1
q(cid:3)q
K0(γ)(η1, η2, . . . , ηq) = (ηq, η1, η2, . . . , ηq−1).
Therefore id − K0(γ) corresponds to the matrix
1 −1
0
0
...
0
−1
0 · · ·
1 −1 · · ·
1 · · ·
0
...
...
. . .
0 · · ·
0
0
0 · · ·
0
0
0
0
0
0
...
...
1 −1
0
1
.
and one checks that its image is contained in ker(id − K0(γm)) for all m
q(cid:3) to ker(id − K0(γ)),
The map η → (η, η, . . . , η) is an isomorphism from Z(cid:2) 1
such that 0 ≤ m ≤ q − 1. Therefore this map is an isomorphism from Z(cid:2) 1
q(cid:3)
q−1\m=0
The computation of the Cuntz semigroup now follows from Remark 2.13
by observing that V (Dq) is the positive part of K0(Dq) and the uniqueness
of the tracial state on Dq implies that LAff b(T (Dq))++ = (0, ∞).
(cid:3)
ker(id − K0(γm)).
to
Proposition 4.9. Let Cq and γ : Zq → Aut(Cq) be as in Definition 4.7.
Then Dq = Cq ⋊γ Zq satisfies the Universal Coefficient Theorem.
K-THEORY
17
Proof. The C*-algebra Cq = Aq ⊗ Bq absorbs the q∞ UHF algebra. More-
over, by Lemma 4.6 the action γ has the Rokhlin property, so Proposition
3.9(2) implies that γ has the continuous Rokhlin property. Using Proposi-
tion 3.8, we conclude that Cq ⋊γ Zq satisfies the Universal Coefficient The-
orem.
(cid:3)
5. The main step
Let Dq = Cq ⋊ Zq be as in Definition 4.7, and let τ be its unique tracial
state. In this section we show that if q is an odd prime such that −1 is not
a square mod q, or if q = 4, then πτ (Dq)′′ is not isomorphic to its opposite
algebra. This is the main step in proving that Dq, as well as the tensor
product E ⊗ Dq for suitable E, is not isomorphic to its opposite algebra.
The following result belongs to the theory of cocycle conjugacy, but we
have not found a reference in the literature.
Lemma 5.1. Let M be a factor and let n ∈ Z>0. Let α, β : Zn → Aut(M )
be actions of Zn on M . Write the elements of Zn as 0, 1, . . . , n−1, so that, for
example, the automorphisms generating the actions are α1 and β1. Suppose
that there is a unitary y ∈ M such that β1 = Ad(y) ◦ α1. Then there is an
isomorphism ϕ : M ⋊β Zn → M ⋊α Zn which intertwines the dual actions,
that is, for all l ∈cZn we have ϕ ◦bβl =bαl ◦ ϕ.
Proof. For k ∈ Z we write αk = αk
1 . (This agrees with the
notation in the statement when k ∈ {0, 1, . . . , n − 1}.) For k ∈ Z>0 define a
unitary yk ∈ M by
1 and βk = βk
yk = yα1(y)α2(y) · · · αk−1(y).
Set y0 = 1, and define yk = αk(y∗
−k) for k < 0. Then one easily checks that
Ad(yk) ◦ αk = βk for all k ∈ Z, and moreover that yjαj(yk) = yj+k for all
j, k ∈ Z.
Since αn = βn = idM and M is a factor, we have yn ∈ C1. So there is a
scalar ζ with ζ = 1 such that yn = ζ n1. For k ∈ Z define zk = ζ −kyk. Then
zk is unitary, and we have Ad(zk)◦αk = βk for all k ∈ Z and zjαj(zk) = zj+k
for all j, k ∈ Z. Moreover, zj = zk whenever n divides j − k.
Let u0, u1, . . . , un−1 be the standard unitaries in the crossed product M ⋊α
k = αk(a)
Zn which implement α, so that for a ∈ M ⊂ M ⋊α Zn we have ukau∗
and
Similarly let v0, v1, . . . , vn−1 be the standard unitaries in M ⋊β Zn which
implement β. Then there is a unique linear bijection ϕ : M ⋊β Zn → M ⋊α Zn
such that ϕ(avk) = azkuk for a ∈ M and k = 0, 1, . . . , n − 1. One checks,
using the properties of (zk)k∈Z, that ϕ is a homomorphism. Moreover,
(cid:0)ϕ ◦bβl(cid:1)(avk) = ϕ(cid:0)e2πikl/navk(cid:1) = e2πikl/nazkuk =bαl(azkuk) = (bαl ◦ ϕ)(avk)
for every a ∈ M and k ∈ {0, 1, . . . , n − 1}. Therefore ϕ ◦bβl =bαl ◦ ϕ.
Lemma 5.2. Let A and B be C*-algebras. Let ρ be a state on A and let ω
be a state on B. Then πρ⊗ω(A ⊗min B)′′ ∼= πρ(A)′′⊗πω(B)′′.
(cid:3)
M ⋊α Zn =(n−1Xk=0
akuk : a0, a1, . . . an−1 ∈ M) .
18
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
Proof. The proof is straightforward (one starts by identifying Hρ⊗ω with
Hρ ⊗ Hω), and is omitted.
(cid:3)
Proposition 5.3. Let q be 4 or any odd prime such that −1 is not a
square mod q. Let Dq = Cq ⋊γ Zq be the C*-algebra of Definition 4.7.
Let τ be the unique tracial state on Dq (Proposition 4.8), and let πτ be
the Gelfand-Naimark-Segal representation associated to τ . Then the von
Neumann algebra πτ (Dq)′′ is not isomorphic to its opposite algebra.
Proof. Set
N =(cid:2)⋆q
1L∞([0, 1])(cid:3) ⋆ L(Zq),
and for k = 1, 2, . . . , q denote by εk : L∞([0, 1]) → N the inclusion of the
k-th free factor L∞([0, 1]) in N . Let u0 be the element of L(Zq) ∼= Cq defined
in (4.1) and let α be the automorphism of N given by
(5.1) α(ε1(g)) = ε2(g), α(ε2(g)) = ε3(g),
. . . , α(εq−1(g)) = εq(g),
(5.2)
α(εq(g)) = Ad(u0)(ε1(g))
for all g ∈ L∞([0, 1]), and
α(u0) = e−2πi/qu0.
(5.3)
Thus, (α)q = Ad(u0). Then N is the weak operator closure of the image of
Aq under the Gelfand-Naimark-Segal representation coming from the unique
tracial state on Aq (see Remark 4.3), and α ∈ Aut(N ) is an extension of
the automorphism α defined in (4.2), (4.3), and (4.4). Let ω be the unique
tracial state on Bq, and let R0 be the weak operator closure (πω(Bq))′′. Then
R0 is isomorphic to the hyperfinite II1 factor R. Denote by β the extension
of β to R0.
The automorphism Ad(w) ◦(cid:0)α ⊗ β(cid:1) generates an action γ of Zq on N ⊗R.
Since β has has period q in Out(R), Corollary 1.14 in [23] (or Theorem
13.1.16 in [22]) implies that γk is outer for k = 1, 2, . . . , q −1. By Proposition
13.1.5(ii) of [22] the crossed product M = (N ⊗R)⋊γ Zq is a factor of type II1.
We want to show that M ∼= πτ (Dq)′′. Let σ be the unique tracial state
on Aq. We have an obvious map Cq = Aq ⊗ Bq → N ⊗R which inter-
twines γ and γ. Lemma 5.2 shows that this map induces an isomorphism
πσ⊗ω(Cq)′′ ∼= N ⊗R. Since the group is finite, taking crossed products by Zq
gives an isomorphism M ∼= πτ (Dq)′′.
To show that M is not isomorphic to its opposite algebra, we give a recipe
which starts with a factor P , just given as a factor of type II1 with certain
properties (see (1), (2), (3), and (4) below), and produces a subset Sq(P ) of
Zq, which we identify with {0, 1, . . . , q − 1}. The important point is that this
recipe does not depend on knowing any particular element, automorphism,
etc. of P . That is, if we start with some other factor of type II1 which is
isomorphic to P , then we get the same subset of {0, 1, . . . , q − 1}, regardless
of the choice of isomorphism. When −1 is not a square mod q, we will show
that the recipe also applies to P op and gives a different subset, from which
it will follow that M op 6∼= M .
We describe the construction first, postponing the proofs that the steps
can be carried out and the result is independent of the choices made. Let P
K-THEORY
19
be a factor of type II1 with separable predual. Let χ(P ) denote the Connes
invariant of P as in Definition 2.7, and assume that P satisfies the following
properties:
(1) χ(P ) ∼= Z
(2) The unique subgroup of χ(P ) of order q is the image of a subgroup
q2.
(not necessarily unique) of Aut(P ) isomorphic to Zq.
(3) Let ρ : Zq → Aut(P ) come from a choice of the subgroup and iso-
Aut(P ⋊ρ Zq) be the dual action. Then for every nontrivial element
ϕ ◦ ψ, in which ϕ is an approximately inner automorphism and ψ is
a centrally trivial automorphism.
morphism in (2). Form the crossed product P ⋊ρ Zq, and letbρ :cZq →
l ∈ cZq, the automorphism bρl ∈ Aut(P ⋊ρ Zq) has a factorization
(4) For any nontrivial element l ∈ cZq and any factorization bρl = ϕ ◦ ψ
as in (3), there is a unitary z ∈ P ⋊ρ Zq such that ψq = Ad(z), and
there is k ∈ {0, 1, . . . , q − 1} such that ψ(z) = e2πik/qz. (See the
obstruction to lifting of Definition 2.9.)
For a type II1 factor P which satisfies (1), (2), (3), and (4), we take Sq(P )
to be the set of all values of k ∈ {0, 1, . . . , q − 1} which appear in (4) for any
subset of Zq in the obvious way.
choice of the action ρ : Zq → Aut(P ), any nontrivial element l ∈ cZq, and
any choice of the factorization bρl = ϕ ◦ ψ as in (3). We think of Sq(P ) as a
We claim that the crossed product P ⋊ρ Zq and the dual action bρ :cZq →
Aut(P ⋊ρ Zq) are uniquely determined up to conjugacy and automorphisms
of Zq. There are two ambiguities in the choice of ρ. If we change the iso-
morphism of Zq with the subgroup of χ(P ) of order q, we are modifying
ρ by an automorphism of Zq. The crossed product M ⋊ρ Zq is the same,
and the dual action is modified by the corresponding automorphism of cZq.
Suppose, then, that we fix an isomorphism of Zq with the subgroup of χ(P )
of order q, but choose a different lift ρ to a homomorphism Zq → Aut(P ).
Then Lemma 5.1 implies that the crossed products are isomorphic and the
dual actions are conjugate. This proves the claim. Since if the dual action
changes by conjugation by an automorphism, the automorphisms in the de-
composition of (3) also change by conjugation by an automorphism, and the
obstruction to lifting is unchanged by conjugation, it follows that changing
the dual action by conjugation leaves Sq(P ) invariant. This show that Sq(P )
can be computed by fixing a particular choice of ρ : Zq → Aut(P ).
Next we check that if P satisfies (1), (2), (3), and (4), then so does P op.
For this purpose, we use the von Neumann algebra P c described in Defini-
tion 2.1 which is isomorphic to P op by Remark 2.2. Scalar multiplication
enters in the definition of Sq(P ) in only two places. The first is the defini-
tion of the dual actionbρ :cZq → Aut(P ⋊ρ Zq). However, the change is easily
undone by applying the automorphism l 7→ −l of cZq. The other place is in
the definition of the obstruction to lifting. So P c satisfies the conditions (1),
(2), (3), and (4), and we get
(5.4)
Sq(P c) = {−l : l ∈ Sq(P )},
20
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
where we are treating Sq(·) as a subset of Zq.
In the rest of the proof, we show that the II1 factor M = (N ⊗R) ⋊γ Zq
satisfies (1), (2), (3), and (4), and that moreover Sq(M ) can be computed
computing Sq(M ).
using, for each nontrivial element l ∈cZq, just one choice of the factorization
bρl = ϕ ◦ ψ in (3) and one choice of the unitary z in (4). We then finish by
Begin by observing that N ∼= L(Fq ⋆ Zq) is full, which is equivalent to
saying that N has no nontrivial central sequences.
(See Lemma 3.2 and
Remark 3.1 of [40].) It follows from the proof of Proposition 3.5 of [40] that
N ⊗R has no nontrivial hypercentral sequences. The same argument as in
Section 5 of [40] shows that for our II1 factor M the short exact sequence
described in (2.1) is given by
0 → Zq → χ(M ) → Zq → 0
q2, as required for (1). Using similar reasoning
and χ(M ) is isomorphic to Z
to that in the proof of Theorem 5.5 of [40], one can also show that the unique
subgroup of order q in χ(M ) ∼= Z
Aut(M ) obtained as the dual action on M = (N ⊗R) ⋊γ Zq. Thus, there
exists at least one choice for ρ, namely σ composed with some isomorphism
q2 is the image of the action σ : bZq →
Zq →cZq. Therefore M satisfies property (2).
morphism
By Takesaki's duality theory (see Theorem 4.5 of [37]), there is an iso-
M ⋊ρcZq ∼= (N ⊗R) ⊗ B(l2(Zq))
λ(g) denotes the left regular representation of Zq on l2(Zq).
which identifies g 7→bρg with the tensor product g 7→ γg ⊗ Ad(λ(g)∗), where
Now let l ∈cZq. We claim that bρl can be written as ϕ ◦ ψ for an approxi-
mately inner automorphism ϕ and a centrally trivial automorphism ψ, and
that this factorization is unique up to inner automorphisms. This will imply
property (3). We first consider uniqueness, which is equivalent to showing
that every automorphism which is both approximately inner and centrally
trivial is in fact inner. Since, as noted above, N is full, the decomposi-
tion of Lemma 3.6 of [40] can be used to show that every automorphism
of N ⊗R which is both approximately inner and centrally trivial is in fact
inner. Uniqueness now follows because (N ⊗R) ⊗ B(l2(Zq)) ∼= N ⊗R.
For existence, since the approximately inner automorphisms are a normal
subgroup of Aut(cid:0)M ⋊ρcZq(cid:1), it suffices to take l to be the standard generator
of cZq. Equivalently, consider γ ⊗ Ad(λ(1)∗). We will take
ϕ =(cid:0)Ad(w) ◦(cid:0)idN ⊗ β(cid:1)(cid:1) ⊗ Ad(λ(1)∗) and ψ = (α ⊗ idR) ⊗ idB(l2(Zq)).
It is clear that γ ⊗Ad(λ(1)∗) = ϕ◦ψ. The automorphism ϕ is approximately
inner because, by construction, β is approximately inner. (In fact, by Theo-
rem XIV.2.16 of [38], every automorphism of R is approximately inner.) To
see that ψ is centrally trivial, we observe that, by the proof of Proposition 3.5
of [40], every central sequence in N ⊗R has the form (1⊗xn)∞
n=1 for
a central sequence (xn)∞
n=1 in N ⊗R such that
kynk2 = 0. Since N ⊗R has a unique trace, it follows immediately that
lim
n→∞
n=1 in R and a sequence (yn)∞
n=1 +(yn)∞
As observed in Equation (5.4) above, Sq(M op) is then given by
Sq(M ) =(cid:8) − l2 : l ∈ Zq \ {0}(cid:9).
Sq(M op) =(cid:8)l2 : l ∈ Zq \ {0}(cid:9).
K-THEORY
21
kψ(yn)k2 = 0, which implies that ψ is centrally trivial. This proves the
lim
n→∞
claim.
The obstruction to lifting for ψ (as in property (4)) is independent of the
choice of the unitary z implementing ψq because M is a factor. By the proof
of Proposition 1.4 of [8] it is unchanged if ψ is replaced by Ad(y) ◦ ψ for
any unitary y ∈ M ⋊σcZq. The centrally trivial factor in the decomposition
of any automorphism of M ⋊σ cZq, in particular, of (γ ⊗ Ad(λ(1)∗))l, is
determined up to inner automorphisms. Since, moreover, ϕ and ψ commute
up to an inner automorphism, it follows that we can compute Sq(M ) by
simply computing the obstructions to lifting for all powers ψl for a fixed
choice of ψ and for l = 1, 2, . . . , q − 1. We can take
ψ = (α ⊗ idR) ⊗ idB(l2(Zq)),
for which z = u0 ⊗ 1 ⊗ 1 has already been shown to be a unitary with
ψq = Ad(z) and ψ(z) = e−2πi/qz. Now one uses Equations (5.1), (5.2),
and (5.3) to check that
(ψl)q = Ad(zl) and ψl(zl) = e−2πil2/qzl.
Therefore (identifying {0, 1, . . . , q − 1} with Zq in the usual way)
If q is an odd prime, then we are assuming −1 is not a square mod q. If
q = 4, then one easily checks that −1 is not a square mod q. In either case,
we have Sq(M op) 6= Sq(M ), whence M op 6∼= M .
(cid:3)
6. C*-algebras not isomorphic to their opposite algebras
We now use the result of Section 5 to produce a number of examples
of simple exact C*-algebras not isomorphic to their opposite algebras and
which satisfy the Universal Coefficient Theorem.
Lemma 6.1. Let A and B be unital C*-algebras and assume that B has
a unique tracial state τ . Then the map σ 7→ σ ⊗ τ is an affine weak*
homeomorphism from the tracial state space T (A) of A to the tracial state
space T (A ⊗ B) of A ⊗ B.
We can't use Proposition 6.12 in [9], because (see Proposition 2.7 in [9])
it assumes that T (B) is finite dimensional.
Proof. It is easy to check that the map σ 7→ σ ⊗ τ is injective, by considering
(σ ⊗ τ )(a ⊗ 1) for a ∈ A.
We prove surjectivity. Let ρ be a tracial state on A ⊗ B. Define a tracial
state σ on A by σ(a) = ρ(a ⊗ 1) for a ∈ A. We claim that σ ⊗ τ = ρ. It
suffices to verify equality on a ⊗ b for a ∈ A+ and b ∈ B. So let a ∈ A+.
Define a tracial positive linear functional νa : B → C by νa(b) = ρ(a ⊗ b) for
b ∈ B. By uniqueness of τ , there is λ(a) ≥ 0 such that νa = λ(a)τ . Then
λ(a) = νa(1) = ρ(a ⊗ 1) = σ(a).
22
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
Thus for all b ∈ B, we have
(σ ⊗ τ )(a ⊗ b) = λ(a)τ (b) = νa(b) = ρ(a ⊗ b).
This completes the proof of surjectivity, and shows that the inverse map is
given by ρ 7→ ρA⊗1B .
It is obvious that ρ 7→ ρA⊗1B is affine and is continuous for the weak*
topologies. Since both tracial state spaces are compact and Hausdorff, it
follows that σ 7→ σ ⊗ τ is an affine homeomorphism.
(cid:3)
Proposition 6.2. Let q be 4 or any odd prime such that −1 is not a square
mod q. Let Dq be the C*-algebra of Definition 4.7. Let E be a simple
separable unital nuclear stably finite C*-algebra. Then E ⊗ Dq is exact and
E ⊗ Dq 6∼= (E ⊗ Dq)op.
Proof. Exactness follows from Proposition 7.1(iii) of [24].
Let τ be the unique tracial state on Dq (Proposition 4.8). Let R be the
hyperfinite II1 factor. We claim that
(6.1)
R⊗πτ (Dq)′′ ∼= πτ (Dq)′′.
To prove the claim, let ω be the unique tracial state on Bq. By Proposi-
tion 4.8, there is an isomorphism ϕ : Dq → Bq ⊗ Dq. Since (ω ⊗ τ ) ◦ ϕ is a
tracial state on Dq, we have (ω ⊗ τ ) ◦ ϕ = τ . Therefore πω⊗τ (Bq ⊗ Dq)′′ ∼=
πτ (Dq)′′. Since πω⊗τ (Bq ⊗ Dq)′′ ∼= πω(Bq)′′⊗πτ (Dq)′′ by Lemma 5.2, and
πω(Bq)′′ ∼= R, the claim follows.
We now claim that we may assume that E ⊗ Bq ∼= E. Indeed, Bq ⊗ Dq ∼=
Dq by Proposition 4.8, so that (E ⊗ Bq) ⊗ Dq ∼= E ⊗ Dq. Accordingly, we
may assume that E is also infinite dimensional. By Corollary 9.14 of [14],
there is a tracial state on E. So the Krein-Milman Theorem provides an
extreme tracial state σ on E. We have
(6.2)
πσ(E)′′ ∼= R.
Using Lemma 5.2 at the first step, (6.2) at the second step, and (6.1) at
the third step, we get
(6.3)
πσ⊗τ (E ⊗ Dq)′′ ∼= πσ(E)′′⊗πτ (Dq)′′ ∼= R⊗πτ (Dq)′′ ∼= πτ (Dq)′′.
Now suppose that there is an isomorphism ψ : (E ⊗ Dq)op → E ⊗ Dq. It
follows from Lemma 6.1 that σ ⊗ τ is an extreme tracial state on E ⊗ Dq.
Therefore (σ ⊗ τ ) ◦ ψ is an extreme tracial state on (E ⊗ Dq)op ∼= Eop ⊗ Dop
q .
Lemma 6.1 provides now an extreme tracial state ρ on Eop such that (σ ⊗
τ ) ◦ ψ = ρ ⊗ τ op. The state ρop is clearly extreme. Using (6.3) at the first
step, Lemma 2.4 at the fourth step, Lemma 5.2 at the fifth step, and (6.3)
with ρop in place of σ at the sixth step, we therefore get
πτ (Dq)′′ ∼= πσ⊗τ (E ⊗ Dq)′′ ∼= π(σ⊗τ )◦ψ((E ⊗ Dq)op)′′
= πρ⊗τ op ((E ⊗ Dq)op)′′ ∼= [πρop⊗τ (E ⊗ Dq)′′]op
∼= [πρop (E)′′⊗πτ (Dq)′′]op ∼= [πτ (Dq)′′]op.
This contradicts Proposition 5.3.
(cid:3)
K-THEORY
23
We use Proposition 6.2 to give many examples of simple separable exact
C*-algebras not isomorphic to their opposite algebras. Many other varia-
tions are possible. The ones we give are chosen to demonstrate the pos-
sibilities of nontrivial K1, of K0 being the same as that of many different
UHF algebras, of real rank one rather than zero, and of having many tracial
states.
Theorem 6.3. Let p be 2 or an odd prime such that −1 is not a square
mod p. Then there exists a simple separable unital exact C*-algebra A
not isomorphic to its opposite algebra which is approximately divisible and
stably finite, has stable rank one, tensorially absorbs the p∞ UHF algebra
and the Jiang-Su algebra, and has the property that traces determine the
order on projections over A. In addition, A has the following properties:
(2) K1(A) = 0.
(1) K0(A) ∼= Z(cid:2) 1
(3) W (A) ∼= Z(cid:2) 1
p(cid:3) with [1A] 7→ 1 and K0(A)+ → Z(cid:2) 1
p(cid:3)+ ∐ (0, ∞).
(4) A has real rank zero.
(5) A has a unique tracial state.
(6) A satisfies the Universal Coefficient Theorem.
p(cid:3) ∩ [0, ∞).
Proof. Take q = 4 if p = 2, and otherwise take q = p. Take A to be the
C*-algebra Dq of Definition 4.7. Then A 6∼= Aop follows from Proposition 5.3
(or equivalently from Proposition 6.2 with E = C). All the other properties
follow from Proposition 4.8 and Proposition 4.9, in the case p = 2 using
(cid:3)
B4 ∼= B2 and Z(cid:2) 1
4(cid:3) = Z(cid:2) 1
2(cid:3).
Theorem 6.4. Let p be 2 or an odd prime such that −1 is not a square
mod p. Let B be any UHF algebra whose "supernatural number" is divisible
by arbitrarily large powers of p. Then there exists a C*-algebra A as in
Theorem 6.3, except that (1) and (3) are replaced by:
(1) K0(A) ∼= K0(B) as a scaled ordered group.
(3) W (A) ∼= K0(B)+ ∐ (0, ∞).
Proof. Take q = 4 if p = 2, and otherwise take q = p. Let Dq be as in
Definition 4.7. Take A = B ⊗ Dq. Then exactness of A and A 6∼= Aop follows
from Proposition 6.2. Since Dq satisfies the Universal Coefficient Theorem,
and B belongs to the nuclear bootstrap category, A satisfies the Universal
Coefficient Theorem.
The condition on B implies that K0(B) ⊗Z Z(cid:2) 1
1(cid:0)K∗(B), Z(cid:2) 1
q(cid:3) ∼= K0(B). Moreover,
q(cid:3)(cid:1) is clearly zero. Since B is in the bootstrap class, the
TorZ
Kunneth formula of [35] gives K0(A) ∼= K0(B) and K1(A) = 0.
It is obvious that A is separable and unital. Simplicity of A follows from
simplicity of B and Dq and nuclearity of B, by the corollary on page 117
in [36]. (We warn that this reference systematically refers to tensor prod-
ucts as "direct products".) Since Dq has a unique tracial state (by Propo-
sition 4.8), Lemma 6.1 implies that A has a unique tracial state. Combined
with simplicity, this gives stable finiteness. The algebra A absorbs both
the UHF algebra Bq and Z because Dq does (by Proposition 4.8). Since
A is stably finite, Bq is a UHF algebra, and Bq ⊗ A ∼= A, Corollary 6.6
24
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
of [31] implies that tsr(A) = 1. The algebra Dq is approximately divisible
by Proposition 4.8. So A = B ⊗ Dq is approximately divisible. Since A is
simple, approximately divisible, exact, and has a unique tracial state, it has
real rank zero by Theorem 1.4(f) of [2]. It follows from Proposition 2.6 of [29]
that the order on projections over A is determined by traces. The compu-
tation of W (A) follows from the computation of K0(A) above, Z ⊗ A ∼= Z,
and Remark 2.13.
(cid:3)
Theorem 6.5. Let p be 2 or an odd prime such that −1 is not a square
mod p. Let G be any countable abelian group. Then there exists a C*-
algebra A as in Theorem 6.3, except that (2) is replaced by:
Proof. Choose, using Theorem 4.20 of [11], a simple unital AH algebra E
p(cid:3).
(2) K1(A) ∼= G ⊗Z Z(cid:2) 1
with a unique tracial state, such that K0(E) ∼= Z(cid:2) 1
such that K1(E) ∼= G. Take q = 4 if p = 2, and otherwise take q = p. Let
Dq be as in Definition 4.7. Take A = E ⊗Dq. Using E in place of B, proceed
as in the proof of Theorem 6.4. The only difference is in the computation
of K∗(A). We have
p(cid:3), with [1E ] 7→ 1, and
and
K0(E) ⊗Z Z(cid:2) 1
q(cid:3) ∼= Z(cid:2) 1
q(cid:3) ⊗Z Z(cid:2) 1
q(cid:3)(cid:1) = 0 and TorZ
q(cid:3) ∼= Z(cid:2) 1
q(cid:3),
q(cid:3)(cid:1) = 0.
1(cid:0)K1(E), Z(cid:2) 1
1(cid:0)K0(E), Z(cid:2) 1
TorZ
So the Kunneth formula of [35] implies that K∗(A) is as claimed.
(cid:3)
Theorem 6.6. Let p be 2 or an odd prime such that −1 is not a square
mod p. Let ∆ be any Choquet simplex with more than one point. Then
there exists a C*-algebra A as in Theorem 6.3, except that (3), (4), and (5)
are replaced by:
(4) A has real rank one.
(5) T (A) ∼= ∆.
Proof. Using Theorem 3.9 of [39], choose a simple unital AI algebra E such
(3) W (A) ∼= Z(cid:2) 1
that K0(E) ∼= Z(cid:2) 1
p(cid:3)+ ∐ LAff b(∆)++.
p(cid:3), with [1E] 7→ 1, and T (E) ∼= ∆. Take q = 4 if p = 2,
and otherwise take q = p. Let Dq be as in Definition 4.7. Take A =
E ⊗ Dq. Using E in place of B, proceed as in the proof of Theorem 6.4.
The differences are as follows. Here, since Dq has a unique tracial state (by
Proposition 4.8), Lemma 6.1 gives an affine homeomorphism from T (E) ∼=
∆ to T (A). The computation of W (A) is as before, but the answer is
different because T (A) ∼= ∆ instead of being a point. Since there is only
one state on the scaled ordered group K0(A), all tracial states must agree
on all projections in A. Since ∆ has more than one point, the projections
in A do not distinguish the tracial states. So A does not have real rank zero
by Theorem 1.4(e) in [2] and Theorem 5.11 in [13]. However, we still get
tsr(A) = 1, so A has real rank at most 1 by Proposition 1.2 of [3].
(cid:3)
Remark 6.7. Each of Theorem 6.4, Theorem 6.5, and Theorem 6.6 (sepa-
rately) gives uncountably many mutually nonisomorphic C*-algebras satis-
fying the Universal Coefficient Theorem.
K-THEORY
25
7. Open questions
Question 7.1. Let q be an odd prime such that −1 is a square mod q. Is
is still true that Dq, as in Definition 4.7, is not isomorphic to its opposite
algebra?
The invariant we use, the obstruction to lifting, no longer distinguishes
Dq and (Dq)op, but this does not mean that they are isomorphic.
Even if Dq ∼= (Dq)op, different methods might give a positive answer to
the following question.
Question 7.2. Let p be an odd prime such that −1 is a square mod p. Does
there exist a simple separable unital exact stably finite C*-algebra A not
isomorphic to its opposite algebra such that K0(A) ∼= Z(cid:2) 1
Of course, we would really like to get all the other properties in The-
orem 6.3 as well, in particular, unique tracial state, real rank zero, and
Bp ⊗ A ∼= A.
p(cid:3) and K1(A) = 0?
Question 7.3. Does there exist a simple separable unital exact stably finite
C*-algebra A not isomorphic to its opposite algebra such that K0(A) ∼= Z,
with [1A] 7→ 1, and K1(A) = 0?
Such an algebra would have no nontrivial projections.
Question 7.4. Does there exist a simple separable unital exact purely in-
finite C*-algebra A not isomorphic to its opposite algebra?
Acknowledgments
Some of this work was carried out during a summer school held at the
CRM in Bellaterra, during a visit to the Fields Institute in Toronto, and
during a visit by the second author to the University of Oregon. Both
authors are grateful to the CRM and the Fields Institute for their hospitality.
The second author would also like to thank the University of Oregon for its
hospitality.
References
[1] D. Avitzour, Free products of C*-algebras, Trans. Amer. Math. Soc. 271 (1982),
423 -- 435.
[2] B. Blackadar, A. Kumjian, and M. Rørdam, Approximately central matrix units and
the structure of non-commutative tori , K-Theory 6 (1992), 267 -- 284.
[3] L. G. Brown and G. K. Pedersen, C*-algebras of real rank zero, J. Funct. Anal. 99
(1991), 131 -- 149.
[4] N. P. Brown, F. Perera, and A. S. Toms, The Cuntz semigroup, the Elliott conjecture,
and dimension functions on C*-algebras, J. Reine Angew. Math. 621 (2008), 191 -- 211.
[5] A. Connes, Almost periodic states and factors of type III1, J. Funct. Anal. 16 (1974),
415 -- 445.
[6] A. Connes, Sur la classification des facteurs de type II , C. R. Acad. Sci. Paris S´er. A
281 (1975), 13 -- 15.
[7] A. Connes, A factor not anti-isomorphic to itself , Ann. Math. (2) 101 (1975), 536 --
554.
[8] A. Connes, Periodic automorphisms of the hyperfinite factor of type II1, Acta Sci.
Math. (Szeged) 39 (1977), 39 -- 66.
26
N. CHRISTOPHER PHILLIPS AND MARIA GRAZIA VIOLA
[9] J. Cuntz and G. K. Pedersen, Equivalence and traces on C*-algebras, J. Funct. Anal.
33 (1979), 135 -- 164.
[10] K. Dykema, Exactness of reduced amalgamated free product C*-algebras, Forum Math.
16 (2004), 161 -- 180.
[11] G. A. Elliott and G. Gong, On the classification of C*-algebras of real rank zero, II ,
Ann. Math. (2) 144 (1996), 497 -- 610.
[12] E. Gardella, Crossed products by compact group actions with the Rokhlin property,
preprint (arXiv:1408.1946v3 [math.OA]).
[13] U. Haagerup, Quasitraces on exact C*-algebras are traces C. R. Math. Rep. Acad.
Sci. Canada 36 (2014), 67 -- 92.
[14] U. Haagerup and S. Thorbjørnsen, Random matrices and K-theory for exact C*-
algebras, Documenta Math. 4 (1999), 341 -- 450 (electronic).
[15] I. Hirshberg and W. Winter, Rokhlin actions and self-absorbing C*-algebras, Pacific
J. Math. 233 (2007), 125 -- 143.
[16] T. Houghton-Larsen and K. Thomsen, Universal (co) homology theories, K-theory
16 (1999), 1 -- 27.
[17] M. Izumi, Finite group actions on C*-algebras with the Rokhlin Property, I Duke
Math. J. 122 (2004), 233 -- 280.
[18] M. Izumi, Finite group actions on C*-algebras with the Rohlin property. II , Adv.
Math. 184 (2004), 119 -- 160.
[19] X. Jiang and H. Su, On a simple unital projectionless C*-algebra, Amer. J. Math.
121 (1999), 359 -- 413.
[20] V. F. R. Jones, Notes on the Connes invariant χ(M ), unpublished notes.
[21] V. F. R. Jones, A II1 factor anti-isomorphic to itself but without involutory antiau-
tomorphism, Math. Scand. 46 (1980), 103 -- 117.
[22] R. V. Kadison and J R. Ringrose, Fundamentals of the Theory of Operator Algebras,
Vol. II. Advanced theory, Graduate Studies in Mathematics, 16, American Mathe-
matical Society, Providence, RI, 1997.
[23] R. R. Kallman, Jones index theory by Hilbert C*-bimodules and K-theory, Duke
Math. J. 36(1969), 781 -- 789.
[24] E. Kirchberg, Commutants of unitaries in UHF algebras and functorial properties of
exactness, J. reine angew. Math. 452 (1994), 39 -- 77.
[25] E. Kirchberg and N. C. Phillips, Embedding of exact C*-algebras in the Cuntz algebra
O2, J. reine angew. Math. 525 (2000), 17 -- 53.
[26] A. Kishimoto, Outer automorphisms and reduced crossed products of simple C*-
algebras, Commun. Math. Phys. 81 (1981), 429 -- 435.
[27] H. Lin, Tracially AF C*-algebras, Trans. Amer. Math. Soc. 353 (2001), 693 -- 722.
[28] H. Osaka and N. C. Phillips, Crossed products by finite group actions with the Rokhlin
property, Math Z. 270 (2012), 19 -- 42.
[29] N. C. Phillips and M. G. Viola, A simple separable exact C*-algebra not anti-
isomorphic to itself , Math. Ann. 355 (2013), 783 -- 799.
[30] M. Rørdam, The stable and the real rank of Z-absorbing C*-algebras, Internat. J.
Math. 15 (2004), 1065 -- 1084.
[31] M. Rørdam, On the structure of simple C*-algebras tensored with a UHF-algebra,
J. Funct. Anal. 100 (1991), 1 -- 17.
[32] M. Rørdam, A short proof of Elliott's Theorem: O2 ⊗ O2
∼= O2, C. R. Math. Rep.
Acad. Sci. Canada 16(1994), 31 -- 36.
[33] J. Rosenberg, Appendix to O. Bratteli's paper on "Crossed products of UHF algebras",
Duke Math. J. 46 (1979), 25 -- 26.
[34] J. Rosenberg and C. Schochet, The Kunneth theorem and the Universal Coefficient
Theorem for Kasparov's generalized K-functor , Duke Math. J. 55 (1987), 431 -- 474.
[35] C. Schochet, Topological methods for C*-algebras II: geometric resolutions and the
Kunneth formula, Pacific J. Math. 98 (1982), 443 -- 458.
[36] M. Takesaki, On the cross-norm of the direct product of C*-algebras, Tohoku Math.
J. (2) 16 (1964), 111 -- 122.
K-THEORY
27
[37] M. Takesaki, Duality for crossed products and the structure of von Neumann algebras
of type III , Acta Math. 13 (1973), 249 -- 310.
[38] M. Takesaki, Theory of Operator Algebras III , Springer-Verlag, Berlin, etc., 2003.
[39] K. Thomsen, Inductive limits of interval algebras: The tracial state space, Amer. J.
Math. 116 (1998), 605 -- 620.
[40] M. G. Viola, On a subfactor construction of a factor non-antiisomorphic to itself ,
Internat. J. Math. 15 (2004), 833 -- 854.
[41] S. Wassermann, Tensor products of *-automorphisms of C*-algebras, Bull. London
Math. Soc. 7 (1975), 65 -- 70.
Department of Mathematics, University of Oregon, Eugene OR 97403-1222,
USA
E-mail address: [email protected]
Department of Mathematical Sciences, Lakehead University-Orillia, 500 Uni-
versity Avenue, Orillia ON L3V 0B9, Canada
E-mail address: [email protected]
|
1008.0632 | 1 | 1008 | 2010-08-03T19:55:43 | Complex Hadamard matrices of order 6: a four-parameter family | [
"math.OA"
] | In this paper we construct a new, previously unknown four-parameter family of complex Hadamard matrices of order 6, the entries of which are described by algebraic functions of roots of various sextic polynomials. We conjecture that the new, generic family G together with Karlsson's degenerate family K and Tao's spectral matrix S form an exhaustive list of complex Hadamard matrices of order 6. Such a complete characterization might finally lead to the solution of the famous MUB-6 problem. | math.OA | math |
COMPLEX HADAMARD MATRICES OF ORDER 6:
A FOUR-PARAMETER FAMILY
FERENC SZ OLL OSI
Dedicated to Professor Uffe Haagerup on the occasion of his 60th birthday
Abstract. In this paper we construct a new, previously unknown four-parameter family
of complex Hadamard matrices of order 6, the entries of which are described by algebraic
functions of roots of various sextic polynomials. We conjecture that the new, generic family
G(4)
form an
exhaustive list of complex Hadamard matrices of order 6. Such a complete characterization
might finally lead to the solution of the famous MUB-6 problem.
together with Karlsson's degenerate family K (3)
6 and Tao's spectral matrix S(0)
6
6
2010 Mathematics Subject Classification. Primary 05B20, secondary 46L10.
Keywords and phrases. Complex Hadamard matrices, mutually unbiased bases, MUB
1. Introduction
Complex Hadamard matrices form an important family of orthogonal arrays with the
additional unimodularity constraint imposed on their entries. These matrices obey the alge-
braic identity HH ∗ = nIn where ∗ stands for the Hermitean transpose, and In is the identity
matrix of order n. They appear in various branches of mathematics frequently, including
linear algebra [6], coding- and operator theory [7, 15] and harmonic analysis [13, 19]. They
play an important role in quantum optics, high-energy physics, and they are one of the key
ingredients to quantum teleportation- and dense coding schemes [20] and mutually unbiased
bases (MUBs) [2]. An example of complex Hadamard matrices is the Fourier matrix Fn,
well-known to exists for all n. It is natural to ask how does a "typical" complex Hadamard
matrix of order n look like, and a satisfying answer to this question can be given provided
we have a complete characterization of Hadamard matrices of order n at our disposal. These
types of problems, however, are notoriously difficult even for small n. Naturally, one is inter-
ested in the essentially different matrices only, and we identify two matrices H and K and
say that they are equivalent if H = P1D1KD2P2 for some unitary diagonal matrices D1, D2
and permutational matrices P1, P2. Recall that a complex Hadamard matrix is dephased,
if all entries in its first row and column are equal to 1. While studying and classifying real
Hadamard matrices is naturally a discrete, finite problem which can be handled by deep
algebraic methods and sophisticated computer programs to some extent, the complex case,
however, behaves essentially different. In particular, due to the appearance of various para-
metric families one cannot hope for a finite list of inequivalent matrices, but rather for a
finite list of constructions, each of them leading to an infinite family of complex Hadamard
matrices.
The complete classification of complex Hadamard matrices is available up to order n = 5
only. It is trivial that Fn is the unique complex Hadamard matrix for orders n ≤ 3. The case
Date: August, 2010., Preprint.
This work was supported by the Hungarian National Research Fund OTKA K-77748.
1
2
FERENC SZ OLL OSI
n = 4 is still elementary, and it was shown by Craigen that all complex Hadamard matrices
of order 4 belong to an infinite, continuous one-parameter family [4]. In order 5 we have
uniqueness again, a result which is absolutely non-trivial already. In particular, Lov´asz was
the first who showed [12] that F5 is the only circulant complex Hadamard matrix in this
order, and a decade later Haagerup managed to prove the uniqueness of F5 by discovering
an algebraic identity (cf. formula (17)) relating the matrix entries in a surprising way [7].
In order 6 various one- [1, 5, 14, 21], two- [11, 17] and three-parameter families [9, 10] have
been constructed recently and it is conjectured that these are part of a more general, four-
parameter family of complex Hadamard matrices, yet to be discovered [2]. This conjecture
is supported by overwhelming numerical evidence [16], however so far only a fairly small
subset of it was described by closed analytic formulae, including an isolated matrix S(0)
6 and
a three-parameter matrix K (3)
6
[9, 10].
The reason why the 6 × 6 case received significant attention in the past couple of years
is the fact that complex Hadamard matrices are closely related to mutually unbiased bases.
Recall that two orthonormal bases of Cd, B1 and B2 are unbiased if for every e ∈ B1, f ∈ B2
we have he, fi2 = 1/d. A family of orthonormal bases is said to be mutually unbiased if
every two of them are unbiased. The famous MUB-6 problem asks for the maximal number
of mutually unbiased bases in C6. On the one hand this number is at least 3, as there
exists various infinite families of triplets of MUBs in this order [8, 17, 21], on the other
hand it is well-known that it cannot be larger than 7 (cf. the references of [2]).
In fact,
it is conjectured that a triplet is the best one can come up with in dimension 6 [21]. The
connection between MUBs and Hadamard matrices of order 6 has been exploited in [8] very
recently, where a discretization scheme was offered to attack the problem and it was proved
by means of computers, however, in a mathematically rigorous way, that the members of the
two-parameter Fourier family F (2)
6 (a, b) and its transpose cannot belong to a configuration of
7 MUBs containing the standard basis in dimension 6. One reasonable hope to finally settle
the MUB-6 problem is to give a complete characterization of complex Hadamard matrices
of order 6 and apply the same technique to them.
The goal of this paper is to propose a general framework towards the complete classification
of complex Hadamard matrices of order 6. In particular, by characterizing the orthogonal
triplets of rows in complex Hadamard matrices we generalize an observation of Haagerup [7]
to obtain a new algebraic identity relating the matrix entries in an unexpected way. This is
an essentially new tool to study complex Hadamard matrices of small orders, and one of the
main achievements of this paper. We apply this result to obtain complex Hadamard matrices,
moreover we conjecture that the the construction we present here reflects the true nature
of complex Hadamard matrices of order 6. It has the following three features: Firstly, it is
general in contrast with the earlier attempts where always some additional extra structure
was imposed on the matrices including self-adjointness [1], symmetry [14], circulant block
structure [17] or H2-reducibility [9]. Secondly, it has 4 degrees of freedom and thirdly all the
entries of the obtained matrices can be described by algebraic functions of roots of various
sextic polynomials. This suggests on the one hand the existence of a four-parameter family
of complex Hadamard matrices of order 6 and reminds us on the other hand the fact that
the desired algebraical description where the entries are expressed by radicals might not be
possible at all. However, from the applicational point of view, and in particular, to utilize the
computer-aided attack of [8] to the MUB-6 problem we shall need these matrices numerically
anyway.
and attempt to embed it into a complex Hadamard matrix of order 6
(2)
G(4)
6 (a, b, c, d) :=
≡(cid:20) E B
C D (cid:21) .
E(a, b, c, d) :=
1 1
1 a
1
c
1 g h ∗
1 t1
t3 ∗
1 t2
t4 ∗
1
1
1
b
e s1 s2
d f s3 s4
∗
∗
∗
∗
∗
∗
1 1 1
1 a b
1 c d
1
COMPLEX HADAMARD MATRICES OF ORDER 6
3
The outline of the paper is as follows.
In Section 2 we briefly discuss the main ideas
of the construction to motivate the various auxiliary results we prove there. The excited
reader might want to skip this section at first, and jump right ahead to Section 3, where the
construction of the new family is presented from a high-level perspective. In Section 4 we
analyze the construction thoroughly.
2. Preliminary results
In this section we present the ingredients necessary to construct our new family of complex
Hadamard matrices of order 6, including a characterization of the mutual orthogonality of
three rows in Hadamard matrices (cf. Theorem 2.4). First, however, we would like to motivate
these efforts by describing the main ideas of the construction.
We start with a submatrix
(1)
with 3× 3 blocks E, B, C and D in two steps, as follows. First we construct the submatrices
B and C featuring unimodular entries to obtain three orthogonal rows and columns of G6.
Secondly we find the unique lower right submatrix D to get a unitary matrix. Should the
entries of this matrix become unimodular, we have found a complex Hadamard matrix. We
conjecture that the submatrix E can be chosen, up to equivalence, in a way that there will
be only finitely many candidates for the blocks B and C and therefore we can ultimately
decide whether the submatrix E can be embedded into a complex Hadamard matrix. The
resulting matrix G6 can be thought as the "Hadamard dilation" of the operator E.
We shall heavily use the following through the paper without any further comment: the
conjugate of a complex number of modulus 1 is its reciprocal, and hence the conjugate of
a multivariate polynomial with real coefficients depending on indeterminates of modulus 1
is just the polynomial formed by entrywise reciprocal of the aforementioned indeterminates.
We computed various Grobner bases [3] in this paper with the aid of Mathematica. The
reader is advised to use a computer algebra system for bookkeeping purposes and consult
[18] for the standard notations for well-known complex Hadamard matrices such as S(0)
6 , K (3)
6 ,
etc.
We begin with recalling two elementary results from the existing literature.
Lemma 2.1. Suppose that we have a partial row (1, a, b, e,∗,∗) composed from unimodular
entries. Then one can specify some unimodular numbers s1 and s2 in place of the unknown
numbers ∗ to make this row orthogonal to (1, 1, 1, 1, 1, 1) if and only if
(3)
1 + a + b + e ≤ 2.
4
Proof. To ensure orthogonality, we need to have 1 + a + b + e + s1 + s2 = 0 from which it
follows that 1 + a + b + e = s1 + s2 ≤ 2. It is easily seen geometrically, that in this case
we can define the unimodular numbers required.
(cid:3)
FERENC SZ OLL OSI
The missing coordinates featuring in Lemma 2.1, s1 and s2, can be obtained algebraically
through the well-known
[14])). Suppose that the rows (1, 1, 1, 1, 1, 1) and
Lemma 2.2 ((Decomposition formula,
(1, a, b, e, s1, s2) containing unimodular entries are orthogonal. Let us denote by Σ := 1 + a +
b + e, and suppose that 0 < Σ ≤ 2. Then
Σ
(4)
2 ± i
Σr1 − Σ2
4
If Σ = 0 then s1 is independent from a, b, e but s2 = −s1.
Proof. Clearly s1 and s2 are the unimodular numbers with s1 + s2 = −Σ.
s1,2 = −
Σ
(cid:3)
.
Now we proceed by investigating the orthogonality of triplets of rows. In order to do this,
the following is a crucial
Definition 2.3 ((Haagerup polynomial)). The Haagerup polynomial H associated to the
rows (1, 1, 1, 1, 1, 1), (1, a, b, e,∗,∗) and (1, c, d, f,∗,∗) of a complex Hadamard matrix read
H(a, b, c, d, e, f ) := (1 + a + b + e)(1 + c + d + f )(1 + ca + db + f e).
The first result of ours is the following
Theorem 2.4. Suppose that we have the partial rows (1, a, b, e,∗,∗) and (1, c, d, f,∗,∗),
composed from unimodular entries. Then one can specify some unimodular numbers s1, s2, s3
and s4 in place of the unknown numbers ∗ to make these rows together with (1, 1, 1, 1, 1, 1)
mutually orthogonal if and only if
(5)
with
(6)
H(a, b, c, d, e, f ) = 4 − 1 + a + b + e2 − 1 + c + d + f2 − 1 + ca + db + f e2
H(a, b, c, d, e, f ) ≤ 8.
Proof. First we start by proving that (5) holds. To do this, we utilize Haagerup's idea [7] as
follows: by pairwise orthogonality, we find that
1 + a + b + e = −s1 − s2
1 + c + d + f = −s3 − s4
1 + ca + db + f e = −s3s1 − s4s2
Now, by multiplying these three equations together we find that (6) follows and further
H = −(s1 + s2)(s3 + s4)(s3s1 + s4s2) = −(s1s3 + s2s4 + s1s4 + s2s3)(s3s1 + s4s2) =
−s1s3 + s2s42 − (s1s4 + s2s3)(s3s1 + s4s2) = −s1s3 + s2s42 − 2ℜ(s1s2 + s3s4).
To conclude the proof, we need to show that
2ℜ(s1s2 + s3s4) = 1 + a + b + e2 + 1 + c + d + f2 − 4
holds, however this follows from the Decomposition formula easily.
COMPLEX HADAMARD MATRICES OF ORDER 6
5
To see the converse direction, we need to show that (5) essentially encodes orthogonality.
Let us use the notations Σ := 1 + a + b + e, ∆ := 1 + c + d + f , Ψ := 1 + ca + db + f e. With
this notation condition (5) boils down to
H = Σ∆Ψ = 4 − Σ2 − ∆2 − Ψ2.
(7)
Clearly, if Σ ≤ 2 and ∆ ≤ 2 hold, then by the decomposition formula we can find s1, s2,
s3 and s4 to ensure orthogonality to row (1, 1, 1, 1, 1, 1). Now observe, that the mutual
orthogonality of rows (1, a, b, e, s1, s2) and (1, c, d, f, s3, s4) reads
(8)
Ψ + s3s1 + s4s2 = 0.
Suppose first that we have the trivial case Σ = ∆ = 0. Then, by the decomposition formula
we have s2 = −s1 and s4 = −s3, and (7) implies that Ψ = 2. Therefore, if we set the
unimodular number s3 := −Ψs1/2 the orthogonality equation (8) is fulfilled.
Suppose secondly, that we have ∆ = 0, but Σ 6= 0. Then we have s4 = −s3, and from (7)
it follows that Σ ≤ 2, and in particular
(9)
Now we can use the Decomposition formula to find out the values of s1 and s2 and the
orthogonality equation (8) becomes
Ψ =p4 − Σ2.
Σr1 − Σ2
Σ
Ψ + s3 −2i
4 ! = 0.
(10)
This holds, independently of s3, if Σ = 2, as by (9) Ψ = 0 follows. Otherwise, set the
unimodular number
to ensure the orthogonality through (10).
s3 := −i
ΣΨ
ΣΨ
Finally, let us suppose that Σ 6= 0 and ∆ 6= 0. Now observe that in this case the value of
Ψ needed for formula (8) can be calculated through (7). The other ingredient, namely the
value of s3s1 + s4s2 can be established through the Decomposition formula, once we derive
the required bounds Σ ≤ 2 and ∆ ≤ 2. Depending on the value of H, we treat several
cases differently.
CASE 1: Suppose that −Ψ2 ≤ H. This implies, by formula (7), that Σ2 + ∆2 ≤ 4,
Suppose first that H ≥ 0. Hence, after taking absolute values, (7) becomes
and in particular Σ ≤ 2, ∆ ≤ 2 hold. Next we calculate Ψ from (7).
and the only non-negative root is
(11)
Ψ2 + Σ∆Ψ + Σ2 + ∆2 − 4 = 0,
Ψ = −Σ∆ +p(4 − Σ2)(4 − ∆2)
Ψ2 − Σ∆Ψ + Σ2 + ∆2 − 4 = 0,
2
.
Now suppose that −Ψ2 ≤ H < 0. Hence, after taking absolute values, (7) becomes
(12)
and we find that the only non-negative root we have under the assumption Σ2 + ∆2 ≤ 4
reads
(13)
Ψ = Σ∆ +p(4 − Σ2)(4 − ∆2)
2
.
6
FERENC SZ OLL OSI
CASE 2: Suppose now that H < −Ψ2. This implies that Σ2 + ∆2 > 4, and we do not
have a priori the bounds Σ ≤ 2, ∆ ≤ 2. Nevertheless, we derive equation (12) again, and
we find that the values of Ψ can be any of
(14)
Ψ1,2 = Σ∆ ±p(4 − Σ2)(4 − ∆2)
2
,
provided that the roots are real, namely we have either Σ > 2 and ∆ > 2 or Σ ≤ 2 and
∆ ≤ 2. The first case is, however, not possible, as it would imply Ψ > 2 contradicting the
crucial condition (6).
Once we have established the bounds Σ ≤ 2, ∆ ≤ 2 and the value(s) of Ψ has been
found, we are free to use the Decomposition formula to obtain the values of s1, s2, s3 and
s4. Clearly, we can set s1 with the + sign, while s2 with the − sign as in formula (4), up to
equivalence. However, we do not know a priori how to distribute the signs amongst s3 and
s4, and to simplify the notations we define
(15)
s3 = −
∆
2 ± i
∆
∆r1 − ∆2
4
,
s4 = −
∆
2 ∓ i
∆
∆r1 − ∆2
4
.
In particular, by using (7) and (15) we find that the orthogonality equation (8) becomes
(16)
4 − Σ2 − ∆2 − Ψ2
Σ∆
+
Σ∆
2 ± 2
Σ∆
Σ∆r1 − Σ2
4 r1 − ∆2
4
= 0,
where the ± sign agrees with the definition of s3. To conclude the theorem plug in all of the
possible values of Ψ as described in (11), (13), (14) into (16) to verify that for some choice
of the sign it holds identically.
(cid:3)
Remark 2.5. The two possible signs described by formula (14) can be realized. In particular,
there are two different orthogonal triplet of rows composed of sixth roots of unity where
Σ = ∆ = √3 in both cases, however, in one of the cases Ψ = 1 while Ψ = 2 in the
other.
Corollary 2.6 ((Haagerup's trick, [7])). Suppose that the rows (1, 1, 1, 1, 1, 1), (1, a, b, e, s1, s2)
and (1, c, d, f, s3, s4) composed of unimodular entries are mutually orthogonal. Then
(17)
H(a, b, c, d, e, f ) ∈ R.
Haagerup used the property (17) to give a complete characterization of complex Hadamard
matrices of order 5, or equivalently, describe the orthogonal maximal abelian ∗-subalgebras
of the 5 × 5 matrices [7]. Since then it was used in [1] and [14] to construct new, previ-
ously unknown complex Hadamard matrices of order 6 as well. However, to guarantee the
mutual orthogonality of three rows the necessary condition (17) should be replaced by the
more stronger identity (5). Nevertheless, (17) will play an essential role in this paper too.
These type of identities are extremely useful as they feature less variables than the standard
orthogonality equations considerably simplifying the calculations required.
Solving the system of equations (5) -- (17) is the key step to obtain the submatrices B and
C of G6. Once we have three orthogonal rows and columns we readily fill out the remaining
lower right submatrix D. This is explained by the following two lemmata, the first of which
being a special case of a more general matrix inversion
COMPLEX HADAMARD MATRICES OF ORDER 6
7
Lemma 2.7. If U and V are n × n matrices then
(In + UV )−1 = In − U (In + V U)−1 V
provided that one of the matrices In + UV or In + V U is nonsingular.
Proof. By symmetry, we can suppose that the matrix In + V U is nonsingular. Then, we have
(In + UV )(In − U(In + V U)−1V ) = In + UV − U(In + V U)(In + V U)−1V = In.
(cid:3)
Lemma 2.8. Suppose that we have a 6 × 6 partial complex Hadamard matrix consisting of
three orthogonal rows and columns, containing no vanishing 3 × 3 minor. Then there is a
unique way to construct a unitary matrix containing these rows and columns as a submatrix.
Proof. Let U be a 6 × 6 matrix with 3 × 3 blocks A, B, C and D, as the following:
U =(cid:20) A B
C D (cid:21) .
By the orthogonality of the first three rows and columns and using the fact that the entries
are unimodular, we have
(18)
(19)
AA∗ + BB ∗ = 6I3
A∗A + C ∗C = 6I3
To ensure orthogonality in-between the first three and the last three rows we need to have
AC ∗ + BD∗ = O3, the all 0 matrix. As B is nonsingular by our assumptions we can define
(20)
D := −CA∗(B−1)∗.
Now we need to show that the last three rows are mutually orthogonal as well. Indeed, by
using (20) and (18) we have
CC ∗ + DD∗ = C(cid:0)I3 + A∗(BB ∗)−1A(cid:1) C ∗ = C(cid:0)I3 + A∗ (6I3 − AA∗)−1 A(cid:1) C ∗,
which, by Lemma 2.7 and (19) is
C I3 +
1
6
A∗(cid:18)I3 −
1
6
AA∗(cid:19)−1
A! C ∗ = C(cid:18)I3 −
1
6
A∗A(cid:19)−1
C ∗ = 6C (C ∗C)−1 C ∗ = 6I3.
(cid:3)
We do not state that the obtained unitary matrix U is Hadamard, which is not true in
general. Recall that our goal is to embed the submatrix E into the matrix G6 (cf. (1)-(2)).
We have the following trivial
Lemma 2.9. Suppose that a submatrix E can be embedded into a complex Hadamard matrix
G6 of order 6 in which the upper right submatrix B is invertible. Then for some unimodular
submatrix C for which the first three columns of G6 are orthogonal the lower right submatrix
D = −CE ∗(B−1)∗ is unimodular.
Proof. Indeed, this is exactly what embedding means.
(cid:3)
In particular, if the submatrices B and C are chosen carefully, the unimodular property
of D follows for free.
8
FERENC SZ OLL OSI
Corollary 2.10. Start from a submatrix E and suppose that there are only finitely many
(invertible) candidate matrices B ∈ SOLB and C ∈ SOLC such that the first three rows and
columns of the matrix G6 are orthogonal. Then E can be embedded into a complex Hadamard
matrix of order 6 if and only if there is some B ∈ SOLB and C ∈ SOLC such that the matrix
D = −CE ∗(B−1)∗ is unimodular.
Note that due to the finiteness condition in Corollary 2.10 once we have all (but finitely
many) candidate matrices B and C we can decide algorithmically whether the submatrix E
can be embedded into a complex Hadamard matrix.
The next step is to characterize 6 × 6 complex Hadamard matrices with vanishing 3 × 3
minors. To do this we need two auxiliary results first.
Lemma 2.11. Suppose that in a dephased 6 × 6 complex Hadamard matrix there exist a
noninitial row (or column) containing three identical entries x. Then x = ±1 and this row
(or column) reads (1, 1, 1,−1,−1,−1), up to permutations.
Proof. Suppose to the contrary, that there are three nonreal numbers x in a row (column).
Then the sum of these numbers x together with the leading 1 read 1+3x2 = 10+6ℜ(x) > 4,
and hence, by Lemma 2.1 this row (column) cannot be orthogonal to the first row (column),
a contradiction. To ensure orthogonality to the first row (column) we should specify the
remaining three entries to −x and hence the last part of the statement follows.
(cid:3)
Recall that the core of a dephased complex Hadamard matrix of order n is its lower right
(n − 1) × (n − 1) submatrix. A vanishing sum of order k is a k-term sum adding up to 0.
The following breakthrough result was obtained very recently.
Theorem 2.12 ((Karlsson, [9, 10])). Let H be a dephased complex Hadamard matrix of
order 6. Then the following are equivalent:
(a) H belongs to the three-parameter degenerate family K (3)
6 ;
(b) H is H2-reducible;
(c) The core of H contains a −1;
(d) Some row or column of H contains a vanishing sum of order 2;
(e) Some row or column of H contains a vanishing sum of order 4.
In particular, Theorem 2.12 gives a characterization of complex Hadamard matrices of
order 6 containing F2 as a submatrix. The term H2-reducibility refers to the beautiful
structure of these matrices: they have a canonical form in which all 9 of their 2×2 submatrices
are complex Hadamard. Part (c) and (d) of Theorem 2.12 allow us to quickly recognize if
a matrix belongs to the family K (3)
6 , and we shall heavily use these conditions through our
paper. Part (e) tells us that once we have four entries in a row or column of a matrix which
lies outside the family K (3)
the Decomposition formula readily derives the unique remaining
6
two values through (4). As the family K (3)
forms a three-parameter subset, the matrices it
6
contains are atypical, hence the adjective "degenerate".
Corollary 2.13. Suppose that in a dephased 6×6 complex Hadamard matrix H there exist a
noninitial row (or column) containing three identical entries. Then H belongs to the family
K (3)
6 .
Lemma 2.14. Suppose that a 6 × 6 complex Hadamard matrix H has a vanishing 3 × 3
minor. Then H belongs to the family K (3)
6 .
COMPLEX HADAMARD MATRICES OF ORDER 6
9
Proof. Suppose that H has a vanishing 3×3 minor, say the upper left submatrix E(a, b, c, d),
as in formula (1). Such an assumption can be made, up to equivalence. As det(E) =
b + c − a − d + ad − bc = 0, we find that if any of the indeterminates a, b, c, d is equal to 1
then E contains a noninitial row (or column) containing three 1s, and therefore by Corollary
2.13 we conclude that this matrix belongs to the family K (3)
6 . Otherwise, we can suppose
that none of a, b, c, d is equal to 1 and hence we find that d = (a + bc − b − c)/(a− 1), which
should be of modulus one. To ensure this solve the equation dd − 1 = 0 to find that either
b = a or c = a should hold, but then we have either d = c or d = b as well. In particular, we
find that E has two identical rows (or columns). After enphasing the matrix, again, we find
that there is a full column (or row) of entries 1 and a reference to Corollary 2.13 concludes
the lemma.
(cid:3)
Therefore to investigate those matrices which lie outside the family K (3)
6 we can safely use
Lemma 2.8 and in particular the inversion formula (20).
It turns out, that the isolated matrix S(0)
6
is featured in the following
(cf. [18]) requires a special treatment as well. It
Lemma 2.15. Suppose that in a 6 × 6 dephased complex Hadamard matrix H there is a
noninitial row and column composed of cubic roots of unity. Then H is either equivalent to
S(0)
6
or belongs to the family K (3)
6 .
Let us denote by ω the principal cubic root of unity once and for all, that is ω := e2πi/3.
Proof. First suppose that the cubic row and column meet in a common 1. Then our matrix
looks like as the matrix H on the left below, up to equivalence:
H =
, H ′ =
1
1
1
1
1
1
1
1 ω2
1 ω2
1
1
1 ω ω ω2 ω2
d
ω a
b
∗
ω ∗ ∗
∗
∗ ∗
∗
∗ ∗
c
∗
∗
∗
1
1
1 1
1 1
1 a
1 ∗
1 ∗ ω2
1 ∗ ω2
1
1
ω ω ω2 ω2
d
1
b
∗
ω ∗
∗
∗
∗
∗
c
∗
∗
∗
.
Now by orthogonality of the first three rows we find that a + b = −ω and c + d = −1. Hence,
we can assume, up to equivalence, that a = 1, b = −ω2, and c = ω, d = ω2. But then, we
can fill out the fourth row, and the third and fourth column as well. We conclude that the
obtained matrix is equivalent to S(0)
6 .
Secondly, let us suppose that the cubic row and column meet in a common ω. This matrix
H ′ is depicted on the right above. Then, by calculating the orthogonality equations, we find
that either a = −1, and hence the matrix (if it can be completed to a Hadamard at all)
belongs to the family K (3)
6 , or a = ω, b = ω2, and, up to equivalence c = ω, d = ω2. This
implies that the third row and third column feature cubic entries only meeting in a common
1 therefore reducing the situation to the first case. The third case, namely when the cubic
row and column meet in a common ω2 can be treated similarly.
(cid:3)
Now we turn to the presetting of the submatrix E(a, b, c, d) (see (1)). In order to avoid
the case when the system of equations (5) -- (17) is linearly dependent we need to exclude
various input quadruples (a, b, c, d). However, it shall turn out that we are free to do such
10
FERENC SZ OLL OSI
restrictions, up to equivalence. Before proceeding further, let us define the following two-
variable function mapping T2 to C as follows:
E(x, y) := x + y + x2 + y2 + xy2 + x2y.
We say that y is an elliptical pair of x, if E(x, y) = 0. Observe that for a given x 6= −1
the sum of its elliptical pairs read y1 + y2 = −(1 + x2)/(1 + x). The following is a strictly
technical
Proposition 2.16 ((Canonical transformation)). Suppose that we have a complex Hadamard
matrix H inequivalent from S(0)
and any of the members of the family K (3)
6 . Then H has a
6
3 × 3 submatrix E(a, b, c, d) as in formula (1), up to equivalence, satisfying
(b − 1)(c − 1)(b − d2)(c − d2)(b − c)(bc − d)E(b, d)E(c, d) 6= 0.
(21)
Proof. The strategy of the proof is the following: first we pick a "central element" d from
the core of the matrix and then we show that b and c can be set satisfying (21). Recall, that
by Lemma 2.15 there is no a noninitial row and column composed from cubic roots of unity
in the matrix.
First let us assume that there is a 1 in the core somewhere. Suppose that there is a, to
say, row full of cubics. In this case set d = ω, and c = ω2. Now in the column containing d
there is a non-cubic entry γ, and we are free to set b = γ. If there is neither a full row nor a
column of cubics in the matrix, then set d = 1, and choose a non-cubic c from its row. Note
that there cannot be a further noninitial 1 in the row or column of d by Corollary 2.13. Now
observe that as the elliptical pairs of 1 are ω and ω2, we can choose a suitable b from the
column of d unless all entries there are members of the set {ω, ω2, c, c}. Note that ω together
with ω2 cannot be in the column of d at the same time, and from this it is easily seen that
we cannot define the value of b only if the column of d is one of the following four cases, up
to permutations: (1, 1, ω, c, c, c), (1, 1, ω, c, c, c), (1, 1, ω2, c, c, c), (1, 1, ω2, c, c, c). However,
by normalization and by orthogonality, the sum of the entries in this column should add
up to 0, and we find in all cases that the unimodular solution to c is a cubic root of unity,
contradicting the choice of c. Therefore one of the entries in the column of d is different from
ω, ω2, c, c which will be chosen as b.
Secondly, let us suppose, that there is no 1 in the core. In particular, all entries in the
core are different rowwise and columnwise.
Pick any d from the core of the matrix. Let us denote by c1 and c2 the elliptical pairs
of d (maybe c1 = c2, or they are undefined). Now we have several cases depending on the
appearance of these values in the row and column containing d.
CASE 1: c1 and c2 present in both the row and column containing d. Hence, in the row
and column containing d there are the entries 1, d, c1 and c2 already, the remaining two,
α and β are uniquely determined by the Decomposition formula. Note that α 6= d2, as
otherwise we would have β = −(2d + d2 + d3)/(1 + d) (as the sum of all entries in a noninitial
row add up to 0, and the sum of the elliptical pairs is known), which is unimodular if and
only if d = ±i or d = ω or d = ω2. In the first case we have β = ∓i and we are dealing with
a member of the family K (3)
6 . The second and third case imply that we have a full row and a
full column of cubics, a contradiction. If β 6= d/α then by picking c = α from the row we can
set b = β in the column. Otherwise, in case we have β = d/α, then reset the central element
d to the α which is in the same row as d. Now observe that after this exchange we should
met the requirements of Case 1 again (otherwise we are done here), and hence the elliptical
COMPLEX HADAMARD MATRICES OF ORDER 6
11
pairs of α, α1 and α2, should present in the row and column of α, which therefore contain
exactly the same entries. However α1,2 6= d, as otherwise orthogonality with the condition
E(α, d) = 0 would imply d = ±1, a contradiction; α1,2 6= d/α, as otherwise d = ω or d = ω2
would follow from the same argument implying that the row containing α has a noninitial
1, a contradiction. Therefore, the only option left is that α1,2 = c1,2. But then we can set
c = d 6= α2 and b = d/α 6= α2, and we are done.
CASE 2: c1 and c2 present in (to say) the row containing d, but only one of these values
(say c1) is present in the column of d. Let us denote by α and β the two further entries in
this row which, again, are different from d2. In the column of d there is already 1, d and
c1, and observe that the remaining three entries cannot be (d2, α, β) as this would imply
d2 = c2, contradicting our case-assumption. Therefore one of the three unspecified entries γ
is different from d2, α and β. Now if γ 6= d/α then set c = α, b = γ otherwise set c = β,
b = γ. We are done.
CASE 3: Only the value c1 is in (to say) the row containing d and in the column of d as
well. This is a tricky case, as it might happen that the undetermined triplet in both the
d-th row and column is precisely (d2, α, d/α), and therefore we cannot ensure condition (21).
However, from the orthogonality equation 1 + d + d2 + c1 + α + d/α = 0, from its conjugate,
and the elliptical condition E(c1, d) = 0 we can form a system of equations, the solution of
which can be found by computing a Grobner basis. By investigating the results, we find
that either α = −1 or α + d = 0 leading us to the family K (3)
or c1 = d. In the last case,
however, we can calculate the values of d and α explicitly. In particular, we find that the
values of d and α are given by some of the unimodular roots of the following polynomials
1 + 2d + 2d3 + d4 = 0 and 1 + 4α− 2α2− 8α3− 8α4 − 8α5− 2α6 + 4α7 + α8 = 0. Now reset the
"central" entry d to α, and observe that the elliptical pairs of α are not present in its row,
and therefore the conditions of this subcase are no longer met. Otherwise we can suppose
that the undetermined triplet in the column of d is not (d2, α, d/α). Set c = α 6= d2. Pick γ
from the column which is different from d2, α, d/α, set b = γ and we are done.
CASE 4: Only the value c1 is in (to say) the row containing d and in the column of d
there is the other elliptical value c2 6= c1. We can suppose that two of the undetermined
entries in the row of d satisfy α 6= d2 and β 6= d2 and set c = α. Now if in the column the
undefined triplet is precisely (α, d/α, d2), then observe that the same triplet cannot appear
in the row, as otherwise c1 = c2 would follow. Therefore we can reset c to a value different
from α, d/α, d2, and set b = α. Otherwise there is an entry in the column which can be set
to b, we are done.
6
CASE 5: In the column of d there is no elliptical value at all. Pick any c = α 6= d2 from
the row. Now in the column there are four unspecified entries. Clearly, one of them, say γ
will be different from d2, α and d/α. Set b = γ. We are done.
(cid:3)
Not every submatrix E can be embedded into a complex Hadamard matrix of order 6.
To offer a necessary condition, let us recall first that an operator A is called a contraction,
if kAk2 ≤ 1, where k.k2 denotes both the Euclidean norm on C6 and the induced operator
norm on the space of 6 × 6 matrices. We have the following
Lemma 2.17. If A is any 3× 3 submatrix of a complex Hadamard matrix H of order 6 then
A/√6 is a contraction.
12
FERENC SZ OLL OSI
Proof. Clearly, we can assume that this submatrix A is the upper left of the matrix H, which
we will write in block form, as follows:
H =(cid:20) A B
C D (cid:21) .
Now suppose, to the contrary that there is some vector s, such that kAsk2 > √6ksk2 and
consider the block vector s′ := (s, 0)T ∈ C6. We have
kHs′k2 =(cid:13)(cid:13)(As, Cs)T(cid:13)(cid:13)2 ≥(cid:13)(cid:13)(As, 0)T(cid:13)(cid:13)2 = kAsk2 > √6ksk2 = √6 ks′k2 = kHs′k2 ,
where in the last step we used that the matrix H/√6 is unitary.
(cid:3)
In particular, we have the following
Corollary 2.18. If the submatrix E can be embedded into a complex Hadamard matrix of
order 6, then every eigenvalue λ of the matrix E ∗E satisfy λ ≤ 6.
Corollary 2.18 is a useful criterion to show that a matrix E cannot be embedded into
a complex Hadamard matrix, however, it is unclear how to utilize it for our purposes. In
particular, we do not know how to characterize those 3 × 3 matrices which satisfy its con-
ditions. Also it is natural to ask whether the presence of the large eigenvalues is the only
obstruction forbidding the submatrix E to be embedded. The answer to this question might
depend on the dimension, as it is easily seen that while every 2× 2 matrix can be embedded
into a complex Hadamard matrix of order 4, only a handful of very special 2 × 2 matrices
can be embedded into a complex Hadamard matrix of order 5 due to the finiteness result of
Haagerup [7].
Now we are ready to present a new, previously unknown family of complex Hadamard
matrices. The next section gives an overview of the results.
3. The construction: A high-level perspective
Here we describe the generic family G(4)
6
from a high-level perspective. In particular, we
outline the main steps only, and do not discuss some degenerate cases which might come up
during the construction. The next section is dedicated to investigate the process in details.
The main result of this paper is the following
Construction 3.1 ((The Dilation Algorithm)). Do the following step by step to obtain
complex Hadamard matrices of order 6.
#1: INPUT: the quadruple (a, b, c, d), forming the upper left 3× 3 submatrix E(a, b, c, d), as
#2: Use Haagerup's trick to the first three rows of G(4)
(see (2)) to obtain a quadratic
6
in formula (1).
equation to f :
(22)
(23)
F1 + F2f + F3f 2 = 0,
where the coefficients F1, F2 and F3 depend on the parameters a, b, c, d and the inde-
terminate e, and derive the following linearization formula from it:
f 2 = −F1
F3 − F2
F3
f.
#3: Use Theorem 2.4 to obtain another quadratic equation to f :
COMPLEX HADAMARD MATRICES OF ORDER 6
13
G1 + G2f + G3f 2 = 0,
where, again, the coefficients G1, G2 and G3 depend on the parameters a, b, c, d and the
indeterminate e. Plug the linearization formula (23) into (24) and rearrange to obtain
the companion value of e f = F (e), where
(24)
(25)
F (e) := −F3G1 − F1G3
F3G2 − F2G3
.
#4: As f = 1 should hold, calculate the sextic polynomial G(e) coming from the equation
F (e)F (e) − 1 = 0 and solve it for e.
#5: Amongst the roots of G find all unimodular triplets (e, s1, s2) satisfying e + s1 + s2 =
−1 − a − b, calculate the companion values f = F (e), s3 = F (s1) and s4 = F (s2)
through formula (25) and store all sextuples (e, s1, s2, f, s3, s4) in a solution set called
SOLB.
#6: Repeat steps #2 -- #6 to the transposed matrix (i. e. to the first three columns), mutatis
mutandis to obtain the solution set SOLC.
#7: For every pair of sextuples from SOLB and SOLC construct the submatrices B and
C, check if the first three rows and columns are mutually orthogonal and finally use
Lemma 2.8 to compute the lower right submatrix D through formula (20).
#8: OUTPUT:all unimodular matrices found in step #7.
Construction 3.1 gives the essence of the new family discovered, and in the next section
we shall give it a mathematically rigorous, low-level look.
4. The construction: The nasty details
Here we investigate the steps of Construction 3.1 in details.
STEP #1: Choose a quadruple (a, b, c, d) in compliance with the Canonical Transformation
described by Proposition 2.16 as the INPUT, and form the submatrix E. Check if it meets
the requirements of Corollary 2.18; if yes, then proceed, otherwise conclude that it cannot be
embedded into a complex Hadamard matrix of order 6. Experimental results show that once
three out of the four parameters are fixed the last one can be easily set to a value such that
the quadruple (a, b, c, d) leads to a complex Hadamard matrix. Heuristically this means that
there is nothing "mystical" in the choice of the initial quadruple and hence the parameters
should be independent from each other.
STEP #2: To obtain formula (23) we need to see that F3 6≡ 0, independently of e. Indeed,
suppose otherwise, which means that the following system of equations (where the last two
are the conjugate of the first two, up to some irrelevant constant factors)
abc + a2bc + abd + ab2d + a2bcd + ab2cd ≡ 0,
≡ 0,
≡ 0,
≡ 0,
are fulfilled. We compute a Grobner basis and find that the polynomial
b2c + ab2c + a2d + a2bd + acd + bcd
a2b + ab2 + bc + b2c + ad + a2d
a + b + ac + abc + bd + abd
bc(1 + c2)(c − d)d(1 + d2)(c2 + d2)(1 + d + d2)
14
FERENC SZ OLL OSI
is a member of it. After substituting back into the original equations we find that there is
either a vanishing sum of order 2 in E or a = b = 1 and therefore the whole family is a
member of K (3)
3 but these matrices have b = c which however
is not allowed by the Canonical Transformation.
6 , or we have E = F3 or E = F ∗
It might happen that F3 6≡ 0 but there is a unimodular e making it vanish, which cannot
be anything else, but
(26)
e =
a2b + a2d + ab2 + ad + b2c + bc
abc + abd + ac + a + bd + b
.
Nevertheless, we can suppose that in case of one of the pairs (e, f ), (s1, s3) and (s2, s4) we do
not set e as above, otherwise we would have e = s1 = s2 which, by Lemma 2.11 would imply
e = s1 = s2 = −1, obtaining some member of the family K (3)
6 by Corollary 2.13. Hence,
we can suppose that e is different than the value described by formula (26) above, and we
conclude that F3 6= 0.
STEP #3: Clearly, one cannot expect to recover a unique f from e in general, as formula
(25) might suggests. Indeed, there are complex Hadamard matrices in which s1 = e, but
s3 6= f . The reason for this phenomenon is that formulas (22) and (24) might be linearly
dependent. After plugging (23) into (24) we obtain the expression
which can lead us to one of the following three cases:
F3G1 − F1G3 + (F3G2 − F2G3) f = 0,
CASE 1: Both the polynomials F3G1 −F1G3 and F3G2 −F2G3 vanish identically, indepen-
dently of e, meaning that in this case we do not have another condition on f . Luckily this
can never happen, as by calculating a Grobner basis (again, to speed up the computations
we have added the conjugates of the equations as well) we find that the polynomial
(b − 1)b2(b − c)c(bc − d)(1 + c + d)(b − d2)E(c, d)
is a member of the basis. Therefore we have either c + d = −1 or one of the degenerate cases
described in the Canonical Transformation. The case c + d = −1 implies that the set {c, d}
consists of nontrivial cubic roots only. By directly solving the corresponding equations, we
find that the quadruples (1, 1, ω, ω2), (ω2, ω, ω, ω2) and (1, 1, ω2, ω), (ω, ω2, ω2, ω) can vanish
both polynomials, however these cases were excluded by the Canonical Transformation.
CASE 2: Both the polynomials F3G1 − F1G3 and F3G2 − F2G3 vanish for some e = 1,
meaning that in this case we do not have another condition on f . However, having these
numbers e at our disposal we can recover the two possible values of f from (23). Once we
have the candidate pairs (e, f1) and (e, f2) we readily calculate the remaining pairs (s1, s3)
and (s2, s4) through the Decomposition Formula. Store all suitable sextics (e, s1, s2, f, s3, s4)
in the solution set SOLB. Proceed to step #6.
CASE 3: One cannot set (or have not set in Case 2) a unimodular e to make these two
polynomial vanish at the same time. Hence we can derive formula (25) for F (e). Proceed to
step #4.
STEP #4: Next we need to ensure that f is of modulus one. To do this, we calculate the
fundamental polynomial
Pa,b,c,d(e) ≡ F3G1 − F1G32 − F3G2 − F2G32 .
After some calculations, it will be apparent that P has the following remarkable structure:
P ≡ a8b8c6d6P + a8b8c6d6P (2 + 2a + 2b)e + a8b8c6d6Qe2 + Re3 + Qe4 + P (2 + 2a + 2b)e5 + P e6
COMPLEX HADAMARD MATRICES OF ORDER 6
15
where the coefficients P, Q and R depend on the quadruple (a, b, c, d) only. If P ≡ 0 then
the construction fails. Otherwise we find all possible roots of P(e) of modulus one.
STEP #5: If the number on the right hand side of (26) is not of modulus one, then the
role of the pairs (e, f ), (s1, s3) and (s2, s4) is symmetric, and from all of the roots of P of
modulus one we select all possible triplets (e, s1, s2) satisfying e + s1 + s2 = −1− a− b, which
is needed to ensure orthogonality of the first two rows. From (25) we compute the unique
companion values f = F (e), s3 = F (s1) and s4 = F (s2).
Otherwise, should the number on the right hand side of (26) is of modulus one, then for
every root e of P we calculate its unique companion value f = F (e), and then we use the
Decomposition formula to determine the pairs (s1, s3) and (s2, s4).
If no unimodular roots are found at this point then the matrix E(a, b, c, d) cannot be
embedded into a complex Hadamard matrix of order 6, and we do not proceed any further.
At the end of this step we store all obtained sextics (e, s1, s2, f, s3, s4) in the solution set
SOLB. Typically two sextics are found.
STEP #6: In this step one constructs the first three columns along the lines of steps #2 -- #5
described above and obtain the set SOLC in a similar way.
STEP #7: For every candidate solution from SOLB and SOLC we check if the first three
rows and columns are orthogonal, disregard those cases in which the submatrix B is singular
and finally use Lemma 2.8 to obtain the lower right submatrix D. Note that by Lemma 2.14
we disregard members of the family K (3)
only. We check if D is composed of unimodular
6
entries.
STEP #8: Finally, we OUTPUT all unimodular matrices found during the process. We remark
here that by Corollary 2.10 if no unimodular matrices were found then the submatrix E
cannot be embedded into any complex Hadamard matrices of order 6. If unimodular matrices
are found, then typically we find two matrices, as the solution set SOLB and SOLC contains
two suitable sextics each, however, experimental results show that for each sextic in SOLB
there is a unique sextic in SOLC making D unimodular as required.
We have finished the discussion of Construction 3.1. The results are summarized in the
following
Theorem 4.1. Start from a submatrix E as in (1) and suppose that there are only finitely
many (invertible) candidate submatrices B and C such that the first three rows and columns
of the matrix G6 (see (2)) are orthogonal. Then Construction 3.1 gives an exhaustive list of
all complex Hadamard matrices of order 6, up to equivalence, containing E as a submatrix.
The interested reader might want to see an example of generic Hadamard matrices which
can be described by closed analytic formulae, that is for which the fundamental polynomials
Pa,b,c,d and Pa,c,b,d are both solvable. Such a matrix can be obtained when we choose the input
quadruple (a, a, c, a) where the real part of a is the unique real solution of 4ℜ[a]3−2ℜ[a]+1 =
0 and c = (−a3 + a2 + a + 1)/(a4 + a3 + a2 − a). It is easily seen that the matrices we obtain
starting from the submatrix E(a, a, c, a) are inequivalent from S(0)
6 and do not belong to the
family K (3)
6 .
Remark 4.2. When P ≡ 0 then the main difficulty we are facing with is that we have
infinitely many candidate submatrices B. In this case we have the trivial restriction (3) on
e, while the companion value f coming from (25) is unimodular unconditionally. Although
16
FERENC SZ OLL OSI
in principle we can find three orthogonal rows through the Decomposition formula for every
suitable e, we do not know which one to favourize in order to obtain a unimodular submatrix
D via formula (20). Also, it might happen that the polynomial Pa,c,b,d obtained during step
#6 shall vanish as well bringing another free parameter into the game making things even
more complicated. In contrast, if both Pa,b,c,d 6≡ 0 and Pa,c,b,d 6≡ 0, then we have a finitely
many choices for the submatrices B and C and we can use Corollary 2.10 to conclude the
construction.
Remark 4.3. The polynomial P formally can vanish when we have F3G1 − F1G3 ≡ F3G2 −
F2G3 ≡ 0, however this is excluded by the Canonical Transformation and explained in details
in Case 1 of step #3. It might vanish for some other, non-trivial quadruples as well making
the whole construction process fail. In theory, the common roots of the coefficients of P can
be calculated by means of Grobner bases, but as these coefficients are rather complicated
obtaining such a basis turned out to be a task beyond our capabilities. Nevertheless, we
conjecture that the case P ≡ 0 can be excluded completely in a similar fashion as we
disregarded various quadruples during the Canonical Transformation. This would mean
that all complex Hadamard matrices of order 6, except from S(0)
6 , can be recovered
from Construction 3.1.
6 and K (3)
It is reasonable to think that every complex Hadamard matrix of order 6 has some 3 × 3
In particular, we do not
6 ) which
submatrix E leading to nonvanishing fundamental polynomials.
expect any complex Hadamard matrices of order 6 (except maybe S(0)
6
cannot be recovered from Construction 3.1. Therefore we formulate the following
and K (3)
Conjecture 4.4. The list of complex Hadamard matrices of order 6 is as follows: the isolated
matrix S(0)
6 and the four-parameter generic family
G(4)
6 , the three-parameter degenerate family K (3)
6 as described above.
It would be nice to understand the structure of G(4)
6 more thoroughly and express the
entries of these matrices by some well-chosen trigonometric functions in a similar fashion as
K (3)
is described, however, as we have encountered sextic polynomials already the appearance
6
of such formulas is somewhat unexpected. Also, it is natural to ask which matrices satisfy the
conditions of Corollary 2.18. An algebraic characterization of these matrices might lead to a
deeper understanding of the generic family G(4)
6 and hopefully to the desired full classification
of complex Hadamard matrices of order 6.
References
[1] K. Beauchamp and R. Nicoara, Orthogonal maximal abelian ∗-subalgebras of the 6 × 6 matrices,
[2] I. Bengtsson, W. Bruzda, A. Ericsson, J.-A. Larsson, W. Tadej and K. Zyczkowski, Mu-
Linear Algebra and its Applications, 428:8 -- 9 (2008), 1833 -- 1853.
tually unbiased bases and Hadamard matrices of order six, J. Math. Phys. 48, 052106 (2007)
[3] B. Buchberger, An Algorithm for Finding the Basis Elements in the Residue Class Ring Modulo
a Zero Dimensional Polynomial Ideal., PhD Thesis, Mathematical Institute, University of Innsbruck,
Austria, 1965.
[4] R. Craigen, Equivalence classes of inverse orthogonal and unit Hadamard matrices, Bull. Austral.
Math. Soc., 44 (1991), 109 -- 115.
[5] P. Dit¸a, Some results on the parametrization of complex Hadamard matrices, J. Phys. A, 20 (2004),
5355 -- 5374.
[6] C. Godsil and A. Roy, Equiangular lines, mutually unbiased bases, and spin models, European
Journal of Combinatorics 30 (2009), 246 -- 262.
COMPLEX HADAMARD MATRICES OF ORDER 6
17
[7] U. Haagerup, Orthogonal maximal Abelian ∗-subalgebras of n×n matrices and cyclic n-roots, Operator
[8] Ph. Jaming, M. Matolcsi, P.M´ora, F. Szollosi and M.Weiner, A generalized Pauli problem
Algebras and Quantum Field Theory (Rome), MA International Press, (1996), 296 -- 322.
and an infinite family of MUB-triplets in dimension 6 , J. Phys. A: Math. Theor. 42 245305
[9] B. R. Karlsson, H2-reducible Hadamard matrices of order 6, Preprint, arXiv:1003.4133v1 [math-ph]
(2010)
[10] B. R. Karlsson, Three-parameter complex Hadamard matrices of order 6, Preprint, arXiv:1003.4177v1
[math-ph] (2010)
[11] B. R. Karlsson, Two-parameter complex Hadamard matrices for N = 6, J. Math. Phys. 50, 082104
(2009)
[12] L. Lov´asz, Unpublished, (1984).
[13] M. N. Kolountzakis and M. Matolcsi, Complex Hadamard matrices and the spectral set conjecture,
Collectanea Mathematica, Vol. Extra (2006), 281 -- 291.
Systems & Information Dynamics, 15:2 (2008), 93 -- 108.
[14] M. Matolcsi and F. Szollosi, Towards a classification of 6 × 6 complex Hadamard matrices, Open
[15] S. Popa, Orthogonal pairs of ∗-subalgebras in finite von Neumann algebras, J. Operator Theory, 9
[16] A. J. Skinner, V. A. Newell and R. Sanchez Unbiased bases (Hadamards) for six-level systems:
(1983), 253 -- 268.
Four ways from Fourier J. Math. Phys. 50, 012107 (2009)
[17] F. Szollosi, A 2-parameter family of complex Hadamard matrices of order 6, induced by hypocycloids,
Proc. Amer. Math. Soc. 138:3 (2010) 921 -- 928.
[18] W. Tadej and K.
Zyczkowski, A concise guide to complex Hadamard matrices, Open Syst. Inf.
Dyn., 13 (2006), 133 -- 177.
[19] T. Tao, Fuglede's conjecture is false in 5 and higher dimensions, Math Res. Letters, 11 (2004), 251 -- 258.
[20] R. F. Werner, All teleportation and dense coding schemes, J. Phys. A, 34 (2001), 7081 -- 7094.
[21] G. Zauner, "Quantendesigns: Grundzauge einer nichtkommutativen Designtheorie", (German) [Quan-
tumdesigns: The foundations of a noncommutative design theory], Ph.D thesis, Universitat Wien,
(available at http://www.mat.univie.ac.at/∼neum/ms/zauner.pdf), 1999.
Ferenc Szollosi: Department of Mathematics and its Applications, Central European
University, H-1051, N´ador u. 9, Budapest, Hungary.
E-mail address: [email protected]
|
1306.1915 | 1 | 1306 | 2013-06-08T13:13:08 | Perturbations of intermediate C*-subalgebras for simple C*-algebras | [
"math.OA"
] | We study uniform perturbations of intermediate C*-subalgebras of inclusions of simple C*-algebras. If a unital simple C*-algebra has a simple C*-subalgebra of finite index, then sufficiently close simple intermediate C*-subalgebras are unitarily equivalent. These C*-subalgebras need not to be nuclear. The unitary can be chosen in the relative commutant algebra. An imediate corollary is the following: If the relative commutant is trivial, then the set of intermediate C*-subagebras is a finite set. | math.OA | math |
Perturbations of intermediate C∗-subalgebras for
simple C∗-algebras
SHOJI INO AND YASUO WATATANI
Abstract. We study uniform perturbations of intermediate C∗-subalgebras of inclusions of
simple C∗-algebras.
If a unital simple C∗-algebra has a simple C∗-subalgebra of finite index,
then sufficiently close simple intermediate C∗-subalgebras are unitarily equivalent. These C∗-
subalgebras need not to be nuclear. The unitary can be chosen in the relative commutant
algebra. An imediate corollary is the following: If the relative commutant is trivial, then the set
of intermediate C∗-subagebras is a finite set.
1. Introduction
The study of the uniform perturbation theory of operator algebras was started with [11] by
Kadison and Kastler in 1972. They introduced a metirc on the set of C∗-subalgebras of a fixed
C∗-algebra by the Hausdorff distance between the unit balls. A conjugacy by a unitary near to
the identity gives close C∗-subalgebras. They conjectured that suitably close C∗-algebras must
be unitarily equivalent. Although Choi and Christensen gave counterexamples to the conjecture
in [2], the conjecture has been verified in various situations. The problem was solved positively
when one algebra is separable and AF in Chiristensen [5] and separable C∗-algebras of continuous
trace by Phillips-Raeburn [16]. Khoshkam [13] showed that sufficiently close nuclear C∗-algebras
have isomorphic K-groups. Recently Christensen, Sinclair, Smith, White and Winter have solved
it completely when one algebra is separable and nuclear in [6].
In this paper we study uniform perturbations of intermediate C∗-subalgebras of inclusions of
simple C∗-algebras.
If a unital simple C∗-algebra has a simple C∗-subalgebra of finite index,
then sufficiently close simple intermediate C∗-subalgebras are unitarily equivalent. Thanks to
Izumi's result in [9], we do not need to assume the existence of the conditional expectations onto
intermediate C∗-subalgebras apriori. Furthermore, if the relative commutant is trivial, then the
set of intermediate C∗-subagebras is a finite set.
In the case of subfactor theory of Jones [10], the lattices of intermediate subfactors with their
finiteness were studied in Popa [17], Watatani [20], Teruya-Watatani [18], Longo [15], Khoshkam-
Mashhood [14], Grossman-Jones [8], Grossman-Izumi [7] and Xu [21] for example.
In these
study, the k · k2-perterbation technique of von Neumann algebras developed by Christensen [4]
are essentially used.
We prepare to consider uniform perturbations of intermediate C∗-subalgebras of inclusions
of not necessarily simple C∗-algebras. But in general, we need to assume the existance of the
conditional expectations onto intermediate C∗-subalgebras to get a positive solution for the
moment.
2. Metric and index for C∗-subalgebras
Let C be a C∗-algebra. We denote by C1 the unit ball of C. We recall a metric on the set of all
subalgebras of a C∗-algebra C after Kadison and Kastler [11]. Let A and B be C∗-subalgebras
of C. Then the distance d(A, B) between A and B is defined by the Hausdorff distance between
1
the unit balls of A and B, that is,
d(A, B) = max(cid:26) sup
a∈A1
inf
b∈B1 ka − bk , sup
b∈B1
a∈A1 kb − ak(cid:27) .
inf
Therefore if d(A, B) < γ, then for each x in the unit ball of either A or B, there exists y in the
unit ball of the other algebra with kx − yk < γ.
We need the following known fact to show that a desired inclusion is onto.
Lemma 2.1. Let A and B be C∗-subalgebras of a C∗-algebra. If A ⊂ B and d(A, B) < 1, then
A = B.
The next proposition records some standard estimates.
Lemma 2.2. Let A be a unital C∗-algebra.
(1) Let x ∈ A satisfy that kx − 1k < 1 and u ∈ A be the unitary in the polar decomposition
x = ux. Then
ku − 1k ≤
√2kx − 1k.
(2) Let p ∈ A be a projection and a ∈ A be self-adjoint. Assme that δ := ka − pk < 1/2. Let
q = χ[1−δ,1+δ](a). Then q is a projection in C ∗(a, I) such that kq − pk ≤ 2ka − pk < 1.
(3) Let p and q be projections in A with kp − qk < 1. Then there exists a unitary w ∈ A
such that
wpw∗ = q,
and
kw − 1k ≤
√2kp − qk.
We recall some basic facts on index for C∗-subalgebas in [19].
Definition 2.3. Let B be a C∗-algebra and A a C∗-subalgebra of B with a common unit. Let
E be a conditional expectation of B onto A. We say that E is of finite index if there exists a
finite set {u1, . . . , uN} ⊂ B, called a (quasi-)basis for E, such that
N
b =
Xi=1
uiE(u∗
i b) ,
for any b ∈ B.
When E is of finite index, then the index of E is defined by
Index E =
N
Xi=1
uiu∗
i .
The value Index E is in the center of B and does not depend on the choice of a basis for E.
We can choose a basis for E in the unit ball of B if it is necessary. In fact, choose a positive
Moreover Index E is positive invertible operator in B. In fact Index E ≥ I.
integer K such that K ≥ max{ku1k, . . . ,kuNk}. Define
v(i−1)K+j =
ui,
for i = 1, 2, . . . , N, j = 1, 2, . . . , K.
1
√K
Then {v1, . . . , vKN} is a desired basis.
tion. Define a A-valued inner product on B by
Next, we recall the C∗-basic construction. Let E : B → A be a faithful conditional expecta-
hx, yiA = E(x∗y) , x, y ∈ B.
2
We denote by E the completion of B. Then E becomes a Hilbert A-module. Let η : B → E
be the natural inclusion map. Thus kη(x)k = kE(x∗x)k1/2. Let LA(E) be the set of bounded
A-module maps on E with adjoints. Let KA(E) be the set of "compact" operators on E.
For b ∈ B, define λ(b) ∈ LA(E) by
λ(b)η(x) = η(bx) , x ∈ B.
Then λ : B → LA(E) turns out to be an injective ∗-homomorphism. Define the Jones projection
eA ∈ LA(E) by
eAη(x) = η(E(x)) , x ∈ B.
The C∗-basic construction C ∗hB, eAi is defined by the closure of the linear span of {λ(x)eAλ(y)x, y ∈
B}.
Moreover λ and eA satisfy the following:
(1) (covariant relation) eAλ(b)eA = λ(E(b))eA for b ∈ B.
(2) Let b ∈ B. Then b is in A if and only if eAλ(b) = λ(b)eA.
Definition 2.4. Let D be a C∗-algebra and C a C∗-subalgebra of D with a common unit.
Assume that the inclusion C ⊂ D has a conditional expectation ED
C : D → C of finte index. We
denote by IMS(C, D, ED
C ) the set of all intermediate C∗-subalgebra A between C and D with
A = ED
a conditional expectation ED
C ,
where EA
C to A.
C ◦ ED
C A : A → C is the coditional expectation defined by the restriction of ED
A : D → A satisfying the compatibility condition EA
C := ED
Let A be in IMS(C, D, ED
If there exists another conditional expectation F D
C ).
satisfying the compatibility condition EA
and a ∈ A, we have that ED
A (ax)) = ED
C (F D
C ◦ F D
C (ax). Hence
A = ED
C , then F D
A = ED
A : D → A
A . In fact for any x ∈ D
ED
C (aF D
A (x)) = ED
C (aED
A (x)).
C (ax) = ED
A (x) − ED
A (x))∗. Since ED
C is of finite index, ED
C
Then ED
is faithful. This implies that F D
A (x) − ED
C (a(F D
A (x))) = 0. Put a = (F D
For any A in IMS(C, D, ED
index. In fact, let {u1, . . . , uN} ⊂ D be a basis of ED
is a basis of EA
C = ED
C A : A → C is of finite
A (ui). Then {v1, . . . , vN} ⊂ D
But we should be careful that an intermediate C∗-subalgebra A between C and D may not
C . If ui is in D1, then vi is in A1.
A = ED
A .
C ), the conditional expectation EA
C . Put vi = ED
have a conditional expectation ED
A : D → A in general.
Example 2.5. Let D = C([0, 1], M2(C)) be the algebra of 2 × 2 matrix valued continuous
functions on the unit interval [0, 1]. Consider a C∗-subalgebra C = C([0, 1], CI) ⊂ D. Then
there exist a conditional expectation ED
C (f ))(x) =
tr(f (x))I for f ∈ D and x ∈ [0, 1], where tr is the normalized trace on M2(C). Define an
intermediate C∗-subalgebra A between C and D by
C : D → C of finte index. In fact, define (ED
A := (cid:26)f ∈ C([0, 1], M2(C)) (cid:12)(cid:12)(cid:12)
f (cid:18) 1
2(cid:19) ∈ CI(cid:27)
Then there exist no conditional expectation of D onto A. In fact, suppose that there were a
conditional expectation ED
A : D → A. We can choose a ∈ C([0, 1], C) such that a(1/2) = 0 and
3
A (cid:18)(cid:18)a 0
A (cid:18)(cid:18)1 0
A (cid:18)(cid:18)1 0
a(x) 6= 0 for any x 6= 1/2. Since (cid:18)a 0
0 0(cid:19)(cid:19) = ED
(cid:18)a 0
0 0(cid:19) = ED
0 0(cid:19) ,(cid:18)a 0
A (cid:18)(cid:18)1 0
0 a(cid:19) ∈ A, we have
0 0(cid:19)(cid:18)a 0
1/2. By the continuity, ED
0 0(cid:19)(cid:19)(cid:18)a 0
0 a(cid:19) ,
0 0(cid:19)(cid:19) = (cid:18)1 0
0 0(cid:19). But (cid:18)1 0
0 0(cid:19)(cid:19) (x) = (cid:18)1 0
where 1 is an identity of C([0, 1], C). This shows that ED
0 a(cid:19)(cid:19) = ED
A (cid:18)(cid:18)1 0
0 0(cid:19) for any x 6=
0 0(cid:19) /∈ A. This is a contradiction.
There is a relation between d(A, B) and the norm estimate keA − eBk of Jones projections
eA and eB for intermediate C∗-subalgebras A, B ∈ IMS(C, D, ED
C ). More precisely, let E be the
completion of D by the C-valued inner procuct ED
C (x∗y). Then E becomes a Hilbert C-module.
Let η : D → E be the natural inclusion map. We can define Jones projections for intermediate
C ∗-subalgebras by eAη(x) = η(ED
B (x)) for x ∈ D. These projections
also enjoy similar properties with a usual Jones projection eC . For example, eA commute with
the left multiplication operator λ(a) for a ∈ A.
Lemma 2.6. Let D be a C∗-algebra and C a C∗-subalgebra of D with a common unit and
ED
C : D → C a conditional expectation of finite index. Then for A, B ∈ IMS(C, D, ED
C ) we have
that
A (x)) and eBη(x) = η(ED
d(A, B) ≤ kIndex ED
C kkeA − eBk.
C k. Then for any a ∈ A1, we have kη(a)k ≤ kak ≤ 1. By the Pimsner-
Proof. Put c = kIndex ED
Popa inequality, E(x∗x) ≥ c−1x∗x of Proposition 2.6.2 in [19],
≥ kη(a − ED
B (a)))k1/2
1
B (a))k1/2 =
cka − ED
keA − eBk ≥ kη(ED
= kED
≥
A (a) − ED
kη(a)k
C ((a − ED
B (a))∗(a − ED
B (a))∗(a − ED
1
ck(a − ED
B (a))k
B (a))k
B (a)k.
Therefore for any a ∈ A1, we can find b := ED
symmmetric argument, we have that d(A, B) ≤ ckeA − eBk.
B (a) ∈ B1 such that ka − bk ≤ ckeA − eBk. By a
(cid:3)
We begin with some elementary estimations.
3. Perturbations
Lemma 3.1. Let A and D be C∗-algebras. Let ϕ : A → D be a contractive positive map and
ψ : A → D be a ∗-homomorphism. Then for any x, y ∈ A
kϕ(xy) − ϕ(x)ϕ(y)k ≤ 3kϕ − ψkkxkkyk.
Proof. Approximate ϕ by ψ.
Lemma 3.2. Let D be a C∗-algebra and A, B be C∗-subalgebras of D. Let EB : D → B be
a conditional expectation. Consider the restriction map EBA : A → D and an inclusion map
ιA : A → D. Then we have
(cid:3)
kEBA − ιAk ≤ 2d(A, B),
4
and for any x, y ∈ A,
kEB(xy) − EB(x)EB(y)k ≤ 6d(A, B)kxkkyk.
Proof. For any ǫ > 0 and a ∈ A1, there exists b ∈ B1 such that ka − bk ≤ d(A, B) + ε/2. Then
kEB(a) − ιA(a)k ≤ kEB(a − b)k + ka − bk ≤ 2d(A, B) + ε.
(cid:3)
Hence kEBA − ιAk ≤ 2d(A, B). The rest is clear.
We shall show that two close intermediate C∗-subalgebras A, B ∈ IMS(C, D, ED
equivalent. We need the following two key lemmas:
Lemma 3.3. Let D be a C∗-algebra and C a C∗-subalgebra of D with a common unit. Let
ED
C : D → C be a conditional expectation of finite index with a basis {u1, . . . , uN} in D1. For
any A, B ∈ IMS(C, D, ED
C ), if d(A, B) < (24N 2)−1, then there exists a unital ∗-homomorphism
ψ : A → B such that ψC = idC and
B A − ψk ≤ 8√3Npd(A, B).
kED
Proof. Let E be the Hilbert B-module completion of D using ED
map from D to E. Define an injective ∗-homomorphism λ : D → LC(E) by
B and η the natural inclusion
C ) are unitarily
λ(d)η(x) = η(dx) , d, x ∈ D.
Then for any b ∈ B, we have λ(b)eB = eBλ(b). The map B ∋ b 7→ λ(b)eB ∈ LC(E) is a injective
∗-homomorphism and kλ(b)eBk = kλ(b)k. For any z ∈ D, we have eBλ(z)eB = ED
B (z)eB. Hence
for any x, y ∈ D,
B (x)ED
B (y)k = kλ(ED
B (xy) − ED
kED
= kλ(ED
= keBλ(xy)eB − eBλ(x)eBλ(y)eBk = keBλ(x)(I − eB)λ(y)eBk.
B (xy) − ED
B (y))eBk
B (xy))eB − λ(ED
B (x))eBλ(ED
B (y))eBk
B (x)ED
C be a restriction of ED
Let EA
in A1. Define an operator
C to A and put vi = ED
A (ui). Then {v1, . . . , vN} is a basis for EA
C
t =
N
Xi=1
λ((Index EA
C )−1)λ(vi)eBλ(v∗
i ) ∈ LC(E).
Recall that Index EA
C is in the center of A. For any a ∈ A
λ(a)t =
=
=
=
N
Xi=1
Xi=1
N
N
Xj=1
N
Xj=1
λ((Index EA
C )−1)λ(avi)eBλ(v∗
i )
N
Xj=1
λ((Index EA
C )−1)λ(vj EA
C (v∗
j avi))eBλ(v∗
i )
λ((Index EA
C )−1)λ(vj)eB
N
Xi=1
λ(EA
C (v∗
j avi)v∗
i )
λ((Index EA
C )−1)λ(vj)eBλ(v∗
j )λ(a) = tλ(a).
5
Since Index EA
N
t −
N
i )eB(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
C )−1)λ(vi)λ(v∗
λ((Index EA
i and Lemma 3.2,
C = Pi viv∗
Xi=1
kλ((Index EA
i ) − λ(v∗
kt − eBk = (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
Xi=1
≤
≤ NkeBλ(v∗
= Nk(I − eB)λ(v∗
= N · max {k(I − eB)λ(v∗
= N · maxnkeBλ(vi)(I − eB)λ(v∗
= N · maxnkED
B (vi)ED
≤ Np6d(A, B) <
i ) − ED
.
i )eB − eBλ(v∗
B (viv∗
C )−1)kkλ(vi)kkeBλ(v∗
i )eBk
i )(I − eB)k
i )eBk , keBλ(v∗
1
2
i ) − λ(v∗
i )eBk
i )(I − eB)k}
i )(I − eB)λ(vi)eBk1/2o
i )eBk1/2 , keBλ(v∗
i vi) − ED
i )k1/2, kED
B (v∗
B (v∗
i )ED
B (v∗
B (vi)k1/2o
Put δ = kt − eBk = k(t + t∗)/2 − eBk. Let q = χ[1−δ,1+δ]((t + t∗)/2). By Lemma 2.2, q is a
projection in C ∗(t, IE ) and commutes with λ(a) for any a ∈ A, and kq − eBk ≤ 2kt − eBk < 1.
Therefore there exists a unitary w ∈ C ∗(t, eB, IE ) ⊂ LB(E) such that weBw∗ = q and kw−1Ek ≤
√2kq − eBk. Define ψ′ : A → LB(E) by
ψ′(a) = w∗qλ(a)qw = eBw∗λ(a)weB , a ∈ A.
Since the projection q commutes with λ(a), it is clear that ψ′ is a ∗-homomorphism.
The unitary w is in C ∗(t, eB, IE ) ⊂ C ∗(λ(A), eB, IE ) , and eBw∗λ(a)weB ∈ λ(B)eB for a ∈ A.
Therefore, ψ′(A) ⊂ λ(B)eB ⊂ LB(E). Let E ′ be a closure of η(B) in E. Define an injective
∗-homomorphism λ′ : B → LB(E ′) by
and a surjective ∗-isomorphism ι : λ(B)eB → λ′(B) by
λ′(b)η(x) = η(bx) , b, x ∈ B,
ι(λ(b)eB) = λ′(b) , b ∈ B.
Then λ(B)eB is isomorphic to λ′(B). Thus, we can define a ∗-homomorphism ψ = (λ′)−1◦ι◦ψ′ :
A → B, that is, λ(ψ(a))eB = ψ′(a). Then for any contraction a ∈ A,
B A(a) − ψ(a)))eBk = keBλ(ED
B A(a) − ψ(a)k = keB(λ(ED
B (a))eB − ψ′(a)k
kED
Therefore
c ∈ C. Hence
= keBλ(a)eB − eBw∗λ(a)weBk
= keBλ(a)(I − w∗)eB + eB(I − w∗)λ(a)weBk ≤ 2kw − Ik.
B A − ψk ≤ 2kw − Ik ≤ 8√3Npd(A, B).
kED
Since w is contained in C ∗(t, eB, IE ), C ⊂ A and C ⊂ B, we have that λ(c)w = wλ(c) for
ψ′(c) = eBw∗λ(c)weB = eBλ(c)eB = λ(c)eB
Therefore ψ(c) = c = idC (c) for any c ∈ C.
Next Lemma shows how to find an intertwiner for close ∗-homomorphisms.
6
(cid:3)
Lemma 3.4. Let D be a C∗-algebra and C a C∗-subalgebra of D with a common unit and
ED
C : D → C be a conditional expectation of finite index with a basis {u1, . . . , uN} in D1. For
any A ∈ IMS(C, D, ED
C ), If φ1, φ2 : A → D are unital ∗-homomorphisms such that φ1C = idC =
φ2C and kφ1 − φ2k < 1/N , then there exists a unitary u ∈ D such that φ1 = Ad(u) ◦ φ2 and
ku − IDk < √2Nkφ1 − φ2k.
Proof. Let EA
C be the estriction of ED
C to A and vi = ED
A (ui). Put
s =
N
Xi=1
Since k(Index EA
C )−1k ≤ 1,
φ1((Index EA
C )−1)φ1(vi)φ2(v∗
i ).
ks − IDk = (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
N
N
s −
Xi=1
Xi=1
kφ1((Index EA
≤
≤ Nkφ1 − φ2k < 1.
φ1((Index EA
C )−1)φ1(vi)φ1(v∗
i )(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)
C )−1)φ1(vi)kkφ2(v∗
i ) − φ1(v∗
i )k
Therefore the unitary u in the polar decomposition s = us lies in D and satisfies ku − IDk ≤
√2Nkφ1 − φ2k by Lemma 2.2. Furthermore, for any a ∈ A,
C )−1)φ1(avi)φ2(v∗
i )
φ1((Index EA
φ1(a)s =
N
Xi=1
Xi=1
N
=
=
=
=
N
Xj=1
φ1((Index EA
C )−1)φ1(vjEA
C (v∗
j avi))φ2(v∗
i )
N
N
Xi=1
Xj=1
N
N
Xj=1
Xi=1
φ1((Index EA
C )−1)φ1(vj)EA
C (v∗
j avi)φ2(v∗
i )
φ1((Index EA
C )−1)φ1(vj)φ2(EA
C (v∗
j avi)v∗
i )
N
Xj=1
φ1((Index EA
C )−1)φ1(vj)φ2(v∗
j a) = sφ2(a).
Taking adjoints gives
Therefore,
s∗φ1(a) = φ2(a)s∗ , a ∈ A.
Since sφ2(a) = φ2(a)s, φ1(a)u = uφ2(a). Therefore we have that φ1 = Ad(u) ◦ φ2.
s∗sφ2(a) = s∗φ1(a)s = φ2(a)s∗s , a ∈ A.
(cid:3)
The following proposition shows that sufficiently close intermediate subalgebras with condi-
tional expectations are unitarily equvalent.
7
Proposition 3.5. Let D be a C∗-algebra and C a C∗-subalgebra of D with a common unit and
ED
C : D → C a conditional expectation of finite index. Then, there exists a positive constant
γ satisfying the following: For any A, B ∈ IMS(C, D, ED
C ), if d(A, B) < γ, then there exists a
unitary u ∈ C ∗(A, B) such that uAu∗ = B. We can choose the unitary in the relative commutant
C ′ ∩ D.
Proof. Let N be the number of a finite basis for ED
there exists a unital ∗-homomorphism ψ : A → B such that ψC = idC and
B A − ψk ≤ 8√3Npd(A, B) < 8√3(100N )−1.
kED
B Ak + kED
B A − idAk < 8√3(100N )−1 + 2(10N )−4 <
C in D1. Put γ := (10N )−4. By Lemma 3.3,
kψ − idAk ≤ kψ − ED
Since
1
N
by Lemma 3.2, there exists a unitary u ∈ C ∗(A, B) such that ψ = Ad(u) and ku − 1Dk ≤
√2N (8√3(100N )−1 + 2(10N )−4) by Lemma 3.4. That is uAu∗ ⊂ B. For any b ∈ B1, there
exists a ∈ A1 such that kb − ak ≤ γ by d(A, B) < γ. Then
kb − uau∗k ≤ kb − ak + ka − uau∗k
≤ γ + 2kakku − 1Dk
≤ (10N )−4 + 2√2N (8√3(100N )−1 + 2(10N )−4)
≤ (4√2 + 1)(10)−4 + 16√6(10)−2 < 1;
therefore, d(uAu∗, B) < 1. By Lemma 2.1, we have that uAu∗ = B.
(cid:3)
The following is the main theorem of the paper. Thanks to Izumi's result in [9], we do not
need to assume the existence of the conditional expectations onto intermediate C∗-subalgebras
apriori. We also need the notion minimality of conditional expecations in [12]. Let D be a
simple C∗-algebra and C a simple C∗-subalgebra of D with a common unit and ED
C : D → C
a conditional expectation of finite index. Then there there exists a unique minimal conditional
expectation E0 : D → C, that is, Index E0 ≤ Index E for any conditional expectation E : D → C
of finite index.
Theorem 3.6. Let D be a simple C∗-algebra and C a simple C∗-subalgebra of D with a common
unit and ED
C : D → C a conditional expectation of finite index. Then, there exists a positive
constant γ satisfying the following: For any simple intermediate C∗-subalgebras A and B for
C ⊂ D, if d(A, B) < γ, then there exists a unitary u ∈ C ∗(A, B) such that uAu∗ = B. We can
choose the unitary in the relative commutant C ′ ∩ D.
Proof. We may assume that ED
C : D → C is a minimal conditional expectation of finite index
by replacing the original conditional expectation by the minimal one, if necesary. Let A be a
simple interemediate C∗-subalgebra such that C ⊂ A ⊂ D. Since ED
C satisfies Pimsner-Popa
inequality, so does the restriction EA
C : A → C. By Corollary 3.4 in Izumi [9], EA
C also has a
finite basis. Let {u1,··· , uN} be a basis for EA
C . Moreover there exists a conditional expectation
ED
A : D → A of finite index by Proposition 6.1 in Izumi [9]. In fact, we can define ED
A by
ED
A (x) = (cid:0)Index EA
C(cid:1)
−1
N
uiED
C (u∗
i xuj)u∗
j , x ∈ D.
Xi,j=1
8
C and ED
A by minimal conditional expectations. Then the conposition EA
Replace EA
A is
also a minimal conditional expectation by Theorem 3 in [12]. Since the minimal conditional
expectation is unique, EA
C . Hence we may assume that the conditional expectation
ED
(cid:3)
Remark 3.7. Even if we do not assume that an intermediate C∗-subalgebra A is not simple,
we can prove a similar fact. But the constant γ above depends on the choice of A:
A satisfies the compatibility condition. Therefore Proposition 3.5 can be applied.
C ◦ ED
C ◦ ED
A = ED
Let D be a simple C∗-algebra and C a simple C∗-subalgebra of D with a common unit and
ED
C : D → C a conditional expectation of finite index. For any intermediate C∗-subalgebras
A, there exists a positive constant γ satisfying the following: For any otheir intermediate C∗-
subalgebras B for C ⊂ D, if d(A, B) < γ, then there exists a unitary u ∈ C ∗(A, B) such that
uAu∗ = B. We can choose the unitary in the relative commutant C ′ ∩ D. In fact, all we need
is that the existance of the conditinal expectation ED
A and a finite basis for EA
C .
In the case of subfactor theory of Jones [10], the lattices of intermediate subfactors with their
finiteness were studied in Popa [17], Watatani [20], Teruya-Watatani [18], Longo [15], Khoshkam-
Mashhood [14], Grossman-Jones [8], Grossman-Izumi [7] and Xu [21] for example.
In these
study, the k · k2-perturbation technique of von Neumann algebras developed by Christensen [4]
are essentially used or motivated.
Since we can choose the implementing unitary u in the relative commutant, we immediately
get the following finiteness of the intermediate subalgebras for both simple C∗-algebras and
factors. A bound of the number is obtained as in Longo [15].
Corollary 3.8. Let D be a simple C∗-algebra and C a simple C∗-subalgebra of D with a common
unit and ED
C : D → C a conditional expectation of finite index. If the relative commutant C ′∩ D
is trivial, then the number of intermediate C∗-subalgebras is finite.
Proof. Since ED
C : D → C is a conditional expectation of finite index , there exists a dual
conditional expectation of finite index ED : C ∗hD, eCi → D. Thus, ED ◦ ED
C : C ∗hD, eCi → C
is a conditional expectation of finite index. Since the commutant C ′ ∩ C is finite-dimensional,
the relative commutant C ′ ∩ C ∗hD, eCi is also finite-dimensional by Proposition 2.7.3 in [19].
Therefore the set
P := {p ∈ C ′ ∩ C ∗hD, eCi p is a projection }
is a compact Hausdorff space with respect to the operator norm topology. Let N be the number
C k)−1. By the compactness, there
of a finite basis for ED
exists a finite open covering by ε-open balls. For any intermediate C∗-subalgebras A and B,
their Jones projection eA and eB are in C ′ ∩ C ∗hD, eCi and satisfies
C in D1. Let ε = (2(10N )4kIndex ED
d(A, B) ≤ kIndex ED
C kkeA − eBk
by Lemma 2.6. If two Jones projection eA and eB are in one of these ε-open balls, then d(A, B) <
(10N )−4. By Theorem 3.6, there exists a unitary u in C ′ ∩ D such that B = uAu∗. Since
C ′ ∩ D = CI, u is a scalar. Therefore B = A. This shows that each ε-open ball of the cover
contains at most one Jones projection for some intermediate C∗-subalgebra. This completes the
proof.
(cid:3)
Since any subfactor of a type II1 factor has a conditional expectation, we can also apply the
same method in this case.
Corollary 3.9. Let M be a type II1 factor and N a subfactor of finite index. If the relative
commutant N ′ ∩ M is trivial, then the set of intermediate subfactors is a finite set.
9
References
[1] D. Bisch, A note on intermediate subfactors, Pacific J. Math., 163 (1994), 201-216.
[2] M. D. Choi and E. Christensen, Completely order isomorphic and close C ∗-algebras need not be ∗-isomorphic,
Bull. London Math. Soc., 15 (1983), 604-610.
[3] E. Christensen, Perturbations of type I von Neumann algebras, J. London Math. Soc., 9 (1974/75), 395-405.
[4] E. Christensen, Subalgebras of a finite algebras, Math. Ann., 243 (1979), 17-29.
[5] E. Christensen, Near inclusions of C ∗-algebras, Acta. Math., 144 (1980), 249-265.
[6] E. Christensen, A. M. Sinclair, R. R. Smith, S. A. White and W. Winter, Perturbations of nuclear C ∗-
algebras, Acta. Math., 208 (2012), 93-150.
[7] P. Grossman and M. Izumi, Classification of noncommuting quadrilaterals of factors, Internat. J. Math., 19
(2008), 557-643.
[8] P. Grossman and V. Jones, Intermediate subfactors with no extra structure, J. Amer. Math. Soc., 20 (2007),
219-265.
[9] M. Izumi, Inclusions of simple C ∗-algebras, J. Reine. Angew. Math., 547 (2002), 97-138.
[10] V. Jones, Index for subfactors, Inv. Math. 72 (1983), 1-25.
[11] R. V. Kadison and D. Kastler, Perturbations of von Neumann algebras. I. Stability of type, Amer. J. Math.,
94 (1972), 38-54.
[12] S. Kawakami and Y. Watatani, The multiplicativity of the minimal index of simple C ∗-algebras, Proc. Amer.
Math. Soc., 123 (1995), 2809-2813.
[13] M. Khoshkam, Perturbations of C ∗-algebras and K-theory, J. Operator Theory, 12 (1984), 89-99.
[14] M. Khoshkam and B. Mashood, On finiteness of the set of intermediate subfactors, Proc. Amer. Math., 132
(2004), 2939-2944.
[15] R. Longo, Conformal subnets and intermediate subfactors, Comm. Math. Phys., 237 (2003), 7-30.
[16] J. Phillips and I. Raeburn, Perturbations of C ∗-algebras II., Proc. London Math. Soc., 43 (1981), 46-72.
[17] S. Popa, Correspondence, preprint.
[18] T. Teruya and Y. Watatani, Lattices of intermediate subfactors for type III factors, Arch. Math., 68 (1997),
454-463.
[19] Y. Watatani, Index for C ∗-subalgebras, Mem. Amer. Math. Soc., 424 (1990).
[20] Y. Watatani, Lattices of intermediate subfactors, J. Func. Anal., 140 (1996), 312-334.
[21] F. Xu, On representing some lattices as lattices of intermediate subfactors of finite index, Adv. Math. 220
(2009), 1317-1356.
(Shoji Ino) Department of Mathematical Sciences, Kyushu University, Motooka, Fukuoka, 819-
0395, Japan.
E-mail address: [email protected]
(Yasuo Watatani) Department of Mathematical Sciences, Kyushu University, Motooka, Fukuoka,
819-0395, Japan.
E-mail address: [email protected]
10
|
1807.11020 | 2 | 1807 | 2018-08-20T14:30:42 | On the C*-algebra of matrix-finite bounded operators | [
"math.OA",
"math.FA"
] | Let $H$ be a separable Hilbert space with a fixed orthonormal basis. Let $\mathbb B^{(k)}(H)$ denote the set of operators, whose matrices have no more than $k$ non-zero entries in each line and in each column. The closure of the union (over $k\in\mathbb N$) of $\mathbb B^{(k)}(H)$ is a C*-algebra. We study some properties of this C*-algebra. We show that this C*-algebra is not an AW*-algebra, has a proper closed ideal greater than compact operators, and its group of invertibles is contractible. | math.OA | math |
ON THE C ∗-ALGEBRA OF MATRIX-FINITE BOUNDED OPERATORS
V. MANUILOV
Abstract. Let H be a separable Hilbert space with a fixed orthonormal basis. Let
B(k)(H) denote the set of operators, whose matrices have no more than k non-zero entries
in each line and in each column. The closure of the union (over k ∈ N) of B(k)(H) is a
C ∗-algebra. We study some properties of this C ∗-algebra. We show that this C ∗-algebra
is not an AW ∗-algebra, has a proper closed ideal greater than compact operators, and
its group of invertibles is contractible.
Introduction
Hilbert spaces often come out with a fixed special orthonormal basis, e.g. l2(X) for a
countable discrete space X, and the correspondence between operators and their matrices
makes more sense. We got interested in the operators, whose matrices with respect to
the fixed basis satisfy the following property: there is some k ∈ N such that each line and
each column of the matrix has no more than k non-zero entries. The norm closure of the
union (over k ∈ N) of these operators turns out to be a C ∗-algebra, denoted by Bf (H).
We establish some properties of this C ∗-algebra, which is, in some aspects, similar to the
algebra B(H) of all bounded operators on the Hilbert space H, e.g. the Kuiper theorem
on contractibility of the group of invertibles holds also for invertibles in Bf (H). On the
other hand, Bf (H) is not a von Neumann algebra, and even not an AW ∗-algebra, and
the quotient Bf (H)/K(H) is not simple.
1. Matrix-finite bounded operators on a Hilbert space
Let H be a separable Hilbert space with a fixed orthonormal basis {en}n∈N. For k ∈ N,
denote by B(k)
C (H)) the set of all bounded operators on H such that each
line (resp. each column) of their matrix (with respect to the fixed basis) contains no more
than k non-zero entries. Note that
L (H) (resp. B(k)
B(k)
L (H) ⊂ B(l)
L (H) when k < l, and B(k)
C (H) = (B(k)
L (H))∗.
Set also
B(k)(H) = B(k)
L (H) ∩ B(k)
Let a, b ∈ B(H), A = (aij), B = (bij) their matrices.
L (H).
L (H) then a + b ∈ B(2k)
Lemma 1.1. If a, b ∈ B(k)
Proof. Obvious.
C (H).
L (H) then ab ∈ B(k2)
Lemma 1.2. If a, b ∈ B(k)
Proof. Let cil = Pj∈N aijbjl. Fix i. There exist j1, . . . , jk ∈ N such that aij = 0 if
j /∈ {j1, . . . , jk}. For each jm, m = 1, . . . , k, there exist l(m)
∈ N such that
bjml = 0 if l /∈ {l(m)
n,m=1, hence the i-th line contains
no more than k2 non-zero entries.
k }. So cil = 0 for l /∈ {l(m)
n }k
, . . . , l(m)
, . . . , l(m)
L (H).
k
1
1
1
(cid:3)
(cid:3)
2
V. MANUILOV
Set B(∞)(H) = ∪k∈NB(k)(H). This algebra was defined and studied in [5], Abadie-
Cortinas.
Corollary 1.3. B(∞)(H) is a unital ∗-subalgebra in B(H).
Note that the correspondence between operators and their matrices is more clear for
operators in B(∞)(H).
Lemma 1.4. Let (aij)i,j∈N be a matrix such that each line and each column contains no
more than k non-zero entries, and there exists C such that aij < C for any i, j ∈ N.
Then the operator a given by this matrix is bounded.
Proof. Let the non-zero elements in the i-th line are ai,j(i)
a(1), . . . , a(k) by
1
, . . . , ai,j(i)
k
. Define matrices
Then each matrix a(m) has a single non-zero element in each line, and a = a(1) +· · · + a(k).
(a(m))ik =( ai,j(i)
Let ξ = (ξn)n∈N ∈ H. Then a(m)ξ = (a1,j(1)
ka(m)ξk2 = a1,j(1)
m 2 + a2,j(2)
ξj(1)
, a2,j(2)
m 2 + · · · < C 2(ξj(1)
ξj(2)
0
ξj(1)
m
m
m
m
m
if k = j(i)
m
if k 6= j(i)
m .
ξj(2)
m
m
, . . .), hence
m 2 + ξj(2)
m 2 + · · · ) ≤ C 2kXj∈N
ξj2,
(cid:3)
(cid:3)
as each j appears not more than k times among the numbers j(1)
m , . . ..
Thus ka(m)k2 < C 2k. As kak ≤ ka(1)k + · · · + ka(k)k, we get kak < Ck3/2.
m , j(2)
Denote by Bf (H) the norm closure of B(∞)(H). Then it is a C ∗-algebra.
Lemma 1.5. Mm(Bf (H)) ∼= Bf (H) for any m ∈ N.
Proof. Obvious.
Let K(H) denote the C ∗-algebra of compact operators on H.
Lemma 1.6. K(H) ⊂ Bf (H).
Proof. Any compact operator can be approximated by operators with only finite number
of non-zero entries.
(cid:3)
The following examples show that the C ∗-algebra Bf (H) is smaller than B(H), but not
2. Difference from B(H)
too much smaller.
Proposition 2.1. Define the vectors an ∈ H, n ∈ N, by
a1 = (1, 0, 0, . . .); a2 = (0,
, 0, 0, . . .); a3 = (0, 0, 0,
1
√2
,
1
√2
1
√3
,
1
√3
,
1
√3
, 0, 0, . . .); . . .
Define an isometry v by v(en) = an. Then v /∈ Bf (H).
Proof. If v ∈ Bf (H) then for any ε > 0 there exists k ∈ N and b ∈ B(k)
f (H) such that
kv − bk < ε. In particular, k(v − b)enk < ε for any n ∈ N. The vector ben has no more
than k non-zero coordinates, hence kan − benk2 ≥ n−k
2 and n > 2k, we get
a contradiction.
n . Taking ε < 1
(cid:3)
ON THE C ∗-ALGEBRA OF MATRIX-FINITE BOUNDED OPERATORS
3
The same argument can be used in the next example:
Proposition 2.2. Let L ⊂ H be the closed subspace spanned by a1, a2, . . ., and let b1, b2, . . .
be an orthonormal basis for L⊥. Set u(e2n−1) = an, u(e2n) = bn. Then u is a unitary,
and u /∈ Bf (H).
Proposition 2.3. Let H = ⊕n∈NCn with the basis consisting of the standard bases of Cn,
n ∈ N. Let pn = 1
n! denote the projection onto the line (x, . . . , x), x ∈ C, in Cn.
...
...
Set p = ⊕n∈Npn. Then p ∈ Bf (H). Note that we have p = vv∗, where v ∈ B(H) \ Bf (H)
is the isometry from Proposition 2.1.
...
1
n ···
n ···
1
n
1
Proof. The proof uses expander graphs [8] and essentially is contained in [6], where it is
shown that the direct sum of one-dimensional projections ⊕n∈Npmn lies in the uniform Roe
algebra (hence in Bf (H), cf. Lemma 5.3 below) for certain increasing sequences {mn} of
sizes. We only add a remark that one may take mn = n.
Let X = (V, E) be a d-regular graph, without loops and multiple edges, with the set V
of vertices and the set E of edges. Let V = m. The m-dimensional Hilbert space l2(V )
is endowed with the standard basis consisting of characteristic functions of the vertices.
The Laplacian on l2(V ) is the positive operator L given by the matrix (Lv,w)v,w∈V , where
Set H = ⊕n∈Nl2(Vn) with the basis obtained by uniting the bases of each l2(Vn), and
let L = ⊕n∈NLn.
Note that each Ln has no more than d + 1 non-zero entries in each line and in each
column. By Lemma 1.1 and Lemma 1.2, we conclude that for any s ∈ N there exists
k = k(s) ∈ N such that each line and each column of fs(Ln) contains no more than k
non-zero entries. Therefore, fs(L) = ⊕n∈NLn ∈ B(k)(H). As the spectrum of L lies in
{0}∪ [δ, 2], the operators fs(L) converge to the spectral projection p′ = ⊕n∈Np2n onto the
kernel of L:
s→∞kfs(L) − p′k = 0,
lim
hence p′ ∈ Bf (H).
Similarly, p′′ = p1 ⊕ (⊕n∈N(p2n ⊕ p2n ⊕ 0)) ∈ Bf (H). Note that
(cid:13)(cid:13)(cid:13)(cid:13)
0(cid:19)(cid:13)(cid:13)(cid:13)(cid:13)
p2n+1 −(cid:18)p2n 0
2n + 1(cid:18)−p2n
1(cid:19)(cid:13)(cid:13)(cid:13)(cid:13)
=(cid:13)(cid:13)(cid:13)(cid:13)
2n + 1
≤
l∗
0
1
l
1
(kp2nk + 2klk + 1) =
2 + 2√2n
2n + 1
,
if v = w;
if v, w are adjacent;
1,
− 1
d,
0, otherwise.
Lv,w =
It is known that kLk ≤ 2 for any m ∈ N. It is also easy to see that the matrix L has no
more than d + 1 non-zero entries in each line and in each column for any m ∈ N.
The smallest eigenvalue λ0 of L is zero, and the corresponding eigenfunctions are con-
stants, so the projection onto the kernel of L is the spectral projection onto this eigenspace,
and its matrix equals pm.
2 -expander graph Xn = (Vn, En) with
Vn = 2n ([8], Proposition 1.2.1), and it follows ([8], Section 4.2) that the second smallest
eigenvalue λ1 of each Laplacian Ln of Xn satisfies λ1 ≥ δ for some δ > 0.
If d ≥ 5 then for each n ∈ N there exists a 1
Let {fs}s∈N be a sequence of polynomials such that
• fs(0) = 1 for any s ∈ N;
• fs(t) ≤ 1
s for any t ∈ [δ, 2].
4
V. MANUILOV
where l is the column of 2n units. So, p − p′′ ∈ K(H), and, by Lemma 1.6, p ∈ Bf (H).
(cid:3)
Note that, by spectral theorem, any selfadjoint a ∈ B(H) can be approximated by linear
combinations of projections, so if all projections in H would lie in Bf (H) then a ∈ Bf (H),
hence any b ∈ B(H) must lie in Bf (H), which is false. Therefore, there exists a projection
in B(H) \ Bf (H). Here is an explicit example.
Proposition 2.4. Let u be the unitary from Proposition 2.2 on a Hilbert space H, and
let p = 1
the two copies of H). Then p /∈ Bf (H ⊕ H).
Proof. This follows from Lemma 1.5 and from u /∈ Bf (H).
2(cid:18) 1 u
u∗ 1(cid:19) be a projection in H ⊕ H (with the basis obtained by uniting bases of
(cid:3)
It is known that K(H) is the unique closed two-sided ideal in B(H). This is not true
for Bf (H).
3. Ideals
Let a ∈ Bf (H), (aij)i,j∈N the matrix of a (recall that the basis is fixed). Set
I(H) = {a ∈ Bf (H) : lim
n→∞
sup
i,j≥naij = 0}.
Obviously, K(H) ⊂ I(H). It is easy to see that the projection p from Proposition 2.3 lies
in I(H) and is not compact, hence I(H) is strictly bigger than K(H).
A similar ideal in uniform Roe algebras was studied, e.g. in [11], under the name of the
ideal of ghost operators.
Theorem 3.1. I(H) is the maximal closed two-sided proper ideal in Bf (H).
Indeed, suppose that (a(ι))ι∈I converges to a ∈
Proof. First, note that I(H) is closed.
Bf (H), and a(ι) ∈ I(H) for any ι ∈ I. Fix ε > 0, then there is κ ∈ I such that
ka(ι) − ak < ε for any ι ≥ κ. Therefore, a(ι)
ij − aij < ε for any i, j ∈ N. Then
aij ≤ a(κ)
ij + aij − a(κ)
ij < a(κ)
ij + ε.
As a(κ) ∈ I(H), there exists N such that a(κ)
any i, j > N.
ij < ε for any i, j > N. Then aij < 2ε for
Now let a ∈ I(H), b ∈ B(k)(H) for some k. Let us check that ab ∈ I(H). Due to the
involution, this would imply ba ∈ I(H). Set cil = Pj∈N aijbjl. Take ε > 0 and find N
such that aij < ε for i, j > N. After fixing N, we can find M such that bjl = 0 for any
j ≤ N and any l > M (see Lemma 1.2). Then, for i > N and l > M we have
< εkkbk
cil =(cid:12)(cid:12)(cid:12)Xj∈N
aijbjl(cid:12)(cid:12)(cid:12)
=(cid:12)(cid:12)(cid:12)Xj>N
aijbjl(cid:12)(cid:12)(cid:12)
(we use here that the sum contains not more than k non-zero summands and that bjl ≤
kbk for all j, l ∈ N). Thus I(H) is a two-sided ideal in Bf (H).
It remains to show that I(H) is the maximal proper closed two-sided ideal in Bf (H).
Assume that there exists an ideal J and a selfadjoint a ∈ J \ I(H). We are going to
construct a subspace L ⊂ H and an isometry u : H → L in two different cases. First,
consider the case when there exists a sequence {in}n∈N such that kain,ink > δ for some
δ > 0. As B(∞)(H) is dense in Bf (H), there exists k ∈ N and a selfadjoint a(k) ∈ B(k)(H)
ON THE C ∗-ALGEBRA OF MATRIX-FINITE BOUNDED OPERATORS
5
such that ka − a(k)k < δ/4. Then a(k)
in,in > 3δ/4. Let us pass to a subsequence {inm}m∈N
inductively: set n1 = 1, and if mn is already fixed then let nm+1 be defined by the
condition
in1 ,l, . . . , a(k)
inm+1 > l for any l ∈ N such that at least one of a(k)
(for each n, there is not more than k non-zero entries a(k)
in,j). Then let L ⊂ H be a closed
subspace spanned by the vectors einm , m ∈ N, and let pL ∈ Bf (H) be the projection onto
L. By (1), pLa(k)L is a diagonal operator:
= 0 if m 6= m′,
inm ,inm > 3δ/4.
and a(k)
inm ,l is non-zero
a(k)
inm ,in
(1)
m′
Then kpLa(k)ξk > 3δ
4 kξk and
kpLaξk ≥ kpLa(k)ξk − kpL(a(k) − a)ξk >
3δ
4 kξk −
δ
4kξk =
δ
2kξk
for any ξ ∈ L. Let u : H → L be the isometry defined by u(em) = einm . Then u ∈ Bf (H).
Otherwise, assume that limi→∞ aii = 0. As a /∈ I(H), there exists some δ > 0 and
sequences {in},{jn}n∈N such that in+1, jn+1 > max(in, jn), jn 6= in, and ain,jn > δ for
any n ∈ N. As in the first case, find a selfadjoint a(k) ∈ B(k)(H) such that ka−a(k)k < δ/6.
Then in the two-by-two matrices
a(k)
in,in a(k)
jn,in a(k)
a(k)
jn,jn!
in,jn
(2)
the diagonal entries satisfy a(k)
i,i < δ/6 for sufficiently great n, while the off-diagonal
entries have modulus greater than 5δ/6, hence there exists some n0 ∈ N such that the
matrices (2) are bounded from below by 2δ/3 for n > n0.
Once again let us pass inductively to a subsequence {nm}m∈N. Set n1 ≥ n0, and if nm
is already fixed then let nm+1 be defined by the condition
min(inm+1, jnm+1) > l
in1 ,l, . . . , a(k)
for any l ∈ N such that at least one of a(k)
Let L be a closed subspace spanned by einm , ejnm , m ∈ N. Then pLa(k) is a block-
3 for
diagonal operator with two-dimensional blocks of the form (2), hence kpLa(k)ξk > 2δ
any ξ ∈ L. Then
jn1 ,l, . . . , a(k)
jnm ,l is non-zero.
inm ,l, a(k)
kpLaξk ≥ kpLa(k)ξk − kpL(a(k) − a)ξk >
2δ
3 kξk −
δ
6kξk =
δ
2kξk
for any ξ ∈ L. Similarly to the first case, let u : H → L be the isometry defined by
u(e2m−1) = einm , u(e2m) = ejnm . Again we have u ∈ Bf (H).
The further argument is standard [4]. In both cases, for any ξ ∈ H we have ku∗pLauξk >
2kξk for any ξ ∈ H, hence u∗pLau is invertible, therefore, 1 ∈ J and J = Bf (H).
δ
(cid:3)
4. Similar C ∗-algebras
We know three C ∗-algebras, which seem similar to Bf (H). We don't know if they all
are the same or different, but it is easy to see that Bf (H) lies in all of them.
In [12] the following C ∗-subalgebra of B(H) was defined. Let a ∈ B(H), (aij)i,j∈N its
matrix. The operator a is called l1-bounded if there is m ∈ (0,∞) such thatP∞
i=1 aij < m
for any j ∈ N and P∞
j=1 aij < m for any i ∈ N. It is shown in [12] that l1-bounded
6
V. MANUILOV
operators form a ∗-algebra, and its norm-closure is a C ∗-algebra. Let us denote it by
BT (H).
Lemma 4.1. BF (H) ⊂ BT (H).
Proof. Let a ∈ B(k)(H). Then P∞
j=1 aij ≤ kkak.
(cid:3)
Another C ∗-subalgebra of B(H) can be defined as follows.
Let ω be a free ultrafilter on N. Set
H = {ξ = (ξn)n∈N : sup
n kξnk < ∞},
hξ, ηi = lim
ω hξn, ηni,
ξ, η ∈ H,
Then H ω is a (non-separable) Hilbert space, and B(H) is obviously represented on H ω by
H0 = {ξ ∈ H : hξ, ξi = 0}, H ω = H/H0.
π(a)ξ = (aξn)n∈N,
π : B(H) → B(H ω).
As H ω contains a copy of H consisting of constant sequences, π is injective.
For k ∈ N let H(k) denote the set of all sequences ξ ∈ H such that not more than k
entries are non-zero. Although each H(k) is not a linear subspace, their union
H(∞) = ∪k∈NH(k)
is. Then H(∞)/H0 is a linear subspace of H ω. Denote its closure by H ω
subspace of H ω.
Lemma 4.2. Let a ∈ B(H). If a ∈ Bf (H) then H ω
Proof. Let a ∈ B(k)(H), ξ = (ξn)n∈N ∈ H(l). Then aξn has not more than kl non-zero
coordinates, hence (aξn)n∈N ∈ H(kl), hence π(a) maps H(k) +H0 to H(kl) +H0, thus leaves
H(∞)/H0 invariant.
f is invariant for π(a).
f . It is a proper
(cid:3)
B(H).
Set Bω(H) = {a ∈ B(H) : π(a)(H ω
f }. It is a C ∗-algebra, and Bf (H) ⊂ Bω(H) ⊂
One more C ∗-algebra is the multiplier C ∗-algebra M(I(H)) of the ideal I(H). As
f ) ⊂ H ω
It is easy to see that the isometry from Proposition 2.1 lies in neither of the three
K(H) ⊂ I(H) ⊂ B(H), we have M(I(H)) ⊂ B(H).
C∗-algebras.
Problem 4.3. Are the C ∗-algebras Bf (H), BT (H), Bω(H), M(I(H)) different?
5. Embeddings
We do not know if any separable C ∗-algebra A admits an injective ∗-homomorphism
into Bf (H), but here are four easy cases when the answer is positive.
Lemma 5.1. Let a discrete group Γ act on a countable discrete space X. Then the reduced
crossed product Cb(X) ⋊r Γ by the algebra Cb(X) of bounded functions on Γ admits an
injective ∗-homomorphism into Bf (H). In particular, the reduced group C ∗-algebra C ∗
r (Γ)
admits an injective ∗-homomorphism into Bf (H).
ON THE C ∗-ALGEBRA OF MATRIX-FINITE BOUNDED OPERATORS
7
Proof. Set H = l2(X) with the basis of the characteristic functions δx of individual points
x ∈ X. Let a ∈ C[Γ], a = λ1g1 + . . . + λrgr, be a finite linear combination of group
elements g1, . . . , gr ∈ Γ, λ1, . . . , λr ∈ C. As gδx = δgx, g−1δx = δg−1x, the matrix of the
operator on l2(Γ) given by a has no more than r non-zero entries in each line and in
each column. This proves inclusion of C ∗
r (Γ) in Bf (H). Operators of multiplication by
functions on X are diagonal, hence trivially lie in Bf (H). The case X = Γ proves Lemma
for C ∗
r (Γ).
(cid:3)
Lemma 5.2. Let Fn be a free group on n generators. Then the full group C ∗-algebra
C ∗(Fn) admits an injective ∗-homomorphism into Bf (H).
Proof. It is shown in [8], Proposition 2.3, that the universal norm on C[Fn] equals the
supremum of the norms of representations of Fn which factor through a finite quotient.
By Lemma 5.1, C ∗(Fn) embeds into Bf (H).
(cid:3)
Let X be a metric space of bounded geometry, which means that, for any R > 0, the
number of points in each ball of raduis R is uniformly bounded. The uniform Roe algebra
C ∗(X) [10] is generated on H = l2(X) by operators a ∈ B(H), whose matrices satisfy
axy = 0 if d(x, y) > R, R > 0 (here x, y ∈ X, axy = hδx, aδy).
Lemma 5.3. The uniform Roe algebra C ∗(X) for metric spaces of bounded geometry
admits an injective ∗-homomorphism into Bf (H).
Proof. Obvious.
(cid:3)
Let V = N, and let G = (V, E) be an infinite, uniformly locally finite (ULF) graph, i.e.
E denotes the set of edges of G, two vertices v, w ∈ V are adjacent (v ∼ w) if there is an
edge e ∈ E that connects v and w, each vertex v has a finite number deg v of adjacent
vertices, and these numbers deg v are uniformly bounded on V . Let AG = ((aG)v,w)v,w∈V
be the adjacency matrix, i.e. the infinite matrix defined by (aG)v,w = (cid:26) 1,
if v ∼ w;
0, otherwise.
Each line and each column of the matrix AG has no more than k non-zero entries for some
k ∈ N. By Lemma 1.4, this matrix determines a bounded operator aG, called adjacency
operator. Let B(G) = C ∗(l∞(V ), aG) be the C ∗-subalgebra of B(l2(V )) generated by
(operators of multiplication by) l∞(V ) and by aG.
Lemma 5.4. The C ∗-algebra B(G) defined above admits an injective ∗-homomorphism
into Bf (H).
Proof. Obvious.
(cid:3)
Problem 5.5. Are there separable C ∗-algebras that do not admit an injective ∗-
homomorphism into Bf (H)?
6. Contractibility of the group of invertibles
Theorem 6.1. The group GLf (H) of invertible elements of Bf (H) is contractible.
Proof. The proof follows the original proof of Kuiper [7]. Let f : Sn → GLf (H) be a
continuous map from a sphere. A preliminary step is to deform f to a map f1 such
that f1(Sn) is contained in the linear span of some g1, . . . , gN ∈ GLf (H). After a small
8
V. MANUILOV
l ∩(cid:0)∩N
deformation, we may assume that there exists k ∈ N such that gj ∈ GL(k)(H), j =
1, . . . , N, where GL(k)(H) = GLf (H) ∩ B(k)(H).
i, and of finitedimensional subspaces
Ai ⊂ H with the following properties:
Then we have to find a sequence of unit vectors ai, a′
• Ai = Span(ai, g1(ai), . . . , gN (ai), a′
i);
• a′
i ⊥ ai, g1(ai), . . . , gN (ai), and a′
i ⊥ Al for l < i;
• ai ∈ ∩i−1
l )(cid:1)(cid:3).
l=1(cid:2)A⊥
We claim that if gj ∈ GL(k)(H) then the vectors ai, a′
i, i ∈ N, can be taken from the
basis {en}n∈N of H.
Indeed, one can start with a1 = e1. As gj ∈ GL(k)(H), there exists r1 such that
a1, g1(a1), . . . , gN (a1) ∈ Lr1, where Ln denotes the linear span of the vectors e1, . . . , en.
Then one can take a′
i−1 are already fixed, and a1, . . . , ai−1 ⊂ Ln
for some n ∈ N. Then we can find ai ∈ L⊥
n , j = 1, . . . , N. Indeed,
consider g1(em), m > n. There are no more than k non-zero coordinates among the first
n coordinates of g1(em) for each m. Then, as each line of the matrix of g1(em) cannot
contain more than k non-zero entries, the number of m's (i.e. the number of columns) for
which there is a non-zero coordinate among the first n coordinates of g1(em) is finite (not
greater than kn). The same holds for each j = 1, . . . , N, so there is some m such that the
first n coordinates of gj(em), j = 1, . . . , N, are all zeroes.
1 = er1+1, and we have A1 ⊂ Lr1+1.
n such that gj(ai) ∈ L⊥
Now assume that a1, . . . , ai−1 and a′
j=1g−1
j (A⊥
1, . . . , a′
The next step of homotopy in [7] is rotation of the two-dimensional subspace spanned
by g(ai) and a′
i, where g ∈ Span(g1, . . . , gN ), and then another rotation in the two-
i and ai. Note that, although B(k)(H) is not a linear
dimensional subspace spanned by a′
space, we have g ∈ GL(kN )(H). Both rotations involve only finite number of coordinates,
hence this part of the homotopy lies in GLm(H) for some m ∈ N and connects f1 with f2
such that f2(s)ai = f1(s)aiai, s ∈ Sn, i ∈ N.
Let H ′ = Span(a1, a2, . . .), H1 = (H ′)⊥. The next step of homotopy connects f2 with
f3 such that (f3)H1 = (f2)H1 and f3(s)H ′ = id. This homotopy only proportionally
changes lengths of f2(s)ai, so does not change the number of non-zero entries. Let p′ and
p1 denote the projections onto H ′ and H1 respectively. Then f3(s) = p′ + f2(s)p1. The
next step of homotopy connects f3(s) with f4(s) = p′ + p1f2(s)p1, and does not increase
the number of non-zero entries.
Recall that H ′ is spanned by an infinite set of the vectors from the basis, so we can
write it as an infinite sum of infinitedimensional subspaces H ′ = H2 ⊕ H3 ⊕ · · · , each
of which is spanned by an infinite set of the vectors of the basis. Then, with respect to
the decomposition H = H1 ⊕ H2 ⊕ H3 ⊕ · · · , we can write f3(s) = u
...! with u
having in each line and in each column mot more that C non-zero entries for some C.
The standard homotopy that connects ( u
1 ) (applied block-diagonal-wise)
increases the number of non-zero entries in each line and in each column only twice, so the
last step of homotopy that connects f4 with f5, where f5(s) = 1, also lies within GLf (H).
(cid:3)
u−1 ) with ( 1
1
1
Corollary 6.2. K0(Bf (H)) = K1(Bf (H)) = 0. In particular, the class of the projection
from Proposition 2.3 is trivial.
Proof. This follows from Lemma 1.5 and from the isomorphism Ki(A) ∼= πi±1(GL∞(A)),
i ∈ Z/2, [13], where A is a C ∗-algebra, GL(A) denotes the group of invertibles in A, and
GL∞(A) = limn→∞ GL(Mn(A)).
ON THE C ∗-ALGEBRA OF MATRIX-FINITE BOUNDED OPERATORS
9
(cid:3)
7. Miscellanea
Lemma 7.1. The C ∗-algebra Bf (H) has no polar decomposition, hence is not an AW ∗-
algebra.
Proof. Let v ∈ B(H) denote the isometry from Proposition 2.1. Let h be the compact
operator given by h(en) = λnen, with λn 6= 0 for any n ∈ N and limn→∞ λn = 0. Set
a = vh, where v was defined in Proposition 2.1. Compactness of a implies that a ∈ Bf (H),
and Ker h = 0 means that v is determined in a unique way. As v /∈ Bf (H), there is no
polar decomposition within Bf (H). Therefore Bf (H) is not a Rickart algebra, hence not
an AW ∗-algebra [2].
(cid:3)
Lemma 7.2. The stable topological rank of Bf (H) is ∞.
Proof. As in B(H), the unit projection 1 is the sum of the projections onto the subspaces
spanned by odd and by even vectors of the basis, each of which is Murray -- von Neumann
equivalent to 1, and both isometries v1 and v2 given by v1(en) = e2n−1, v2(en) = e2n,
n ∈ N, lie in Bf (H), hence Bf (H) is properly infinite, hence has infinite stable topological
rank [9].
(cid:3)
Problem 7.3. Calculate the real rank of Bf (H).
Acknowledgement. The author is grateful to E. Troitsky for fruitful discussions.
References
[1] B. Abadie, G. Cortinas. Homotopy invariance through small stabilizations. J. Homotopy Relat.
Struct. 10 (2015), 459 -- 493.
[2] P. Ara. Left and right projections are equivalent in Rickart C ∗-algebras. J. Algebra 120 (1989),
433 -- 448.
[3] N. P. Brown, N. Ozawa. C ∗-algebras and Finite-dimensional Approximations. Amer. Math. Soc.,
2008.
[4] J. W. Calkin. Two-sided ideals and congruences in the ring of bounded operators in Hilbert space.
Ann. Math. 42 (1941), 839 -- 873.
[5] G. Cortinas. Cyclic homology, tight crossed products, and small stabilizations. J. Noncommut. Geom.
8 (2014), 1191 -- 1223.
[6] N. Higson, V. Lafforgue, G. Skandalis. Counterexamples to the Baum -- Connes conjecture. Geom.
Funct. Anal. 12 (2002), 330 -- 354.
[7] N. Kuiper. The homotopy type of the unitary group of Hilbert space. Topology 3 (1965), 19-30.
[8] A. Lubotzky. Discrete groups, expanding graphs and invariant measures. Birkhauser, 1994.
[9] M. A. Rieffel. Dimension and stable rank in the K-theory of C ∗-algebras. Proc. London Math. Soc.
46 (1983), 301 -- 333.
[10] J. Roe. Lectures on Coarse Geometry. University Lecture Series. Amer. Math. Soc., 31, 2003.
[11] J. Roe, R. Willett. Ghostbusting and property A. J. Funct. Anal. 266 (2014), 1674 -- 1684.
[12] B. Tanbay. Pure state extensions and compressibility of the l1-algebra. Proc. Amer. Math. Soc. 113
(1991), 707 -- 713.
[13] R. Wood. Banach algebras and Bott periodicity. Topology 4 (1966), 371 -- 389.
Moscow State University, Leninskie Gory 1, Moscow, 119991, Russia
E-mail address: [email protected]
|
1808.07807 | 1 | 1808 | 2018-08-23T15:31:44 | Ample groupoids: equivalence, homology, and Matui's HK conjecture | [
"math.OA",
"math.KT"
] | We investigate the homology of ample Hausdorff groupoids. We establish that a number of notions of equivalence of groupoids appearing in the literature coincide for ample Hausdorff groupoids, and deduce that they all preserve groupoid homology. We compute the homology of a Deaconu{Renault groupoid associated to k pairwisecommuting local homeomorphisms of a zero-dimensional space, and show that Matui's HK conjecture holds for such a groupoid when k is one or two. We specialise to k-graph groupoids, and show that their homology can be computed in terms of the adjacency matrices, using a chain complex developed by Evans. We show that Matui's HK conjecture holds for the groupoids of single vertex k-graphs which satisfy a mild joint-coprimality condition. We also prove that there is a natural homomorphism from the categorical homology of a k-graph to the homology of its groupoid. | math.OA | math | AMPLE GROUPOIDS: EQUIVALENCE, HOMOLOGY,
AND MATUI'S HK CONJECTURE
CARLA FARSI, ALEX KUMJIAN, AND DAVID PASK AND AIDAN SIMS
Abstract. We investigate the homology of ample Hausdorff groupoids. We establish
that a number of notions of equivalence of groupoids appearing in the literature coincide
for ample Hausdorff groupoids, and deduce that they all preserve groupoid homology.
We compute the homology of a Deaconu -- Renault groupoid associated to k pairwise-
commuting local homeomorphisms of a zero-dimensional space, and show that Matui's
HK conjecture holds for such a groupoid when k is one or two. We specialise to k-graph
groupoids, and show that their homology can be computed in terms of the adjacency ma-
trices, using a chain complex developed by Evans. We show that Matui's HK conjecture
holds for the groupoids of single vertex k-graphs which satisfy a mild joint-coprimality
condition. We also prove that there is a natural homomorphism from the categorical
homology of a k-graph to the homology of its groupoid.
8
1
0
2
g
u
A
3
2
]
.
A
O
h
t
a
m
[
1
v
7
0
8
7
0
.
8
0
8
1
:
v
i
X
r
a
1. Introduction
The purpose of this paper is to investigate the homology of ample Hausdorff groupoids,
and to investigate Matui's HK-conjecture for groupoids associated to actions of Nk by local
homeomorphisms on locally compact Hausdorff zero-dimensional spaces. Ample Hausdorff
groupoids are an important source of examples of C ∗-algebras. They provide models for
the crossed-products associated to Cantor minimal systems [20], Cuntz -- Krieger algebras
and graph C ∗-algebras and their higher-rank analogues [44, 28, 27], and recently models
for large classes of classifiable C ∗-algebras [6, 14, 41].
It is therefore very desirable to
develop techniques for computing the K-theory of the C ∗-algebra of an ample Hausdorff
groupoid. Unfortunately, there are relatively few general techniques available.
In a series of recent papers [32, 33, 34], Matui has advanced a conjecture that if G is
a minimal effective ample Hausdorff groupoid with compact unit space then K0(C ∗
r (G))
is isomorphic to the direct sum of the even homology groups H2n(G) of the groupoid as
defined by Crainic and Moerdijk [11], and K1(C ∗
r (G)) is isomorphic to the direct sum of
the odd homology groups H2n+1(G). He has verified this conjecture for a number of key
classes of groupoids, including finite cartesian products of groupoids associated to shifts
of finite type, transformation groupoids for Cantor minimal systems, and AF groupoids
with compact unit spaces. He has also developed tools for computing the homology of
ample Hausdorff groupoids, including a spectral sequence that relates the homology of a
groupoid G endowed with a cocycle c taking values in a discrete abelian group H with the
Date: August 24, 2018.
2010 Mathematics Subject Classification. Primary 37B05, 22A22, 46L80, 19D55.
Key words and phrases. Ample groupoid, groupoid equivalence, homology of ´etale groupoids, K-theory
of C ∗-algebras, Deaconu-Renault groupoid.
C.F. was partially supported by Simons Foundation Collaboration grant 523991. A.K. was partially
supported by Simons Foundation Collaboration grant 353626. This research was supported by the Aus-
tralian Research Council.
1
2
C. FARSI, A. KUMJIAN, AND PASK, SIMS
homology of H with values in the homology groups of the skew-product groupoid G ×c H.
Other authors have subsequently verified Matui's conjecture for Exel-Pardo groupoids
and certain graded ample Hausdorff groupoids (see [38] [21]).
We begin Section 3 by investigating the many notions of groupoid equivalence in the
literature in the context of arbitrary ample Hausdorff groupoids. Crainic and Moerdijk
focus on the notion of Morita equivalence of groupoids (see [11, 4.5]) while Matui employs
the notions of similarity (see [32, Definition 3.4]) and Kakutani equivalence (see [32, Def-
inition 4.1]). Similarity of groupoids was previously studied by Renault [44] and Ramsay
[43]. In the setting of ample Hausdorff groupoids with σ-compact unit spaces, it follows
from [32, Theorem 3.6] that Kakutani equivalence implies similarity. We show that sim-
ilarity, Kakutani equivalence, Renault's notion of groupoid equivalence [45, 37], and the
notion of groupoid Morita equivalence of Crainic and Moerdijk (as well as a number of
other notions) all coincide for ample Hausdorff groupoids with σ-compact unit spaces (see
Theorem 3.12).
In Section 4 we recall the definition of homology for an arbitrary ample Hausdorff
groupoid from [11, 3.1] (see also [32, Definition 3.1]) and we appeal to a theorem of
Crainic and Moerdijk to observe that groupoid equivalence preserves groupoid homology
for arbitrary ample Hausdorff groupoids (see [11, Corollary 4.6]). Matui also proved that
similar Hausdorff ´etale groupoids have isomorphic homology groups (see [32, Proposition
3.5]), and this formulation allows us to give an explicit description of the isomorphism
when the equivalence arises from a similarity.
In Section 5 we introduce Matui's HK
conjecture, and extend his proof that AF groupoids with compact unit space satisfy the
HK conjecture to the case of non-compact unit spaces.
Vn Zk ⊗Cc(X, Z) and the boundary maps are built from the forward maps σn
Our main computations of groupoid homology are in Section 6, where we investigate
the homology of Deaconu -- Renault groupoids G(X, σ) associated to actions σ of Nk by
local homeomorphisms on totally disconnected locally compact Hausdorff spaces X. We
adapt techniques developed by Evans [16] in the context of K-theory for higher-rank
graph C ∗-algebras to construct a chain complex Aσ in which the n-chains are elements of
∗ on Cc(X, Z)
that satisfy σn
∗ (1U ) = 1σn(U ) whenever U ⊆ X is a compact open set on which σn is
injective. Our main result, Theorem 6.5, gives an explicit computation of the homology
groups Hn(G(X, σ)): we prove that Hn(G(X, σ)) is canonically isomorphic to Hn(Aσ
∗ ). We
then show that, if c : G(X, σ) → Zk is the canonical cocycle, then the homology groups
H∗(Aσ) also coincide with the homology groups H∗(Zk, K0(C ∗(G(X, σ)×c Zk))) appearing
in Kasparov's spectral sequence for the double crossed product C ∗(G(X, σ) ×c Zk) ⋊ Zk.
Since this double crossed product is Morita equivalent to C ∗(G(X, σ)) by Takai duality,
this provides a useful tool for calculating the K-theory of C ∗(G(X, σ)).
In Theorems
6.7 and 6.10 we calculate both the K-groups of C ∗(G(X, σ)) and the homology groups
of G(X, σ) explicitly for k = 1, 2, and in particular prove that ample Deaconu -- Renault
groupoids of rank at most 2 satisfy Matui's HK conjecture. We also discuss the differences
between Kasparov's spectral sequence and Matui's for k ≥ 3 and indicate where one might
look for counterexamples to Matui's conjecture amongst such groupoids.
Finally, in Section 7, we specialise to k-graphs. The k-graph groupoid GΛ of a row-
finite k-graph Λ with no sources is precisely the Deaconu -- Renault groupoid G(Λ∞, σ)
associated to the shift maps on the infinite-path space of Λ. We begin the section by
linking the homology of the k-graph groupoid with the categorical homology of the k-graph
by constructing a . natural homomorphism from H∗(Λ) to H∗(GΛ). We then investigate
AMPLE GROUPOID HOMOLOGY
3
how to apply the results of Section 6 in the specific setting of k-graph groupoids. We prove
∗ associated to (Λ∞, σ) as in Section 6 has the same homology
that the chain complex Aσ
as the much simpler chain complex DΛ
∗ described by Evans in [16]. This provides a very
concrete calculation of the homology of a k-graph groupoid. It follows that the homology
of GΛ does not depend on the factorisation rules in Λ. We use this and the preceding
section to establish an explicit description of the homology of 1-graph groupoids and
2-graph groupoids and to see that these groupoids satisfy Matui's conjecture. We also
prove that for arbitrary k, if Λ is a k-graph with a single vertex such that the integers
Λe1 − 1, . . . , Λek − 1 have no nontrivial common divisors, then both the homology of GΛ
and the K-theory of C ∗(Λ) are trivial, and in particular GΛ satisfies the HK-conjecture.
Acknowledgements: C.F. thanks A.K. for his hospitality during her visit to UNR.
A.K. thanks his co-authors for their hospitality and support on his visits to Boulder and
Wollongong.
2. Background
2.1. Groupoids and their C ∗-algebras. We give some brief background on groupoids
and their C ∗-algebras and establish our notation. For details, see [17, 44, 49]. A groupoid
is a small category G with inverses. We write G(0) for the set of identity morphisms of G,
called the unit space, and we write r, s : G → G(0) for the range and source maps. We
write G(2) for the set {(γ1, γ2) ∈ G × G : s(γ1) = r(γ2)} of composable pairs in G. The
groupoid G is a topological groupoid if it has a locally compact topology under which
all operations in G are continuous and G(0) is Hausdorff in the relative topology. If the
topology on all of G is Hausdorff, we call G a Hausdorff groupoid. An ´etale groupoid is a
topological groupoid in which G(0) is open, and r, s : G → G(0) are local homeomorphisms
(in [44] such a groupoid is called r-discrete with Haar system). An open subset U ⊆ G is
said to be an open bisection if both rU and sU are homeomorphisms onto their ranges.
Given u ∈ G(0) we write Gu for {γ ∈ G : r(γ) = u}, Gu for {γ ∈ G : s(γ) = u} and
u = Gu ∩ Gu.
Gu
A groupoid G is ample if it is ´etale and G(0) is zero dimensional; equivalently, G is ample
if it has a basis of compact open bisections.
The orbit of a unit u ∈ G(0) is the set [u] := r(Gu) = s(Gu). A subset U ⊆ G(0) is full
if U ∩ [u] 6= ∅ for every unit u. We say that G is minimal if the only nontrivial open
invariant subset of G(0) is G(0); equivalently, G is minimal if the closure of [u] is equal to
G(0) for every u ∈ G(0).
The isotropy of G is the setSu∈G(0) Gu
u of elements of G whose range and source coincide.
A groupoid G is said to be effective 1 if the interior of its isotropy coincides with its unit
space G(0)
Let A be an abelian group and let c : G → A be a 1-cocycle. Then we may form the skew
product groupoid G ×c A which is the set G × A with structure maps r(γ, a) = (r(γ), a),
s(γ, a) = (s(γ), a + c(γ)) and (γ, a)(η, a + c(γ)) = (γη, a) (see [44, Definition I.1.6]). There
is a natural action α of A on G ×c A given by αb(γ, a) = (γ, a + b).
1Matui uses the term essentially principal, and the term topologically principal has also been used
elsewhere in the literature.
(f ∗ g)(γ) = Xγ=γ1γ2
f (γ1)g(γ2)
and
f ∗(γ) = f (γ−1).
4
C. FARSI, A. KUMJIAN, AND PASK, SIMS
Given a Hausdorff ´etale groupoid G, the space Cc(G) of continuous compactly supported
functions from G to C becomes a ∗-algebra with operations given by
f (γ)δγη. The reduced C ∗-algebra C ∗
The groupoid C ∗-algebra C ∗(G) is the universal C ∗-algebra generated by a ∗-representation
of Cc(G) (cf. [44, II.1.5]). For each unit u ∈ G0 there is a ∗-representation πu : Cc(G) →
r (G) is the
B(ℓ2(Gu)) given by πu(f )δη = Pγ∈Gr(η)
closure of the image of Cc(G) under the representationLu∈G(0) πu.
A cocycle c : G → Zk determines an action of Tk by automorphisms of Cc(G) given
by (z · f )(γ) = zc(γ)f (γ), and this extends to an action of Tk by automorphisms on
r (G). There is an isomorphism C ∗(G ×c Zk) ∼= C ∗(G) ⋊ Tk that
each of C ∗(G) and C ∗
carries a function f ∈ Cc(G × {n}) ⊆ Cc(G ×c Zk) to the function z 7→ (g 7→ znf (g, n)) ∈
r (G ×c Zk) ∼=
C(Tk, C ∗(G)) ⊆ C ∗(G)⋊Tk. This isomorphism descends to an isomorphism C ∗
C ∗
r (G) ⋊ Tk.
We will be particularly interested in Deaconu -- Renault groupoids associated to actions
of Nk, which are defined as follows. Let X be a locally compact Hausdorff space, and
let σ be an action of Nk on X by surjective local homeomorphisms. The associated
Deaconu -- Renault groupoid2 G = G(X, σ) is defined by
G = {(x, p − q, y) ∈ X × Zk × X : σp(x) = σq(y)}
(cf. [13, 18]). We identify X with the unit space via the map x 7→ (x, 0, x). The structure
maps are given by r(x, n, y) = x, s(x, n, y) = y and (x, m, y)(y, n, z) = (x, m + n, z). A
basis for the topology on G is given by subsets of the form U × {p − q} × V where U, V
are open in X and σp(U) = σq(V ).
There is a natural cocycle c : G(X, σ) → Zk given by c(x, n, y) := n. We can then
form the skew-product groupoid G ×c Zk. With our conventions, in this groupoid we
have r((x, n, y), p) = (x, p), s((x, n, y), p) = (y, p + n) and ((x, n, y), p)((y, m, z), p + n) =
((x, m + n, z), p).
There is an actioneσ of Nk on eX = X × Zk by surjective local homeomorphisms given
byeσq(x, p) = (σq(x), p + q). Moreover there is an isomorphism G ×c Zk ∼= G(eX,eσ) given
((x, m, y), p) 7→ ((x, p), m, (y, p + m)).
(2.1)
by
The full and reduced C ∗-algebras of a Deaconu -- Renault groupoid coincide (see for
example [51, Lemma 3.5]).
2.2. k-graphs, their path groupoids, and their C ∗-algebras. For k ≥ 1, a k-graph
is a non-empty countable small category equipped with a functor d : Λ → Nk that satisfies
the following factorisation property. For all λ ∈ Λ and m, n ∈ Nk such that d(λ) = m + n
there exist unique µ, ν ∈ Λ such that d(µ) = m, d(ν) = n, and λ = µν. When d(λ) = n
we say λ has degree n, and we write Λn = d−1(n). The standard generators of Nk are
denoted ε1, . . . , εk, and we write ni for the ith coordinate of n ∈ Nk. We define a partial
order on Nk by m ≤ n if mi ≤ ni for all i ≤ k.
If Λ is a k-graph, its vertices are the elements of Λ0. The factorisation property implies
that these are precisely the identity morphisms, and so can be identified with the objects.
2sometimes referred to as a transformation groupoid (see [18])
AMPLE GROUPOID HOMOLOGY
5
For λ ∈ Λ the source s(λ) is the domain of λ, and the range r(λ) is the codomain of
λ (strictly speaking, s(λ) and r(λ) are the identity morphisms associated to the domain
and codomain of λ). Given λ, µ ∈ Λ and E ⊆ Λ, we define
λE = {λν : ν ∈ E, r(ν) = s(λ)}, Eµ = {νµ : ν ∈ E, s(ν) = r(µ)},
and
λEµ = (λE)µ = λ(Eµ).
In particular, for v ∈ Λ0 and n ∈ Nk, we have vΛn = {λ ∈ Λn : r(λ) = v}.
We say that the k-graph Λ is row-finite if vΛn < ∞ for each n ∈ Nk and v ∈ Λ0, and
has no sources if 0 < vΛm for all v ∈ Λ0 and m ∈ Nk.
Let A be an abelian group. Given a functor c : Λ → A, we may form the skew-product
k-graph Λ ×c A which is the set Λ × A endowed with structure maps given by r(λ, a) =
(r(λ), a), s(λ, a) = (s(λ), a + c(λ)), (λ, a)(µ, a + c(λ)) = (λµ, a), and d(λ, a) = d(λ) (see
[27, Definition 5.1]). There is a natural A-action α on Λ×c A given by αb(λ, a) = (λ, a+b).
Examples 2.1. (a) A 1-graph is the path category of a directed graph (see [27]).
(b) Let Mor Ωk = {(m, n) ∈ Nk × Nk : m ≤ n}, and Obj Ωk = Nk. Define r, s : Mor Ωk →
Obj Ωk by r(m, n) = m, s(m, n) = n, and for m ≤ m ≤ p ∈ Nk define (m, n)(n, p) =
(m, p) and d(m, n) = n − m. Then (Ωk, d) is a k-graph. We identify Obj Ωk with
{(m, m) : m ∈ Nk} ⊆ Mor Ωk.
Let Λ be a row-finite k-graph with no sources. The set Λ∞ = {x : Ωk → Λ
x is a degree-preserving functor} is called the infinite path space of Λ. For v ∈ Λ0, we put
vΛ∞ = {x ∈ Λ∞ : x(0, 0) = v}.
For λ ∈ Λ, let Z(λ) = {x ∈ Λ∞ : x(0, d(λ)) = λ}. Then {Z(λ) : λ ∈ Λ} forms a basis
of compact open sets for a topology on Λ∞. For p ∈ Nk, the shift map σp : Λ∞ → Λ∞
defined by (cid:0)σp(x))(m, n) = x(m + p, n + p) for x ∈ Λ∞ is a local homeomorphism (for
more details see [27, Remark 2.5, Lemma 2.6]).
Following [27, Definition 2.7] we define the k-graph groupoid of Λ to be the Deaconu --
Renault groupoid
(2.2) GΛ := G(Λ∞, σ) = {(x, m − n, y) ∈ Λ∞ × Zk × Λ∞ : m, n ∈ Nk, σm(x) = σn(y)}.
The sets Z(µ, ν) := {(µx, d(µ) − d(ν), νx) : x ∈ Z(s(µ))} indexed by pairs µ, ν ∈ Λ
with s(µ) = s(ν) form a basis of compact open bisections for a locally compact Hausdorff
topology on GΛ. With this topology GΛ is an ample Hausdorff groupoid (see [27, Propo-
sition 2.8]). The sets Z(λ) = Z(λ, λ) form a basis for the relative topology on G(0)
Λ ⊆ GΛ.
We identify G(0)
Λ = {(x, 0, x) : x ∈ Λ∞} with Λ∞. The groupoid GΛ is minimal if and only
if Λ is cofinal [27, Proof of Proposition 4.8].
As in the proof of [27, Theorem 5.2], there is a bijection between Λ∞×Zk and (Λ×dZk)∞
given by (x, p) 7→ ((m, n) 7→ (x(m, n), m + p)). After making this identification, we
obtain an isomorphism of the groupoid GΛ×dZk of the skew-product k-graph with the
skew-product groupoid GΛ ×c Zk corresponding to the canonical cocycle c(x, n, y) = n via
the map ((x, p), m − n, (y, q)) 7→ ((x, m − n, y), p).
Let Λ be a row-finite k-graph with no sources. A Cuntz -- Krieger Λ-family in a C ∗-
algebra B is a function s : λ 7→ sλ from Λ to B such that
(CK1) {sv : v ∈ Λ0} is a collection of mutually orthogonal projections;
(CK2) sµsν = sµν whenever s(µ) = r(ν);
(CK3) s∗
λsλ = ss(λ) for all λ ∈ Λ; and
6
C. FARSI, A. KUMJIAN, AND PASK, SIMS
(CK4) sv =Pλ∈vΛn sλs∗
λ for all v ∈ Λ0 and n ∈ Nk.
The k-graph C ∗-algebra C ∗(Λ) is the universal C ∗-algebra generated by a Cuntz -- Krieger
Λ-family. There is an isomorphism C ∗(Λ) ∼= C ∗(GΛ) satisfying sλ 7→ 1Z(λ,s(λ)) (see [27,
Corollary 3.5(i)]).
As discussed above, we have GΛ×dZk ∼= GΛ ×c Zk, and so
C ∗(GΛ ×c Zk) ∼= C ∗(GΛ×dZk ) ∼= C ∗(Λ ×d Zk).
These are approximately finite dimensional (AF) algebras by [27, Lemma 5.4].
2.3. K-theory for C ∗-algebras. This paper is concerned primarily with calculating
groupoid homology, but it is motivated by the relationship between this and K-theory of
groupoid C ∗-algebras, and some of our key results concern K-theory. Readers unfamiliar
with C ∗-algebras and their K-theory, and who are primarily interested in groupoids and
groupoid homology will not need to know more about C ∗-algebraic K-theory than the
following points: C ∗-algebraic K-theory associates two abelian groups K0(A) and K1(A)
to each C ∗-algebra A; these groups are invariant under Morita equivalence, and continuous
with respect to inductive limits; the K0-group is the Grothendieck group of a semigroup
consisting of equivalence classes of projections in matrix algebras over A; the K-groups
of a crossed-product of a C ∗-algebra A by Z are related to those of A by the Pimsner --
Voiculescu exact sequence [40]; and the K-groups of a crossed-product of a C ∗-algebra A
by Zk fit into a spectral sequence, due to Kasparov [23], involving the homology groups
of Zk with values in the K-groups of A. For more background on C ∗-algebraic K-theory,
we refer the interested reader to [55], [46] or [2].
2.4. c-soft sheaves. Let X be a locally compact Hausdorff space. By a sheaf of abelian
groups, or simply a sheaf, over X, we mean a (not necessarily Hausdorff) ´etale groupoid
F with unit space F (0) = X in which r = s, so every element belongs to the isotropy,
and in which each isotropy group Fx = F x
x is abelian. We think of F as a group bundle
over X with bundle map r = s. Given a subset W ⊆ X, we write Γ(W, F ) for the set
{t : W → F : t(w) ∈ Fw and t is continuous} of continuous sections of F over W . A
sheaf F over X is said to be c-soft if the restriction map Γ(X, F ) → Γ(K, F ) is surjective
for any compact set K ⊆ X (see e.g. [22, Definition 2.5.5] or [4, II.9.1]); that is, if every
continuous section of Γ over a compact subset of X extends to a continuous section over
all of X.
The property of c-softness is a key hypothesis for results of Crainic and Moerdijk (see
[11]) that we will need in our study of the homology of ample Hausdorff groupoids.
Lemma 2.2. Let X be a 0-dimensional topological space, and let K ⊆ X be compact.
If W is a compact open subset of K (in the relative topology on K), then there exists a
compact open set V ⊆ X such that W = V ∩ K.
(in the relative topology), There is a finite subset F ⊆ U ′ such that
K := {U ∩ K : U ∈ U ′} is an open cover of the compact set W ⊆ K
Proof. Since W is open in the relative topology, W = bV ∩ K, with bV open in X. Let U be
a basis of compact open sets for the topology on X; so U ′ := {U ∈ U : U ⊆ bV } satisfies
bV =SU ∈U ′ U. Since U ′
Now V :=SU ∈F U is compact open, and W = V ∩ K.
W = [U ∈F
(U ∩ W ).
(cid:3)
AMPLE GROUPOID HOMOLOGY
7
Proposition 2.3. Let X be a 0-dimensional topological space. Then the constant sheaf
F on X with values in a discrete abelian group A is c-soft.
Proof. Since F is the constant sheaf, for every W ⊆ X we have Γ(W, F ) ∼= C(W, A). Let
K ⊆ X be compact and fix f ∈ C(K, A). Then f (K) is a compact subset of the discrete
group A, and hence finite. For a ∈ f (K) we let Ua = f −1(a). Since f is continuous each
Ua ⊆ K is clopen and since K is compact Ua is compact and open. By Lemma 2.2, for
each a ∈ A there exists a compact open set Va ⊆ X such that Ua = Va ∩ K. Fix a total
order ≤ on f (K) and for each a ∈ f (K), define V ′
a is a
compact open subset of X since the Va are compact open. Moreover, since the Ua are
mutually disjoint, we have V ′
a := Va\Sb<a Vb. Then each V ′
a ∩ K = Va ∩ K for all a. Hence the formula
defines a continuous extension f of f to X.
f (x) :=(cid:26) a
0
for x ∈ V ′
a
otherwise,
(cid:3)
3. Equivalence of ample Hausdorff groupoids
There are a number of notions of equivalence of groupoids that are relevant to us here,
and we need to know that they all coincide for ample Hausdorff groupoids. The notions
that we consider are Morita equivalence [11], groupoid equivalence [37] (see also [45]),
equivalence via a linking groupoid, equivalence via isomorphic ampliations, similarity
[43, 44, 32], Kakutani equivalence [32] and stable isomorphism [8]. We show that the
first four of these notions coincide for all Hausdorff ´etale groupoids (Proposition 3.10),
and that they all coincide for ample Hausdorff groupoids with σ-compact unit spaces
(Theorem 3.12).
The following notion of similarity, recorded by Matui [32, Definition 3.4] and called
homological similarity there, appears in the context of algebraic groupoids in Renault's
thesis [44], and earlier still in [43] -- where Ramsay in turn attributes it to earlier work of
Mackey. An adaptation of this notion to the situation of Lie groupoids also appears in
[36] where it is called strong equivalence.
Definition 3.1 ([44, Definition I.1.3], [32, Definition 3.4]). Let G, H be Hausdorff ´etale
groupoids and let ρ, σ : G → H be continuous homomorphisms. We say that ρ is similar
to σ if there is a continuous map θ : G(0) → H such that
θ(r(g))ρ(g) = σ(g)θ(s(g))
for all g ∈ G.
We say that G and H are similar groupoids if there exist ´etale homomorphisms ρ : G → H
and σ : H → G such that σ ◦ ρ is similar to idG and ρ ◦ σ is similar to idH. In this case,
each of the two maps, ρ and σ, is called a similarity.
Remark 3.2. It is not stated in [44] or in [32] that similarity of groupoids is an equivalence
relation, but this is standard (it is essentially the argument that category equivalence is an
equivalence relation). It is also easy to give a direct argument: suppose that σ : G → H
and ρ : H → G implement a similarity, and that α : H → K and β : K → H also
implement a similarity. We aim to show that α ◦ σ and ρ ◦ β implement a similarity. By
symmetry it suffices to find κ : G(0) → G such that
κ(r(g))(cid:0)ρ ◦ β ◦ α ◦ σ(cid:1)(g) = gκ(s(g))
for all g.
8
C. FARSI, A. KUMJIAN, AND PASK, SIMS
Since ρ ◦ σ ∼ idG and β ◦ α ∼ idH, there are θ : G(0) → G and η : H(0) → H such that
θ(r(g))ρ(cid:0)σ(g)(cid:1) = gθ(s(g))
and
η(r(h))β(cid:0)α(h)(cid:1) = hη(s(h)).
Define κ(x) = θ(x)ρ(cid:16)η(r(σ(x)))(cid:17). Then using that r(σ(r(g))) = r(σ(g)), and that ρ is a
homomorphism, we compute
κ(r(g))(cid:0)ρ ◦ β ◦ α ◦ σ(cid:1)(g) = θ(r(g))ρ(cid:0)η(r(σ(g))(cid:1)β(cid:0)α(σ(g)))(cid:1)
= θ(r(g))ρ(cid:0)σ(g)(cid:1)ρ(cid:0)η(s(σ(g)))(cid:1) = gθ(s(g))ρ(cid:0)η(s(σ(g)))(cid:1) = gκ(s(g)).
Thus (ρ ◦ β) ◦ (α ◦ σ) is similar to idG and by symmetry (α ◦ σ) ◦ (ρ ◦ β) is similar to idK.
Remark 3.3. Let ρ, σ : G → H be continuous groupoid homomorphisms between Haus-
dorff ´etale groupoids. Then ρ and σ both induce well-defined orbit maps [u] 7→ [ρ(u)]
and [u] 7→ [σ(u)]. Suppose that ρ and σ are similar; then [ρ(u)] = [σ(u)] for all u ∈ G(0),
and so the orbit maps induced by ρ and σ are equal. It follows that every similarity of
groupoids induces a bijection between their orbit spaces.
Definition 3.4 ([11, Section 4.5]). Let G, H be groupoids. A continuous functor ϕ : G →
H is a weak equivalence if
(i) the map from G(0) ∗H(0) H := {(u, γ) ∈ G(0) × H : ϕ(u) = r(γ)} to H(0) given by
(u, γ) 7→ s(γ) is an ´etale surjection, and
(ii) G is isomorphic to the fibred product
G(0) ∗H(0) H ∗H(0) G(0) = {(u, γ, v) ∈ G(0) × H × G(0) : r(γ) = ϕ(u), s(γ) = ϕ(v)}.
If such a ϕ exists, then we write G ∼→ H. The groupoids G and H are Morita equivalent
if there is a groupoid K that admits weak equivalences G ∼← K ∼→ H.
Remark 3.5. It is not stated explicitly in [11] that Morita equivalence is in fact an equiv-
alence relation, though this is certainly standard. In any case it follows from Theorem 3.12
below.
In [36, Proposition 5.11] it is shown that the analogue of similarity for Lie groupoids
implies the analogue of Morita equivalence. We briefly indicate how to modify their
argument to obtain the same result for Hausdorff ´etale groupoids.
Lemma 3.6. Let G, H be Hausdorff ´etale groupoids. If G and H are similar, then they
are Morita equivalent.
Proof. Suppose that σ : G → H and ρ : H → G constitute a similarity of groupoids.
We claim that the map σ is a weak equivalence. It is easy to show that (u, h) 7→ s(h)
is a surjection from G(0) ∗G(0) H to H(0) since any v ∈ H(0) is the image of (ρ(v), v). We
claim that it is an ´etale map. For this, fix (u, h) ∈ G(0) ∗G(0) H, and use that σ is an ´etale
map to choose a neighbourhood U of u such that σU is a homeomorphism onto its range.
Pick a bisection neighbourhood B of h. By shrinking if necessary, we can assume that
σ(U) = rH(B). Then s ◦ π2 is a homeomorphism on U ∗ B ∼= B.
Since ρ, σ constitute a similarity of groupoids it is straightforward to see that the map
g 7→ (r(g), σ(g), s(g)) is a bijection from G to G(0) ∗H(0) H ∗H(0) G(0). For g ∈ G we use
that r, σ, and s are ´etale maps to find a neighbourhood U of g on which they are all
homeomorphisms, and observe that then g 7→ (r(g), σ(g), s(g)) is a homeomorphism onto
r(U) ∗H(0) σ(U) ∗H(0) s(U). So g 7→ (r(g), σ(g), s(g)) is continuous and open.
(cid:3)
AMPLE GROUPOID HOMOLOGY
9
The third notion of equivalence that we consider is the one formulated by Renault (see
[45, Section 3]) and studied in [37]. Given a locally compact Hausdorff groupoid G, we
say that a locally compact Hausdorff space Z is a left G-space if it is equipped with a
continuous open map r : Z → G(0) and a continuous pairing (g, z) 7→ g · z from G ∗ X to
X such that r(g · z) = r(g) and (gh) · z = g · (h · z) and such that r(z) · z = z. We say
that Z is a free and proper left G-space if the map (g, x) 7→ (g · x, x) is a proper injection
from G ∗ X to G × X. Right G-spaces are defined analogously.
Definition 3.7 ([37, Definition 2.1],[45, §3]). The groupoids G and H are equivalent if
there is a locally compact Hausdorff space Z such that
(1) Z is a free and proper left G-space with fibre map r : Z → G(0),
(2) Z is a free and proper right H-space with fibre map s : Z → H(0),
(3) the actions of G and H on Z commute,
(4) r : Z → G(0) induces a homeomorphism Z/H → G(0), and
(5) r : Z → H(0) induces a homeomorphism G\Z → H(0).
The fourth notion of equivalence we need to discuss is the generalisation of Kakutani
equivalence developed by Matui [32, Definition 4.1] in the situation of ample Hausdorff
groupoids with compact unit spaces, and extended to non-compact unit spaces in [8].
This notion has previously been discussed only for ample groupoids, but it makes sense
for general Hausdorff ´etale groupoids, and in particular weak Kakutani equivalence is
a fairly natural notion in this setting (though in this more general setting it is not an
equivalence relation, see Example 3.13 ).
Definition 3.8. The Hausdorff ´etale groupoids G and H are weakly Kakutani equivalent
∼= HY . They are
if there are full open subsets X ⊆ G(0) and Y ⊆ H(0) such that GX
Kakutani equivalent if X and Y can be chosen to be clopen sets.
For ample Hausdorff groupoids with σ-compact unit spaces, [8, Theorem 3.2] shows
that weak Kakutani equivalence and Kakutani equivalence both coincide with groupoid
equivalence, and with a number of other notions of equivalence. Our next two results
show first that for Hausdorff ´etale groupoids, Morita equivalence and equivalence in the
sense of [37] are equivalent to the existence of a linking groupoid, and to existence of
isomorphic ampliations of the two groupoids in the following sense.
If G is a Hausdorff ´etale groupoid, X is a locally compact Hausdorff space, and ψ :
X → G(0) is a local homeomorphism, then the ampliation (also known as the blow-up [57,
§3.3]) Gψ of G corresponding to ψ is given by
Gψ = {(x, γ, y) ∈ X × G × X : ψ(x) = r(γ) and ψ(y) = s(γ)}
with (cid:0)(x, γ, y), (w, η, z)(cid:1) ∈ (Gψ)(2) if and only if y = w, and composition and inverses
given by (x, γ, y)(y, η, z) = (x, γη, z) and (x, γ, y)−1 = (y, γ−1, x). This is a Hausdorff
´etale groupoid under the relative topology inherited from X × G × X.
Example 3.9. Let X and Y be locally compact Hausdorff spaces and ψ : Y → X be a
local homeomorphism. Then we may regard
R(ψ) := {(y1, y2) ∈ Y × Y : ψ(y1) = ψ(y2)}
as an Hausdorff ´etale groupoid (see [26]). Note that R(ψ) is the ampliation of the trivial
groupoid X corresponding to ψ.
10
C. FARSI, A. KUMJIAN, AND PASK, SIMS
Proposition 3.10. Let G and H be Hausdorff ´etale groupoids. The following are equiv-
alent:
(1) G and H are Morita equivalent;
(2) there is a Hausdorff ´etale groupoid L and a decomposition L(0) = X ⊔ Y of L(0)
into complementary full clopen subsets such that LX
(3) G and H are equivalent in the sense of Renault; and
(4) G and H admit isomorphic ampliations.
∼= G and LY
∼= H;
If G and H are weakly Kakutani equivalent, then they satisfy (1) -- (4).
Proof. For (1) =⇒ (2), suppose that φ : G → H is a weak equivalence in the sense of
Definition 3.4. Then G is isomorphic to the fibred product
G(0) ∗ H ∗ G(0) := {(x, η, y) ∈ G(0) × H × G(0) : φ(x) = r(η) and φ(y) = s(η)}.
Using the first condition of Definition 3.4 it is straightforward to check that under the
natural operations and topology, the disjoint union
L := (G(0) ∗ H ∗ G(0)) ⊔ (G(0) ∗ H) ⊔ (H ∗ G(0)) ⊔ H
is a Hausdorff ´etale groupoid satisfying (2) with respect to X = G(0) and Y := H(0).
For (2) =⇒ (3), one checks that given L as in (2), the subspace
Z := {z ∈ L : r(z) ∈ X and s(z) ∈ Y },
under the actions of G and H by multiplication on either side, is a G -- H-equivalence as
in (3).
For (3) =⇒ (4), suppose that Z is a G -- H-equivalence; to avoid confusion, we will
write ρ : Z → G(0) and σ : Z → H(0) for the anchor maps. Since Z/H ∼= G(0) and
since the right H-action is free, if x, y ∈ Z satisfy ρ(x) = ρ(y), then there is a unique
element [x, y]H of H satisfying x · [x, y]H = y. By [37, §2] the map [·, ·]H is continuous
(see also [50, Lemma 2.1]). Similarly there is a continuous pairing (x, y) 7→ G[x, y] from
{(x, y) ∈ Z 2 : σ(x) = σ(y)} to G such that G[x, y] · y = x. Consider the ampliations
Gρ := {(x, γ, y) : x, y ∈ Z, γ ∈ G, ρ(x) = r(γ), ρ(y) = s(γ)}
Hσ := {(x, η, y) : x, y ∈ Z, η ∈ H, σ(x) = r(η), σ(y) = s(η)}.
and
If (x, γ, y) ∈ Gρ, then ρ(γ · y) = r(γ) = ρ(x), and so we can take the pairing [x, γ ·
y]H to obtain an element Θ(x, γ, y) := (x, [x, γ · y]H, y) ∈ Hσ.
It is routine to check
that this is a continuous groupoid homomorphism. Symmetrically, we see that Θ′
:
(x, η, y) 7→ (x, G[x · η, y], y) is a continuous groupoid homomorphism from Hσ to Gρ. A
simple calculation using the defining properties of G[·, ·] and [·, ·]H shows Θ and Θ′ are
mutually inverse. So Θ is an isomorphism, giving (4).
For (4) =⇒ (1), fix ampliations Gφ and Hψ and an isomorphism Θ : Gφ → Hψ. Write
πG for the canonical map (x, γ, y) 7→ γ from Gφ to G, and πH for the corresponding
map from Hψ to H. We obtain continuous groupoid homomorphisms φ : Gφ → G and
ψ : Gφ → H by φ := πG and ψ := πH ◦ Θ. It is routine to check that this determines a
Morita equivalence G
ψ
→ H.
φ
← Gφ
For the final statement, observe that if U is a full open subset of G0, then the argument
of [9, Lemma 6.1] shows that GU = {g ∈ G : s(g) ∈ U} is a G -- GU -equivalence under the
actions determined by multiplication in G. So, writing ∼R for equivalence in the sense
AMPLE GROUPOID HOMOLOGY
11
of Renault, if G and H are weakly Kakutani equivalent, say GU
G ∼R GU
∼= HV , then we have
∼= HV ∼R H. Since ∼R is an equivalence relation, we deduce that G ∼R H. (cid:3)
(i) Recall the definition of R(ψ) from Example 3.9. Proposition 3.10
shows that R(ψ) and X are equivalent, and so C ∗(R(ψ)) is Morita equivalent to
C0(X).
Remarks 3.11.
(ii) It follows from the proof of Proposition 3.10 that if G and H admit isomorphic
ampliations, then there exist a locally compact Hausdorff space X, local homeomor-
phisms φ : X → G(0) and ψ : X → H(0), and an isomorphism Θ : Gφ → Hψ such
that Θ(x, φ(x), x) = (x, ψ(x), x) for all x ∈ X.
(iii) Let G and H be minimal Hausdorff ´etale groupoids which are equivalent in the
sense of Renault. Then with notation as in the Proposition 3.10(2), we may identify
G = LX and H = LY where L is a Hausdorff ´etale groupoid and X and Y are
complementary full clopen subsets of L(0). Let U ⊆ L be an open bisection such
that r(U) ⊆ X and s(U) ⊆ Y . Since L is minimal, both r(U) and s(U) are full
∼= Hs(U ) and so G and H are weakly Kakutani
open subsets. It follows that Gr(U )
equivalent.
Our next result shows that the notions of equivalence in Proposition 3.10 are further
equivalent to a number of additional conditions, including similarity, in the special case
of ample Hausdorff groupoids with σ-compact unit spaces. We write R for the (discrete)
full equivalence relation R = N × N.
Theorem 3.12. Let G and H be ample Hausdorff groupoids with σ-compact unit spaces.
Then the following are equivalent:
(1) G and H are similar;
(2) G and H are Morita equivalent;
(3) there exist an ample Hausdorff groupoid L and a decomposition L(0) = X ⊔ Y of
L(0) into complementary full clopen subsets such that LX
(4) G and H are equivalent in the sense of Renault;
(5) G and H admit isomorphic ampliations;
(6) G × R ∼= H × R;
(7) G and H are Kakutani equivalent; and
(8) G and H are weakly Kakutani equivalent.
∼= G and LY
∼= H;
Proof. (1) =⇒ (2) follows from Lemma 3.6.
Proposition 3.10 shows that (2) -- (5) are equivalent, and [8, Theorem 3.2] shows that(4),
(6), (7) and (8) are equivalent. In particular, we have (1) =⇒ (2) =⇒ · · · =⇒ (8).
For (8) =⇒ (1), suppose that U ⊆ G(0) and V ⊆ H(0) are full open sets with GU
∼=
HV . Matui proves in [32, Theorem 3.6(2)] that G is similar to GU and H is similar to
HV . Since any isomorphism of groupoids is a similarity, and since similarity of groupoids
is an equivalence relation (Remark 3.2), it follows that G and H are similar.
(cid:3)
To close the section, we present an example to show that groupoid equivalence does
not imply either similarity or weak Kakutani equivalence in general. We also show that
weak Kakutani equivalence is not an equivalence relation.
Example 3.13. Let Y := R and X := S1 and define ψ : Y → X by ψ(y) = e2πiy. Then
ψ is a local homeomorphism and the groupoid R(ψ) (see Example 3.9) is equivalent to
the trivial groupoid X (see Remarks 3.11(1)).
12
C. FARSI, A. KUMJIAN, AND PASK, SIMS
We claim that there is no similarity ρ : X → R(ψ).
Indeed suppose that such a ρ
exists. Since ρ is a groupoid map, we have ρ(X (0)) ⊆ R(ψ)(0). Identifying R(ψ)(0) = R
and X (0) = S1 we obtain a continuous map ρ : S1 → R. Since ρ is a similarity it induces
a bijective map on orbits (see Remark 3.3). Since the orbits in X are singletons, this
implies that ρ is injective which is impossible.
We also claim that X and R(ψ) are not weakly Kakutani equivalent. To see this,
suppose that U is a full open subset of X (0). Since X is trivial, we have U = X (0) = S1,
which is not homeomorphic to any open subset of R = R(ψ)(0). So there is no full open
∼= R(ψ)V , and so the two groupoids are not weakly
subset V ⊆ R(ψ)(0) such that XU
Kakutani equivalent.
Consider the local homeomorphism ϕ : X ⊔ Y → X given by
ϕ(z) :=(z
ψ(z)
if z ∈ X,
if z ∈ Y.
Then X and R(ψ) are each weakly Kakutani equivalent to R(ϕ) but as shown above X
is not weakly Kakutani equivalent to R(ψ).
4. Crainic -- Moerdijk -- Matui homology for ample Hausdorff groupoids
Crainic and Moerdijk introduced a compactly supported homology theory for Hausdorff
´etale groupoids in [11]. Matui reframed the theory for ample Hausdorff groupoids (though
he did not explicitly require this; see [32, Definition 3.1]). To use the results of [11] we
must ensure that the standing assumptions of [11, Section 2.5]) are satisfied. We therefore
require that all groupoids we consider henceforth are locally compact, Hausdorff, second
countable, and zero dimensional.
For the reader's convenience we recall Matui's definition of homology for an ample
Hausdorff groupoid G (see [32, Section 3.1]). Since a locally constant sheaf over such a
groupoid with values in a discrete abelian group is c-soft (see Section 2.4), this agrees
with the definition given by Crainic and Moerdijk [11, Section 3.1] under our standing
assumptions.
We first need to establish some notation. Given a locally compact Hausdorff zero-
dimensional space X and a discrete abelian group A, let Cc(X, A) denote the set of
compactly supported A-valued continuous (equivalently, locally constant) functions on
X. Then Cc(X, A) is an abelian group under pointwise addition. Given a (not neces-
sarily surjective) local homeomorphism ψ : Y → X between two such spaces, as in [32,
Section 3.1] we define a homomorphism ψ∗ : Cc(Y, A) → Cc(X, A) by
(4.1)
ψ∗(f )(x) := Xψ(y)=x
f (y)
for all f ∈ Cc(Y, A), and x ∈ X.
If U ⊆ Y is compact open and ψU is injective, then ψ∗(1U ) = 1ψ(U ), where 1U is the
indicator function of U.
Recall that for n > 0, the space of composable n-tuples in a groupoid G is
(4.2)
G(n) = {(g1, . . . , gn) ∈ Gn : s(gi) = r(gi+1) for 1 ≤ i < n},
di(g1, . . . , gn) :=
(g2, . . . , gn)
(g1, . . . , gigi+1, . . . , gn) 1 ≤ i ≤ n − 1,
(g1, . . . , gn−1)
i = n.
i = 0,
while G(0) is the unit space. For n ≥ 2 and 0 ≤ i ≤ n we define di : G(n) → G(n−1) by
AMPLE GROUPOID HOMOLOGY
13
Note that G(n) is 0-dimensional and each di is a local homeomorphism.
Definition 4.1. Let G be a second-countable ample Hausdorff groupoid. For n ≥ 1 define
∂n : Cc(G(n), A) → Cc(G(n−1), A) by
∂1 = s∗ − r∗
and
∂n :=
(−1)i(di)∗
for n ≥ 2,
nXi=0
and define ∂0 to be the zero map from Cc(G(0), A) to 0. Routine calculations show that
this defines a chain complex (Cc(G(∗), A), ∂∗). We define the homology of G with values
in A to be the homology of this complex, denoted H∗(G, A). If A = Z we simply write
H∗(G).
Remark 4.2. An ample groupoid with one unit is just a discrete group. In this instance
the groupoid homology just defined coincides with group homology, see [12, Section 2.22],
and also [5].
Matui shows in [32, Proposition 3.5] that if G and H are similar, then H∗(G, A) ∼=
H∗(H, A) for any discrete abelian group A. So Theorem 3.12 implies that if G and H are
ample and have σ-compact unit spaces and are equivalent via any of the eight notions of
equivalence listed in the statement of the theorem, then their homologies coincide. We
will also an explicit description of the isomorphism.
Lemma 4.3. Let G and H be ample Hausdorff groupoids with σ-compact unit spaces. If
the pair G and H satisfies any of the eight equivalent conditions in Theorem 3.12, then
Hn(G) ∼= Hn(H). In particular, if ρ : G → H is a similarity (see Definition 3.1) then it
induces an isomorphism ρ∗ : Hn(G) ∼= Hn(H). If X is a full open subset of G(0) then the
inclusion GX ⊆ G is a similarity and induces an isomorphism H∗(GX) ∼= H∗(G).
Proof. Crainic and Moerdijk show that a Morita equivalence between Hausdorff ´etale
groupoids induces an isomorphism between their homology groups (see [11, Corollary
4.6]). The proof of [32, Proposition 3.5] shows that if ρ : G → H is a similarity then ρ
induces an isomorphism Hn(G) ∼= Hn(H) for all n ≥ 0. Let X be a full open subset of G(0)
then the argument of [32, Theorem 3.6] proves that the inclusion GX ⊆ G is a similarity.
Hence, the inclusion map induces an isomorphism H∗(GX) ∼= H∗(G).
(cid:3)
Proposition 4.4. Let X be a 0-dimensional space. If we regard X as an ample groupoid
with X (0) = X and with trivial multiplication then
Hn(X) =(Cc(X, Z)
0
if n = 0,
otherwise.
Proof. The boundary maps for the groupoid X are all trivial and there are no nondegen-
erate n-chains for n ≥ 1.
(cid:3)
14
C. FARSI, A. KUMJIAN, AND PASK, SIMS
Remark 4.5.
(1) Following [32, Definition 3.1], if G is an ample Hausdorff groupoid,
then there is a natural preorder on H0(G) determined by the cone
H0(G)+ := {[f ] : f ∈ Cc(G(0), Z) and f (x) ≥ 0 for all x ∈ G(0)}.
(2) For any 0-dimensional space X regarded as a groupoid as in Proposition 4.4, we
have C ∗(X) ∼= C0(X), and
K0(C0(X)) ∼= H0(X) ∼= Cc(X, Z),
via an isomorphism that carries the positive cone of K0(C0(X)) to H0(X)+.
(3) The notion of a type semigroup for the transformation group (X, Γ), where X is a
Cantor set and Γ is discrete was introduced in [47]. This idea was generalised by
Rainone and Sims in [42, Definition 5.4], and independently by Bonicke and Li [3],
who introduced the type semigroup S(G) of an ample Hausdorff groupoid G. The
map [1U ]H0(G) 7→ [1U ]G(S(G)) induces an isomorphism of the homology group H0(G)
onto the Grothendieck group of S(G). This isomorphism carries H0(G)+ to the
image of S(G) in its Grothendieck group. In particular, the coboundary subgroup
HG of [42, Definition 6.4] is exactly im ∂1 as defined above.
Remark 4.6. Homology for ample Hausdorff groupoids is functorial in the following
sense. Let G and H be ample Hausdorff groupoids and let φ : G → H be an ´etale
groupoid homomorphism (so in particular, φ is a local homeomorphism). Then as Crainic
and Moerdijk observe (see [11, 3.7.2]), the maps φ(n)
: Cc(G(n), Z) → Cc(H(n), Z) induce
∗
homomorphisms on homology which we denote by φ∗ : Hn(G) → Hn(H), and φ 7→ φ∗
preserves composition. If G is an open subgroupoid of an ample Hausdorff groupoid H,
then G is also an ample Hausdorff groupoid and the inclusion map ι : G → H is an
´etale groupoid homomorphism. Hence ι induces a map ι∗ : H∗(G) → H∗(H) satisfying
ι∗[1U ]Hn(G) = [1U ]Hn(H).
One key point of the functoriality of homology described in the preceding remark is that
it leads to the following notion of continuity for homology of ample Hausdorff groupoids.
Proposition 4.7 (cf.
{Gi} be an increasing sequence of open subgroupoids of G. Then
[38] Lemma 1.5). Let G be an ample Hausdorff groupoid and let
H∗(G) ∼= lim−→ H∗(Gi).
Proof. For each i the inclusion map Gi ֒→ G induces a homomorphism ιi : H∗(Gi) → H∗(G)
by Remark 4.6. So the universal property of the direct limit yields a homomorphism
ι∞ : lim−→ H∗(Gi) → H∗(G). This homomorphism is injective because if ι∞(a) is a boundary,
say ι∞(a) = ∂n(f ), then f ∈ Cc(G(n+1)
) for large enough i, and then a = ∂n(f ) belongs to
,
are open and nested, it follows that U is a compact open subset of G(n)
(cid:3)
Bn(Gi). It is surjective because if U is a compact open subset of G(n), then U ⊆Si G(n)
i
i
and since the G(n)
for large i. Hence every generator of H∗(G) belongs to the image of ι∞.
i
i
Lemma 4.8. Let X, Y be locally compact Hausdorff spaces, and let ψ : Y → X be a local
homeomorphism. Suppose that there exists a continuous open section ϕ : X → Y of ψ.
Then the groupoid maps ρ : R(ψ) → X given by ρ(y1, y2) = ψ(y1) and σ : X → R(ψ)
Indeed, σ ◦ ρ is similar to idR(ψ)
given by σ(x) = (ϕ(x), ϕ(x)) are both similarities.
and ρ ◦ σ is similar to idX . Moreover, the induced maps ρ∗ : H∗(R(ψ)) → H∗(X) and
σ∗ : H∗(X) → H∗(R(ψ)) are inverse to each other.
AMPLE GROUPOID HOMOLOGY
15
Proof. The first assertion follows from the second. To prove the second assertion, define
θ : R(ψ)(0) → R(ψ) by θ(y, y) = (ϕ ◦ ψ(y), y). Then
σ ◦ ρ(y1, y2)θ(y2, y2) = (ϕ ◦ ψ(y1), ϕ ◦ ψ(y2))(ϕ ◦ ψ(y2), y2) = (ϕ ◦ ψ(y1), y2)
= (ϕ ◦ ψ(y1), y1)(y1, y2) = θ(y1, y1) idR(ψ)(y1, y2).
Hence, σ ◦ ρ is similar to idR(ψ). Since ρ ◦ σ = idX, it follows that ρ ◦ σ is similar to idX.
The last assertion now follows from [32, Proposition 3.5].
(cid:3)
Definition 4.9 (cf. [32, Definition 2.2]). An ample groupoid G is said to be elementary
if it is isomorphic to the groupoid R(ψ) of Example 3.9 for some local homeomorphism
ψ : Y → X between 0-dimensional spaces. An ample groupoid G is said to be AF if it
can be expressed as a union of open elementary subgroupoids.
The only point of difference between Definition 4.9 and Matui's [32, Definition 2.2] is
that we allow non-compact unit spaces.
For the following result, recall that if ψ : Y → X is a local homeomorphism then the
homomorphism ψ∗ : Cc(Y, Z) → Cc(X, Z) is given in (4.1). There is also an inclusion
ι : C0(Y ) ֒→ C ∗(R(ψ)) induced by the homeomorphism Y ∼= R(ψ)(0), and this induces a
homomorphism ι∗ : Cc(Y, Z) → K0(C ∗(R(ψ)).
Theorem 4.10. Let X, Y be locally compact Hausdorff spaces, and let ψ : Y → X be a
local homeomorphism. Then Y is an R(ψ) -- X equivalence with anchor maps id : Y →
R(ψ)(0) and ψ : Y → X, right action of X given by y · ψ(y) = y, and left action given by
(x, y) · y = x. Hence H∗(R(ψ)) ∼= H∗(X).
If Y is σ-compact and totally disconnected, then the map ψ admits a continuous open
section and the map ρ : R(ψ) → X given by ρ(y1, y2) = ψ(y1) is a similarity and thus
induces the isomorphism H0(R(ψ)) ∼= H0(X) = Cc(X, Z) determined by [1U ] 7→ ψ∗(1U )
for U ⊆ Y compact and open. We have Hn(R(ψ)) = 0 for n ≥ 1.
The groupoid C ∗-algebra C ∗(R(ψ)) is an AF algebra, the map ψ∗ induces an isomor-
phism K0(C ∗(R(ψ))) → Cc(X, Z) such that the diagram
Cc(Y, Z)
ι∗
K0(C ∗(R(ψ)))
ψ∗
∼=
Cc(X, Z)
commutes, and we have K1(C ∗(R(ψ))) = {0}.
Proof. For the first statement, one just checks directly that the maps described satisfy
the axioms for an equivalence of groupoids. The isomorphism H∗(R(ψ)) ∼= H∗(X) follows
from Lemma 4.3.
Now suppose that Y is σ-compact and totally disconnected. Then X is also σ-compact
i=1 Ui of Y by countably many compact
open sets. Since ψ is a local homeomorphism and the Ui are compact, each Ui is a finite
union of compact open sets on which ψ is injective, so by relabelling we may assume
that the Ui have this property. For each i, let Vi
and totally disconnected. Choose a cover Y =S∞
j=1 ψ−1(ψ(Uj))(cid:1). Then
X =Fi ψ(Vi), and the Vi are compact open sets on which ψ restricts to a homeomorphism
:= Ui \(cid:0)Si−1
ψi : Vi → ψ(Vi). So we can define a continuous section ϕ for ψ by setting ϕψ(Vi) = ψ−1
:
i
16
C. FARSI, A. KUMJIAN, AND PASK, SIMS
ψ(Vi) → Vi. Hence Lemma 4.8 yields a similarity ρ : R(ψ) → X such that the restriction
of ρ∗ to Cc(Y, Z) coincides with ψ∗ : Cc(Y, Z) → Cc(X, Z).
The equivalence Y of groupoids determines a Morita equivalence between C ∗(R(ψ))
and C0(X) [37, Theorem 2.8]. Since approximate finite dimensionality is preserved by
Morita equivalence and C0(X) is AF, we see that C ∗(R(ψ)) is AF, and the Morita equiv-
alence induces the desired isomorphisms in K-theory. To see that the diagram commutes,
suppose that U ⊆ Y is compact open and that ψU is injective. Then ι∗(1U ) = [1U ] ∈
K0(C ∗(R(ψ))), and this is carried to 1ψ(U ) by the isomorphism K0(C ∗(R(ψ))) → Cc(X, Z)
just described. This is precisely ψ∗(1U ).
(cid:3)
5. Matui's HK Conjecture
In [34, Conjecture 2.6] Matui posed the HK conjecture for a certain class of ample
Hausdorff groupoids. Recall that an ´etale groupoid G is said to be effective if the interior
of its isotropy coincides with its unit space G(0) and minimal if every orbit is dense.
Matui's HK Conjecture. Let G be a locally compact Hausdorff ´etale groupoid such that
G(0) is a Cantor set. Suppose that G is both effective and minimal. Then for j = 0, 1 we
have
(5.1)
Kj(C ∗
r (G)) ∼=
H2i+j(G).
∞Mi=0
We are interested in the extent to which the isomorphism (5.1) holds amongst groupoids
that do not necessarily have non-compact unit space and are not necessarily minimal or
effective. To streamline our discussion, we make the following definition.
Definition 5.1. We define M to be the class of ample Hausdorff groupoids for which the
isomorphism (5.1) holds.
Matui proves in [32, Theorem 4.14] that the groupoids associated to shifts of finite type
belong to M and in [32, Theorems 4.10 and 4.11] that AF groupoids with compact unit
space belong to M. He shows in [34, Proposition 2.7] that M is closed under Kakutani
equivalence, and he shows in [34, Theorem 5.5] that M contains all finite cartesian prod-
ucts of groupoids associated to shifts of finite type. He proves in [32, Section 3.1] that M
contains the transformation groupoids of topologically free and minimal actions of Z on
the Cantor set.
In [21] Hazrat and Li verify (5.1) for j = 0 in the setting of groupoids of row-finite
In [38]
1-graphs with no sinks. We complete the analysis for j = 1 in Theorem 6.7.
Ortega shows that the Katsura-Exel-Pardo groupoid GA,B associated to square integer
matrices with A ≥ 0 belongs to M. Here we consider Deaconu-Renault groupoids and
thereby study higher dimensional aspects not present in other cases.
There are examples of ample Hausdorff groupoids G that belong to M but either do not
have compact unit spaces or which are not necessarily effective or minimal. For example,
if X is any noncompact totally disconnected space, then the groupoid X × Z satisfies none
of these conditions, but belongs to M. It is also easy to show that Zn is in M for all n.
Let Fn denote the free group on n letters. Then by [2, 10.8.1] and [5, II.4.2] we have
K0(C ∗
r (Fn)) ∼= H0(Fn) ∼= Z, K1(C ∗
r (Fn)) ∼= H1(Fn) ∼= Zn
and Hi(Fn) = 0 for all i > 1. Hence Fn lies in M. The integer Heisenberg group also
belongs to M -- see [24, Corollary 1] and [25, Example 8.24]. On the other hand, not
AMPLE GROUPOID HOMOLOGY
17
every ample Hausdorff groupoid belongs to M:
finite cyclic group.
for example, M contains no nontrivial
Here we expand the class of groupoids known to belong to M. We show that all
AF groupoids, all Deaconu -- Renault groupoids associated to actions of N or N2 on 0-
dimensional spaces, and path groupoids associated to many one-vertex k-graphs belong
to M.
By Proposition 4.4 and Remark 4.5, any 0-dimensional space X regarded as a trivial
groupoid belongs to M. More generally, the following corollary to Theorem 4.10 shows
that all AF groupoids belong to M.
Corollary 5.2. Let G be a groupoid that can be expressed as a direct limit G = lim−→
Gn of
open subgroupoids each of which is isomorphic to R(ψn) for some local homeomorphism
ψn : G(0) → Xn. Suppose that G(0) is totally disconnected. Then there are maps ϕn :
Xn → Xn+1 such that ϕn ◦ ψn = ψn+1 for all n. We have Hn(G) = 0 for n ≥ 1, and
H0(G) ∼= lim−→(H0(Xn), (ϕn)∗).
open U ⊆ G(0), and we have K1(C ∗(G)) = {0}. In particular, G belongs to M.
There is an isomorphism K0(C ∗(G)) ∼= H0(G) that carries [1U ]0 to [1U ] for each compact
Proof. Each R(ψn)(0) is totally disconnected because it is an open subspace of G(0), and
so Theorem 4.10 shows that Hp(R(ψn)) = 0 for p ≥ 1 and all n, and that H0(R(ψn)) ∼=
Cc(Xn, Z) ∼= K0(C ∗(R(ψn))) with both isomorphisms induced by (ψn)∗. Since homology
and K-theory are continuous with respect to inductive limits, the result follows.
(cid:3)
6. Deaconu -- Renault groupoids
In this section we first show that the homology of an ample higher-rank Deaconu-
Renault groupoid G(X, σ) is given by H∗(Zk, H0(G(X, σ) ×c Zk)) using spectral sequence
arguments of Matui. In Theorem 6.5 we describe a complex which allows us to compute
H∗(Zk, H0(G(X, σ) ×c Zk) by adapting techniques from [16]. We then use this description
and Kasparov's K-theory spectral sequence to prove that G(X, σ) belongs to M when the
rank is either one or two (see Theorems 6.7, 6.10). Furthermore we also give formulas for
computing the K-theory of C ∗(G(X, σ)) in these cases.
Recall from Section 2 that if σ is an action of Nk on a locally compact Hausdorff space
defines an isomorphism of the skew-product groupoid G(X, σ) ×c Zk corresponding to the
X by local homeomorphisms, then we write eX := X × Zk, and there is an action σ of Nk
by local homeomorphisms on eX given by σq(x, p) = (σq(x), p + q). Equation (2.1) then
cocycle c(x, m, y) = m onto the Deaconu -- Renault groupoid G(eX, σ).
Our first result shows that G(X, σ) ×c Zk is equivalent to c−1(0); this in turn allows us
to compute its homology.
Lemma 6.1. Let X be a locally compact Hausdorff totally disconnected space, and let
σ be an action of Nk on X by local homeomorphisms. The set X × {0} ⊆ X × Zk
groupoid.
is a clopen G(eX, σ)-full subspace of G(eX, σ)(0). The map (x, 0, y) 7→ ((x, 0), 0, (y, 0)) is
an isomorphism of c−1(0) ⊆ G(X, σ) onto G(eX, σ)X×{0}, and G(X, σ) ×c Zk is an AF
There is an isomorphism of H∗(c−1(0)) onto H∗(G(eX, σ)) that carries [1U ] to [1U ×{0}]
for every compact open U ⊆ X.
18
C. FARSI, A. KUMJIAN, AND PASK, SIMS
(x, 0) ∈ X × {0} and s(γ) = (y, n), proving the claim.
The map (x, 0, y) 7→ ((x, 0), 0, (y, 0)) is clearly an injective homomorphism. To show
We claim that it is full. Fix (x, n) ∈ X × Zk and write n = n+ − n− where n+, n− ∈ Nk.
Then there exists y ∈ X such that σn−(y) = σn+(x). Set γ = ((x, 0), n+ − n−, (y, n)).
Proof. It is clear that X × {0} is a clopen subset of X × Zk, the unit space of G(eX, σ).
By construction we have σn+(x, 0) = σn−(y, n), and so γ ∈ G(eX, σ). Furthermore r(γ) =
that it is surjective, let γ = ((x, m), m − n, (y, n)) ∈ G(eX, σ)X×{0}. Then m = n = 0, so
Since G(X, σ) ×c Zk ∼= G(eX, σ) via the isomorphism given in (2.1), and since X ×
{0} is G(eX, σ)-full, it follows from Theorem 3.12 that G(X, σ) ×c Zk is equivalent to
G(eX, σ)X×{0}. By the preceding paragraph G(eX, σ)X×{0}
∼= c−1(0) . Since c−1(0) can be
written as an increasing union of the elementary groupoids R(σn), it is AF (see Corol-
lary 5.2), and so G(X, σ) ×c Zk is an AF groupoid also.
γ is in the range of c−1(0).
The final statement follows from Lemma 4.3.
(cid:3)
To compute the homology of c−1(0), we decompose it as the increasing union of the
subgroupoids R(σn) as n ranges over Nk.
Lemma 6.2. Let X be a totally disconnected locally compact Hausdorff space, and let σ
be an action of Nk on X by surjective local homeomorphisms. There is an isomorphism
lim−→(Cc(X, Z), σn
(1U ) to [1U ] for every compact open U ⊆
X. We have
∗ ) → H0(c−1(0)) that takes σ0,∞
∗
Hq(G(X, σ) ×c Zk) ∼=(lim−→n∈Nk(Cc(X, Z), σn
0
∗ )
if q = 0
otherwise.
∗ ) induced by the σn
∗ .
The isomorphism H0(G(X, σ) ×c Zk) ∼= lim−→n∈Nk (Cc(X, Z), σn
∗ ) intertwines the action of Zk
on H0(G(X, σ) ×c Zk) given by p · ((x, m, y), n) = ((x, m, y), n + p) with the action of Zk
on lim−→n∈Nk(Cc(X, Z), σn
Proof. If U, V ⊆ X are compact open sets on which σn is injective, and if we have σn(U) =
σn(V ), then we have [1U ] = [1V ] in H0(R(σn)). Since every W ⊆ X can be expressed
as a finite disjoint union W =Fj σn(Uj) where each Uj is compact open and each σnUj
is injective, it follows that there is a unique homomorphism ϕn : Cc(X, Z) → H0(R(σn))
such that ϕn(σn
∗ (1U )) = [1U ] for all compact open U. This ϕn is the map induced by the
weak equivalence of groupoids ϕn : R(σn) → X (see Remark 3.11), so Proposition 3.10
and Lemma 4.3 imply that it is an isomorphism. For m, n ∈ Nk, let ιm,n be the inclusion
map R(σm) ֒→ R(σm+n), and let (ιm,n)∗ : H0(R(σm)) → H0(R(σm+n)) be the induced
map in homology. Then we have a commuting diagram
Cc(X, Z)
σn
∗
Cc(X, Z)
ϕm
ϕm+n
H0(R(σm))
(ιm,n)∗
H0(R(σm+n))
AMPLE GROUPOID HOMOLOGY
19
Since c−1(0) is the increasing union of the closed subgroupoids R(σn), continuity of ho-
mology gives H0(c−1(0)) ∼= lim−→(H0(R(σm)), ι∗). So the universal properties of the direct
limits prove the first statement.
The isomorphism (2.1) of G(X, σ) ×c Zk with G(eX,eσ) and Lemma 6.1 show that
H∗(G(X, σ) ×c Zk) ∼= H∗(c−1(0)). So the preceding paragraph proves that H0(G(X, σ) ×c
Zk) ∼= lim−→n∈Nk(Cc(X, Z), σn
∗ ). We saw in the preceding paragraph that c−1(0) is an AF
groupoid, so Corollary 5.2 proves that Hq(G(X, σ) ×c Zk) = 0 for q ≥ 1. The final
statement follows from direct computation of the maps involved.
(cid:3)
Matui's spectral sequence [32, Theorem 3.8(1)] relates H∗(G(X, σ)) to the homology of
Zk with coefficients in H∗(G(X, σ) ×c Zk). Since Hq(G(X, σ) ×c Zk)) = 0 for q ≥ 1 (it is
AF by Lemma 6.2), the spectral sequence collapses: It follows that
Hq(G(X, σ)) ∼= Hq(Zk, H0(G(X, σ) ×c Zk)) for 0 ≤ q ≤ k.
The proof of the following lemma is based on the technique developed in [16, Section 3].
elements εi1 ∧ · · · ∧ εip indexed by integer tuples 1 ≤ i1 < i2 < · · · < ip ≤ k. We define
Recall that for 1 ≤ p ≤ k, the module Vp Rk is the free R-module generated by the
V0 Rk := R.
Lemma 6.3. Let A be an abelian group, and suppose that σ1, . . . , σk are pairwise com-
muting endomorphisms of A (i.e. an action of Nk). For 1 ≤ p ≤ k, define
on spanning elements (εi1 ∧ · · · ∧ εip) ⊗ a in which the ij are in strictly increasing order by
∂p :Vp Zk ⊗ A →Vp−1 Zk ⊗ A
1 ⊗ (id −σi1)a
for p > 1,
for p = 1.
∂p(εi1 ∧ · · · ∧ εip ⊗ a) =(Pj(−1)j+1εi1 ∧ · · · ∧bεij ∧ · · · ∧ εip ⊗ (id −σij )a
Then ∂p−1 ◦ ∂p = 0 for each p, so that (V∗ Zk ⊗ A, ∂∗) is a complex. For each 0 ≤ i ≤ k,
the homomorphism id ⊗σi :V∗ Zk ⊗ A →V∗ Zk ⊗ A commutes with ∂∗, and the induced
Proof. Direct computation on a spanning element (εi1 ∧ · · · ∧ εip+1) ⊗ a ofV∗ Zk ⊗ A using
For the final statement, we first claim that for x ∈Vp Zk ⊗ A and 1 ≤ l ≤ k, we have
that the σi commute shows that ∂p+1 ◦ ∂p = 0. Clearly the id ⊗σi commute with ∂∗.
map (id ⊗σi)∗ in homology is the identity map.
∂p+1(εl ∧ x) = −εl ∧ ∂p(x) + (id ⊗(id −σl))(x).
To prove this, it suffices to consider x = (εi1 ∧ · · · ∧ εip) ⊗ a where 1 ≤ i1 < · · · < ip ≤ k.
So fix such an x and fix l ≤ k. We consider two cases.
Case 1: l 6= ih for all h. Then there exists 0 ≤ j ≤ p such that ih < l for h ≤ j and
ih > l for h > j. Then, using at both the first and the third steps that εl ∧ εi1 ∧ · · · ∧ εin =
20
C. FARSI, A. KUMJIAN, AND PASK, SIMS
(−1)nεi1 ∧ · · · ∧ εin ∧ εl for any i1 < i2 < · · · < in, we calculate:
∂p+1(εl ∧ x)
= (−1)j∂p+1(εi1 ∧ · · · ∧ εij ∧ εl ∧ εij+1 ∧ · · · ∧ εp ⊗ a)
+
=
+ (−1)j+2εi1 ∧ · · · ∧ εij ∧ εij+1 ∧ · · · ∧ εp ⊗ (id −σl)a
= (−1)jh jXh=1
(−1)h+1εi1 ∧ · · · ∧bεih ∧ · · · ∧ εij ∧ εl ∧ εij+1 ∧ · · · ∧ εp ⊗ (id −σih)a
(−1)h+2εi1 ∧ · · · ∧ εij ∧ εl ∧ εij+1 ∧ · · · ∧bεih ∧ · · · ∧ εp ⊗ (id −σih)ai
pXh=j+1
(−1)hεl ∧ εi1 ∧ · · · ∧bεih ∧ · · · ∧ εp ⊗ (id −σih)a
(−1)hεi1 ∧ · · · ∧bεih ∧ · · · ∧ εp ⊗ (id −σih)a!
pXh=1
= εl ∧ pXh=1
+ (id ⊗(id −σl))(εi1 ∧ · · · ∧ εip ⊗ a)
+ εi1 ∧ · · · ∧ εp ⊗ (id −σl)a
= −εl ∧ ∂p(x) + (id ⊗(id −σl))(x).
Case 2: l = ih for some h. Then ∂p+1(εl ∧ x) = ∂p+1(0) = 0. So we must show that
εl ∧ ∂p(x) = (id ⊗(id −σl))(x). We have
pXj=1
The terms corresponding to ij 6= l are zero, so this collapses to
εl ∧ ∂p(x) =
εl ∧ (−1)j+1εi1 ∧ · · · ∧bεij ∧ · · · ∧ εp ⊗ (id −σij )(a).
εl ∧ ∂p(x) = (−1)h+1εl ∧ εi1 ∧ · · · ∧bεih ∧ · · · ∧ εip ⊗ (id −σih)(a)
= εi1 ∧ · · · ∧ εip ⊗ (id −σl)(a)
= (id ⊗(id −σl))(x).
This completes the proof of the claim.
We now prove the final statement. Fix x ∈ Vp Zk ⊗ A such that ∂p(x) = 0, and fix
l ≤ p. We just have to show that (id ⊗(id −σl))(x) ∈ image(∂p+1). Since ∂p(x) = 0 and
using the claim, we see that
(id ⊗(id −σl))(x) = (id ⊗(id −σl))(x) − εl ∧ ∂p(x) = ∂p+1(εl ∧ x),
and the result follows.
(cid:3)
Lemma 6.4. Let A be an abelian group, and suppose that σ1, . . . , σk are pairwise com-
muting endomorphisms of A (i.e. an action of Nk). Let ∂p :V∗ Zk ⊗ A →V∗ Zk ⊗ A be as
in Lemma 6.3. Let eA := lim−→Nk(A, σn). For i ≤ k let σi be the automorphism of eA induced
by σi, and let ∂p :V∗ Zk ⊗eA →V∗ Zk ⊗eA be the boundary map obtained from Lemma 6.3
applied to eA and the σi. Then the canonical homomorphism σ0,∞ : A → eA corresponding
to the 0th copy of A induces an isomorphism H∗(V∗ Zk ⊗ A) ∼= H∗(V∗ Zk ⊗ eA). Moreover,
σ extends to an action of Zk on eA, and H∗(V∗ Zk ⊗ eA) ∼= H∗(Zk, eA).
AMPLE GROUPOID HOMOLOGY
21
Proof. Since the homology functor is continuous (see, for example, [52, Theorem 4.1.7])
For the second statement, we follow the argument of [16, Lemma 3.12] (see also [29,
Theorem 5.5]). Let G := Zk = hs1, . . . , ski and let R := ZG. For p ≥ 2, we define
we have H∗(V∗ Zk ⊗ eA) ∼= lim−→(cid:0)H∗(V∗ Zk ⊗ A), (id ⊗σn)∗(cid:1). Lemma 6.3 shows that this is
equal to lim−→(cid:0)H∗(V∗ Zk ⊗ A), id(cid:1) = H∗(V∗ Zk ⊗ A).
∂p :Vp Rk →Vp−1 Rk by
Define ∂1 : V1 Rk → R by ∂1(εj) := 1 − sj, and let η : R → Z be the augmentation
(−1)j+1(1 − sij )εi1 ∧ · · · ∧bεij ∧ · · · ∧ εip.
∂p(εi1 ∧ · · · ∧ εip) =Xj
homomorphism determined by η(si) = 1 for each i. Then
η
−→ Z
is a free resolution of Z. Hence, by definition of homology with coefficients in the Zk-
0 −→Vk Rk ∂k−→ · · · ∂2−→V1 Rk ∂1−→ R
module eA [5, Equation III(1.1)], we have H∗(Zk, eA) ∼= H∗(V∗ Rk ⊗R eA). As a group we
haveVp Rk ⊗R eA ∼=V∗ Zk ⊗ eA, and this isomorphism intertwines the boundary maps in
the complex defining H∗(V∗ Rk ⊗R eA) with the maps e∂p.
We can now state our main theorem for this section, which is a computation of the
homology of the Deaconu -- Renault groupoid G(X, σ) associated to an action of Nk by
surjective local homeomorphisms of a totally disconnected locally compact space X.
(cid:3)
Theorem 6.5. Let X be a second-countable totally disconnected locally compact space,
and let σ be an action of Nk by surjective local homeomorphisms σp : X → X. For
1 ≤ p ≤ k, let
and let Aσ
p = {0} for p > k. For p ≥ 1, define ∂p : Aσ
p → Aσ
p−1 by
Aσ
p =Vp Zk ⊗ Cc(X, Z)
id ⊗(id −σ
ei1
∗ )f
pXj=1
0
∂p(ǫi1 ∧ · · · ∧ ǫip ⊗ f ) =
(−1)j+1ǫi1 ∧ · · · ∧bǫij ∧ · · · ∧ ǫip ⊗ (id −σ
eij
∗ )f
if p = 1,
if 2 ≤ p ≤ k,
if p ≥ k + 1.
Then (Aσ
∗ , ∂∗) is a complex, and
H∗(G(X, σ)) ∼= H∗(Zk, H0(G(X, σ) ×c Zk)) ∼= H∗(Aσ
∗ , ∂∗).
We have Hp((Zk, H0(G(X, σ) ×c Zk)) = 0 for p > k.
Proof. Lemma 6.3 implies that (Aσ
∗ , ∂∗) is a complex.
The automorphisms αp : ((x, m, y), n) 7→ ((x, m, y), n + p) of G(X, σ) ×c Zk induce an
actioneαp of Zk on H∗(G(X, σ) ×c Zk). This actioneα makes each Hp(G(X, σ) ×c Zk) into a
Zk-module, so it makes sense to discuss the homology groups H∗(Zk, Hq(G(X, σ) ×c Zk))
of Zk with coefficients in these modules. By [32, Theorem 3.8(1)] applied to the cocycle
c : G(X, σ) → Zk, there is a spectral sequence Er
p,q converging to Hp+q(G(X, σ)) satisfying
E2
p,q = Hp(Zk, Hq(G(X, σ) ×c Zk)). Lemma 6.2 shows that Hq(G(X, σ) ×c Zk) is zero
for q 6= 0, and therefore the differential maps on the E2 page and above of the spectral
22
C. FARSI, A. KUMJIAN, AND PASK, SIMS
sequence are trivial. Hence E∞
Hp(G(X, σ)) ∼= E2
p,0
∼= Hp(Zk, H0(G(X, σ) ×c Zk)) for all p.
p,q = E2
p,q for all p, q and so [32, Theorem 3.8(1)] shows that
The last statement of Lemma 6.2 gives H∗(Zk, H0(G(X, σ)×cZk)) ∼= H∗(Zk, lim−→(X, σn
Lemma 6.4 shows that this is isomorphic to H∗(Aσ
∗ , ∂∗).
∗ )).
(cid:3)
Our next two results show that G(X, σ) belongs to M if k ≤ 2. For k = 1 we use the
Pimsner -- Voiculescu sequence [40, Theorem 2.4]. To prove them, we need the following
lemma, which follows from a standard argument.
Lemma 6.6. Let X be a second-countable locally compact totally disconnected space. Let σ
be an action of Nk on X by surjective local homeomorphisms. Let c : G(X, σ) → Zk be the
cocycle c(x, m, y) = m. Then the action of Zk on G(X, σ) ×c Z given by αp((x, m, y), n) =
((x, m, y), n + p) induces an action ¯α of Zk on C ∗(G(X, σ) ×c Z). The crossed product
C ∗(G(X, σ) ×c Z) ⋊¯α Zk is stably isomorphic to C ∗(G(X, σ)).
Proof. Every automorphism of a groupoid induces an automorphism of its C ∗-algebra, and
then simpler calculations establish the first statement. Let γ : Tk → Aut(C ∗(G(X, σ)))
be the gauge action γz(f (x, m, y)) = zmf (x, m, y). Then there is an isomorphism θ :
C ∗(G(X, σ) ×c Zk) → C ∗(G(X, σ)) ×γ Tk such that for f ∈ Cc(G(X, σ) × {m}) ⊆
Cc(G(X, σ) ×c Zk), we have
(cid:0)θ(f )(z)(cid:1)(g) = zmf (g, m).
This θ intertwines ¯α and the dual actionbγ on C ∗(G(X, σ)) ×γ Tk, and so Takesaki -- Takai
duality implies that C ∗(G(X, σ) ×c Zk) ⋊¯α Zk is stably isomorphic to C ∗(G(X, σ)) (see [54,
Theorem 4.5], [53, Theorem 3.4], and [7, Theorem 1.2]).
(cid:3)
Theorem 6.7. Let X be a second-countable locally compact totally disconnected locally
compact space, let σ : X → X be a surjective local homeomorphism, and let σ∗ :
Cc(X, Z) → Cc(X, Z) be the induced map. Then
K0(C ∗(G(X, σ)) ∼= H0(G(X, σ)) ∼= coker(id −σ∗),
K1(C ∗(G(X, σ)) ∼= H1(G(X, σ)) ∼= ker(id −σ∗),
and Hn(G(X, σ)) = 0 for n ≥ 2.
In particular, G(X, σ) belongs to M.
Proof. We first calculate the K-theory of C ∗(G(X, σ)). Lemma 6.6 applied with k = 1
shows that K∗(C ∗(G(X, σ))) ∼= K∗(C ∗(G(X, σ) ×c Z) ⋊¯α Z). Corollary 5.2 shows that
K1(C ∗(G(X, σ) ×c Z)) = 0, and so exactness of the Pimsner -- Voiculescu sequence [40,
Theorem 2.4] implies that
(6.1)
K0(C ∗(G(X, σ))) ∼= coker(id −¯α∗)
and K1(C ∗(G(X, σ))) ∼= ker(id −¯α∗).
We now compute the homology of G(X, σ). Theorem 6.5 shows that Hp(G(X, σ)) ∼=
Hp(Z, H0(G(X, σ) ×c Z)) for all p. Moreover Hp(Z, H0(G(X, σ) ×c Z)) = 0 for p ≥ 2 by
Theorem 6.5.
Let α∗ be the action of Z on H0(G(X, σ) ×c Z) induced by α. By [5, Example III.1.1],
(6.2)
H0(Z, H0(G(X, σ) ×c Z)) ∼= ker(id −α∗),
H1(Z, H0(G(X, σ) ×c Z)) ∼= coker(id −α∗).
and
AMPLE GROUPOID HOMOLOGY
23
Recall from Lemma 6.1 that G(X, σ)×c Z is an AF groupoid. The isomorphism between
H0(G(X, σ) ×c Z) and K0(C ∗(G(X, σ) ×c Z)) supplied by Corollary 5.2 intertwines the Z-
actions α∗ and ¯α∗. Thus coker(id −¯α∗) ∼= coker(id −α∗) and ker(id −¯α∗) ∼= ker(id −α∗).
Hence by (6.1) and (6.2),
K0(C ∗(G(X, σ)) ∼= H0(G(X, σ))
and K1(C ∗(G(X, σ)) ∼= H1(G(X, σ)).
Since k = 1 here, the complex Aσ
∗ of Theorem 6.5 reduces to
0 −→ Cc(X, Z) −→ Cc(X, Z) −→ 0
where the central map is id −σ∗. Then we have
H0(G(X, σ)) ∼= H0(Aσ
H1(G(X, σ)) ∼= H1(Aσ
∗ , ∂∗) ∼= coker(id −σ∗), and
∗ , ∂∗) ∼= ker(id −σ∗)
as required.
(cid:3)
Remark 6.8. Matui's paper [32], together with recent results by Hazrat and Li [21]
and Ortega [38] suggest that the groupoids of (not necessarily finite) 1-graphs with no
sources belong to M. This now follows from Theorem 6.7, since graph groupoids are, by
definition, rank-1 Deaconu -- Renault groupoids.
We now discuss Kasparov's K-theory spectral sequence for C ∗(G(X, σ)). Lemma 6.6
shows that C ∗(G(X, σ)) is stably isomorphic to the crossed product C ∗(G(X, σ) ×c Zk) ׯα
Zk. Hence Theorem 6.10 of [23] (see also [16, §3]) shows that there is a spectral sequence
Er
p,q converging to K∗(C ∗(G(X, σ))) with E2-page given by
p,q =(Hp(Zk, Kq(C ∗(G(X, σ)) ×c Zk))
0
E2
if q is even, and 0 ≤ p ≤ k
otherwise.
The differential maps in the spectral sequence are maps dr
then for any p, q at least one of Er
only for 0 ≤ p ≤ k. Hence dr
can improve on this:
because Er
particular, if k = 2, then we have E∞
p,q is trivial for r > k. Thus E∞
if r = k is even, then at least one of Er
p,q is nontrivial only for q even, and it follows that E∞
p,q and Er
p,q = E2
p,q.
p−r,q+r−1. If r > k,
p,q is nontrivial
p,q . If k is even, we
p−r,q+r−1 is trivial
p,q for all p, q. In
p,q = Ek+1
p,q and Er
p,q = Ek
p,q : Er
p,q → Er
p−r,q+r−1 is trivial, because Er
For our next theorem, we need the well-known fact that if X is locally compact Haus-
dorff space, then Cc(X, Z) is a free abelian group. We provide a proof for completeness.
Lemma 6.9. Let X be a second-countable locally compact Hausdorff space. Then Cc(X, Z)
is a free abelian group.
Proof. First note that if X is not compact, then C(X ∪ {∞}, Z) ∼= Cc(X, Z) ⊕ Z via
f 7→ (f − f (∞)1, f (∞)), so it suffices to prove the result for X compact.
So suppose that X is compact. Since X is metrisable by the Urysohn metrisation
theorem (see [56, Theorems 23.1 and 17.6(a)]), the Alexandroff -- Hausdorff theorem (see
[56, Theorem 30.7]) shows that there is a continuous surjection φ : {0, 1}∞ → X. Hence
φ∗ : C(X, Z) → C({0, 1}∞, Z) is an injective group homomorphism. Since subgroups
of free abelian groups are themselves free abelian,
it therefore suffices to show that
C({0, 1}∞, Z) is free abelian.
For this, let {0, 1}∗ denote the collection of all finite words in the symbols 0, 1, including
the empty word ε. Let I = {ε} ∪ {ω1 : ω ∈ {0, 1}∗} denote the subset of {0, 1}∗
24
C. FARSI, A. KUMJIAN, AND PASK, SIMS
consisting of the empty word and all nontrivial words that end with a 1. We claim
that B := {1Z(ω) : ω ∈ I} is a family of free abelian generators of C({0, 1}∞, Z). To
see this, we first argue by induction on n that spanZ{1Z(ω) : ω ∈ I and ω ≤ n} =
spanZ{1Z(ω) : ω ∈ {0, 1}n} for all n. The containment ⊆ is trivial. The containment ⊇
is also trivial for n = 0, and if it holds for n = k, then for each ω = ω′0 ∈ {0, 1}k+1
that ends in a 0, we have 1Z(ω) = 1Z(ω′) − 1Z(ω′1). We have ω′1 ∈ I with ω′1 = k + 1,
and 1Z(ω′) ∈ spanZ{1Z(ω) : ω ∈ I and ω ≤ k} by the inductive hypothesis. So the
containment ⊇ also holds for n = k + 1. Hence B generates C({0, 1}∞, Z) as a group.
To see that B is a family of free generators, suppose for contradiction that F ⊆ I is a
of minimal length. Then µ000 · · · ∈ {0, 1}∞ belongs to Z(µ) but not to Z(ω) for any
finite set and {aω : ω ∈ F } are nonzero integers such thatPω∈F aω1Vω = 0. Fix µ ∈ F
ω ∈ F \ {µ}. Hence 0 =(cid:0)Pω∈F aω1Vω(cid:1)(µ000 · · · ) = aµ contradicting the assumption that
the aω are nonzero.
(cid:3)
Theorem 6.10. Let X be a second-countable locally compact totally disconnected space.
Let σ be an action of N2 on X by surjective local homeomorphisms. Define d2 : Cc(X, Z) →
Cc(X, Z) ⊕ Cc(X, Z) by d2(f ) = ((σe2
∗ )f ) and define d1 : Cc(X, Z) ⊕
Cc(X, Z) → Cc(X, Z) by d1(f ⊕ g) = (id −σe1
∗ − id)f, (id −σe1
∗ )f + (id −σe2
∗ )g. Then
K0(C ∗(G(X, σ)) ∼= H0(G(X, σ)) ⊕ H2(G(X, σ)) ∼= coker(d1) ⊕ ker(d2),
K1(C ∗(G(X, σ)) ∼= H1( G(X, σ)) ∼= ker(d1)/ image(d2).
and
In particular, G(X, σ) belongs to M.
Proof. Let A := K0(C ∗(G(X, σ) ×c Z2)) = H0(G(X, σ) ×c Z2). Lemma 6.6 applied with
k = 2 shows that C ∗(G(X, σ)) is stably isomorphic to C ∗(G(X, σ) ×c Z2) ׯα Z2 for the
action ¯α induced by translation in Z2 in the skew-product G(X, σ) ×c Z2.
We follow the argument of [16, 29]. As discussed above, Kasparov's spectral sequence
[23, Theorem 6.10] for K∗(C ∗(G(X, σ) ×c Z2) ׯα Z2) converges on the second page, and
we deduce that K0(C ∗(G(X, σ)) is an extension of E2
2,0 while K1(C ∗(G(X, σ)) is
isomorphic to E2
0,2 by E2
0,1.
As discussed prior to the statement of the theorem, E2
p,q is isomorphic to the homology
group Hp(Z2, K0(C ∗(G(X, σ))×c Zk)) for q even and is zero for q odd. Corollary 5.2 shows
that K0(C ∗(G(X, σ)) ×c Z2) ∼= H0(G(X, σ) ×c Z2), and that this isomorphism intertwines
the actions of Z2 on the two groups induced by translation in the second coordinate. It
therefore follows from Theorem 6.5 that for q even, we have E2
∗ , ∂∗). Since
p,q
Cc(X, Z) is free abelian by Lemma 6.9, so is the subgroup H2(Aσ
∗ , ∂∗) = ker(∂2). Hence
2,0 splits, and we obtain K0(C ∗(G(X, σ)) ∼=
the extension K0(C ∗(G(X, σ)) of E2
H0(Aσ
∗ , ∂∗). The result then follows
j) for
from Theorem 6.5 because the obvious identificationsVj Z2 ⊗ C(X, Z) ∼= C(X, Z)(2
∗ , ∂∗) and K1(C ∗(G(X, σ))) ∼= H1(Aσ
0 ≤ j ≤ 2 intertwine ∂∗ with d∗.
∗ , ∂∗) ⊕ H2(Aσ
∼= Hp(Aσ
0,2 by E2
(cid:3)
It follows that the path groupoid of a 2-graph belongs to M since it is a rank-2 Deaconu-
Renault groupoid (see Corollary 7.7).
Question 6.11. Our proof Theorem 6.10 uses that H2(Aσ
∗ ) is a free abelian group so
that the extension 0 → H0(G(X, σ)) → K0(C ∗(G(X, σ))) → H2(G(X, σ)) → 0 splits.
Hence the map from H0(G(X, σ)) ⊕ H2(G(X, σ)) to K0(C ∗(G(X, σ))) that we obtain is
not natural. An interesting question arises: can the isomorphism (5.1) be chosen to be
AMPLE GROUPOID HOMOLOGY
25
natural in some sense for elements of M in general, and for rank-2 Deaconu -- Renault
groupoids in particular?
Remark 6.12. The proof of Theorem 6.10 is special to the situation k = 2, and issues
arise already when k = 3. In this situation, the groups on the E3-page of Kasparov's
spectral sequence coincide with those on the E2-page, but the E3-page has potentially
nontrivial differential maps, d3
∗ ) if q is even, and is 0
if q is odd, and the E3-page has the following form.
0,2l+2. So E3
p,q = Hp(Aσ
3,2l : E3
3,2l → E3
...
0
...
0
...
0
...
0
H0(Aσ
∗ )
H1(Aσ
∗ )
H2(Aσ
∗ )
H3(Aσ
∗ )
0
0
d3
3,0
0
0
H0(Aσ
∗ )
H1(Aσ
∗ )
H2(Aσ
∗ )
H3(Aσ
∗ )
...
0
0
0
0
. . .
. . .
. . .
. . .
The sequence converges on the E4-page, and so we have exact sequences
0 → coker(d3
0 → H1(Aσ
3,0) →K0(C ∗(G(X, σ))) → H2(Aσ
∗ ) →K1(C ∗(G(X, σ))) → ker(d3
∗ ) → 0
3,0) → 0.
and
3,0 is trivial, there is no reason to expect that coker(d3
∗ ); and even if d3
Unless d3
3,0
is trivial, there is no reason to expect that H2(Aσ
∗ ) is free abelian, so the extension defining
K0(C ∗(G(X, σ))) need not split. This suggests rank-3 Deaconu -- Renault groupoids as a
potential source of counterexamples to Matui's HK-conjecture.
3,0) ∼= H0(Aσ
Remark 6.13. If the groups H∗(Aσ
∗ ) are finitely generated, and the natural homomor-
phism H0(G(X, σ)) → K0(C ∗(G(X, σ))) is injective, then one would expect d3
3,0 to be
trivial, and then in the rank-3 case, G(X, σ) would satisfy Matui's conjecture up to sta-
bilisation by Q (the so-called rational HK-conjecture). This suggests that it would be
worthwhile to investigate when the homomorphism H0(G(X, σ)) → K0(C ∗(G(X, σ))) is
injective.
7. k-graphs
In this section, we first establish the existence of a natural map from the homology
of a k-graph to the homology of its groupoid. We show that this homomorphism is in
general neither injective nor surjective. We then apply the results of Section 6 to see
that all 1-graph and 2-graph groupoids belong to M. Finally, we restrict our attention to
k-graphs with one vertex, and demonstrate that for any such k-graph in which gcd(Λe1−
1, . . . , Λek − 1) = 1, the corresponding k-graph groupoid belongs to M.
26
C. FARSI, A. KUMJIAN, AND PASK, SIMS
7.1. A map from the categorical homology of a k-graph to the homology of its
groupoid. To define the categorical 3 homology groups H∗(Λ) for a k-graph Λ, we use
the following notation.
Given a k-graph Λ, let
For n ≥ 2, and for i ∈ {0, . . . , n} define a map di : Λ∗
n → Λ∗
n−1 by
Λ∗ n :=(Λ0
{(λ1, . . . , λn) ∈Qn
di(λ1, . . . , λn) :=
i=1 Λ s(λi) = r(λi+1)}
if n = 0
otherwise.
(λ2, . . . , λn)
(λ1, λ2, . . . , λiλi+1, . . . , λn)
(λ1, . . . , λn−1)
if i = 0,
if 0 < i < n,
if i = n;
Definition 7.1. (cf. [19, Remark 2.14]) Let Λ be a k-graph. For n ≥ 0, let Cn(Λ) :=
ZΛ∗ n, the free abelian group with generators indexed by Λ∗ n.
Identifying elements of
Λ∗ n with the corresponding generators of Cn(Λ), we regard the boundary maps di as
: Cn(Λ) → Cn−1(Λ). We obtain homomorphisms ∂n : Cn(Λ) →
homomorphisms di
i=1(−1)idi for n ≥ 2. Regarding s, r as maps from C1(Λ) to C0(Λ)
define ∂1 : C1(Λ) → C0(Λ) by ∂1 := s − r and ∂0 : C0(Λ) → {0} to be the zero map.
Standard calculations show that (C∗(Λ), ∂∗) is a complex. The resulting groups Hn(Λ) :=
ker(∂n)/ image(∂n+1) are called the categorical homology groups of Λ.
Cn−1(Λ) by ∂n :=Pn
Note that one may define a more general homology theory with coefficients in an abelian
group A as in [19] but we do not need this level of generality.
Example 7.2. Let Λ be the 1-graph with vertex connectivity matrix ( 5 2
2 3 ). By [19,
Theorem 6.2] the categorical homology of Λ coincides with its cubical homology, which
can computed as follows. Since Λ is connected we have H0(Λ) ∼= Z; since Λ is finite its first
homology group is free abelian with rank equal to its Betti number p = Λ1−Λ0+1 = 11.
Hence H1(Λ) ∼= Z11. Moreover, Hn(Λ) = 0 for all n > 1. We will return to this example in
Remark 7.8 after establishing how to compute the homology of the associated groupoid.
Remark 7.3. One can check that C0(Λ) together with the subgroups of the Cn(Λ) for
n ≥ 1 generated by elements (λ1, · · · , λn) in which each λi 6∈ Λ0 form a subcomplex under
the same boundary maps ∂n, and that the homology of this subcomplex is isomorphic to
H∗(Λ).
Recall that if Λ is a k-graph and λ, µ ∈ Λ satisfy s(λ) = s(µ), then the cylinder set
Z(λ, µ) is defined as Z(λ, µ) = {(λx, d(λ) − d(µ), µx) : x ∈ s(λ)Λ∞} ⊆ GΛ, and is a
compact open subset of GΛ.
Definition 7.4. Let Λ be a k-graph. Let GΛ be the associated groupoid (see (2.2))
and for each n let G(n)
Λ be the collection of composable n-tuples in GΛ as in (4.2). For
(λ1, . . . , λn) ∈ Λ∗ n ⊆ Cn(Λ), define
Y (λ1, . . . , λn) :=(cid:0)Qn
i=1 Z(λi, s(λi))(cid:1) ∩ G(n)
Λ .
Let Ψ∗ : C∗(Λ) → Cc(G(∗)
Λ , Z) be the homomorphism such that Ψ0(v) = 1Z(v) ∈
Λ , Z) for v ∈ Λ0, and Ψn(λ1, . . . , λn) = 1Y (λ1,...,λn) for n ≥ 1 and (λ1, . . . , λn) ∈ Λ∗ n.
Cc(G(0)
3so called because it matches with the categorical cohomology of Λ defined in [30].
AMPLE GROUPOID HOMOLOGY
27
Theorem 7.5. Let Λ be a k-graph. The maps Ψ∗ : C∗(Λ) → Cc(G(∗)
comprise a chain map, and induce a homomorphism Ψ∗ : H∗(Λ) → H∗(GΛ).
Λ , Z) defined above
Proof. It suffices to prove that Ψ∗ intertwines the boundary maps on generators of C∗(Λ).
Fix λ ∈ Λ = Λ∗ 1 ⊆ C1(Λ). Then
∂1(Ψ1(λ)) = ∂1(1Y (λ)) = s∗(1Y (λ)) − r∗(1Y (λ)) = 1Z(s(λ)) − 1Z(r(λ))
= Ψ0(s(λ)) − Ψ0(r(λ)) = Ψ0(∂1(λ)).
For n ≥ 2, it suffices to prove that given an element (λ1, . . . , λn) ∈ Λ∗ n and any 0 ≤ i ≤ n,
we have
(7.1)
(di)∗(Ψn(λ1, . . . , λn)) = Ψn(di(λ1, . . . , λn)).
In the following calculation, given sets Z1, . . . , Zn ⊆ GΛ we define
Z1 ∗ Z2 ∗ · · · ∗ Zn := (Z1 × Z2 × · · · × Zn) ∩ G(n)
Λ .
Note that Y (λ1, . . . , λn) = Z(λ1, s(λ1)) ∗ Z(λ2, s(λ2)) ∗ · · · ∗ Z(λn, s(λn)). First suppose
that 1 ≤ i ≤ n − 1. Then
(di)∗(Ψn(λ1, . . . , λn)) = (di)∗(cid:0)1Y (λ1,...,λn)(cid:1) = (di)∗(cid:0)1Z(λ1,s(λ1))∗Z(λ2,s(λ2))∗···∗Z(λn,s(λn))(cid:1)
= 1Z(λ1,s(λ1))∗···∗Z(λi,s(λi))Z(λi+1,s(λi+1))∗···∗Z(λn,s(λn)),
and
Ψn(di(λ1, . . . , λn)) = Ψn(λ1, . . . , λiλi+1, . . . , λn) = 1Y (λ1,...,λiλi+1,...,λn)
= 1Z(λ1,s(λ1))∗···∗Z(λiλi+1,s(λi+1))∗···∗Z(λn,s(λn)).
Since the cylinder sets Z(λi, s(λi)) and Z(λi+1, s(λi+1)) are both bisections and since
s(Z(λi, s(λi))) = Z(s(λi)) = r(Z(λi+1, s(λi+1))), we have Z(λi, s(λi))Z(λi+1, s(λi+1)) =
Z(λiλi+1, s(λi+1)), and (7.1) follows.
Now consider i = 0 (the case i = n is very similar). We have
(d0)∗(Ψn(λ1, . . . , λn)) = (d0)∗(cid:0)1Y (λ1,...,λn)(cid:1)
= (d0)∗(cid:0)Z(λ1, s(λ1)) ∗ Z(λ2, s(λ2)) ∗ · · · ∗ Z(λn, s(λn))(cid:1)
= 1Z(λ2,s(λ2))∗···∗Z(λn,s(λn))
= 1Y (λ2,··· ,λn)
= Ψn(λ2, · · · , λn)
= Ψn(d0(λ1, . . . , λn)).
(cid:3)
In general the map Ψ∗ is neither injective nor surjective: see Remark 7.8.
7.2. The HK conjecture for 1-graph groupoids and 2-graph groupoids. In this
subsection we apply the results of §6 to groupoids associated to 1-graphs and 2-graphs.
Recall from §2 that if Λ is a k-graph, then there is an action σ of Nk by endomorphisms
on its infinite-path space Λ∞ and that the k-graph C ∗-algebra coincides with the C ∗-
algebra C ∗(G(Λ∞, σ)) of the associated Deaconu -- Renault groupoid (see [27]). We begin
this section by showing that the homology of G(Λ∞, σ) as computed in Theorem 6.5
coincides with the homology of the complex DΛ
∗ used by Evans [16] to compute the K-
theory of C ∗(Λ).
The complex (DΛ
∗ , ∂∗) is given as follows. We write ε1, . . . , εk for the generators of Zk,
and we write εv for the generators of ZΛ0. We write Mj ∈ MΛ0(Z) for the vertex matrix
28
C. FARSI, A. KUMJIAN, AND PASK, SIMS
given by Mj(v, w) = vΛejw, which we regard as an endomorphism of ZΛ0. For p ≥ 0 we
define DΛ
p = {0} for p > k. For p ≥ 1
we define ∂p : DΛ
p =Vp Zk ⊗ ZΛ0 and we define DΛ
p−1 by ∂p = 0 if p > k,
0 = ZΛ0, and DΛ
p → DΛ
pXj=0
∂p(εi1 ∧ · · · ∧ εip ⊗ εv) =
if 2 ≤ p ≤ k and
(−1)j+1εi1 ∧ · · · ∧bεij ∧ · · · ∧ εip ⊗ (I − M t
ij )εv
∂1(εi ⊗ εv) = (I − M t
i )εv.
∗ and the homology of the complex Aσ
In the following proposition, we establish an isomorphism between the homology of
Evans' complex DΛ
∗ associated to the shift maps
σn on Λ∞. We could obtain the existence of isomorphisms H∗(DΛ
∗ ) using
∗ ) ∼= H∗(Zk, K0(C ∗(Λ ×d Zk))), that by Matui's results,
that, by Evans' results, H∗(DΛ
∗ ) ∼= H∗(Zk, H0(GΛ ×c Zk)), and then by identifying C ∗(Λ ×d Zk) with C ∗(GΛ ×c Zk)
H∗(Aσ
and applying the HK conjecture for AF groupoids as stated in Corollary 5.2. However,
we have chosen to present a more direct proof, which also has the advantage that it shows
that the natural inclusion ZΛ0 ֒→ Cc(Λ∞, Z) induces the isomorphism.
∗ ) ∼= H∗(Aσ
Proposition 7.6. Let Λ be a row-finite k-graph with no sources. Let GΛ = G(Λ∞, σ) be the
associated groupoid, and let (Aσ
∗ ) be the complex of Theorem 6.5. Then the homomorphism
ι : ZΛ0 → Cc(Λ∞, Z) determined by ι(1v) = 1Z(v) induces an isomorphism H∗(Aσ
∗ ) →
H∗(DΛ
∗ ). In particular, H∗(G(Λ∞, σ)) ∼= H∗(DΛ
∗ ).
direct limit of the Aσ
inclusion of Aσ
∗ under the maps induced by the σ∗. Lemma 6.4 shows that the
Proof. Corollary 5.2 shows that we can express the complex V∗ Zk ⊗ H0(c−1(0)) as the
∗ inV∗ Zk ⊗ H0(c−1(0)) induces an isomorphism in homology. Similarly, in
[16, Theorem 3.14] Evans proves that one can express the complexV∗ Zk⊗K0(C ∗(Λ×dZk))
as the direct limit of DΛ
∗ that
takes 1v to the class of pv,0 in the direct limit induces an isomorphism in homology. By
[27, Theorem 5.2], there is an isomorphism C ∗(Λ×d Zk) ∼= C ∗(G(Λ∞, σ)×c Zk) that carries
p(v,0) to 1Z(v)×{0}. So Corollary 5.2 shows that there is an isomorphism K0(C ∗(Λ×d Zk)) ∼=
H0(G(Λ∞, σ)×cZk) that takes [p(v,0)] to [1Z(v)×{0}]. Lemma 6.1 therefore implies that there
is an isomorphism K0(C ∗(Λ ×d Zk)) ∼= H0(c−1(0)) induced by the map that carries the
class of pv,0 to 1Z(v).
∗ under the maps induced by the M t
∗, and the inclusion of DΛ
Given v ∈ Λ0 and n ∈ Nk, we have
(7.2)
σn
∗ (ι(1v)) = σn
1Z(s(λ)) = ι(M t
n(1v)).
∗ (1Z(v)) = Xλ∈vΛn
σn
∗ (1Z(λ)) = Xλ∈vΛn
Hence ι induces a map of complexes ι∗ : DΛ
n)∗.
The same computation combined with the universal property of the direct limit shows
that we obtain the following commuting diagram.
∗ that intertwines the σn
∗ with the (M t
∗ → Aσ
M t
∗
σn
∗
DΛ
∗
ι∗
Aσ
∗
DΛ
∗
ι∗
Aσ
∗
. . .
. . .
V∗ Zk ⊗ H0(c−1(0))
V∗ Zk ⊗ H0(c−1(0))
ι∞
∗
AMPLE GROUPOID HOMOLOGY
29
By definition, ι∗([1v]) = [1Z(v)] for every v ∈ Λ0. Write (M t
copy of DΛ
∗ intoV∗ Zk ⊗ H∗(c−1(0)) and write σn,∞
toV∗ Zk ⊗ H∗(c−1(0)). Then (M t
(ι∗(1v)) = σn,∞
(1Z(v)) = σn,∞
σn,∞
∗
(σn
∗ (1Z(µ))) = σn,∞
∗
∗
∗
n,∞)∗(1v) = [1Z(µ)] for any µ ∈ Λnv. Since
n,∞)∗ for the map from the nth
for the map from the nth copy of Aσ
∗
∗
(1Z(µ)) = [1Z(µ)],
we see by commutativity of the diagram that ι∞
1Z(µ) generate Cc(Λ∞, Z), we deduce that ι∞
∗
the identity map in homology. Since the maps H∗(DΛ
H∗(Aσ
the functoriality of homology implies that ι∗ induces an isomorphism H∗(DΛ
∗ ) → H∗(V∗ Zk ⊗ H0(c−1(0))) and
∗ ) → H∗(V∗ Zk ⊗ H0(c−1(0))) are isomorphisms, and the diagram above commutes,
∗ (1Z(µ)) = [1Z(µ)] for all µ. Since the
is the identity map, and therefore induces
∗ ) → H∗(Aσ
The final statement follows from Theorem 6.5.
∗ ).
(cid:3)
Though we already know that graph groupoids belong to M by Remark 6.8, the fol-
lowing result goes a step further, computing the homology of the 1-graph and 2-graph
groupoids in terms of the vertex matrices of the 1-graph or 2-graph. Recall that given
a k-graph Λ and given i ≤ k we write Mi for the Λ0 × Λ0 integer matrix given by
Mi(v, w) = vΛeiw. If Λ is the path category of a directed graph E, then M1 is just the
usual adjacency matrix AE of E.
Corollary 7.7 (see [16, Proposition 3.16]).
(1) Let E be a row-finite graph with no
sources. Then
K0(C ∗(E)) ∼= H0(GE) ∼= coker(I − At
K1(C ∗(E)) ∼= H1(GE) ∼= ker(I − At
E).
E)
and
(2) Let Λ be a row-finite 2-graph with no sources. Then
2 − I) ∩ ker(I − M t
1),
K0(C ∗(Λ)) ∼= H0(GΛ) ∼= ZΛ0/(cid:0)(I − M t
and
1, I − M t
K1(C ∗(Λ)) ∼= H1(GΛ) ∼= ker(I − M t
1, I − M t
2)(ZΛ0)2(cid:1) ⊕ ker(M t
2)/(cid:0) (M t
1)(cid:1)ZΛ0.
2−I)
(I−M t
In particular, graph groupoids and 2-graph groupoids belong to M.
Proof. Theorems 6.7 and 6.10 establish the isomorphisms between homology and K-
theory. The descriptions of the homology groups follow from Proposition 7.6 and the
definition of the complex DΛ
∗ .
(cid:3)
Remark 7.8. The strongly connected 1-graph Λ described in Example 7.2 has homology
given by
By Corollary 7.7 the homology of GΛ is
Z
Z11
0
if n = 0,
if n = 1,
otherwise.
Hn(Λ) =
Hn(GΛ) =(Z/2Z ⊕ Z/2Z if n = 0,
0
otherwise.
So the map Ψ0 : H0(Λ) → H0(GΛ) of Theorem 7.5 is neither surjective nor injective.
30
C. FARSI, A. KUMJIAN, AND PASK, SIMS
Remark 7.9. In [16, Proposition 3.18], Evans shows that if Λ is a 3-graph and {(I−M t
i )a :
1 ≤ i ≤ 3, a ∈ ZΛ0} generates ZΛ0, then the K-theory of C ∗(Λ) is equal to the homology
of DΛ. So the groupoids of such k-graphs belong to M.
We also see that if Λ is a 3-graph or a 4-graph for which the page 3 differentials in
Kasparov's sequence are zero and H2(DΛ) (and H3(DΛ) in the case of a 4-graph) are
free abelian, then K∗(C ∗(Λ)) is determined by H∗(DΛ) and so GΛ belongs to M; but of
course the hypothesis on the differential maps in Kasparov's sequence are not checkable
in practice.
Remark 7.10. Suppose that Λ is a finite 3-graph. Then the homology groups H∗(DΛ
∗ )
are finite rank, and H3(DΛ
∗ ) is free abelian. Consequently, if it were possible to construct
an example of a finite 3-graph for which the page 3 differential d3
3,0 in Kasparov's spectral
sequence was nontrivial, then consideration of the ranks of the groups involved would show
that the associated groupoid did not satisfy the HK conjecture, even up to stabilisation
by Q.
7.3. One vertex k-graphs. In this section we will show that if Λ is a 1-vertex k-graph
in which each Λei ≥ 2 and in which gcd(Λe1 − 1, . . . , Λek − 1) = 1, then K∗(C ∗(Λ)) =
H∗(GΛ) = 0. We work with row-finite k-graphs throughout, but we include a comment at
the end of the section indicating how to extend our K-theory calculation to non-row-finite
k-graphs. A similar result has been proved in [1, Theorem 6.4(a)] under the assumption
that the elements of {Λei : 1 ≤ i ≤ k} are pairwise relatively prime.
The key point is the following consequence of Matui's Kunneth formula for the groupoid
homology of an ample Hausdorff groupoid (see [34, Theorem 2.4]).
Theorem 7.11. Let Λ be a row-finite single-vertex k-graph with at least two edges of each
colour, and write Ni := Λei − 1 for each i ≤ k. Then
Hn(GΛ) ∼=((Zgcd(N1,...,Nk))(k−1
0
n )
if 0 ≤ n ≤ k − 1
otherwise.
Proof. We proceed by induction. This follows from Corollary 7.7(1) if k = 1. Suppose
it holds for k = K − 1 and that Λ is a K-graph with one vertex and at least two edges
of each colour. Since the complex DΛ
∗ is independent of the factorisation rules in Λ,
Proposition 7.6 shows that H∗(GΛ) is independent of the factorisation rules. So we can
theorem gives a split exact sequence
i=1 GBNi+1 and H = GBNK +1 so that GΛ
assume that Λ = BN1+1 × · · · × BNk+1 and so GΛ =Qk
Write G = QK−1
Hi(G) ⊗ Hj(H) −→ Hn(G × H) −→ Mi+j=n−1
0 → Mi+j=n
i=1 GBNi+1.
∼= G × H. Matui's Kunneth
Tor(Hi(G), Hj(H)) → 0,
and since H∗(H) = (ZNK , 0, 0, . . . ) this collapses to a split exact sequence
0 → Hn(G) ⊗ ZNK −→ Hn(G × H) −→ Tor(Hn−1(G), ZNK ) → 0.
Since the sequence splits, the middle term is the direct sum of the two ends. Write γ :=
(K−2
n )
gcd(N1, . . . , NK−1). Then the inductive hypothesis gives Hn(G) = Z
γ
and Hn−1(G) =
(K−2
n−1)
Z
γ
. Also, gcd(γ, NK) = gcd(N1, . . . , NK). As ⊗ and Tor are both additive in the first
AMPLE GROUPOID HOMOLOGY
variable and Zl ⊗ Zm = Zgcd(l,m) = Tor(Zl, Zm), we have
Hn(G × H) = (Zgcd(γ,NK ))(K−2
n ) ⊕ (Zgcd(γ,NK ))(K−2
n−1)
= (Zgcd(N1,...,NK ))(K−2
= (Zgcd(N1,...,NK ))(K−1
n ).
n )+(K−2
n−1)
31
(cid:3)
We deduce that the groupoids of 1-vertex k-graphs in which there are at least two edges
of each colour, and in which gcd(Λe1 − 1, . . . , Λek − 1) = 1 belong to M.
Corollary 7.12. If Λ is a single-vertex k-graph with at least two edges of each colour,
and gcd(Λe1 − 1, . . . , Λek − 1) = 1, then K∗(C ∗(Λ)) = H∗(GΛ) = 0. In particular, if
C ∗(Λ) is simple then it is isomorphic to O2.
Proof. Theorem 7.11 shows that H∗(GΛ) = 0. It follows that the groups E2
p,q in Matui's
spectral sequence are all zero. Since the terms F 2
p,q in Kasparov's spectral sequence [23]
(see also [16]) are given by E2
p,0 if q is even, and 0 if q is odd, we deduce that the F 2
p,q
are all zero. So Evans' spectral sequence collapses, and we obtain K∗(C ∗(Λ)) = 0 as well.
The final statement follows from [48, Proposition 8.8 and Corollary 8.15].
(cid:3)
Remark 7.13. Unfortunately, if gcd(N1, . . . , Nk) > 1, we can conclude little new about
the HK conjecture. The problem is that in K-theory, with the exception of tensor prod-
ucts, we only obtain from Evans' spectral sequence that the K-groups have filtrations of
length at most k − 1 with subquotients equal to direct sums of copies of Zgcd(N1,...,Nk).
Remark 7.14. The above discussion deals only with row-finite k-graphs. We can extend
the K-theory calculation to non-row-finite examples as follows. If Λ is any 1-vertex k-
Γ (Nk) ∼= Λ and Γek+1 is infinite, then
graph, and Γ is a 1-vertex (k + 1)-graph such that d−1
as in [6] we can make the identification
C ∗(Γ) ∼= Tℓ2(C ∗(Λ))C∗ (Λ),
and deduce from Pimsner's [39, Theorem 4.4] that the inclusion C ∗(Λ) ֒→ C ∗(Γ) deter-
mines a KK-equivalence, so K∗(C ∗(Γ)) ∼= K∗(C ∗(Λ)).
Applying this iteratively, we can compute the K-theory of the C ∗-algebra of a 1-vertex,
not-necessarily-row-finite k-graph as the K-theory of the C ∗-algebra of the subgraph con-
sisting only of those coordinates in which there are finitely many edges.
References
[1] S. Barlak, T. Omland, and N. Stammeier, On the K-theory of C ∗-algebras arising from integral
dynamical systems, Ergod. Thy. & Dynam. Sys. 38 (2018), 832 -- 862.
[2] B. Blackadar, K-theory for operator algebras. MSRI Publications vol. 5, Cambridge University Press,
1998.
[3] C. Bonicke and K. Li, Ideal sructure and pure infiniteness of ample groupoid C ∗-algebras, Ergodic
Theory Dynam. Systems, to appear.
[4] G.E. Bredon, Sheaf theory, Second edition. Graduate Texts in Mathematics, 170. Springer-Verlag,
New York, 1997.
[5] K.S. Brown, Cohomology of groups, Graduate Texts in Mathematics, vol. 87, Springer-Verlag, New
York, 1994, Corrected reprint of the 1982 original.
[6] J.H. Brown, L.O. Clark, A. Sierakowski and A. Sims, Purely infinite simple C ∗-algebras that are
principal groupoid C ∗-algebras, J. Math. Anal. Appl. 439 (2016), 213 -- 234.
32
C. FARSI, A. KUMJIAN, AND PASK, SIMS
[7] R. Brown, P. Green and M. Rieffel, Stable isomorphism and strong Morita equivalence of C ∗-algebras
Pacific J. Math., 71 (1977), 349 -- 363.
[8] T. Carlsen, E. Ruiz, and A. Sims, Equivalence and stable isomorphism of groupoids, and diagonal-
preserving stable isomorphisms of graph C ∗-algebras and Leavitt path algebras, Proc. Amer. Math.
Soc. 145 (2017), 1581 -- 1592.
[9] L.O. Clark and A. Sims, Equivalent groupoids have Morita equivalent Steinberg algebras, J. Pure
Appl. Algebra 219 (2015), 2062 -- 2075.
[10] G. Cornelissen, O. Lorscheid, and M. Marcolli, On the K-theory of graph C ∗-algebras, Acta Appl.
Math. 102 (2008), 57 -- 69.
[11] M. Crainic and I. Moerdijk, A homology theory for ´etale groupoids, J. reine angew. Math. 521 (2000),
25 -- 46.
[12] M. Crainic, Cyclic cohomology of ´etale groupoids: the general case, K-Theory 17 (1999), 319 -- 362.
[13] V. Deaconu, Groupoids associated with endomorphisms, Trans. Amer. Math. Soc. 347 (1995), 1775 --
1786.
[14] R. Deeley, I.F. Putnam and K. Strung, Constructing minimal homeomorphisms on point-like spaces
and a dynamical presentation of the Jiang -- Su algebra, J. reine. angew. Math., to appear.
[15] A. Dimca, Sheaves in topology, Universitext. Springer-Verlag, Berlin, 2004.
[16] D.G. Evans, On the K-theory of higher rank graph C ∗-algebras, New York J. Math. 14 (2008), 1 -- 31.
[17] R. Exel, Inverse semigroups and combinatorial C ∗-algebras, Bull. Braz. Math. Soc. (N.S.) 39 (2008),
191 -- 313.
[18] R. Exel and J.N. Renault, Semigroups of local homeomorphisms and interaction groups., Ergod. Thy.
& Dynam. Sys., 27 (2007), 1737 -- 1771.
[19] E. Gillaspy and J. Wu, Isomorphism of the cubical and categorical cohomology groups of a higher-rank
graph, preprint 2018 (arXiv:1807.02245 [math.OA]).
[20] T. Giordano, I.F. Putnam and C.F. Skau, Topological orbit equivalence and C ∗-crossed products, J.
reine angew. Math. 469 (1995), 51 -- 111.
[21] R. Hazrat and H. Li, Homology of ´etale groupoids a graded appropach, preprint 2018
(arXiv:1806.03398 [math.RA]).
[22] M. Kashiwara and M., Schapira, Sheaves on Manifolds, Grund. der math. Wiss., 292, Springer,
Berlin Heidelberg New York (1990).
[23] G.G. Kasparov, Equivariant KK-theory and the Novikov conjecture, Invent. Math. 91 (1988), 147 --
201.
[24] K. Kodaka, K-theory for the C ∗-algebras of the discrete Heisenberg group, Tokyo J. Math. 9 (1986),
365 -- 372.
[25] K. Knudson, Homology of linear groups, Progress in Mathematics, 193. Birkhauser Verlag, Basel,
2001.
[26] A. Kumjian, Preliminary algebras arising from local homeomorphisms, Math. Scand. 52 (1983),
269 -- 278.
[27] A. Kumjian and D. Pask, Higher rank graph C ∗-algebras, New York J. Math. 6 (2000), 1 -- 20.
[28] A. Kumjian, D. Pask, I. Raeburn and J.N. Renault, Graphs, groupoids, and Cuntz -- Krieger algebras,
J. Funct. Anal. 144 (1997), 505 -- 541.
[29] A. Kumjian, D. Pask and A. Sims, C ∗-algebras associated to coverings of k-graphs, Doc. Math. 13
(2008), 161 -- 205.
[30] A. Kumjian, D. Pask and A. Sims, On twisted higher rank graph C ∗-algebras, Trans. Amer. Math.
Soc. 367 (2015), 5177 -- 5216.
[31] A. Kumjian, D. Pask, and A. Sims, Homology for higher-rank graphs and twisted C ∗-algebras, J.
Funct. Anal 263 (2012), 1539 -- 1574.
[32] H. Matui, Homology and topological full groups of ´etale groupoids on totally disconnected spaces,
Proc. Lond. Math. Soc. (3) 104 (2012), 27 -- 56.
[33] H. Matui, Topological full groups of one-sided shifts of finite type, J. reine angew. Math. 705 (2015),
35 -- 84.
[34] H. Matui, ´Etale groupoids arising from products of shifts of finite type, Adv. Math. 303 (2016),
502 -- 548.
AMPLE GROUPOID HOMOLOGY
33
[35] S. MacLane, Homology, Reprint of the first edition. Die Grundlehren der mathematischen Wis-
senschaften, Band 114. Springer-Verlag, Berlin-New York, 1967.
[36] I. Moerdijk and J. Mrcun, Introduction to foliations and Lie groupoids, Cambridge Studies in Ad-
vanced Mathematics, 91 Cambridge University Press, Cambridge, 2003. x+173 pp.
[37] P.S. Muhly, J.N. Renault and D.P. Williams, Equivalence and isomorphism for groupoid C ∗-algebras,
J. Operator Theory 17 (1987), 3 -- 22.
[38] E. Ortega, Homology of
the Katsura -- Exel -- Pardo groupoid, preprint 2018 (arXiv:1806.09297
[math.OA]).
[39] M.V. Pimsner, A class of C ∗-algebras generalizing both Cuntz -- Krieger algebras and crossed products
by Z, Fields Inst. Commun., 12, Free probability theory (Waterloo, ON, 1995), 189 -- 212, Amer.
Math. Soc., Providence, RI, 1997.
[40] M.V. Pimsner and D. Voiculescu, Exact sequences for K-groups and Ext-groups of certain cross-
products of C ∗-algebras, J. Operator Theory 4 (1980), 93 -- 118.
[41] I.F. Putnam, Some classifiable groupoid C ∗-algebras with prescribed K-theory, Math. Ann. 370
(2018), 1361 -- 1387.
[42] T. Rainone and A. Sims, A dicothomy for groupoid C ∗-algebras, Ergod. Thy. & Dynam. Sys., to
appear.
[43] A. Ramsay, Virtual groups and group actions, Advances in Math. 6 (1971) 253 -- 322.
[44] J.N. Renault, A groupoid approach to C ∗-algebras, Lecture Notes in Mathematics, 793, Springer,
Berlin, 1980.
[45] J.N. Renault, C ∗-algebras of groupoids and foliations. Operator algebras and applications, Part I
(Kingston, Ont., 1980), pp. 339 -- 350, Proc. Sympos. Pure Math., 38, Amer. Math. Soc., Providence,
R.I., 1982.
[46] M. Rørdam, F. Larsen and N. Laustsen, An introduction to K-theory for C ∗-algebras, Cambridge
University Press, Cambridge, 2000, xii+242.
[47] M. Rørdam and A. Sierakowski, Purely infinite C ∗-algebras arising from crossed products, Ergod.
Theory & Dynam. Sys. 32 (2012), 272 -- 293.
[48] A. Sims, Gauge-invariant ideals in the C ∗-algebras of finitely aligned higher-rank graphs, Canad. J.
Math. 58 (2006), 1268 -- 1290.
[49] A. Sims, Hausdorff ´etale groupoids and their C ∗-algebras, to appear in Operator algebras and dy-
namics: groupoids, crossed products and Rokhlin dimension (F. Perera, Ed.) in Advanced Courses
in Mathematics. CRM Barcelona, Birkhauser.
[50] A. Sims and D. Williams, Renault's equivalence theorem for reduced groupoid C ∗-algebras, J. Operator
Theory 68 (2012), 223 -- 239.
[51] A. Sims and D.P. Williams, The primitive ideals of some ´etale groupoid C ∗-algebras, Algebr. Repre-
sent. Theory 19 (2016), 255 -- 276.
[52] E.H. Spanier, Algebraic topology, McGraw-Hill Book Co., New York, 1966.
[53] H. Takai, On duality for crossed products of C ∗-algebras, J. Funct. Anal, 19 (1975) 25 -- 39.
[54] M. Takesaki, Duality for crossed products and the structure of von Neumann algebras of type III
Acta Math. 131 (1973), 249 -- 310.
[55] N.E. Wegge-Olsen, K-theory and C ∗-algebras, A friendly approach, The Clarendon Press Oxford
University Press, New York, 1993, xii+370.
[56] S. Willard, General topology, Reprint of the 1970 original
[Addison-Wesley, Reading, MA;
MR0264581], Dover Publications, Inc., Mineola, NY, 2004, xii+369.
[57] D.P. Williams, Haar systems of equivalent groupoids, Proc. Amer. Math. Soc. (Ser B) 3 (2016), 1 -- 8.
34
C. FARSI, A. KUMJIAN, AND PASK, SIMS
(C.F.) Department of Mathematics, Campus Box 395, Boulder, Colorado 80309-0395,
USA
E-mail address: [email protected]
(A.K) Department of Mathematics (084), University of Nevada, Reno, NV 89557-0084,
USA
E-mail address: [email protected]
(D.P, &A.S.) School of Mathematics and Applied Statistics, University of Wollongong,
NSW 2522, Australia
E-mail address: [email protected], [email protected]
|
1810.02650 | 2 | 1810 | 2018-12-13T11:32:51 | Simulating acculturation dynamics between migrants and locals in relation to network formation | [
"cs.MA"
] | International migration implies the coexistence of different ethnic and cultural groups in the receiving country. The refugee crisis of 2015 has resulted in critical levels of opinion polarization on the question of whether to welcome migrants, causing clashes in receiving countries. This scenario emphasizes the need to better understand the dynamics of mutual adaptation between locals and migrants, and the conditions that favor successful integration. Agent-based simulations can help achieve this goal. In this work, we introduce our model MigrAgent and our preliminary results. The model synthesizes the dynamics of migration intake and post-migration adaptation. It explores the different acculturation outcomes that can emerge from the mutual adaptation of a migrant population and a local population depending on their degree of tolerance. With parameter sweeping, we detect how different acculturation strategies can coexist in a society and in different degrees among various subgroups. The results show higher polarization effects between a local population and a migrant population for fast intake conditions. When migrant intake is slow, transitory conditions between acculturation outcomes emerge for subgroups, e.g., from assimilation to integration for liberal migrants and from marginalization to separation for conservative migrants. Relative group sizes due to speed of intake cause counterintuitive scenarios, such as the separation of liberal locals. We qualitatively compare the processes of our model with the German portion sample of the survey Causes and Consequences of Socio-Cultural Integration Processes among New Immigrants in Europe (SCIP), finding preliminary confirmation of our assumptions and results. | cs.MA | cs | Simulating Acculturation Dynamics Between Migrants and Locals in Relation to
Network Formation
Rocco Paolillo
Bremen International Graduate School of Social Sciences
University of Bremen & Jacobs University Bremen
Germany
Wander Jager
Groningen Center for Social Complexity Studies
University College Groningen, University of Groningen
the Netherlands
Abstract
International migration implies the coexistence of different ethnic and cultural groups in the receiving
country. The refugee crisis of 2015 has resulted in critical levels of opinion polarization on the
question of whether to welcome migrants, causing clashes in receiving countries. This scenario
emphasizes the need to better understand the dynamics of mutual adaptation between locals and
migrants, and the conditions that favor successful integration. Agent-based simulations can help
achieve this goal. In this work, we introduce our model MigrAgent and our preliminary results. The
model synthesizes the dynamics of migration intake and post-migration adaptation. It explores the
different acculturation outcomes that can emerge from the mutual adaptation of a migrant population
and a local population depending on their degree of tolerance. With parameter sweeping, we detect
how different acculturation strategies can coexist in a society and in different degrees among various
subgroups. The results show higher polarization effects between a local population and a migrant
population for fast intake conditions. When migrant intake is slow, transitory conditions between
acculturation outcomes emerge for subgroups, e.g., from assimilation to integration for liberal
migrants and from marginalization to separation for conservative migrants. Relative group sizes due
to speed of intake cause counterintuitive scenarios, such as the separation of liberal locals. We
qualitatively compare the processes of our model with the German portion sample of the survey
"Causes and Consequences of Socio-Cultural Integration Processes among New Immigrants in
Europe" (SCIP), finding preliminary confirmation of our assumptions and results.
Keyword list: acculturation, migration, tolerance, polarization, agent-based simulation
1
Introduction
International migration implies individuals or entire populations crossing international boundaries
(International Organization for Migration [IOM], 2017). It is a complex phenomenon common in
human history that connects people from different cultures and historical backgrounds.
Understanding the consequences of international migration for receiving societies has become more
important considering the increase in migration flows to Europe and other global destinations since
2015, namely due to the refugee crisis (European Commission [EC], 2015; Organization for
Economic Co-operation and Development [OECD], 2015). At the peak of the crisis, as reported by
the European Commission (2015), the United Nations High Commissioner for Refugees [UNHCR]
confirmed that 1,015,078 migrants had reached Europe via the Mediterranean Sea in 2015. As
Eurostat (2017) has reported, EU countries approved 307,650 asylum requests during 2015, and the
number increased in 2016 to 672,890 requests accepted. In 2016, Germany approved 433,905
requests, while Sweden approved 66,585 and Italy 35,405. Although the peak of the refugee crisis is
likely in the past (UNHCR, 2017), factors such as climate change might increase migration pressure
in the future due to floods, droughts and associated conflicts. For example, Kelley, Mohtadi, Cane,
Saeger and Kushnir (2015) have linked the conflict in Syria and the associated refugee crisis to
unprecedented droughts in the region.
The consequences of such migration for receiving countries are evident. A survey by the Pew
Research Center (Wike, Stokes & Simmons, 2016) has demonstrated opinion polarization in
European countries due to the refugee crisis. For example, the discussions on Brexit focused on,
among other topics, the sovereignty of the UK in the context of refugee acceptance. On the one hand,
many Europeans consider refugees as unwilling to integrate and as representing a threat to the stability
and economy of their nations (Wike et al., 2016). On the other hand, many support the urgent calls to
open borders and accept refugees as a form of humanitarian aid (Wike et al., 2016). This internal
chasm regarding the acceptance of migrants is more evident than at earlier stages of European history
(OECD, 2015). Additionally, many asylum seekers who have been granted the status of humanitarian
migrants are likely to settle in Europe in the long term. As social scientists, we are interested in the
consequences of this scenario as regards the cohesion of the receiving societies and the integration of
migrants.
Facing these issues requires one to consider the dynamics of acculturation in the context of continuous
first-hand contact between members of different cultures (Redfield, Linton, & Herskovits, 1936). A
prominent model in the field of acculturation studies is the fourfold model proposed by Berry (1984,
2005), which identifies acculturation as a mutually adaptive process between a local community and
migrant minorities. It assumes that acculturation occurs at two levels: (1) a cultural-group level made
up of norms and (2) a psychological-individual level concerning the individual experience of
adaptation (Berry, 1984, 2005). The model is based on the orientations people adopt to face issues
related to maintaining their own cultural identities and participating in the larger society. Four
orientations emerge from the intersection of the two distinct dimensions:
2
●
Integration: high maintenance of one's own culture, frequent interactions with the other group
● Assimilation: low maintenance of one's own culture, frequent interactions with the other
group
● Separation: high maintenance of one's own culture, infrequent interactions with the other
group
● Marginalization: low maintenance of one's own culture, infrequent interactions with the other
group
The most recent elaborations of the model have associated societal ideologies with the different
orientations (Berry, 2005, 2017). Integration is associated with multicultural societies and people able
to navigate the features of different cultures. Assimilation entails melting pot policies and
incorporation within the receiving society. Separation means segregation between ethnic groups,
while marginalization is associated with exclusion. Regardless of the original focus on the
orientations, Berry's categories are universally recognized as a typological model describing the
range of possibilities regarding how people situate themselves and participate in different cultural
groups (Nguyen & Benet-Martinez, 2013).
Figure 1. Acculturation orientations and societal profiles. Adapted from Berry (2005)
Many social science studies have explored what processes favor one strategy over another and how
desired integration outcomes can be reached (Ward & Kus, 2012). The dynamics of acculturation are
influenced by many factors. In this paper, we focus on the role of tolerance for both the receiving
community and migrants. Apparently simple in its meaning, the term tolerance, when applied to
integration in a diverse context, is of interest in domains ranging from social science research (Brewer
& Pierce, 2005) to policy design (Verkuyten, Yogeeswaran, & Adelman, 2018). In this context,
tolerance implies acceptance despite recognition and disapproval of diversity (van Doorn, 2014).
Nevertheless, as Verkuyten et al. (2018) have highlighted, in most social sciences fields, tolerance is
synonymous with openness to cultural others or a generalized "positive attitude" (p.10) towards them.
Both meaning highlight the role of tolerance as an antecedent of network formation in a multicultural
context due to migration flows. Roccas and Brewer (2002) have demonstrated that low levels of
tolerance towards outgroups (e.g., migrants and those of different race) are associated with
conservatism values, which motivate people to avoid uncertainty and ambiguity. On the other hand,
tolerance can cause group cohesion to fracture. On a personal level, tolerant people might prefer
tolerant people of other groups over conservative members of their own group (Verkuyten, 2010).
3
Conversely, conservative people might reject tolerant members of their same group for holding
different values (Verkuyten, 2010). Moreover, conservative people might openly condemn the
acceptance of cultural others by members of their own group out of a sense of ethnic loyalty (Padilla
& Perez, 2003).
Even if tolerance is accepted as a possible predictor of migrant integration, an important question is
how it actually modulates the mutual adaptation between locals and migrants. The contact hypothesis,
stating how positive contact between groups reduces prejudice, can fit this interest (Allport, 1954).
Although this assumption is not completely new within acculturation studies (Berry, 1984), it has
received more emphasis in recent years (Berry, 2017). Even though reciprocity in social interactions
is recognized as a fundamental element of network formations and their consolidation (Stark, 2015),
the dynamics of the process are complex and conclusions not obvious. Research has clearly
demonstrated that positive intergroup interaction reduces prejudice towards outgroups (Dovidio,
Eller, & Hewstone, 2011). Nevertheless, Riek, Mania and Gaertner (2006) have shown that negative
intergroup experiences, either real or imagined, increase negative attitudes towards outgroups. Due
to this complexity, many scholars have highlighted the need to more effectively address the
interactions occurring in the ecological contexts of acculturation and the ways in which they condition
integration processes (Ward & Geeraert, 2016; Berry, 2017).
We believe that knowledge of how tolerance influences the social dynamics of mutual adaptation
between locals and migrants in relation to network formation is crucial for understanding how
polarizing opinions on out-groups can influence future integration scenarios. Agent-based modelling
can be useful in this regard for several reasons. First, agent-based modelling allows researchers to
translate their own hypotheses on social processes into distributed actions of individuals in artificial
societies (Macy & Willer, 2002). For this reason, agent-based modelling is the only method able to
simulate and explore the emergence of complex social phenomena linking the micro-level of
individual behavior with the macro-level of observed social phenomena (Edmonds & Meyer, 2017;
Gilbert & Troitzsch, 2005). Second, model building allows us to compare different types of artificial
societies, so to embrace the complexity necessary to disentangle conditions and processes underlying
social phenomena such as the contact hypothesis. Lastly, what-if scenarios allow us to connect
different processes that can contribute to integration outcomes in a sequential order, a feat that
variable-based models are not capable of (Squazzoni, Jager & Edmonds, 2013). As in the case of
acculturation studies, the clarification of mechanisms requires examining the relationship among
migration flows, speed of intake, network formation and changes in tolerance as an effect of
reciprocity.
In sum, following Edmonds (2017), we propose using agent-based modelling to explain some
plausible social mechanisms underlying the emergence of acculturation within Berry's typological
model in a migration context. To that end, we developed the MigrAgent model, which this paper
presents. In our model, we focus on the role of tolerance and reciprocity in intergroup network
formation. In the first experiments described in this paper, we compare different societies in terms of
their level of conservatism and simulate migration flows using varying speed of intakes. Our aim is
to determine what type of acculturation emerges for different strata of society from these scenarios.
In this preliminary work, we qualitatively compare the assumptions of our model and observed
processes with data from the survey Causes and Consequences of Socio-Cultural Integration
Processes among New Immigrants in Europe (SCIP) collected in Germany.
4
Model Description
MigrAgent was built using NetLogo 6.04, and it is currently available at the CoMSES Computational
Model Library, see section Additional Information. In the model, the world is split between a home
country and a host country. Local agents (blue color tag) reside in the host country, and migrant
agents (green color tag) reside in the home country. Once the simulation starts, migrant agents move
from the home country to the host country according to speed of intake, which represent the random
probability of an agent to move to the host country. No agent, whether local or migrant, can leave the
host country. The three main dynamics of the simulation occur in the host country in a circular way.
The first dynamic is group aggregation based on agents' attachment preferences; together with the
speed of intake, this variable influences spatial sorting and the probability of interaction between
agents. The second dynamic is the acceptance or rejection of intergroup interactions based on the
tolerance of the receiver. The last dynamic is the change in tolerance as a function of reciprocity (see
Figure 2). Table 1 summarizes the parameters and the attributes of the agents.
Following Berry's description of acculturation as involving group-level norms (2005), we included a
group-level dimension of conservatism to simulate tolerance. The collective conservatism of each
ethnic group ranges from -1 and +1. The individual level of conservatism of each agent at the start of
the simulation runs (time step 0) is determined according to a normal distribution with mean equal to
the collective conservatism of the own ethnic group and standard deviation of 0.45. The individual
level of conservatism of each agent then changes as a dynamic variable according to their experience
of rejection or acceptance in intergroup interactions, as described below. Values of individual
conservatism below 0 denote liberal agents (bright color tag); values of conservatism equal to or
higher than 0 denote conservative agents (dark color tag). In using the terms conservative and liberal
agents, we do not refer to any political or philosophical interpretation, but to agents with a low level
of tolerance (conservatives) or a high level of tolerance (liberals). Thus, four types of agents interact
in our simulation: liberal locals (bright blue), conservative locals (dark blue), liberal migrants (bright
green) and conservative migrants (dark green). Ethnicity is a fixed state, and conservatism is a
dynamic state.
5
Parameter
Type
Range
Function
number_local
integer
[0,1000]
number of local agents
conservatism_local
continuous [-1,1]
mean in normal distribution of
local agents' conservatism.
standard deviation = 0.45
number_migrant
integer
[0,1000]
number of migrant agents
conservatism_migrant
continuous [-1,1]
speed_intake
integer
[1,100]
mean in normal distribution of
migrant agents' conservatism.
standard deviation = 0.45
speed of intake: percentage of
migrant agents moving
simultaneously to host country
Agents attributes
ethnicity
Type
fixed
Range
Tag and Action
[local, migrant] blue color: local
conservatism
dynamic
[-∞,+∞]
happy?
boolean
[true, false]
green color: migrant
liberal agents [< 0]:
bright ethnicity color
conservative agents [≥ 0]:
dark ethnicity color
liberal agents: fraction liberal
agents in-radius 1.5 ≥ 0.5
conservative agents: fraction
conservative same ethnicity in-
radius 1.5 ≥ 0.5
Table 1. Parameters of MigrAgent model
We designed our model based on Schelling's dynamics of segregation (1971), although we did not
literally copy the rules of his model. We took the core assumption that individual homophily
preferences can lead to high levels of segregation. Although Schelling's model is usually applied to
spatial segregation, we applied it to network aggregation. Our idea was that people's segregation
under certain homophily preferences limits individuals' opportunities to interact with members of
other groups, regardless of their willingness to engage in such interactions. In our adaptation, liberal
agents consider as similar other agents based on their attitude, not their ethnicity. They prefer for their
proximal neighborhood to be home to agents with low conservatism, independent of their ethnicity,
and for that neighborhood to exclude conservative agents, including those of their own group.
Conservative agents instead consider as similar agents sharing both their ethnicity and their high level
of conservatism. Thus, conservative agents reject both liberals of their own ethnic group and agents
of different ethnicities. They prefer being in neighborhoods hosting only conservative agents of their
6
own ethnicity. In Schelling's original model, agents are happy with their neighborhood when it
includes a certain fraction of desired agents ranging from 0 to 1. To control parameter sweeping, we
kept the desired fraction of agents fixed at 0.5. This means that agents are happy with their
neighborhood if those considered as similar are not in a minority condition; moreover, an agent could
be happy if alone. Neighborhood space is calculated as an in-radius of 1.5. With Schelling's
dynamics, agents do not move if a satisfactory proportion of desired agents live in their neighborhood.
We differentiated our model in this regard. In our simulation, agents move within a neighborhood if
they are satisfied with its composition; if dissatisfied, they relocate elsewhere. We made this change
to increase the likelihood of intergroup interaction and to simulate how people's sense of security
(Berry, 2017) influences their exploration of the world. Due to this decision, we do not envision the
model achieving a stable equilibrium, and the model does not feature a stop rule. The formula below
represents the behavior of agents:
where:
(cid:3004) =
𝑓(cid:3036)
(cid:3284)∈(cid:3300)
∑ ((cid:3004)⊂(cid:3006)(cid:3284))
(cid:3015)(cid:3284)∈(cid:3300)
(cid:3013) =
𝑓(cid:3036)
∑ ((cid:3013))
(cid:3284)∈(cid:3300)
(cid:3015)(cid:3284)∈(cid:3300)
(cid:3004) = fitness of conservative agent i
𝑓(cid:3036)
(cid:3013) = fitness of liberal agent i
𝑓(cid:3036)
𝑖 ∈ 𝑦 = neighborhood y of agent i
𝐶 ⊂ 𝐸(cid:3010) = conservative agents of the ethnic group E of agent i
𝐿 = liberal agents of both ethnic groups
𝑁(cid:3036)∈(cid:3052) = total number of agents in neighborhood y of agent i
If 𝑓(cid:3036) ≥ 0.5, agent i moves within the neighborhood
If 𝑓(cid:3036) < 0.5, agent i relocates elsewhere
Following Berry's definition of a psychological-individual level of adaptation in acculturation
processes (2005), the initial conservatism of agents varies as a function of reciprocity or rejection in
intergroup interactions. The segregation dynamics in the host country that have been described
influence the probability of interaction. At each step, agents propose an interaction with others in the
neighborhood (in-radius 1.5) regardless of their ethnicity or conservatism. In making this choice, we
assumed that any person is potentially connected with others in his or her spatial proximity. Ingroup
interactions occur by default; intergroup interactions are accepted by liberal agents and rejected by
conservative agents. In our view, this assumption distinguishes between the behavior expected from
liberal versus conservative agents in an intercultural context. The individual level of conservatism of
each agent increases following their experiences of intergroup rejection and decreases due to
7
intergroup acceptance. For experiences of both acceptance and rejection, the highest value is kept in
memory is as follows:
𝑐(cid:3036),(cid:3047) = 𝑐(cid:3036),(cid:3047)(cid:2879)(cid:2869) + 𝑚𝑎𝑥(cid:3036)
(cid:3045)(cid:3032)(cid:3037) − 𝑚𝑎𝑥(cid:3036)
(cid:3028)(cid:3030)(cid:3030)
(cid:3045)(cid:3032)(cid:3037) = maximal experience of rejection of agent i
(cid:3028)(cid:3030)(cid:3030)= maximal experience of acceptance of agent i
where:
𝑐(cid:3036),(cid:3047) = conservatism of agent i at time step t
𝑐(cid:3036),(cid:3047)(cid:2879)(cid:2869)= conservatism of agent i at time step t-1
𝑚𝑎𝑥(cid:3036)
𝑚𝑎𝑥(cid:3036)
Note that 𝑐(cid:3036),(cid:3047) at time step 0 follows the distribution of collective conservatism of ethnic group of agent
i
Once agents are no longer in the same neighborhood due to their movements, the local interaction
breaks, although the maximal experience of rejection and maximal experience of acceptance are kept
in the memory of the agent.
Based on the dynamics thus described, the acculturation categories in Berry's model are defined as
the types of direct links that an agent receives; these links simulate participation in each ethnic group
(see Figure 3):
● Integration: The agent receives interactions from the ingroup and out-group
● Assimilation: The agent receives interactions from the out-group, but not from the ingroup
● Separation: The agent receives interactions from the ingroup, but not from the out-group
● Marginalization: The agent receives interactions from neither of the groups
Figure 2. Prototype of MigrAgent model
8
Figure 3. Stylized acculturation outcomes simulated in MigrAgent. Acculturation refers to the agent
in dashed circle: A integration, B assimilation, C separation, D marginalization. Images are
elaborated for grey-scale printing.
Experiments and Results
We used MigrAgent to explore the polarization dynamics and associated acculturation outcomes
emerging from the interaction between a migrant and a local population in relation to network
formation combining different levels of conservatism in both populations. We were additionally
interested in how these outcomes varied depending on the speed of intake. To this end, we performed
parameter sweeping, varying the parameters of the collective conservatism of the local population,
the conservatism of the migrant population and the speed of intake. For the level of collective
conservatism, we selected five conditions in both the migrant and the local population. The first
condition was that of an equal distribution between liberals and conservatives (conservatism = 0).
The other conditions were liberal societies (conservatism = -0.25), extremely liberal societies
(conservatism = -0.75), conservative societies (conservatism = 0.25) and extremely conservative
societies (conservatism = 0.75). For speed of intake, we selected condition 1 (minimum speed = 1%
of the migrant population moves to the host country at each step) and 100 (maximal speed = 100%
of the migrant population moves to the host country at each step). The number of agents was kept
constant in all conditions: 500 local agents and 500 migrant agents. We computed a total of 50
combinations and counted 1,000 time steps for each simulation. Each simulation was repeated 20
times. Table 2 summarizes the parameter sweeps and experimental conditions. To avoid incremental
effects of feedback in intergroup interactions, we put a limit to the change of conservatism of agent,
equal to +1 for the upper level and -1 for the lower level. Agents change their individual level of
conservatism as long it falls into the range. Outlier agents who fall out of the limit in the initial
distribution at time step 0 do not change their individual level of conservatism, thus serving as noise
to test the robustness of the model across the conditions. Results refer to the averaged scores out of
the 20 repetitions.
9
Parameters
Values
conservatism_local
-0.75
Extremely
liberal
conservatism_migrant -0.75
speed_intake
number_local
number_migrant
Time steps
Repetition runs
-0.25
liberal
-0.25
liberal
0
equally
distributed
0
equally
distributed
0.25
conservative
0.25
conservative
0.75
Extremely
conservative
0.75
extremely
conservative
100
fast
extremely
liberal
1
slow
500
500
1000
20
Table 2. Parameter sweeps and experimental conditions
Describing our results, we first show a polarization effect at the macro-level of our observations.
Then, we illustrate the acculturation processes along the simulation for substrata of society (liberal
migrants, conservative migrants, liberal locals, conservative locals). Figure 4 serves as a baseline
indicating the initial values of the simulation at time step 0 for each condition. Values reported in
Figure 4 refer both to the local and migrant populations. Conservatism informs the mean level of
conservatism in each population (i.e., how conservative or liberal the population is). The other two
components of Figure 4 (rectangles) inform the fraction of liberals and migrants (i.e., how many
agents of the category are in the population).
10
Figure 4. Baseline of conservatism mean, fraction of conservative agents and liberal agents at the
start of the simulations (time step 0)
Figure 5 depicts the level of conservatism in each condition at the end of the simulation runs (final
time step 1,000). The figures compare the migrant population (on the top) and local population (on
the bottom), and for each observation, the slow intake speed and the fast intake speed are compared.
Slow intake implies that approximately 250 migrants have entered the receiving society at time step
1,000, whereas fast intake means all 500 migrants have entered. Thus, the different intake speeds are
a proxy for relative group sizes, with the migrant group in a minority condition in the slow intake
condition. On the x-axis, the range of collective local conservatism at time step 0 is reported; on the
y-axis, the range of collective migrant conservatism at time step 0 is reported. The cells in the heatmap
report the average level of conservatism for that population. At first glance, for both of the
populations, it is clear how polarization effects emerge with the tendency towards greater
conservatism for conditions on the upper right-hand side of the diagonal and towards lower
conservatism for conditions on the lower left-hand side of the diagonal. The heatmap provides an
indication of how resilient one group is to the rejection of the other group. When more tolerant
societies meet, they both shift to lower levels of conservatism (higher tolerance). Likewise,
interactions between more conservative societies produce a shift towards greater conservatism for
both. The local population is more sensitive to the degree of conservatism of the migrants than vice
versa. The speed of intake of migrants has a stronger effect on the conservatism of the local
population than on the migrants' degree of conservatism. For the migrant population, fast intake
seems associated with a larger polarization effect. Figure 6 illustrates the acculturation outcomes that
emerge at the global level for each population at time step 1,000. Values very close to 0 do not appear
in the figures. The results highlight that integration and separation are the dominant scenarios.
Conditions with convergence towards lower conservatism show integration as the dominant outcome,
whereas for conditions with convergence towards greater conservatism, separation is the dominant
outcome.
11
Figure 5. Conservatism mean at time step 1,000 for migrant and local population within slow and
fast intake
Figure 6. Acculturation outcomes at time step 1,000
Figure 7 reports the change in the fraction of liberals and conservatives in the migrant and local
populations over the 1,000 time steps. Grey color refers to local population, black color refers to
migrant population, while solid lines represent the fraction of liberal agents in each population and
dotted lines conservative one.
12
Figure 7. Emerging fraction liberal and conservative agents in local and migrant population over
1,000 time steps
In Figures 8,9,10 and 11, we explore the acculturation categories for each substratum of society over
1,000 time steps that can be compared with the change in the fraction of liberal or conservative agents
as in Figure 7. The same graph is repeated for liberal migrants, conservative migrants, liberal locals
and conservative locals. Each line represents the fraction of agents of the substratum that engages in
that acculturation outcome.
As regards liberal migrants (Figure 8), assimilation is more evident in the early time steps of the
simulation, but eventually it is overcome by integration. This outcome reflects how liberal migrants
benefit from the availability of liberal locals they have more chance to interact with when in the
condition of minority. The trend is more evident in the row for extremely conservative migrant
population (conservatism = 0.75) in the slow intake condition. Comparing with Figure 7, we observe
how integration becomes eventually the dominant outcome as long as at least a small part of liberals
survives in both groups. When liberal agents turn conservative due to rejection by conservatives,
surviving liberal migrants can still engage in assimilation for conditions of low conservatism in the
local population (conservatism below 0). When the availability of liberal locals is not possible in the
more conservative local population, separation emerges since liberal migrants can only find similar
in their own ethnic group. In the fast intake condition, the change towards conservatism as effect of
rejection is more drastic, with the extinction of liberal migrants when collective conservatism of either
of the populations is above 0.
13
Conservative migrants (Figure 9) exhibit similar patterns but with opposite acculturation outcomes.
In the early time steps of simulation runs, when conservative migrants are in the minority,
marginalization arises for all combinations. The trend is more evident for slow intake conditions. This
outcome happens because the conservative migrants, who are in the minority, refuse to interact with
both the local population and liberal migrants. When more conservative agents enter the receiving
society, separation is more prominent. In the fast intake condition, a trend toward separation emerges
continuously and steadily already from the first time steps of the simulation. In the condition of
extremely liberal societies (i.e., conservatism for both populations equal to -0.75), the conservative
migrants disappear; they become liberals as effect of intergroup acceptance.
For liberal locals (Figure 10), an increase in separation appears for the first time steps, then overcome
by integration. This result might seem counterintuitive, especially when extremely liberal migrant or
extremely liberal local populations are involved. Similar to the trend of assimilation of liberal
migrants, the reason is the unbalanced group sizes, which results in more opportunities for locals to
interact with liberal members of the own group rather than migrants. As migrants enter the host
society, integration steadily increases. When intake is fast, the convergence towards integration is
more rapid, although at a lower rate than liberal migrants Interestingly, Figure 7 shows that the
fractions of liberal migrants and liberal locals are almost identical. This similarity seems not reflected
in the acculturation outcomes of the two populations comparing figure 8 and 10, and more evidently
for the interaction of extremely liberal local population and extremely conservative migrant
population. In this condition, liberal locals show integration at a much lower level than liberal
migrants, both for slow intake and fast intake. The reason is likely to be again the relative group sizes.
Since ingroup interactions occur by default, liberal locals have a higher probability to interact with
other liberals of the own ethnic group than with liberal migrants. Additionally, although liberal locals
would not aggregate with conservative locals, they have a high probability to relocate close to them
due to the simple effect of larger group size. Therefore, diverse conditions favor segregation of liberal
locals as simple effect of higher availability of co-ethnics.
Finally, the results for conservative locals (Figures 11) are not unexpected. They show more
separation, and the trend follows the line of their distribution within the local population as shown in
Figure 7. We can conclude that this outcome is a simple effect of unbalanced groups: the local
population represents the majority, making it relatively easy for conservative locals to connect with
others similar to themselves. As such, the lowest degree of separation for conservative locals is
reached for conditions of fast intake and extremely liberal migrant societies, when they turn liberal
because of the effect of intergroup acceptance
14
Figure 8. Acculturation outcomes for liberal migrants
15
Figure 9. Acculturation outcomes for conservative migrants
16
Figure 10. Acculturation outcomes for liberal locals
17
Figure 11. Acculturation outcomes for conservative locals
Our findings have implications regarding acculturation processes. First, the simulations demonstrate
that different acculturation processes can coexist for different strata of society, and even within the
same stratum. Slow intake influences acculturation processes through effects on relative group sizes.
Acculturation strategies can emerge as a transitory stage, as seen with the liberal migrants' shift from
assimilation to integration and the conservative migrants' shift from marginalization to separation.
Counterintuitive scenarios, such as the separation of liberal locals in conditions where conservative
agents represent the majority, are also possible. Furthermore, the simulations identified the effects of
reciprocity due to acceptance and rejection between liberal and conservatives of the two populations.
As for liberal migrants, reciprocity causes convergence to low levels of integration and assimilation
in conditions of great conservatism among locals and migrants. Conversely, liberal locals who
initially engage in integration shift to separation once they become conservative.
To check the effect sizes of the parameters involved in our simulations, we ran multiple linear
regressions for each substratum of the population. Analyses were run in R. To ensure the readability
of the text, a summary table of the multiple linear regressions is in the supplemental material. The
dependent variables were the four acculturation outcomes. As for the predictors, we selected the
emerged percentage of conservative locals, the percentage of conservative migrants, both over the
1,000 time steps, and the speed of intake as a factor variable. The regression models allowed us to
compare the direction of causality of the predictors on the observed behavior and to calculate Cohen's
18
(cid:3046)(cid:3045)2
1(cid:2879)(cid:3019)2 , where 𝑠𝑟2 is the squared semi-partial
𝑓2 for the effect sizes. The index is computed as 𝑓2 =
correlation and 𝑅2the multiple R2 of the regression model. As a rule of thumb, we maintain that an
𝑓2 ≥ 0.02 denotes a small effect size, 𝑓2 ≥ 0.15 a medium effect size, and 𝑓2 ≥ 0.35 a large effect
size (Kabacoff, 2011). The results are noteworthy for the consistent acculturation outcomes in each
subgroup, as observed in Figures 8,9,10 and 11. As regards the integration of liberal migrants, the
regression models confirm a negative effect of the percentage of conservative locals and conservative
migrants, which are the groups that tend to isolate liberal migrants. The speed of intake is positively
related to the integration of liberal migrants and negatively related to their assimilation. The Cohen's
𝑓2 for intake speed is 𝑓2 = 0.28 for integration of liberal migrants and 𝑓2 = 0.32 for assimilation of
liberal migrants, close to a large effect size. As for conservative migrants, the percentage of
conservative locals is negatively related to separation and positively related to marginalization. In
contrast, the percentage of conservative migrants is positively linked to separation and negatively
linked to marginalization.
The effect of conservative locals on separation of conservative migrants seems counterintuitive, and
it might depend on the role of the other types of agents and reciprocity effects. On the contrary, the
other results fit the theoretical assumptions of the model. Increased number of conservative locals
favors their chance to cluster, with the exclusion of conservative migrants and the increase in their
marginalization. At the same time, the large number of conservative migrants increases the chance of
interaction with other similar migrants, with the result to increase separation and buffer against
marginalization. As for speed of intake, larger group size is positively associated with the separation
of conservative migrants and negatively associated with their marginalization, because of the
influence on the probability of interaction. For both cases, Cohen's 𝑓2 indicates a large effect size
(𝑓(cid:2870) = 0.40). As for liberal locals, an interesting result is that an increased percentage of conservative
locals is negatively related to integration at a high rate, due to the rejection of liberals by conservatives
because of the different attitude. The speed of intake exhibits large effect sizes for integration
(𝑓(cid:2870) = 1.72) and marginalization (𝑓(cid:2870) = 0.43), although R² for integration outcome is suspiciously
high and more likely due to a simple effect. Finally, for conservative locals, the observations can be
interpreted as a simple effect of their distribution, and indeed, the results of regression models are
spurious.
Comparison with Empirical Data
We qualitatively compared the assumptions and observations of our model with data from the second
wave of the Causes and Consequences of Socio-Cultural Integration Processes among New
Immigrants in Europe (SCIP) survey in Germany (Diel et al., 2015). The survey is a panel study
interested in the antecedents of the adaptation of first-generation migrants in four European countries:
Germany, the Netherlands, Ireland, and the UK. We focused on the second wave (2012/2013) after
one year of residence in the country. Although one year of adaptation might not be a long enough
time period when considering acculturation, the items of the survey seemed to be appropriate proxies
for the parameters of our model. We selected the German sample because of consistency in the
number of participants as compared to other countries' samples; moreover, the data collected best fit
our model parameters. After approximating normality assumptions to run multiple linear regression
19
models, we were left with a final sample of 1,224 migrants of Polish and Turkish origins. Analysis
can be reproduced via the R-code provided.
Table 3 illustrates the items we selected as proxies for the parameters in our model and their ranges.
As for the typology of networks as the dependent variable, we selected the frequency of interactions
with locals and co-ethnics. As for the predictors in our model, we selected the experience of rejection
as a proxy for rejection in the host country, the experience of hospitality as a proxy for acceptance
and the importance of the country of birth as a proxy for the conservatism of individuals.
Proxy for
Item
Range Scale
Conservatism
IPIDCB: How important is the following to
your sense of who you are: the country where
you were born?
1: very important
4: not important at
all
Rejection
Acceptance
DSCRFREQ: […] How often do you think CO
people are discriminated against in RC?
1: very often
5: never
HOSP_RC: In general, RC is a
hospitable/welcoming country for [CO
people/pl]?
1: strongly agree
5: strongly
disagree
Interactions
with locals
PPRC: How often do you spend time with [RC
people/pl]?
1: every day
6: never
Interactions
with co-ethnics
PPCO: How often do you spend time with [CO
people/pl]?
1: every day
6: never
RC: receiving country; CO: country of origin
Table 3. Items from SCIP survey used as proxy in our analyses
We ran multiple linear regressions to detect the direction of causality between the predictors and types
of networks and then checked if the results fit the observed processes of our model. For the
interpretation of the results, one should refer to the scale range of the items in Table 3. As for
interactions with locals (Table 4), the negative coefficient b for conservatism implies that the more
important participants considered their country of birth (1 = very important) in defining themselves
(proxy for conservatism), the fewer interactions they had with locals of other ethnicity (6 = never).
Conversely, the less important people considered their country of birth in defining themselves (4 =
not important), the more likely they were to interact with locals (1 = every day). As for rejection, the
same interpretation of the negative coefficient b applies. People who reported high levels of rejection
(1 = very often) were less likely to interact with locals (6 = never), whereas people who reported
lower levels of rejection (5 = never) were more likely to interact with locals (1 = every day). The
same held true for the positive coefficient b for acceptance: Migrants who felt accepted in the host
20
country (1 = strongly agree) were more likely to connect with locals (1 = every day), whereas people
who did not feel accepted (5 = strongly disagree) were less likely to connect with them (6 = never).
As for interactions with co-ethnics (Table 5), the positive direction of coefficient b for conservatism
fits our model, meaning that migrants considering their country of birth as very important to their
self-definition (1 = very important) were more likely to interact with co-ethnics (1 = every day), and
those who did not value their country of birth as much (4 = not important at all) were less likely to
interact with co-ethnics (6 = never). We can expect rejection by the local population to be positively
related to interactions with co-ethnics, as happens in our model with segregation patterns due to
effects of intergroup rejection. Although coefficient b is positive and effect size strong (𝑓(cid:2870) = 0.57),
its magnitude is close to 0 and p-value not significant, so that our hypothesis cannot be confirmed.
As regards the effect of acceptance by locals on interactions with co-ethnics, we cannot formulate
firm hypothesis considering ingroup interactions happen by default in our model, and unless attitude
of migrants and spatial sorting are taken into consideration. As a matter of fact, coefficient b for
experience of acceptance by locals denotes a null correlation, and a not significant p-value.
In sum, the regression analyses confirmed the plausibility of our model's results, in particular for
interactions with locals with significant results. Yet, the magnitude of coefficient b, as compared to
the simple regression and partial R², was not very large. Namely for interactions with co-ethnics, the
results did not strongly support our model. Additionally, the effect sizes (Cohen's 𝑓2) were strong for
the effects of rejection and acceptance on interactions with locals and co-ethnics. Nevertheless,
predicted interactions with co-ethnics were not associated with significant p-values for all predictors,
except for the proxy for conservatism.
21
Interactions with locals
Predictor
b
partial R² r
sr²
Cohen f² p-value
conservatism
-0.275 0.020
-0.15 0.08 0.09
3.01e-06***
experience rejection
-0.169 0.012
-0.14 0.63 0.66
0.0017**
experience
acceptance
(Intercept)
R²
Adj R²
0.126
0.008
0.12 1.27 1.33
0.0204*
2.812
0.046
0.043
< 2e-16***
3.089e-10***
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Table 4. Summary multiple linear regression model with SCIP data for interactions with locals
Interactions with co-ethnics
Predictor
b
partial R² r
sr²
Cohen f² p-value
conservatism
0.179
0.013
0.11 0.00 0.01
0.000368***
experience rejection
0.051
0.001
0.04 0.56 0.57
0.268089
experience
acceptance
(Intercept)
R²
Adj R²
0.018
-0.002
0.00 0.45 0.46
0.701909
1.640
0.014
0.011
1.83e-12***
0.002499**
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Table 5. Summary multiple linear regression model with SCIP data for interactions with co-ethnics
22
Limitations and Conclusions
In this preliminary work, we have presented our MigrAgent model and demonstrated how different
acculturation processes can emerge and coexist for different groups in society in a stylized scenario.
We have also explored these processes' association with polarization dynamics. We think our work
shows that agent-based modelling can be a valuable contribution to the study of acculturation, but we
recognize much must be done to increase the validity of our model. A main challenge in the work
presented herein was the comparison with real data. We recognize that although the selected empirical
data reasonably support our model, they are not very strong for multiple reasons. First of all, these
were secondary data we adopted as proxies for our parameters, and so the meanings of the items
might not fit those of the model. This could explain the lack of consistency for interactions with
locals. Even if we could access data fitting the meanings of the parameters of our model, we would
need historical data to validate the processes in MigrAgent. Additionally, the sample of SCIP do not
include locals, who are a class of actors included in our model. A survey specifically designed to feed
our simulation might be envisioned in the future, testing the hypothesis generated by the processes
observed in MigrAgent and addressing specifically the interaction between locals and migrants over
time. On the theoretical level, one issue regards the definition of acculturation. Migrant integration is
a complex phenomenon that can be studied from a psychological, structural or intergenerational
perspective (Rumbaut, 2015). At this stage, our model allows us to differentiate between an individual
and a collective level, which is the main purpose of agent-based modelling (Squazzoni, Jager, &
Edmonds, 2013). Although this feature can be an asset in the study of acculturation processes, it is
still necessary to consider what specific areas of the literature might benefit from the model and how
its contribution can be best oriented. Here, we have focused on networks in terms of the participation
of individuals from different cultures (Berry, 2005). This approach fits our research question, but
many other dimensions should be included, depending on the definition of acculturation. A critical
one is the dimension of power. The asymmetrical power relations between locals and migrants have
been widely recognized (Verkuyten, Yogeeswaran, & Adelman, 2018), which justifies the focus on
exploring the dynamics of opinion polarization in receiving countries. However, adding a dimension
of power would imply the inclusion of additional parameters and different mechanisms of interaction
between the two populations. For the sake of the model's simplicity at this stage, we avoided such
considerations in this study. Finally, the results of our simulations illustrate strong polarization
between integration and separation, which might have obscured more nuanced marginalization and
assimilation conditions. We theorize these results were due to the choice to keep the desired fraction
of similar others at 0.50, which means that agents would relocate as soon as those considered as
similar are in the minority condition. This scenario increases chances of spatial segregation of agents
and consequently influences formation of agents' network in their proximity. Nevertheless, our
parameter sweeping already included too many conditions to compare, and we opted for this choice.
Sweeping the fraction of desired others would be a feasible area of investigation in the future, so to
better understand how spatial sorting influences the network formation. Further studying could also
include thorough tests on the robustness of the model and boundary effects on the distribution of
conservatism in the groups.
Even recognizing certain limits due to the preliminary status of our work, we think MigrAgent can
provide a useful basis for the study of acculturation dynamics and the conditions under which mutual
acceptance can emerge or fail, resulting in different scenarios for societal network structures. The
23
main contribution of the model is that it compares different types of societies and strata for multiple
groups, in this case along the dimension of conservatism/liberalism. In the future, societies or their
strata might differ along other dimensions, such as the internal distribution of wealth within groups
(Bergreen & Nilsson, 2003) or the similarity of cultural models and practices (Hofstede, 2001). An
asset of MigrAgent is its ability to translate these social dynamics into scenarios linking migration
flows and post-migration adaptation. We aim in the future to parameterize MigrAgent using available
data to further improve its empirical grounding and conduct studies on specific cases of different
migration scenarios.
Author Information
Rocco Paolillo, M.A., is a EU COFUND BIGSSS-departs PhD fellow at the Bremen International
Graduate School of Social Sciences, University of Bremen & Jacobs University Bremen (Germany).
His work for the project here presented is related to a research stay at the University College
Groningen (the Netherlands) funded by the Foundation "Franco e Marilisa Caligara" in Torino (Italy)
in 2015 and with the support of University College Groningen.
email: [email protected]
Wander Jager, Ph.D., is an associate professor of Social Complexity at University College Groningen
and managing director of the Groningen Center for Social Complexity Studies, both at the University
of Groningen (the Netherlands).
email: [email protected]
Additional Information
NetLogo model and R-code for data analysis can be found at:
Paolillo, Rocco, Jager, Wander (2018, November 28). "MigrAgent" (Version 1.2.0). CoMSES
Computational Model Library. Retrieved from: https://www.comses.net/codebases/a6fc3cc6-bd5b-
4cd6-8300-75978ea1e362/releases/1.2.0/.
Empirical data from SCIP project is available at GESIS Data Archive (study nr. ZA5956):
Diehl, C., Gijsberts, M., Güveli, A., Koenig, M., Kristen, C., Lubbers, M., McGinnity, F., Mühlau,
P., Platt, L., & Van Tubergen, F. (2016). Causes and consequences of socio-cultural integration
processes among new immigrants in Europe (SCIP). GESIS Data Archive, Cologne. ZA5956 Data
file
from
https://dbk.gesis.org/dbksearch/sdesc2.asp?no=5956&db=e&doi=10.4232/1.12341
doi:10.4232/1.12341.
Retrieved
Version
1.0.0,
24
References
Allport, G. W. (1954). The nature of prejudice. Reading, MA: Addison-Wesley.
Berggren, N., & Nilsson, T. (2003). Does economic freedom foster tolerance? Kyklos, 66(2), 177 --
207.
Berry, J. W. (1984). Multicultural policy in Canada: A social psychological analysis. Canadian
Journal of Behavioral Science, 16(4), 353 -- 370.
Berry, J. W. (2005). Acculturation: Living successfully in two cultures. International Journal of
Intercultural Relations, 29, 697 -- 712.
Berry, J. W. (2017). Theories and models of acculturation. In S. J. Schwartz & J. Unger (Eds.).
The Oxford handbook of acculturation and health (pp. 1-24). Oxford, UK: Oxford
University Press.
Brewer, M. B., & Pierce, K. P. (2005). Social identity complexity and outgroup tolerance.
Personality and Social Psychology Bulletin, 31(3), 428 -- 437.
Diehl, C., Gijsberts, M., Güveli, A., Koenig, M., Kristen, C., Lubbers, M., McGinnity, F., Mühlau,
P., Platt, L., & Van Tubergen, F. (2016). Causes and consequences of socio-cultural
integration processes among new immigrants in Europe (SCIP). GESIS Data Archive,
Cologne. ZA5956 Data file Version 1.0.0, doi:10.4232/1.12341. Retrieved from
https://dbk.gesis.org/dbksearch/sdesc2.asp?no=5956&db=e&doi=10.4232/1.12341
Dovidio, J.F., Eller, A., & Hewstone, M. (2011). Improving intergroup relations through direct,
extended and other forms of indirect contact. Group Processes & Intergroup Relations,
14(2), 147 -- 160.
Edmonds, B., & Meyer, R. (2017). Simulating social complexity (2nd ed.). Heidelberg, Germany:
Springer-Verlag.
Edmonds, B. (2017). Different Modelling Purposes. In B. Edmonds, & R., Meyer (Eds.), Simulating
social complexity (pp. 39-58). Heidelberg, Germany: Springer-Verlag.
European Commission. (2015). Communication from the commission to the European parliament,
the Council, the European Economic and Social Committee and the Committee of the
Regions. A European agenda on migration (No. COM(2015) 240 final). Brussels, Belgium.
Eurostat. (2017). First instance decisions on applications by citizenship, age and sex. Annual
from:
aggregated
http://appsso.eurostat.ec.europa.eu/nui/submitViewTableAction.do
Retrieved
data
(rounded).
Gilbert, N., & Troitzsch, K. G. (2005). Simulation for the social scientist (2nd ed.). Maidenhead,
UK: McGraw-Hill Education.
Hofstede, G.H. (2001). Culture's consequences: Comparing values, behaviors, institutions, and
organizations across nations. London, UK: SAGE.
International Organization for Migration. (2018, November 20). Key migration terms.
25
https://www.iom.int/key-migration-terms#Migration.
Kabacoff, R. I. (2011). R in action. Data analysis and graphics with R., Shelter Island, NY:
Manning Publications Co.
Kelley, C.P., Mohtadi, C.P., Cane, M.A., Seager, R., Kushnir, Y. (2015). Climate change in the
fertile crescent and implications of the recent Syrian drought. PNAS 112(11), 3241−3246.
Macy, M. W., & Willer, R. (2002). From factors to factors: computational sociology and
agent-based modeling. Annual Review of Sociology, 28(1), 143-166.
Nguyen, A. M. D., & Benet-Martínez, V. (2013). Biculturalism and adjustment: A meta-analysis.
Journal of Cross-Cultural Psychology, 44(1), 122 -- 159.
Organization for Economic Co-operation and Development. (2015). Is this humanitarian migration
crisis different? Migration Policy Debates, 7.
Padilla, A. M., & Perez, W. (2003). Acculturation, social identity, and social cognition: A new
perspective. Hispanic Journal of Behavioral Sciences, 25(1), 35-55.
Redfield, R., Linton, R., & Herskovits, M. J. (1936). Memorandum for the study of acculturation.
American Anthropologist, 38(1), 149 -- 152
Riek, B. M., Mania, E. W., & Gaertner, S. L. (2006). Intergroup threat and outgroup attitudes: A
meta-analytic review. Personality and Social Psychology Review, 10(4), 336 -- 353.
Roccas, S., & Brewer, M. B. (2002). Social identity complexity. Personality and Social Psychology
Review, 6(2), 88 -- 106.
Rumbaut, R. G. (2015). Assimilation of immigrants. International Encyclopedia of the Social and
Behavioral Sciences, 2(2), 81-87.
Schelling, T. C. (1971). Dynamic models of segregation. Journal of Mathematical Sociology,
1(2),143 -- 186.
Squazzoni, F., Jager, W., & Edmonds, B. (2013). Social simulation in the social sciences: A brief
overview. Social Science Computer Review, 20(10), 1 -- 16.
Stark, T. H. (2015). Understanding the selection bias: Social network processes and the effect of
prejudice on the avoidance of outgroup friends. Social Psychology Quarterly, 78(2),
127 -- 150.
United Nations High Commissioner for Refugees. (2017). Operational portal. Refugee situations.
Retrieved from: http://data2.unhcr.org/en/situations/mediterranean.
Van Doorn, M. (2014). The nature of tolerance and the social circumstances in which it emerges.
Current Sociology, 62(6), 905-927.
Verkuyten, M. (2010). Multiculturalism and tolerance. An Intergroup Perspective. In R. Crisp (Ed.),
The psychology of social and cultural diversity (pp. 147 -- 170). Chichester, UK: Blackwell
Publishing Limited.
26
Verkuyten, M., Yogeeswaran, K., & Adelman, L. (2018). Intergroup toleration and its implications
for culturally diverse societies. Social Issues and Policy Review, 00(0), 1-30.
Ward, C., & Geeraert, N. (2016). Advancing acculturation theory and research: The acculturation
process in its ecological context. Current Opinion in Psychology, 8, 98 -- 104.
Ward, C., & Kus, L. (2012). Back to and beyond Berry's basics: The conceptualization,
operationalization and classification of acculturation. International Journal of Intercultural
Relations, 36(4), 472-485.
Wike, R., Stokes, B., & Simmons, K. (2016). Europeans fear wave of refugees will mean more
terrorism, fewer jobs sharp ideological divides across EU on views about minorities,
diversity
from
http://www.pewglobal.org/2016/07/11/europeans-fear-wave-of-refugees-will-mean-more-
terrorism-fewer-jobs/
Retrieved
and
national
identity.
27
Supplemental Material
Multiple Linear Regression Models Simulations
R²: Multiple R²; Adj R²: Adjusted R²; b: regression coefficient; r: correlation coefficient; sr²:
squared semi-partial correlation; Cohen f²: effect size
Integration Liberal Migrants
Predictor
b
partial R² r
sr² Cohen f²
p-value
% conservative locals
-0.363 0.016
-0.67 0.00 0.00
<2e-16 ***
% conservative migrants -0.151 0.003
-0.66 0.00 0.00
<2e-16 ***
Speed intake
(Intercept)
R²
Adj R²
0.197 0.130
0.42 0.12 0.28
<2e-16 ***
0.809
0.570
0.570
<2e-16 ***
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Assimilation Liberal Migrants
Predictor
b
partial R² r
sr² Cohen f² p-value
% conservative locals
0.197 0.005
0.47 0.00 0.00
< 2e-16 ***
% conservative migrants 0.065 0.001
0.45 0.00 0.00
3.72e-07 ***
-0.252 0.212
-0.51 0.20 0.32
< 2e-16 ***
Speed intake
(Intercept)
R²
0.254
0.388
0.388
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
< 2e-16 ***
28
Multiple Linear Regression Models Simulations
R²: Multiple R²; Adj R²: Adjusted R²; b: regression coefficient; r: correlation coefficient; sr²:
squared semi-partial correlation; Cohen f²: effect size
Separation Liberal Migrants
Predictor
b
partial R²
r
sr² Cohen f² p-value
% conservative locals
0.158
0.011
0.58 0.00 0.00
<2e-16 ***
% conservative migrants 0.088
0.003
0.58 0.00 0.00
<2e-16 ***
Speed intake
(Intercept)
R²
0.061
-0.066
0.374
0.048
0.04 0.04 0.07
<2e-16 ***
<2e-16 ***
0.374
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Marginalization Liberal Migrants
Predictor
b
partial R²
r
sr² Cohen f² p-value
% conservative locals
0.008
0.001
0.13
0.00 0.00
1.28e-09 ***
% conservative migrants
-0.002 0.000
0.12 0.00 0.00
0.0883 .
Speed intake
-0.005 0.012
-0.14 0.01 0.01
< 2e-16 ***
(Intercept)
R²
0.004
0.029
0.029
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
< 2e-16 ***
29
Multiple Linear Regression Models Simulations
R²: Multiple R²; Adj R²: Adjusted R²; b: regression coefficient; r: correlation coefficient; sr²:
squared semi-partial correlation; Cohen f²: effect size
Separation Conservative Migrants
Predictor
b
partial R²
r
sr²
Cohen f² p-value
% conservative locals
-0.117 0.002
0.01 0.00 0.00
<2e-16 ***
% conservative migrants 0.189
0.005
0.03 0.00 0.00
<2e-16 ***
Speed intake
(Intercept)
R²
0.312
0.634
0.303
0.294
0.54 0.28 0.40
<2e-16 ***
<2e-16 ***
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
0.303
Marginalization Conservative Migrants
Predictor
b
partial R²
r
sr²
Cohen f² p-value
% conservative locals
0.117
0.002
-0.01 0.00 0.00
<2e-16
% conservative migrants
-0.189 0.005
-0.03 0.00 0.00
<2e-16
-0.312 0.294
-0.54 0.28 0.40
<2e-16
Speed intake
(Intercept)
R²
0.366
0.303
0.303
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
<2e-16
30
Multiple Linear Regression Models Simulations
R²: Multiple R²; Adj R²: Adjusted R²; b: regression coefficient; r: correlation coefficient; sr²:
squared semi-partial correlation; Cohen f²: effect size
Integration Liberal Locals
Predictor
b
partial R²
r
sr²
Cohen f² p-value
% conservative locals
-0.803
0.19
-0.92 0.01 0.06
< 2e-16 ***
% conservative migrants
-0.059
0.11
-0.90 0.00 0.00
3.03e-15 ***
Speed intake
(Intercept)
R²
0.153
0.774
0.881
0.22
0.881
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
0.35 0.21 1.72
< 2e-16 ***
< 2e-16 ***
Assimilation Liberal Locals
Predictor
b
partial R²
r
sr²
Cohen f² p-value
% conservative locals
-0.001 0.00
-0.29 0.00 0.00
1.14e-12 ***
% conservative migrants
0.000
0.00
-0.28 0.00 0.00
2.28e-06 ***
0.25 0.04 0.05
< 2e-16 ***
< 2e-16 ***
31
Speed intake
(Intercept)
R²
0.001
0.002
0.124
0.05
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
0.124
Multiple Linear Regression Models Simulations
R²: Multiple R²; Adj R²: Adjusted R²; b: regression coefficient; r: correlation coefficient; sr²:
squared semi-partial correlation; Cohen f²: effect size
Separation Liberal Locals
Predictor
b
partial R²
r
sr²
Cohen f² p-value
% conservative locals
-0.197 0.01
-0.27 0.00 0.00
< 2e-16 ***
% conservative migrants 0.060 0.00
-0.27 0.00 0.00
8.74e-16 ***
Speed intake
(Intercept)
R²
-0.153 0.21
-0.39 0.20 0.28
< 2e-16 ***
0.223
0.271
< 2e-16 ***
0.271
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Marginalization Liberal Locals
Predictor
b
partial R²
r
sr²
Cohen f² p-value
% conservative locals
0.000 0.00
-0.16 0.00 0.00
<2e-16 ***
% conservative migrants -0.001 0.33
-0.18 0.00 0.00
<2e-16 ***
-0.001 0.30
-0.51 0.28 0.43
<2e-16 ***
Speed intake
(Intercept)
R²
0.001
0.331
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
0.331
<2e-16 ***
32
Multiple Linear Regression Models Simulations
R²: Multiple R²; Adj R²: Adjusted R²; b: regression coefficient; r: correlation coefficient; sr²:
squared semi-partial correlation; Cohen f²: effect size
Separation Conservative Locals
Predictor
b
partial R²
r
sr²
Cohen f²
p-value
% conservative locals
0.995
1.000
1.00 0.04 NA
<2e-16 ***
% conservative migrants -0.002
0.994
0.98 0.00 NA
<2e-16 ***
Speed intake
(Intercept)
R²
-0.002
0.000
1.000
0.257
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
1.000
-0.18 0.24 NA
<2e-16 ***
<2e-16 ***
Marginalization Conservative Locals
Predictor
b
partial R²
r
sr²
Cohen f²
p-value
% conservative locals
0.005
0.079
0.87 0.00 0.02
<2e-16 ***
% conservative migrants 0.002
0.817
0.87 0.00 0.00
<2e-16 ***
0.09 0.24 1.38
<2e-16 ***
Speed intake
(Intercept)
R²
0.002
0.000
0.822
0.257
Adj R²
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
0.822
<2e-16 ***
33
|
1802.06108 | 3 | 1802 | 2019-12-31T20:08:44 | Modeling the Formation of Social Conventions from Embodied Real-Time Interactions | [
"cs.MA",
"cs.AI",
"cs.GT",
"q-bio.NC",
"stat.ML"
] | What is the role of real-time control and learning in the formation of social conventions? To answer this question, we propose a computational model that matches human behavioral data in a social decision-making game that was analyzed both in discrete-time and continuous-time setups. Furthermore, unlike previous approaches, our model takes into account the role of sensorimotor control loops in embodied decision-making scenarios. For this purpose, we introduce the Control-based Reinforcement Learning (CRL) model. CRL is grounded in the Distributed Adaptive Control (DAC) theory of mind and brain, where low-level sensorimotor control is modulated through perceptual and behavioral learning in a layered structure. CRL follows these principles by implementing a feedback control loop handling the agent's reactive behaviors (pre-wired reflexes), along with an adaptive layer that uses reinforcement learning to maximize long-term reward. We test our model in a multi-agent game-theoretic task in which coordination must be achieved to find an optimal solution. We show that CRL is able to reach human-level performance on standard game-theoretic metrics such as efficiency in acquiring rewards and fairness in reward distribution. | cs.MA | cs |
MODELING THE FORMATION OF SOCIAL CONVENTIONS FROM
EMBODIED REAL-TIME INTERACTIONS
A PREPRINT
Institute for Bioengineering of Catalonia
Ismael T. Freire
SPECS Lab
Barcelona, Spain
Clement Moulin-Frier
DTIC
Universitat Pompeu Fabra
Barcelona, Spain
Marti Sanchez-Fibla
DTIC
Universitat Pompeu Fabra
Barcelona, Spain
[email protected]
[email protected]
[email protected]
Institute for Bioengineering of Catalonia
Institute for Bioengineering of Catalonia, ICREA
Xerxes D. Arsiwalla
SPECS Lab
Barcelona, Spain
[email protected]
Paul F.M.J. Verschure
SPECS Lab
Barcelona, Spain
[email protected]
January 3, 2020
ABSTRACT
What is the role of real-time control and learning in the formation of social conventions? To answer
this question, we propose a computational model that matches human behavioral data in a social
decision-making game that was analyzed both in discrete-time and continuous-time setups. Further-
more, unlike previous approaches, our model takes into account the role of sensorimotor control
loops in embodied decision-making scenarios. For this purpose, we introduce the Control-based
Reinforcement Learning (CRL) model. CRL is grounded in the Distributed Adaptive Control (DAC)
theory of mind and brain, where low-level sensorimotor control is modulated through perceptual and
behavioral learning in a layered structure. CRL follows these principles by implementing a feedback
control loop handling the agent's reactive behaviors (pre-wired reflexes), along with an adaptive layer
that uses reinforcement learning to maximize long-term reward. We test our model in a multi-agent
game-theoretic task in which coordination must be achieved to find an optimal solution. We show that
CRL is able to reach human-level performance on standard game-theoretic metrics such as efficiency
in acquiring rewards and fairness in reward distribution.
Keywords Cognitive Modeling · Multi-Agent Reinforcement Learning · Game Theory · Social Decision-Making ·
Embodied Cognition · Artificial Intelligence · Behavior-based Robotics
1
Introduction
Our life in society is often determined by social conventions that affect our everyday decisions. From the side of the
road we drive on to the way in which we greet each other, we rely on social norms and conventions to regulate these
interactions in the interest of group coordination. But what is a convention, how is it formed and maintained over time?
What cognitive system does an individual need in order to form and maintain such apparently complex behavior? If we
aim to integrate robots or intelligent machines into our daily lives, we have to provide them with cognitive models that
are able to learn and adapt to our social norms and practices.
After decades of research on the topic, the boundaries and relationship between the different categories of social norms
are still under debate [1]. However, both recent [2] and classic [3] literature do agree on their definition: conventions are
patterns of behavior that emerge within a group to solve a repeated coordination problem. More concretely, conventions
exhibit two characteristic features [3]: (i) they are self-sustaining and (ii) they are largely arbitrary. Self-sustaining, in
A PREPRINT - JANUARY 3, 2020
the sense that a group of agents in a given population will continue to conform to a particular convention as long as
they expect the others to do so; and arbitrary, in the sense that there are other equally plausible solutions to solve the
same problem. Identifying the set of conditions that lead to the formation of such conventions is still an open question,
traditionally studied through coordination games, a sub-domain of game theory [4].
A typical case for a coordination game is the so-called 'Choosing Sides'. This game proposes a situation in which two
drivers meet in the middle of a narrow road. Both drivers must make a decision to avoid a fatal collision: whether
to turn right or left. If both choose to turn to the same side they will manage to dodge each other, but if they choose
differing maneuvers they will collide. The payoff matrix of Table 1 represents numerically this situation.
Left
10, 10
0, 0
Right
0, 0
10, 10
Left
Right
Table 1: Choosing Sides payoff matrix
The solutions to this type of matrix-form game satisfy the two criteria of a convention. First, there are more that one
possible solution (both players choose Left or both choose Right), so choosing one or the other is an arbitrary decision.
And once a solution is reached, it is more optimal for each player to keep their current decision than to change it
unilaterally, so it is also self-sustaining.
Although classical matrix-form games have been extensively investigated in literature over the past decades [5, 6],
several studies point to the fact that coordination in realistic social circumstances usually requires a continuous exchange
of information in order for conventions to emerge [5, 7, 8]. Precisely, this is a feature that classical matrix-form
games lack because they are based on discrete-time turns that impose a significant delay between actions [9, 8, 10].
In order to address this problem, recent literature has devised ways to modify standard game theoretic discrete-time
tasks into dynamic versions where individuals can respond to the other agent's actions in real or continuous-time
[11, 12, 13, 14, 15, 16]. Their results point out that cooperation can be more readily achieved in the dynamic version of
the task due to the rapid flow of information between individuals and their capacity to react in real-time [12, 17].
A recent example of such an ecological approach can be found in [18], where Hawkins and Goldstone show that
continuous-time interactions help to converge to more stable strategies in a coordination game (the Battle of the Exes)
compared to the same task modeled in discrete-time. They also show that the involved payoffs affect the formation of
social conventions. According to these results, they suggest that real-life coordination problems can be solved either by
a) forming a convention or b) through spontaneous coordination. And, critically, the solution depends on what is at
stake if the coordination fails. To illustrate this point, they suggest two real-life examples of a coordination problem:
On the one hand, when we drive a car, the stakes are high because if we fail to coordinate the outcome could be fatal, so
we resort to a convention -- e.g. to drive on the right side of the road. On the other hand, when we try to avoid people in
a crowded street, we do it "on the fly" because the stakes are low, so it is not risky to rely on purely reactive behaviors
(e.g. avoidance behavior) to solve it.
However, despite the substantial amount of behavioral studies and the advances in the development of more ecologically
valid setups, a complete theory of how embodied agents coordinate in real time is still missing. One fundamental step
in this direction will be a model that can account for how lower-level dynamic processes interact with higher-level
(strategic) cognitive processes.
In this work we formalize high-level strategic mechanisms as a model-free RL algorithm, and we formalize low-level
sensorimotor mechanisms as simple control loops. Finally, we develop a model, called Control-based Reinforcement
Learning (CRL), that integrates these two mechanisms as layers in a larger architecture. To do so, we draw upon the
Distributed Adapative Control theory (DAC) [19, 20, 21], that proposes that cognition is based on several control layers
operating at different levels of abstraction. DAC makes explicit the distinction between real-time control on the one
hand (Reactive layer) and perceptual and behavioral learning on the other hand (Adaptive layer). It is, therefore, an
adequate theoretical framework for understanding the specific roles of these two principles (low-level sensorimotor
control and high-level strategic learning) in the formation of social conventions, which is the aim of this paper.
In summary, we introduce a novel two-layered cognitive architecture -CRL- that integrates a low-level reactive control
loop to manage within-round conflicts on rapid time scales, along with a policy learning algorithm to acquire across-
round strategies over longer time scales. We implement this computational model in embodied cognitive agents
involved in a social decision-making task called the Battle of the Exes. We compare performance metrics of the CRL
model to results of human behavioral data published in [18]. We run simulations showing that the modeled cognitive
agents rely more on high-level strategic mechanisms when the stakes of the game are higher. We also show that
low-level sensorimotor control helps enhance performance in terms of efficiency and fairness. These results provide
a computational hypothesis explaining key aspects of the emergence of social conventions such as how cognitive
2
A PREPRINT - JANUARY 3, 2020
processes operating across different temporal scales interact. Finally, we also provide new experimental predictions to
be tested on human behavioral studies.
1.1 Related Literature
As for computational modeling of game-theoretical tasks, there is an extensive body of literature where the study
of the emergence of conflict and cooperation in agent populations has been addressed, especially through the use of
Multi-Agent Reinforcement Learning (for extensive reviews, check [22, 23, 24]). In this direction, a lot of focus has
been recently directed towards developing enhanced versions of the Deep Q-Learning Network architecture proposed
in [25], particularly on their extensions to the social domain [26, 27, 28, 29]. This architecture uses a reinforcement
learning algorithm that extracts abstract features from raw pixels through a deep convolutional network. Along those
lines, some researchers [26, 27, 28] are modeling the type of conflicts represented in the classic game-theoretic tasks
into more ecologically valid environments [26] where agent learning is based on deep Q-networks [27, 28]. For instance,
agents based on this cognitive model are already capable of learning how to play a two-player video game such as Pong
from raw sensory data and achieve human-level performance [25], both in cooperative and competitive modes [29].
Other similar approaches have focused on constructing agent models that achieve good outcomes in general-sum games
and complex social dilemmas, by focusing on maintaining cooperation [30], by making an agent pro-social (taking into
account the other's rewards) [31] or by conditioning its behavior solely on its outcomes [32].
However, in all of the above cases, the games studied involve social dilemmas that only provide one single cooperative
equilibrium, whereas the case we study in this paper provides several ones, a prerequisite (arbitrariness) for studying
the formation of conventions. Also, the abovementioned examples relax one key assumption of embodied agents, that
is, that sensory inputs must be obtained through one's own bodily sensors. Agents in previous studies gather their
sensory data from a third person perspective. They are trained using raw pixel data from the screen, in either completely
observable [29, 30, 31, 32] or partially observable [27, 28] conditions. Another point of difference between previous
approaches and the work presented here relates to the continuity of the interaction itself. Most of the work done so
far in multi-agent reinforcement learning using game theoretical setups has been modeled using grid-like or discrete
conditions [26, 27, 28, 30, 31, 32]. Although there has been progress insofar many of these studies provide a spatial
and temporal dimension (situatedness) to many classical games, they still lack continuous time properties of real-world
interactions.
Still, there are a few recent cases where the coordination task has been modeled in real-time and the agents are situated
[29, 33, 34]. However, these models suffer from the so-called sample-inefficiency problem due to the huge amount
of episodes they require to reach human level performance. A recent review on Deep Reinforcement Learning [35]
points out that one way to solve the sample-inefficiency problem would be to integrate architectural biases that help
to bootstrap the learning mechanisms by providing some pre-wired adaptation to the environment the agent will live
in. To tackle this issue, the Control-based Reinforcement Learning (CRL) model we introduce in this paper integrates
lower-level sensorimotor control loops that help to bootstrap policy learning on the higher level of the cognitive
architecture. Moreover, we show that the CRL model is sample efficient, by comparing our results to the experimental
human data collected in [18].
In order to do that, in the next section first we begin by describing the benchmark task, a coordination game called the
Battle of the Exes [18]. After that, we present the CRL architecture and its two layers: one dealing with the low-level
intrinsic behaviors of the agent and another based on model-free reinforcement learning, allowing the agents to acquire
rules for maximizing long-term reward [36]. In the Results section, we first compare the results of our model against
the benchmark human data and then we show the contribution of each layer by performing several ablation studies.
Finally, we conclude this paper by discussing the main implications of our findings, and also comment on limitations
and possible extensions of the current model.
2 Methods
2.1 Behavioral Benchmark
The Battle of the Exes is a coordination game similar to the classic Battle of the Sexes [37], that imposes the following
social scenario: A couple just broke up and they do not want to see each other. Both have their coffee break at the same
time, but there are only two coffee shops in the neighborhood: one offers great coffee whereas the other, average coffee.
If both go to the great coffee shop they will come across each other and will not enjoy the break at all. Therefore, if
they want to enjoy their coffee break, they will have to coordinate in a way that they avoid each other every day. This
situation can be modeled within the framework of game theory with a payoff relation such as a > b > 0; where a is the
3
A PREPRINT - JANUARY 3, 2020
payoff for getting the great coffee, b the payoff for the average coffee and 0 the payoff for both players if they go to the
same location.
In [18], Hawkins and Goldstone perform a human behavioral experiment based on the above-mentioned game to
investigate how two factors -- the continuity of the interaction (ballistic versus dynamic) and the stakes of the interaction
(high versus low condition) -- affect the formation of conventions in a social decision-making task. Concerning the
stakes of the interaction, the payoff matrix is manipulated to create two different conditions: high and low, based on
a bigger and smaller difference between rewards, respectively. The payoff matrices in Figure 1 illustrate these two
conditions.
Figure 1: Payoff matrices of the original "Battle of the Exes" game. The numbers indicate the reward received
by each player (red and blue). Reproduced from [18].
As for the continuity of the interaction, the experiment has a ballistic and a dynamic condition. In the ballistic
condition, as in classical game theory, the players can only choose an action at the beginning of every round of the
game, without any further control on the outcome. However, in the dynamic condition, the players can freely change
the course of their avatars until one of them reaches a reward (for a visual example of the difference between conditions,
check the original videos here). In both conditions, the round ends when one of the players reaches one of the reward
spots that represent the coffee shops. Altogether, this results in four conditions: two for the stakes of the interaction
(high vs. low) combined with two for the continuity of the interaction (ballistic vs. dynamic). For the experiment, they
pair human players in dyads that depending on the payoff condition, play 50 (high) or 60 (low) consecutive rounds
together. In order to analyze the coordination between the players of each dyad, they use three measures -efficiency,
fairness, and stability- based on Binmore's three levels of priority [38]:
• Efficiency -- It measures the cumulative sum of rewards that players were able to earn collectively in each
round, divided by the total amount of possible rewards. If the efficiency value is 1, it means that the players
got the maximum amount of reward.
• Fairness -- It quantifies the balance between the earnings of the two players. If the fairness value is 1, it means
that both players earned the higher payoff the same amount of times.
• Stability -- It measures how well the strategy is maintained over time. In other words, it quantifies how
predictable the outcomes are of the following rounds based on previous results by "using the information-
theoretic measure of surprisal, which Shannon defined as the negative logarithm of the probability of an event"
[18].
In other words, Efficiency measures utility maximization, Fairness measures the amount of cooperation, and Stability
measures the speed and robustness of conventions formed. The results show that players in the dynamic condition
achieve greater efficiency and fairness than their counterparts in the ballistic condition, both in the high payoff and
low payoff setups. However, their key finding is that in the dynamic condition, the players coordinate more "on the
fly" (i.e. without the need of a long-term strategy) when the payoff is low, but when the payoff is high, the participants
coordinate into more stable strategies. Namely, they identified the stakes of the interaction as a crucial factor in the
formation of social conventions when the interaction happens in real-time.
4
A PREPRINT - JANUARY 3, 2020
2.2 Control-Based Reinforcement Learning
In this section, we introduce our Control-based Reinforcement Learning (CRL) model. The CRL is composed of two
layers, a Reactive and an Adaptive layer. The former governs sensorimotor contingencies of the agent within the rounds
of the game, whereas the latter is in charge of learning across rounds. This is an operational minimal model, where
reinforcement learning interacts with a feedback controller by inhibiting specific reactive behaviors. The CRL is a
model-free approach to reinforcement learning, but with the addition of a reactive controller (for model-based adaptive
control see [39]).
2.2.1 Reactive Layer
The Reactive Layer (RL) represents the agent's sensorimotor control system and is supposed to be pre-wired (typically
from evolutionary processes in a biological perspective [35]). In the Battle of the Exes game that we are considering
here, we equip agents with two predefined reactive behaviors: reward seeking and collision avoidance. This means that,
even in the absence of any learning process, the agents are intrinsically attracted to the reward spots and avoid colliding
between each other. This intrinsic dynamic will bootstrap learning in the Adaptive Layer, as we shall see.
To model this layer, we follow an approach inspired by Valentino Braitenberg's Vehicles [40]. These simple vehicles
consist of just a set of sensors and actuators (e.g. motors) that, depending on the type of connections created between
them, can perform complex behaviors. For a visual depiction of the two behaviors (reward seeking and collision
avoidance), see this video.
• The reward seeking behavior is made by a combination of a crossed excitatory connection and a direct
inhibitory connection between the reward spot sensors (s) and the motors (m), plus a forward speed constant
f set to 0.3,
mlef t = f + sX
mright = f + sX
right − sX
lef t − sX
right
lef t
(1)
(2)
where sX
lef t is the sensor positioned on the left side of the robot indicating the proximity of a reward spot,
and X is either the high (H) or the low reward (L) sensor. The sensors perceive the proximity of the spot.
The closer the reward spots, the higher the sensors will be activated. Therefore, if no reward spot is detected
(sX
right = 0), the robot will go forward at speed f. Otherwise, the most activated sensor (left or right)
will make the robot turn in the direction of the corresponding reward spot.
lef t = sX
• The collision avoidance behavior is made by the opposite combination: a direct excitatory connection and a
crossed inhibitory connection, but in this case between the agent sensors (sA) and the motors (m),
mlef t = f + sA
mright = f + sA
lef t − sA
right − sA
right
lef t
(3)
(4)
where sA
lef t is the sensor positioned on the left side of the robot indicating the proximity of the other agent.
The closer the other agent, the higher the sensors will be activated. In this case as well, if no agent is detected
(sA
right = 0), the robot will go forward at the speed f. Otherwise, the most activated sensor will make
the robot turn in the opposite direction of the other agent, thus avoiding it.
lef t = sA
2.2.2 Adaptive Layer
The agent's Adaptive layer (AL) is based on a model-free reinforcement learning algorithm that endows the agent with
learning capacities for maximizing long-term reward. Functionally, it determines the agent's action at the beginning of
the round, based on the state of the previous round and its policy. The possible states S are three: high, low and tie; and
they indicate the outcome of the previous round for each agent. That is, if an agent got the high reward on the previous
round, the state is high; if it got the low reward, the state is low; and if both agents went to the same reward, the state is
tie. The actions A are three as well: go to the high, go to the low and none.
The Adaptive Layer implements reinforcement learning for maximizing accumulated reward over rounds through action,
similar to the one implemented in [41] and adapted to operate on discrete state and action spaces. More specifically, we
use an Actor-Critic Temporal Difference Learning algorithm (TD-learning), which is based on the interaction between
two main components:
5
A PREPRINT - JANUARY 3, 2020
Figure 2: Representation of the Control-based Reinforcement Learning (CRL) model. On top, the Adaptive
layer (reinforcement learning control loop) composed of a Critic or value function (V ), an Actor or action policy
(P ), and an inhibitor function (i). At the bottom, the Reactive layer (sensorimotor control loop), composed
of three sets of sensors sH, sL, sA (corresponding to High/Low reward and the other Agent, respectively),
three functions f H, f L, f A (corresponding to High/Low reward seeking and collision avoidance behaviors,
respectively) and two motors ml, mr (corresponding to the left and right motors). The action selected by the AL
is passed through the inhibitor function that will turn off one of the attraction behaviors of the RL depending on
the action selected. If the action is go to the high, the low reward seeking reactive behavior will be inhibited. If
the AL selects go to the low, the RL will inhibit its high reward seeking behavior. If the AL selects none, the RL
will act normally without any inhibition.
• an Actor, or action policy , which learns the mapping from states (s ∈ S) to actions (a ∈ A) and defines what
the action (a) is, based on a probability (P ), to be performed in each state (s);
(5)
(6)
• and a Critic, or value function Vπ(s), that estimates the expected accumulated reward (E[R]) of a state (s)
π(as) = P (a = ats = st−1)
π : S × A → [0, 1]
following a policy;
γir(st+i+1)]
where γ ∈ [0, 1] is the discount factor, and r(si) is the reward at step i.
Vπ(st) = E[R] = E[
∞(cid:88)
i=0
(7)
The Critic also estimates if the Actor performed better or worse than expected, by comparing the observed reward with
the prediction of Vπ(s). This provides a learning signal to the actor for optimizing it, where actions performing better
(resp. worse) than expected are reinforced (resp. diminished). This learning signal is called the temporal-difference
error (TD error). The TD error e(st−1) is computed as a function of the prediction from value function Vπ(s) and the
currently observed reward of a given state r(st),
(8)
where γ is a discount factor that is empirically set to 0.40. When e(s) > 0 (respectively e(s) < 0), this means that the
action performed better (resp. worse) than expected. The TD error signal is then sent both to the Actor and back to the
Critic for updating their current values.
e(st−1) = r(st) + γVπ(st) − Vπ(st−1)
6
A PREPRINT - JANUARY 3, 2020
Figure 3: Panel A: Top view of an agent's body, as represented by the dark-blue large circle. Here the agent is
facing the top of the page. The two thin black rectangles at the sides represent the two wheels, controlled by their
speed. On its front, the agent is equipped with three types of sensors. A: agent sensors (sensing the proximity of
the other agent), L: low reward sensors, and H: high reward sensors. For each type, the agent is able to sense
the proximity of the corresponding entity both on its left and right side (hence six sensors in total). Panel B:
Screenshot of the experimental setup (top view). In blue, the two cognitive agents in their initial position at the
start of a round. In green, the two reward spots; the bigger one representing the high reward and the smaller, the
low reward (i.e. lower payoff). In white, the circles that delimit the tie area.
The Critic (value function) is updated following,
Vπ(st−1) = Vπ(st−1) + ηe(st−1)
(9)
where η is a learning rate that is set to 0.15.
The update of the Actor is done in two steps. First, a matrix C(at, st−1), with rows indexed by discrete actions and
columns by discrete states, is updated according to the TD error,
C(at, st−1) = C(at, st−1) + δe(st−1)
(10)
where δ is a learning rate that is set to 0.45, at is the current action and st−1 the previous state. C(at, st−1) integrates
the observed TD errors when executing the action at in the state st−1. It is initialized to 0 for all at, st−1 and kept to a
lower bound of 0. C(at, st−1) is then used for updating the probabilities by applying Laplace's Law of Succession [42],
P (A = atS = st−1) =
(cid:0)(cid:80)
a∈A C(at, st−1)(cid:1) + k
C(at, st−1) + 1
(11)
where k is the number of possible actions.
Laplace's Law of Succession is a generalized histogram (frequency count) where it is assumed that each value has
already been observed once prior to any actual observation. By doing so it prevents null probabilities (when no data has
been observed, it returns a uniform probability distribution). Therefore, the higher C(at, st−1), the more probable at
will be executed in st−1. Using these equations, actions performing better than expected (e(s) > 0) will increase their
probability to be chosen the next time the agent will be in state st−1. When e(s) < 0, the probability will decrease. If
this probability distribution converges for both agents, we consider that a convention has been attained.
2.3 Multi-Agent Simulations
We follow, as in the Battle of the Exes benchmark [18], a 2x2 between-subjects experimental design. One dimension
represents the ballistic and dynamic versions of the game, whereas the other dimension is composed of the high and
low difference between payoffs. Each condition is played by 50 agents who are paired in dyads and play together 50
rounds of the game if they are in one of the high payoff conditions (ballistic or dynamic), or 60 rounds if they are
in one of the low payoff conditions. Regarding the task, we have developed the two versions (ballistic and dynamic)
of the Battle of the Exes in a 2D simulated robotic environment (see Figure 3B for a visual representation). The
source code to replicate this experiment is available online at: https://gitlab.com/specslab/neurorobotics/
control-reinforcement-learning.
7
A PREPRINT - JANUARY 3, 2020
In the ballistic condition, where there is no possibility of changing the action chosen at the beginning of the round,
agents only use the Adaptive layer to operate. The two first actions (high and low) will take the agent directly to the
respective reward spots, while the none action will choose randomly between them. In each round, the action at chosen
by the AL is sampled according to P (A = atS = st), where st is the actual state observed by the agent.
In the dynamic condition, the agent uses the whole architecture, with the Adaptive and the Reactive layer working
together (see Figure 2). As in the previous condition, the agent's AL chooses an action at the beginning of the round,
based on the state of the previous round and its policy. This action is then signaled to the RL, that will inhibit the
opposite reward-attraction reactive behavior according to the action selected by the AL. In the case that the AL chooses
the action go to the high, the RL will inhibit the low reward seeking behavior, allowing the agent to focus only on the
high reward. Conversely, if the AL chooses the action go to the low, the reactive attraction to the high reward will be
inhibited. In both cases, the agent avoidance reactive behavior still operates. Finally, if the action none is selected,
instead of choosing randomly between the other two actions as in the ballistic condition, the AL will rely completely on
the behaviors of the RL to play that round of the game.
The rules of the game are as follows: A round of the game finishes when one of the agents reaches a reward spot. If
both agents are within the white circle area when this happens, the result is considered a tie, and both get 0 points. The
small spot always gives a reward of 1, whereas the big spot gives 2 or 4 depending on the payoff condition (low or high
respectively, see Figure 1). The reward spots are allocated randomly between the two positions at the beginning of each
round.
3 Results
We report the main results of our model simulations in relation to human performance in the Battle of the Exes task
[18], which are analyzed using: efficiency, fairness, and stability [38]. For each of these measures, we report the results
of the model and plot them in contrast with human data from [18]. Then, we interpret those results and analyze the role
of each layer of the CRL architecture in relation to the data obtained in each condition.
Figure 4: Results of Control-based Reinforcement Learning and TD-learning compared to human performance
in the Battle of the Exes game, measured by Efficiency (left), Fairness (center) and Stability (right). The top
panel shows the results on the high-payoff condition. The bottom panel shows the results on the low-payoff
condition. Within each panel, blue bars represent the results in the ballistic condition, and red bars represent the
results in the dynamic condition. Human data from [18]. All error bars reflect standard errors.
Regarding the efficiency scores on the low-payoff condition (see Figure 4, bottom-left panel), first, a non-parametric
Kruskal-Wallis H-test was performed, showing a statistically significant difference between groups (H(3) = 98.9, p <
.001). Post-hoc Mann-Whitney U-tests showed that there were significant differences in efficiency (p < .001) between
8
A PREPRINT - JANUARY 3, 2020
humans playing the ballistic conditions (M = 0.70) of the game and the TD-learning benchmark algorithm (M = 0.45).
However, there were no significant differences (p = .34) between human scores in the dynamic condition (M = 0.85)
and the scores achieved by the CRL model (M = 0.86). The same statistical relationships are maintained in the
high-payoff condition (H(3) = 102.29, p < .001), where human ballistic scores (M = 0.69) and TD-learning scores
(M = 0.46) were significantly different (p < .001), while the CRL model (M = 0.88) shows no statistical difference
(p = .26) with human dynamic scores (M = 0.84).
As for the fairness scores on the low-payoff condition (see Figure 4, bottom-center panel), a non-parametric Kruskal-
Wallis showed no statistically significant difference between groups (H(3) = 5.35, p < .001), which means that both
TD-learning (M = 0.61) and the CRL model (M = 0.69) matched human scores on this metric in its respective
ballistic and dynamic conditions (M = 0.61, M = 0.69). The result is similar for the high-payoff condition (see Figure
4, top-center panel). Although this time the Kruskal-Wallis H-test showed a significant difference between groups
(H(3) = 18.74, p < .001), the post-hoc analysis showed no statistical difference (p = .78) between human ballistic
condition (M = 0.50) and TD-learning (M = 0.50), nor between human dynamic condition (M = 0.69) and CRL
(M = 0.68, p = .04).
On the stability metric, the results of the four conditions showed a non-Gaussian distribution, so a non-parametric
Kruskal-Wallis H-test was performed that showed a statistically significant difference between groups (H(3) =
2385.35, p < .001)). The post-hoc Mann-Whitney U-tests showed that both the differences between human ballistic
condition (M = 0.61) and TD-learning (M = 1.18), and between human dynamic condition (M = 0.61) and
CRL model (M = 1.17), were statistically significant (p < .001 on both cases). On the high-payoff condition, a
Kurskal-Wallis also showed significant differences among all stability scores (H(3) = 2569.62, p < .001). Post-hoc
Mann-Whitney U-tests confirmed the statistical difference (p < .001) between human ballistic scores (M = 0.61) and
TD-learning (M = 1.16). Similarly, human dynamic scores (M = 0.56) were significantly smaller (p < .001) than
the ones obtained by the CRL model (M = 1.09).
3.1 Analysis
Overall, the model achieved a good fit with the benchmark data. Like in the human experiment, we observe that the
dynamic (real/continuous-time) version of the model achieves better results in efficiency and fairness and that this
improvement is consistent regardless of the manipulation of the payoff difference.
Figure 5: Top panel: Outcomes of two dyads of CRL agents (dyad 45 on the left, dyad 25 on the right) in the
high dynamic condition, showing the formation of turn-taking (left) and pure dominance (right) equilibria. Each
bar represents the outcome of a round of the game. A red bar means that player 1 got the high reward, and a
blue bar means that player 2 got the high reward. Black bars represent ties. Bottom panel: Surprisal measure
over rounds of play. When a convention is formed, the surprise drops down because the outcomes start to be
predictable.
9
A PREPRINT - JANUARY 3, 2020
The remarkable results in efficiency of the CRL model are due to the key role of the Reactive Layer in avoiding
within-round conflict when both agents have chosen to go to the same reward, a feature that a ballistic model such as
TD-learning lacks. The reactive behavior exhibited by the CRL model represents a kind of 'fight or flight' response that
can be triggered to make the agent attracted or repulsed to other agents, depending on the context that it finds itself in.
In this case, due to the anti-coordination context presented in the Battle of the Exes, the reactive behavior provides the
agent with a fast (flight) mechanism to avoid conflict. But in a coordination game like the Battle of the Sexes, this same
reactive behavior could be tuned to provide an attraction (fight) response towards the other agent. Future work will
extend this model to observe how the manipulation of this reactive behavior can be learned to help the agent in both
cooperative and competitive scenarios.
As for the results in stability, the model was overall less stable than the human benchmark data, although it reflected
a similar relation between payoff conditions: an increase in stability in the high dynamic condition (M = 1.09 and
M = 1.17) compared to the low dynamic (see Figure 4, right panels). Nonetheless, our results show that social
conventions, such as turn-taking and dominance, can be formed by two CRL agents, as shown in Figure 5. The examples
shown in the figure illustrate how these two conventions were formed in the dynamic high condition, where these
type of equilibria occurred more often and during more rounds than in the other three conditions, thus explaining the
higher stability in this condition. Overall, this results are consistent with human data in that dynamic, continuous-time
interactions help converge to more efficient, fair and stable strategies when the stakes are high.
Model Comparison
Now we analyze the specific contributions of the each layer to the overall results of the CRL architecture. In order
to do that, we perform two model-ablation studies, where we compare the results of the whole CRL model against
versions of itself operating with only one of its two layers. In the first model ablation, we deactivate the Adaptive layer,
so the resulting behavior of the agents is entirely driven by the Reactive layer. In the second model ablation, we do the
opposite so the only layer working is the Adaptive Layer. As in the main experiment, there are two payoff conditions
(high and low) and 50 dyads per condition.
Figure 6: Top panel: Results of the model-ablation experiment compared to the complete CRL results. Red
bars shows the results of the high-payoff conditions, whereas the orange bars refer to the low-payoff conditions.
The ablated model operates using only the Reactive layer's sensorimotor control. Bottom panel: Results of the
adaptive-only model compared to the complete CRL results. Dark blue shows the results of the high-payoff
conditions, whereas the light blue bars refer to the low-payoff conditions. The adaptive-only operates using
only the Adaptive layer's TD-learning algorithm. All results are represented in terms of Efficiency (left panel),
Fairness (center) and Stability (right panel). Note that stability is measured by the level of surprisal, which
means that lower surprise values imply higher stability. All error bars reflect standard errors.
10
A PREPRINT - JANUARY 3, 2020
Agents exclusively dependent on the one layer perform worse overall, with a significant drop in efficiency. This drop is
caused by a higher amount of rounds that end up in ties, in which both agents do not receive any reward. The results in
Fairness are comparable to the ones of CRL model. However, note that these results are computed from fewer rounds,
precisely due to the high amount of ties obtained (fairness computes how evenly the high reward is distributed among
agents). Regarding stability, we observe that it is lower than that obtained by the full CRL model, as demonstrated by
higher values in surprise in Figure 6. In summary, we find that any of the layers working alone leads to more unstable
and less efficient results.
In a way, these reactive-only and adaptive-only versions of the model instantiate two different approaches of modeling
cognition and artificial intelligence [43, 44]. On the one side we have the adaptive-only model implementing a TD-
learning algorithm. It represents the symbolic AI tradition, since it works with symbols on a discrete state space [45, 46].
On the other side, the reactive-only model instantiates a pure embodied approach. This model relies only on low-level
sensorimotor control loops to guide the behavior of the agent. Therefore, it represents the bottom-up approach to
cognition of the behavior-based robotics tradition [47, 48]. But, as we have seen, these models alone are not sufficient
to reach human-level performance. Moreover, none of them can match the combination of high-level strategy learning
and embodied dynamics shown by the complete CRL model.
Agents rely more on the Adaptive Layer when stakes are high
We now analyze the participation of each CRL layer across different payoff conditions through the measurement of the
"none" action, which refers to the case when the Adaptive Layer is not used during that trial. Based on the results of the
benchmark and the CRL model in the dynamic condition, where higher payoff differences helped to achieve higher
stability, we expect that the more we increase this difference between payoffs, the more the agents will rely on the
Adaptive layer. For testing this prediction, we have performed a simulation with six different conditions with varying
levels of difference between payoffs (high vs. low reward value), from 1-1 to 32-1. To measure the level of reliance on
each layer, we logged the number of times each agent outputted a none action, that is the action in which the agent
relies completely on the Reactive layer to solve the round.
Figure 7: Mean of the percentage of adaptive layer actions (ie. go to the high and go to the low actions) selected
by the agents plotted against 6 conditions with an increasing difference between high and low payoffs. Bars
reflect standard errors.
Considering that there are only 3 possible actions ('go high', 'go low', 'none'), if the Adaptive layer is randomly
choosing the actions, we should observe that the agent selects each action, on average, the same amount of times. That
means that prior to any learning, at the beginning of each dyad, the reliance on the Reactive layer would be 33% and
the reliance on the Adaptive layer 66%. Starting from this point, if our hypothesis is correct, we will expect to observe
an increase in the reliance on the Adaptive layer as the payoff difference increases. As expected, the results confirm,
as seen in Figure 7, that there is a steady increase in the percentage of selection of the Adaptive layer as the payoff
difference augments.
4 Discussion
We have investigated the role of real-time control and learning on the formation of social conventions in a multi-agent
game-theoretic task. Based on principles of distributed adaptive control theory, we have introduced a new Control-based
11
A PREPRINT - JANUARY 3, 2020
Reinforcement Learning (CRL) cognitive architecture. The CRL model uses a model-free approach to reinforcement
learning, but with the addition of a reactive controller. The CRL architecture is composed of a module based on an
actor-critic TD learning algorithm that endows the agent with learning capacities for maximizing long-term reward, and
a low-level sensorimotor control loop handling the agent's reactive behaviors. This integrated cognitive architecture
is applied to a multi-agent game-theoretic task, the Battle of the Exes, in which coordination between two agents can
be achieved. We have demonstrated that real-time agent interaction does affect the formation of more stable, fair and
effective social conventions when compared to the same task modeled in discrete-time. The results of our model are
consistent with those of Hawkins and Goldstone obtained with human subjects in [18].
Interpreting our results in the context of a functional cognitive model we have elucidated the role of reactive and adaptive
control loops in the formation of social conventions and of spontaneous coordination. We found that the Reactive layer
plays a significant role in avoiding within-round conflict (spontaneous coordination), whereas the Adaptive layer is
required to achieve across-round coordination (social conventions). In addition, the CRL model supports our hypothesis
that higher payoff differences will increase the reliance on the Adaptive layer. Based on the differences obtained
between the ballistic and dynamic conditions, our results might also suggest that initial successful interactions solved
primarily by reactive sensorimotor control can speed up the formation of social conventions.
In our simulations, we have also modeled extensions of experimental conditions (such as increasing differences between
payoffs, presented in Figure 7) which affect task outcomes as well as functionality of each control loop. These results
allow us to make predictions that can later be tested in new human experiments. More concretely, based on our
simulations, we predict that an increased difference in value between the two rewards will promote a faster convergence
towards a convention in cooperation games such as The Battle of the Exes. At the cognitive level we suggest that this
increase in convention formation could be linked to a higher level of top-down cognitive control, as predicted by the
increase in activation of the Adaptive layer of the CRL model.
Furthermore, there is a biological correspondence of the functions identified by modules of the CRL architecture.
Computations described by temporal difference learning have been found in the human brain, particularly in the ventral
striatum and the orbitofrontal cortex [49]. It has also been shown that premotor neurons directly regulate sympathetic
nervous system responses such as fight-or-flight [50]. The top-down control system of the brain has been identified in
the dorsal posterior parietal and frontal cortex, and shown to be involved in cognitive selection of sensory information
and responses. On the other hand, the bottom-up feedback system is linked to the right temporoparietal and ventral
frontal cortex and is activated when behaviorally relevant sensory events are detected [51, 52, 53].
To the best of our knowledge, this is the first embodied and situated cognitive model that is able to match human
behavioral data in a social decision-making game in continuous-time setups. Moreover, unlike previous attempts, we
take into account the role of sensorimotor control loops in solving social coordination problems in real-life scenarios.
This is arguably a fundamental requirement for the development of a fully embodied and situated AI.
Regarding the limits of the model, we observe that the CRL model still does not reach human level performance in
terms of stability. Although the model is quite sample-efficient and reaches good performance in very few trials, it is
clear that it does not learn at the same rate as humans. This is because the model-free algorithm implemented at the
Adaptive layer obviously does not capture the cognitive complexity of strategic human behavior. For instance, people
can inductively abstract to a 'turn-taking' strategy while the adaptive layer would have to separately learn policies for
the 'up' and 'down' states from scratch. This can be clearly seen when the model is compared to human performance in
the ballistic conditions, where only the Adaptive layer is active.
In the end, the CRL model represents a very minimally social model of convention formation (almost as minimal as the
naive RL model of [54]). The only way in which the existence of other agents is incorporated into decision-making is
through the reactive layer's avoidance mechanism, since its modulated by the presence of the other agent. At most, since
the state variable (s) depends on the actions selected by each agent in the previous round of the game, one could argue
that this variable implicitly takes into account information about the opponent. Besides that, the agents are agnostic
to where the rewards are coming from, and certainly not representing and updating the other agents' latent policy or
engaging in any kind of social cognition when planning.
But, how much cognitive sophistication is really needed to find solutions to social coordination problems? Our results
suggest that we do not need to invoke any advanced social reasoning capabilities for achieving successful embodied
social coordination. At least, on this type of coordination problems. Arguably in more complex social scenarios (e.g.
generalizing or transferring conventions from one environment to another) a certain level of social representation may
become necessary.
For future work, there are several directions in which we can continue to develop reserach presented in this paper.
One obvious extension would be to make the CRL model social. This could be done by representing its partner as an
12
A PREPRINT - JANUARY 3, 2020
intentional agent or trying to predict and learn what its partner is going to do (as inverse RL, or Bayesian convention
formation theories do).
Another possibility is the addition of a memory module to the CRL architecture. As discussed in [20, 36], this will
facilitate the integration of sensory-motor contingencies into a long-term memory that allows for learning of rules.
This is important for building causal models of the world and taking into account context in the learning of optimal
action policies. The goal of such extensions can be to build meta-learning mechanisms that can identify the particular
social scenario in which an agent is placed (i.e., social dilemmas, coordination problems, etc.) and then learn the
appropriate policy for each context. Extending our model with such functionality could enable solving more diverse
and complicated social coordination problems, both at the dyadic and at the population levels.
Lastly, another interesting avenue concerns the emergence of communication. We could extend our model by adding
signaling behaviors to agents and testing them in experimental setups similar to the seminal sender-receiver games
proposed by Lewis [3]. One could also follow a more robot-centric approach such as that of [55, 56]. These approaches
enable one to study the emergence of complex communicative systems embedding a proto-syntax [41, 57].
Put together, our model in this paper along with recent related work (see [44]) helps towards advancing our understanding
of a functional embodied and situated AI that can operate in a multi-agent social environment. For this purpose, we
plan to extend this model to study other aspects of cooperation such as in wolf-pack hunting behavior [58, 59], and also
aspects of competition within agent populations as in predator-prey scenarios. In ongoing work, we are developing
a setup in which embodied cognitive agents will have to compete for limited resources in complex multi-agent
environments. This setup will also allow us to test the hypothesis proposed in [60, 61, 62] concerning the role of
consciousness as an evolutionary game-theoretic strategy that might have resulted through natural selection triggered by
a cognitive arms-race between goal-oriented agents competing for limited resources in a social world.
Acknowledgements
This project has received funding from the European Union's Horizon 2020 research and innovation program under
grant agreement ID:820742 and ID:641321.
Conflict of interest
The authors declare that they have no competing interests.
References
[1] Robert XD Hawkins, Noah D Goodman, and Robert L Goldstone. The emergence of social norms and conventions.
Trends in cognitive sciences, 2018.
[2] H Peyton Young. The evolution of social norms. economics, 7(1):359 -- 387, 2015.
[3] David Lewis. Convention: a philosophical study. Cambridge, Harvard university press, 1969.
[4] John Von Neumann and Oskar Morgenstern. Game theory and economic behavior. Joh Wiley and Sons, New York,
1944.
[5] Kevin C. Clements and David W. Stephens. Testing models of non-kin cooperation: mutualism and the Prisoner's
Dilemma. Animal Behaviour, 50(2):527 -- 535, 1995.
[6] Christina Riehl and Megan E Frederickson. Cheating and punishment in cooperative animal societies. Philo-
sophical transactions of the Royal Society of London. Series B, Biological sciences, 371(1687):20150090, feb
2016.
[7] Richard C. Connor. Altruism among non-relatives: alternatives to the 'Prisoner's Dilemma'. Trends in Ecology &
Evolution, 10(2):84 -- 86, 1995.
[8] Ronald Noë. Cooperation experiments: coordination through communication versus acting apart together. Animal
Behaviour, 71(1):1 -- 18, 2006.
[9] Noam Miller, Simon Garnier, Andrew T Hartnett, and Iain D Couzin. Both information and social cohesion
determine collective decisions in animal groups. Proceedings of the National Academy of Sciences of the United
States of America, 110(13):5263 -- 8, mar 2013.
[10] Michael Taborsky, Joachim G. Frommen, and Christina Riehl. Correlated pay-offs are key to cooperation: Table 1.
Philosophical Transactions of the Royal Society B: Biological Sciences, 371(1687):20150084, feb 2016.
13
A PREPRINT - JANUARY 3, 2020
[11] Maria Bigoni, Marco Casari, Andrzej Skrzypacz, and Giancarlo Spagnolo. Time Horizon and Cooperation in
Continuous Time. Econometrica, 83(2):587 -- 616, 2015.
[12] G Sander van Doorn, Thomas Riebli, and Michael Taborsky. Coaction versus reciprocity in continuous-time
models of cooperation. Journal of theoretical biology, 356:1 -- 10, sep 2014.
[13] Daniel Friedman and Ryan Oprea. A Continuous Dilemma. American Economic Review, 102(1):337 -- 363, feb
2012.
[14] Curtis Kephart and Daniel Friedman. Hotelling revisits the lab: equilibration in continuous and discrete time.
Journal of the Economic Science Association, 1(2):132 -- 145, dec 2015.
[15] Ryan Oprea, Gary Charness, and Daniel Friedman. Continuous time and communication in a public-goods
experiment. Journal of Economic Behavior & Organization, 108:212 -- 223, 2014.
[16] Ryan Oprea, Keith Henwood, and Daniel Friedman. Separating the Hawks from the Doves: Evidence from
continuous time laboratory games. Journal of Economic Theory, 146(6):2206 -- 2225, 2011.
[17] Robert X D Hawkins. Conducting real-time multiplayer experiments on the web. Behavior research methods,
47(4):966 -- 76, dec 2015.
[18] Robert X. D. Hawkins and Robert L. Goldstone. The Formation of Social Conventions in Real-Time Environments.
PLOS ONE, 11(3):e0151670, mar 2016.
[19] Paul F. M. J. Verschure. Synthetic consciousness: the distributed adaptive control perspective. Philosophical
Transactions of the Royal Society of London B: Biological Sciences, 371(1701), 2016.
[20] Paul F M J Verschure, Cyriel M A Pennartz, and Giovanni Pezzulo. The why, what, where, when and how of
goal-directed choice: neuronal and computational principles. Philosophical transactions of the Royal Society of
London. Series B, Biological sciences, 369(1655), nov 2014.
[21] Paul F. M. J. Verschure, Thomas Voegtlin, and Rodney J. Douglas. Environmentally mediated synergy between
perception and behaviour in mobile robots. Nature, 425(6958):620 -- 624, oct 2003.
[22] L. Busoniu, R. Babuska, and B. De Schutter. A Comprehensive Survey of Multiagent Reinforcement Learning.
IEEE Transactions on Systems, Man, And Cybernetics-Part C: Applications and Reviews, 38 (2), 2008, 2008.
[23] Caroline Claus and Craig Boutilier. The dynamics of reinforcement learning in cooperative multiagent systems.
AAAI/IAAI, 1998:746 -- 752, 1998.
[24] Ming Tan. Multi-agent reinforcement learning: Independent vs. cooperative agents. In Proceedings of the tenth
international conference on machine learning, pages 330 -- 337, 1993.
[25] Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare, Alex
Graves, Martin Riedmiller, Andreas K. Fidjeland, Georg Ostrovski, Stig Petersen, Charles Beattie, Amir Sadik,
Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, Shane Legg, and Demis Hassabis. Human-
level control through deep reinforcement learning. Nature, 518(7540):529 -- 533, feb 2015.
[26] Max Kleiman-Weiner, Mark K. Ho, Joseph L. Austerweil, Michael L. Littman, and Joshua B. Tenenbaum.
Coordinate to cooperate or compete: Abstract goals and joint intentions in social interaction. COGSCI, 2016.
[27] Joel Z. Leibo, Vinicius Zambaldi, Marc Lanctot, Janusz Marecki, and Thore Graepel. Multi-agent Reinforcement
Learning in Sequential Social Dilemmas. In Proceedings of the 16th Conference on Autonomous Agents and
MultiAgent Systems, pages 464 -- 473. International Foundation for Autonomous Agents and Multiagent Systems,
2017.
[28] Julien Pérolat, Joel Z. Leibo, Vinicius Zambaldi, Charles Beattie, Karl Tuyls, and Thore Graepel. A multi-agent
reinforcement learning model of common-pool resource appropriation. In I Guyon, U V Luxburg, S Bengio,
H Wallach, R Fergus, S Vishwanathan, and R Garnett, editors, Advances in Neural Information Processing Systems
30, pages 3646 -- 3655. Curran Associates, Inc., jul 2017.
[29] Ardi Tampuu, Tambet Matiisen, Dorian Kodelja, Ilya Kuzovkin, Kristjan Korjus, Juhan Jaan Aru, Juhan Jaan Aru,
and Raul Vicente. Multiagent Cooperation and Competition with Deep Reinforcement Learning. PLOS ONE,
12(4):e0172395, nov 2015.
[30] Adam Lerer and Alexander Peysakhovich. Maintaining cooperation in complex social dilemmas using deep
reinforcement learning. Arxiv, abs/1707.0, jul 2017.
[31] Alexander Peysakhovich and Adam Lerer. Consequentialist conditional cooperation in social dilemmas with
imperfect information. Arxiv, abs/1710.0, oct 2017.
[32] Alexander Peysakhovich and Adam Lerer. Prosocial learning agents solve generalized Stag Hunts better than
selfish ones. Arxiv, abs/1709.0, sep 2017.
14
A PREPRINT - JANUARY 3, 2020
[33] Max Jaderberg, Wojciech M Czarnecki, Iain Dunning, Luke Marris, Guy Lever, Antonio Garcia Castaneda,
Charles Beattie, Neil C Rabinowitz, Ari S Morcos, Avraham Ruderman, et al. Human-level performance in 3d
multiplayer games with population-based reinforcement learning. Science, 364(6443):859 -- 865, 2019.
[34] Bowen Baker, Ingmar Kanitscheider, Todor Markov, Yi Wu, Glenn Powell, Bob McGrew, and Igor Mordatch.
Emergent tool use from multi-agent autocurricula. arXiv preprint arXiv:1909.07528, 2019.
[35] Mathew Botvinick, Sam Ritter, Jane X Wang, Zeb Kurth-Nelson, Charles Blundell, and Demis Hassabis. Rein-
forcement learning, fast and slow. Trends in cognitive sciences, 2019.
[36] Clément Moulin-Frier, Xerxes D Arsiwalla, Jordi-Ysard Puigbò, Martì Sanchez-Fibla, Armin Duff, and Paul FMJ
Verschure. Top-down and bottom-up interactions between low-level reactive control and symbolic rule learning in
embodied agents. In CoCo@ NIPS, 2016.
[37] Drew. Fudenberg, Jean. Tirole, Drew Fudenberg, and Jean Tirole. Game theory, volume 1. MIT Press, 1991.
[38] Ken Binmore. Natural justice. Oxford University Press, 2005.
[39] Ivan Herreros, Xerxes Arsiwalla, and Paul Verschure. A forward model at purkinje cell synapses facilitates
cerebellar anticipatory control. In Advances in Neural Information Processing Systems, pages 3828 -- 3836, 2016.
[40] Valentino Braitenberg. Vehicles: Experiments in synthetic psychology. MIT press, 1986.
[41] Clement Moulin-Frier, Marti Sanchez-Fibla, and Paul F.M.J. Verschure. Autonomous development of turn-
taking behaviors in agent populations: A computational study. In 2015 Joint IEEE International Conference on
Development and Learning and Epigenetic Robotics (ICDL-EpiRob), pages 188 -- 195. IEEE, aug 2015.
[42] E.T. Jaynes and G.L. Bretthorst. Probability theory : the logic of science. Cambridge University Press, 2003.
[43] George Lakoff and Mark Johnson. Philosophy in the Flesh. New york: Basic books, 1999.
[44] Clément Moulin-Frier, Jordi-Ysard Puigbo, Xerxes D Arsiwalla, Martì Sanchez-Fibla, and Paul FMJ Verschure.
Embodied artificial intelligence through distributed adaptive control: An integrated framework. arXiv preprint
arXiv:1704.01407, 2017.
[45] Allen Newell. Unified theories of cognition. Harvard University Press, 1994.
[46] John E Laird, Allen Newell, and Paul S Rosenbloom. Soar: An architecture for general intelligence. Artificial
intelligence, 33(1):1 -- 64, 1987.
[47] Rodney Brooks. A robust layered control system for a mobile robot. IEEE journal on robotics and automation,
2(1):14 -- 23, 1986.
[48] Rodney A Brooks. Intelligence without representation. Artificial intelligence, 47(1-3):139 -- 159, 1991.
[49] John P O'Doherty, Peter Dayan, Karl Friston, Hugo Critchley, and Raymond J Dolan. Temporal difference models
and reward-related learning in the human brain. Neuron, 38(2):329 -- 337, 2003.
[50] Arthur SP Jansen, Xay Van Nguyen, Vladimir Karpitskiy, Thomas C Mettenleiter, and Arthur D Loewy. Central
command neurons of the sympathetic nervous system: basis of the fight-or-flight response. Science, 270(5236):644 --
646, 1995.
[51] Maurizio Corbetta and Gordon L Shulman. Control of goal-directed and stimulus-driven attention in the brain.
Nature reviews neuroscience, 3(3):201, 2002.
[52] Etienne Koechlin, Chrystele Ody, and Frédérique Kouneiher. The architecture of cognitive control in the human
prefrontal cortex. Science, 302(5648):1181 -- 1185, 2003.
[53] Yuko Munakata, Seth A Herd, Christopher H Chatham, Brendan E Depue, Marie T Banich, and Randall C
O'Reilly. A unified framework for inhibitory control. Trends in cognitive sciences, 15(10):453 -- 459, 2011.
[54] Dale J Barr. Establishing conventional communication systems: Is common knowledge necessary? Cognitive
science, 28(6):937 -- 962, 2004.
[55] L. Steels. Language games for autonomous robots. IEEE Intelligent Systems, 16(5):16 -- 22, 2001.
[56] Luc Steels. Evolving grounded communication for robots. Trends in Cognitive Sciences, 7(7):308 -- 312, jul 2003.
[57] Clément Moulin-Frier and Paul F.M.J. Verschure. Two possible driving forces supporting the evolution of animal
communication. Physics of Life Reviews, 16:88 -- 90, mar 2016.
[58] C. Muro, R. Escobedo, L. Spector, and R.P. Coppinger. Wolf-pack (Canis lupus) hunting strategies emerge from
simple rules in computational simulations. Behavioural Processes, 88(3):192 -- 197, nov 2011.
[59] Alfredo Weitzenfeld, Alberto Vallesa, and Horacio Flores. A Biologically-Inspired Wolf Pack Multiple Robot
Hunting Model. In 2006 IEEE 3rd Latin American Robotics Symposium, pages 120 -- 127. IEEE, oct 2006.
15
A PREPRINT - JANUARY 3, 2020
[60] X D Arsiwalla, I Herreros, C Moulin-Frier, M Sanchez, and P F Verschure. Is consciousness a control process.
Artificial Intelligence Research and Development, pages 233 -- 238, 2016.
[61] Xerxes D Arsiwalla, Ivan Herreros, Clement Moulin-Frier, and Paul Verschure. Consciousness as an evolutionary
game-theoretic strategy. In Conference on Biomimetic and Biohybrid Systems, pages 509 -- 514. Springer, 2017.
[62] Xerxes D. Arsiwalla, Ricard Sole, Clement Moulin-Frier, Ivan Herreros, Marti Sanchez-Fibla, and Paul Verschure.
The Morphospace of Consciousness. Arxiv, abs/1705.11190, 2017.
16
|
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.